Alignment Change in Iranian Languages

Geoffrey L.J. Haig

Iran ian dialects are subgroups of Indo-Iranian dialects and are arranges dependent on their course of events and locale. All Iranian dialects of the center and current occasions have some normal highlights. The plain word request is generally action word last, and the time framework depends on two action word lines, present and past, no matter what. While the current stem proceeds with the Old Iranian present that was acquired straightforwardly fro m Indo-European, the previous stem depends on a participatory type of the action word finishing off with-ta. The Iran ian dialects are probably going to be spoken by in excess of 80 million individuals in a wide zone fro m Turkey with Zaza as the western most to China with Sarikoli as the western that the easternmost Iranian lan guage and primarily cover the entirety of Iran, Afghanistan and Tajikistan. The advancement of the Iranian dialects can be analyzed in three significant chronicled periods: Old Iranian (up to the fourth/third century BC), Central Iran ian (fro m the fourth/t hird century BC to the eighth/ninth century AD) and new Iranian language (since 900 AD). Two of the antiquated Iranian dialects are known and archived, Avestan and Old Persian. The Central Iranian d ialects (appro x. 300 BC-AD 950) are significantly mo re various. They are part itioned into t wo principle gatherings, western and eastern. Current Iranian d ialects fall into two enormous "Eastern" and "Western" gatherings, with "Northern" and "Southern" sub-gatherings, separately. The qualificat ion between the eastern and western Iranian d ialects lies in the fundamental geological dispersion of the princip le speakers of these dialects in the east or wes t of Lut deserts in Iran.

Aspects of Iranian Linguistics introduces readers to recent research into various properties of a number of Iranian languages. The volume consists of twenty chapters that cover a full range of Iranian linguistics, including formal theoretical perspectives (from a syntactic and morphological point of view), typological and functional perspectives, and diachronic and areal perspectives. It also contains papers on computational linguistics and neurolinguistics, as well as the modern history of lexicography in Iran. Various Iranian languages are discussed in this volume, including Hawrami and Kermanji, two of the major dialects of Kurdish, Medival, Classical and Modern Persian, Balochi, Taleshi and Pamir. With the exception of Persian, other Iranian languages had not received much attention in the past. Thus this work, as the first volume ever published on various aspects of these languages and their linguistic properties, is a valuable contribution to our understanding of a less common...

Content and English Summary of "Die Entwicklung der Ergativkonstruktion im Alt- und Mitteliranischen - Eine korpusbasierte Untersuchung zu Kasus, Kongruenz und Satzbau"

Alignment change in Iranian languages A Construction Grammar approach Geoffrey L.J. Haig 2008 Mouton De Gruyter Berlin • New York Acknowledgements Work on this book goes back almost a decade, and has been dependent on the support of a large number of people and institutions. It is my great pleasure to acknowledge their various contributions here, in very rough chronological order. Before doing so, let me first express my gratitude to those native speakers of Iranian languages who with unfailing patience and good humour so willingly shared their knowledge with me: Ismet Ramm, Sadiq Basid, Behrooz Shojai, Parwin Mahmutweyssi, Abdullah Incekan, Ergin Opengin, Mehmet Şerif Derince, and many others who prefer to remain anonymous. A two-year research position (1999–2001) funded by the Deutsche Forschungsgemeinschaft (Grant number Mo 728/2-1) enabled me to begin seriously investigating alignment in Iranian. I am deeply indebted to the project supervisor, Ulrike Mosel, for her encouragement, rigour and unfailing spirit, which have been a continuing source of inspiration ever since. A short visiting fellowship at the Research Centre for Linguistic Typology (La Trobe University) in 2003 helped me clarify many of the ideas behind chapter two; I would like to thank Alexandra Aikhenvald and Bob Dixon for their generosity in providing such a stimulating research environment and infrastructure. The fruits of those earlier endeavours were submitted to the Philosophische Fakultät of the University of Kiel as a Habilitationsschrift (Haig 2004a), and I am indebted to the members of the commission for their critical input: Ulrike Mosel, Ludwig Paul, Harald Thun and Jarich Hoekstra. In the intervening years I began to delve deeper into the history of the Iranian languages in search of answers to some of the unresolved issues. However, historical Iranian philology is extremely rugged terrain for the uninitiated; without the assistance of several experienced guides I would doubtless have gone hopelessly astray. In particular, Ludwig Paul has been, and continues to be a bountiful source of knowledge on historical Iranian linguistics. Bo Utas, Agnes Korn and Prods Oktor Skjærvø also made very significant contributions in the interpretation of the historical data, which I gratefully acknowledge here. The real breakthrough in bringing this book into its final form was made possible through a six-month research fellowship from the Swedish Collegium for Advanced Study (SCAS) in Uppsala (2005–2006). At that wonderful institution I not only found the time, but also the company and the vi Acknowledgements resources to extend the data base to include a reasonable selection of West Iranian languages, and partial coverage of Middle Iranian. In particular, I would like to thank Björn Wittrock, Barbro Klein and all the staff of SCAS for providing such a marvellously invigorating environment, and Lars Johanson and Éva Csató for organizing the interdisciplinary research group. During my time at SCAS, Don Stilo initiated me into the study of the minority languages of Iran, and has been my mentor on more matters than I care to name since. Among my other colleagues at SCAS, Prods Oktor Skjærvø, Christiane Bulut, Bo Utas, Carina Jahani and Serge Axonov also contributed in different ways to shaping the ideas put down here, though they may not recognize their respective contributions, or even agree with what I have made of them. Later versions of the manuscript benefited enormously from the critical input of Nicole Nau, Szymon Słodowicz, Agnes Korn, Thomas Juegel, Shahar Shirtz, Denise Bailey, Parwin Mahmutweyssi, Gerardo De Caro, Behrooz Shojai and an anonymous reviewer from Mouton de Gruyter. Parts of the book were presented at various venues, and I would like thank the audiences there for much critical discussion: the Research Centre for Linguistic Typology, La Trobe University, Melbourne (March 2003); LENCA 2, Kazan (May 2004); the Workshop on the Evolution of Syntactic Relations, Mainz (Feb. 2004); the Workshop on Case at Uppsala University (Jan. 2006); the First International Conference on Iranian Linguistics in Leipzig (June 2005), and the Second International Conference on Iranian Linguistics in Hamburg (August 2007). The staff at Mouton de Gruyter were extremely helpful throughout the publishing process, in particular Ursula Kleinhenz and Wolfgang Konwitschny. Christoph Eyrich mastered countless difficulties in the final formatting with admirable patience and professionalism, and Allison Peirse and Stefan Schnell did a fine job of proofreading the manuscript at very short notice. Finally, a word of gratitude to my family for their unflagging support, a source of great comfort across many years of what at times has been fairly solitary labour. Contents Acknowledgements v Abbreviations ix 1 1.1 1.2 1.3 1.4 Introduction Aims and assumptions . . . . . The Iranian languages . . . . . . Alignment in the Iranian context Constructions and syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 4 6 15 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 Alignment in Old Iranian The manā kartam construction . . Implications for diachronic syntax What is a passive? . . . . . . . . . Re-assessing the m. k. construction The semantics of the Genitive . . . Summing up the alternatives . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 26 32 36 41 55 82 86 3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 Western Middle Iranian Middle Iranian . . . . . . . . . . . Past Transitive Constructions . . . The case system . . . . . . . . . . Case and person . . . . . . . . . . Pronominal clitics . . . . . . . . Clitics expressing core arguments . Past transitive verbs . . . . . . . . Summary of Middle Iranian . . . . . . . . . . . . 89 89 92 95 101 105 112 117 128 4 4.1 4.2 4.3 Case systems in West Iranian Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . Three processes . . . . . . . . . . . . . . . . . . . . . . . . . . Innovated object markers . . . . . . . . . . . . . . . . . . . . . 131 131 133 152 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii Contents 4.4 4.5 4.6 4.7 4.8 The Tatic-type languages Explanations for change . Case and animacy . . . . Towards a solution . . . Summary of case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 172 179 193 197 5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 Kurdish (Northern Group) Introduction . . . . . . . . . . . . . Overview of the morphosyntax . . . The canonical ergative construction . Deviations from canonical ergativity Summary of deviations . . . . . . . Evidence from Badı̄nānı̄ . . . . . . Summary of the Northern Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 201 204 213 224 255 256 273 6 6.1 6.2 6.3 6.4 6.5 6.6 The Central group Introduction . . . . . . . . . . . . . . . . . . Suleimani morphosyntax . . . . . . . . . . . Past transitive constructions . . . . . . . . . . Aligning case and agreement . . . . . . . . . Summary of the Central Group . . . . . . . . Desire, Obligation, Possession, and Ergativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 277 277 288 301 304 305 7 7.1 7.2 7.3 7.4 Conclusions A brief synopsis . . . . . . . . . . . Areal pressure and alignment change Alignment in Indo-European . . . . On explanations for change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311 311 317 320 323 . . . . . . . . . . . . . . . . . . . . Appendix 333 A.1 Case in Old Persian . . . . . . . . . . . . . . . . . . . . . . . . 333 A.2 Changing rules of clitic placement . . . . . . . . . . . . . . . . 334 References Subject index Index of Iranian languages Index of non-Iranian languages 339 359 365 367 Abbreviations A A-past A-pres Subject of transitive verb A of past tense verb form A of present tense verb form ABS Absolute ACC Accusative Adj Adjective ADP Adposition Badı̄n. Badı̄nānı̄ (dialect of Kurdish, Northern Group) CLC Clitic COP Copula DAT Dative DEF Definite Det. Determiner DIR Direct (case) DIREC Directional DOM Differential Object Marking EXCL Exclamatory F Feminine FUT Fut GEN Genitive IMP Imperative IND Indicative INDEF Indefinite I NN O BJ Innovated object marker INTERR Interrogative IRR Irrealis IZ Izafe particle IZF Feminine Izafe particle IZM Masculine Izafe particle IZP Plural Izafe particle Kurm. Kurmanji (alternative name for Northern Group, Kurdish) LOC M MED MHG m. k. NEG NCS NHG NP N O O-past O-pres OBL PTC PL PLUP PP PST PTCPL PROG RECIPR REFL s S SAP TAM TSA V VP 1s/2s/3s Locative Masculine Medium (Middle Voice) Middle High German manā kartam (construction) Negation Non-Canonical Subject New High German Noun phrase Noun Object of transitive verb O of past tense verb form O of present tense verb form Oblique Past Transitive Construction Plural Pluperfect Prepositional phrase Past Participle Progressive Reciprocal Reflexive Singular Subject of intransitive verb Speech Act Participant Tense, Aspect and Modality Tense Sensitive Alignment Verb Verb phrase First person singular/ Second . . . /Third . . . Chapter 1 Introduction 1.1 Aims and assumptions The Iranian languages are currently spoken across a vast stretch of Asia, ranging from the westernmost provinces of China to Central Anatolia in Turkey. Their speakers inhabit several distinct geographic and cultural areas and have been in long-standing contact with numerous genetically diverse languages. Typologically, the languages are also highly divergent; Ossetic, for example, is analyzed as having nine nominal cases (Thordarson 1989: 469), while the Central Group of Kurdish has no nominal case marking (see Chapter six). Yet despite the deep typological rifts cross-cutting the family, there is a striking grammatical property common to the vast majority of Iranian languages: the morphosyntax associated with past transitive verbs differs from that associated with all other verbs in the language concerned. Iranian wasn’t always like this. Some two to three thousand years ago, in the Old Iranian period, the case marking and agreement patterns associated with past transitive verbs were identical to those of the present tenses. At that time, the language had, in the terminology adopted throughout this study, a unified accusative alignment in all tenses. The first issue to be addressed in this book then, is this: why is it that throughout the entirety of Iranian, the alignment associated with past transitive clauses came adrift from the rest of the grammar, and what were the actual mechanisms involved? After the initial shift away from accusative alignments in past tenses, Iranian languages have since undergone a broad spectrum of further changes. Some have completely ironed out the wrinkles in their past tenses and returned to unified accusative alignments (Persian, for example). But most have retained various kinds of hybrid constructions, still different to those of the present tenses, but which do not fit any of the various labels commonly conferred by linguists (see Section 1.3.4). A second major aim of this book, then, is to investigate the array of non-accusative alignments found in past transitive constructions in Iranian with the aim of reconstructing the pathways down which they have progressed, and identifying the underlying principles that guided their developments. 2 Introduction These questions have of course been addressed before. In the literature on alignment and diachronic syntax, a number of important studies have been devoted to explaining “the emergence of ergativity”. Obviously, the Iranian languages offer unique potential for this endeavour: alignment shifts from accusative to ergative, and back to accusative, can actually be traced in historically documented texts over some 2500 years. Furthermore, Iranian languages past and present have been richly documented across more than a century of Iranian philology, yielding a data base of staggering proportions, though of somewhat uneven quality and accessibility. Yet surprisingly, the existing accounts of alignment change in Iranian are based on a minute subset of the available data. Consequently, many of them are overly simplistic, or simply wrong. Nevertheless, most scholars within both typology and diachronic syntax have been content to accept them, and they are duly repeated in standard works on diachronic syntax (e.g., Harris and Campbell (1995), cf. Chapter two for discussion). There has been a persistent reluctance to reassess these accounts against a more representative data base, or in the light of alternative syntactic theories. This is all the more surprising given that the other well-documented case of alignment change in Indo-European, the rise of ergativity in Indo-Aryan, continues to attract intense attention from linguists of all persuasions.1 More than 20 years ago, in his pioneering study of Differential Object Marking in Iranian, Bossong (1985: 118) concludes (my translation) “the problem of ergativity [in Iranian] is in need of thorough analysis”. Ten years later Bubenik (1994: 121) urges that “more work has to be done on the morphosyntactic history of the ergative Iranian languages (Pashto, Kurmanji, Tati).” Yet even today, relatively little progress has been made in that direction. The present study is intended to go at least some way towards closing the gap by providing a more substantive treatment of alignment change in Iranian, drawing on both historical as well as comparative data from a large variety of Iranian languages. As regards the scope of the data considered, I should point out from the outset that there is a heavy bias towards the western Iranian branch of the 1 See, among many others, Klaiman (1987), Peterson (1998), Bubenik (2001), Butt (2001) and Deo and Sharma (2006). Aims and assumptions 3 family, while Eastern Iranian languages have been woefully neglected. I am painfully aware of these shortcomings, but my conviction is that genuine insights into alignment systems can only be obtained through in-depth analysis of individual languages. Undertaking a truly representative study with the depth required would have more than doubled the size of this book. Despite these gaps, I have nevertheless formulated some explicit hypotheses on the mechanisms of alignment change in Iranian. No doubt some of them will require modification when they are scrutinized against additional data. This does not, however, detract significantly from their value. We need explicit hypotheses for defining research goals, the questions to be asked when we approach the next data set. And they permit the results to be related to more general theories of syntax, and language change. Over and beyond their relevance for reconstructing the history of Iranian syntax, the hypotheses formulated here have implications for understanding the alignment of proto-IndoEuropean, the evolution of case and agreement systems, and more generally, for understanding how typology constrains the outcome of morphosyntactic change. The book is organized along the following lines: chapter one begins with an overview of the Iranian languages and the relevant features of their morphosyntax, introduces the terminology and theoretical background, and provides initial exemplification of alignment in the Iranian context. In Chapter two, alignment in Old Iranian is examined, based largely on the Old Persian texts. I argue at length against the theory that ergative alignments emerged from an agented passive construction. An alternative proposal is formulated, according to which ergativity emerged through the extension of a pre-existing non-canonical subject construction. In Chapter three, some of the developments in the Middle Iranian period are discussed, whereby the treatment is, for practical reasons, largely confined to western Middle Iranian. The focus here is on the role that pronominal clitics played in defining a specifically Iranian brand of alignment, still widespread in many Iranian languages. Chapter four presents a comprehensive account of the evolution of case systems in western Iranian, and proposes certain universal principles that shaped the developments. The discussion centres on the relative importance of different functional principles (case discrimination, cross-system harmony) and the role of animacy. Chapters five and six present in-depth case studies of alignment in Kurdish languages, where an exceptionally broad range of align- 4 Introduction ment types is concentrated in a close-knit genetic grouping. The cross-section of currently observable variation in these languages reveals uncanny echoes of processes that have been playing themselves out in the history of Iranian across two millennia. Finally, in Chapter seven I present a synopsis of the historical developments, as I conceive them, and broach a number of broader issues of diachronic syntax and typology. 1.2 The Iranian languages The Iranian languages constitute one branch of the Indo-European language family. Their closest genetic relatives are the Indo-Aryan languages; there is little doubt that the two are offshoots of a common Indo-Iranian protolanguage. The language of the oldest attested variety of Iranian, the Gatha Avestan texts, is so close to the oldest attested Indo-Aryan texts, Vedic Sanskrit, that the two may be considered almost dialectal variants (Beekes 1988: xvi, Mayrhofer 1989). The dating of the oldest attested Iranian texts, the Old Avestan (or Gatha-Avestan) texts remains controversial, with estimates ranging from the fourteenth to the fourth centuries BC (cf. Chapter two). The Old Iranian stage is also attested in the Old Persian texts, which can be reliably dated to 6–4c. BC. A fundamental distinction is generally drawn between West and East Iranian languages, but it is uncertain whether Old Persian and Avestan should be considered to be the respective predecessors of these two branches. Traditionally, an additional distinction is drawn among the West Iranian languages between Northwest and Southwest Iranian respectively. Recently, however, there has been some debate on the relevance of the Northwest vs. Southwest distinction. Paul (1998a) suggests that it is not a matter of clear-cut genetic grouping, but involves a continuum of overlapping areal and genetic isoglosses (see also Paul (2003b), and Korn (2003) for a different view). In fact, much of the sub-grouping of Iranian languages remains highly controversial, and some of the issues are broached in the relevant sections below. At this stage, a very much simplified overview is provided in Table 1. Only a small selection of the better-known languages and language groups are included in the table – more detailed accounts are available in Sims-Williams (1998) and Schmitt (2000). The Iranian languages 5 Table 1. Overview of the Iranian languages (simplified) Historical stages Major attested languages Old Iranian Old Persian (6–4c.BC) Old Avest., Younger Avest. (14c.–6c.BC?) Western Iranian Mid. Persian, Parthian Eastern Iranian Sogdian, Khotanese Persian Kurdish Balochi etc. Pashto Pamir Group Ossetic etc. Middle Iranian (4/3c.BC–8/9c.AD) Modern Iranian The data for this investigation comes from a varied selection of published materials; details on the sources are provided in the relevant chapters. The transcription systems used in the sources are extremely heterogenous, which has posed a number of practical difficulties of presentation. The default procedure adopted throughout is to follow the source conventions, where this is technically possible and if the transcription is Roman-based. Where the original is in another script (e.g., Arabic or Cyrillic), standard procedures of transliteration have been applied. The drawback of this procedure is that data from one and the same language may appear in different transcriptions, depending on the respective sources. However, the alternative, namely to attempt to create a cross-language standard transcription (or use the IPA) would have raised more problems than it would have solved. And in a book concerned with syntax and morphology, rather than phonology, some degree of inconsistency in the transcriptions appears tolerable. A second problem is the rendering of language names; here I have attempted to impose some standards by adopting wherever possible a single form, generally the orthographically simplest. 6 Introduction 1.3 Alignment in the Iranian context The term alignment is used here in the sense of Nichols (1992), or Harris and Campbell (1995), as a cover term encompassing labels such as ergative, accusative, active etc. Each of the latter is thus considered to be a distinct type of alignment. The choice of ‘alignment’ over the equally widely-used ‘grammatical relations’ is justified because the latter is often used to refer to individual relations (e.g. ‘subject’) rather than the specific bundling of features denoted by ‘alignment’. Furthermore, ‘alignment’ enjoys wider currency than alternatives such as ‘actancy schemata’ (Lazard 1998) or ‘argument linking’ (Stiebels 2002). There is a vast literature available on alignment, most of it skewed towards ergativity (see Dixon (1994) and Manning (1996) for booklength surveys), and it is unnecessary to recapitulate common knowledge here. At this stage I will introduce a minimum of basic concepts and definitions, while reserving more in-depth discussion to later sections. Different alignments can be defined using the following three parameters: Case The case marking of core arguments, restricted here to just subjects and direct objects (see below). Agreement The formal means of cross-referencing core arguments outside of the NPs coding those arguments. Agreement is usually manifested on the verb, but in Iranian, agreement with core arguments may be via clitics on other constituents. Syntactic processes Processes involving syntactic rules which can only be formulated with reference to specific core arguments. Typical examples are Equi-NP deletion, relativization, or control of reflexive pronouns. The core arguments relevant to determining alignment are, in traditional terminology, subject and direct object. More recently, the concept of alignment has been extended to indirect objects (Siewierska 2004: 57, Croft 2001: 142– 147), but this avenue will not be pursued here. A further distinction between transitive and intransitive subjects can also be drawn, yielding three core arguments. Following Dixon (1994), they are abbreviated as follows: S=Subject of intransitive verb A=Subject of transitive verb O=Object of transitive verb S, A and O correspond to S, A and P in Comrie (1978), or X, Y and Z in Alignment in the Iranian context 7 Lazard (1998); my choice of symbol is dictated by mere force of habit.2 The ergative alignment type is defined as one in which the morphosyntactic properties associated with an O are identical to those associated with an S, while those associated with an A are distinct from either. Accusative alignment on the other hand involves identical properties of S and A, while O is distinct. Active alignment, also known as split-S or fluid-S (Dixon 1994), is but marginally relevant in the Iranian context (the exception being East Iranian Wakhi, see Bashir 1986) and will play no significant role in this study. Examples from modern standard Persian illustrating accusative alignment are: (1) man ruznāme-rā mi-xān-am 1 S newspaper-ACC PROG-read:PRES-1 S ‘I(=A) am reading the newspaper(=O)’ (2) man be šahr mi-rav-am 1 S to town PROG-go:PRES-1 S ‘I(=S) am going to town’ Here S and A share the same case form (the morphologically unmarked, or Nominative case), and both determine agreement on the verb. The O on the other hand is marked with an additional Accusative case marker, and plays no role in person agreement with the verb. As preliminary illustration of ergative alignment, consider the following examples from Zazaki (West Iranian, Central Eastern Turkey): (3) ti kām ē? 2 S:DIR who COP:PRES:2 S ‘Who are you?’ (Paul 1998b: 72) (4) wexto ki to āw-i šimit-ā at.time that 2 S:OBL water-F:DIR drink:PST-F:3 S ‘When you drank the water [. . .].’ (Paul 1998b: 91) 2 I remain non-committal on the universal status of S, A, and O (see Mithun and Chafe (1999) for critical discussion), though I am most sympathetic to the view of Du Bois (1985: 357), according to whom they are “simply a convenient but fictive intermediate level of analysis”. 8 Introduction Here it will be seen that the S in (3) and the O in (4) share a common case, the Direct, and both determine agreement on the verb. The A in (4), on the other hand, is in a distinct case, the Oblique, and does not determine agreement on the verb. A distinction is commonly drawn between morphological ergativity, and syntactic ergativity (also called inter-clausal ergativity in Dixon (1994), but this terminology is not used here). Both types of ergativity imply the unity of S and O, as opposed to A, but they refer to different formal properties of the constituents concerned. Morphological ergativity is concerned solely with those properties that are systematically reflected in inflectional morphology, for example case marking, or agreement. Syntactic ergativity on the other hand concerns the properties as they relate to syntactic rules. To say that a language has syntactic ergativity implies that in a particular formally defined environment S and O exhibit identical properties with reference to certain syntactic rules. In general, establishing the presence of syntactic ergativity for any given language is a much more challenging task than identifying morphological ergativity. Indeed, for many languages, controversy on their syntactic alignment continues unabated (see Chapter 1 of Manning (1996) for a balanced discussion). Fortunately, we are spared this particular can of worms because, according to general consensus, no Iranian language exhibits clear evidence of syntactic ergativity.3 However, the lack of syntactic ergativity, not only in the modern languages but in all earlier attested stages of Iranian, is nevertheless of considerable significance when it comes to evaluating theories of the emergence of ergativity in Iranian. Having introduced the basic concepts and definitions related to alignment, we are now in a position to introduce four features that are characteristic of alignment in Iranian: Tense-Sensitive Alignment, lexical transitivity, the polyfunctional Oblique case, and the proliferation of hybrid alignment types. 3 The only exception known to me is Northwest Iranian Vafsi, where one textually very rare version of the ergative construction has the word order OAV, rather than the usual word-order in Vafsi transitive clauses, AOV. However, even the change in word order does not seem to affect underlying syntactic relations, although Don Stilo informs me that he does not have sufficient data to fully analyze the inter-clausal syntax associated with this construction. Alignment in the Iranian context 9 1.3.1 Tense-Sensitive Alignment (TSA) Throughout the entirety of the Iranian language family, ergative, or more generally non-accusative alignments, are almost completely restricted to a single formally defined environment: clauses headed by verb forms built from the past stem of transitive verbs. In all other environments we find accusative alignment. The sole noteworthy exception to this generalization occurs in certain languages where alignment with verbs of sensory perception, desire, and obligation pattern with past transitive verbs; this group is examined in Section 6.6. The situation found in many Iranian languages is often referred to as “split ergativity” (Dixon 1994), but the term is misleading. It is not ergativity that is split, but alignment: Accusative alignment is found in one part of the grammar, non-accusative alignment in another. Furthermore, the alignment associated with past-tense verb forms is in many cases not straightforward ergativity, but some brand of non-accusative alignment (see below). There are thus good reasons to reject the term ‘split-ergativity’ as a general characteristic of Iranian languages. Instead, I will refer to Tense-Sensitive Alignment (TSA) as the defining feature of the majoriy of Iranian languages.4 The restriction to past-tense environments conforms to the well-known universal regarding such tense-based splits: in a tense/aspect-based split, ergative alignment is invariably associated with past or perfective verb forms. There have been attempts to explain this distribution in terms of universal semantic/pragmatic principles, as in for example Dixon (1994: 98–99). However, it needs stressing that in Iranian it is not primarily some semantic notion of ‘pastness’ or ‘perfectivity’ that is crucial to triggering ergativity, but the historical link between certain verb forms, and a particular alignment type. The same general point is made by Anderson (1992: 355), and I will briefly reinforce it here. In the vast majority of modern Iranian languages, each verb has two stems, generally referred to as past and present (or past and non-past) respectively. Each of the two acts as the basis for a variety of different tenses and moods. Although this basic binary distinction is blurred and cross-cut by secondary distinctions in many of the languages, it is nevertheless a remark- 4 This terminology is reminiscent of Klaiman (1987)’s term TACS (Tense-Aspect Conditioned Split), which she applies to Indo-Aryan. 10 Introduction ably stable characteristic, one of the deepest traces of genetic unity across the family. The historical origins of this fundamental opposition in the verb system will be discussed in Chapter three. For the time being, it needs to be emphasized that ergative, or more generally non-accusative, alignments in Iranian languages are always associated with the past stems of transitive verbs. Now past stems are generally also associated with the semantic notions of pastness and perfectivity, while present tenses are generally associated with present and future meanings. But in some languages, the expected correlations do not always hold. Yet crucially, the link between ergative alignment and past stems continues to obtain, even when the semantics do not match up. For example, in the Awroman dialect of Gorani, a west Iranian language spoken in the Iranian province of Kurdistan, all verb forms are based on either the past, or the present stem of the verb. And typically, non-accusative alignments are found exclusively with verb forms based on past stems. Now Awroman also has a secondary development, a tense referred to by MacKenzie (1966: 38) as the Imperfect, which is based on the present stem. Crucially, the Awroman Imperfect, even when it has clear past-tense reference, still has accusative alignment rather than the non-accusative alignments associated with past stems of the verb. Similarly, the Badı̄n. dialect of Northern Kurdish discussed in Chapter five has a past irrealis (‘I would have done X’) based on the suffix da- plus the present stem of the verb. Again, alignment is accusative, despite the past-time reference. A very similar development is noted in Talyshi, where an imperfect tense with past tense reference is found, but based on the present stem of the verb. And here as well, alignment with these verb forms conforms with expected alignments with present-stem verb forms, despite the actual past reference (Don Stilo, p. c.). In sum, although the verb stems I am calling ‘past stems’ do, in general, express past-tense meanings, it is not past-time reference in itself which acts as the trigger for non-accusative alignments. It is ultimately a matter of the origins of particular verb forms, their links to the historical reflex of what was in fact once a participle. Alignment in the Iranian context 11 1.3.2 Lexical transitivity Along with the past stem of a verb, a second triggering feature for nonaccusative alignments is the transitivity of the verb. And here again, it is not transitivity in a semantic sense that is crucial. This fact can be observed most clearly when we examine the alignment associated with various types of complex predicate consisting of a non-verbal element and a transitive light verb. These complex predicates may express states of affairs hardly expected of transitive verbs. Consider the following examples, from three different West Iranian languages: (5) Bihar-ê dest pê kir-i-ye spring-OBL hand to.it do:PST-PTCPL-3 S ‘Spring has begun’ (lit. Spring has put hand to-it) (Northern Group of Kurdish, Haig 2002a: 18) (6) kāgı̄-ā bāl ku crow-OBL flying do:PST:3 S ‘The crow flew’ (Balochi from Karachi, Farrell 2003: 198) (7) tani há=s kærd He:OBL running=CLC:3 S do:PST ‘He ran away’ (Vafsi, Stilo forthc. b) All three clauses refer to states of affairs involving a single participant (‘spring beginning’, ‘flying’, ‘running away’), and we would probably hesitate to refer to them as ‘transitive’. But the case marking in the examples is nonaccusative: the A is in the Oblique case, which it would not be for an intransitive verb in any of the languages concerned. The reason is simply that the lexical verbs on which these clauses are based, ‘do’, ‘make’ and ‘give’, are lexically specified as transitive, and hence in past tense forms take the non-accusative alignment appropriate for the class of transitive verbs. In the same connection Matras (1992/1993: 152) draws attention to certain modals in Northern Kurdish, which do not require a NP as direct object, but nevertheless trigger ergative case marking on their subjects in the past tenses. Thus it is not primarily a semantic, or “whole-clause” notion of transitivity (in the sense of Hopper and Thompson 1980) that is relevant, but the lexically determined class of the verb. 12 Introduction However, the semantics of the entire clause do in fact affect the way alignment patterns are distributed, and there are examples of etymologically transitive verbs shifting class under semantic pressure, as I will show. In Southern Balochi (Farrell 1995: 232), there are also some indications of whole-clause semantics impacting on the strictly lexical specification of transitivity. Likewise, there are interesting interactions between main verbs and auxiliaries, in those languages which have them. At this point, however, I am merely clarifying my terminology by stressing that when I use the term ‘transitive’ I am referring to a particular class of verb lexemes, rather than syntactic or semantic properties of actual clauses. In particular, it must be noted that ‘transitive’ in this sense does not necessarily imply the presence of a syntactically realized direct object (see Haig (2002a) for extensive justification of this standpoint). 1.3.3 The polyfunctional Oblique case Another cross-linguistically unusual feature of Iranian is that the A of a Past Transitive Construction is often marked with a general-purpose Oblique case. The same case marker is also generally used to mark the O in present tenses. These facts are illustrated by past (ergative) and present (accusative) transitive clauses from Southern Balochi: (8) jınık-a b@cık ja girl-OBL boy:DIR hit:PST:3 S ‘The girl hit the boy.’ (Farrell 1995: ex.(7)) (9) b@cık jınık-a j@-ã boy:DIR(PL) girl-OBL hit:PRES:3 PL ‘The boys hit the girl.’ (Farrell 1995: ex.(15)) It is well-known that for languages with an ergative construction, the case of the A is often formally identical with other markers, for example an Instrumental (see Palancar (2002) for a recent survey). Næss (2007: 169–170) notes that such functional polysemy is generally semantically justified: cases that share some values on features such as volitionality or affectedness are more likely to be marked by one and the same form. For example, Recipient markers may be identical to Possessors, Instruments to Ergatives or Locatives to Alignment in the Iranian context 13 Goals. But what is found in Iranian, namely formal identity between an Ergative marker and an Accusative marker is, as Bossong (1985: 118–121) points out, a genuine typological rarity. The reasons behind this state of affairs are best understood against the background of the history of the case system, which is laid out in Chapter four. Note, however, that identity between the Ergative marker and the Accusative marker does not pose a problem for differentiating core arguments in a transitive clause, because the two functions have strictly defined functions according to the verb stem in a given clause: only with past stems is the Oblique interpreted as an Ergative marker (barring a number of complications to be discussed in Chapter four.) For the time being, then, readers should be alerted to the fact that nowhere in Iranian is there a specific ‘Ergative’ case marker. Thus the A of an ergative construction in an Iranian language is generally glossed simply with OBL ( IQUE ), barring certain Tatic-type languages discussed in Section 4.4. 1.3.4 The proliferation of past-tense alignments For Iranian languages, alignment patterns are defined mostly in terms of two parameters: case marking, and verbal agreement. Assuming that there are two structural cases, Direct and Oblique, then we have four different possible constellations of case on A and O. Now the verb may, in theory, agree with either the A or the O. Or it may not agree with either. Thus we have three possible agreement constellations. These can be combined with the above-mentioned four possible constellations of case-marking to yield a total of twelve logically possible alignment types. Ergative and accusative alignments, given in Table 2, represent thus merely two of these twelve possibilities. In a detailed study, Dorleijn (1996) investigates past transitive alignments in the Northern Group of Kurdish (see Section 5.4 for details). Remarkably, Dorleijn (1996: 118) finds all twelve logically possible alignments attested Table 2. Two possible alignment types (two cases, no pronominal clitics) ACCUSATIVE ALIGNMENT: E RGATIVE ALIGNMENT: A=Dir. A=Obl. O=Obl. O=Dir. Verb agrees with A Verb agrees with O 14 Introduction in her data. Although certain combinations are vastly more preferred than others, her findings nicely illustrate a recurrent issue: there is nothing particularly privileged about ergative alignment, a point also brought home by Lazard (1999) and Payne (1980). Nor is it, in its pure form, even very common in Iranian. Rather, what we find time and time again are various types of hybrid alignments: they differ significantly from the accusative alignment regularly found in the present tenses, but are not pure ergative. Matters are actually much more complicated than the Kurdish example suggests. In many other Iranian languages, Past Transitive Constructions crucially involve a set of pronominal clitics as part of their agreement configuration. An examples has already been given in (7) from Vafsi, where a pronominal clitic obligatorily cross-references the A. When the complications of the pronominal clitics are also taken into account, the number of theoretically possible alignment types increases dramatically. In fact, every single language investigated in this study has several alternative alignments of case and agreement in its past transitive clauses. For example, Vafsi has three (Stilo 2004b: 243), as does the Baraki-Barak variety of Ōrmur.ı̄ (Kieffer 2003: 186), and the languages examined in later chapters confirm this fact. The degree of variation found in past transitive constructions contrasts bluntly with the almost total lack of comparable variation in other environments, where we find the monotony of accusative alignment broken only by Differential Object Marking in some languages (see Section 4.3). There has been a quite unwarranted tendency to concentrate research on one single alignment type, the ergative alignment. When other possibilities are mentioned, they tend to be discarded as unstable transitional phases which languages pass through on their implacable progression from ‘ergative to accusative’. But the widespread distribution and frequency of the supposedly intermediate forms suggests that they represent viable systems in their own right, of considerable stability and time-depth.5 In this study, I have avoided coining new taxonomic labels for the various hybrid alignments (cf. for example “anti-absolutive”, “inverse”, “superabsolutive” introduced in Bubenik 1989). Instead, I have found it more insightful to decompose alignment into 5 Note that the facts from Iranian make a mockery of attempts to reduce ergativity to a “Parameter setting”, with the values on/off, as claimed for example in Baker (2001). Constructions and syntax 15 its component sub-systems: case, agreement, and clitic pronouns. Each subsystem can be shown to follow its own trajectory of historical development, which, to a quite remarkable extent, is independent of the others. The advantage of this perspective is that it yields a natural explanation for the attested variation in alignment types: the various alignments emerge as contingent combinations of case, agreement and clitic pronouns; in fact, as epiphenomenal. To briefly recapitulate the main points of this section, it is a basic fact of the syntax of modern Iranian languages that their Past Transitive Constructions (PTCs) display a variety of non-accusative alignments. Outside the PTC, simple clauses are quite uniformly accusative. It is this fact that is central to Iranian, not the presence of ergativity in some of the languages. Ergativity, I contend, is but one of the possible results of partially independent changes in case and agreement patterns, and it is with these that a historical account of alignment changes in Iranian must be primarily concerned. 1.4 Constructions and syntax The data presented in this book are intended to be accessible to linguists, present and future, regardless of their theoretical preferences. For this reason, no special formalism is presupposed beyond what can be reasonably accommodated in the text. Nevertheless, the book is not ‘theory-neutral’: it has become increasingly evident that certain theories are much better equipped to explain the Iranian phenomena than others. The most promising candidate is Construction Grammar, for reasons that I will spell out now. A more detailed theoretical assessment of the results is deferred to the final chapter. The term Construction Grammar now covers a family of theories, but I will concentrate here on the versions in Goldberg (1995) and Goldberg (2006). In Construction Grammar, syntactic constructions, to the extent that they carry non-predictable semantics, are taken as grammatical primitives. This can be contrasted with the orthodox view underlying what Culicover and Jackendoff (2005) refer to as Mainstream Generative Grammar. Here, there has traditionally been a strict division drawn between a lexicon, and an autonomous rule component, working on variables fed into it from the lexicon. On this account, then, the syntactic constructions that make up actual speech 16 Introduction are merely the epiphenomenal results of the application of rules. A further claim central to the Mainstream Generative Grammar, and indeed many other theories such as LFG, is the conviction that the structure of a clause is determined by the lexically-specified argument structure of its predicate (cf. the Projection Principle, Chomsky 1981: 38). The widespread acceptance of the Projection Principle has resulted in a good deal of effort going into developing a series of “universal, or near-universal” (Goldberg 1995: 8) linking algorithms mediating between the lexically-determined argument structure of the verb, and the syntactic structure it projects to. This approach has much to commend it, but it also suffers from significant drawbacks. One major drawback is that if different syntactic structures are associated with one and the same verb, as they very often are, then corresponding differences must be sought in the lexical semantics of the verb concerned. Goldberg (1995: 11) gives the following eight syntactic structures associated with the verb kick: (10) a. Pat kicked the wall b. Pat kicked Bob black and blue c. Pat kicked the football into the stadium d. Pat kicked at the football e. Pat kicked his foot agaist the chair f. Pat kicked Bob the football g. The horse kicks h. Pat kicked his way out of the operating room Now a lexical approach to argument structure is obliged to account for all these distinct clause types by stipulating distinct argument structures, in effect, distinct senses of the verb kick. But apart from the proliferation of verb senses that would result, this approach seems to miss an important intuition, namely that there is a basic semantic unity in each sense of kick across all the above examples. An alternative way of accounting for this – actually very commonplace – situation is to admit the possibility that the different constructions in which kick occurs have an independent existence as grammatical units in their own right. Thus the clause structures found are not merely derivatives of lexically projected argument structures, but basic and independent entities Constructions and syntax 17 of language. And like other units of languages – morphemes, words – constructions have meaning, meaning that is to some extent independent of the actual items that instantiate a given construction. The different constructions associated with kick are thus explained in terms of a fusion (see Goldberg (1995: 50–52) for formalization of this notion) of the verb meaning with the meaning implied by a particular construction. The meaning and structure of a particular clause is thus not determined solely in a bottom-up manner, i.e., as projection from the lexicon into the syntax, but by a combination of lexical and constructional (top-down) processing.6 Relaxing the Projection Principle thus allows syntax to be co-determined by the lexical frame of the verb, and the semantics of the construction. A logical consequence of this move is that “a construction can add roles not contributed by the verb.” (Goldberg 1995: 54) The arguments that Goldberg rallies in support of this claim stem largely from English ‘Caused Motion Constructions’, such as the following (Goldberg 1995: 152): (11) They laughed the poor guy out of the room. It is difficult to maintain that a verb like laugh is lexically specified for an object, or for a component of ‘caused motion’. It is, nevertheless, like a large number of other verbs, compatible with the construction illustrated in (11). To account for these phenomena, Goldberg posits a construction in the grammar of English, a particular pairing of form and meaning, with the argument frame [Subj V Obj PP] (see Goldberg (1995: 160) for a fuller representation of the syntax and semantics.) Notice that this construction is posited for the grammar of English; there is no implication that it is part of the constructional inventory of other languages, and indeed, good evidence that it is not. Because constructions are basic units of language, we expect to find crosslinguistic variation, just as we do in the lexicon generally: just as a language 6 Of course even Construction Grammar must account for the very simple fact that not all verbs are compatible with all constructions, and it is fairly obvious that the lack of free combinability must be due to lexically specified constraints. In fact, lexically-specified argument structure creeps in the back door through the notion of “lexical profiling of participants” (Goldberg 1995: 52). But there is nevertheless a fundamental difference between Goldberg’s notion of fusion of lexical meaning with constructional meaning, and projection of argument structure from the lexicon to the syntax. 18 Introduction may or may not have a lexical ‘have’ verb, a language may or may not have a particular construction. Another example of arguments contributed by a particular construction comes from German. Consider the following: (12) a. Das Wasser ist kalt the water is cold ‘The water is cold.’ b. Das Wasser ist zu kalt the water is too cold ‘The water is too cold.’ c. Mir ist das Wasser zu kalt to.me is the water too cold ‘The water is too cold for me/I find the water too cold.’ d. *Mir ist das Wasser kalt to.me is the water cold Intended: ‘The water is cold for me/I find the water cold.’ The Dative-pronoun mir is fine in (12c), but is not possible in (12d). How are we to account for this? Surely we would not wish to claim that the judgementparticle zu has somehow contributed an extra argument? The more reasonable explanation is that the constructional semantics of (12c) arise from the combination of zu with an adjective. This construction implies the existence of an Experiencer, which is coded via a fronted Dative, as in a whole family of related constructions with similar semantics in German. Further evidence that supports the notion of argument structures independent of lexical verbal semantics comes from English, where almost any noun can be verbed. From this simple fact the following question arises: what is the source of knowledge which allows speakers to deploy a noun for the first time as a verb, governing arguments of its own? A syntax-from-the-lexicon approach obliges us to assume that each English noun (or adjective) is somehow fitted out with a ‘potential argument structure’, to be implemented when the noun is deployed as a verb. Construction Grammar would, simplifying somewhat, claim that the knowledge a speaker commands of the noun’s semantics, and of the meaning in the constructions available in her lexicon, are sufficient to allow her to decide on how to combine the new verb with an appropriate construction. Constructions and syntax 19 In the context of this study, the notion of non-lexically specified arguments is most relevant in connection with non-canonical subjects. They are provisionally defined here as clause constituents which differ in their morphology from the subjects of simple intransitive clauses, yet display at least some of the syntactic features generally associated with these (and other) subject constituents in the language concerned. Non-canonical subjects have been discussed under a variety of different labels in the literature (oblique subject, non-nominative subject, quirky subject, dative subject etc.) Several languages are said to have some type of non-canonical subject (see the papers in Aikhenvald et al. (2001) and Bhaskararao and Subbarao 2004). A fairly uncontroversial example of non-canonical subjects is provided by Icelandic. The following examples contrast a canonical (a) and a non-canonical (b) subject in Icelandic (from (SigurDsson 2002: 692, 711), glosses slightly modified): (13) a. Hún var fáklædd she:NOM was scantily.dressed ‘She was scantily dressed’ b. Henni var kalt she:DAT was cold ‘She was cold.’ Despite the fact that the first NP in (13b) is not in the Nominative, NPs in this configuration still exhibit largely the same set of syntactic properties that canonical Nominative subjects in Icelandic do (SigurDsson 2002; 2004: 693). Icelandic thus illustrates the phenomenon of NPs which lack the morphological properties of full subjects (in particular, Nominative case), yet share most of their syntactic properties. In other well-studied languages like Japanese or Russian, the analysis, and indeed the existence of non-canonical subjects remains highly contentious – cf. for example the ongoing debate on Russian (Moore and Perlmutter 2000; SigurDsson 2002), or Japanese (Shibatani 2001; Kishimoto 2004). These controversies are symptomatic for the problems posed by non-canonical subjects for mainstream generative syntax: on the one hand they undoubtedly exhibit argument-like properties, on the other it is difficult to demonstrate that they are lexically licensed by the predicate, as they are often optional. 20 Introduction The most common solution is simply to list the relevant verbs in the lexicon as licensing exceptional, non-structural case marking of their arguments. Linking rules then account for which (if any) of the arguments is assigned to the subject position. Lightfoot (1999: 130) advocates this approach for Old English verbs such as lician ‘like’. But the lexicon-approach is much less appealing for a living language like Icelandic which, according to the figures mentioned in SigurDsson (2004), has well over 500 verbs that trigger such constructions. Furthermore, there are sometimes different patterns associated with one and the same verb. The lexical-listing approach also misses a fairly obvious commonality across non-canonical subject constructions wherever they are found: they express a very typical cluster of meanings, centering on mental and physical perception, desire, need, and possession. From a Construction Grammar perspective, non-canonical subjects have a natural explanation: they are arguments contributed by the construction, not the verb. More importantly, the Construction Grammar approach allows for an elegant and intuitively simple explanation of the semantic commonalities common to non-canonical subjects: they are part of the constructional semantics. For Construction Grammar, the raw material for the analysis is the construction itself, both its form and its meaning. Certain aspects of the construction’s meanings can be attributed to the construction without being derived secondarily from the verb’s thematic grid. And certain aspects of the formal properties of the non-canonical subjects (their subject-like properties) is also assignable to the construction itself, rather than via special linking rules. Croft (2001) has recently extended this line of thought. He argues that subjects are always construction-specific, rather than language specific. Each construction defines a specific set of properties that its particular ‘subject’ will possess. The relevance of this approach is that we are obliged to recognize distinct grades of subjecthood, rather than a single monolithic “grammatical relation”. A construction-specific approach to subjecthood will turn out to have considerable implications in explaining the diachronic developments in Iranian. The final point to be emphasized in this section is that Construction Grammar rejects a derivational approach to syntax. As mentioned, constructions, to the extent that they exhibit non-predictable aspects of form and meaning, are taken as primitives in their own right. There is no attempt to derive one from another, nor to set up more abstract representations based on semantic Constructions and syntax 21 equivalence. But constructions are not simply isolated elements of an unstructured inventory. Instead, they form a structured network, linked to those other surface constructions with which they share aspects of form and meaning. There is an extremely important principle at work here, which Goldberg (2006: 25) formulates as follows: Surface Generalization Hypothesis: There are typically broader syntactic and semantic generalizations associated with a surface argument structure form than exist between the same surface form and a distinct form that it is hypothesized to be syntactically or semantically derived from. In essence, Goldberg argues here for the primacy of surface form in explaining the properties of a given syntactic structure. In the context of different alignments, this principle has far-reaching implications. Effectively, each distinct alignment, for example an accusative alignment associated with a present-tense verb, and the ergative alignment associated with the same verb’s past tense, are distinct constructions. There is no requirement to derive one from the other. Consequently, there is no necessity to postulate additional mechanisms to account for the non-identical case and agreement constellations across the two constructions (no Exceptional Case Marking, no additional Infl-nodes etc.) Derivational approaches are faced with intractable problems when confronted with the numerous hybrid types of alignment introduced in the preceding sections, and taken up below. I would, incidentally, apply a similar non-derivational approach to the analysis of the passive in many languages, including English, but this is not the place to expand on that possibility. A Construction Grammar approach on the other hand, is well placed to cope with these phenomena. If we assume that Past Transitive Constructions are constructions in their own right, then it is perfectly natural that they should have developed in highly divergent ways, while the corresponding present tense constructions remained unchanged over millennia. Croft (2001: 168) explicitly endorses the view that distinct alignments are distinct constructions in languages with alignment splits, and this is also the approach adopted here. The Past Transitive Construction is of course closely related to the present transitive construction through the common semantics, and shared aspects of syntax, but it is a relationship of loose interdependence, comparable to that existing in a lexical network, rather than a derivational relationship. Fur- 22 Introduction thermore, the construction approach paves the way for exploring the links between Past Transitive Construction, and other types of construction in the language, in particular non-canonical subject constructions. Finally, like other items in the lexicon, constructions display polysemy. Goldberg (1995: 31) suggests that constructions are “typically associated with a family of closely related senses rather than a fixed, abstract sense.” She nevertheless proposes that a construction will have a central sense, but allows for radial extensions to related ones. The notion of constructional polysemy has considerable explanatory power in the area of syntactic change. It has long been recognized that one of the preconditions for semantic change of individual words is their polysemy (see for example Evans and Wilkins (2000). Constructional polysemy, if accepted, provides a simple explanation for changes in the way a particular construction is used. The central meaning of a construction shifts to one of its radial extensions, in a manner paralleling semantic change in words. This is then a further important consequence of adopting a constructional view of syntax: the differences between syntactic change and lexical change, a difference still axiomatic for many generative accounts of syntactic change (Kroch 2001: 699), are radically reduced. In sum, a Construction Grammar approach offers a number of significant advantages over alternative models. Of course this is not to suggest it will provide answers to all the questions raised here, but taken on balance, it appears much better equipped for explaining what is, as we shall see, best conceived of as the history of a particular construction in a group of related languages. Chapter 2 Alignment in Old Iranian The stage of Iranian referred to as Old Iranian is attested in two major bodies of texts, the oldest surviving records of any Iranian language: Avestan and Old Persian.1 The Avestan texts are religious in nature, and are traditionally divided into two groups: Old Avesta (sometimes also called Gatha Avesta), and Young Avesta. The texts were first written down around 600 AD, but prior to this they were transmitted orally “by specially trained priests” (Skjærvø 2003: xiii) over several centuries. Their precise ancestry and place of origin remain controversial. However, on both archeological and comparative linguistic evidence – the Old Avestan texts are linguistically extremely close to the language of the oldest parts of the R.gveda – the earliest Avestan poems can probably be dated to around 1500 BC (Skjærvø 2003). Geographically, it is assumed that Avestan reflects an Iranian language spoken in what is now northeastern Iran. Young Avestan refers to a body of texts differing linguistically quite sharply from Old Avestan. The language is quite close to Old Persian (see below) and is dated at a similar period. Young Avestan does not comprise a homogenous body of texts; the consistency with which the texts may be considered to faithfully reflect the language of the time varies according to the date at which the texts were committed to writing, and the respective knowledge of the scribes. Given the difficulties in dating, in distinguishing corrupt scribal practices from genuine linguistic features, along with the highly arcane religious nature of the content, the interpretation of Avestan texts is an incredibly complex undertaking, mastered by only a handful of specialists world-wide. For all these reasons I have made but sparse reference to Avestan data, concentrating instead on the other well-attested variety of Old Iranian, Old Persian. Unlike Avestan, the extant records of Old Persian are firmly rooted in time and place. The texts are written in a cuneiform script inscribed in stone 1 At least two other ancient Iranian languages are known to have existed in the first millennium BC, Median and Scythian. However, as nothing is known of their syntax, they will be ignored henceforth. 24 Alignment in Old Iranian and dated 5–4c. BC. The language itself is considered to have been spoken in what is now southwestern Iran and was presumably the vernacular of the rulers of the Achaemenian dynasty, whose deeds are recounted in the inscriptions. According to Schmitt (2000: 30–31), however, Old Persian played no role in the administration of the Empire, where Aramaic continued to be dominant. The longest texts, the Behistān inscriptions, are generally accompanied by translational equivalents in Elamite and Accadian, in some cases also by a version in Egyptian hieroglyphics. Although the syntactic interpretation of Old Persian is, on the whole, more straightforward than that of Avestan, three main difficulties remain: First, the corpus is quite restricted in size. Second, the cuneiform script coupled with damage to parts of the inscriptions renders some passages difficult to interpret. Finally, the texts are written in a stylised and formulaic register. As Schmitt (2000: 30) notes, the inscriptions were not intended primarily to be read – some were inscribed on inaccessible cliff faces, or built into the foundations of buildings. The main purpose of the inscriptions appears to have been representational rather than communicative. This raises doubts with regard to the extent to which the syntax of the inscriptions reflect the syntax of ‘normal’ spoken language. Clearly then, when interpreting the syntax of Old Persian, a good deal of caution is required. Despite these difficulties, most previous research on the evolution of alignment in Iranian has taken as its starting point certain features of Old Persian, and I will be continuing that tradition here. The examples cited here are based on the version of the text corpus provided in Kent (1953). Although Kent’s readings have in some cases since been superseded (cf. Brandenstein and Mayrhofer (1964), Schmitt (1990) and Schmitt (1999) for more recent interpretation), for the sake of consistency, all examples follow Kent’s transcription and system of cross-referencing, unless indicated otherwise. Thus I continue to write kartam instead of the now more usual kr.tam. Where more recent scholarship has shown Kent’s readings to be mistaken, supplementary notes and references have been added. The examples have been supplied with a highly simplified morphological glossing: only those inflectional categories considered relevant for the syntactic analysis of each example have been included. Furthermore, both active Aorist and Imperfect are given a unified gloss as PAST, because the difference in meaning is slight and does not appear to be relevant here (see Kent (1953: 90–91) on the use of the two tenses). Alignment in Old Iranian 25 Alignment in Old Persian was accusative throughout all tenses, and this can be assumed to be representative of Old Iranian generally. S and A took a uniform case, the Nominative, and the verb agreed with them, while O was marked with a special case, the Accusative. The following examples of a transitive clause (14) and an intransitive clause (15) demonstrate the formal identity of S and A, both in case marking and agreement on the verb, and the accusative marking of O: (14) pasāva adam(A) kāram(O) frāišayam Bābirum thereupon 1 S:NOM army:ACC send:PST:1 S to.Babylon ‘Thereupon I(A) sent an army(O) to Babylon’ (Kent 1953: DB III,84) (15) adam(S) xšāyaTiya abavam 1 S:NOM king become:PST:1 S ‘I(S) became King’ (Kent 1953: XPf,36–37) Likewise in Old Avestan we find in all tenses accusative alignment, shown in the case marking and verbal agreement of the following past-tense transitive clause: (16) Twā fšuyantaē=cā vāstrāi=cā at zı̄ ¯ and indeed 2 S:ACC(O) cattle.breeder:DAT=and herdsman:DAT=and Twōr@šta tatašā fashioner:NOM(A) has.created(3 S) ‘And indeed the Fashioner(A) has created you(O) for the benefit of the cattle-breeder and the herdsman.’ (Old Avestan, Yasna 29,6) However, in many modern Iranian languages, alignment in past tenses is no longer accusative. Past transitive clauses with the accusative alignment found in (14), or (16), were, by Middle Persian, simply no longer part of the grammar, and indeed in many modern languages they remain a syntactic impossibility. In this chapter I will be developing proposals to explain the loss of this type of construction, and the complementary spread of non-accusative alignments in past transitive clauses. 26 Alignment in Old Iranian 2.1 The manā kartam construction It is generally assumed that the ergative alignment of past tenses in modern Iranian did not develop from a finite transitive construction such as that illustrated in (14). Rather, the source is a construction headed by a resultative participle, rather than a finite verb form. I refer to this construction as the manā kartam construction, abbreviated to m. k. construction. An example is the following: (17) ima tya manā kartam pasāva yaTā xšāyaTiya that which 1 S:GEN do:PTCPL after when king abavam become:PST:1 S ‘This (is) that (which) was done by me after (I) became king’ (Kent 1953: DB I,28–29) This particular phrase, with minor variations, is repeated in the texts at least twenty times, suggesting a strongly formulaic character.2 The m. k. construction consists of: 1. a NP in the nominative case (here the ‘relative article’ tya, Nominative Neuter Singular) 2. a NP in the Genitive case expressing an Agent (here manā ‘first person singular Genitive’) 3. a resultative participle in −ta, here kartam, from kar- ‘do, make’, carrying Nominative Singular Neuter ending in agreement with tya. Optionally, the participle may be extended with a form of the copula verb, for example astiy in (18). However, it is unclear what factors influence the presence or absence of the copula. A second possibility for expressing the Agent-phrase was through a clitic form of the Genitive pronoun, as in (18) and (19): 2 Schmitt (1999: 103) suggests that all m. k. constructions occur in relative clauses. However, examples such as (19) are not relative clauses. The manā kartam construction 27 (18) utā=maiy aniyasçiy vasiy astiy kartam and=1 S:GEN much else COP:PRES:3 S do:PTCPL ‘and much else was done by me’ (Kent 1953: DB IV,46) (19) avaTā=šām hamaranam kartam thus=3 PL:GEN battle do:PTCPL ‘thus by them battle was done’ (Kent 1953: DB III,18–19), cf. also DB III,40, 47–48,63–64,68–69;DB II,27,42,47,56,98 The importance of clitic expression of the Agent-phrase cannot be overemphasized: it has proved to be one of the most resilient features of Iranian syntax and it remains a characteristic feature of the majority of modern Iranian languages to this day. In Section 2.4.3 below the features of Old Persian cliticization are taken up once more, and the histories of clitic Agent-marking will be picked up at various points in the book (see Appendix, A.2 for an overview). For the time being, we should simply note that expressing the Agent through a clitic pronoun was already in Old Persian at least as widespread as full NP or free pronoun Agents. Although the basic outline of the m. k. construction is well-established, its precise analysis and status within Old Persian syntax remains very controversial. In the next sections I will review some of the more important literature on the topic before going on to present my own analysis. 2.1.1 The possessive interpretation Traditionally, the m. k. construction had been interpreted as a passive, but this view was challenged in a short paper by Benveniste, first published in 1952 (Benveniste 1952/1966). Benveniste’s article is perhaps the most widely cited contribution to the debate on the m. k. construction, but in evaluating his proposals it should be noted that the bulk of the article is concerned with broader parallels between possession and perfectivity from a cross-language perspective. His claims on Iranian take up just four pages, and are of a brief and programmatic character. Benveniste claims two things: (i) the m. k. construction displays clear parallels with a type of possessive construction, also involving a fronted Genitive. The parallels extend to the use of a clitic pronoun to express the possessor. Compare the use of the clitic pronoun =taiy ‘second 28 Alignment in Old Iranian person singular’ to express a possessor in (20), and the clitic pronoun =šām ‘third person plural’ expressing Agent in (19), a phrase attested many times in the corpus and repeated here for convenience: (20) Possessor as clitic utā=taiy tauhmā vasiy biyā and.also=2 S:GEN seed much may.be ‘and may you have much seed (offspring)’ (lit: ‘and may your seed be much/to you may be much seed’) (Kent 1953: DB IV,75) (19) Agent-phrase as clitic avaTā=šām hamaranam kartam thus=3 PL:GEN battle do:PTCPL ‘thus by them battle was done’ On the basis of these parallels, Benveniste suggests that the m. k. construction is essentially possessive in nature. He draws further parallels to comparable structures in other Indo-European languages (e.g., the Latin mihi est type), concluding that the m. k. construction is “un parfait actif d’expression possessive” (Benveniste (1952/1966: 180), original emphasis). Note that in Benveniste’s formulation the term ‘active’ already occurs, and this leads us to Benveniste’s second claim: (ii) the m. k. construction is not a passive. The argumentation in support of this claim runs as follows: according to Benveniste, a construction can only be considered ‘passive’ if the verb form is clearly marked morphologically as a passive. In fact, Old Persian did have such verb forms: (21) upariy avām Tikām hadiš frāsahya on that rubble palace construct:PASS:PST ‘On that rubble the palace was constructed.’ (Kent 1953: DSf,27) In order to avoid confusion, I will refer to passive verb forms of this type, i.e., finite verb forms with overt passive morphology distinguishing them from the active forms of the same verb lexeme, as synthetic passives. These are to be distinguished from participles, which are not (originally at least) finite verb forms. Now when Old Persian synthetic passives occur with an overt Agentphrase, they are marked with the preposition hacā. According to Benveniste, The manā kartam construction 29 this makes the Agent-phrase with hacā a defining feature of the passive construction in Old Persian. Therefore, because the apparent Agent-phrase in the m. k. construction is not of this type, but a Genitive, as in (17), the m. k. construction cannot be considered a passive. There are several immediate difficulties with Benveniste’s claims. For a start, it is unclear in what sense the m. k. construction is ‘possessive’. Semantically, it does not express ‘possession’, at least not in the narrower sense of the word, but quite simply ‘completion of an activity’. Now the links between perfective aspect and possession are pervasive, and have been discussed in connection with the rise of have-perfects many times. Likewise, the link between expressions of possession and expressions of agentivity are widely attested – cf. for example Allen (1964), and further discussion below. Benveniste went on to note historical parallels in the expression of possession and the development of the perfect tenses in many languages, and these certainly appear well-founded. However, his blunt statement that the m. k. construction ‘is’ possessive was bound to evoke negative reactions, in particular as he makes no effort to define just what he means by a possessive construction. The point that Benveniste was really making is not so much that the construction ‘is’ possessive; rather, the point is that the use of the so-called ‘Genitive’ (a misnomer, as I will argue below) in the m. k. construction parallels its use in possessive constructions. And in this respect, I believe he is fully correct; these issues will be dealt with in some detail below. However, the bulk of the criticism against Benveniste’s proposals has been less concerned with the issue of possession, but has targetted his claims regarding the non-passive nature of the m. k. construction. 2.1.2 The passive interpretation A number of authors (cf. Bynon 1979, Bynon 1980, Payne 1980, Cardona 1970, Statha-Halikas 1979, Skjærvø 1985, Payne 1998, Bubenik 1989, Harris and Campbell 1995: 243–244) have rejected Benveniste’s claims, arguing for a return to the traditional passive interpretation.3 Before addressing their 3 Although Benveniste’s proposals have largely gone out of favour in diachronic syntax, 30 Alignment in Old Iranian case, it is necessary to briefly mention some differences in terminological convention. Several scholars from within Iranian studies also use the term “passive” when referring to the m. k. construction, and indeed to its later progeny, the ergative and other non-accusative constructions in Middle Iranian (e.g., Heston 1976) and modern Iranian languages (e.g. Yarshater (1970) on Tati dialects). But often the use of the term “passive” is mere terminological convention, implying no commitment to a particular syntactic analysis (in fact, few of these authors are particularly concerned with syntactic analysis at all). The comments in this section, then, are restricted to those authors who not only use the terminology “passive” for the m. k. construction, but explicitly endorse its syntactic consequences. The earliest refutation of Benveniste’s interpretation is Cardona (1970), which I will be discussing in connection with the typology of Agent-markers in Section 2.5.5. But perhaps the most telling evidence in favour of a passive interpretation of the m. k. construction comes from the fact that participial constructions occur in Old Persian without any form of overt agent at all. This point is made most clearly by Skjærvø (1985), and independently by Statha-Halikas (1979). Skjærvø (1985) draws attention to examples of participial usage such as the following: (22) xšaçam tya hacā amāxam taumāyā parābartam kingdom which from our family taken.away:PTCPL āha be:PST:3 S ‘the kingdom which was taken away from our family’ (Kent 1953: DB I,61–62) a minority of linguists accept his position (see especially Anderson (1977) and Trask (1979) for a positive assessment). Significantly, Theodora Bynon, one of the most influential proponents of the passive-origins view, has now completely abandoned her earlier standpoint. In Bynon (2005) she advocates an interpretation much closer to that of Benveniste and my own, although she does not explicitly renounce her earlier claims. Bynon’s more recent work came to my attention too late to receive detailed treatment, but it is briefly discussed in fn. 29 below. The manā kartam construction 31 (23) vasiy aniyašciy naibam kartam anā Pārsā much other good do:PTCPL in Persepolis ‘much other good (construction) was built in Persepolis’ (Kent 1953: XPa,13–14) (24) . . . tya bardiya avajata . . . that Smerdis slay:PTCPL ‘. . . that Smerdis had been slain’ (Kent 1953: DB I,32) In these constructions, the same participles (in −ta) occur as in the m. k. construction. Likewise, they occur with or without copula support (in (22) with copula āha ‘was’, in (23) and (24) without the copula). Constructions of this type translate quite naturally into an English agentless passive, which is also based on a past participle. On the assumption that they are indeed analytic passive constructions, then the most obvious interpretation of the m. k. constructions such as (17) is that they are merely the agented versions of such passives. Proponents of the passive analysis thus argue that the basic construction is the agentless passive, such as (23). Constructions of this type are attested with a variety of different verbs in Old Persian, whereas the m. k. construction is attested with only a single verb, kar- ‘make, do’ (Skjærvø 1985: 219).4 The agentless passive with the participle is also solidly attested in Vedic and Avestan, as well as in other branches of Indo-European (Statha-Halikas 1979: 355–356). Viewed from this perspective, the m. k. construction is simply a straightforward participle-based passive, typical of many Indo-European languages, which has been extended by the addition of a (facultative) Agentphrase. Skjærvø (1985: 218) goes as far as to suggest that the m. k. construction shows all the features generally associated with an Agented passive construction: 1) the predicate is a form of the verb with passive meaning, 2) the direct object of the corresponding active structure becomes grammatical subject of the 4 The example with the reconstructed participle xšnūtam ‘heard’, quoted by Benveniste (1952/1966: 178) and repeated by Statha-Halikas (1979: 350), is, according to Skjærvø (1985), not reliable. However, there is another candidate for an m. k. construction with a different verb, which Skjærvø (1985) does not mention: DSf,19–20, with the participle framātam from ‘command’. 32 Alignment in Old Iranian passive one, 3) the grammatical subject of the corresponding active structure if present in the passive one, becomes agent and is usually marked by a special morpheme [. . .]. Not all scholars have embraced the passive interpretation, and alternatives are discussed in Section 2.6, but it has nevertheless remained highly influential. In the following section, I take up some of the implications this analysis entails for theories of the emergence of ergativity in Iranian. 2.2 The implications of the passive-theory for diachronic syntax The assumption that the m. k. construction is an agented passive entails certain assumptions concerning its syntax. In particular, it implies that the Agentphrase is an optional, syntactically peripheral constituent, with little or no access to major syntactic rules (a more precise definition of passive is presented in the following section). The subject, both morphologically and syntactically, is the Patient. In the ergative construction, on the other hand, the status of the A and the O is very different: the A is quite evidently the syntactic subject, as shown by Haig (1998) for Northern Kurdish, and indeed this seems to be the case throughout Iranian (cf. discussion in Section 1.3). Now if we assume that these ergative constructions emerged from a passive construction, then we must account for (among other things) how the erstwhile syntactically peripheral Agent-phrase of the passive came to be the syntactic subject of the ergative construction. This is not a trivial matter. Nevertheless, a number of linguists are prepared to assume that such a change in the construction-internal syntax must have occurred. The theoretical prereqisites for these theories can be traced back to the seminal insights of Keenan (1976). Keenan recognized that cross-linguistically, the concept of subjecthood can be most meaningfully investigated in terms of a number of logically independent properties. Based on the investigation of a large number of languages, Keenan identified some 30 properties relevant to defining “subjecthood”, generally referred to as subject properties. He distinguished two types, coding and behavioural properties, a distinction that approximates the one used to distinguish the properties relevant for morphological and syntactic ergativity (see Section 1.3). Furthermore, Keenan made the crucial obser- Implications for diachronic syntax 33 vation that different subject properties can apply to different extents, or that they may be distributed across more than one NP in the clause. This notion of graded subjecthood has been pivotal in much subsequent work on diachronic syntax. For if subjecthood is a multi-factorial rather than absolute property, then diachronically, we can expect gradual shifts in the degree of subjecthood exhibited by a particular constituent. The explanatory potential these insights carry for diachrony were outlined with particular reference to ergativity by Comrie (1978), and by Cole et al. (1980) from a broader perspective. Cole et al. developed a theory according to which subject properties could gradually transfer from a subject to a non-subject NP, such that the latter would eventually become the full subject of the construction. Much of the empirical support for this notion, which I will refer to as transfer of subject properties, stems from the Germanic languages, which have since become a textbook example.5 Cole et al. also formulated principles of syntactic change which they considered to be universal. The most important of these claims are that (a) the acquisition of subjecthood always begins with an NP with no subject properties (Cole et al. 1980: 742); and (b) syntax before morphology, that is, the transfer of subject properties begins with behavioural properties, and is then followed by coding properties. In other words, no case is attested of a language where a non-subject acquires the coding properties associated with subjects of that language before it has acquired the behavioural properties associated with subjects. Thus the diachronic acquisition of subjecthood encompasses a fixed developmental sequence. The most detailed application of this type of explanation in alignment change is from Estival and Myhill (1988), who developed a general theory for the emergence of ergativity.6 Basically, they claimed that ergative constructions arose from passives through the mechanism of transfer of subject properties. A broad synopsis of this view is given below (here I use the terms ‘syntactic subject properties’ for ‘behavioural properties’, while ‘morpholog5 6 More recently, however, Eythórsson and BardDal (2003) and BardDal and Eythórsson (2003) have rejected Cole et al.’s interpretation of the Germanic data. There are striking parallels to the Iranian situation, which unfortunately cannot be taken up here. Estival and Myhill (1988) claim that all ergative constructions evolved from passives, a claim that has since been refuted (Dixon 1994: 189). 34 Alignment in Old Iranian ical subject properties’ corresponds to ‘coding properties’). The division of the process into single stages is largely for the sake of clarity; it should not be understood as implying the existence of a specific number of discrete steps in the process: Stage 1 A deverbal adjective with resultative/passive sense acquires the ability to take an Agent-phrase: the window was broken ⇒ the window was broken by me Stage 2 The Agent-phrase, presumably via some form of topicalization, begins to acquire syntactic (i.e., behavioural) subject properties. Likewise, the deverbal adjective acquires some of the properties of a finite verb form (integration into tense/aspect oppositions, expression of mood etc., no longer reliant on copula support etc.): by me – the window (was) broken Stage 3 The Agent-phrase acquires more syntactic subject properties, concommitantly the O loses them (control of reflexives, coreferential deletion, target of imperatives etc.) Stage 4 by-me the window broken is now the formally unmarked and pragmatically neutral way of saying ‘I broke the window’. The A NP has full syntactic subject properties, and may acquire some morphological ones (e. g case and agreement). Stage 5 (possible further stage) the O NP acquires overt Object-properties, for example Accusative case marking. At this stage, then, we have a construction of the Nominative-Accusative type. Estival and Myhill (1988: 478) cite the Iranian case as “the only historically documented example of the full development [. . .] from passive to ergative and then accusative”. For the authors, it is thus beyond doubt that the m. k. construction is an agented passive, hence at Stage 1. Stage 4 would be represented by, for example, the ergative construction in Kurmanji (Section 5.3). The possible development at Stage 5, the acquisition of morphological object properties by the O, is represented by Persian (cf. ex. (1) in Chapter 1). The scheme outlined above is conceptually elegant, and is congruent with a view of syntactic change as a gradual process involving small but incremental steps. Although not all scholars adopt the ‘transfer-of-subject-properties’ terminology, essentially the same processes are assumed by all scholars who accept the origin of ergativity to be an agented passive. For example Harris Implications for diachronic syntax 35 and Campbell (1995: 243–245) refer to the changes in terms of a reanalysis, (in keeping with the major thesis of their book, according to which reanalysis is the primary mechanism of syntactic change). Thus on their account, the passive m. k. construction is ‘reanalyzed’ as an ergative construction. As they are the most prominent proponents of the passive-to-ergative view in diachronic syntax, it is worth quoting their account of the Iranian data in some detail (Harris and Campbell 1995: 243–245): The perfect passive participle construction [=the m. k. construction, G.H.] was later reanalyzed, and this brought about the following changes in the grammar: (i) a new rule of case marking was introduced for the past-tense system only (case alignment has been reanalyzed as ergative) [. . .] (ii) A new rule of agreement is introduced for the past-tense system, for while verbs in the present tense continue to agree with their subjects, those in the past now agree with their absolute arguments. This account actually does not explain anything at all. The supposedly “new” rules of case assignment (Agent takes an Oblique case) and for agreement (predicate agrees with the Patient) simply describe the case marking and agreement in the m. k. construction itself, which were there from the outset, hence prior to the supposed reanalysis. The questions of how this construction ultimately became an ergative construction, and how it came to be dominant throughout the past tenses, are left unanswered. The brief account of Payne (1998: 556), who likewise assumes a passiveto-ergative shift, is similarly non-committal on the actual pathways of development: The reinterpretation of these constructions as active rather than passive, and the collapse of the genitive and dative into a single oblique case, gave rise to ergative constructions in which the original passive agent was reinterpreted as an A in the oblique case, while the original passive subject was reinterpreted as an O in the absolute case. The verb agreement was then also oriented towards the O . Although Payne here refers to a “reinterpretation”, his account is otherwise indistinguishable from that of Harris and Campbell (1995) quoted above. Both accounts merely list the morphological features of the m. k. construction, before invoking an unspecified process (reanalysis, or reinterpretation) 36 Alignment in Old Iranian which somehow effects the necessary changes towards the ergative construction. No attempt is made to spell out the sequence of sub-changes, nor is any evidence provided for the intermediate changes that presumably accompanied such construction-internal shifts. Rather, we are simply provided with examples from the two putative endpoints of the development (the Old Persian m. k. construction, and an ergative construction from a modern Iranian language), while actual evidence of the restructuring itself is not provided. The question of how intermediate stages might have looked, and why none have been attested to date, is a serious drawback with theories that assume a passive origin, and will be taken up in Section 2.5.6. However, at this point I am more concerned with the presumed origins of the construction, and its status within the grammar of Old Persian. It will be evident that the assumption of a passive origin of the ergative construction has had considerable repercussions for how theories of the emergence of ergativity are conceived. In fact, it is fair to say that it essentially pre-determines the formulation of such theories. From this assumption it follows that the central facts to be accounted for are construction-internal syntactic changes, hence the emphasis on reanalysis, or the postulation of a transfer of subject properties. But there are other mechanisms involved in syntactic change, which may in fact provide a better explanation, and may not even require the kind of large-scale syntactic restructuring assumed by the passive-origins theory. Thus if it can be shown that the m. k. construction was not a passive, then the necessity for assuming massive construction-internal change may disappear, clearing the way for alternative, and ultimately empirically sounder conceptions of the change. In the following sections, we will re-assess the nature of the m. k. construction against a typologically bettergrounded notion of passive. 2.3 What is a passive? The claim that ergative alignment in Iranian (or any language) arose from a passive presupposes that passives and ergatives can be reliably distinguished from one another. However, in practice the distinction is not always clearcut: for certain constructions, linguists remain divided in their views (see esp. Manning (1996: Chap. 1) for discussion of these issues). In response to these difficulties, Comrie (1988) proposes a scalar approach to the distinction, What is a passive? 37 according to which a particular construction can be evaluated along several related, but independent, parameters as being closer to the ergative prototype, or the passive prototype respectively (see Shibatani (2004) for a more recent prototype approach to passive). Below, the four most relevant parameters for identifying a passive cross-linguistically are provided, whereby this formulation differs to some extent from that of Comrie: 1. Argument structure of the verb form: A passive verb form licenses a single core argument, semantically a Patient or Theme. The single core argument is a full subject, i.e., possesses all of the subject properties generally associated with an S in the language concerned. Semantically, however, a passive verb form implies the influence of some external agency in bringing about the event (prototypically an Agent). But expression of the Agent in an argument position is suppressed in the passive verb form.7 2. Markedness of the verb form: Within the paradigmatic system of verb forms available in the language, a passive verb form is the marked member of a voice opposition, contrasting with the morphologically less marked active, and derivable from the latter via a productive morphosyntactic process. 3. Syntactic status of Agent-phrase: The implied Agent may be expressed through an Agent-phrase (though in many languages it can not). Agent-phrases are always optional; the construction is fully grammatical without one. In the terms of Comrie (1988: 12–14), the Agentphrase displays a low degree of “integration into clause syntax”. There are few, if any, syntactic rules which make direct reference to it. 4. Pragmatic-Semantic configuration: Passive constructions primarily serve the function of backgrounding an Agent. Thus passives are used when the Agent is, relative to the Patient, either non-topical, or low in Animacy, or when its identity is unknown/communicatively irrelevant. The approach outlined above identifies prototypical passive values on four distinct parameters, but of course a particular construction may not yield identical values on all four parameters. This type of multi-featural approach 7 It is of course a moot point whether passive verb forms include an Agent in their thematic structures (cf. Kroeger 2004: 86, fn. 6). 38 Alignment in Old Iranian thus explicitly endorses the possibility of a construction being more or less passive-like, or more or less ergative-like. The typological investigation of voice systems and alignment in fact suggests that such a conclusion is wellfounded. Let us briefly consider how this approach could be applied to one of the best known cases of an ergative construction, that of Dyirbal (Dixon 1972). In the construction under consideration, the A is in an oblique case, Ergative, the O is unmarked (Absolutive). The verb shows no agreement with core arguments. An example is the following (the digraph <dj> is used here to indicate a laminal stop, for which Dixon uses a different symbol): (25) balan djugumbil baNgul yaóa-Ngu balgan class2:ABS woman class1:ERG man-ERG hit ‘(The) man is hitting (the) woman’ (Dixon 1972: 59) (25) appears in many respects to be the functional equivalent of its English translation. However, it differs from the English construction in some crucial respects. First, the O, rather than the A, controls various syntactic processes, which I will not go into here – see Dixon (1977: 367–376) for a useful summary. Second, the A can be omitted, giving the clause a reading often translateable with a passive in English: (26) balan djugumbil balgan class2:ABS woman hit ‘(Someone) is hitting the woman, the woman is being hit’ (Dixon 1972: 70) (25) differs from (26) only in the presence vs. absence of the A baNgul yaóaNgu. Furthermore, I have already mentioned that it is the O rather than the A which controls most syntactic processes, such as coreferential deletion in coordination. Given just these two facts, one is tempted to interpret (25) as a passive: the O is the subject, and the A is merely some kind of peripheral by-phrase equivalent. Indeed, some linguists have opted for precisely this analysis (see Comrie (1988) or Manning (1996: 37) for discussion). But there are other facts which militate against such an analysis. First, despite its omissability, the A can control certain syntactic processes in Dyirbal, for example coreferential deletion when the verb is modified via the suffix −Nura (Dixon 1972: 77–79). Second, the passive interpretation would essentially What is a passive? 39 imply that there are no transitive verbs in Dyirbal. But a number of other features of Dyirbal suggest that a transitive/intransitive distinction on the verbs is fundamental to the language (Manning 1996: 68–70). The final and most convincing argument against a passive analysis of Dyirbal is the fact that the verb form found in (25) represents the unmarked voice in the Dyirbal verb system, contrasting with the marked anti-passive. Referring back to the parameters introduced above, we may summarize our assessment of Dyirbal as follows, whereby only the first three parameters are considered here: Parameter 1: The construction is passive, on the assumption that it has a single core argument, the O, which appears to share most of the properties of an S in Dyirbal. Parameter 2: The construction is ergative, not passive, because the verb is an unmarked form (and contrasts with the marked anti-passive, not discussed here). Parameter 3: Evaluating the construction on this parameter depends crucially on how one weights the omissability of the A, and certain facts of clause combining. It will be seen that Parameters 1 and 2 yield unequivocal, but mutually incompatible results, while Parameter 3 remains open to interpretation. These mixed results are reflected in three decades of controversy surrounding the interpretation of Dyirbal’s alignment. Nevertheless, probably the majority of scholars, including Dixon himself, have opted for the ergative interpretation of Dyirbal, which suggests that Parameter 2, the markedness of the verb form, is privileged in some sense (see Dixon (1977) for an explicit statement to this effect). The optionality of the A (a sub-aspect of Parameter 3), on the other hand, appears to be less decisive. The analysis of other languages widely considered to have ergativity lends further support to this view. For example, in Samoan (Oceanic), transitive verbs are characterized as follows: 1. Ergative verbs [=transitive verbs, G.H.] do not distinguish between active and passive voice. 2. The argument expressing the actor (the ergative noun phrase or its pronominal equivalent) is always optional. (Mosel and Hovdhaugen 1992: 104, emphasis added) 40 Alignment in Old Iranian The following examples illustrate the optionality of the A: (see Mosel (1991: 182–184) for further discussion of the optionality issue): (27) Na sasa e le teine le maile i le lā‘au PST hit ERG ART girl ART dog LOC :DIREC ART tree ‘The girl hit the dog with a stick’ (Mosel and Hovdhaugen 1992: 416) (28) Na sasa le tama PST hit ART boy ‘The boy was hit’ (Mosel and Hovdhaugen 1992: 415) Contrasting the optionality of the O with that of the A, Mosel (1987: 458) notes that a clause such as (28) is fully grammatical, even in isolation. With no additional cues from the context, it is simply interpreted as ‘the boy was hit’, ‘someone hit the boy’. However, without the O, the clause is not readily interpretable. A clause such as (29) ’Ua fasi e le tama PERF hit ERG ART boy ‘The boy hit’ (Mosel 1987: 458) requires additional contextual information on who was hit before it can be interpreted (see Mosel and Hovdhaugen (1992: 700–704) for further discussion).8 Again, from an Indo-European perspective one might be tempted to analyze the constructions shown as passives, and the A phrase would then be a peripheral type of by-phrase. But on the basis of extensive syntactic analysis, Mosel and Hovdhaugen (1992) conclude that the construction is better considered an active one, an ergative construction, and the A, despite its optionality, is a core argument. An important factor in the analysis is the fact that (27) is the unmarked type of expression for transitive predications – in fact in Samoan, there is no active/passive distinction. Other Oceanic languages which have been analyzed as having ergativity also share the feature of an optional A, for example Tuvaluan (Besnier 2000: 126–129). 8 See Croft (2001: 273) for discussion and references on this type of contextuallydetermined omission. Re-assessing the m. k. construction 41 The conclusion to be drawn from these brief typological findings is that in characterizing a particular construction as ergative or passive, a number of distinct parameters need to be carefully distinguished. Thus rather than accepting a simple either/or classification of the m. k. construction as passive, it would be more enlightening to investigate it across distinct parameters in order to arrive at a more objective verdict. Finally, the cross-linguistic evidence suggests that, in evaluating a particular construction as passive or not-passive, the optionality of the A weighs less heavily than the systemic status of the verb form (marked vs. unmarked). This fact will have considerable relevance for our evaluation of the Iranian data. 2.4 Re-assessing the m. k. construction In this section, a more detailed investigation of the m. k. construction is undertaken on the basis of the four parameters just introduced: the argument structure of the verb, the systemic status of the verb form, the syntactic status of the Agent-phrase, and the pragmatic configuration of the clause (relative topicality/animacy of Agent and Patient). 2.4.1 The argument structure of the participles Evaluating the first parameter, the argument structure of the verb, is fairly straightforward. The verb forms found in the m. k. construction are formally the Old Persian participles in −ta, reflexes of Indo-European participles in *−to. They were originally verbal adjectives, with a resultative sense, and I will follow Nedjalkov (2001) in using the term ‘resultative’ participle to refer to them. This type of resultative participle can be derived from both transitive and intransitive verbs. But the participle itself is uniformly intransitive in the sense that unlike a finite verb form, it is incapable of assigning accusative case directly to its Patient argument. To this extent at least one can claim that the resultative participles of Old Iranian, and indeed their reflexes in most other Indo-European languages, are intransitive. Essentially, they shared many of the distributional features of adjectives, and in Old Persian they could still be used as attributive adjectives. However, they were 42 Alignment in Old Iranian already being used predicatively, often without copula support (see above), and this of course was the beginning of their career as verb forms. However, at the Old Iranian stage they undoubtedly still shared with adjectives the feature of intransitivity. Furthermore, these resultative participles, again as they do in countless languages, express the property of having undergone a particular state of affairs (hence resultative), rather than the property of having effected a particular state of affairs. They are thus both intransitive, and oriented towards the state of the Patient, rather than an Agent. Therefore, on parameter one, the argument structure of the verb, they are typical of passive verb forms, and this is undoubtedly one justification for characterizing the entire construction as passive. 2.4.2 The systemic status of the verb form The Old Iranian verb preserved most of the categories reconstructed for IndoEuropean: ideally, three stems (Present, Perfect, Aorist) plus a participle in −ta, formed directly from the root. The stems formed the basis for further verbal categories, e.g., active and passive forms, distinct moods, and of course person paradigms. But in Old Persian, the Perfect stem was already almost defunct (Drinka 2003: 83, fn.5), and the Aorist was very thinly attested. Thus at this stage already the verbal system had become simplified in comparison to what can be assumed for earlier stages of Iranian and Indo-European. In the transition to Middle Iranian, the Aorist stem went out of use, leaving the (West) Iranian verbal system based on a single fundamental opposition between forms built from the present stem, and those built from the participle.9 The Old Persian verbal system was thus in a transitional stage between the three-stem complexities of the ancient languages, and the binary opposition (participle vs. present stem) in most of Middle Iranian and later languages. The system of tense and voice oppositions in Old Persian has been reconstructed by Skjærvø (1985). Table 1, slightly modified from Skjærvø (1985: 221), summarizes the facts (the terms ‘Habitual Past’ and ‘Completed Past’ 9 In East Middle Iranian, e.g., Sogdian, additional stems can be identified (Sims-Williams 1989: 187). Re-assessing the m. k. construction 43 Table 1. Voice distinctions available in different tenses (Old Persian) Tense Present Imperf. Habitual past Completed past Perfect Active Passive + + + + + + − + Participle are used by Skjærvø (1985) to describe Tense/Aspect nuances over and above the more widely recognized tenses Aorist and Imperfect. I include them here for the sake of accuracy, but they do not impinge on the arguments at stake.) Table 1 shows that, for the Present and Imperfect tenses, both passive and active forms are attested. Likewise, Skjærvø (1985) finds a passive equivalent for the ‘Completed Past’. For the (rare) ‘Habitual Past’, the only attested verb forms are active. But when we turn to the Perfect, the sole representative of this category is the participle used in the m. k. construction (with or without the copula). And the participle does service both in simple, apparently active clauses (i.e., with a participle based on an intransitive verb: is gone) and in apparently passive clauses (i.e., based on a transitive verb: is closed). In other words, at this point in the verb system, there simply is no active-passive opposition. Speakers (or writers) who wished to express events in the Perfect had no choice but to use the participles. Skjærvø (1985: 222) himself concludes that there is no active-passive opposition in the perfect, but rather one of participles with an agent versus participles without an agent.10 Thus on the typological framework sketched in Section 2.3, it can be argued that the m. k. construction is not a passive because the relevant verb form is not the marked member of a voice opposition; in the perfect tense, the participle is the only verb form available. 10 Note that this statement is difficult to reconcile with the author’s own earlier claim (Skjærvø 1985: 218), cited above, according to which the m. k. construction “has all the marks usually associated with [agented] passive structures”. 44 Alignment in Old Iranian However, some scholars nevertheless maintain the view that the participle is the marked member of a voice opposition in Old Persian. These authors consider it to be the passive equivalent of the active Imperfect. For example, a construction widely attested in Old Persian is the following: (30) ima tya adam akunavam this which 1 S do:PST:1 S ‘this is that (which) I did’ (Kent 1953: DB IV,5–6) It will be noted that (30) is largely parallel with (17), both verb forms being inside relative clauses, and both expressing a similar content. Kent (1953: 88) interprets (17) as the passive version of (30), although strictly speaking, the tenses are different. How one interprets these facts will depend crucially on how one delineates the categories within which markedness relations are to hold. If, for example, we were to collapse the two categories of Perfect and Imperfect into a broader semantic category of ‘Past’, then we would find that the m. k. construction does indeed contrast with active verb forms (this is the standpoint adopted by Bynon 1980: 152). However, it is not a neat binary opposition, because, as Table 1 shows, the Imperfect had its own synthetic passive, so the participle cannot simply be considered to stand in a voice opposition to the Imperfect Active. The issue of the systemic status of the verb forms in the m. k. construction is thus disputable. However, it is nevertheless quite clear that within the tense identified as Perfect, there was no formal voice contrast available on the verb forms, a fact which Skjærvø (1985) recognizes quite clearly. Independently of Skjærvø (1985), Lazard (1984: 242) comes to a similar conclusion: Le cas du participe en −ta est tout différent [from that of the synthetic passive, G.H.]. Dans le tour qui constitue le nouveau parfait, le parfait vivant (manā krtam), il ne s’oppose a rien: il n’est donc ni actif ni passif; ou plutôt, puisqu’il est seul et conséquent non marqué au point de vue de la diathèse, il se range du côté de l’actif, terme non marqué de l’opposition de diathèse. Thus as far as the systemic status of the participles is concerned, both Skjaervø and Lazard converge in their verdict that the m. k. construction is essentially voice-neutral, hence the m. k. construction falls short of a prototypical passive on this parameter. Re-assessing the m. k. construction 45 2.4.3 The status of the Agent-phrase In this section, I will investigate the third parameter, the syntactic status of the Agent-phrase (in what follows I will sometimes abbreviate Agent-phrase to ‘A’, which should not be taken as a commitment to the core-argument status of the phrase). If the A can be shown to be a peripheral element, to which no syntactic rules make reference, then we would have strengthened the case for treating the m. k. construction as a passive. If, on the other hand, it can be demonstrated that the Agent-phrase possesses at least some properties of a core argument, then the passive interpretation is weakened. In the analysis of case systems, it is commonplace to draw a distinction between structural (also termed grammatical, or syntactic) and semantic (also termed concrete, or inherent) cases. The former are restricted to the expression of grammatical relations, such as Subject, Direct Object, and Indirect Object (Blake 1994: 32–34). They are required by specific syntactic configurations, and hence do not have a one-to-one correspondence with semantic relations. Semantic cases, on the other hand, express more directly specific semantic roles, such as Instrument, or Location. It would be naive to assume that we can always split the case system of a given language neatly along these lines. But despite numerous problematic instances – such as the one to be discussed shortly – the fundamental validity of this distinction (under whichever label) is accepted by most grammarians. The relevance of this to the issue of the syntactic status of the Agent-phrase is that in a prototypical passive, the Agent-phrase is in a non-structural case, often a local case form or an adposition. Thus if it can be shown that the Old Persian Genitive was a semantic case, then the Agent-phrase is more likely to be a purely peripheral constituent, and the m. k. construction would look more like a prototypical agented passive. The Old Persian genitive can best be understood against the background of its history. The traditional label ‘Genitive’ refers to the etymology of the actual morph, which is the reflex of the Indo-European Genitive. In early Old Iranian, this Genitive case stood in opposition to a Dative, still attested in Avestan. But in the Old Persian texts, the Dative is entirely absent, its functions having been absorbed by the Genitive. This explains why, from a purely functional, synchronic perspective, the Old Persian Genitive exhibits the typical properties of a Dative. Some researchers prefer the label Genitive/Dative, 46 Alignment in Old Iranian Table 2. Pronominal special clitics in Old Persian (Brandenstein and Mayrhofer 1964: 66–67) S ING . 1 2 ACC . =mā — G EN . =maiy =taiy P LURAL 3 1 2 =šim =dim =šaiy — — — — 3 =šiš =diš =šām which I will use in some contexts, but generally I abide by the traditional label. The first feature of the Genitive I wish to examine is its involvement in the cliticization process. In Old Persian, the oblique forms of the personal pronouns could be realized as clitics (clitic forms of the Nominative do not exist). Table 2 gives an overview.11 The syntactic rule determining the placement of these clitics is that they attach to the first word of the clause of which they are syntactically constituents, regardless of the syntactic category of that word (i.e., a Wackernagel position). Examples of Genitive clitics are found in (19), repeated here for convenience, and (31), while examples of Accusative clitic pronouns are provided in (32) and (33). (19) 11 avaTā=šām hamaranam kartam thus=3 PL:GEN battle do:PTCPL ‘thus by them battle was done’ Table 2 does not include the Ablative clitic pronouns because unlike the Genitive and Accusative pronouns, the Ablative pronoun clitics are not special clitics (in the sense of Zwicky 1977): they cliticize directly to their adjacent governing constituent, generally a preposition, rather than moving to a special position in the phrase. Re-assessing the m. k. construction 47 (31) aita=maiy Auramazdā dadātuv this=1 S:GEN Ahuramazda may.give ‘may Ahuramazda give this to me’ (Kent 1953: DNa,53–55), cf. also DPd,23–24; DPh,8; DNa,50–51,54–55 (32) pasāva=dim manā frābara after.that=3 S:ACC 1 S:GEN bestow:PST:3 S ‘after that (he) bestowed it on me’ (Kent 1953: DNa,33), cf. also DB I,60–61; DPd,7–8,13–14; DNa,33 (33) abara yātā Bābirauv kāra hya ATuriya hau=dim people which Assyrian DEM=3 S:ACC brought to Babylon ‘The Assyrian people – they brought it to Babylon’ (Kent 1953: DSf, 32–33) The last example shows that cliticization is sensitive to syntax rather than pragmatics: The phrase kāra hya ATuriya is a fronted topic, external to the clause, and is thus not treated as a first constituent for the purposes of cliticization. The Accusative clitic attaches to the first grammatical constituent of the clause, the subject pronoun hau (cf. Hale (1988: 35–36) for discussion of clitic placement). The cliticization rule illustrated in the preceding examples is restricted to Accusative and Genitive pronouns. We thus have a well-defined syntactic rule which must be formulated to include only these two cases. If we recall the characterization of a prototypical Agent-phrase given in Section 2.3, one of the most prominent properties is that “few if any syntactic rules refer to the A (agent phrase).” (Comrie 1988: 16). But as we have seen in this section, the Genitive A of the m. k. construction is subject to a major syntactic rule, that of cliticization. In other words, the cliticization rule defines a class of cases which includes the – clearly structural — Accusative, and the Genitive, but which excludes the other cases (recall that although there are also Ablative clitics, they are not special clitics in the sense defined above, that is, they do not undergo movement to clause-initial position). This constitutes one piece of evidence in favour of considering the Genitive a structural rather than a semantic case, and hence against considering the A of the m. k. construction to be an Agent-phrase of a prototypical passive. Let us consider now the functions of the Genitive. As can be seen from the preceding examples, the Old Persian Genitive was a very versatile case. 48 Alignment in Old Iranian It was, among other things, the normal case for what can loosely be termed Indirect Objects. The following examples illustrate this function, as does (32) above: (34) manā bājim abarā 1 S:GEN tribute bear:PST:3 PL ‘to me (they) bore tribute’ (Kent 1953: DB I,19) (35) manā hau=diš frābara 1 S:GEN it=3 PL:ACC bestow:PST:3 S ‘on me he bestowed them’12 (Kent 1953: DSs,6–7) If the Indirect Object is a pronoun, it may also undergo cliticization as described above, and indeed this option is probably more frequent: (36) aita=maiy Auramazdā dadātuv this=1 S:GEN Ahuramazda may.give ‘may Ahuramazda give this to me’ (Kent 1953: DNa,53–55), cf. also DPd,23–24; DPh,8; DNa,50–51,54–55 (37) Auramazdā=maiy upastām abara Ahuramazda=1 S:GEN aid bear:PST:3 S ‘Ahuramazda bore me aid’ (Kent 1953: DB I,87–88) The Genitive is also the normal case for Addressees of verbs of speech: (38) avaTā=šaiy aTaham thus=3 S:GEN say:PST:1 S ‘thus (I) said to him’ (Kent 1953: DB II,30) (39) avaTā=šām aTaham thus=3 PL:GEN say:PST:1 S ‘thus (I) said to them’ (Kent 1953: DB III,85) These examples will suffice as initial illustration – the polyfunctional Genitive will be taken up again below. In the present context, it is sufficient to note that the Genitive is the normal option for Addressees of verbs of speech, 12 In this example the Accusative clitic pronoun is not in the expected Wackernagel position; I have no explanation for this. Re-assessing the m. k. construction 49 or Recipients of verbs of transfer. To the extent that one wishes to consider Indirect Object a grammatical relation, then one could argue that the Genitive in Old Persian is a structural rather than a semantic case, because it regularly expresses a grammatical relation. To briefly recapitulate the arguments of this section, we have established that the Genitive, the case of the Agent-phrase in the m. k. construction, is subject to a major syntactic rule, the cliticization rule. Furthermore, it is the normal case for expressing Indirect Objects. Taken together, these two facts indicate that the Genitive case should rather be considered a structural rather than a semantic case. Now if we recall Parameter three above, the Agentphrase of the prototypical passive is normally in a semantic case, and is not subject to syntactic processes. In this respect, the Agent-phrase of the m. k. construction falls short of what one would expect of the Agent-phrase in a prototypical passive.13 2.4.4 The semantics and pragmatics of passives The fourth and final parameter to be investigated concerns the prototypical semantic and pragmatic constellation associated with passive constructions. Cross-linguistically, “a prototypical passive is used in contexts where the A is relatively low in topicality with respect to the P [=O]” (Payne 1997: 207). In addition to the discourse-related factor of topicality, the use of passives is often linked to animacy: when the A is high in animacy (e.g., a first or second person pronoun), the active form is normal. With low-animacy A’s, on the other hand, a passive form will be preferred. These tendencies can readily be observed in English. In (40a), the NP highest in animacy is cast in the subject role, which conforms with our expectations, and the sentence is correspondingly pragmatically neutral. In (40b), however, the NP lower in animacy is the subject, resulting in a slightly odd, though still grammatical, type of expression. (40c), on the other hand, where the most animate participant is demoted 13 Passives with Agent-phrases in the Dative are nevertheless attested in some languages, but crucially, they are practically unattested in Indo-European; see Section 2.6 for references. 50 Alignment in Old Iranian to a by-phrase, is distinctly unidiomatic, and would require a complex and unusual context to be appropriate: (40) a. I was stung by a bee. b. A bee stung me. c. ?A bee was crushed by me. While in English the correlation between animacy and syntactic function is at best a statistical tendency, holding at the level of frequency counts in texts, there are languages in which these factors actually determine the grammaticality of the relevant constructions. Comrie (1989: 192–193) discusses the interaction of voice and animacy in Southern Tiwa (Tanoan, Mexico and Arizona). If the A is first or second person, thus higher than or equal to the O (e.g., ‘I saw him/you’), then an active transitive construction must be used. If, on the other hand, the O is higher than the A in terms of animacy (e.g., ‘He saw me/you’), a passive construction is obligatory. A similar phenomenon is noted by Mithun (2004) for Yana (Hokan stock, California): there is a regular passive construction, but it cannot be used when the Agent is first or second person, i.e., maximally high in animacy. The above tendencies are reflected in a regularly observed cross-linguistic regularity: the subject of active transitive clauses (i. e. the A) is generally topical, and highly animate. Now both high topicality and high animacy converge in the personal pronouns, particularly of the first and second person, and as Du Bois (1987) has noted, an A is, statistically at least, overwhelmingly pronominal (or zero in languages that make regular use of zero-anaphora) rather than a full NP. Du Bois refers to this tendency as “Avoid lexical A’s”.14 There are thus well-established correlations between syntactic function on the one hand, and animacy and topicality on the other. The choice of passive versus active can to a large extent be motivated by the need to preserve these correlations. The observed tendencies are summed up in Table 3. 14 Related tendencies have been noted by numerous authors under various guises, many of them drawing on seminal work by Chafe and Givón (compare Mithun’s “Cross-linguistic subject selection tendencies” (Mithun 2004), or Lambrecht (1994)’s Topic acceptability scale. Du Bois’s formulation is important in the present context because he restricts his observations to transitive clauses. Re-assessing the m. k. construction 51 Table 3. The interaction of voice with animacy and topicality A ACTIVE PASSIVE High animacy/topicality Low animacy/topicality O Low animacy/topicality High animacy/topicality Let us return now to Old Persian. It is remarkable that in all the attested m. k. constructions, the Agent-phrase is maximally topical and animate: either a personal pronoun, most commonly first person singular manā (or clitic =maiy), or third person plural (clitic =šām, referring to humans), or (in one example) a kinship term ‘my father’. This distribution of animacy and topicality is precisely what one would expect of the A of an active transitive construction. It is precisely what one would not expect of the Agent phrase of a passive construction. Now one can of course argue that the correlations between animacy, topicality and syntactic relations are at best statistical tendencies. However, it seems distinctly odd, to say the least, that in Old Persian, the sole attested examples of a putative agented passive construction are used exclusively in contexts which are, cross-linguistically, typical of active constructions. Semantically and pragmatically, the apparent Agent-phrases are simply not what one expects to find in a passive construction. As mentioned above, it was Du Bois (1987) who pointed out that transitive subjects (A) are generally topical, hence definite and often pronominal, or omitted through zero-anaphora. Thus they are seldom lexical NPs. Obliques, however, are often an entry point for new referents into discourse, hence they are often indefinite, and full lexical NPs. In general, Du Bois’ claims have been confirmed in more recent cross-linguistic studies (cf. the papers in Du Bois et al. 2003). Yet remarkably, Harris and Campbell (1995: 253) take issue with these claims, not on empirical grounds, but because they go “against the grain” of their assumed passive-to-ergative shift. They argue that in the passive, the Agent-phrase is an oblique, hence, according to Du Bois, will display a different pragmatic configuration to the A of an active transitive clause. Therefore, “the fairly frequent shift from passive to ergative does not conform to what Du Bois’ account would lead us to expect.” Thus Harris and Campbell (1995) actually reject Du Bois’ claims on the grounds that they do not fit with their passive-to-ergative theory. But in fact Du Bois’ claims are 52 Alignment in Old Iranian fully correct: the Agent-phrase of a true passive is oblique, and is likely to be indefinite. The most obvious conclusion to draw from this would be that there is something definitely amiss in the claim that passive constructions are a “fairly frequent” source for the ergative construction. Yet Harris and Campbell, because of their commitment to the passive-to-ergative scenario, prefer instead to reject Du Bois’ empirically well-justified correlations. Further evidence that the A of the m. k. construction is not merely a syntactically peripheral by-phrase equivalent comes from their role in what appears to be control of deletion across coordinate clauses. Consider the following example: (41) a. avaTā=šami hamaranan kartam utā and=3 PL:GEN battle do:PTCPL and utā b. 0/ i avam Vahyazdātam agarbāya 0/ that Vahyazdata take.prisoner:PST:3 PL and fratamā anušiyā āhata c. 0/ i martiyā tya=šaiy 0/ men who=3 S:GEN foremost followers were agarbāya take.prisoner:PST:3 PL (a) ‘then battle was fought by themi and’ (b) ‘0/ i that Vahyazdata took prisoner and’ (c) ‘0/ i the men who were his foremost followers took prisoner’ (Kent 1953: DB III,47–49) A more natural translation of (41) is the following: (42) ‘they fought battle and (they) took that V. prisoner and (they) took [. . .] his foremost followers prisoners’ It will be seen that in (41), the first mention of the Agent is through a pronominal clitic, the A of a m. k. construction. In the following two clauses, the same referent is expressed through zero-anaphora, in what appears to be a typical case of coreferential deletion in coordinate clauses. This type of syntactic behaviour is highly typical of syntactic subjects, whereas the Agent-phrases of passive clauses are incapable of controlling coreferential deletion (Kazenin 2001: 912). Consider in this connection (17), repeated here for convenience and supplied with indices: Re-assessing the m. k. construction 53 (17) ima tya manāi kartam pasāva yaTā 0/ i xšāyaTiya that which 1 S:GEN do:PTCPL after when 0/ king abavam become:PST:1 S ‘This (is) that (which) was done by mei after 0/ i became king’ Here too the A of the m. k. construction is resumed via zero-anaphora in the following phrase. Compare now (17) with (43). The latter expresses essentially the same proposition, with the difference that the first clause is not an m. k. construction but an active one with a subject in Nominative (adam): (43) ima tya adam akunavam vašnā Auramazdāha hamahyāyā that which 1 S do:PST:1 S by.favour of.Ahuramazda same Tarda pasāva yaTā xšāyaTiya abavam in.year after when king become:PST:1 S ‘This (is) what Ii did by the favour of A. in one and the same year after 0/ i became king’ (Kent 1953: DB IV,3–5) It will be seen that in both (17) and (43) the pronoun is omitted in the second clause. In other words, the A is capable of controlling deletion across coordinate clauses, regardless of whether it is a Nominative subject, as in (43) or the Genitive of a m. k. construction, as in (43). I would urge caution in interpreting these examples at this stage, because it is uncertain whether subject omission in Old Persian is grammatically or pragmatically driven.15 However, it should be clear that pragmatically, the A of (41a) is topical enough to serve as the start of a deletion-chain, just as the Nominative subject of (43) is. In other words, the A in all the attested m. k. constructions exhibits the semantic and pragmatic features typical for A’s in active transitive constructions. 15 Thomas Juegel has pointed out another example, XPf 21–24, where a non-Nominative NP appears to control subject deletion in subsequent clauses. He refers to the “morphologische Unempfindlichkeit des Pivots” (p. c.). Again, it is extremely difficult to decide whether we are dealing with a more general, pragmatically driven type of NP-omission (in which case pivot is probably not the appropriate term), or a syntactic rule of coreferential deletion. 54 Alignment in Old Iranian 2.4.5 Summary of the four parameters The results for the evaluation of the m. k. construction across the four parameters introduced in Section 2.3 can be summed up as follows: Argument structure of the verb form The participles can be considered intransitive (unable to assign accusative case), and govern a single argument, Patient or Theme. On this parameter, the m. k. construction is a typical passive. The systemic status of the verb Debatable. Although the participle can be considered the marked member of a voice opposition, contrasting with the Aorist, this is a rather forced interpretation. Generally, it must be accepted that in that part of the verbal tense/aspect paradigm covered by the participle, there is no voice opposition. Thus the m. k. construction is not a typical passive on this parameter. The syntactic status of the Agent-phrase Again, the results are mixed. The Genitive case exhibits properties of a structural case in Old Persian (Indirect Object function, cliticization rule applies to it), thus making it certainly a less than typical type of Agent-phrase. Pragmatic-semantic configuration The attested m. k. constructions are pragmatically and semantically utterly atypical of passives, but conform fully to the statistically preferred form for active transitive clauses: highly topical (overwhelmingly pronominal) and highly animate A. They are arguably also the controllers of coreferential deletion across coordinate clauses, again a feature hardly compatible with a peripheral by-phrase status. On three of the four parameters, the m. k. construction diverges from a prototypical passive. In fact, its claim to being a passive rests largely on the argument structure of the verb form (Parameter one), and on the omissability of the Agent-phrase, a sub-aspect of Parameter three. But as noted in Section 2.3, the latter feature is compatible with constructions found in other languages considered to be ergative, not passive, by scholars of the relevant languages. And the evaluation of parameters two and four cast grave doubts on a passive interpretation. In addition to the difficulties with establishing the passive origins, there are also empirical problems in tracing the later developments that presumably followed the passive stage; these are taken up in Section 2.5.6. The semantics of the Genitive 55 While it is almost impossible to ‘prove’ the issue one way or the other, normal scientific procedure would be to maintain the passive analysis just as long as it appears to be the most plausible of the available alternatives. However, if an alternative is available which is consonant with the data, typologically well-founded, provides a solution for the problems discussed above, and avoids the empirical issues to be discussed in Section 2.5.6, then it would obviously be preferable. In fact, such an alternative is available, and will be outlined in the next sections. 2.5 The semantics of the Genitive The analysis I have in mind represents a return to the spirit, if not the letter, of Benveniste’s proposal discussed above. To justify it will require, however, much more than the four pages Benveniste devoted to the topic. To appreciate why Benveniste was largely correct, it is first necessary to take a more detailed look at the nature of the Old Persian Genitive. In this section I will be concerned solely with Genitives as clausal constituents, rather than with adnominal Genitives, which I discuss in Section 2.5.3. In Section 2.4.3 we have already noted the use of the Genitive to encode Addressees and Recipients. But the functions of the Old Persian Genitive also included a remarkably broad, though not fully arbitrary, spectrum of other semantic roles. For reasons that will become apparent later in the section, I consider the core function of the Genitive to be Benefactive (including Malefactive; I will simply refer to Benefactive henceforth), and the other attested functions can best be interpreted as radial extensions from this meaning. The following examples are fairly typical of Benefactive usage: (44) ava=maiy visam ucāram āha that=1 S:GEN all successful be:PST:3 S ‘all that was successful for me’ (Kent (1953: DSj,4), cf. also DSl,4– 5) (45) ava=taiy Auramazdā ucāram kunautuv that=2 S:GEN Ahuramazda successful may.make:3 S ‘that for thee may Ahuramazda make successful’ (Kent 1953: DB IV,76) 56 Alignment in Old Iranian (46) aniyahyā asam frānayam rest:PL:GEN horses buy:PST:1 S ‘for the rest (I) bought horses’ (Kent 1953: DB I,87) (47) hya šiyātim adā martiyahyā who happiness create:PST:3 S man:GEN ‘who created happiness for man’ (Kent 1953: DNa,3–4) (48) Auramazdā=taiy jatā biyā Ahuramazda=2 S:GEN smiter may.be:3 S ‘may Ahuramazda be a smiter unto thee’ (Kent 1953: DB IV,78–79) (49) ava=taiy Auramazdā nikatuv that=2 S:GEN Ahuramazda may.destroy:3 S ‘that may Ahuramazda for thee destroy’ (Kent 1953: DB IV,79–80) The Benefactive usage of the Old Persian Genitive continues the use of the Dative in Avestan (recall that the Old Persian Genitive took on the functions of the Dative after the latter disappeared), as in the following Gatha (Old) Avestan examples: (50) nōit=mōi vāsta xšmat anyō ¯ ¯ NEG =1 S:DAT herdsman 2 S:ABL other ‘There is no herdsman for me other than you [. . .]’ (Old Avestan, Yasna 21, Vs. 1) (51) at-cā=v¯@ mı̄žd¯@m aNhat ¯ ¯ thus-and=2 PL:DAT fruit become:PRES:SUBJ:3 S ‘Thus and fruit will arise for you:PL [. . .]’ (Old Avestan, Yasna 53, Vs. 7) Many of the attested Genitives do not readily lend themselves to classification in terms of the conventional inventory of semantic roles. Often it is impossible to decide whether a Benefactive or a Possessive meaning is intended. For instance, (47) above can equally plausibly be interpreted in a possessive manner (‘man’s happiness’). The following examples further illustrate the fluid transition between Benefactive and Possessor: The semantics of the Genitive 57 (52) imā dahyāva tyai=šām adam xšāyaTiya abavam these countries which=3 PL:GEN 1 S king become:PST:1 PL ‘these are the countries of/to/for which I became king’ (Kent 1953: DSm,5–6), cf. also XPh,14–15 (53) imā dahyāva tyā manā patiyāiša these province:PL which 1 S:GEN come:PST:3 PL ‘these (are) the provinces which came unto me’ (Kent 1953: DB I,13,18)16 (54) pasāva dahyāuš manā abava after province 1 S:GEN become:PST:3 S ‘afterwards the province became mine / for me’ (Kent 1953: DB V,13–14) (55) avam kāram Bābiruviyam jatā hya manā that:ACC army:ACC Babylonian smite:IMP:2 PL which 1 S:GEN naiy gaubātaiy NEG call:MED :3 S ‘smite that Babylonian army that does not call itself (=is not called) mine’ (Kent 1953: DB III,85–86), cf. also DB II,21,84 Besides Benefactives and Possessors, the Genitive also expressed the Experiencer, as in: (56) karāhyā naiy azdā abava people:GEN NEG known be:PST:3 S ‘(it) was not known to the people’ (Kent 1953: DB I,31–32) (57) ada=taiy azdā bavātiy then=2 S:GEN known be:PRES:3 S ‘then (it) is known to you’ (Kent 1953: DNa,43) Again, the Experiencer reading can be traced back to the functions of the old Dative, as shown for example in the following Old Avestan example: 16 Schmitt (1990: 11–12) confirms the reconstruction of the verb form. 58 Alignment in Old Iranian (58) aTā=mōi sa˛ sta vohū vāstryā thus=1 S:DAT show:AOR:2 PL good:PL pasturePL:ACC ‘Thus (you:PL) must show me good pastures’ (Old Avestan, Yasna 21, Vs. 1) 2.5.1 Indirect Participation One could of course describe the function of the Genitive in terms of a disjoint list of semantic functions (Recipient, Benefactive, Experiencer etc.). Kent (1953: 80–81) actually lists 11 different uses of the Genitive. But there are good reasons to believe that the list-format for describing the functions of the Genitive is missing some important generalizations. First, the list of functions that emerges turns out to be remarkably similar to the range of functions covered by Datives in a number of languages, in particular Indo-European languages, so it is unlikely to be a purely arbitrary collection. Second, in many cases it is almost impossible to classify particular examples of the Genitive in terms of the categories on the list, a fact that Kent (1953) duly acknowledges. Thus one must question whether the individual categories (Benefactive, Indirect Object etc.) are really relevant at all. Third, the bundle of functions covered by the Old Persian Genitive has proven to be remarkably persistent in the daughter languages, and where changes (functional expansion or reduction) occur, they tend to follow regular paths. All this suggests that rather than treat the Genitive in terms of a disjoint and ultimately arbitrary list, it would be more insightful to attempt a more coherent analysis. What I will be proposing is that the function of the Genitive can best be captured in terms of the concept of Indirect Participation, as opposed to Direct Participation. In terms of traditional semantic roles, the core of this notion is the Benefactive: a sentient being who is neither Agent, nor Patient, but whose interests are affected by, or perceived to be affected by, the event expressed in the verb. The notion I have in mind has been discussed by a number of scholars under various labels. Fried (1999: 500), discussing the Dative of external possession in Czech refers to a “a particular kind of affectedness, here called ‘interest’ ”. Similar formulations are found in the contribution to External Possession in Payne and Barshi (1999a), although the terminology differs. In Shibatani (1995) a similar concept is developed under the label of The semantics of the Genitive 59 “relevance”. The notions of interest, relevance, or indirect participation, are generally very similar, as I will point out below. The problem with them is that they are not readily compatible with accounts of semantic roles that work on a one-dimensional scale of control and affectedness. A typical example of one such hierarchy is the scale of the Thematic relations in Van Valin (2001: 31): (59) Agent > Experiencer > Recipient > Stimulus > Theme > Patient But with roles such as Benefactive, affectedness and control are not necessarily in a relationship of inverse proportion. For example a Benefactive may, depending on the particular context, be characterized by both a lack of control and a lack of affectedness. An attempt to capture this type of indirect participation is the model of Lehmann et al. (2004: 7–9), which I will briefly outline here. The authors distinguish two dimensions of participation in an event. First, the familiar dimension of direct participation, where participants can be distinguished in terms of their relative control over the event (ability to willfully instigate and conclude the event), and their affectedness by the event. But in addition, a second dimension of indirect participation is recognized.17 The rationale behind this second dimension lies in the fact that events are not solely definable in terms of the participants directly involved, but also in terms of those participants for whom the event is relevant in one way or another. The typical representative of the dimension of indirect participation is the Benefactive, who may actually not be directly involved in the event at all. Nevertheless, the event is highly relevant for the Benefactive, and it is certainly no accident that many languages have more or less grammaticalized means for expressing this role. Similarly, Experiencers and Possessors may or may not be direct participants; neither are necessarily affected, nor have control, but the event is often very relevant for them. Finally, Recipients and Addressees may also, depending on the circumstances, be considered indirect participants. Notice how the notion of Indirect Participation is prototypically restricted to sentient 17 The authors actually refer to this dimension as “involvement”, but I consider this term misleading, as it would also cover direct participation (a Patient is undeniably “involved” in a state of affairs). For this reason I prefer to distinguish between Direct and Indirect Participation 60 Alignment in Old Iranian beings, most commonly humans. It implies subjective perception, a consciousness of personal interests, hence the participants characterized by high Indirect Participation will generally be human. 18 Figure 1 gives a simplified overview of the semantic roles situated in regard to the two dimensions of direct and indirect participation. I have added the role of Causer, mainly to show how the two dimensions can be fruitfully applied in describing this otherwise rather difficult role: it shares high Control and low Affectedness with the Agent, but differs from the Agent in terms of lower directness of participation. As Lehmann et al. (2004) note, in the centre of the model the participant roles “become indistinct”: with regard to the Recipient role, there are obviously grounds for considering it both a direct and an indirect participant, hence its position at the intersection of the two axes. Likewise, different orderings of the roles Benefactive, Experiencer and Possessor than the one provided in Figure 1 would be quite possible. The crucial insight of this model is that there is a bundle of semantic roles defined precisely by the dimension of Indirect Participation, and this dimension intersects with Direct Participation at the level of Recipient. And the Old Persian Genitive covers that area characterized by maximum indirect participation. It is, for example, no accident that the Genitive is not used to express Instrument, or Patient, both roles of Direct Participation. The Genitive can thus be characterized as the case of the Indirect Participant, a function encompassing the traditional roles of Recipient, Benefactive, Possessor and Experiencer. If pushed for a statement, I would consider the Benefactive as the central meaning, as it provides the most natural semantic core from which the other attested functions can be derived. But in the context 18 Very recently, Næss (2007) has proposed a feature-based account of case-marking working on three semantic features: [±Volitionality(VOL)], [±Instigation(INST)] and [±Affectedness(AFF)]. The notion of Indirect Participation developed in my own work on Iranian languages would be captured in her system with the feature combination [+VOL, -INST, +AFF]. As she notes, this is the combination typically displayed by socalled ‘Datives’ cross-linguistically, generally covering the roles of “recipients, benefactives and experiencers” with a “fairly frequent” extension to possessor (Næss 2007: 199). Her findings provide strong typological confirmation of my proposals concerning the nature of Indirect Participation, but Næss (2007) unfortunately appeared too late to receive the more detailed treatment it deserves. The semantics of the Genitive 61 Low affectedness High control ✛ AGENT ✬ High affectedness ... R ECIPIENT E XPERIENCER ✻ B ENEFACTIVE P OSSESSOR C AUSER ✡ ✩ DIRECT PARTICIPATION ✡ ✡ ✡ ✫ ✡ ✣ ✡ Functional domain of the Old Persian Genitive ... Low control ✲ PATIENT ✪ L OCAL CASES ✈ INDIRECT PARTICIPATION Figure 1. The function of the Old Persian Genitive of Old Persian grammar I suspect the traditional labels such as Benefactive etc. have little relevance; they are just arbitrary divisions of the broad category of Indirect Participation which, in Old Persian, was grammaticalized into a single case. 2.5.2 External Possession In fact it is cross-linguistically not unusual that a single case expresses what I have termed Indirect Participation. The case commonly found in this function is the Dative, and the functions of such Datives very often include the coding of an External Possessor. According to Haspelmath (1999: 109), an External Possessor Construction (EPC) involves a possessive modifier that “does not occur as a dependent constituent of the modified NP, but NP-externally as a constituent of the clause.” The crucial aspect of this definition is the syntactic 62 Alignment in Old Iranian part: the Possessor is a clause-level constituent, rather than a sub-constituent of a NP. Note that the precise status of the External Possessor within the clause is left open, and it is indeed highly controversial, as we shall see. Many modern Indo-European languages have EPCs with Dative Possessors. The following example from Colloquial German is fairly typical: (60) Ihm ist der Vater gestorben 3 S:DAT is ART:NOM father die:PTCPL ‘His father died’ (lit. to-him is father died) As far as the semantics of the Dative cases involved in External Possession are concerned, Haspelmath (1999: 126) provides a list of functions typical for such cases. The resulting ‘semantic map’ covers the following categories: Predicative Possessor, Direction, Recipient/Addressee, Experiencer, Benefactive, External possessor and Judicantis.19 With the exception of Judicantis, for which I have found no example in Old Persian, this list could be applied verbatim to the Old Persian Genitive. Thus functionally, the Old Persian Genitive was quite unremarkable, exhibiting a profile very typical of the Datives found in numerous modern Indo-European languages. Likewise, Old Persian had an EPC, for example the following, where the Possessor is expressed as a clitic pronoun: (61) utā=taiy tauhmā vasiy biyā and.also=2 S:GEN seed much may.be ‘and may you have much seed (offspring)’ (lit: ‘and may to you/for you much seed be’) (Kent 1953: DB IV,75) (62) utā=taiy taumā mā biyā and=2 S:GEN seed NEG may.be:3 S ‘and may you have no offspring’ (‘and may to you/for you offspring not be’) (Kent 1953: DB IV,58–59), cf. also 73–74 It is worth pointing out that I do not consider the Genitive External Possessors as formally distinct from, say the Benefactives in for example (46), repeated here for convenience: 19 The Dative of ‘judgement’, typically associated with words such as ‘enough’, or ‘too’. The semantics of the Genitive 63 (46) aniyahyā asam frānayam rest:PL:GEN horses buy:PST:1 S ‘for the rest (I) bought horses’ (Kent 1953: DB I,87) Rather, I prefer to see the External Possessor Construction as one type of construal made available by the semantics of Indirect Participation. Thus the term EPC, for Old Persian at least, can be read as a label for describing those instances of clause-level Genitives which can be – semantically – interpreted as Possessors. But syntactically, there does not appear to be any clear distinction between clausal Genitives expressing external Possessors, or Benefactives, or Experiencers. For this reason, I reject an account of External Possession in terms of possessor-raising, according to which the external possessor is ‘raised’ from an underlying adnominal possessor. This type of account does not work for the free Datives of German, either, as shown by the following (adapted from Hole 2005): (63) a. Frank ist vor fünf Jahren gestorben. Frank is before five years died ‘Frank died five years ago.’ b. Jetzt ist seine Frau gestorben. Now is his wife died ‘Now his wife has died.’ die Frau gestorben. c. *Jetzt ist ihm Now is 3 S:DAT the wife died Attempted reading: ‘Now his wife has died.’ In the context given, the adnominal possessive seine in the (b)-version cannot be paraphrased with a free Dative (cf. the (c)-version), because the free Dative would imply not merely possession, but crucially, some form of affectedness (in German, often adversative). For obvious reasons, the connotation of affectedness is not available in this context. In the absence of data to the contrary, I will therefore assume that in Old Iranian the Genitive/Dative also coded the Possessor in an EPC. Note that the Possessor-reading is considered to be one construal of the broader category of Indirect Participant. Comparative evidence from the earliest attested stages of Indo-European also suggests that EPCs with Dative possessors were widespread. As Drinka (1999: 472) states, most Indo-European languages 64 Alignment in Old Iranian had a “mihi est construction (‘to me there is’) using ‘be’ plus a genitive or a dative NP to express possession.” (see also Baldi (2002: 86) for the Dative of possession in Latin, and Bauer (1999) for more general discussion.) 2.5.3 Adnominal versus External Possession At this point I would like to address a related issue, namely that of attributive Genitives in Old Iranian. It is this function which the label “Genitive” most obviously implies, and there is little doubt that an attributive use of Genitives as adnominal possessors was available. However, I strongly suspect that this usage was much more restricted than it has traditionally been portrayed, and that many of the supposed attributive Genitives are not in fact attributive (i.e., sub-constituents of an NP), but clause-level Genitives. In the many cases where a distinction is difficult to draw, the adnominal interpretation has been given preference, and Hale (1988) has actually attempted to extend it to further examples (see below). Hale’s account rests crucially on the assumption of a VP-constituent in Old Persian, to which certain constituents either belong or do not belong. Precisely this assumption makes it impossible to come to terms with the clause-level Genitives, for reasons I will attempt to lay out here. As a first example, consider (64), given with Kent’s translation (the transcription of the name Aspačanā follows Brandenstein and Mayrhofer (1964) rather than Kent): (64) Aspačanā vaçabara Dārayavahauš xšāyaTiyahyā isuvām Aspathines bow-bearer Darius:GEN king:GEN battle-axe:ACC dārayatiy holds ‘Aspathines, bow-bearer, holds the battle-axe of Darius the King’ (Kent 1953: DNd) On Kent’s reading, the Genitive here is adnominal, i.e., we have a structure as follows (I avoid commitment to a VP-structure here): [Aspathines] [Darius King:GEN battle-axe:ACC]N P [holds]V ERB The semantics of the Genitive 65 But equally plausible is a reading with an External Possessor along the following lines: [Aspathines] [Darius King:GEN]N P [battle-axe:ACC]N P [holds]V ERB On this interpretation, the Genitive is a clause-level constituent, and as mentioned, the intended reading as regards the Benefactive vs. Possessor distinction remains typically vague (in fact, the context implies that both are intended). On this interpretation (64) and the so-called ‘Benefactives’ etc. discussed above are all instantiations of clausal Genitives expressing Indirect Participation. On grounds of simplicity alone, this appears to be a desirable outcome. Consider a further example, where both an adnominal or an external possessor reading would be possible:20 (65) pasāva diš Auramazdā manā dastayā akunauš afterwards 3 PL:ACC Ahuramazda 1 S:GEN hand:DAT do:PST:3 S ‘afterwards Ahuramazda put them in my hand’ (Kent 1953: DB IV,35) Again, I see no principled reason for favouring the adnominal interpretation (and of course many modern Indo-European languages would prefer the External Possessor here, cf. German mir in die Hand geben). Consider how the External Possessor interpretation also provides a natural explanation for examples such as the following, where the Genitive actually occurs after its possessed: (66) hya adadā šiyātim martiyahyā who created happiness:ACC man:GEN ‘who created happiness for man’ (Kent 1953: DNb,1–2) On my account, the Genitive, being a clause-level constituent, enjoys some measure of word-order freedom. Thus I entirely concur with the Benefactive translation provided for this example by Kent (1953). Hale (1988), however, takes issue with Kent’s Benefactive reading of this, and other examples, on syntactic grounds. Briefly, Hale assumes a VP-constituent for Old Persian, 20 In Kent’s rendering of (65), the third person plural accusative clitic diš is written separately from the preceding word, a fact which clearly runs counter to our expectations. This requires verification against more recent text interpretations, but does not immediately impinge on the arguments at hand. 66 Alignment in Old Iranian and furthermore, that both Direct and Indirect Objects are part of the VP. The order of constituents in the VP is, according to Hale, verb-final. He then postulates certain types of extraction and movement processes (e.g., Topicalization), by means of which constituents may be moved out of their phrases. However, he stipulates that such processes may only occur once in a given sentence (the rationale for this stipulation is unclear). This means, according to Hale, that in (66), only one of the post-verbal nouns can be an object, as only one constituent can be moved out of the VP. Thus, according to the logic of this account, martiyahyā ‘man:GEN’ cannot be an indirect object (Hale’s interpretation of the Benefactive). Instead, it must be an attributive Genitive to šiyātim ‘happiness:ACC’. The argumentation is frankly contorted, resting as it does on at least two doubtful assumptions. The vastly simpler account is in terms of a clauselevel Genitive expressing Indirect Participation. Whether or not this type of Genitive is part of a VP (and indeed, whether Old Persian should be analyzed as having a VP), simply cannot be decided. A further drawback of Hale’s account is that it leaves the supposedly adnominal Genitive in an unusual post-nominal position. Hale’s analysis of the following example is likewise dubious: (67) tyām manā Auramazdā frābara which:ACC 1s:GEN Ahuramazda bear:PST:3 S ‘which Ahuramazda bore to me’ (Kent 1953: DPd,7–8) Now according to Hale, the Genitive manā is an Indirect Object, and must therefore have been extracted out of the VP to its current position. Apparently, it cannot have been topicalized (I can only presume that Hale considers tyām to be topicalized, and further topicalization would be ruled out by Hale’s ‘one-constituent-only’ rule; given that tyām is bound to sentenceinitial position anyway, I find this line of argumentation obscure, but see no other explanation for Hale’s proposals). Hale (1988: 33) goes on to suggest that the position of māna “can only be explained if we take the form to be enclitic (it is outside the VP but not in a topicalized position)” . Thus in order to explain the position of māna, Hale is obliged to assume that it has undergone cliticization. But the distinction between full forms of the pronouns and clitic pronouns is elsewhere consistently observed in the corpus, so it seems quite arbitrary to declare some examples of the full forms of The semantics of the Genitive 67 the pronouns to be enclitic merely to save a postulated rule of the grammar. In fact, there is simply no necessity for such ad hoc solutions to examples such as (67). The clause-level analysis of the Genitive proposed here readily accounts for these and many other cases, and is perfectly compatible with the typological findings regarding the syntax and semantics of External Possessors. And when one begins to examine the context more closely, numerous examples of apparent adnominal Possessors in Old Persian appear in a different light. Typically quoted examples are often provided out of context, for example manā pitā ‘my father’ (Kent 1953: DB I,4), a common phrase from the genealogies. But in context, it also can be considered a clausal Genitive: (68) manā pitā Vistāspa 1 S:GEN father V. ‘(says D. the King:) my father (was) V.’ (Kent 1953: DB I,4), translation from Kent Rather than interpreting manā pitā as a NP, as it appears to be when taken out of context, manā could be construed as an External Possessor, yielding something like ‘to-me (the) father (was) V. ’. Likewise, the phrase manā badaka ‘my subject’ is also open to a different interpretation, when the context is taken into account: (69) Vidarna nāma Pārsa manā badaka Vidarna by.name Persian 1 S:GEN subject ‘A Persian by name Hydarnes, my subject’ (Kent 1953: DB II,19–20), translation from Kent Kent translates Possessor and Possessed with an appositional phrase. But the entire example could be considered an independent clause with zero copula: ‘A Persian by name H. (was) subject to me’, leaving the Genitive māna as an External Possessor, perfectly comparable with the other examples above. A similar conclusion can be drawn from the following: (70) karā hya manā people which 1 S:GEN ‘people which (are) of me’ (=my people) (Kent 1953: DB II,27) Let us consider finally the sole example of the m. k. construction in which the Agent-phrase is apparently not pronominal: 68 Alignment in Old Iranian (71) yaTā adam xšāyaTiya abavam vasiy tya fraTaram when 1 S king become:PST:1 S much which excellent akunavam tya=maiy piça kartam āha ava adam make:PST:1 S that=1 S:GEN father:GEN do:PTCPL was that 1 S apayaiy protect:PST:1 S ‘when I became king I built much that was excellent. What had been built by my father, that I protected . . . ’ (Kent 1953: XPf,36–39), cf. also 47; XPa,19–20 The standard analysis is to see the sequence tya=maiy piça as consisting of the complementizer plus a possessive NP, with the possessor having cliticized on to the complementizer: that=my father. But equally plausible is a different reading: the cliticized Genitive can be taken as a Benefactive, a clausal constituent, giving the reading ‘all that father built in my interest/for me, that I protected.’ I see no reason to rule the latter interpretation out; it is perfectly compatible with all the other examples of cliticized Genitives seen up until now, where the indeterminacy between the Benefactive and Possessor reading is absolutely characteristic. A Benefactive reading is also available for the following example, and others like it:21 (72) yaTā=maiy pitā Darayavauš gāTavā ašiyava after=1 S:GEN father Darius throne:LOC go:PST ‘after my father D. went from the throne (=died)’ (Kent 1953: XPf,32– 34) Although there is often no ready means of deciding how a particular Genitive should be interpreted, there is one good reason to favour a clausal Genitive interpretation for many cases: the facts of cliticization, discussed above. If we recall that the general rule for clitics is to float to the leftmost constituent of their governing domain, then it is clear that we are obliged to treat the relevant domain for the Genitive as the clause, rather than a NP, because 21 According to the more recent interpretation of Schmeja (1982), this example would be more appropriately translated with ‘went to the place (=throne) of the Gods’, i.e., with a Goal reading of gāTavā. This does not affect the reading of the Genitive, however. The semantics of the Genitive 69 the Genitive clitic, like the Accusative clitic, regularly floats to the first constituent of the clause (commonly a complementizer). Below I have repeated some of the preceding examples with Genitive clitics expressing Possession, e.g., (61), Indirect Object or Addressee (37–39), and Benefactive (45). It will be observed that in all such examples, the syntax of cliticization is basically identical, regardless of the precise nature of the semantic role coded by the clitic. (61) utā=taiy tauhmā vasiy biyā and.also=2 S:GEN seed much may.be ‘and may you have much seed (offspring)’ (lit: ‘and may to you/for you much seed be’) (Kent 1953: DB IV,75) (37) Auramazdā=maiy upastām abara Ahuramazda=1 S:GEN aid bear:PST:3 S ‘Ahuramazda bore me aid’ (Kent 1953: DB I,87–88) (39) avaTā=šām aTaham thus=3 PL:GEN say:PST:1 S ‘thus (I) said to them’ (Kent 1953: DB III,85) (45) ava=taiy Auramazdā ucāram kunautuv that=2 S:GEN Ahuramazda successful may.make:3 S ‘that for thee / of yours may Ahuramazda make successful’ (Kent 1953: DB IV,76) The simpler analysis is this: the syntax of cliticization applies identically to Genitive pronouns, regardless of semantic function. Thus it applies equally to Genitives when used as Benefactives, Indirect Objects, Experiencers, or External Possessors. Those examples of apparent adnominal Genitives which involve a clitic Possessor are thus best analyzed as examples of External Possession; the Genitive is not a sub-constituent of a NP, but a clause-level constituent. Ultimately, we have to admit the possibility that in Old Persian, adnominal Genitives and clausal Genitives were quite simply indistinguishable in many contexts. Notice, however, that historically the adnominal Genitives have disappeared entirely in some West Iranian languages, their functional role now filled by the Izafe-construction, the precursor of which we have seen 70 Alignment in Old Iranian in (70).22 Prenominal Genitives do remain in formulaic phrases such as the Middle Iranian šāh-ān šāh ‘of-kings king’, i.e., King of Kings, but outside of such fixed expressions, there is little evidence for a robust prenominal Genitive construction in Early Middle Persian. In the Pahlavi and Early New Persian texts (pre 1000AD) examined by Heston (1976: 19–24), the prenominal Genitive, although attested, is much less frequent than the Izafe-construction. The suggestion that I am making is simply that even in Old Persian, the demise of the adnominal Genitive may have already been well under way. Interestingly, the best candidates for adnominal Genitives in Old Persian are in expressions of quantity, e.g., partitives such as māhyā ‘(14 days) of the month’, where the Genitive expresses a non-human entity. It is possible that differentiation was already under way along an animacy-based distinction, but this will need verification. 2.5.4 From indirect to direct participant While there is little doubt that Old Persian made widespread use of External Possessor constructions, this is not of itself sufficient to prove that the EPC provided the model for the m. k. construction. For such a shift would have involved an expansion of the semantics of the Genitive to include agentive readings. In other words, this would mean the step from expressing indirect to direct participation. What it entails is a chain of extensions something along the lines of Benefactive > Possessor > Agent. In fact, the semantic proximity of possession and agency has been repeatedly pointed out in the literature (Palancar 2002: 225–227). The central commonality between Benefactives, Possessors and Agents is semantically rooted in the feature of sentience (or simply [+human]), which is intimately bound to all three roles (less so for Agent than the other two). One area where 22 It is notable, however, that in a considerable number of modern West Iranian languages prenominal genitives are well-attested, for example Balochi and Talyshi. Just how the current split between West Iranian languages with postnominal Izafe-based Possessors, and those with prenominal Genitives developed historically remains one of the most puzzling aspects of comparative Iranian syntax – see Stilo (2005) for discussion of the current areal dimensions of this divide. The semantics of the Genitive 71 a marker of Recipient/Benefactive takes on a more agentive role is the use of the Dative to express the Causee in causative constructions in languages such as Turkish: the Causee is of course not a pure Agent, but is nevertheless (potentially) an instigator and controller of his or her sub-event in the overall causation state of affairs. Although it is probably more common for erstwhile Datives to develop into object markers (often restricted to animate objects), it is nevertheless attested, particularly in Indo-European, for Datives to develop new functions in the other direction, that is, to acquire agentive functions. Hettrich (1990) documents this tendency in a number of Indo-European languages, noting that wherever such Datives of agency occur, they generally combine with various types of non-finite verb forms – gerunds, participles, and infinitives – which is perfectly in line with the participial verb forms of the m. k. construction.23 Because of the high degree of similarity in the construction, and the fact that it is attested across so many old Indo-European languages, Hettrich (1990: 14) assumes that it reflects a “syntactic type” that must have been present in the proto-language. Constructions with non-finite verb forms often expressed not merely a state of affairs, but also implied a modal notion of obligation, ability or necessity; in these cases, the Obligee (i.e., the person under obligation to carry out the action expressed by the verbal noun) would be in the Dative. Precisely this type of construction is attested in Young Avestan, where, Datives of agency were found with “verbal adjectives” (Skjærvø 2003: 132) expressing agency coupled with obligation. The best-known construction of this type is the Latin gerundivum, proba- 23 Hettrich (1990: 7–8) notes that generally, Agented passives of any description were very rare in the earliest texts. However, they were vastly more frequent in conjunction with analytical passive-like constructions than with finite (medio-)passives. Hettrich quotes figures from the comedies of Plautus, in which just five Agent-phrases are found with passive verbs in −r (e.g., laudatur), whereas in the same texts, analytical passives (of the type laudatus est) occur with Agent-phrases 28 or 30 times. Similar proportions are reported for Hittite, Tocharian and Vedic (cf. Hettrich (1990: 8–9) for further references, and Luraghi (2003: 65) for Ancient Greek). These differences are too great and too consistent across the various languages to be pure chance, and further underscore the fundamental difference between the analytical and the finite passive forms – which of course reinforces one of Benveniste’s central claims, namely that the m. k. construction needed to be distinguished from a synthetic passive. 72 Alignment in Old Iranian bly related to the −ta participles of Old Persian (see also Van Langendonck (1998: 238) for similar examples): (73) adeundus mihi illic est homo go:PTCPL 1s:DAT there is man ‘I must go to the man there’ (Hettrich 1990: 13), Plaut.Rud.1298 Another example, this time from Old High German with an infinitive as verb form, is the following: (74) sint kind ze bëranne uns 1 PL:DAT are children to bear ‘to-us are children to bear (i.e., ‘we can bear children’) (Hettrich 1990: 13) On Hettrich’s view, the use of the Dative in this type of construction is an extension of the Benefactive/Malefactive function of the Dative, well attested in all the relevant languages. Thus he does not assume a distinct ‘Agentfunction’ of the Dative case; rather, the Dative covers a wide array of meanings, referred to by Hettrich as a ‘dative continuum’ (Hettrich 1990: 23). The precise interpretation of any particular instantiation of the dative therefore emerges from the construction in which it occurs.24 Needless to say, Hettrich’s notion of “Dative continuum” covers essentially the same grounds as what I have called the semantics of Indirect Participation. And in many of the ancient Indo-European languages, the Dative continuum was extended to include agentive meanings in constructions headed by non-finite verb forms. This is, I suggest, precisely the extension that must also have taken place in Old Persian, when clausal Genitives were combined with participles. From the general meaning of Indirect Participation, which included Possessor, the step to agency is actually quite a small one. Just how close Possessor and Agent meanings can be is revealed in one of the most frequently attested m. k. constructions, (19), repeated here for convenience: 24 Unfortunately, Hettrich says little on the syntax of the Dative in such constructions. Thus we do not come any closer to the issue of the syntactic status of these Free Datives. The Old Irish example quoted by Hettrich (1990: 13, ex. (18)) contains a form of reflexive pronoun which appears to be controlled by the Dative. However, this must be checked against the original sources. The semantics of the Genitive 73 (19) avaTā=šām hamaranam kartam thus=3 PL:GEN battle do:PTCPL ‘thus by them battle was done’ (translation from Kent 1953) One can well imagine that this sentence was ambiguous between an ExternalPossessor reading: ‘and then their battle was done’, and the agentive meaning given. Such ambiguities could have functioned as the bridging context spanning the gap from possession to agency. Viewed from this angle, there is really nothing spectacular in the suggestion that the m. k. construction emerged from the extension of an existent EPC; all the ingredients for this development can be found in other branches of Indo-European. The real question is: why did the m. k. construction in Iranian not simply remain an occasional, perhaps stylistically specialized syntactic variant, but ultimately became the sole means for expressing all past-tense transitive clauses? The reasons for the special development in Iranian are to be sought in the changes in the tense system of Iranian, which are taken up in Section 2.6.1. Let us return now to the arguments in connection with the proposed passive origins. I have suggested that this view is misguided, and claimed instead that the m. k. construction was an extension of an EPC. But to what extent does merely replacing the agented-passive account by an EPC-account actually solve any of the problems discussed in Section 2.4.5? It will be recalled that a major obstacle faced by the claim of an agented-passive origin lies in explaining the change in syntactic and pragmatic status of the Agentphrase: from syntactically peripheral, and associated with low topicality and low animacy, to syntactic subject, associated with high topicality and animacy. As noted in Section 2.4.4, the attested m. k. constructions exhibit the semantic and pragmatic properties typical of active transitive constructions, but completely atypical of passive constructions. Herein lies the principle advantage of the EPC analysis: External Possessors, unlike the Agent-phrase of a passive construction, already exhibit many of the typical features of transitive subjects: Possessors are typically topical and high in animacy (hence frequently pronominal). For example, in the data analyzed by Du Bois (1987: 822, 841), Possessors pattern on most criteria most closely with the A of an active transitive clause (percentage of animate versus inanimate, percentage of full NP versus pronominal mention). 74 Alignment in Old Iranian Table 4. Person and syntactic function for clitics in Old Iranian (A DP C OMP=Complement of adposition) SAP T HIRD P ERSON T OTAL I ND PART A DP C OMP / D IRO BJ TOTAL 19 2 21 6 10 16 25 12 37 Thus both External Possessors and transitive Subjects exhibit essentially the same clustering of high animacy and topicality. Furthermore, these parallels extend to the Indirect Participant role: Indirect Participants are with much greater than chance frequency Speech Act Participant (SAP) pronouns. A count of the clitics expressing Indirect Participant in the Old Avestan texts in the corpus Old Iranian Online25 confirmed this; the relevant figures are given in Table 4.26 Now even allowing for the restricted and formulaic nature of the corpus, there seems little doubt that these figures reflect a significant general finding: Indirect Participants are almost five times more frequently first and second person than third person. In the direct Object role, this trend is reversed. The Old Iranian data actually reflect a universal tendency, according to which SAP-pronouns are the most frequent exponents of the the Indirect Participant roles such as Recipients and Benefactives, while they are relatively infrequent in the Direct Object role – see Haspelmath (2002) for discussion. The important point in the present connection is that Indirect Participants are like Agents in that they are typically maximally animate and topical, hence the concentration of SAP-pronouns in these roles. 25 26 A collection of Old Iranian texts with glosses and grammatical analysis, compiled by the Linguistics Research Center, University of Texas, see: www.utexas.edu/cola/depts/ lrc/eieol. Of a total number of 45 clitic pronouns counted, eight proved to be of uncertain interpretation and were therefore not included in the figures given in Table 4. Note that in the text version used, the third person singular dative pronoun/demonstrative hōi is glossed as a pronoun, rather than an enclitic. I checked the six attested examples and found them to be consistently in a position compatible with a clitic analysis, and therefore counted these six forms as examples of clitic third person singular pronouns. But the interpretation of the third person pronouns (demonstratives?) remains undeniably problematic. The semantics of the Genitive 75 The proximity of possessor and agent is put very clearly by Comrie (1989: 225): Note further that, in the possessive construction, the possessor is usually high in animacy, while the possessed noun phrase is typically low. Given the correlation between high animacy and topichood, and between topichood and subjecthood, this provides a promising base for the shift of subject properties to the more animate noun phrase [. . .]. While Comrie in the above quote still refers to a shift of subject properties, his quote merely underscores the main advantage of the EPC-analysis when compared to the Agented-passive analysis: A Possessor is, from the outset, a much better candidate for subjecthood. Indeed, it is not altogether far-fetched to suggest that such External Possessors, and with them the Agent of the m. k. construction, were close to syntactic subjects already in Old Persian. Cross-linguistically, the Possessor in an EPC often displays syntactic properties of core, rather than peripheral arguments (Payne and Barshi 1999b: 3), so subjecthood is a possibility that cannot be dismissed offhand. Such a suggestion, while probably impossible to verify, would be tantamount to suggesting that the A of the m. k. construction was a kind of noncanonical subject. As Faarlund (2001: 102) notes, establishing the existence of non-canonical subjects in extinct languages is difficult, “since the decision to some extent must depend on negative data”. Nevertheless, data suggestive of subjecthood was introduced in connection with examples like (41) above, so the conclusion that the Agent of the m. k. construction was already in Old Persian rather subject-like does not seem at all unreasonable. It is conducive with the data, and with what is typologically plausible. As mentioned, the existence and analysis of non-canonical subjects remains controversial even in well-studied living languages like Russian, so a unanimously accepted solution for Old Persian is unlikely to be forthcoming. The problem of establishing the subjecthood of various types of non-canonical subjects is aptly summed up by König (2001: 977) in his discussion of External Possessors: No convincing and generally accepted solution, however, has so far been offered for the problems raised by the existence of external possessors for syntactic theory. In particular, external possessor constructions challenge the notion that clause-level syntax depends directly on the argument structure or valence of individual verbs, a notion that is part of many syntactic theories. 76 Alignment in Old Iranian Various syntactic, semantic and pragmatic explanations have been offered for the apparent mismatch between the argument structure of the verb and the extra (‘unlicensed’) argument found in external possessor constructions, none of which provides a convincing solution for all problems [. . .]. With the m. k. construction, we have an Agent, presumably not licensed by the verb because the latter is a resultative participle hence inherently intransitive. For most syntactic theories, which work on some version of the Projection Principle, argument status is thus excluded on principle. However, as mentioned in Section 1.4, there are viable syntax models which do not require all arguments of a clause to be licensed by the verb. Such an approach provides the most compelling account of the Old Persian m. k. construction. Furthermore, the notions of ‘constructional polysemy’, and ‘inheritance’ can clearly be applied to the scenario I have suggested in the preceding sections: Old Persian had a basic construction, let us call it the External Possessor Construction, whereby ‘Possessor’ here would include Benefactive readings (recall that in many examples, it is not possible to distinguish a Possessor from a Benefactive reading). The construction was characterized by an argumentlike (possibly subject-like) Genitive, linked to a predicate expressing an event which was maximally relevant to the Genitive-coded participant. The core type of relevance involved benefaction or possession. A radial extension of that notion was agency. The agentive reading, the m. k. construction, was probably in Old Persian an occasionalism, but it nevertheless crucially inherited the structural features of the original construction: argument, perhaps subject, status of the Genitive. 2.5.5 Variation and consistency in Agent-markers This section takes up the arguments of Cardona (1970) against Benveniste’s possessive analysis of the m. k. construction. Benveniste (1952/1966) had pointed out that the Agent-phrase in the m. k. construction was a Genitive, whereas the Agent-phrase of a true synthetic passive in Old Persian was a prepositional phrase with hacā. For Benveniste, this was a highly significant distinction, and it meant that the m. k. construction could not be a passive, but something else. Unfortunately, Benveniste did not get his facts entirely right. The semantics of the Genitive 77 As it turns out, at least one example of a synthetic passive with a Genitive Agent-phrase is attested, namely (21) above, repeated here for convenience: (21) avaiy Ūvjiyā arikā āha uta=šām Auramazdā naiy those Elamites faithless COP:3 PL and=3 PL:GEN Ahuramazda NEG ayadiya worship:PST:PASS ‘those Elamites were faithless and by them Ahuramazda was not worshipped.’ (Kent 1953: DB V,15–16) Compare (21) with the following, in which the synthetic passive verb form, ayadiya ‘was worshipped’ occurs without an Agent-phrase: (75) yadāyā paruvam daivā ayadiya ... where previously Daivas worship:PST:PASS ‘where previously the Daivas were worshipped . . . ’ (Kent 1953: XPh, 39–40) It would seem that the Genitive in (21) encodes a clearly peripheral by-phrase type of constituent.27 On the basis of this example, Cardona (1970) claimed that the m. k. construction, like synthetic passives, was simply a passive, and Benveniste had overstated the differences between them. However, there are problems with Cardona’s argumentation. First, the rarity of examples like (21) make it difficult to judge the interpretation of the Genitive (perhaps a Benefactive, or general case of Indirect Participation?). But even if the Genitive was used occasionally as an Agent-phrase with a synthetic passive in Old Persian, it is a usage that is not attested anywhere else in Iranian, where Agent-phrases, if used at all, are always coded with some type of adposition, or innovated case marker (see Chapter four on innovated case markers). And as noted above in connection with the Dative agents of ancient Indo-European, it is typical of Indo-European that such Datives are overwhelmingly associated with non-finite verb forms, rather than finite passives such as in (21). Thus the example is both language-internally, as well as cross-linguistically, highly unusual. But in fact, a closer look at the Agent-phrases of synthetic passives actually underscores Benveniste’s basic claim that the Agent-phrase of the 27 See Lazard (1984: 243–244) for discussion of the pragmatics of these examples. 78 Alignment in Old Iranian m. k. construction is a fundamentally different type of constituent to the Agentphrase of a synthetic passive. Let us briefly recapitulate the available evidence from Agent-phrases with synthetic passive verb forms in Old Persian. First, we know that an agent could be expressed with the preposition hacā followed by an Ablative. An example is the following: (76) tya=šām hacā=ma aTahya that=3 PL:GEN from=1 S:ABL command:PST:PASS ‘what was said(=commanded) to them by me’ (Kent 1953: DB I,19– 20) However, note that this preposition is attested only in combination with the verb form ‘was commanded’, and it is not unreasonable to link this particular verb with the ablative sense of the preposition hacā. Kent (1953: 87) implies just this when he refers to the Ablative expressing, for example, the person “from whom commands proceed (=agent)”. In other words, the use of hacā could be interpreted semantically, i.e., as linked to the meaning of this particular verb, rather than as a general marker of Agents with passives.28 Second, we know that the postposition rādiy is attested in Agent-phrases, although according to Skjærvø (1985: 215), there is some dispute regarding the agentive reading; an alternative (rejected by Skjaervø) is the reading ‘on account of’. Third, the Genitive can be used, as in (21), for which again only a single verb is attested. All in all, the evidence relevant to agented synthetic passives is thin. However, even in the small number of agented synthetic passives attested we find three distinct means of coding the Agent. What this suggests is that in Old Persian, the expression of the Agent with synthetic passive forms was only very weakly grammaticalized, that is, semantic or other factors mediated the choice of forms, rather than there being a strictly syntactic-based assignment. Such a finding is fully compatible with that established for the other Ancient Indo-European languages, where agented passives were both extremely rare (cf. Hettrich 1990), and were characterized by a lack of uniformity of expression. And of course the situation is to some extent mirrored by modern 28 Gerardo De Caro (p. c.) has suggested to me that the use of hacā with this verb could be motivated by the fact that it selects a Genitive for the Addressee, hence a Genitive phrase for the Agent would presumably be inappropriate for reasons of ambiguity avoidance. The semantics of the Genitive 79 languages such as German where there is often some choice in Agent-phrases (durch, von, mittels etc.). Now contrast this situation to that obtaining in the m. k. construction. Here we find in all attested examples, regardless of context, a single uniform expression of the Agent: the Genitive. Again, this points to a very different status of the Genitive in the m. k. construction than merely some kind of byphrase equivalent. The differences in consistency of expression are perfectly matched by the cross-linguistic data on differences between the Agent-phrase of a passive construction, and the A of an ergative construction. In his typological study, Palancar comes to the following conclusion: For Passive Agents, it is often the case that the language has more than one marker available to encode the NP expressing the Agent. [. . .] In contrast, ergative languages often only have one Ergative marker to encode a given NP in A function. (Palancar 2002: 23) Again, we find that the Genitive of the m. k. construction exhibits properties typically associated with arguments, including subjects (for example the A in an ergative construction), rather than properties expected of a purely peripheral by-phrase type of constituent. Thus in general, the arguments of Cardona (1970) do not detract from the conclusions drawn here and in fact, once the full range of data on Agentphrases with synthetic passives is taken into consideration, the picture that emerges provides further support for my conclusions. A much more damaging piece of evidence would have been if we had found an m. k. construction with a different type of Agent-phrase (e.g., with the preposition hacā). Significantly, this has not been forthcoming. 2.5.6 The missing link in the passive analysis As discussed, the passive-origins theory presupposes a major constructioninternal change, one that presumably will have involved a number of intermediate steps (see the discussion in Section 2.2 for details of the steps entailed). One of the few explicit statements on what kind of changes such a shift must have entailed in Iranian comes from Bynon (1980). She assumes, following Cardona (1970), that the origin of the ergative construction was 80 Alignment in Old Iranian a passive. Furthermore, she considers that a passive involves a “topicalization of the logical object” (153). Such a topicalization would, the argumentation continues, have been reflected in word order patterns, because as a rule, “topic precedes comment” (159). This leads Bynon (159) to the conclusion that “simple transitive sentences in (early) Middle Iranian might ideally have looked as follows:” Present: agent + object + verb Past: object + agent + verb According to Bynon, at this (hypothetical) stage, the past tense construction would have been both morphologically (case and agreement) and pragmatically a passive, in that it involved topicalization of the Object. Bynon then surmises that this apparently passive construction must have undergone a shift in topic assignment, and consequently word-order: the Agent became the Topic, and was fronted, giving rise to the AOV word order of most of modern Iranian transitive clauses in both tenses. The problems with this scenario are legion. First, even in Old Iranian, as discussed, the m. k. construction did not involve the topicalization of the O. The impression that it did presumably arises from the particular type of construction in which many manā kartam constructions occur, namely as relative clauses. Recall the now familiar example (17), repeated here for convenience: (17) ima tya manā kartam pasāva yaTā xšāyaTiya that which 1 S:GEN do:PTCPL after when king abavam become:PST:1 S ‘This (is) that (which) was done by me after (I) became king’ Superficially, one might claim that this construction serves to topicalize ‘that which was done’, and indeed, that is probably the case. But it is not the manā kartam construction which achieves the topicalization, but the relative clause construction within which it is embedded. The proof of this is provided by the fact that essentially the same topicalization is achieved in an active construction, as in (30) from above: The semantics of the Genitive 81 (30) ima tya adam akunavam this which 1 S do:PST:1 S ‘this is that (which) I did’ But the main drawback with Bynon’s proposals is the complete lack of empirical support for an initial OAV word order, subsequently changed through a shift in topicalization. Rather, we find remarkably consistent preference for the A (if present at all; probably more commonly it was omitted, or cliticized) to occur clause-initially, and not a scrap of evidence that this was ever any different. Bynon’s “hypothetical” early Middle Iranian nevertheless has some value in that it spells out just what the passive-origins theory entails. Empirically, however, it is without support.29 Most other proponents of the passive-origins theory remain silent on the intermediate stages, or assign them to the dark period between Old and Middle Iranian (some 5–6 centuries, depending on which texts are considered). Of course this is a considerable time-span, and perhaps the entire bundle of changes that must have accompanied a passive-to-ergative shift really did happen during this period, leaving no trace in any of the later languages. I would be prepared to accept that as a working hypothesis, if there were no alternatives. But as pointed out in the preceding sections, there is an alternative theory which has several other advantages over the passive-origins theory, and does not require the assumption of a “dark period” in which major restructuring occurred. 29 In a more recent paper, mostly concerned with Indo-Aryan, Bynon (2005) abandons her earlier passive-origins view in favour of a Possessor Raising analysis. While much closer to my own view than her previous work, two points of disagreement remain. First, I consider it unnecessary to invoke Possessor Raising. The External Possessor is, on my view, not raised out of an embedded possessed NP; it is merely a contextually-driven construal of the broader notion of Indirect Participation (see Section 2.5.2 for the arguments in support of this view). Second, Bynon relates the developments to an assumed element of evidentiality in the verb forms concerned, for which I see no evidence in Iranian. But her revised proposals are in many respects compatible with mine. 82 Alignment in Old Iranian 2.6 Summing up the alternatives In the preceding sections I have presented the justification for abandoning the theory that ergativity in Iranian arose from an agented-passive construction. I have also developed an alternative, according to which the m. k. construction is an extension of an External Possessor construction, already present in Old Iranian, from which the m. k. construction inherited crucial structural features. In particular, the semantic, pragmatic and syntactic constellation associated with Possessors in EPCs coincide with those associated with Agenthood, thus easing the shift from EPC to m. k. construction, and ultimately an ergative construction. The restricted nature of the Old Iranian data means that proving these claims in any absolute sense is probably impossible. Rather, we are obliged to weigh up the relative advantages of the competing hypotheses. The criteria by which such hypotheses can be judged are as follows: compatibility with the available data, typological plausibility, and considerations of simplicity and coherence of the arguments. In all three respects, the present theory offers significant advantages over the agented-passive theory. Let me briefly sum up those advantages: 1. The fact that the A is in the Genitive (functionally a Dative), poses a problem for the passive analysis, because Dative Agent-phrases are typologically very unusual in Indo-European (though attested outside the family): of the 30 Indo-European languages investigated in Palancar (2002), none show evidence for a Dative source of their Agent marker in the passive. But the Genitive/Dative form of the Agent follows quite naturally from the assumption that the construction was originally an EPC. 2. The fact that the verb form is voice-neutral poses a serious problem for the passive interpretation: it is not the marked member of a voice opposition. For the EPC interpretation, however, this is not a problem. EPCs typically involve intransitive verbs, and the participles (with or without copula support) are typical of the types of predicates found in EPCs. 3. The Agents in m. k. constructions are maximally high in topicality and animacy, both features compatible with subject status, but not with status as the by-phrase of an agented passive. 4. The variation in form typically associated with agent-phrases of pas- Summing up the alternatives 83 sive constructions (and also found in Old Persian) contrasts bluntly with the fully uniform expression of the A in the m. k. construction (cf. Section 2.5.5). 5. The present theory assumes that the A of the m. k. construction already exhibited features typical of subjects, certainly high topicality and animacy, and possibly syntactic subject features as well. These are fully compatible with what is known about Possessors of EPCs. Thus the theory has a significant advantage in terms of simplicity: according to the passive-origins theory, the Agent-phrase must have traversed the entire scale from syntactically peripheral to the syntactically most privileged clausal constituent, the subject. The present theory, on the other hand, postulates no such massive construction-internal change because it assumes that the A was further down the road to subjecthood from the outset. 6. The theory outlined here is supported by what can be found in later stages of Iranian, while for the agented-passive theory, no clear support is attested (cf. Section 2.5.6). 2.6.1 The triggering factor As mentioned, the m. k. construction is only marginally attested in Old Persian, restricted to at most two verb lexemes in fairly formulaic contexts.30 30 To avoid misunderstandings, I am referring to uncontroversial examples with an overt Agent-phrase. Participial forms from transitive verbs without such an accompanying nominal or clitic Agent-phrase were very much more widespread; Thomas Juegel (p. c.) has pointed out to me around ten transitive verbs for which such forms are attested. It may in fact be possible to interpret some of them as m. k. constructions with discoursedeleted As, which would significantly increase the frequency of the construction. However, as we shall see in the discussion of Middle Iranian, it is often very difficult to draw a clear distinction between a m. k. construction with a discourse-deleted A, and the simple stative/passive, and agentless, use of the participle. I have therefore opted for the more conservative solution of only counting those examples with an overt A as m. k. constructions. Even if the number should turn out to be greater, this would not alter the main thrust of the arguments here; it would merely suggest that the developments were already further underway in Old Persian than has generally been assumed, a conclusion that I find quite attractive. 84 Alignment in Old Iranian This may of course be an artefact of a small and not particularly representative corpus. It is quite possible that it may already have been much more widespread in the vernacular of the time, but of course we will probably never know. Whatever its status may have been at the time the Old Persian texts were inscribed, within the space of a few centuries its progeny had become the sole means available for expressing past transitive constructions in Middle Iranian. The question that remains is therefore: What drove this massive expansion? Note that the main ingredients of the m. k. construction (EPCs with Dative Possessors exhibiting some degree of subjecthood, Datives with agent semantics linked to non-finite verb forms) are readily observable in other Indo-European languages (e. g. Icelandic). But within Indo-European, only in Indo-Aryan and Iranian do we find ergative alignment developing in past tenses (although parallels are found in Lithuanian, which definitely require closer investigation). In other words, the factors outlined above were undoubtedly a necessary condition for the development of ergativity, but they are not a sufficient one. We must therefore seek other factors specific to the Iranian and Indo-Aryan branches which triggered the wholesale alignment changes. The most obvious solution has very little to do with syntax, but a lot to do with historical morphology. Consider for a moment the array of different constructions available to express past transitive events in Old Persian: (77) a. AORIST: avaTā hamaranam akumā then battle do:AOR:1 PL ‘then (we) made battle’ (Kent 1953: DB I,90) b. I MPERFECT: avadā hamaranam akunauš there battle do:IMPF:3 S ‘there (he) made battle’ (Kent 1953: DB II,23) c. PARTICIPLE, cf. (19) avaTā=šām hamaranam kartam thus=3 PL:GEN battle do:PTCPL ‘thus by them battle was done’ Summing up the alternatives 85 In (77), three distinct constructions are found all expressing, as far as the limited data allow us to interpret, essentially the same event (though with differing Agents), and with a similar temporal reference (past). But while (77a) and (77b) have finite verb forms, (77c) uses the participle in a m. k. construction. Crucially, the type of finite past tense verb forms found in the first two of these examples died out, although the reasons for the demise of these tenses is ultimately beyond our knowledge. The consequences of this loss of tense morphology for the syntax, however, were momentous. Essentially, it left the language bereft of finite verb forms to express past tenses. The sole option available in the grammar was thus the participle, and as the parallels in (77) suggest, and I have argued for throughout this chapter, it was already in Old Persian a viable alternative to the finite forms. It is thus not surprising that with the demise of the finite past tenses, constructions based on the participle stepped in. What we are primarily dealing with is thus a pull-chain development, with the initial catalyst coming from changes in the verb morphology. But while the semantic notion of ‘pastness’ could be expressed by the participle, a serious problem arose with transitive verbs: a participle cannot assign accusative case, and neither can the copula it combined with. Therefore, there was simply no verb form available in past tenses which could have assigned an accusative case in the same manner which present-tense verb forms could. To express transitive propositions in the past tenses, then, a different construction to the Accusative-construction of the present tenses was required.31 The m. k. construction was the obvious alternative, because it permitted the expression of two arguments, it already had the pragmatic constellation typical of transitive clauses, and the Agent of the m. k. construction may well already have had subject properties at the Old Iranian stage. We can assume that it gradually extended the range of verbal lexemes it was associated with, successively compensating for the loss of the finite tense forms, until it became the sole available means of expressing past transitive propositions 31 Many western Indo-European languages which developed an analytical perfect tense based on the resultative participle were faced with the same problem: the participles themselves cannot assign accusative case. The solution followed for most of these languages is to use a transitive auxiliary (usually have or its equivalent) with participles from transitive verbs. 86 Alignment in Old Iranian throughout Iranian. The mechanism involved in this process is thus primarily one of extension, driven ultimately by changes in verbal morphology, rather than syntactic reanalysis of the construction itself. In a sense, the alignment change characteristic of the syntax of past transitive constructions in Iranian (and presumably Indo-Aryan, though I make no specific claims here) is merely a by-product of the loss of finite past-tense morphology. Notice that locating the triggering factor in deep features of historical verbal morphology has the advantage of explaining why all members of a particular branch were affected, regardless of their areal location. Verb morphology, to a much stronger degree than syntax, is a trait that is deeply linked to genetic relatedness, so fundamental changes in the ancestor language are likely to be inherited by all its descendants. Just what triggered the changes in the verb system that led to the wholesale loss of finite past-tense verb forms is a matter for historical phonology and morphology, beyond the scope of this book. 2.7 Conclusions In this chapter a theory was spelled out for accounting for the initial change from accusative to ergative alignment in past tenses in Iranian. On my account, a number of ingredients contributed to the wholesale abandonment of accusative alignment in past tenses of Iranian. First, I assume the prior existence of an External Possessor Construction, with the Possessor coded as a syntactically clause-level NP in the Genitive case, arguably a structural case in Old Iranian. ECPs are widely attested in ancient Indo-European (cf. Hettrich 1990; Bauer 2000), and their existence in Old Iranian is hardly controversial. The second ingredient is the semantics of the Old Persian Genitive case, covering the domain of Indirect Participation: Benefactive, Experiencer, Possessor, Recipient and Indirect Object. Third, the extension of the ECP to constructions with participles as predicates: the m. k. construction. The semantic shift implied here is primarily from Possessor to Agent, undoubtedly a plausible one, and well-supported by the data from other ancient Indo-European languages, where Datives frequently occur with nonfinite verb forms in Agentive meanings. The shift was eased by the properties of high topicality and animacy, common to both Possessors and Agents, and Conclusions 87 both conducive of subjecthood. Finally, the spread of the m. k. construction to become the general means of expressing past transitive propositions with all verbs was driven by the loss of the appropriate finite past tense morphology in the verbal system. I have argued at length against the assumption that the m. k. construction was an agented passive by pointing to a number of serious drawbacks, both theoretical and empirical, which this theory entails. More generally, in the theory outlined here, there is simply no necessity to postulate a wholesale “transfer of subject properties” (Estival and Myhill 1988), or a “reanalysis” (Harris and Campbell 1995: 243–245). In fact, as we shall see in the next chapter, the Middle Iranian ‘ergative construction’ does not demonstrably differ syntactically from the Old Persian m. k. construction. The differences are located primarily in the massively increased frequency of the construction, which by Middle Iranian had advanced to the sole means of expressing past transitive propositions. But that is not a difference best captured in terms of construction-internal changes, but rather one in terms of markedness and status within the entirety of the forms available in the grammar at any given time. One of the problems in previous analyses of the emergence of ergativity in Iranian lies in the constraints of the theoretical straitjacket provided by the “ergativity paradigm”, as outlined in for example Dixon (1994) or Manning (1996). In fact, I believe that deeper insights into the Iranian developments can be gained through the literature on External Possessors, and on noncanonical subjects. The assumption that this type of construction provided the point of departure for the later developments allows for a vastly simpler and more natural account of the changes. In a sense, the story of the emergence of ergativity in Iranian can be seen as the extension of non-canonical subjects to a specific, morphologically-defined environment: the past tense of transitive verbs. In other languages which have non-canonical subjects, they are generally restricted to a semantically-defined group of predicates, in particular psych-verbs or verbs of possession. The shift to the past transitive environment in Iranian was, as I have argued, largely motivated by highly specific developments in the verbal morphology, which were largely restricted to the Indo-Iranian branch of Indo-European and are therefore rare elsewhere (although similar tendencies have been observed for Balto-Slavic, cf. Ambrazas (2004) for references). 88 Alignment in Old Iranian Non-canonical subjects are messy. Even for well-researched living languages such as Japanese or Russian, the interpretation of this phenomenon continues to be highly controversial. For poorly-attested, extinct languages, the difficulties of interpretation increase exponentially. It is simply not possible to provide the kind of in-depth syntactic analysis which would be necessary to establish the extent of subjecthood commanded by the relevant constituents. For this reason, the theory I have set out in this chapter must remain to some degree speculative. However, I believe the typological parallels, the supporting evidence from related languages, and the evidence from the daughter languages (see especially Section 5.6) all point towards a construction of this nature as the most plausible origin of ergativity in Iranian. Chapter 3 Western Middle Iranian This chapter looks at the major developments in alignment over the Middle Iranian period, focussing on the western branch of Iranian. All the languages attested from this period onwards were characterized by what I have termed Tense-Sensitive Alignment (TSA): The alignment of past transitive clauses differs systematically from the alignment found in all other environments. Even though some some of the modern languages do not have TSA (e. g. Persian), the historical evidence shows quite clearly that they must have passed through such a stage. In the grammar of Past Transitive Constructions, clitic pronouns came to play a pivotal role, so this chapter also lays the foundations for understanding the developments in the system of clitic pronouns that underly Past Transitive Constructions in numerous modern Iranian languages. 3.1 Middle Iranian The term Middle Iranian is loosely used to refer to Iranian languages attested from the period between the collapse of the Achaemenidian Empire (4–3 century BC .) and the Arabic-Islamic conquest of Iran (8–9 century AD .). The languages identified as Middle Iranian were spoken (written) across a vast area, stretching as far to the east as Central Asia and Pakistan, and over a time span of more than a thousand years. Unsurprisingly, they differ considerably, and the nature of the relationships between them remains in many cases controversial. Traditionally, the languages are divided into two groups, Western Middle Iranian and Eastern Middle Iranian. The best-attested languages of Eastern Middle Iranian are Sogdian, Choresmian and Khotanese, but additionally some other languages are recognized such as Bactrian and Sarmatian. Western Middle Iranian is attested in Middle Persian1 and Parthian. This 1 In some publications (e.g. Heston 1976), reference is made to “Pahlavi”, but the term Pahlavi is probably better reserved for a particular script, and texts written in (some variant) of that script, than for a group of languages. In citing Heston (1976), I have used 90 Western Middle Iranian chapter draws almost exclusively on Western Middle Iranian data; it was not feasible to undertake adequate coverage of the relevant topics in Eastern Middle Iranian, which remains an urgent topic for future research. As mentioned, Western Middle Iranian is conventionally divided into two main languages, Parthian, and Middle Persian. The differences between the two are, in terms of the morphosyntactic phenomena under investigation here, not great. Indeed, earlier researchers did not consider them to be two languages, but two dialects (Sundermann 1989c: 106). Both were literary languages, used in administrative and sacred texts across large areas, hence it is problematic to identify them with specific, geographically defined vernaculars. Middle Persian is the forerunner of Modern Persian, while Parthian represents a language, or group of languages, spoken in an area roughly covering the province of Khorasan in today’s Iran, the region Gorgan, and Turkmenistan. Middle Iranian texts have survived in a large variety of different scripts, mostly based on some version of the Aramaic script, a fact that reflects the role of Aramaic as the language of administration in the Achaemenidian Empire. The scripts contained, to varying degrees at different stages, fossilized Aramaic elements, referred to by Schmitt (1989) as “heterograms”. They did not represent directly the phonology of the relevant Iranian language, but rather stood for particular meaningful elements. Heterograms could be modified by a phonetic component indicating, for example, inflectional morphology. However, the mixed nature of the script undoubtedly complicates the interpretation of much of Middle Iranian morphosyntax. For example, some forms of the first and second person pronouns, and indeed the very existence of a Direct form of the second person pronoun, remain controversial due to problems of interpretation of the writing system (see Sims-Williams (1981: 166) on Inscriptional Middle Persian and Sundermann (1989b: 131) on Parthian). In conventional transcription, heterograms are rendered with capital letters, and this convention is followed here when citing the sources. “Middle Persian” where she refers to “Pahlavi”. An alternative, and perhaps preferable basic distinction, could be drawn between Inscriptional Parthian and Book Pahlavi. Middle Iranian 91 An important body of texts is related to Manichaeism, the religion founded by Mani (born 216 AD, see Boyce (1975: 1–14) for a concise overview). The use of Parthian as the sacred language of Manichaean religious groups survived in Central Asia up to the 13 century AD (Sundermann 1989b: 114). The Middle Iranian Manichaean manuscripts were discovered in the ruins of Manichaean monasteries in Chinese Turkestan, but all the extant manuscripts “have been badly mutilated [. . .]. Very few single pages even survive intact and most of the material is in small, damaged fragments.” (Boyce 1975: 14) Thus the difficulties of interpretation engendered by the mixed script are further compounded by the fragmentary nature of much of the available texts. Coupled with these practical issues of text reconstruction, the arcane religious nature of the content render interpreting the texts an extraordinarily delicate task. Some of the difficulties confronting a syntactic analysis of Western Middle Iranian should now be apparent. Nevertheless, there is a reasonable body of texts available, and a considerable number of reliable secondary sources and commentaries. In writing this chapter, I have drawn most heavily on the following text editions: Boyce (1975), MacKenzie (1999c), Sundermann (1981), Durkin-Meisterernst (2006) and Sundermann (1973). A valuable source is also Heston (1976), which is directly concerned with grammatical analysis of four Middle Iranian languages. For comparison with more recent stages of Persian, Paul (2002), Lazard (1963) and in particular, Williams (1990a;b) have been referred to. In citing examples, I have followed the conventions used in the respective sources. If a text edition has been supplied with both a transliteration from the original script and a transcription, I quote the transcription, as these are more transparently readable. Nevertheless, some inconsistencies are unavoidable, but in a work concerned with syntax rather than phonology, I believe they remain within tolerable limits. Having outlined the background and sources, I will proceed to map out some of the most important changes between Old and (Western) Middle Iranian with regard to alignment. The changes can only be properly understood against the background of general changes which affected the inflectional morphology of Iranian languages. As I argued in the last chapter, what appear to be alignment changes are more profitably seen as by-products of changes to the verbal and nominal inflection. 92 Western Middle Iranian 3.2 Past Transitive Constructions The most fundamental change between Old Iranian and Western Middle Iranian was the loss of much inflectional morphology. The case system underwent massive syncretism, which deprived most nouns of any systematic morphological case differentiation (see Appendix, A.1 for an overview of case in Old Persian, and Windfuhr (1992) for an overview of historically attested changes). Thus a major diagnostic for determining alignment, that of nominal case, is of only very restricted validity for Western Middle Iranian. The main burden in distinguishing different alignments is carried by agreement morphology, and by the use of pronominal clitics (see Section 3.6). Verbal agreement in Past Transitive Constructions was generally, if not consistently, with an O-past. Sundermann (1989b: 129) provides the following Parthian examples of transitive clauses in the present and past tenses respectively (glosses added): (78) P RESENT T RANSITIVE : hawı̄n abgundām DEM :PL uncover:PRES:1 S ‘(I) uncover them’ (79) PAST T RANSITIVE : man abgust (a)hēnd 1 S uncover:PTCPL COP:3 PL ‘I uncovered them’ In (78), the verb agrees with the (contextually recoverable) A, while in (79), the predicate consists of the old participle plus an auxiliary, the latter showing agreement with the O. However, with a third person singular O, there is generally no overt copula, hence no overt agreement (cf. (81) below). Verbal agreement in the Past Transitive Construction, although far from consistent as we shall see, nevertheless differs significantly from present tense agreement patterns. The strong tendency towards agreement with the O is definitely suggestive of ergativity in the past tenses, or at least of non-accusativity.2 2 In addition to the verb forms based on the old participle plus copula auxiliary, two other types of predicate appear in Western Middle Iranian: a non-finite form in −išn, and com- Past Transitive Constructions 93 Let us turn now to the differences between the Middle Iranian Past Transitive Construction and what is presumably its predecessor in Old Persian, the manā kartam construction. The Old Persian example in (80) is familiar from the discussion in Chapter 2; I have reduced it here for ease of comparison with the Middle Persian construction in (81): (80) ima tya manā kartam that which 1 S:GEN/DAT do:PTCPL ‘This (is) that (which) was done by me’ (Kent 1953: DB I,28–29) (81) dēn ı̄g man wizı̄d religion which 1 S choose:PTCPL ‘the religion which I chose’ [. . .]’ (Boyce 1975: a,1) In both examples we find superficially the same features: the verb is etymologically the old participle, and agrees with the O rather than the A (in Old Persian, also still in gender). The A appears in Old Persian in the Gen./Dat. case, the reflex of which is still found in the pronoun in the Middle Persian example. The word order in this example is likewise preserved, nor does there appear to be any major change in word order between Old and Middle Persian. On the whole, one must conclude that the two constructions are extremely similar. Viewed from this perspective, it seems rather odd that the Old Persian construction has often been analyzed as a passive (see discussion in the preceding chapter), while the Middle Persian equivalent is regularly referred to as an ergative.3 There are two good reasons for abandoning a passive analysis of the Middle Persian construction, though neither has very much to do with syntax. The 3 binations of participle with an auxiliary (?) ēstēd. In both cases, alignment is basically ergative. However, a full discussion of these matters is beyond the scope of this chapter. I am grateful to Shahar Shirtz for drawing my attention to these forms. In fact, the terminology used for Middle Iranian Past Transitive Constructions such as (79) is anything but consistent. As Sundermann (1989c: 109) notes, different authors refer to them variously as “passive, possessive, ergative, or agential”, often without any discernible difference (my translation). To this list we can add “impersonal”, used for example by Brunner (1977) and Boyce (1975). The terminological confusion is partly merely a reflection of differing descriptive traditions, hence of no particular significance. In part, however, it reflects a justifiable reluctance on the part of the scholars concerned to commit themselves to either an ergative, or a passive interpretation. 94 Western Middle Iranian first concerns the verbal system as a whole: The participle used in the Old Persian construction was an optional choice; it was still also possible to express two-participant clauses in the past tenses using a fully-finite verb form, for example the Imperfect, as shown in example (30). In Middle Persian, on the other hand, such an opposition was simply not available in past tenses. The result was that constructions such as (81) were the only available means of expressing past transitive propositions. From a markedness perspective, then, it becomes hardly plausible to maintain the passive analysis. Furthermore, whereas the Old Persian manā kartam construction is a fairly marginal one, attested uncontroversially with just one verb, the corresponding construction in Western Middle Iranian is quite simply the regular form for past tense narratives, hence attested with numerous transitive verbs. The second reason concerns the case system. In the manā kartam construction, the A is in the Genitive/Dative case, in Old Persian still contrasting with a Nominative, an Accusative, an Ablative etc. Thus the case used for the A is clearly oblique, and could be analyzed as a kind of by-phrase, as scholars such as Cardona (1970) proceeded to do. In Middle Persian, on the other hand, these case oppositions had largely disappeared. Thus to claim that the construction in (81) is ‘passive’ would entail the claim that the ‘Agent-phrase’ is in a case-form basically identical to the case used for subjects of intransitive verbs, or indeed for subjects of any other finite verb forms. This appears to be stretching the by-phrase-analysis too far, and is doubtless a major reason for abandoning it with regard to Middle Persian. These then are the two main reasons behind the differing analyses of the Old and Middle Persian constructions. But note that neither are directly related to the syntax of the construction itself. And indeed, it has never been satisfactorily demonstrated that there are significant differences in this respect. The simplest option appears to me, then, to assume that syntactically they have not changed (although I will have cause to qualify this statement somewhat in Section 3.5). The kind of continuity in syntax evident between the manā kartam construction of Old Persian and the Past Transitive Construction of Middle Iranian is also evident in the use of pronominal clitics to express the A. As mentioned in Section 2.4.3, in Old Persian the A of a manā kartam construction could be expressed through a pronominal clitic, as in (82), repeated from the previous chapter: The case system (82) 95 avaTā=šām hamaranam kartam thus=3 PL battle do:PTCPL ‘thus by them battle was done’ (Kent 1953: DB III,18–19) Precisely this usage continues through Middle Iranian, as in the following example from late Middle Persian: (83) u=š ēn=iz guft and=3 S this=too speak:PTCPL ‘and he said this too’ (Williams 1990a: 47.5) Throughout Middle Iranian, the use of pronominal clitics is by far the most frequently employed strategy for expressing an A-past. The syntax of the clitic pronouns will be dealt with in more detail in subsequent sections; here I would simply stress that the basic syntax of the structure replicates that of the Old Persian one very closely. To recapitulate, in Western Middle Iranian, Past Transitive Constructions differed systematically in their alignment from the rest of the grammar. Syntactically, the structures found are clearly the continuation of the Old Persian manā kartam constructions. They differ from the latter, however, in that whereas in Old Persian, the construction was a marginally used alternative, by Middle Iranian they had become the only available means of expressing transitive clauses in the past tenses. Furthermore, the case form of the A was, in Middle Iranian, ‘less’ of an Oblique than the Genitive/Dative of Old Persian, simply because in much of the Middle Iranian grammar (see below for details) there simply were no morphological case distinctions, a single form of the noun doing service in S, A and O functions. The differences between the Old Iranian manā kartam construction and the Past Transitive Constructions of Middle Iranian are thus primarily rooted in differences in the organization of the grammar as a system of paradigms and oppositions; they do not necessarily imply changes in the syntactic organization of the structures concerned, and indeed, such changes have yet to be convincingly demonstrated. 3.3 The case system Already in late Old Iranian, the rich system of nominal case distinctions was undergoing massive syncretism and by Middle Iranian, only isolated rem- 96 Western Middle Iranian nants of a single case opposition remained: an unmarked case, the Direct (the continuation of the Old Iranian Nominative), contrasted in some environments with a marked case, the Oblique (most probably a reflex of the Old Iranian Genitive). The Direct/Oblique case distinction is absolutely fundamental for understanding the character of alignment splits in Iranian, and in this section the historical and functional features of the system will be laid out. The old Direct/Oblique distinction is still evident in modern western Iranian languages, though in some even this has been lost entirely. However, many of the languages have also recruited additional case markers from various sources (often grammaticalized adpositions), which are used, for example, to create a new ‘Genitive’ case, or a new ‘Accusative’ case (see Bossong (1985) for discussion, and Chapter four). Thus from a synchronic perspective, many languages, for example the much-discussed Balochi (Jahani 2003; Korn forthc. a), appear to have richer case systems than the two-way opposition outlined above. But in the vast majority of languages investigated, a distinction can be made between the reflexes of the old Oblique/Direct distinction, and a later overlay of innovated case markers. The difference is not merely one of etymology; the two types of case marking often differ significantly in their distribution and function. The development of innovated cases, and their interaction with the inherited system, will be taken up in the next chapter. For the time being it is only necessary to establish a terminological distinction between what I will term the Oblique case, an inherited reflex of the Old Iranian Genitive case, and the various types of innovated case markers. In the discussion in this chapter, I am solely concerned with the inherited case system. Wherever a western Iranian language retains the Direct/Oblique opposition, it can be shown that the two cases show striking parallels in their distribution. This characteristic distribution is referred to by Skjærvø (1983: 47) as “our ‘intuitive’ expectations based on the general knowledge of Iranian languages”. For those readers who may not have acquired an ‘intuitive’ guide to Iranian case systems, it is worth spelling out the crucial aspects here, as they are repeated in one language after another throughout Iranian. However, I will be restricting my treatment to the distribution of cases across the core argument functions of S, A and O; the various uses of the cases with complements of adpositions, or as Possessors, or adverbial usage, will not be The case system 97 Table 1. The core functions of inherited cases in Early Western Middle Iranian P RESENT S Dir A Dir PAST O Obl S Dir A Obl O Dir discussed here as they lie outside the discussion of alignment in the narrow sense. It will, however, be necessary to distinguish between S, A and O in the present, and in the past tenses, because in Middle Iranian and beyond, past tenses often have a different alignment to present tenses. The basic distribution of case over syntactic function is provided in Table 1. The crucial point about Table 1 is of course the difference between the past and the present tenses regarding the case marking of A and O. The coding of S, however, is uniformly Direct throughout the system, and this feature holds good throughout all of Iranian, past and present (with the notable exception of Eastern Iranian Wakhi, which has developed differential subject marking of intransitive verbs, see Bashir 1986). Along with the uniform treatment of S, a second generally stable feature of the system is the Oblique marking of the A-past. In other words, wherever an inherited Oblique is maintained in the system, it is generally (though not without exceptions) deployed in the A-past function. Let us return now to the actual case system of Western Middle Iranian. It has been shown that for some of the earliest attested texts, the Direct/Oblique case distinction was preserved in only certain parts of the grammar (SimsWilliams 1981; Skjærvø 1983). Specifically, two distinct case forms are attested for the following: First person singular pronoun; Possibly for the second person singular; A small set of kinship terms (‘father, brother, mother, son, daughter’); The plural forms of nouns and adjectives. Outside of these areas, uncontroversial evidence for morphological case in Western Middle Iranian is not available.4 The small set of kinship terms had 4 The existence of a distinct Oblique form for nouns was argued by Nyberg (1974: 277– 278), who reconstructs an Oblique singular suffix −*ē. This view was sharply criticized 98 Western Middle Iranian a unique singular Oblique ending, found only with these words. I will not pursue the details here, but merely note that a unique singular Oblique form (−r, or variants thereof) for certain kinship terms is still found in some western Iranian languages, for example the Tati dialects investigated in Yar-Shater (1969: 86–95). Having set out the the key aspects of the functions of the cases, I will now turn to look at two important aspects of the system: the interaction of case and number, and of case and person. 3.3.1 Case and number A peculiarity of Early Western Middle Iranian, still preserved in some modern Iranian languages (e.g., Turkmen Balochi, or the Northern Group of Kurdish) concerns the interaction of case and number. Setting aside the kinship terms for a moment, the system with nouns and third person pronouns functioned as follows: nouns and adjectives in the singular have no case distinctions, hence remain unmarked regardless of syntactic function. In the plural, when used in functions associated with the Direct case (see Table 1), they were also formally unmarked. Thus in the Direct case, nouns showed no overt distinction between Singular and Plural. Nouns/Adjectives in plural and in Oblique functions (see Table 1) on the other hand were marked with a suffix, transliterated with −’n, and traditionally transcribed −ān.5 This system is summed up in Table 2. The typological oddity preserved in this system should be immediately apparent: as a rule, we expect at least as many inflectional distinctions to be available to the unmarked member of an opposition as to the marked one 5 by Back (1978: 35–37). Today, scholars are divided in their views on this. Prods Oktor Skjærvø, Professor of Iranian languages at Harvard, considers the evidence from texts insufficient to postulate such a form (p.c., 15.05.2006). Other scholars, e.g., Korn (forthc. a), assume the existence of an Oblique singular case marker. While the textual evidence may not support the existence of such a form in Middle Iranian, many of the later languages do in fact preserve an Oblique ending on singular nouns, so it can, therefore, be reconstructed. A reflex of the Proto-Iranian Genitive Plural *-ānām, cf. the Old Persian Gen./Dat. Plural of the −ā- and −o-stems. The case system 99 Table 2. Case and number on nouns in Western Middle Iranian D IRECT O BLIQUE S INGULAR P LURAL unmarked unmarked unmarked marked (-ān) (for example, at least as many gender distinctions in singular numbers as in non-singular, cf. Greenberg’s Universal #37). With respect to the distribution of number distinctions across cases, “we expect to find more number values for subject than for object, more for direct cases than for oblique.” (Corbett 2000: 274) In the Middle Iranian system summed up in Table 2, however, nouns in Oblique functions apparently express more number distinctions than nouns in direct functions. Conversely, we could expect that in the singular number, at least as many case oppositions would be available as in the plural, and this appears to be generally the case cross-linguistically.6 But again, the Western Middle Iranian system runs counter to this prediction, because a case distinction is made in the plural, but not the singular. It is uncertain how diachronically stable the system outlined above actually was. Already in Middle Iranian, for example in the Pahlavi Psalter texts investigated by Skjærvø (1983: 177), overtly-plural marked nouns occurred in functions associated with the Direct case. Likewise, Nyberg (1974: 278) notes that while the Direct/Oblique case distinction was “still well distinguished” in the plural, already in “good classical texts there occur instances of −ān” in Direct functions. Thus already in the Middle Iranian stage, the Plural Oblique marker was extended to usage as a simple marker of plural number, used in all syntactic functions. Consider the following examples from a Manichaean Middle Persian text (Mani’s šābuhragān), in the edition of MacKenzie 1999c), contrasting the noun yzd ‘God’ in the singular and 6 However, Dahl and Koptjevskaja-Tamm (2006) note counter-examples for some Swedish dialects, where more case distinctions are made in the plural than in the singular. There are suggestive parallels to the Iranian data here: the authors suggest that one reason behind the Swedish situation is the existence of a relatively uniform, phonetically ‘heavy’ Dative Plural, which tends to be maintained when other case morphology is lost. Much the same applies to the Iranian Plural Oblique. 100 Western Middle Iranian Table 3. Case and Number in Western Middle Iranian Stage 1, reconstructible for earliest WMI S ING . PL. D IR O BL ø -ān ø *-ē Stage 2, attested Early WMI, cf. Table 2 S ING . PL. D IR O BL -ø -ān -ø -ø Stage 3, Late W. Middle Iranian S ING . PL. D IR O BL -ān -ān -ø -ø plural, both in direct functions (segments in brackets indicate that they are restorations, line numbers refer to MacKenzie’s edition): (84) (’)wd (’y)myš’n yzd-’n ky and these god-PL which ‘and these gods, which [. . .]’ (are lord of the house etc.) (l. 30) (85) tw yzd ẅ ’nwšg hy you god and immortal be:PRES:2 S ‘you are god and immortal’ (l. 88) In later Middle Iranian texts, as in for example the texts of Williams (1990a), the Plural Oblique ending −ān is used consistently as a general marker of plurality, thereby ironing out one of the typological oddities in the system illustrated in Table 2. These developments are reproduced schematically in the form of three stages in Table 3, where the items in bold type show the respective innovations. The intriguing point about the historical developments is the following: the system changed in a manner which actually reduced the degree to which Case and person 101 Direct and Oblique cases are distinguished. At Stage 1, Direct and Oblique are consistently distinguished throughout, while at Stage 3, they are no longer distinguished at all, and this system is still found in a number of modern Iranian languages. What has increased, on the other hand, is the systematic marking of the singular/plural distinction. Originally, plurality was only systematically indicated in the Oblique case, but this changed very early so that the opposition singular/plural became regularly expressed, regardless of case. It is worth noting that every Iranian language I am aware of marks plural number via bound morphology (as does perhaps every Indo-European language?), despite the otherwise impoverished nominal inflection in some of the languages. The general principle at work here appears to be: when there is only little inflectional morphology available, whatever is left will be deployed for the singular/plural distinction – even if this entails sacrificing a case distinction. Of course many languages have compensated by recruiting adpositions for marking the Direct/Oblique distinction. But we know from the history of Persian that the use of the postposition −rā to consistently mark (definite) direct objects did not emerge until centuries after the loss of the old inherited Oblique marker. This type of ‘delayed functional change’ will be a recurrent theme throughout the book. 3.4 Case and person As mentioned, Oblique case marking in the singular is only attested for parts of the nominal lexicon, and only in early Middle Iranian texts. In the first and second person pronouns (Speech Act Participant, SAP), a distinct Oblique form was found for the first person singular (in Parthian Direct az and Oblique man). For the second person singular, such a distinction has been suggested, but it remains controversial (Sundermann 1989b: 131). An overview of the personal pronouns Parthian is given in Table 4, based on Sundermann (1989b), with missing forms supplied from the glosses in Durkin-Meisterernst (2006). The ‘third person’ pronouns are demonstratives, but were also used in pronominal functions. In later Middle Iranian texts, the old Direct/Oblique distinction on the first and second person singular is often absent, leaving but a single form for all functions (for example, Early New Persian). But while the loss of the 102 Western Middle Iranian Table 4. Case on personal pronouns in Parthian S INGULAR 1 P. 2 P. 3 P. P LURAL D IRECT O BLIQUE az tū(?) man tō hō D IRECT O BLIQUE amāh ašmāh hawı̄n case distinction apparently parallels the loss of the case distinction already discussed for nouns, the development with SAP-pronouns is nevertheless different: with the pronouns, it is the old Direct form that disappears, while the Oblique form survives as the sole pronoun (for example, first singular man). With singular nouns, on the other hand, it is the Oblique form that disappears, leaving the Direct as the single form. Furthermore, the Direct/Oblique distinction itself definitely survived longer on the first person singular than elsewhere (see next chapter). These facts bring home an important principle underlying the grammar of case in general, and in particular, in their histories: the existence of parallel case systems within one and the same language. Such a situation is so commonplace that we tend to overlook its significance, referring to languages as ‘having’ a particular number of cases. But the basis for claiming that a language has an x-term case system is seldom made explicit. Consider modern standard German. It is generally agreed that German has a four-case system: Nominative, Accusative, Dative, Genitive. Yet only a single class of words in German regularly distinguish all four cases: masculine determiners in singular number (definite articles, indefinite articles, demonstratives etc.). Personal pronouns of the first and second person singular (ich, du) arguably also do so, but the status of the genitive forms of these words is dubious. Outside of these two groups, German case marking observes maximally three, in many environments just two, and on the vast majority of feminine nouns in singular, no overt case distinctions whatsoever. English is another case in point: formally, only (some) personal pronouns show any case distinctions (maximally two forms), and for most grammarians, this would not be considered sufficient to say that English has a two-term case system (the status of the Genitive-’s also remains controversial). But the difference between German and English is really one of degree rather than Case and person 103 kind, and there does not seem to be an established set of criteria for deciding on how many cases a particular language ‘has’. The rough rule of thumb adopted by most grammarians is to take the maximum number of case distinctions drawn in any particular domain of the grammar (for example an inflectional class), and to assume it represents the case system of the language as a whole. If, however, a domain is very small, as in the case of English pronouns, it is ignored, whereby what counts as ‘very small’ is left open to interpretation. While this rule of thumb may be harmless enough as shorthand for language description, I believe it is a grave mistake to assume that because a particular number of case distinctions is systematically reflected in one part of the grammar, we must necessarily assume that the same case distinctions are also in any sense relevant in those parts of the grammar which show no formal expression of these distinctions. Dahl and Koptjevskaja-Tamm (2006: 59) refer to the same point. They prefer to analyze a case system in terms of a “set of paradigms”, rather than characterizing a language in terms of single monolithic case system, and to analyze the interactions between the various sub-paradigms. This is also the procedure adopted here. The most glaring manifestations of such disjunctions in case systems of individual languages are found in languages with so-called ‘animacy’ or ‘person’-based split alignments. Probably the most widely discussed example of a language of this type is Dyirbal (Dixon 1972), which is taken up in connection with animacy-based splits in 4.6.1 below. The necessity to distinguish distinct sub-systems within the case system is never more pressing than in connection with diachronic developments. It can be shown time and time again that the case systems found in different domains of the grammar develop — at least to some extent – independently of one another. Case distinctions may disappear on nouns, but be retained on some pronouns, they may disappear in plural number, but continue in singular number, and so on. Differences between personal pronouns of the first and second person, and other pronouns and nouns are ubiquitous crosslinguistically. In discussing change, then, it is not sufficient to refer to ‘the case system’. Rather, the systems in the individual domains found to be distinct in a particular language or genetically related group of languages must be investigated. Of course the distinct subsystems are not fully independent; rather there appears to be a kind of loose interdependence which constrains 104 Western Middle Iranian the possible range of changes, stopping complete divergence. We find various cross-overs, analogy, and levelling out of differences between the various subsystems, in some cases leading to full parallelism. For example, in New Persian, a single Accusative case marker, −rā (with various phonetic variants), applies across the board to personal pronouns, and singular and plural nouns. But elsewhere in Iranian, the first and second person, and the plural number, repeatedly show case marking patterns which systematically diverge from each other. These simple facts, readily observable in countless languages, make it probably more realistic to examine the history of case systems in terms of a loosely-connected web of parallel sub-systems, rather than a single monolithic set of oppositions. 3.4.1 Summary of case in Western Middle Iranian In Western Middle Iranian we find maximally two morphological cases, the Direct and the Oblique. In singular nouns, in plural nouns, and in the first person singular, the old Oblique marker developed in three quite distinct ways: the old Oblique marker for singular nouns simply disappeared completely, leaving nouns in an unmarked form in all core functions. The old plural Oblique on the other hand was actually extended to a general plural marker, regardless of syntactic function. Finally, the old Oblique of the first person singular became the general form of the pronoun, while the original direct form disappeared without trace.7 Thus while the net result of change in the case system is undoubtedly towards reducing the number of inherited case distinctions, the different sub-systems of the nominal lexicon do not all work in synchronization with each other, and indeed loss affects the different sub-systems in different ways. It is these differences in trajectory of change 7 While not all the daughter languages exhibit all three tendencies, it is undeniably true that wherever changes in these three sub-systems (singular nouns, plural nouns, first person singular) are found, they generally follow the patterns outlined. For example, several languages have actually retained both inherited Direct and Oblique forms of the first person singular. But crucially, no language is attested which has retained the old Direct, but lost the old Oblique completely (though in some cases the Oblique is augmented by additional Innovated case markers; see the Tatic-type languages in Chapter four). Pronominal clitics 105 Table 5. Possible functions of pronominal clitics in different tenses T ENSE : A O I ND . PARTIC . A DP.C OMPL . A DN .P OSS . P RESENT PAST − + + − + + + + + + across individual subsystems that are primarily responsible for the apparent complexity evident in many western Iranian case systems. 3.5 Pronominal clitics In Middle Iranian syntax, the pronominal clitics are omnipresent. In fact, one might suggest that the simplification of the case system between Old and Middle Iranian was, to some extent, compensated for by the massive increase in the use of clitics. The clitic-system that crystallized for Middle Iranian has also remained a striking feature of the grammars of numerous modern West Iranian languages. Despite many minor differences, the clitic systems of all languages that have them share a number of striking similarities, and it is appropriate to treat them in their entirety at one point in the book. Pronominal clitics in Western Middle Iranian were used as a means of cross referencing constituents in five different syntactic functions: (1) an Apast; (2) an O-present; (3) an Indirect Participant; (4) an adpositional complement; (5) an adnominal Possessor. The way these functions are distributed across different tenses is summed up in Table 5. The important point to note is that the core functions of cross-referencing an A or an O are in complementary distribution: A with the past, O with the present. In this sense, then, Western Middle Iranian exhibits Tense-Sensitive Alignment, because the coding of core arguments differs according to the tense of the verb. The other functions of the pronominal clitics, however, operate independently of the tense of the verb. Thus it is quite possible to have two or even more clitics in the same clause, but crucially, not crossreferencing both an A and an O (see below for a possible counter example and discussion). The system of Tense Sensitive alignment outlined in Table 5 is 106 Western Middle Iranian Table 6. Pronominal clitics in Western Middle Iranian S INGULAR P LURAL 1P 2P 3P 1P 2P 3P =m =t =š =mān =tān =šān absolutely characteristic of West Iranian; in Chapter six essentially the same system re-surfaces in the grammar of the Central Group of Kurdish. Turning now to the the inventory of forms, there was a drastic reduction from Old to Middle Iranian. In Old Iranian a distinction between Accusative, Genitive/Dative, and Ablative forms was still preserved at least in parts (cf. Table 2 in Chapter two), but the clitic pronouns of Western Middle Iranian on the other hand had a single form for all Oblique functions. The forms of the Middle Iranian clitic pronouns are fairly consistent across the attested texts and are provided in Table 6, based on Heston (1976: 142) and Sims-Williams (1981: 171). At this point one can reasonably ask what relationship existed between the clitic pronouns and the full pronouns? Their syntactic distribution was no longer identical, so it really cannot be said that the clitic pronouns are merely pragmatically determined optional variants of the full pronouns, as they were in Old Iranian. In fact, the clitics were already on their way to becoming something else: agreement forms for cross-referencing person features. Already in late Middle Persian, Brunner (1977: 104) notes the occurrence of clitic pronouns used “redundantly” to cross-reference an A-past, by which is meant that the clause also contains a full form of the pronoun (see also pp. 106 and 107 of the same book). This “redundant” usage has became the norm in many West Iranian languages. Thus we find in Gorani: (86) ämín kâghäz=îm kîâst 1 S letter=1 S send:PST:3 S ‘I sent (a) letter’ (Hadank and Mann 1930: 296) Here the use of the clitic pronoun cross-referencing the past A (‘I’) is obligatory; it is required even when an overt subject pronoun is present (in the above example ämín). The use of pronominal clitic as some form of agreement is well-attested in several Iranian languages (see Chapter six). There Pronominal clitics 107 is little doubt that these nascent agreement markers originated in the clitic pronouns of Old and Middle Iranian, but the exact chronology of the development remains obscure, and must have proceeded at different rates in different languages. Furthermore, it only affects the clitics in A-past function (see below). Elsewhere, clitics remain in complementary distribution with full pronouns. Finally, as far as clitic placement is concerned, Western Middle Iranian largely continues the tendency documented for Older Iranian: clitics attach to the first element of the clause, in the Middle Persian texts examined by Heston (1976: 90) most commonly a conjunction such as AP ‘and’. However, the rules were already in Middle Iranian not quite identical to those of Old Iranian, and in later languages, placement of pronominal clitics succumbed to the family-wide pressures of rightward drift, and head-attraction. These issues are treated separately in the Appendix A.2, as they do not directly impinge on the issues at hand. 3.5.1 The Indirect Participant role In Section 2.5, I suggested that the most insightful means of capturing the functions of the Old Persian Genitive/Dative case was in terms of a broad semantic role, characterized there as the Indirect Participant. The relevance of this concept and its centrality for understanding the history of Iranian is reinforced by the fact that the same bundling of semantic functions characterizes the usage of the pronominal clitics in Middle Iranian, and indeed throughout Iranian. The specific claim that I am making is that the role of Indirect Participant is the primary function of the clitics, rather than a secondary extension, as is often implied in traditional descriptions. The primary motivation for this claim is the fact that wherever clitic pronouns are found in Iranian languages, they are always used to express Indirect Participants, whereas their other functions (A in the past tense, adnominal possessor) are not always attested. Even in modern Persian, where there is no longer TSA, clitics expressing Indirect Participants can still be seen in idiomatic expressions such as the following: 108 Western Middle Iranian (87) xoš-am mi-yā-yad pleasure-1 S PROG-come:PRES-3 S ‘(I) like (it)’ The first person singular suffix −am, a reflex of the old clitic pronoun, expresses something like an Experiencer (‘liker’). Syntactically, it looks like a Possessor: ‘my pleasure comes’. However, it is not possible to substitute the ending for a full possessive pronoun (*xoš-e man). This kind of usage is found in a number of more or less fixed expressions in modern Persian, for example: man yād=am raft ‘I forgot’ (I, memory=for.me went) man sard=am šod ‘I got cold’ (I, cold=for.me became) man xāb=am gereft ‘I felt sleepy’ (I, sleep=for.me took) In fact, the interpretation of these clitic pronouns is highly controversial in modern Persian linguistics (see Sedighi (2005) for recent discussion), but I believe the issue is less complicated than it appears. From a comparative Iranian perspective, the use of clitics in Modern persian expressions such as those above are quite predictable, as they represent the continuation of the original Indirect Participant function. The problems arise when one attempts to analyze them in terms of the standard catalogue of semantic roles, or in terms of the argument/adjunct distinction. Examples of Indirect Participant clitics from both Parthian and Middle Persian are numerous, though different scholars use different terminology to describe them. For example, Brunner (1977) and Heston (1976) refer to “Indirect object”, “indirect possessor”, “object concerned by the action” (Brunner 1977: 101), or “oblique function” to describe examples I would classify as Indirect Participants. A typical example is the following: (88) ā=šān ān abāyēd ka=šān gyān az tan be then=3 PL this is.necessary that/when=3 PL soul from body out šawēd go:PRES:3 S ‘then this is necessary for them that/when their souls go from their bodies.’ (Williams 1990a: 13b.3) The Indirect Participant clitic pronouns of Middle Iranian are often difficult to interpret. Being special clitics, they will gravitate to a particular position Pronominal clitics 109 in the clause, abandoning the word with which they are most closely associated, and the precise reading of the clitic becomes difficult to recover. As a brief illustration, consider the following examples, given with the translations provided in the sources (glosses added): (89) tw=m’n ’yy xwd’y you=OBL:1 PL COP:2 S lord ‘You are our Lord’ (Parthian, Brunner (1977: 102), original translation) In this example, the clitic pronoun is translated simply as a Possessor (‘our Lord’) although it is not phonologically bound to its possessed. In the next example, the clitic pronoun attaches to what is arguably its possessed (‘his bright day’): (90) rwz spyd=š šb tyrh šd day bright=OBL:3 S night dark became:3 S ‘The bright day became dark night for him’ (Early New Persian, Heston (1976: 88), original translation) The translator here nevertheless opts for a Benefactive/Malefactive: ‘for him’. The following examples are also ambiguous with regard to the Benefactive/Possessor distinction: (91) čē=m zišt pad čašm hē and=1 S hateful to eye be:PRES:2 S ‘and you are hateful to my eyes.’ (Williams 1990a: 18f.4) (92) u=š pas be ō dil rasēd and=3 S then to heart come:PRES:3 S ‘and then it comes to his heart.’ (Williams 1990a: 3.7) (93) ō dēn ı̄ man āyend, u=šān xwad dar ı̄ uzēnišn to religion of me come:PRES:PL and=3 PL self door of salvation bawēd become:PRES:3 S ‘(they) come to my religion, and it verily is their / for them the gate of salvation.’ (Boyce 1975: a,l. 3) It will be evident that these problems exactly replicate those discussed in connection with the interpretation of the Old Persian Genitive/Dative case in 110 Western Middle Iranian Chapter two. There can be little doubt that we are dealing with a continuation of the same broad Indirect Participant category. As with the clause-level Genitive/Datives of Old Iranian, the Indirect Participation reading may overlap with an Agent-reading (at least in connection with past tense verbs). Consider the following, where both Agent, or Possessor readings are possible: (94) ’w=š dst ’hnwnc ny šwst and=3 S hand yet not wash:PTCPL ‘and he had not yet washed (his) hands.’ (Boyce 1975: n1) (95) u=š xwah be ō zan ı̄ dēw dad and=3 S sister to woman of demon give:PTCPL ‘and he gave his sister as wife to the demon.’ (Williams 1990a: 8e.8) In the latter examples, the clitic pronoun could express the possessor of ‘hand’ or ‘sister’, or the Agent of ‘wash’ or ‘give’, or a fusion of both.8 In some cases, only the context can disambiguate. Consider the following example: (96) u=m pānzdah asp kušt ud xward and=1 S 15 horse kill:PTCPL and eat:PTCPL ‘and (he) killed and ate my 15 horses.’ (Williams (1990a: 18f.11), translation from Williams) Outside of this context, however, (96) could undoubtedly mean ‘and I killed 15 horses . . . ’, with the clitic pronoun interpreted as Agent, rather than Possessor (and indeed, the latter interpretation has been suggested in the literature, though Williams argues against it in the notes accompanying this passage, cf. Williams 1990b: 166). The second clause of the next example is also grammatically ambiguous: 8 When two clitic pronouns are required semantically, and the two are distinct persons, then it is possible to have a sequence of two clitics. Boyce (1975: 45) notes in this connection that the A-past clitic regularly precedes any other clitic. Judging from the few examples available, the ordering may also be attributed to person (first person preceding second preceding third). Pronominal clitics (97) 111 qyrd, h’n=t’n mn . . . h’n cy ’šmh ’w dynwr’n ¯ that which you:PL to devout.man:PL do:PTCPL, that=2 PL 1 S . . . qyrd do:PTCPL ‘that which you did to the devout men, that you did to me’ (Boyce 1975: z,9) Syntactically, this could also be read as ‘that I did for you’, with the clitic pronoun read as an Indirect Participant. Boyce translates differently, however, with the clitic pronoun as past Agent, for which doubtless contextual evidence is responsible. But out of context, a clear decision is not possible. A final problem with the interpretation of the clitics is that they may also express the complement of an adposition. Here too they are subject to the placement constraints which means that they will often leave the governing adposition and occur in clause-second position. The same phenomenom recurs today in the Central Group of Kurdish, where the pronominal clitics have retained some of their mobility. The following examples from that language are illustrative: (98) bō=yān bār a-ka-yn for=3 PL loading PROG-do:PRES-1 PL ‘We are loading (something) for them’(Kurdish, Central Group, Friend 1985b: 92) Here the clitic pronoun is attached to the preposition bō. However, under certain conditions to be discussed in more detail in Section 6.2.3, the clitic floats to a constituent to the left: (99) dargā=t bō a-ka-n-awa door=2 S for PROG-do:PRES-3 PL- POSTV ‘They are opening the door for you.’(Kurdish, Central Group, Friend 1985b: 91) Semantically, we have a typical Indirect Participant, a Benefactive. However, it is fairly clear that this pronoun is governed by the preposition bō. But due to the language-specific rules of clitic placement, it has been moved to another constituent. Here then we are faced with a similar problem of classification: Indirect Participant, or Complement of Adposition? In many cases, a 112 Western Middle Iranian decision is not possible on purely syntactic grounds, but can only be reached through interpretation of the broader context, and even then, uncertainty often remains. To sum up, the clitic pronouns of Middle Iranian were used extensively in what I have termed the Indirect Participant construction, the direct continuation of the central function of the Old Iranian Genitive/Dative. They continue to be used in this function in all Iranian languages that have preserved such clitics, including Persian. I consider this to be their core function, and their extension to other uses, for example O-pres and A-past, are radial extensions of this category (see next section). 3.6 Clitics expressing core arguments Besides the Indirect Participant function, the pronominal clitics also occurred in two other functions: the direct object of a present tense verb (O-pres) and the subject of a past transitive verb (A-past). The former usage, that of the Opres, evidently continues the usage of the Accusative clitic pronouns in Old Iranian, as shown for example in (32). Clitics cross-referencing an O-past, on the other hand, are practically non-existent, although this was possible by Early New Persian, cf. Heston (1976: 95), and is quite normal in modern Persian. I said above “practically non-existent”, but one example of a clitic apparently cross-referencing an O-past is in fact attested: (100) u=t=išān kird azad and=2 S=3 PL make:PST:3 S free ‘And you set them free [. . .].’ (Boyce 1975: cd4) In this example, it seems as though the clitic =išān actually cross-references the O of the past-tense verb, something that otherwise is not attested. In fact, another interpretation of =išān is available, suggested to me by Shahar Shirtz (p. c.): if we consider the word azad, translated here as ‘free’, to be a noun meaning ‘freedom’, then the clause could be read something like ‘And you made freedom for them’. On this reading, azad would be the O-past, and =išān an Indirect Participant, hence available to be cross-referenced with the clitic pronoun regardless of tense. Whether this interpretation is tenable de- Clitics expressing core arguments 113 pends crucially on what is known about the syntax and semantics of the word azad, which I am currently unaware of. But this line of thought definitely merits investigation, as it would provide an elegant explanation of what is otherwise a totally anomalous construction in the context of Western Middle Iranian. This example would be only worth a footnote were it not for the fact that Bubenik (1994) takes just this example as a cornerstone for his theory of the loss of ergativity in Middle Iranian, which I will briefly comment on here. Bubenik treats (100), as well as examples containing both an A-past clitic and an Indirect Participant clitic, as cases of subject/object ambiguity. The dire consequences of this state of affairs are spelled out as follows: The system expressing both subject and object by the same oblique pronominal form allowed for serious ambiguities. [. . .] There were essentially two solutions offered to this serious homophonic clash in the history of the Iranian languages: (i) the ergative construction was given up, i.e., the oblique pronominal form could not be used to express the function of subject any more. This was the route taken by Farsi (Modern Persian). (ii) The ergative construction was preserved, i.e., the oblique pronominal form continued expressing the subject. The remaining matters were sorted out in the fashion that the direct forms expressed the object. This was the option exploited by Pashto, Kurmanji and the Southern Tati dialects.[. . .] (Bubenik 1994: 113) A number of objections can be raised against these claims. First, the “serious homophonic clash”, at least as far as A and O are concerned, simply did not exist; as discussed above, (100) is totally isolated within Middle Iranian, where generally the tense-based distribution of A and O marking is consistently observed. Second, the Parthian and Middle Persian system of cliticcross referencing has, despite the “serious homophonic clash”, survived down to the present in numerous West Iranian languages, for example Central and Southern Groups of Kurdish, or the languages like Davani (Southwest Iranian, Fars Province, discussed in Dabir-Moghaddam 2007). There are thus many more “solutions” available than the two mentioned in the above quote. Third, the languages listed as “preserving” ergativity provide no compelling evidence for the existence of the proposed second solution: Kurmanji, unlike Parthian or Middle Persian, has no pronominal clitics at all, but does have nominal case. It cannot be traced back to a system like the Middle Persian one, thus it cannot be interpreted as a solution to the “serious homophony”- 114 Western Middle Iranian problem at all – even if such a problem ever existed. Southern Tati, on the other hand, is not consistently ergative at all, as shown in 4.4. Finally, the relevance of East Iranian Pashto to the historical developments within West Iranian is marginal at best. Aside from the obvious empirical drawbacks with the proposed scenario, I have more general reservations with regard to explanations in terms of “ambiguity avoidance”. It is for example certainly true that the pronominal clitics of Middle Iranian are sometimes difficult to interpret – I discuss several such examples below. But just because researchers experience difficulties in interpreting thousand-year-old text fragments expressing arcane religious content, we cannot assume that the speakers of those languages, intimately familiar with the rich cultural context in which the texts were embedded, had problems of the same magnitude. More generally, natural language abounds with ambiguities, when utterances are artificially taken out of context. Chomsky’s famous flying planes can be dangerous has at least three structural sources of ambiguity: the polyfunctional −ing suffix, the lack of number agreement on the modal can, and the fact that the verb fly can be used transitively or intransitively, without change of form. I am unaware of English changing in the direction of eliminating any one of these features. Newmeyer (2005: 155–160) has recently argued at some length against the view that ambiguity avoidance is a particularly powerful factor in shaping grammars, and I entirely concur with his arguments. These issues are taken up in 4.5; for the time being it needs to be stressed that the Middle Iranian system of clitic-marking of direct objects was in general consistently restricted to the present tense. Examples such as (100) are extremely atypical, and should not be taken as support for a theory of ambiguity avoidance as a driving factor in alignment changes with the clitics. Let us return to the core functions of the pronominal clitics. Below, several examples of clitics in the O-present function are provided: (101) 9 u=m kunēd nām ‘Kerdir and=1 S make:PRES:3 S name Kerdir ‘And (he) names me Kerdir [. . .].’9 (MacKenzie 1999b: l. 25) In MacKenzie’s translation, the verb form is translated with an English past tense. It is a feature of this text that formally present-tense verb forms are used as historical present Clitics expressing core arguments 115 (102) u=m xwad dārēd ud pāyēd and=1 S self keep:PRES:3 S and protect:PRES:3 S ‘and (he) himself keeps and protects me’ (Boyce 1975: b1) (103) yuDyēnd baGān pad hō u=š hamēw bōžēnd strive.for:PRES:PL god:PL for him and=3 S always save:PRES:3 PL ‘The Gods strive for him and always save him’ (Durkin-Meisterernst 2006: M200, Hymn 1) (104) čı̄d=mān pāyēd always=1 PL protect:PRES:3 S ‘(It) always protects us’ (Durkin-Meisterernst 2006: M105a) The most characteristic use of the clitic pronouns in Middle Iranian is that of the A-past, exemplified in (83) above, and in many other examples later in this section. At this point, I will simply repeat the main feature of the clitic system, namely the complimentary distribution of core functions across tense: objects in the present tense, and transitive subjects in the past tenses. What we have, then, is the familiar picture of Tense Sensitive Alignment, but it manifests itself here not in nominal case, but in the distribution of the pronominal clitics. In addition to the functions listed above, Middle Persian apparently used clitic pronouns in a function for which, as far as I can ascertain, no parallels exist in any other attested Iranian language, past or present: as subject crossreferencer with present tense verb forms: (105) AMT=Šn SGTWNd when=3 PL go:PRES:3 PL ‘when they go’ (Heston 1976: 96) I mention this example if only for the fact that it runs counter to all other attested developments. But perhaps significantly, despite the existence of a small number of examples of this type in Middle Iranian, such a system has not achieved any stability or frequency, as it is not attested elsewhere. tenses. Despite its intended reference, it remains a present tense verb form, hence the use of the clitic to refer to the object is possible. The clitic could also be read as a Possessor, yielding ‘he made my name K.’, as Shahar Shirtz (p. c.) points out. 116 Western Middle Iranian 3.6.1 Summary of the clitics The developments in the clitics can be seen as paralleling those in the system of morphological case to some extent: the original range of 4–6 nonnominative cases available in Old Iranian had melded to a single Oblique case by Middle Iranian, which continued to fulfill the functions of the old non-nominative cases. Likewise, of the different case forms of the clitics, only one survived, which again covered all the old functions. But the parallels in the development of the nominal case and the clitic inventory stopped at this point. For whereas in many of the languages, the Direct/Oblique case distinction on nouns was completely lost, the set of clitics remained remarkably stable and continues to be used across most of the western Iranian languages, regardless of whether they have case systems or not. Perhaps more significantly, the clitic pronouns in at least some of the modern languages are no longer merely pragmatically-driven variants of a full pronoun. In the A-past function, their usage is actually obligatory, regardless of the presence or absence of the corresponding full pronoun. Thus what began as pragmatically-determined variant forms of the personal pronouns became, over the course of time, detached from the rest of the nominal system and began to take on characteristics of an agreement system. The changes in the rules regarding clitic placement are discussed in more detail in Appendix A.2. Broadly speaking, while clitic placement in Old Iranian was largely determined by prosodic rules, which left them in the Wackernagel-position regardless of syntactic function, from Middle Iranian there was a gradual shift towards more syntactically-oriented principles, whereby clitics gravitated to their governing heads. For example, clitics expressing adpositional objects stay on the adposition, and clitics expressing an O-pres stay on the verb. This meant that, for example, pronominal clitics on clause-initial complementizers are extremely rare in modern Iranian.10 For clitics express- 10 The Southwest Iranian language Davani (Fars province, south Iran) is one of the rare exceptions to the above generalization, as pointed out by Dabir-Moghaddam (2007), cf. vo=š ga ‘and=he said’, which perfectly replicates the Middle Iranian construction discussed above. Some other Iranian languages do have clause-second clitics, but crucially, they are not pronominal clitics. Rather, they are usually some form of TAM-marker, with a historically quite different source from the pronominal clitics under discussion Past transitive verbs 117 ing the A-past, or Indirect Participants, the question immediately arises as to what the governing head actually is. They are, as I have repeatedly argued, linked to the clause as a whole, though the nature of their relationship is difficult to reduce to one of government by any particular constituent. Hence we would expect the rules concerning their placement to be considerably more complex than in the more straightforward cases of a clitic expressing, for example, the complement of a preposition, and indeed this is the case. In Middle Iranian, however, clitic placement was still largely a matter of clausal prosody, hence the differences between A-past and Indirect Participant clitics on the one hand, and the other types of clitic on the other, was not so immediately apparent, all being placed in the clause-second position. 3.7 Past transitive verbs It will be recalled that the Middle Iranian period was long, and the past transitive constructions attested in the various languages showed a considerable amount of variation. A simple classification of alignment in Middle Iranian, along the lines of ‘ergative’, or ‘accusative’, simply fails to account for the indeterminate nature of much of the data. In this section I will discuss two crucial factors behind the variation found, and attempt to draw some more general conclusions. The first is what I term lingering intransitivity, the second is variation in the agreement pattern on the verb, that is, whether the verb agrees with an O, an A, or with nothing. 3.7.1 Lingering intransitivity The so-called past stems of Middle Iranian verbs go back to the Old Iranian resultative participles. And, not unexpectedly, the traces of their participial origins accompanied them in their new functions as past tense verb forms. here. For example, the East Iranian language Rošani (Pamir Group) has special clitics in clause-second position (Payne 1980: 159), and the Future clitic of Northern Kurdish can likewise be analyzed as clause-second. 118 Western Middle Iranian Basically, two features are characteristic of the Indo-European resultative participles of transitive verbs, and indeed of many languages: they refer to the state of the patient resulting from the action (patient orientation), and they are incapable of assigning accusative case (intransitivity). The latter feature is that which renders them fundamentally different to finite (active) forms of transitive verbs, and which is ultimately responsible for the existence of TSA: a canonical transitive construction, with an Accusative-marked O, is simply not possible with a participial verb form, hence other means are required. The participial origins of the verb forms are still very much apparent in Middle Iranian, and indeed in many of the later languages. For example, the Parthian ‘infinitive’, also based on this stem, had “passive function”, as Sundermann (1989b: 127) notes. In Kurdish of the Northern Group, an infinitive, likewise based on the past stem, is also used with a passive meaning in the so-called periphrastic passive. Consider for example hat kuštin, lit. came kill:INF, ‘s/he was killed’. This can be quite transparently understood if we assume that the so-called infinitive is not equivalent to English ‘to kill’, but is a verbal noun with a meaning something like ‘being killed’. The construction is thus literally ‘come to being killed’, and has since grammaticalized to the regular passive expression in modern Kurdish of the Northern Group. But the participial, hence intransitive, origins of the past tense verb forms are even more clearly evident in Middle Iranian. For such verb forms are attested not only in what I have termed Past Transitive Constructions, but also without an overt, or even an interpretable, A. In such cases, an ‘agentless passive’ interpretation becomes compelling. The following two examples illustrate one and the same verb form (albeit in different transcriptions), the first used in a Past Transitive Construction with overt A, the second used to express an agentless passive: (106) dēn ı̄g man wizı̄d religion which 1 S choose:PTCPL ‘the religion which I chose’ [. . .]’ (Boyce 1975: a,1) (107) prhyd wcydg’n ’wd nysš’g’n wcyd much elect:PL and hearer:PL choose:PTCPL ‘many elect and hearers were chosen’ (Boyce 1975: h,1) Other examples of simple participles used to express what appears to be agentless passives are the following (where a stative reading is perhaps more appropriate): Past transitive verbs 119 (108) u git u pātixšahr u mātiyān čē ōi zamān and testament and contract and document:PL which at.that time apar varhrān šāhān šāh kart under Varhrān of.kings king make:PTCPL ‘and testaments and contracts and documents which at that time under Varhrān, King of Kings, were made’ (Utas 1974: 103) (109) ’sp ’(wd) mrd wnybwt ¯ horse and man destroy:PTCPL ‘horse and man are destroyed [. . .]’ (Sundermann 1973: l. 1645) An agentless-passive reading is also available when the verb form is combined with an auxiliary, as in the following three examples: (110) ’w=š’n hm ’bjyrwng-[’]n ’st ky ’w dd-’n and=3 PL also disciple-PL some(?) which to wild.animal-PL ’bgn(d) ’hynd o ’st ky ’c šhr ’w šhr throw:PTCPL cop:3 PL and some(?) which from country to country ’šqrd ’hynd pursue:PTCPL be:3 PL ‘and they had disciples, some, which were thrown before wild beasts, others who were hunted from country to country [. . .]’ (Sundermann 1981: l. 1154–56) (111) hrēstāgān rōšnān, im rāy kuv̄st hend disciples:PL lightPL this because.of kill:PTCPL be:PRES:3 PL ‘The apostles of light were killed because of him.’ (Durkin-Meisterernst 2006: M200, Hymn 1)11 It is evident that the past tense verb forms, despite the fact that they were regularly used in Past Transitive Constructions, did in fact retain a certain amount of their participial origins. I refer to this feature as the lingering intransitivity of the past-tense transitive verb forms. Evidence of lingering intransitivity is rife throughout Middle Iranian, and is still evident in some modern languages (see Badı̄n. Kurdish in Chapter five). For Early New Persian (pre-1000AD), Heston (1976: 167) notes the 11 Boyce (1975: bf,1) gives a slightly different translation of this passage. 120 Western Middle Iranian existence of what she refers to as an “adjectival passive”, consisting of the participle, plus nominal inflection, together with a form of the copula. She interprets it as a “nominal clause” which “simply uses a participle instead of an adjective or adverb” as the copula complement: (112) pyš "zyn "ydwn gfth bwd before this thus say:PTCPL COP:PST:3 S ‘Before this thus (it) was said’ (Heston 1976: 167, glosses added) However, an ostensibly identical construction also occurred with an overt Agent: (113) kh d"r" bn" krdh "st which Dārā built do:PTCPL COP:PRES:3 S ‘. . . which Dārā has built’ (Heston 1976: 168, glosses added) The latter construction, however, apparently is more common with the short form of the copula as opposed to the full form in bwd in (112). Lingering intransitivity continued to characterize past-tense transitive verbs into Early Judaeo Persian (EJP, 8–12C . AD). Paul (2002: 162) notes that “the passive meaning of certain EJP simple past transitive forms” is still recognizable in some contexts: (114) cyst w’z cy grypt what.is and.from what take:PTCPL ‘what is (it) and from what has (it) been derived (lit. ‘taken’)?’ (Paul 2002: 162, glosses added) The coexistence of these two interpretations for past-tense transitive verbs is a recurrent feature of the Iranian languages. In fact Phillott (1919), commenting on Persian generally, notes that the original participle+copula nwšth "st can mean either ‘(it) is written’ or ‘(he) has written’.12 The fact that one verb form fulfills two functions, a stative-type passive and an active transitive, has led to the terminological confusion surrounding the issue, with different authors referring to the same constructions as passive, ergative, or agential. Paul (2002: 162, fn.4), discussing the situation for Early Judaeo Persian, considers 12 Quoted in Heston (1976: 227,fn. 16); Phillott (1919) was not available to me at the time of writing. Past transitive verbs 121 that the terms ‘passive’ and ‘ergative’ can be used “more or less synonymously”. Although this may not appear to be a very satisfactory position, it is probably a fair reflection of the messy indeterminateness of the verb forms concerned. The typological perspective shows that it is not necessary to force a particular construction into either a passive, or an ergative category, and it would appear justified in the Iranian case to make full use of any available leeway. Lingering intransitivity renders the analysis of the Past Transitive Constructions in Middle Iranian problematic. In the face of the evidence from the preceding examples, it is difficult to decide whether the verb forms they contain should be considered transitive, or intransitive. They seem to lack a clear lexical specification either way. Of course such labile verbs are well known from other languages, and indeed, English has more than its fair share (e.g., break, drop, melt, cf. Haspelmath (1993) for discussion). But while many languages have a lexical class of labile verbs, the Middle Iranian languages are characterized by the lability of a grammatically defined set of verb forms: the past tense forms of all transitive verbs. These forms are, as it were, in limbo between being participles, hence basically resultative verbal adjectives, and being finite transitive verb forms. The evidence from Middle Iranian, and indeed from some of the modern languages, suggests that this indeterminacy, far from being a fleeting transitional phase, characterized the languages for many centuries. In the light of this simple fact, the importance of the Construction Grammar approach should be evident: the transitivity claimed for the Past Transitive Construction is construction-specific, rather than being a consequence of the verb’s argument structure. In the subsequent history of the languages, the verb forms have progressively adapted to the constructions they occur in, and the morphology of the constructions has also changed. The pathways by which these changes have occurred are complex and intertwined, and will be the subject of later chapters. For the time being I would simply like to stress that the lingering intransitivity of the participial verb forms is ultimately behind the difficulties in analysing the Middle Iranian system of past tense verb forms. In practice, most analysts use the diagnostic of ‘presence vs. absence of an overt agent phrase’ to distinguish the Past Transitive Construction from the agentless passive. According to Paul (2002: 162), ‘when “no agent was 122 Western Middle Iranian expressed, a [Middle Persian] ergative past form must be translated with an English passive.” However, lack of an overt agent-phrase may actually have other causes. The most obvious one is that the agent-phrase is coreferential with the syntactic subject of the preceding clause, and has been omitted in the normal course of coreferential deletion. To appreciate this, it is necessary to examine longer sequences of clauses. When looked at in context, it turns out that the interpretation of many verb forms is considerably less straightforward, and in some cases, it is probably impossible to distinguish the agentless passive usage from a Past Transitive Construction with a deleted, but discourse-recoverable A. Consider the following: (115) hynd a. ud če’ōn sārār-ān ahrāft and as leader-PL uplift:PTCPL COP:PL ‘and as the leaders were uplifted’ b. ēg=išān dēn wihurı̄d then=3 PL religion confuse:PTCPL ‘then their religion became confused’ c. ud pad andarz ud kirdagān sust būd and through command and action:PL soft become:PTCPL hēnd COP:PL ‘and in words and deeds (they) have become soft [. . .]’ d. sidı̄g, ku awēšān gyānān pēšēnagān ig pad xwēš third that those=3 PL soul:PL former:PL which in self dēn kirdagān nē hanzaft, religion action not complete:PTCPL, ‘third, that those former souls, by-whom (ı̄g) in their own religion good action was not completed’ (Boyce 1975: a,l. 2–3) In (115a), we appear to have a genuine agentless passive sense of the verb form. In (115b), however, I am less convinced that this is the case. It seems possible that the clitic pronoun interpreted by Boyce as a possessor (‘their religion’) could equally be interpreted as the Agent (‘and then they confused/ spoiled their religion’). Thus it is not readily apparent which of the two readings is intended. Finally, in (115c) the relative clause introduced by ig seems to lack any overt Agent, but Boyce assumes that an Agent is intended, namely Past transitive verbs 123 one that is coreferent with the preceding NP ‘those former souls’. Thus mere lack of an overt A within a particular clause is not in itself a sufficient diagnostic for distinguishing between a Past Transitive reading, and an agentless passive. A parallel set of issues will concern us again in the discussion of certain facts of Kurdish in Section 5.6.2. At this point I will merely stress that the lingering intransitivity of the past-tense verb forms is characteristic of their usage throughout Middle Iranian, leading to the typologically unusual situation of identical past tense verb forms apparently being used in active and passive senses. The distinction is largely effected through the syntax, but given the propensity for Agents to be shared across clause sequences, and consequently to be omitted, it is no by no means always clear how to interpret a particular example. 3.7.2 The realignment of verbal agreement As noted, the facts concerning verbal agreement in past transitive constructions are somewhat murky. Generally, the participial verb forms are accompanied by an auxiliary which agrees with the O. But in many clauses, no auxiliary is used, and it is not clear if the participial verb form actually agrees with anything. Furthermore, agreement with an A is also attested. Thus there are at least three distinct agreement patterns with this type of clause. As I have been unable to find a systematic treatment of these matters in the literature, I am obliged to treat it fairly superficially here, relying mostly on the brief description of Middle Persian (Pahlavi) in Heston (1976: 176–179).13 Let us first consider the first pattern, where the participle is followed by an auxiliary agreeing in person and number with the O: 13 I am very much indebted to Bo Utas for his assistance in interpreting many of the Middle Iranian examples quoted in this section; needless to say, he bears no responsibility for any of the claims made here. In order to follow the analysis of Heston (1976), it is crucial to know that she uses the terms “active” and “passive” in a very restricted sense: Active means simply ‘the verb formally agrees with the A’, while passive means that the verb formally agrees with the O. 124 Western Middle Iranian (116) ME=m l’sd’l YKTLWNt’ HWEnd because=1 S highwayman(PL) kill:PTCPL COP:3 PL ‘Because I killed the highwaymen’ (Heston 1976: 177) Here, the O ‘the highwaymen’ is plural, although it carries no overt signal of plurality, in accordance with the system outlined in Section 3.4.1. The auxiliary, however, shows quite clearly plural agreement with the O. Another clear example of auxiliary agreeing with an O is the following: (117) xyndg bwd hym ’w=t’n dryst (q)yrd hym ill become:PTCPL COP:1 S and=2 PL healthy do:PTCPL COP:1 S ‘(I) was ill and you cured (me).’ (MacKenzie 1999c: 53–54) In the second agreement pattern regularly encountered, the auxiliary is dropped. It is apparently particularly frequent when the the O is third person singular. In the following example, the missing auxiliary is given with “ø”: (118) ME=m gndlp’ BRA ’wct‘ ø because=1 S Gandarw PREV slay:PTCPL ø ‘Because I slew Gandarw.’ (Heston 1976: 177) (119) čē=t ātaxš ı̄ man pus ōzad u=t pahrēz because=2 S fire of me son extinguish:PTCPL ø and=2 S nē kard tending NEG do:PTCPL ø ‘For you extinguished my son Fire and you did not tend (it)’ (Williams 1990a: 18f4) 14 Examples from later texts showing verbs without auxiliaries are legion, and here the O can be in other forms apart from the third person singular, for example (96) above. In the light of examples such as these, Heny (1984) claims that past transitive verbs should be generally interpreted as incapable of supporting agreement. Her view is based on an analysis of the texts in Nyberg (1964), where past “transitive forms so often lack the auxiliary when it should agree with the logical object” (Heny 1984: 84). However, there are 14 The translation of this passage in Heston (1976: 177) differs from the more recent one of Williams, but the differences are not relevant to the point at hand. Past transitive verbs 125 enough examples of auxiliary agreement with an O to make this strong claim untenable. Rather, we must accept the simple fact that there is considerable variation in agreement patterns. When we come to survey modern languages, it will become apparent that variation in the agreement on past transitive verbs is a trait of many languages with TSA, hence we should not be at all surprised to find it in Middle Iranian. A more drastic change, however, is when there is overt agreement, but with an A rather than with the O, the third pattern to be looked at here. This is apparently particularly the case when the past transitive verb is part of a sequence of intransitive, or present tense verbs, which regularly carry agent agreement. Consider the following example: (120) u andaršut-hēnd u=[šān] ō vištāsp-šāh namač burt-hēnd u and enter:PST-PL and=3 PL to Vištāsp-šāh homage pay and fravartak bēdāt-hend[. . .] letter ‘and they entered and payed homage to Vištāsp-šāh and delivered the letter.’ (Utas 1976: 80) In (120), the third person plural agreement on the copula is perfectly regular with the first verb (‘come’), because it is intransitive. With the following two verbs, however, it is unexpected, because they are transitive and should therefore agree with the O. But they clearly do not. I believe that there is a consistent principle at work here: if consecutive verb forms share an S or A, then the grammar avoids having them differ in their overt agreement. In his commentary to this example, Utas (1976: 80) suggests that the plural agreement following burt and bēdāt- may be a later interpolation. However, in the light of evidence from similar clause-linking constructions in later Iranian languages, it is quite plausible that under precisely the conditions outlined here, a past transitive verb could agree with a plural A. This same principle can be observed in connected texts of at least one modern language, the Northern Group of Kurdish, and will be investigated in more detail in Section 5.4.4. To my knowledge, no systematic investigation of agreement patterns across clause sequences has been undertaken for any Middle Iranian language as yet. 126 Western Middle Iranian Not all instances of A-agreement can be linked to pressure from clause linkage. There is no readily-apparent contextual reason for the A-agreement in the following example: (121) LA ME L krt‘ HWEwm no because 1 S:OBL do:PTCPL COP:1 S ‘No, because I did.’ (Heston 1976: 178) Some examples exhibit mixed patterning. The following example combines A-agreement in the first clause with no agreement in the subsequent ones: (122) AP=m HZYTWNt HWEm ‘mhrspnd AP=m HZYTWNt ø and=1 S see:PTCPL COP:1 S Holy.Immortal(s) and=1 S see:PTCPL ø ’p’ryk yzd’n’ Ap=m dyt ø plw’hl Y g’ywmlt other god:PL and=1 S see:PTCPL ø guardian.angel of Gayōmard W zltwhšt and Zoroaster ‘and I saw:1 PS Holy Immortals and I saw other gods and I saw the guardian angel of Gayōmard and Zoroaster.’ (Heston 1976: 178) Finally, one more brand of verbal agreement with past tense verbs should be mentioned. In some texts, examples are found where the verb agrees with neither the A, nor the O, but with an Indirect Participant, for example an Indirect Object or a Benefactive. This phenomenon was first pointed out by David MacKenzie in a paper from 1964, later reprinted as MacKenzie (1999a), and has since been confirmed by Tafazzoli (1986). Considerations of space preclude proper treatment of this topic here, though it will actually re-surface briefly in Chapter six. But it further strengthens the general impression that agreement with past transitive verbs in Western Middle Iranian cannot be simply accounted for in terms of a single grammatical rule. There are at least three possible distinct controlling elements (the A, the O or an Indirect Participant). 3.7.3 Summary of past transitive verbs We have seen in the preceding section two factors that interact in rendering the past transitive construction of Middle Iranian problematic. First, the verb Past transitive verbs 127 Table 7. Patterns of verbal agreement with past transitive verbs Overt A O-Agr. No Agr. A-agr. − − + + + + +/− + − − − +/− − + − − − − − + Exs. (110) (111) (115) (107) (108) (79) (117) (116) (83) (96) (106) (121) (122) forms retain to varying degrees their intransitivity, a legacy of their participial origins. Second, agreement on the predicate was not consistently determined by a single grammatical rule, but is determined by different controllers, with the variation as yet not fully understood. At least four distinct patterns can be found, which are summed up in Table 7 (the first two types differ only to the extent that the agreement on the verb can be considered to be with the O. When the O is third person singular, these two types cannot be distinguished. Agreement with an Indirect Participant is not shown in the table.) Table 7 is not intended to be read as a strict taxonomy, but merely as a preliminary descriptive framework, based on typically occurring feature bundles. It is probable that chronologically the four types given show a developmental sequence, with the verbs gradually moving towards the A-agreement of the accusative construction, as found in Modern Persian. but given the heterogenous sources, it is difficult to construct a single line of chronological development. A factor that we have not yet discussed is the introduction of a new Object marker, rā, which was already attested in tenth century Early New Persian. This marker is the etymological reflex of the Old Persian rādiy, with the sense of ‘due to, because of’, a meaning that is still preserved in Parthian im rāy ‘this because-of’ in (111) above. Its use as a marker of definite direct objects is a genuine innovation, and one which undoubtedly contributed to the abandonment of the tense-sensitive system of argument cross-referencing. Nevertheless, at least for a while the two systems co-existed, as indeed they do in many modern Western Iranian languages to this day. An example from Early New Persian showing an A-past clitic co-occurring with the innovated Object marker is the following (Heston 1976: 94): 128 Western Middle Iranian (123) k=š ’yn d’st’n r’ bgft when=3 S this story ACC tell:PST:IRR ‘When he told this story.’ Here we see a clause-initial clitic, expressing the A-past, typical for the TenseSensitive Alignment in the clitics. But simultaneously, this example also contains the innovated object marker rā (rendered orthographically here as r’). Agreement on the verb is not interpretable. The point here is that the presence of the innovated Accusative marker is generally considered to be the hallmark of full accusativity, while clitic-expression of the A-past is the hallmark of ergative alignment in Middle Iranian. Yet the two could apparently co-exist, although the status and chronology of such hybrid forms is uncertain. 3.8 Summary of Middle Iranian The data from Western Middle Iranian pose massive problems of interpretation, and the results here are correspondingly speculative. Nevertheless, the changes in the inventory of available inflectional morphology are fairly uncontroversial: (i) the loss of all nominal case distictions by late western Middle Iranian, leaving late Middle Persian lacking any overt case morphology; (ii) the loss of finite past tense verb forms, leaving old participles as the sole past tense verb forms; (iii) the expansion of the old Genitive/Dative forms of the pronominal clitics from their original Indirect Participant function to include both the O-pres and the A-past. In the latter function, they became increasingly grammaticalised (in the sense of increasingly obligatory). The relevant verb forms, old participles, although now the unmarked choice for the past tense, still retained clear signs of their participial origins. I call this lingering intransivity, by which I mean that these verb forms could still be used to express a simple agentless passive. The question of verbal agreement on these verb forms is a vexing one: straightforward examples of agreement with the O can be found, as would be expected, but lack of any agreement is also found, as is A-agreement. Given the fragmentary nature of attestation, it is difficult to construct a straight-line chronology for this, but one can assume from what is known from the modern languages that there has always been a tendency to abandon O-agreement in favour of A-agreement. Summary of Middle Iranian 129 Western Middle Iranian confirms the sobering fact that alignment changes do not proceed in neat progressions from, for example, accusative to ergative, interspersed with fleeting transitional phases. What the data actually suggest is that such transitional phases, which fail to conform to conventional alignment typologies, stretched on for centuries. Empirically, there is little reason to see them as in any sense less viable systems than the ‘purer’ alignment types such as ergativity or accusativity. In fact, such mixed systems are vastly more common in the Iranian context, both at the Middle Iranian and the modern stages, than the supposedly purer types, as we shall see in the next chapters. Chapter 4 Case systems in West Iranian 4.1 Introduction The preceding chapter has shown that between Old and western Middle Iranian, the case system underwent massive syncretism, leaving most of the attested western Middle Iranian languages practically bereft of inherited morphological case oppositions. In this chapter we take up the story of the casesystem again, focussing on the case systems of modern West Iranian. With the exception of Persian, no West Iranian language is attested back to any significant time depth, so there is no direct access to the history of these languages. The procedure followed here is essentially one of internal reconstruction: on the basis of the attested current variation, including dialectal variation, and with reference to what is known from the attested earlier sources of related languages, reasonably precise claims can be inferred. Of course this procedure is naturally more speculative than one based on actually attested earlier sources, but applied with due caution, nevertheless yields significant results. As in most of this book, ‘case system’ is used in a very narrow sense, referring to the morphological means for expressing the core syntactic functions of S, A and O on the respective NP. Case marking on nouns is of course only one of the three parameters that contribute to defining alignments in Iranian, the other two being argument cross-referencing through pronominal clitics, and verbal agreement. These two are, however, largely ignored in this chapter. This is not to downplay their importance, but is merely a necessary simplification which treating a large number of individual languages entails. I also believe that these three parameters develop to some extent independently of one another, hence it is justified to single out one of them here for detailed investigation.1 One point that needs to be emphasized is that many of the 1 Gerardo De Caro (p. c.) has suggested a closer interdependence of case and agreement, illustrated in the alignment variation found in Taleshi. He suggests that loss of verbal agreement is generally a prerequisite for the introduction of DOM, and with it the intrusion of accusative case-marking alignment into the past tenses. This is certainly an 132 Case systems in West Iranian languages have evidently preserved the inherited Direct/Oblique opposition, and continue to use it in a fashion that corresponds to what can be assumed for the earliest stages of Middle Iranian. In other words, the developments that were discussed for Middle Iranian in the preceding chapters by no means affected all the contemporary languages in the branch. It is simply an accident of historical documentation that those languages for which older stages are preserved (Middle Persian and Parthian) happen to have undergone radical loss of inherited case distinctions at a chronologically early stage. Other closely related languages, on the other hand, for which, due to entirely extralinguistic factors no historical records exist, retained the Direct/Oblique case distinction on nouns at least one thousand years longer. Just how these latter languages must have looked in the Middle Iranian period is a question that cannot be answered with any certainty. Be that as it may, these modern languages provide invaluable clues as to how the loss of case may have proceeded in earlier times, and how the old case system may have worked. The second extraordinary point about the modern West Iranian languages is the bewildering array of different alignment systems found in case marking. The accusative vs. ergative dichotomy is just as inadequate for these languages as it is for Middle Iranian. The chapter begins by discussing changes in terms of three related, but distinct processes: the loss of the inherited Oblique, the intrusion of an Innovated Object marker, and the revitalization of the inherited Oblique. These processes are exemplified through a case study of some of the most complex case-systems in the branch, the Tatic-type. In Section 4.5, I take up the question of explanations, contrasting the view of change driven by the need to preserve the discriminatory function of case, with the view according to which cross-system harmony is decisive. Finally, in Section 4.7 I propose a set of ranked principles which provide the most compelling explanation for the attested changes in case-marking systems in West Iranian, and indeed in Iranian generally. avenue that needs to be pursued, although Central and Southern Kurdish, as well as some dialects of Zazaki, seem to have largely lost person agreement on the verb without any accompanying development of DOM. Three processes 133 4.2 Three processes The history of the case systems of Iranian has been characterized by Windfuhr (1992: 26) in the following terms: [. . .] Typically, in the history of languages, morphological differentiation is cyclical: a stage with no or minimal differentiation is followed by a stage with increased differentiation, which in turn is followed by a stage of decay of the case system into a minimal binary opposition, direct case vs. oblique case, or total loss of formal differentiation other than by pre- or postpositions and/or word order. These rudimentary stages are then followed by newly differentiated case marking, usually by the development of pre- or postpositions into ending-like or prefix-like case markers. Such cyclical developments are well represented in the history of the Iranian languages, as well, and each of the many Iranian languages of today represents various stages in that cycle. Viewed from a macro-perspective, one that spans, for example, the development from Old to Modern Persian, a cycle of collapse and renewal appears to be an apt metaphor. It is nevertheless an oversimplification to view all changes in these terms, and in fact, many of the smaller changes in individual languages actually seem to run counter to these expectations. For example, the renewal of a case system using novel means often begins before the old one has disappeared completely. Thus we have considerable overlap, leading to doubly complex systems in parts of the grammar. Second, far from dying out, the inherited case markers in some languages have actually expanded their ranges, thereby undergoing some rather interesting changes in morphological type. Finally, the novel case markers, rather than merely replacing the old ones, often show some very significant functional differences, in particular with regard to alignment. In fact, to a certain degree, the non-accusative alignments of past tenses are tied to ancient morphology, both verbal and nominal, and when this disappears, there is a strong, though not exceptionless, tendency for the new system to be fully accusative. Nevertheless, the general drift in Iranian can profitably be seen in the manner depicted in the above quote from Windfuhr, as a cycle of collapse and renewal. But while the cycle undoubtedly moves in a similar direction across all the languages, the pathways taken by the individual languages can differ quite remarkably. 134 Case systems in West Iranian The main types of change can be described in terms of the interaction of the following three processes: (1) Loss of the inherited Direct/Oblique case distinction (2) Intrusion of an Innovated Object marker into the case system (3) Revitalization of the inherited Oblique case While the first two have been noted and discussed under various labels in the literature, the third is somewhat more problematic, and I willingly concede the preliminary nature of the analysis at this stage. Nevertheless, the facts from certain languages warrant assuming it as a possible, and distinct, development in the case systems.2 In this section, I will discuss each process in its own right before tackling the more challenging task of describing the interactions between them. For some languages, the individual processes can be very simply observed, because they proceed in a fairly orderly fashion throughout the entirety of the grammar, and with little overlap. Persian, for example, initially underwent the first process (loss of the inherited Direct/Oblique distinction in the Middle Persian period) throughout the entirety of the grammar, and then underwent the second, likewise throughout the entirety of its grammar. Here we have the paradigm exemplification of the cycle of collapse and renewal, with both processes proceeding in a clear chronological order, and affecting all nouns and pronouns. The Southern Group of Kurdish has also undergone the first process throughout the entirety of its grammar, but without any accompanying intrusion of an Innovated Object marker. Zazaki, on the other hand, has, apart from certain complications, retained the inherited Oblique in a similar fashion to what can be assumed for an attested Proto-Middle Iranian, and has not acquired any Innovated Object marker. Persian, Southern Group of Kurdish, and Zazaki thus represent fairly straightforward examples for how these developments may occur, or may not. Such clear-cut cases of the three processes working in relative isolation probably represent a minority. Most of the West Iranian languages actually show a mix, for example losing their inherited Oblique in parts of the gram- 2 Primarily the data from Balochi and Talyshi convinced me of the importance of this process. I am very grateful to Agnes Korn and to Gerardo de Caro for much challenging discussion on Balochi and Northern Talyshi respectively. Three processes 135 mar, while retaining it in others, and adopting Innovated Object markers likewise only in parts of the grammar. Such grammar-internal, and chronological, overlaps give rise to systems of considerable complexity, and of considerable interest. Nor is there any a priori reason to consider them any less viable or stable than the predictable and ‘symmetrical’ systems of, for example, Persian. Rather, they demonstrate one of the major themes of this book, the potential independence of morphosyntactic change in different parts of the grammar. Nevertheless, it can be shown that the attested combinations of outcomes are not unconstrained, but are shaped by certain principles which I will be formulating towards the end of the chapter. 4.2.1 The loss of the Inherited Oblique In this section I investigate the changes undergone by the inherited case markers. It is expedient to treat these developments in two parts: changes in the third person (nouns and pronouns), and changes on the (free forms) of SAPpronouns. There are obvious functional and semantic differences between expressions uniquely referring to the speaker and the addressee (first and second person, Speech Act Participant (SAP) pronouns), and expressions denoting other persons or objects (third person), and these are duly reflected in numerous formal assymetries in the grammars of the languages of the world (see Siewierska (2004: 5) for discussion). In Iranian too, the SAP-pronouns regularly diverge from the rest of the nominal lexicon, in particular both in the type, and number, of case-distinctions drawn. It is worth noting that for most of the relevant languages, a dedicated ‘third person personal pronoun’ corresponding to English he, she, it etc. does not exist. Instead, forms of distal demonstratives are regularly analyzed as ‘third person personal pronouns’, leading to considerable confusion in many traditional descriptions. Thus in Iranian, the split between SAP and the rest of the grammar begins with the existence of dedicated pronouns for SAP-functions, and a corresponding lack of dedicated personal pronouns in the third person.3 Other minor areas of the 3 Modern Persian is exceptional in this respect; here a third person pronoun u can be distinguished from the distal demonstrative ān. It is notable, however, than in colloquial speech u is hardly ever used, the demonstrative serving as a pronoun in most contexts. 136 Case systems in West Iranian Table 1. Case distinctions on nouns in early Western Middle Iranian D IRECT O BLIQUE S INGULAR P LURAL -0/ *-ē -0/ -ān nominal lexicon, where specific divergent developments are found (for example kinship terminology, or different genders) are mentioned at various times, but cannot be systematically distinguished for all languages. 4.2.1.1 The third person The case system on nouns that can be assumed to underly the West Iranian languages is shown in Table 1 (the table replicates Stage 1 in Table 3 from the preceding chapter). At this stage, TSA was undoubtedly the norm, with an ergative-type alignment in past tenses. The (typologically very unusual) basic system of leaving singular and plural Direct unmarked, while marking the Oblique case of both numbers with a distinct marker, is retained, barring phonological change, in at least the conservative dialects of the Northern Group of Kurdish (cf. Section 5.2.1). In this language we find the simplest type of change to the inherited case system: no change at all. The second type of theoretically possible change is the converse of the first, where all the inherited case markers are abandoned. Such a system would leave all nouns unmarked for case – and crucially, unmarked for number as well. In fact, such a system, with total abandonment of inherited case markers, is not attested anywhere in West Iranian. What in fact happens is rather more subtle. Complete loss of the singular Oblique marking is extremely common, yet the plural Oblique marker is inevitably retained. But while the marker itself is always retained, its function often shifts from being a marker of the Oblique Plural to simply marking the opposition Singular vs. Plural, with no accompanying case component.4 The system resulting 4 Some languages, for example those of the ‘Tatic type’ (see Section 4.4), have undergone a slightly different development in the plural: a (presumably) innovated Direct Plural suffix is added to the system, −e (or similar), while the inherited Oblique Plural −ā(n) Three processes 137 Table 2. Loss of inherited case distinctions in the singular D IRECT /O BLIQUE : S INGULAR P LURAL -0/ −ān from the complete loss of the singular Oblique, with accompanying extension of the Plural Oblique to a general Plural marker, is reproduced in Table 2. Most of Middle Persian showed a system of this kind (cf. Section 3.3), as do many of the Central and Southern Kurdish dialects, and most of the Central Plateau Languages described in Lecoq (2002). Basically, case simply no longer is an inflectional category on nouns in these languages. The change between the presumed ancestral system given in Table 1 and the full loss of case distinctions shown in Table 2, evidently entails two subchanges: (i) the extension of the old Plural Oblique −ān to the Plural Direct; (ii) the loss of the Singular Oblique. In theory, these two changes could have happened independently of one another, and in any order. But actually, there appears to be a universal sequence, with (i) occurring before (ii). In support of this claim I can note that no language is attested which has fully lost the Oblique in the singular, but retained the distinction between unmarked Plural Direct and marked Plural Oblique.5 In other words, change (ii) always presupposes change (i). The converse, 5 is retained in its original function. This does not detract from the general claims being made here, but merely confirms that marking of number is of high priority in the system. An apparent counter-example is the East Iranian language Sarykoli (Pamir group, Payne 1980: 153). Here, the Direct Plural is marked by a clearly innovated suffix, while the Oblique Plural has a different suffix, thus ensuring a case-distinction in the Plural. In the singular, however, a case-distinction is absent entirely. This system is basically comparable to that of the Tatic-type languages discussed in fn. 4, except that in Sarykoli, the case distinctions have been lost in the singular, whereas in the Tatic-type languages, at least some are maintained. But it is not a counter example to the claim made in the text, which would require a language to have a zero (Direct) vs. marked (Oblique) distinction in the Plural, but no case distinction in the singular. Crucially, no language is attested that exhibits this combination although, given the assumed shape of the original system, it is surely a feasible outcome. 138 Case systems in West Iranian Table 3. Case distinctions on nouns in Northern Talyshi (Schulze 2000, De Caro 2005) D IRECT O BLIQUE S INGULAR P LURAL -0/ −(n)i −on however, does not hold. Northern Talyshi, for example, has apparently undergone change (i), but not change (ii). The relevant parts of the system are shown in Table 3. In Northern Talyshi, case distinctions are neutralized in the plural of nouns (in the personal pronouns, matters are more complex, see below). What is evidently the old Oblique plural marker is now merely an indicator of plurality, as it is in countless other Iranian languages. But unlike the system of, for example, Southern Kurdish shown in Table 2, Northern Talyshi has retained a Direct/Oblique distinction in the singular. The inherited Oblique markers of pre-western Middle Iranian thus show rather distinct paths of development in the singular and plural of nouns. In the singular we generally find either (a) complete loss of the actual marker, or (b) its retention (on at least some of the relevant forms), or (c) even extensions into new environments, see below. The plural Oblique marker, on the other hand, is invariably retained, but is commonly functionally co-opted: it no longer indicates the feature bundle [+Plural, +Oblique] but changes to merely [+Plural]. The changes in the singular and plural are, to some extent at least, independent of one another, but it seems that if there are any changes, it is the above-mentioned shift of the plural Oblique to a general plural marker which will occur first. This example illustrates the principle of functional rankings of available morphology. The basic idea is that certain functions take priority over others; when there is little morphology to go round, then the functions with the highest priority are filled first. In this case, the priority ranking is: Number before case. The simplest support for this idea comes from the fact that all West Iranian language have some nominal morphology for distinguishing plural from singular on nouns, whereas many have no nominal morphology for distinguishing Direct from Oblique. The changes thus essentially sacrificed the marking of case to the marking of number. Three processes 139 Table 4. SAP-pronouns in assumed early Western Middle Iranian S INGULAR 1 P. 2 P. P LURAL D IRECT O BLIQUE az *tū man tō D IRECT O BLIQUE amāh ašmāh Table 5. Eight-term system of SAP-pronouns in the Northern Group of Kurdish S INGULAR 1 P. 2 P. 4.2.1.2 P LURAL D IRECT O BLIQUE D IRECT O BLIQUE ez tu min te em hûn me we SAP-pronouns The free forms of the SAP-pronouns for an assumed early western Middle Iranian are given in Table 4, based on evidence from Inscriptional Parthian (cf. Table 4 in Chapter three for the sources). There are two vital differences to note between this system, and that of the nouns discussed in the preceding section. First, the inherited Direct/Oblique distinction on SAP-pronouns was suppletive, rather than affixal. Second, in the plural, no case distinction is made, or at least attested. With regard to the second point, it is notable that in Northern Kurdish at least, a suppletive case distinction is drawn also on the plural of the SAP-pronouns, yielding the eight-term system shown in Table 5. This system represents the maximum array of suppletive distinctions on SAP-pronouns in West Iranian.6 It is a reasonable assumption that a system of this kind in fact pre-dated the one assumed for early Western Middle Iranian in Table 4, but the Direct/Oblique distinction in the plural was lost very early in most languages, with Northern Kurdish being perhaps the only one 6 Presumably under contact pressure, some languages have developed a secondary inclusive/exclusive distinction on the first person plural pronouns (for example Turkmen Balochi, Axonov 2006). These forms are ignored here. 140 Case systems in West Iranian Table 6. Inherited case on SAP-pronouns in Zazaki (Paul 1998b: 67) S INGULAR 1 P. 2 P. D IRECT O BLIQUE ez ti mi(n) to P LURAL D IRECT O BLIQUE mā šimā Table 7. Inherited cases on SAP pronouns in the Abyānei dialect, Central Plateau (Lecoq 2002: 82) S INGULAR 1 P. 2 P. D IRECT O BLIQUE az tö man ta P LURAL D IRECT O BLIQUE hāmā šömā to retain it. In the other West Iranian languages, lack of an inherited case distinction in the plural of the SAP-pronouns is the norm. Typical examples are the system of Zazaki (Table 6) and the Abyānei dialect, Central Plateau (Table 7). A six-term system of this type, with suppletive case distinctions restricted to the singular, is also found in the Tāti dialect of Kajal (Yarshater 1960), although the forms of the second person singular Direct and Oblique are only very weakly differentiated, as pointed out to me by Gerardo De Caro (p. c.). A language where this latter distinction is completely lost, leaving a fiveterm system, is the Tāti dialect of Shāhrud, spoken in Khalkhāl province in southeastern Azerbaijan (Yarshater 1959: 58). The basic unity behind these systems is nevertheless quite remarkable, and permits us to formulate the following statement on the sequence of changes in the inherited case systems on SAP-pronouns in West Iranian languages: (124) Inherited case distinctions on SAP-pronouns are universally lost in the plural before the singular. What is the explanation for (124)? To some extent, the pattern is quite expected: normally we find at least as many inflectional distinctions to be avail- Three processes 141 Table 8. Inherited cases on SAP-pronouns in Northern Talyshi (Schulze 2000: 35) S INGULAR D IRECT 1 P. 2 P. P LURAL O BLIQUE āz m@ t@ D IRECT O BLIQUE ama š@ma able to the unmarked member of an opposition as to the marked one. Thus lack of case distinctions on plural forms, despite their presence on the corresponding singular forms, is no great surprise (but recall the case of Sarykoli in fn.5), and may be accounted for in terms of neutralization of inflectional categories in the marked number value. Alternatively, it may be linked to the (presumably) lower text frequency of the plural forms as opposed to the singular. It is well known that morphological irregularity, and in particular, suppletion, only survives on items of high frequency. Thus differences in frequency of occurrence between singular and plural SAP-pronouns may help to account for the striking differences in retention of suppletive case-marking distinctions on these pronouns. A second striking regularity in the SAP-pronouns is the regular loss of the Direct/Oblique distinction on the second person singular. When this happens, we are left with a system in which the sole inherited case distinction among the SAP-pronouns is on the first person singular. An example of this very common state of affairs is found in Northern Talyshi, shown in Table 8 (see also the Tāti dialect of Xo’ini (Yarshater 2003) for a parallel system). The very frequent loss of case distinctions on the second person singular is somewhat difficult to account for in terms of animacy, or frequency (but compare the comparable situation in English). I suspect we are dealing here with a development specific to Iranian, for which certain phonological factors are responsible. The distinction between an Oblique and a Direct form of this pronoun rested at best on the quality of the stem-final vowel, in Parthian for example presumably Direct *tū vs. Oblique *tō. It is likely that this distinction has proved less resilient than the az/man-distinction in the first person singular, where the phonological contrast between the two members of the opposition is much more salient. Whatever the reason, it is absolutely typ- 142 Case systems in West Iranian Table 9. SAP-pronouns in the Southern Kurdish dialect of Xânaqin, Iraq (Fattah 2000: 276– 277) S INGULAR D IRECT 1 P. 2 P. O BLIQUE mın tu P LURAL D IRECT O BLIQUE ima iwa ical, and quite possibly universal for West Iranian, that the inherited case distinction on the second person singular is lost before that of the first person singular. The final possibility that needs to be noted is the complete loss of inherited case distinctions on SAP-pronouns. This is found, for example, in all the Central Plateau dialects investigated by Lecoq (2002) (except Abyānei, see Table 7), and also in the Southern Group of Kurdish (Fattah 2000). A fairly typical example for such a system is provided in Table 9. The developments outlined so far can be summed up in the form of a sequence of loss of inherited case on SAP-pronouns, given in (125): (125) Sequence of loss of inherited case on SAP-pronouns Plural > 2 S > 1 S We have then the orderly reduction of a presumably eight-term system, still found in Northern Kurdish (Table 5), to a four-term system illustrated in Table 9. Of course full reduction is not obligatory, and most West Iranian languages show some kind of intermediate stage. But what appears to be obligatory is the particular pathway down which a reduction will proceed. Thus we find no language which has a suppletive case distinction somewhere in the plural, but none in the singular. Likewise no language is found with a suppletive case distinction in the second person singular, but not in the first. We are now in a position to sum up the most important findings regarding the loss of the inherited Direct/Oblique case distinction: 1. The inherited Oblique on nouns is frequently lost in the singular. 2. The inherited Plural Oblique on nouns is never lost, but the marker itself is re-deployed as a general marker of plurality, with no casecomponent. Three processes 143 3. The latter development always precedes the complete loss of the Inherited Oblique on the singular of nouns. The appropriate generalization is that if any inherited inflectional nominal morphology is retained, it will be used to distinguish number. 4. The Inherited Oblique distinction on the SAP-pronouns is suppletive rather than affixal. 5. The loss of the Direct/Oblique distinction on the SAP-pronouns follows a strict sequence: plural forms first, then the second person singular, then the first person singular. The relative independence of case loss on the SAP-pronouns from case-loss on the nouns raises the interesting question of whether any regularities can be observed in the relative timing of the two processes. There are languages where the only inherited case distinctions left are found on the SAP-pronouns, for example the Abyānei dialect, Central Plateau, discussed in connection with Table 7. Here the principle is clearly that the inherited case distinctions are maintained longest on the SAP-pronouns. However, in the Gorani dialects investigated by Hadank and Mann (1930) and MacKenzie (1966), the SAP-pronouns have completely lost their old suppletive Oblique/Direct case opposition, while the nouns have retained what is presumably the inherited Oblique case. Thus in principle at least, both orders are possible. However, it is very notable that in those languages where SAP-pronouns lose their inherited suppletive case distinctions before the nouns lose their old Oblique affixes, there is an (almost exceptionless) tendency for innovated case markers (see next sections) to make up the difference on the SAP-pronouns. Thus we do not find a situation where case marking is found somewhere on the nouns, but not with SAP-pronouns (the sole exception I am aware of is AwromanGorani (MacKenzie 1966), see below, but in the closely related dialects case has been restored on the SAP-pronouns). The basic principle is thus as follows: If there is any morphological case (of any provenience) in the grammar, then it will be found on the SAP-pronouns. 144 Case systems in West Iranian 4.2.2 Re-vitalizing the inherited Oblique While the preceding pages may have created the impression that the story of the Inherited Oblique is largely one of gradual loss, in some languages, rather interesting converse developments can be witnessed. Briefly, what occurs is a re-vitalization of the Singular Oblique marker on nouns, leading in some cases to an extension of this marker into other domains, in particular on to the SAP-pronouns. The developments outlined in this section concern solely the inherited Singular Oblique case of nouns, for which we have assumed a protoform *-ē, itself probably a continuation of the Old Iranian Genitive Singular of the −o-stems −ahyā.7 In the modern Iranian languages, we find this suffix represented by a variety of vowel qualities, but generally sharing the features of unrounded, and front. Let me begin by pointing out some fundamental differences between the reflexes of the inherited Oblique case, and its Old Iranian ancestors. The cases of Old Iranian were fusional, like most of the Ancient Indo-European languages, with allomorphs depending on gender, number and declensional class. But although we have followed common opinion in analysing the Oblique marker of, for example, Kurdish (Northern Group) −î/-ê (variants depend on gender) as the descendent of the Old Iranian Genitive case, the distributional properties of the modern Oblique differ quite strikingly from those of its ancestor. In particular, the Oblique case in Kurdish may be separated from the noun stem by other suffixes, something that was quite impossible for the Old Iranian cases. Thus we find the following combinations of Oblique case with the indefiniteness suffix −ek: 7 This etymology appears a likely one given (a) the massive polysemy of this case, hence its high level of frequency (see the discussion on the Old Persian Genitive/Dative in Section 2.5); (b) the phonological similarity to the Genitive of the −ā-stems (see Kent (1953: 60) for details); (c) its comparatively heavy, bisyllabic form, making it a likely candidate to survive, at least in part, the loss of unstressed final syllables; and (d) the fact that this form was already encroaching into other environments, for example the “neologisms” listed by Kent (1953): Dārayavaušahyā, ‘Genitive Sg.’ of the name Dārayavahuš (Darius). Three processes 145 (126) (i) gund (ii) gund-ek (iii) gund-ek-î (i) village (ii) village-INDEF (iii) village-INDEF-OBL On the reasonable assumptions that the indefiniteness marker emerges via grammaticalization of a numeral yek ‘one’, and is hence a comparatively late addition to nominal inflection, the presence of the indefiniteness marker between the stem and the inherited Oblique is rather strange. In languages that exhibit this kind of pattern, there has obviously been some re-organization of available inflectional morphology. The old markers have, to some extent, been de-grammaticalized, in that their degree of bonding to the stem has been weakened, allowing them to stack up after other suffixes in a very agglutinative manner. In some languages, the other widely-retained inherited suffix, the Plural Oblique in −ān, has also acquired (some) secondary agglutinative characteristics. A particularly striking example of how case, number, and definiteness are expressed in a very agglutinative manner is provided by the nominal morphology of Gorani (MacKenzie 1966: 13–15), illustrated in Table 10. Here the Oblique Singular suffix −y can be attached both to the bare indefinite form of the noun, or to the definiteness suffix −(a)ká. In the plural, on the other hand, we find a reflex of the common West Iranian Plural Oblique suffix −ān, but with a deleted final nasal. Although this suffix has in fact retained more of a fusional character, in that it expresses two categories, [+Plural, +Oblique], it is distributionally agglutinative in the sense that it follows a definiteness marker in the Definite Plural Oblique. It is also remarkable that with noun phrases consisting of noun+adjective, linked by an Izafe-type vowel (−a), the definiteness marker plus Oblique case attaches not to the noun, but to the adjective. Consider example (127) from MacKenzie (1966: 17–18) below: Table 10. Case, number and definiteness in Gorani (Awroman dialect) SG. PL. D IRECT O BLIQUE T RANSLATION INDEF . DEF . yāná yān-aka yāna-y yān-aka-y ‘a house’ ‘the house’ INDEF . yāné yān-ak-é yān-ā yān-ak-ā ‘houses’ ‘the houses’ DEF . 146 Case systems in West Iranian Table 11. Secondary Oblique marking of SAP pronouns in the Kandulai dialect of Gorani (Hadank and Mann 1930: 117) S INGULAR 1 P. 2 P. (127) P LURAL O LD O BL . S EC . O BL . O LD O BL . S EC . O BL . ämín tû ämín-î tû-î îmä šümâ îmä-î šümâ-î a. [kitéb]-aká book-DEF b. [kitéb-a siāw]-aká book-IZ black-DEF ‘the black book’ c. [kitéb-a siāw]-aká-y / siāw-ak-́ā book-IZ black-DEF-OBL / black-DEF-OBL:PL ‘the black book / books’ (as Direct Object) The above brief account of the Oblique case in Gorani ignores some of the complex interactions between case morphology, Izafe-vowels, and the possessive suffixes (MacKenzie 1966: 26). But the general tendency is clear: lack of allomorphy, relative ease of segmentation, with a tendency towards phrasal rather than nominal marking. Markers that exhibit this degree of agglutination are good candidates, or at least better than their fusional ancestors, for extending their distribution to classes of words that they were previously not found on. This appears to have happened in the Gorani dialect of Kandulai, as described in Hadank and Mann (1930), where the inherited Oblique has crossed over from the nouns to the SAP-pronouns. The relevant forms are provided in Table 11, where the forms with a revitalized Oblique marker adopted from the nouns are listed as “Secondary Obliques (S EC . O BL)”.8 8 Table 11 ignores some of the phonetic variants listed in the source, and also simplifies the transcription somewhat for technical reasons. The changes are not relevant for the issues at hand. Three processes 147 Table 12. SAP-pronouns in the Awroman dialect of Gorani (MacKenzie 1966: 24) S INGULAR D IRECT 1 P. 2 P. O BLIQUE mIn to P LURAL D IRECT O BLIQUE emá šImá When we compare the Kandulai SAP-pronouns with those of the closelyrelated Awroman-dialect of Gorani, we catch a glimpse of what must have been the sequence of changes that led to the situation in Table 11: complete loss of the old suppletive Direct/Oblique distinction on the SAP-pronouns, before Oblique suffixes were secondarily reinstated in line with the nouns. Table 12 shows the forms for the SAP-pronouns in the Awroman dialect of Gorani, where, according to MacKenzie (1966), no case distinctions are drawn on the pronouns. The spread of the Oblique suffix from the nouns to the SAP-pronouns appears crucially to begin with third-person pronouns/demonstratives, which in both dialects have an Oblique form with −i suffix (e.g., áu-î in Kandulai, cf. Hadank and Mann 1930: 117). Thus these deictic elements appear to act as the thin edge of the wedge in the process of extending the case marker to the SAP-pronouns: from common nouns, to third person deictics, to SAPpronouns. Let us turn now to the functions of such secondary Oblique forms of the pronouns: Whereas the old Oblique generally codes (among other things), the A-past, the new secondary forms consistently mark exclusively Objectfunctions in all tenses, while S and A are coded identically with the unmarked forms. This is illustrated in the following Gorani examples, each showing SAP-pronouns in A-past and O-past functions: (128) a. ämín áu-î=m vêil kärdän, au ämn-î vêil 1 S 3 S-S EC O BL=1 S loose do:PST:3 S(?), 3 S 1 S-OBL loose ní-mä-kärû NEG -PROG -do:PRES:3 S ‘I(A-past) have let him go, but he isn’t letting me(O-pres) go.’ (Hadank and Mann 1930: 304) 148 Case systems in West Iranian b. ämín tû-î=m kûawä dî 1 S 2 S-S EC O BL=1 S on.the.mountain see:PST ‘I(A-past) saw you(O-past) on the mountain.’ (Hadank and Mann 1930: 296) While in terms of case-marking, (128a and b) exhibit straightforward accusative alignment (NPs in S function are also in the Direct case), they nevertheless maintain the typically ergative pattern of pronominal clitics crossreferencing the A-past (=m in both examples). Their occurrence here is symptomatic for the mixture of different alignment patterns in one and the same clause, and the independence of clitic-cross referencing and case marking. However, the more important point in the present context is that when an old Oblique marker from the nouns crosses over to apply to the SAP-pronouns, it works on the pronouns as an Object marker, and is generally not used to code a past A. In Gorani, the revitalized Oblique is applied to a system of SAP-pronouns that has completely lost the old Direct/Oblique distinction, suggesting that the new marker ‘compensates’ for the loss of the old distinction. However, such a neat sequence of developments is not found everywhere. A different language, Northern Talyshi, shows essentially the same development, but here we find a revitalized Oblique marker added to the SAP-pronouns even though the inherited suppletive Direct/Oblique opposition is still present, at least on the first person singular. The resulting system is shown in Table 13. Thus while the renewal of case on the SAP-pronouns can be seen as shoring up a system that has lost its inherited case distinctions, it can also apply apparently superfluously, leading to what De Caro (2005) refers to as a “hyper-marked” form, for example the first person singular in Northern Talyshi. Table 13. SAP-pronouns in Northern Talyshi (Schulze 2000: 35) S INGULAR D IR . 1 P. 2 P. P LURAL O LD O BL . S EC .O BL . D IR ./O BL . S EC .O BL . m@ m@ni t@ni ama š@ma amani š@mani āz t@ Three processes 149 What was said above regarding the function of the secondary Oblique in Gorani also holds true for the majority of examples in Northern Talyshi: it is a general object marker (in both tenses), and both De Caro (2005) and Schulze (2000) refer to the secondary Oblique forms of the pronouns as an “Accusative”. However, the data are not entirely consistent, and matters are further complicated by the continued co-existence in the first person singular of the inherited suppletive opposition, and the secondary Oblique form. Two examples from De Caro (2006) should suffice for the moment (transcriptions have been simplified in accordance with suggestions made to me by Gerardo De Caro): (129) xan mïni voqând-a=š-e Khan 1 S:S EC O BL send:PST-PERF=3 S(pronominal clitic)-3 S ‘The Khan has sent me.’ (De Caro 2006) (130) âz šïmani voqand=ïm-e 1 S 2 PL:S EC O BL send:PST=1 S(pronominal clitic)-3 S ‘I sent you.’ (De Caro 2006) A final illustration of an inherited case marker expanding into new environments is provided by the old kinship-oblique markers. It will be recalled from Section 3.3 that kinship terms in western Middle Iranian had distinct Oblique forms, containing an [-r-]. In most modern West Iranian languages, these distinct Oblique forms are found only in very isolated traces, or not at all. But in some of the Southern Tati dialects investigated by Yar-Shater (1969), this marker has expanded its usage. For example, in the Chali dialect it is now used “with all nouns denoting a human being, except proper nouns” (Yar-Shater 1969: 87), though not in all environments where an Oblique case would be expected. For instance, it is used with non-kinship terms only when they are modified by a genitive attribute (here a personal pronoun): (131) cemā cupun-ar ku 1 PL:GEN shepherd-OBL from ‘from our shepherd’ (Yar-Shater 1969: 88) Otherwise, the general Oblique in −e is used: (132) ku cupun-e shepherd-OBL from ‘from this shepherd’ (Yar-Shater 1969: 88) je DEM 150 Case systems in West Iranian It is fairly obvious how the addition of the Genitive modifier serves to make the noun conceptually closer to a kinship term, hence the extension to nouns in this environment is well-motivated. In other dialects of the same group, however (for example Eshtehardi), the special Oblique in −r is scarcely attested, and then only with the words for ‘sister’, ‘wife’ and ‘daughter’ (YarShater 1969: 91). Zazaki shows a similar expansion of the old kinshipOblique. In this language, there is a “facultative” Oblique singular suffix −er used to mark the Singular Oblique of nouns of feminine gender, most commonly nouns denoting animates (it is found with approximately 50% of the attested animates, but with less than 15% of the attested inanimates, Paul 1998b: 22). The notion of revitalization developed here assumes that an inherited case marker may, under conditions that need to be clarified, acquire an increased degree of agglutinative characteristics, and in the extreme case, may cross the border from the third person to the SAP-pronouns. There appear to be a number of different parameters involved in the revitalization process, including the levelling out of allomorphy, the ability to be separated from the noun stem, and occurrence at the edge of the NP. In some languages, the singular Oblique marker displays almost none of these properties. For example, in Zazaki there is an indefiniteness suffix −ē(n), but unlike the functionally equivalent marker in Northern Kurdish, it cannot intervene between an Oblique case marker and the stem. Instead, it “replaces” (Paul 1998b: 13) the case marker. Generally, the Oblique marker of Zazaki reveals a very archaic character, with a high degree of allomorphy, and no positional freedom. Thus we would be very surprised to find it used with SAP-pronouns, and this prediction is borne out by the data. A summary of some of the languages investigated here with regard to degree of agglutinative character of the Singular Oblique marker is provided in Table 14.9 There are four parameters investigated, which are defined as follows: 9 Abbreviations used in Table 14: Gor-1: Awroman dialect of Gorani (MacKenzie 1966); Gor-2: Gorani dialect of Kandulai (Hadank and Mann 1930). Notes: (1) Postposed genitive attributes regularly take the Oblique case, thus leading to an Oblique at the ‘edge’ of the NP; however, this particular instance of the Oblique does not signal the syntactic function of the NP as a whole, hence is not treated here as Oblique marking of the Three processes 151 Table 14. Properties of sg. Inherited Obliques in some W. Iranian languages (see fn. 9 for notes to Table 14) # Zaz. 1 2 3 4 − − − − N.Kurd − + −(1) − S.Tati − n. a. n. a. − C.Kurd Vafsi N.Tal. − + −(1) − − (3) a. n. a. − + + n. a. + Gor-1 − + +(?) − Gor-2 + (3) +(3) 1. Unified form The marker in question has a single exponent for all nouns, that is, there are no differences according to gender or other types of noun class. 2. Occurrence outside of (in)definiteness markers (only valid for languages with such markers) 3. Occurrence at the edge of the NP, regardless of word class of base (for languages with head-final NPs, this would apply automatically, so these languages receive “n. a.” in the table) 4. Extension to SAP-pronouns It will be seen that the Gorani dialect of Kandulai (Gor-2), and Northern Talyshi, represent extreme cases of agglutinative properties of their inherited Oblique markers. Crossing the border from the third person to SAPpronouns is clearly not particularly common, but is a possibility that cannot be excluded. A further possible candidate showing this type of cross-over is Balochi, but the interpretation of the Balochi data is highly controversial, and cannot be treated in detail here. Nevertheless, most of the languages examined exhibit a mix of fusional and agglutinative characteristics, for example, different Oblique singular markers according to gender, yet positioning outside of (in)definiteness markers (it is interesting to note that gender-related allomorphy of the Oblique marker tends to be neutralized, or at least reduced, when the marker follows an (in)definiteness marker). NP. Postposed adjectives are not Oblique marked. (2) In those cases where a definiteness marker follows the postposed adjective there is Oblique marking of the case relations of the NP at the NP-edge. Whether this is always the case is difficult to ascertain from the data in MacKenzie (1966). (3) Data insufficient. 152 Case systems in West Iranian The notion of revitalization discussed in this section does not lend itself to a water-tight definition. Rather, there are differing degrees of revitalization, and I would be reluctant to specify any particular limit. Nevertheless, some notion along these lines is required to account for the cases of Northern Talyshi and Gorani, where we find the inherited Oblique crossing over to the SAP-pronouns. This is a process obviously distinct from introducing a new Object marker, as discussed in the next section, and demonstrates the fact that old case morphology need not necessarily just fade away. 4.3 Innovated object markers One of the most-discussed aspects of Iranian case systems is the widespread existence of object markers that are etymologically not related to the Inherited Oblique, but derive from other sources. Such markers have a variety of origins, usually adpositions that grammaticalise to object-marking clitics. Two well-known examples are the Old Iranian postposition rādiy, which became the New Persian Object marker −rā (with various colloquial and dialectal variants −o/-ro etc.), and the Old Iranian preposition hacā ‘from’, which grammaticalized to the preposed Object marker a(z) in Pamir languages, and in Vafsi to become the Innovated Object marker on personal pronouns (cf. Table 18). A third, very rare, type is found in Sangesari, where, according to Bossong (1985: 80), the Innovated Object marker goes back to a local postposition dar (see Table 15). Innovated Object markers display, despite their diverse origins, some rather striking commonalities across the languages that use them. The first concerns the diachronic sequence by which Innovated Object markers become established in the case system: If Innovated Object markers enter the system of core case relations, then they do so invariably via the SAP-pronouns, from whence they may be extended to the rest of the nouns. The justification for making this claim is as follows: there are languages that have Innovated Object markers with the SAP-pronouns, but not with nouns (for example, Vafsi). There are also languages with them throughout the grammar (for example Persian). And there are languages with no Innovated Object marker anywhere (for example Zazaki). But what is not attested is a language which uses an Innovated Object marker for nouns, but not with SAP-pronouns. The diachronic sequence just suggested is the simplest explanation for this state Innovated object markers 153 of affairs. A second possible diachronic universal which can be read off the distribution of Innovated Object markers is that in languages with TenseSensitive Alignment, the Innovated Object marker also enters the system in the present tenses, from whence it may spread to past tenses. The same appears to be true of Indo-Aryan (Deo and Sharma 2006: 374), though this needs more thorough investigation. I will begin by investigating the systems of Innovated Object markers found on the SAP-pronouns. There is one additional complication. In those languages where Innovated Object markers are restricted to pronouns (e.g., Vafsi), it turns out that they also extend to third person ‘pronouns’, or demonstratives, or more generally, deictic elements of the third person (and sometimes including interrogative pronouns). Thus it is not strictly speaking correct to refer only to SAP-pronouns in this connection. Nevertheless, for ease of exposition and in the interests of brevity I will restrict my discussion to the SAP-pronouns, because they are always included in the class of forms that take such Innovated Object markers, with the additional mostly unspoken, stipulation that the patterns found there are often extended to various pronominal elements from the third person. 4.3.1 The forms Systems arising from the intrusion of an Innovated Object marker can be quite complex, and in this section, I provide an overview of the attested types. In the following section, I will be discussing the different functions of the forms concerned. In order to grasp the potential interaction of Innovated Object markers and inherited Oblique, it is instructive to begin with a language which quite clearly has retained the inherited Oblique, but augmented it with an Innovated Object marker, −d(e). The language is Sangesari, described in Christensen (1935) and Azami and Windfuhr (1972); I quote mostly from the former, but in the relevant aspects, the two descriptions are very close. The SAP-pronouns are provided in Table 15. As Bossong (1985: 80) stresses, the system shown in Table 15 is highly unusual in the Iranian context. But it is readily explained as one logical outcome of the processes we have been discussing. We find a basic underlay of old suppletive Direct/Oblique distinctions, maintained in the singular but not in the plural (fully in accordance with the universal of case loss on SAP-pronouns expressed as (124) above). 154 Case systems in West Iranian Table 15. Inherited Oblique and Innovated Object on SAP pronouns in Sangesari (transcription slightly simplified, Christensen 1935: 125) S INGULAR 1 P. 2 P. P LURAL D IR . O BL . I NNOV.O BJ . D IR ./O BL . I NNOV.O BJ . a tw o ma ta má-de tá-de ham xā hám-de xā-de Table 16. Inherited Oblique and Innovated Object on SAP pronouns in Chali (Southern Tati, Yar-Shater 1969: 147–152) S INGULAR D IR . 1 P. 2 P. az ta P LURAL O BL . I NNOV.O BJ . D IR ./O BL . I NNOV.O BJ . men cemen eštá amā šömā cemā šömā The basic assymetry between singular and plural is thus the result of a very familiar tendency. It is not actually affected by the addition of the Innovated Object marker, because the latter is added as an additional one-term overlay to all persons and numbers, regardless of the presence or absence of existing distinctions. Another example of a similar is system is the Chali dialect of the Southern Tati dialects, shown in Table 16. Here the three-way distinction is actually restricted to the first person singular. In Chali we find, as predicted above, that the old suppletive Direct/Oblique distinction has been lost entirely, except for the first person singular (closely related dialects still preserve the distinction also on the second person singular). To the remaining forms, generally the old Obliques, additional Innovated Object markers have been added, giving rise to the system shown (the apparent identity of the second person plural forms can be accounted for through secondary assimilation of the Innovated Object marker to the initial sibilant of the pronoun).10 10 The preposed Innovated Object marker can be traced back to a preposition going back to Old Iranian hacā. In in the earlier stages of Chali, it can be reconstructed as having a form something like *(a)tS(a). In the second person plural, the palatal consonant of the clitic Innovated object markers 155 Table 17. Innovated Object on SAP pronouns in Persian S INGULAR 1 P. 2 P. P LURAL O LD O BL . I NNOV.O BJ . O LD O BL . I NNOV.O BJ . man to ma(n)-rā to-rā mā šomā mā-rā šomā-rā Three general principles emerge from these examples: First, Innovated Object markers are used across the board for all SAP-pronouns (and often extend to other deictic elements, see above), unless low-level phonological rules conceal them, as in the second person plural of Chali. Second, the form they attach to is, etymologically, most commonly the old suppletive Oblique form. For some of the forms, this is not readily evident, but in the first person singular, where the distinction between az and man is readily visible, it appears to hold throughout. This is of course what one could expect given (a) the tendency for this form to survive longest anyway; and (b) the origin of the Innovated Object markers themselves (adpositions), which would normally require the Oblique form of their complements. One result of this is that nowhere do we find a first person singular form of the type *az+Innov.Obj. Third, the intrusion of an Innovated Object marker is, in principle, independent, of the loss of the old Direct/Oblique distinction. This final point is necessary to explain cases like Sangesari, or Chali, where both an inherited Oblique and an Innovated Object marker coexist in at least sub-parts of the system. However, it is also certainly true, and probably more common, that the intrusion of the Innovated Object marker occurs chronologically after than the wholesale loss of the inherited Direct/Oblique opposition. This is the case in Persian, where we have the system shown in Table 17. The broad historical sequence that led to the Persian system is well-known: in the earliest Middle Persian period, there was originally a distinction in the first person at least between the old Direct az and the suppletive Oblique man. The former died out, leaving the latter the sole free assimilates completely to the initial /S/ of the pronoun, leaving it apparently unmarked (a similar assimilation is observable with the Vafsi second person plural soan, Stilo 2004b: 227). 156 Case systems in West Iranian Table 18. Innovated Object on SAP-pronouns in Vafsi (Stilo 2004b: 227) S INGULAR 1 P. 2 P. P LURAL O LD D IR . I NNOV.O BJ . O LD D IR (?) I NNOV.O BJ . æz tæ tæmen esdæ awan soan tawan soan pronoun for all three core syntactic functions S, A and O. Later, the Innovated Object marker −rā was used for O-functions (having already moved into IO-functions) giving rise to the system shown in Table 17. Two-term systems, contrasting an old pronoun and a new Innovated Object form, are quite widespread. However, despite some superficial similarities, they are not all the result of the same developments. Consider the forms of the SAP-pronouns in Vafsi, shown in Table 18. While both Vafsi and Persian appear to have straightforward two-term systems, where the marked member is an Innovated Object marker, in Vafsi the opposition is between an old Direct form, æz, and an Innovated Object marker, while in Persian the unmarked member is an inherited Oblique, man. This difference is quite significant, and is faithfully reflected in the different alignment patterns associated with the respective forms in each language. A number of languages of what I term the ‘Tatic-type’ have systems like that of Vafsi, and are treated in Section 4.4 below. To sum up this section, we have seen that Innovated Object markers of various types can be integrated into the case systems of the SAP-pronouns, and in some languages (Sangesari, Persian), extend also to usage with nouns. Innovated Object markers have quite different properties to the Inherited Oblique forms. For one thing, they apply to all the relevant forms in the SAPparadigm, unless obscured by phonological assimilation. Thus Innovated Object markers are regular in a way that the old Direct/Oblique distinction rarely is, because the latter is often restricted to only one or two persons/number combinations. Among the very common two-term systems, two distinct types can be observed. The Persian type, with an opposition between Inherited Oblique and Innovated Object (man/man-rā, and the Tatic-type, with an opposition between an inherited Direct and an Innovated Object marker (az/cemen). These Innovated object markers 157 D IRECT ♣ S I NHERITED O BL . ♣ ♣ ♣ ♣ ♣ ♣ ♣ ❅ ❅ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ❅ ❅ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ❅ ❅ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ❅ ❅ ♣ ❅ ♣ ❅ ♣ A-pres O-past O-pres A-past Figure 1. The archaic system: TSA, no Innovated Object marker (dotted lines indicate semantically/pragmatically determined alternation of Direct/Oblique on direct Objects, DOM) two types differ not only in the etymologies of the forms concerned, but also in the functional distribution of the respective cases. 4.3.2 The functions Having examined some of the inventories of forms, I will now look at the way these forms are deployed in marking core syntactic functions. A good deal of the variation found can be accounted for in terms of two principles, which are illustrated here with fairly straightforward systems. More complex interactions are found in languages of the Tatic-type, which are dealt with in Section 4.4. In representing these systems I have found it convenient to use diagrams with two layers, the top one being the actual case forms themselves, the second representing the different syntactic functions. The mapping of form to functions is indicated by lines linking the two. The first system to look at is the familiar archaic one, where the inherited Direct/Oblique distinction is maintained, and no additional Object markers have entered the system. We still find TSA, with the characteristic use of the cases (Oblique A-past and the Direct O-past). Such systems are found throughout Northern Kurdish and in Zazaki (with the exception of the plural SAP-pronouns). In a number of languages, the system is additionally complicated by the presence of DOM, most commonly in the present tense. When this occurs, we find that the O-pres may also be coded in the Direct case, or an O-past by the 158 Case systems in West Iranian U NMARKED ♣ S I NNOVATED ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ A(past/pres) O BJECT O(past/pres) Figure 2. Persian-type system: no TSA, DOM indicated by dotted lines Oblique case, hence the dotted lines. This happens, for example, in Zazaki in the present tenses, and on nouns in Vafsi in both tenses (pronouns have a different system with an Innovated Object marker, see below). The use of the inherited Oblique with DOM is discussed in detail in Bossong (1985) under the heading “erste etymologische Klasse”, and I take it up in Section 4.6. It is a noteworthy fact that DOM with inherited case markers is most widespread in the present tenses, symbolized by the dotted line between D IRECT and “O-pres” in Fig. 1 (and in some languages, e.g., Zazaki, is restricted to this environment). Leaving aside many complications for a moment, the most important single feature of the archaic type of system is the following: the Inherited Oblique is associated with the A-past function. We can formulate this as the first principle of case functions in Iranian, although we will have cause to modify (133) somewhat later: (133) First Principle (initial formulation): If an inherited Direct/Oblique distinction is maintained anywhere in the grammar, then the functions of the Inherited Oblique case grammar will always include A-past. The second system to be examined is that of Persian. Here we have complete loss of the inherited Direct/Oblique opposition, and the secondary intrusion of an Innovated Object marker. Such systems generally work as in in Fig. 2. With the wholesale loss of the old Direct/Oblique distinction, TenseSensitive Alignment also disappears, and case-marking follows a uniform accusative alignment throughout. This system illustrates the second principle of case functions in Iranian: Innovated object markers 159 D IRECT ♣ I NNOV.O BJ . ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ S A-pres O(past/pres) O LD O BLIQUE A-past Figure 3. Three-term system, (dotted line indicates DOM) (134) Second Principle (initial formulation): An Innovated Object marker is always linked to the Object function, and is always mediated by DOM. The third system illustrates the combination of the first two: an inherited Direct/Oblique distinction augmented by an Innovated Object marker, giving rise to a three-term system. Systems of this kind are found on nouns and pronouns in Sangesari, and with a slight variation, on pronouns in Chali (Southern Tati, Yar-Shater 1969).11 The functions of the respective cases largely reflect (133) and (134), and are shown in Fig. 3. It will be observed that the Persian-type and the three-term systems are identical except for the fact that in the latter, the old Oblique is retained, albeit in a highly restricted function: solely A-past, entirely in accordance with the principle discussed in (133). The functions of the different cases can be illustrated for Sangesari pronouns as follows (for O in the present tense no example was available, but according to the description, it must also take the Innovated Object marker): 11 Other related dialects that are analyzed as having three-term pronoun systems (Shāhrud, Khalkhāl (Yarshater 1959), Kajal (Yarshater 1960) and Xo’ini Yarshater 2003) do not actually belong here, because the Innovated Object marker is only used as a Possessive, hence does not mark a core syntactic function as defined here. 160 Case systems in West Iranian (135) má tá-de bedía 1 S:OBL 2 S-I NN O BJ see:PST ‘I(A-past) saw you(O-past).’ (Christensen 1935: 126) (136) a b@rún razi r É mišúji 1 S:DIR outside garden:OBL way go:PST:1 S ‘I(S) was passing the garden [. . .].’ (Christensen 1935: 135) (137) má-de bedía n É 3 S:OBL 1 S-I NN O BJ see:PST ‘He saw me(O-past).’ (Christensen 1935: 126) e e The systems found in Persian and in Sangesari are in fact, despite superficial differences, very close. It is solely in the A-past that they differ. If the old Oblique form were to disappear, then we would have a straightforward Persian-type system, without TSA and Object-marking through an Innovated Object marker. Such a development appears to be quite likely, given the very low functional load of the old Oblique in the system. A somewhat different and more complex, set of changes must be assumed for a second group of languages, which are referred to here, for want of a better term, as the “Tatic-type”. These will be treated in the next section in some detail. 4.4 The Tatic-type languages The precise genetic affiliation of the languages covered in this section remains controversial. They are certainly West Iranian, but which of them can be meaningfully subgrouped together is unclear, and how such a subgroup would be related to other West Iranian subgroups is simply not established currently (see Stilo (1981) for discussion of these issues). Nevertheless, in those aspects of grammar relevant in the present connection, they show a number of striking commonalities, many of which set them off from the other West Iranian languages. The languages analyzed in this section are: The Southern Tati dialects (9 dialects) as described in Yar-Shater (1969). The Tati dialect of T.ārom (Yarshater 1970) The dialect of Kajal (Yarshater 1960) The Tatic-type languages 161 The Xo’ini dialect (Yarshater 2003) Vafsi (Stilo forthc. b, Stilo 2004b) They are also an excellent target for a comparative study for practical reasons: Thanks to the work of Ehsan Yarshater, short but sound descriptions are available for a large number of these languages and dialects. I was also extremely fortunate in having full access to Don Stilo’s ongoing work on Vafsi syntax (Stilo forthc. b), probably the most detailed description of the syntax of any minority West Iranian language. Don was also unfailingly generous in sharing his vast knowledge of these languages, and assisted me in more questions of interpretation than I can meaningfully acknowledge here. The Tatic-type languages all share certain morpho-syntactic commonalities. It will be useful to establish these from the outset to avoid unnecessary repetition. The important ones in the present connection are the following: 1. Retention of inherited Direct/Oblique distinction: Nouns (but not necessarily pronouns) have retained the old Direct/Oblique distinction, in about half the languages still differentiating the Oblique for Gender. There is some TSA, with the Past Transitive Constructions showing a variety of non-accusative alignments. However, in five of the nine Southern Tati dialects discussed in Yar-Shater (1969), the Direct case is also used for the A-past, resulting in uniform accusative alignment in case marking. A further complication is the use of DOM in both tenses, with numerous dialect-specific complications. 2. The inherited Oblique codes the Possessor with nouns Adnominal possessors are always preposed. For nouns, they are in the inherited Oblique case, as in Ahmad-e galbar ‘Ahmed’s house-door’ (Tākestāni, Yar-Shater 1969: 103). SAP-pronouns pattern somewhat differently. 3. Cross-referencing of A-past with a clitic pronoun All languages make use of this strategy, though dialectal differences are rife in degree of obligatoriness, and in positioning of the clitic. The following example, from West Iranian Gazi, shows clitic-cross referencing of the A in the Past Transitive Constructions hæmé=ž, and accusative case marking with −ra: (138) kelíl=o dæmané-ra hæmé=ž zunašte-biye Kalila=and Dimna-ACC all=3 S know:PST:PLUP ‘(He) knew Kalila va Dimna, all of it.’ (Stilo forthc. a) 162 Case systems in West Iranian 4. Case-split between SAP-pronouns and nouns. All dialects have an Innovated Object marker, derived from a grammaticalized preposition, used solely with pronouns. Its functions always include Possessor, for example čemā pül ‘our(1 PL:I NN O BJ) money’ (Xo’ini dialect, Yarshater 2003: 170), estæ bera ‘your(2 S:I NN O BJ) brother’ (Vafsi); caman dada ‘my(1 S:I NN O BJ) father’ (Kajal, Yarshater 1960: 281). Even where a language has an inherited Oblique form of the pronoun, it is never used for Possessor, whereas the inherited Oblique is the normal case for Possessor with the nouns (see first point above). A subset of the languages also uses the Innovated Object-forms of the pronouns in core syntactic functions, hence their relevance here. The above four points sum up the relevant commonalities across all the languages considered in this section. In what follows, I will be concentrating on the case systems with the pronouns, because it is only here that we find Innovated Object markers entering the system of core syntactic functions, hence giving rise to the most interesting interactions. For ease of exposition, I take the first person singular as representative of the pronoun system, though this is clearly an oversimplification. It is, however, justifiable on the grounds that the old Oblique/Direct distinction is most clearly recognizable on these forms, and also on the grounds that these forms invariably show the highest number of case distinctions. To what extent the findings from these forms can be generalized to the pronominal systems as a whole is a question that requires more detailed study. But even if we reject such a generalization, a comparative study of the syntax of these forms nevertheless yields valuable results and provides pointers and hypotheses for future research. An overview of the actual forms is provided in Table 19, in which the first nine languages all belong to Yarshater’s ‘Southern Tati’ (Yar-Shater 1969). I have slightly adapted the orthography of the Vafsi examples to facilitate comparison with the others. The broad similarities behind the inventory of forms are readily apparent. The primary distinction is that many of the languages have lost their inherited Oblique form, while some have retained it. Diachronically, the development was presumably as follows: Stage 1: An inherited binary, suppletive Direct/Oblique opposition The Tatic-type languages 163 Table 19. First person singular pronouns in Tatic-type languages L ANGUAGE C HALI T ĀKEST ĀNI E SHTEHARDI X I ĀRAJI E BR ĀHIM ĀB ĀDI S AGZ - ĀB ĀDI D ĀNESF ĀNI E SFARVARINI X OZNINI S H ĀHRUDI T. ĀROMI K AJALI X O ’ INI VAFSI O LD D IR . O LD O BL . I NNOV.O BJ a(z) az az a a a(z) az az az az az az/azi az az men − − − − − − − − man men − man − ceme(n) cemé cemen cemé ceme(n) cemé cemé cemen cemen ceman (1)ačem;(2)adem cama(n) čeman tamen Stage 2: Intrusion of an Innovated Object marker, yielding the three-way distinction still found, for example, in Chali.12 Stage 3: Loss of the old Oblique, yielding the binary opposition now found in most of the languages. Although the forms themselves show a high degree of unity, the alignment systems coded by these forms differ. In what follows I will analyze the variation found and suggest some explanations in terms of general developmental processes. Throughout the discussion I will use the abbreviations AZ, MEN and CEMEN as symbols representing in all languages the old Direct, old Oblique, and Innovated Object markers respectively. To begin with, let us consider the system of Stage 2 above. Here the old AZ / MEN -distinction is preserved, while CEMEN is also present. Crucially, the 12 The dialects of T.ārom have undergone a unique innovation in that they introduced not one, but two Innovated Object markers to the old system, giving rise to a four-way distinction on the first person singular, see below for discussion. 164 Case systems in West Iranian AZ S A-pres (Shāhrud only?) ♣♣ cf. (141) ❅ ❘ ♣♣♣ ❅ ♣ ♣♣ ♣♣ O-pres O-past MEN ♣ ♣♣ ♣♣ ♣ ♣ A-past CEMEN Possessor Figure 4. Archaic Tatic system latter is not used for S, A or O, but solely to encode Possessors. I assume that this is the original function of the CEMEN-form because it is found with all languages of this type, suggesting that it is primary here, and then spread (in some languages) to include core syntactic functions. Systems of this kind are attested in Xo’ini, Kajal and Shāhrud. Schematically, it can be displayed as in Fig. 4. While the cases used for S and A are beyond doubt, the data for Xo’ini and Kajali contain no examples of SAP-pronouns in object function, hence the O-pres function cannot reliably be linked to a particular form. The use of AZ for S and A-pres is beyond question, as is the use of MEN for A-past. Two examples from Xo’ini may suffice for illustration: (139) az bar-e dabastarim 1 S door-OBL shut:PRES:1 S ‘I am closing the door’ (Yarshater 2003: 174) [A-pres] (140) man ow=em ente 1 S:OBL water=1 S drink:PST ‘I drank water’ (Yarshater 2003: 170) [A-past] In Shāhrudi, the MEN form is used for the O-past: (141) ave men=iš vinde 3 S:OBL 1 S:OBL see:PST ‘[. . .] he saw me’ (Yarshater 1959: 58) The result is a double-Oblique construction of the type familiar from some of the dialects of Northern Kurdish (see Section 5.4.1). It seems reasonably The Tatic-type languages 165 AZ S A-pres MEN CEMEN ✘ ✘✏ ✘ ✘✏✏ ✘ ✏ ✘ ✏ ✘✏ ✘✏ ✘✏ ✘✘✏ ✘ ✘ ✏ ✘ ✘✘ ✏✏✏ ✏ ✘✘✘ O-pres O-past A-past Figure 5. Intermediate Tatic-type system (T.ārom uses two distinct and O- present respectively, see Section 4.4.1) Possessor CEMEN -forms for O-past certain that the CEMEN-form is not used for objects, as its function is quite succintly described, but no mention of an object-function is made. While the coding of objects is not fully accounted for, the archaic system shown in Fig. 4 illustrates robustly two features: the linking of the MEN-form to the A-past function, and the restriction of the CEMEN-form to the Possessor role. The next type of system encountered is similar to the archaic one, in that the AZ / MEN-distinction is retained, but in this type, CEMEN has begun to encroach into the core system of syntactic functions. This system, illustrated in Fig. 5, is found with minor variations, in two languages: Chali, and the T.ārom dialects. Note how this system leaves the old Oblique functionally ‘stranded’: its sole remaining function is the A-past. Presumably, once an old Oblique is reduced to marking only A-past, its frequency of occurrence will drop quite considerably. Note further that an A-past is often additionally cross-referenced by a pronominal clitic, so a full pronoun is often omitted in this function. In view of these facts, it would appear a quite natural development for the MEN-form to be lost from systems such as Fig. 5, and indeed, in most of the Tatic-type languages, it is no longer present. The resulting two-term systems thus consist of an opposition between AZ and CEMEN. The question that arises is then, which of these two takes over the A-past function which, following the loss of MEN, is left orphaned? Theoretically, there are two possibilities: CEMEN expands its range to include the A-past function, or AZ takes it on. As it turns out, both options are attested, showing once again how difficult it is to make firm predictions on 166 Case systems in West Iranian CEMEN ✟ ❅ (Vafsi “OSV-ergative”) ✟✟ ❅ ❅ ❅ ✟✟ ❅ ✟ ❅✠ ✟ ❅ ✟ ❅ ✟ ❅ ❅ ✟ AZ S A-pres O-past O-pres A-past Possessor Figure 6. AZ / CEMEN-Type 1 the directions alignment changes may follow. The first option, where CEMEN takes on the A-past function, retains Tense-Sensitive Alignment. This type is attested in Vafsi, and in Eshtehardi (Southern Tati). The basic system is shown in Fig. 6. This system retains elements of the archaic one, where AZ is still associated with an O-past, but the link is tenuous. In Vafsi, it is restricted to just one particular, textually very rare construction, termed by Stilo the “OSVErgative”-construction: (142) æz æhmæd-i yédieym 1 S:DIR Ahmed-OBL see:PST:1 S ‘Ahmed saw me.’ This construction, with O-past in the AZ-form, is to my knowledge, unique in the entirety of West Iranian, being the sole instance of consistent OAV-word order. It occurs, however, very rarely in texts but was consistently elicited under certain conditions (Don Stilo, p. c.). The precise pragmatic function of this word-order variant remains uncertain. An A-past, on the other hand, must always be in the CEMEN-form, and with the more common AOV-word order, an O-past is also in the CEMEN form, as in the following example: (143) tæmen estæ ærkošdæy 1 S:I NN O BJ 2 S:I NN O BJ kill:SUBJ:PST:2 S ‘I would have killed you.’ Thus with Past Transitive Constructions, Vafsi makes widespread use of a double-Oblique construction with pronouns, using its Innovated Object-forms The Tatic-type languages 167 for both A-past and O-past. As for Eshtehardi, it exhibits like Vafsi the characteristic of using the CEMEN-form for the A-past: (144) cemen vafra darund 1 S:I NN O BJ snow sweep:PST:3 S ‘I swept the snow.’ (Yar-Shater 1969: 152) But apparently, like in the Vafsi OAV-construction in (142), in Eshtehardi too an O-past is also in the old Direct AZ form. Yarshater provides only a single example for this construction: (145) az=eš bërëstāym 1 S:DIR=3 S send:PST:1 S ‘(He) sent me.’ (Yar-Shater 1969: 155) These systems are extremely unusual in the context of Iranian languages: As repeatedly mentioned, innovated case-markers are invariably Object-markers, and indeed, they are in both Vafsi and Eshtehardi. But in these two languages, they are also markers of an A-past. Vafsi and Eshtehardi have thus transposed their inherited ergative traits onto new morphology, an otherwise exceedingly rare phenomenon. No other language known to me, with the possible exception of Balochi,13 uses an Innovated Object marker to code the A-past. The 13 The interpretation of the case system in Balochi remains a controversial issue. According to Korn (forthc. a), the case used for an A-past is an innovated Object marker, essentially a phonologically reduced version of rā. This is also the view of Bossong (1985). A different view is that it is actually a reflex of an inherited case marker (Axonov 2006: 70), while the innovated Object marker is restricted to the pronouns. I tend towards the latter view, partly because Korn’s analysis leaves Balochi as the sole counter-example against the comparative evidence amassed here for the ways in which case systems in Iranian typically develop (see also Haig (forthc. b) for discussion), but also because the arguments for the Innovated Object marker analysis are not entirely convincing. Along with evidence from historical phonology, Korn’s main argument is that the case marker on Balochi nouns occurs outside of the indefiniteness suffix, hence is more likely a later addition. However, the inherited Oblique case also occurs outside the indefiniteness suffix in Northern Kurdish (cf. (126) above), so this argument is not particularly telling. However, the reader should consult Korn’s work for the full details of her analysis. It is also possible that the case marker in question is a hybrid, a blending of an inherited case-suffix with a (possibly) borrowed case marker, so that an unequivocal solution to its etymology is simply not possible. 168 Case systems in West Iranian CEMEN AZ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ S A O Possessor Figure 7. AZ / CEMEN-Type 2 point is confirmed by Bossong (1985: 118), with specific reference to −rātype Innovated Object markers, but it seems to hold for others too. Given the rarity of this development, it is worthwhile seeking for special conditions which may have contributed towards it. The first thing to note is that, according to the development sketched above, what I have termed the “Innovated Object” marker is not, in the context of these languages, really an “Object” marker in origin: it is a preposed Possessor-marker. It will be recalled from Chapter 2 that Possessors and Agents share a number of properties, so this extension from Possessor to the A-past function may actually replicate the process that led to the emergence of ergativity in the first place. Thus one of the special conditions that contribute to the strange developments in Vafsi and Eshtehardi may have been the presence of a preposed Possessor for the personal pronouns. However, while this may be a necessary condition, it is not a sufficient one, as all languages of the Tatic-type had the same type of preposed possessors, but most of them do not extend it to the A-past. Instead, we get the second type of AZ / CEMEN-system, to which we now turn. This type is much more straightforward, and conforms more closely with our expectations: the A-past function, left stranded after the loss of MEN, is taken over by AZ. The result is the complete abandonment of Tense-Sensitive Alignment on the pronouns, with simple accusative alignment in all tenses. The system is shown schematically in Fig. 7. The system of Fig. 7 is found on all the Southern Tati dialects of Yar-Shater (1969), with the exception of Eshtehardi (just discussed), and Chali (see (131) and the accompanying discussion). Consider the following examples from the Southern Tati dialects (a clear example with an overt A-pres pronoun was not available, but it would The Tatic-type languages 169 undoubtedly be in the AZ-form). The first three are from the Esfarvarini, the last one is from Tākestāni: (146) az menome 1 S:DIR come:NEG:PRES ‘I don’t come.’ (Yar-Shater 1969: 148) [S] (147) az nidiama 1 S:DIR NEG:see:PST ‘I haven’t seen (it).’ (Yar-Shater 1969: 149) [A-past] (148) cemen undi 1 S:I NN O BJ give:IMP ‘Give me (to someone).’ (Yar-Shater 1969: 151) [O-pres] (149) ceme=šun jā feri ānendā 1 S:I NN O BJ=3 PL DEM:OBL boy:OBL give:NEG:PST ‘(They) did not give me to that boy.’ (Yar-Shater 1969: 151) [O-past] Following the loss of the old Oblique pronouns, the A-past function is absorbed, as it were, by the AZ-form, leading to a unified coding of S and A in all tenses. The tendency for SAP-pronouns to realign towards accusativity is in fact very widespread and will be discussed in Section 4.6.1. But as we have seen in connection with Fig. 6, it is not an absolutely necessary development. Under certain conditions, we find a reversal of the otherwise general trend for Innovated Object markers to be restricted to marking Objects. Here we find an Innovated Object marker takes on the A-past function, thus preserving ergativity, but with new morphology. 4.4.1 The dialects of T.ārom At this point it is worth briefly investigating what is probably the most complex case system for pronouns in West Iranian, that of the Tati dialects of T.ārom (Yarshater 1970). Basically, I consider them to be a sub-variant of the intermediate type discussed in connection with Fig. 5, where the Innovated Object marker intrudes onto the Object functions in both tenses. However, in these dialects, the first person singular pronoun has four distinct cases for 170 Case systems in West Iranian Table 20. SAP-pronouns in the Tati dialects of T.ārom (Yarshater 1970: 460–461) S INGULAR D IR . 1 P. 2 P. az ta O BL . I N O BJ -1 men ačem P LURAL ešta I N O BJ -2 D IR ./O BL . I N O BJ . adem amā šemā ačemā ašemā basic syntactic functions, as well as an additional one for possessive. Table 20 gives the relevant forms. We find in the first person the familiar archaic system, with retention of the old Direct/Oblique distinction. To this is added an Innovated Object form, acem, which presumably arises from the same process as the regular Innovated Object forms in closely related Tatic-type languages, the grammaticalization of a preposition going back to Old Iranian hacā. In the T.ārom dialects, however, we seem to have something like *(h)acā+men yielding *acemen>ačem, with truncation of the final syllable. The origin of the second Innovated Object form, adem, is, however, obscure (it may be linked to the postposition ade ‘from’, but this is speculative). Functionally, the two Innovated Object forms are both Object markers, as we would expect on the second principle of case marking (134), but one is reserved for O-past, while the other is reserved for O-pres. An example of the Innovated Object-1, expressing an O-past, is the following: (150) ešta berā ačem=eš bezze 2 S:I NN O BJ brother 1 S:I NN O BJ-1=3 S hit:PST ‘Your brother hit me (O-past).’ (Yarshater 1970: 460) The Tatic-type languages 171 The inherited Oblique, on the other hand, is used solely for the A-past: (151) men i guravi=m vevat 1 S:OBL one sock=1 S weave:PST ‘I (A-past) wove a sock.’ (Yarshater 1970: 460) Regrettably, Yarshater provides no examples for the Innovated Object-2. But despite the addition of a second Innovated Object marker, the T.ārom system is essentially identical to that of Chali: the Innovated Object marker enters the system as an Object marker, leaving the sole function of the inherited Oblique that of marking the A-past. 4.4.2 Summary of the Tatic-type Languages of the Tatic-type show a particularly interesting interaction of the different processes that have been discussed in the course of this chapter. They combine both the relatively archaic inherited Oblique case, often gender-differentiated, with relatively innovative SAP-pronoun systems, where in some cases the old Direct/Oblique distinction has been completely lost. In general, they confirm most of the principles that have been stated for the case systems of West Iranian: Case on SAP-pronouns are never fewer than on nouns; the inherited suppletive Direct/Oblique distinction on SAPpronouns is retained longest on the first person singular; the inherited Oblique is always associated with the A-past function. However, it must be noted that a number of the Southern Tati dialects have almost completely lost any trace of ergativity in case marking, not only on pronouns (as in the languages with the AZ / CEMEN-Type 2), but also on nouns. How alignment in nouns and pronouns interacts will be discussed in Section 4.6. The languages Vafsi and Eshtehardi display a development which appears to be unique in West Iranian: an Innovated Object marker (the CEMEN-form) is regularly used to code an A-past. Note that these forms are also used for Objects in the languages concerned, thus complying with the principle formulated in (134). But the additional use of these innovated Object markers as A-past markers is virtually unparalleled. I would like to suggest that such systems arise only under the condition that the Innovated Object marker in 172 Case systems in West Iranian Table 21. The birth of secondary case-suppletion on pronouns 1. I NHERITED SUPPLETIVE : 2. M IXED SUPPL ./ AFFIXAL : 3. S ECONDARY SUPPLETIVE D IRECT O BLIQUE I NN O BJ . az az az men men — — cemen cemen question was used to encode a preposed Possessor, but this remains speculative. Finally, the Tatic-type illustrate rather neatly how suppletive oppositions in morphology may arise in the first place. In the first person singular, there was an original suppletive case opposition. In the languages of the Tatic-type, this existing two-way distinction was augmented by a transparently affixal Innovated Object marker, leading to a mixed three-term system. Finally, in many languages, one member of the old suppletive distinction disappeared, leaving what is, from the point of view of the naive speaker without knowledge of related dialects, an unanalysable suppletive case opposition, for example between az and cemen. The three stages are shown schematically in Table 21. 4.5 Explanations for change What kind of more general principles lie behind the changes in the casemarking patterns in West Iranian, and how are they related to the broader question of the function of case marking? In the following sections, I will be discussing two potential contributing factors and assessing their relative merits against the data: maintaining the discriminatory function of case, and maintaining cross-system harmony. My general conclusion is that while both appear to be operative, maintaining the discriminatory function of case only really exerts any pressure when changes are viewed over a very long period. In explaining the minor sub-changes, it appears that maintaining cross-system harmony can explain the data more adequately. In Section 4.7, I attempt to reconcile the partially conflicting factors by formulating an account in terms of a set of ranked constraints on change, probably the most developed theory of changes in Iranian case systems to date. Explanations for change 173 Table 22. Tripartite case system F UNCTION C ASE S 1 A 2 O 3 4.5.1 The discriminatory function of case Case-marking in the broader sense serves the purpose of identifying certain semantic roles, for example Instruments, Locations, Goals etc. But with the core arguments S, A and O, there is no straightforward mapping of morphological form to semantic roles, because these cases may express a broad range of different semantic roles.14 Core arguments are required by the predicate, so their semantics can often be read off the predicate. However, with A and O occurring in the same clause, a potential problem arises: how does one know which NP refers to A and which to O? According to one very influential view, this is where case marking steps in. Put most simply, case marking of core NPs primarily serves the purposes of ensuring that A and O be correctly interpreted (see for example Payne (1997: 140) for a fairly typical statement to this effect). Quite apart from its common-sense appeal, this view receives support from the relative frequencies of different case systems in the languages of the world, as pointed out by Comrie (1989: 124–127). From a discriminatory perspective on case marking, then, the primary demand on any system is that A and O are consistently distinguished. S, on the other hand, need not necessarily be distinguished from either, as S will never actually occur in the same clause as A or O. Theoretically, we could of course mark each with a dedicated case marker. Such a system, referred to as a tripartite system, is shown in Table 22, where different numbers indicate non-equivalent, overt case markers. Systems of this type are very rare, but attested, for example in sub-systems of Iranian languages. In the past tense of Sangesari, for example, S, A and O have three distinct markers, cf. Fig. 3. Such a system obviously succeeds in differentiating different syntactic functions, yet cross-linguistically, it is ex14 Of course semantic and pragmatic features may be incorporated into the case marking of S, A and O, for example when factor such as volitionality, indefiniteness, or animacy impact on case marking (e.g., in Hindi). These points are ignored for the moment. 174 Case systems in West Iranian Table 23. Accusative (S=A) case system F UNCTION C ASE S 0 A 0 O 1 Table 24. Ergative (S=O) case system F UNCTION C ASE S 0 A 1 O 0 tremely rare. The reason is presumably that case systems are not only calibrated towards achieving the discriminatory function, but to doing so at the minimal cost. If all that is necessary is to distinguish A from O, then an additional case reserved for S is superfluous; it could be marked in the same manner as either A or O, hence reducing the overall number of distinct forms from three to two. Thus the discriminatory principle needs to be modified by considerations of economy. This latter factor is presumably what is behind the extensive distribution of the accusative and ergative systems in the world’s languages. The two are illustrated Tables 23 and 24 respectively (in the tables, ‘0’ indicates a formally unmarked, or zero case form). Of all the theoretically possible combinations of case markers, these two succeed in discriminating A from O with the most economical means possible. Thus both in terms of discriminating between A and O, and avoiding redundancy, the Accusative and Ergative systems achieve the same goals. This may, according to Comrie (1989), account for the statistical preference for these types of case systems across the languages of the world: they achieve the maximum necessary in discriminatory function with a minimum of means. A number of authors consider that maintaining the discriminatory function of case marking is a powerful force for morphosyntactic change. Hence languages which, through phonological and morphological change, lose case distinctions, are obliged to develop alternative discriminatory strategies. There are several issues at stake here. One of the commonest claims made in this context is the correlation between case-marking and constituent order, going back at least as far back as Greenberg (1966: Universal 41). The generally observable tendency is that OV word order correlates both with the exis- Explanations for change 175 tence of case marking distinguishing A and O, and with a fairly high degree of word order flexibility. When case marking is lacking, however, we find a fairly rigid AVO word order. The lack of overt case marking is thus compensated for by a fixed association of positions relative to the verb. Perhaps the most-discussed instance of this type of compensatory strategy is the shift in word order from OV to VO in Germanic. According to Lightfoot (1999) and in his earlier publications, loss of case marking was instrumental in triggering a parameter shift from OV to VO, the latter reinstating the discriminatory function by positioning A and O on opposite sides of the verb. However, more detailed investigations of the English data have cast serious doubts on these conclusions (see for example Allen (1995), summing up extensive earlier research). On the basis of a sample of Norwegian texts Sundquist (2006) also shows that VO order was actually established long before the loss of case in Norwegian, suggesting that other factors were responsible for the word order change. Furthermore, Icelandic has maintained a case-system essentially similar to that of Old English, yet it has also shifted to a VO word order. Thus the shift from OV to VO cannot be accounted for solely in terms of a simple trade-off in discriminatory strategies (case for VO-word order). Finally, it is worth taking a look at the most comprehensive typological survey available to date on the correlation between AOV and case marking, that of the online Universals Archive, based at the Department of Linguistics (Sprachwissenschaft) at the University of Konstanz. Under Universal No. 8, the correlation is formulated as follows: (152) If in a language the verb follows the nominal subject and the nominal object as the dominant order, the language almost always has a case system. In the comments to this proposed universal, more than 50 genetically diverse languages are listed that have AOV order, yet do not distinguish A from O with case marking. Thus (152) is at best a tendency, not an absolute universal. And in fact, this tendency itself may be simply a derivative effect of other universal tendencies listed in the same source: the tendency for AOV languages to have agglutinative morphology (Universal 11), which is generally suffixal (Universal 1589). Furthermore, we know that case markers are more generally suffixes (Universal 170). Therefore, a language with an above-average probability of possessing suffixal, agglutinative morphology 176 Case systems in West Iranian is also very likely to include case markers among its morphology. Although calculating the combined effects of these various factors has yet to be carried out, I contend that there are more general principles at work behind the tendency for AOV languages to distinguish A from O. Attributing it solely to the workings of the discriminatory function of case marking is undoubtedly an oversimplification. Returning to Iranian, we can note that many languages of the Central and Southern Groups of Kurdish have entirely lost case (on both nouns and pronouns), yet there is absolutely no evidence of a shift away from the generally valid OV word order of West Iranian towards a VO type. Likewise, across centuries of attested Middle Persian, no trace of nominal case marking is found, yet this was not accompanied by a shift in word order. In general, word order in Iranian appears to have been remarkably stable OV for at least 2500 years, regardless of the massive changes in the case system over that period. In fact, although this has never been thoroughly investigated, the word order of the caseless Iranian languages does not appear to differ significantly from those with case. For example, in the Bayray dialect of Southern Kurdish, A and O are not distinguished through case marking: (153) dẅataga gole.(g) čıni ¯ girl flower picked ‘la fille a cuielli une fleur.’ (Fattah 2000: 672) In fact one can legitimately ask why AOV should be any worse an indicator of case than AVO. Comrie (1989: 214) suggests that AOV is often linked to a freer word order. In such a language, OAV would be a “frequent alternative word order for purposes of topicalizing the object or focusing the subject”, hence leading to a potential confusion of A and O. Therefore, case is required to disambiguate in such cases. But in the generally AOV Iranian languages, a frequent alternative is actually not OAV, but OVA, as in the following variant of (153): (154) gole.(g) čıni dẅataga ¯ flower picked girl ‘C’est une fleur qu’a cuelli la fille.’ (Fattah 2000: 672) This type of OVA-word order is also a very common variant in Persian, and indeed quite common in Turkish too, two languages with case-marked ob- Explanations for change 177 jects, suggesting that the type of pragmatically-driven word order variation is not particularly sensitive to the presence or absence of case-marking. Thus the argument from word-order alternatives in AOV languages as an explanation for their apparent predilection for case marking remains rather spurious. Whatever the details of word order variation may be (and this is yet to be empirically investigated for Iranian), the blunt facts from the dominant word orders are uncontroversial: case was lost in many Iranian languages, often over centuries, yet AOV has been, as far as we can ascertain, the dominant order for at least 2500 years. There is no evidence whatsoever for a correlation between case marking and word order change in Iranian. Let us leave word order and return to changes in the case-marking systems themselves. In his pioneering study Bossong assumes that maintaining, or restoring, the discriminatory function of the case system is the primary motor behind the emergence of Innovated Object markers (Bossong 1985: 117), a view also implicit in the discussions of Schulze (2000), and regularly repeated in more general works.15 Yet individual changes are often difficult to account for in these terms. First, there are languages where an Innovated Object marker is added to an existing Direct/Oblique opposition, thus violating the maxime of economy. Second, the caseless West Iranian languages blear blatant witness to the fact that discriminating A from O through nominal case is simply not necessary. The languages of the Central Group of Kurdish, for example, show no signs whatsoever of adopting an object marker; they are perfectly stable systems that get by without any case marking distinguishing A from O, and have probably done so for centuries. Finally, and most damaging, there is the widespread existence of double-Oblique constructions in past tenses (see Section 5.4.1 for a case study). In such systems, A and O are both Oblique-marked. Nor can the double-Oblique system be explained through phonological attrition of existing case markers, as the change involves the addition of a case marker to the previously unmarked Direct. Generally, even when the existence of such systems is acknowledged, they are dismissed as ‘unstable’, or transitory phases. But such statements are highly problematic. First, the diachronic stability of such systems is, in 15 Arguments against the primacy of the discriminatory function in case marking in general are found in Næss (2007), and with specific reference to Iranian in Arkadiev (2004). 178 Case systems in West Iranian many cases, simply unknown, as the languages have not been recorded to any time depth. And second, the sheer frequency of occurrence of the doubleOblique across different Iranian languages cries out for explanation, whatever its stability in the respective languages may be. At the very least, the existence of double-Obliques anywhere must be squarely faced as a serious challenge to the primacy of the discriminatory function in the evolution of case-systems. The more general question raised by these well-attested changes is: why should a system of ergative alignment change at all? It achieves the discrimination of A and O with the minimal means, and is therefore presumably an optimal case system. Yet as we have seen, the ergative constructions tend to depart from their lean efficiency, giving rise to numerous intermediate forms which are less “efficient” in terms of discrimination and economy. The generally implicit answer to this question is that change is initially triggered by “blind” phonological change, which erodes existing morphological markers. Morphosyntax then reacts by shoring up, or compensating for, the distinctions thus lost. The problem with this account is that if functional pressure, such as discriminating between A and O, is indeed a powerful factor in restructuring morphosyntax, why does it not block the loss of case in the first place? If it can cause new morphology to ‘grow back’, or trigger a shift in word order, why can it not prevent morphology from disappearing? Whatever status the pressure to maintain the discriminatory function of case may have in shaping morphosyntactic change, it is clearly subservient to other demands. Of course the view that nominal case is the sole, or even the main strategy for discriminating core arguments is obviously grossly oversimplified. In a language with agreement morphology, at least one of the core arguments can normally be identified. Most West Iranian languages have agreement with at least one argument, and if the pronominal clitics are considered agreement, then in some cases they cross-reference two arguments. And of course in most instances, knowledge of the context and simply world knowledge suffice to sort out who did what. To take up (154) above: we know that girls pick flowers, rather than the other way round, so disambiguating the sentence is hardly problematic, whatever its case and word-order properties may be. The paradox we are faced with when assessing the importance of the discriminatory function is this: from the perspective of very long-term develop- Case and animacy 179 ments (in the order of millennia), many of the changes attested can indeed be linked to a combination of discriminatory function and economy. Likewise, these factors offer the most cogent explanation for the relative frequencies of attested case systems in the world languages, as Comrie (1989) notes. Thus on a global level, and over extended periods of time, the combined effects of the discriminatory function with considerations of economy do indeed work to tilt the distribution of case systems strongly in favour of certain types. On the other hand, certain individual changes, for example the loss of case marking, or the emergence of double-Oblique systems, actually run counter to these expectations. Such systems constitute valid grammars, acquired and deployed by speakers over many generations. If functional factors are indeed behind the changes, and these are ultimately rooted in the communicative needs of actual speakers, how is it that apparently functionally-motivated changes may take many generations to work themselves out, and involve putatively less functional intermediate stages? Discussion of this issue is deferred to Section 7.4. For the time being I would like to note that while the discriminatory function of case marking is clearly important, it fails to account for a number of attested changes. For these, somewhat different explanations are required. 4.6 Case and animacy The way case is deployed on core arguments is not solely determined by predicate-argument relations. It may also be co-determined by the inherent semantics of the NP concerned, and by its pragmatic status in the discourse context. These latter factors are treated here under the – not entirely fitting – umbrella term of animacy (see below). Any account of case systems in Iranian, and of their evolution, must take into account animacy effects in case marking, as they are of considerable import. There are two main areas where animacy impacts on case systems: the first is what are called animacy-based, or NP-semantics-based alignment splits. The second is Differential Object Marking (DOM). The latter has been investigated in detail for Iranian by Bossong (1985), and was discussed in Section 4.3 in connection with Innovated Object markers. The presence of the former in Iranian languages, however, has received scarcely any attention, and is therefore discussed in more 180 Case systems in West Iranian detail here.16 But first, I will briefly introduce those aspects of animacy which appear to be relevant in the present context. Although many linguists have pointed out that the term ‘animacy’ is somewhat misleading (Payne 1997: 150), I continue to use it in the broad sense of Comrie (1989: 185–200). The cross-linguistic investigation of a number of phenomena has revealed that important generalizations can be formulated through reference to an animacy hierarchy. The hierarchy, initially formulated by Silverstein (1976), has been repeated with minor modifications and under various labels in countless publications since. A recent version of is given in (155). The numbers “1,2,3” refer to first, second and third person pronouns. An additional distinction of Definite/Indefinite has been added, which potentially applies to all items except deictic and SAP-pronouns, which are considered to be inherently definite.17 (155) The Animacy Hierarchy (Corbett 2000: 91) 1 > 2 > 3 > kin > human > animate > inanimate D EFINITE > I NDEFINITE Put in the most general terms, (155) expresses a scale of decreasing likelihood of agenthood, hence decreasing propensitiy to occupy the A-function. In natural discourse, the most natural and frequent type of agent is human, or at least animate. Thus NPs with these semantic features will be the most likely candidates for the A-function in any clause. In contrast, inanimates such as sand, stone, chair etc. are unlikely to be the subjects of transitive verbs, and indeed, there are but few transitive verbs available in the lexicon which they could even meaningfully combine with as subject. Of all the different types of entity referred to by language, it is humans which are capable of carrying 16 17 Filiminova (2005) provides a detailed critical account of animacy and alignment, with special relevance for Indo-Iranian, but unfortunately I became aware of her contribution too late to make major revisions to the text. It is a matter of debate whether first and second person should be distinguished from each other, as in (155), or treated alike; see discussion recent discussion in Filiminova (2005). In Iranian, there are good grounds for the favoured position of the first person. A much more complex issue, for which the hierarchy offers no solution, is the differences between plural and singular of nouns and pronouns, typical for Iranian. I have no explanation for these differences. Case and animacy 181 out the largest number of agentive activities, hence most likely to express the subjects of such verbs. Semantics is not the sole factor involved. In connected discourse, humans are also most likely to be singled out as topics, the entities whose actions form the baseline for most narratives. A narrative often recounts a sequence of actions carried out by a particular person or group of persons. Such topical entities are, throughout most of the narrative, coded as definite NPs in the language concerned, and cast in the role of subject. Thus the semantics of humanness correlates with topicality, and both conspire towards putting human, definite NPs into the subject position (cf. for example (Payne 1997: 150) and Du Bois 1987). In particular, we find that discourse is largely concerned with the actions of the speaker herself and her interlocuters, that is, the first and second persons (in spoken English, I and you are the most frequent pronouns, and indeed are among the most frequent of all words; see Leech et al. (2001), and Tomasello (2003: 82) for child language). Thus regardless of minor differences between individual languages, and between different versions of the animacy hierarchy, pronouns of the first and second person invariably enjoy a privileged position at the top of the hierarchy. The combined force of semantics and pragmatics is therefore to make those NPs furthest to the left on the hierarchy the most likely to fill the A function in a transitive clause. According to a broad functionalist perspective on language, that which is pragmatically and semantically unmarked is expected to be formally less marked. Thus in a transitive clause which conforms to the assumed semantic and pragmatic norm, with A higher in animacy and definiteness than O, we will find a minimum of formal markedness. But deviations from the norm are ‘penalised’ through some additional marking. Therefore, participants that rank highly on the Animacy Hierarchy in (155) will tend to be be morphologically unmarked in A-function. Those further to the right on the hierarchy, however, will tend to be overtly marked when they occur in A-function. 4.6.1 Animacy-based splits in Iranian The Iranian languages are best-known for their tense-based alignment splits. But in the typological literature, a second type of alignment split is com- 182 Case systems in West Iranian Table 25. Animacy-based alignment split in Dyirbal (Dixon 1972: 130–137) 1 ST /2 ND PERS . 3 RD PERS . S A O -ø -ø ø −Nu −na -ø Alignment ACCUSATIVE E RGATIVE monly recognized: a split according to the semantics of the arguments of the clause, referred to here as an animacy-based alignment split. Typically, we find alignments differ according to whether the arguments are pronouns, or lexical NPs. Investigation of languages with animacy-based alignment splits has revealed an extremely robust generalization. Stated informally, the general rule is that accusative alignment is found with those items at the left (top) of the hierarchy in (155), while ergative alignment is found at the bottom. A famous example of this kind of split is Dyirbal (Northern Queensland, Australia, see Dixon 1972). In this language, pronouns of the first and second person take Nominative case marking when in S or A function, but Accusative case marking in O function. Thus case-marking alignment is accusative with these pronouns. With all other NPs, however, case-marking follows an ergative pattern: the A takes a distinct Ergative case marker, while S and O are unmarked. We find here then the characteristic restriction of accusative alignment to the top of the hierarchy. The Dyirbal case system, somewhat simplified (not all allomorphs are shown) is reproduced in Table 25. Returning to Iranian, it will be recalled from the preceding sections that a number of languages have significant differences in the number and type of case distinctions available to the SAP-pronouns, and to the nouns (as indeed does English and most languages of Western Europe). Here then we already find an echo of an animacy-based split in case-marking. The question to be addressed in this section is whether these differences result in animacy-based alignment splits in Iranian languages. Consider for a moment a theoretical possibility: A language with the inherited ergative alignment for the past tenses begins to lose its inherited Oblique case. It is lost first on the nouns of the language, leaving them with no case distinction, but it is retained somewhat longer on the SAP-pronouns. Such a state of affairs is actually quite commonplace, for example in the Abyānei dialect, Central Plateau (cf. Table 7), but also in the Sourkhéi and Case and animacy 183 Lāsguerdı̄ dialects (Christensen 1935). In the latter, no case distinctions are preserved, hence A and O are equally unmarked (Christensen 1935: 66). But the SAP-pronouns (singular) have preserved the old suppletive Direct/ Oblique opposition (in Lāsguerdı̄ for example, first person a/mo, second person to/ta). Now on the assumption that the old Oblique remains associated with the old ergative alignment, we could expect to find in these languages a case-marking split along the following lines: ergative case marking with SAP-pronouns (A in the Oblique, O in the Direct), and neutral system on the nouns (because they have no possibility to express a case distinction). Notice how such an alignment split would run counter to the cross-linguistic predictions on alignment splits, because it would leave the items at the top of the Animacy Hierarchy with ergativity, while those further down would have a neutral alignment. Significantly, this scenario does not occur. Under precisely the historical set of circumstances just outlined what regularly happens is that ergative alignment is abandoned on the SAP-pronouns: the inherited Oblique forms are used for Object function in both tenses. Consider the past transitive constructions in Lāsguerdı̄: (156) žo mó bǽdi 3 S 1 S:OBL see:PST:3 S ‘He saw me.’ (Christensen 1935: 70) (157) á tá bǽdi-æ̀m 1 S 2 S:OBL see:PST:1 S ‘I saw you.’ (Christensen 1935: 70) Here we find the old Oblique used for the O-past (first example), but not for the A-past (second example). This is definitely unexpected, because it runs counter to our First Principle in (133): an old Oblique is always associated with the A-past function. Here, however, this is not the case, hence we must modify this principle to account for languages such as Lāsguerdı̄. What we actually find in this language is a split, but one that respects the predictions based on animacy: Nouns have lost case entirely, and therefore show neutral case alignment (they remain unmarked, regardless of their syntactic function) while the SAP-pronouns are marked Oblique in O-function, and Direct in S and A. This of course is precisely the split found in English, or French. The Central Plateau dialect of Abyānei works in an identical 184 Case systems in West Iranian fashion. Here too, the inherited Oblique case has disappeared on all nouns, but a distinction is retained on the SAP-pronouns. And again, the cases with these pronouns do not work on an ergative basis, but on an accusative one: the inherited Oblique marks O in both tenses, and the Direct marks S and A (cf. Le Coq (1903: 85), and extensive examples in texts there, pp. 392–427). In fact, regardless of how the case systems on the SAP-pronouns and those on the nouns develop, it seems that a situation where the SAP-pronouns retain ergative alignment while the nouns lose it does not occur. We never find a situation in which ergative alignment is maintained at the top of the hierarchy after it has been lost on lower levels. Notice too how another common diachronic development, the intrusion of Innovated Object markers (cf. Section 4.3), reinforces the tendency just mentioned. Innovated Object markers are always introduced at the top of the Animacy Hierarchy, that is, with SAP-pronouns. And Innovated Object markers are always used to mark the O-function (even though, as in Vafsi, they may also used be for the A-past). This means that any language that uses such Innovated Object markers automatically marks the O-function when the NP concerned is an SAP-pronoun, regardless of what may happen in the rest of the grammar. What we find then is that the move towards overt marking of the O-function – essentially the essence of accusativity – is concentrated at the top of the Animacy Hierarchy. Thus introduction of the Innovated Object marker automatically reinforces the predicted effects of animacy because it strengthens accusativity at the top of the hierarchy. Further evidence for the relevance of animacy in splits comes from the Hawrami dialect of Gorani, as described in MacKenzie (1966). In this language, the sequence of loss of the Inherited Oblique has been the exact converse of Lāsguerdı̄ or Abyānei: it is the SAP-pronouns that have lost the Direct/Oblique distinction, while nouns have retained it. Such a sequence runs counter to the general principle that inherited case distinctions are maintained longest on SAP-pronouns, and to my knowledge, this particular dialect is the only language to exhibit such a situation (note that the neighbouring dialect of Kandulai (Hadank and Mann 1930) has ‘rectified’ the situation by innovating secondary Oblique markers on the SAP-pronouns, cf. discussion in connection with Table 11). The Hawrami system is illustrated in Table 26. But what is particularly telling about the Hawrami dialect is how the inherited Oblique case is used on nouns. Generally, past transitive constructions Case and animacy 185 Table 26. Case and nouns and SAP-pronouns in the Hawrami dialect of Gorani D IRECT 1st. Pers. 2nd. Pers. N OUNS (‘donkey’) O BLIQUE mIn to har har-i very rarely have an A in the Oblique case, thus ergative alignment is significantly weakened. The exception to this is apparently with an agent that is inanimate, for example ‘the heat’, ‘illness’, ‘thirst’ (MacKenzie 1966: 51), which may be in the Oblique case. There could hardly be a better example of the relevance of the Animacy Hierarchy than this: precisely those nouns which are semantically least likely to be Agents receive overt marking when they are in A-function, while those higher are unmarked. Another language lending support for the principles discussed above is Southern Balochi, as described in Farrell (1995).18 With all NPs in the third person, Southern Balochi shows the typical Iranian alignment split between present and past tenses (called by Farrell “Perfective”). Present tenses have predictable accusative alignment, with A in the Direct and O in the Oblique (there is the added complication of DOM, but this need not concern us here): (158) b@cık jınık-a j@-ã boy:DIR(PL) girl-OBL hit:PRES:3 PL ‘The boys hit the girl.’ (Farrell 1995: ex. 15) Intransitive clauses have, equally predictably, the S in the Direct case regardless of tense: (159) 18 b@cık r@o boy:DIR go:PRES:3 S ‘The boy goes.’ (Farrell 1995: ex. 14) Two changes have been made to the transcription: (i) The closed-omega symbol, used by Farrell (presumably) for the near-close, near-back rounded vowel, has been replaced by the current IPA-symbol for this vowel, [U]. (ii) The symbol [S] has been replaced by “š”. 186 Case systems in West Iranian (160) jınık šU girl:DIR go:PST:3 S ‘The girl went.’ (Farrell 1995: ex. 6) When we turn to past-tense clauses with SAP-pronouns as core arguments, however, alignment changes to consistent accusative. The relevant examples are the following, with the glosses adapted in line with the present analysis: (161) m@n t@-ra gir-ã 1 S 2 S-I NN O BJ catch:PRES-1 S ‘I will catch you.’ Farrell 1995: ex.(11) (162) m@n t@-ra gıtt 1 S 2 S-I NN O BJ catch:PST ‘I caught you.’ Farrell 1995: ex.(12) (163) ma šUma-ra tacent 1 PL 2 S-I NN O BJ chase.off:PST ‘We chased you off.’ Farrell 1995: ex.(13) These examples demonstrate that the case-marking of first/second person pronouns in A and O functions remains identical, regardless of tense. What has happened in Balochi is in certain respects similar to the situation in described above for the Hawrami dialect of Gorani: the inherited case distinction has been lost on SAP-pronouns, but retained on nouns (this assumption is controversial; cf. discussion in fn. 13). But while the SAP-pronouns of Hawrami are simply case-neutral, in Balochi Innovated Object markers have been introduced, which are deployed on an accusative pattern. The result is ergativity on nouns, and accusativity on pronouns. The preceding discussion has shown that on the whole, animacy-based alignment splits in Iranian conform to the typological expectations: accusative alignment is associated with the top end of the hierarchy. More significantly, this tendency is operative even when the chronology of changes in the case system (the inherited Oblique is lost on nouns before pronouns) actually work towards splits in the unexpected direction (ergativity high on the hierarchy). It seems, then, that the factor of animacy is indeed a powerful force in shaping the outcome of alignment shifts. Case and animacy 187 4.6.2 The Ergativity Continuum It will have become apparent that in their past transitive constructions, very few Iranian languages, from Middle Iranian right down to the present, have anything like pure or consistent ergativity. Instead we find a variety of mixed systems, which render the conventional ergative/accusative dichotomy of limited applicability. Rather than introduce further taxonomic labels to capture the various individual bundles of case and agreement, it would be desirable to identify more primitive principles underlying different alignment types. In this section, I introduce a principle that can profitably be used to describe most of the attested variation: the grammaticalization of the unmarked Object. It is important to note that in this context, the term markedness refers to the formal markedness (German Merkmalhaftigkeit) of the O relative to the formal markedness of the A. Looked at from this perspective, the commonest attested alignment types in modern Iranian languages can be situated on a continuum between ergative, with a maximal discrepancy between a completely unmarked O (Direct case) and the marked A (Oblique case), and accusativity, where the discrepancy holds in the other direction, between an unmarked A and a marked O.19 Among Iranian languages, at least four grades of grammaticalization of the unmarked O can be identified. These are shown in The Ergativity Continuum given in (164): (164) The Ergativity Continuum of case-marking in Iranian a. E RGATIVE : O is unmarked, A is marked. b. N EUTRAL : O is unmarked, A is unmarked. c. D OUBLE O BLIQUE /T RIPARTITE : O is marked, A is marked. d. ACCUSATIVE : O is marked, A is unmarked. To forestall any false impressions, the Ergativity Continuum in (164) is not intended to be read as a diachronic sequence. For example, a neutral stage 19 In Don Stilo’s ongoing work on Vafsi syntax, similar views on the graded nature of ergativity are formulated to capture the different types of Past Transitive Construction in Vafsi. But whereas the present one considers only the parameter of nominal case marking, Stilo’s continuum takes more factors into account (word order, pronominal clitics) and pursues somewhat different aims. 188 Case systems in West Iranian Table 27. Alignment splits in Past Transitive Constructions in Iranian L ANGUAGE SAP- PRONOUNS N OUNS P ERSIAN K URDISH (N. G ROUP ) K URDISH (S. G ROUP ) A BY ĀNEI S OUTHERN BALOCHI H AWRAMI G ORANI VAFSI (AOV) VAFSI (OAV) T. ĀROM ROŠANI (PAMIR ) Accusative Ergative Neutral Accusative Accusative Neutral Double Obl. Ergative Tripartite Double Obl. Accusative Ergative Neutral Neutral Ergative Ergative (restricted) Ergative / Double Obl. Ergative Ergative / Double Obl. Neutral U NATTESTED U NATTESTED U NATTESTED Neutral Ergative Ergative Accusative Double Oblique Accusative is often not attested for languages that have moved from ergative to doubleOblique. The Ergativity Continuum turns out to be relevant in understanding the animacy-based splits on case-marking in individual languages, such as those that were discussed in the preceding sections. It will be recalled that a robust generalization is that ergative case marking is associated with the lower section of the Animacy Hierarchy. With the aid of the Ergativity Continuum, we are now in a position to refine this generalization along the following lines: the highest level of ergativity (in the sense of the Ergativity Continuum outlined above) is associated with the lowest positions on the Animacy Hierarchy given in (155), for example, inanimate nouns. Restricting ourselves to a comparison of the case marking on SAP-pronouns with that of nouns, Table 27 gives a selection of attested combinations, all of which support the above-mentioned generalization. Three logically possible, but to my knowledge unattested, types of split are shown at the bottom of the table. The findings illustrated in Table 27 can be expressed in the form of the following universal, which replaces the First Principle expresssed in (133) above: Case and animacy 189 (165) Universal of Animacy-based splits in Iranian The alignment of SAP-pronouns is never higher on the Ergativity Continuum (164) than the alignment found on nouns. To my knowledge, (165) is the clearest reflection of how functional principles have shaped the development of the Iranian case systems. The general idea is that the unmarked O is only tolerated under rather specific conditions, that is, it presupposes certain alignment constellations in certain parts of the grammar. To this extent, ergativity is a more marked alignment option than accusativity, because it is dependent on certain features being present in certain parts of the grammar.20 In much of this book I have been obliged to recognize a high degree of historical contingency in the way alignment has developed, an often remarkably arbitrary series of independent developments in morphology and phonology that give rise to an almost endless number of distinct combinations. However, the distribution of alignment patterns on the pronouns and nouns quite clearly respects the functional principles outlined at the outset of this section: SAP-pronouns are, in comparison to nouns, the more natural and likely Agents, and the less natural and likely Patients. Correspondingly, the grammar works consistently towards avoiding the formal marking of the likely. The universal stated in (165) is to some extent due to the workings of DOM in Iranian. DOM, the systematic marking of at least a subset of Objects, is in a sense the antithesis to ergative alignment – if the latter is defined as the systematic non-marking of the Object, as it is above. As Bossong (1985) has shown, the SAP-pronouns are the domain of DOM par excellence. Thus as soon as DOM enters a system, its first effect is to introduce Object-marking with SAP-pronouns. In doing so, it draws the case-systems on the SAP pronouns down to the lowest levels on the Ergativity Continuum (either c. or d.), regardless of any other other developments. But as the example from Abyānei above shows, the regular distribution of alignment types expressed in (165) cannot be attributed solely to the workings of DOM. Rather, even when only inherited case markers are available, alignment may change in order to avoid a situation that would run counter to the universal in (165). 20 See Dik (1978) for earlier discussion of ergativity and markedness. 190 Case systems in West Iranian Finally, consider how the Universal of Animacy-based splits provides a natural explanation for one of the apparent anomalies encountered above in the discussion of the Tatic-type languages. It will be recalled that many of these languages have just two SAP-pronouns for core syntactic functions: an old Direct, symbolized with AZ, and an Innovated Object marker, symbolized with CEMEN, referred to as the AZ / CEMEN languages above. Among these languages, two distinct patterns emerged, shown in Fig. 6 and 7 respectively. Basically, the patterns differ according to whether the Innovated Object form CEMEN takes on the A-past function (Vafsi and Eshtehardi), or is used solely as an Object marker (most of the languages in Yar-Shater 1969). The use of CEMEN for the A-past is, I suggested, highly unusual in the Iranian context, because it is clearly an innovated case form, yet it has taken on the A-past function. In the other languages, despite having the same inventory of forms at their disposal, this does not happen. The above discussion on animacybased splits provides a possible explanation for this state of affairs. In both Vafsi and Eshtehardi, case alignment on nouns is fairly robustly ergative in the past tenses, shown for Eshtehardi in (166) and Vafsi in (167), also showing cross referencing of the A through a pronominal clitic on the verb: (166) a. šir-e gāwa beškiast lion-OBL cow(DIR) crush:PST:3 S ‘The lion (A) crushed the cow (O)’ (Yar-Shater 1969: 97) b. gāwā šir beškiast cow:OBL lion(DIR) crush:PST:3 S ‘The cow (A) crushed the lion (O)’ (Oblique case through lengthening the final vowel on gāwa) (Yar-Shater 1969: 97) (167) vǽrg-i luás b-is-di wolf-M:OBL fox(M:DIR) TAM- PART.-CLC:3 S- see:PST:3 S ‘A wolf (A) saw a fox (O)’ (Don Stilo, p. c.) Thus with nouns, both languages exhibit fairly robust ergativity, with regular Oblique marking of the A-past. In Vafsi, however, definiteness and animacy of the O trigger Oblique marking of the O as well, though I cannot do justice to the complexities of this system here. With the other languages of the AZ / CEMEN-type, however, ergativity is largely lost on nouns. The A-past is unmarked, while a definite O is Obliquemarked. The rules determining the case-marking of objects are complex (see Case and animacy 191 Yar-Shater (1969: 99-100) for discussion). Unfortunately, a clear example of an Oblique-marked O-past was not available in the data, but Yar-Shater (1969: 101) states quite clearly that for all dialects, an O-past is “generally treated as an ordinary object except in Eshtehardi”, for which see above. An example from the Dānesfāni dialect is the following: (168) Hasan nun=eš bexwa Hasan(DIR) bread=3 S eat:PST:3 S ‘Hasan(A) ate bread(O)’ (Yar-Shater 1969: 114) An explanation for why Vafsi and Eshtehardi retain ergative, or at least Double Oblique, alignment on the SAP-pronouns, while the other dialects of Yarshater’s Southern Tati abandon it, can is to be sought in the interaction of alignment and animacy: those languages which have all but lost ergativity on nouns cannot maintain it on pronouns, because this would run counter to the universal given in (165), according to which SAP-pronouns can only exhibit ergative alignment when lower levels on the hierarchy also do. This explains why, despite the surface similarities in the inventory of the SAPpronouns themselves, we find two distinct alignments associated with them. Ergative alignment on pronouns needs to be ‘covered’ by an alignment higher in ergativity on the nouns, either ergative, or neutral. But in languages like Dānesfāni, nouns exhibit fairly regular accusative alignment in past tenses, so SAP-pronouns cannot have anything less than accusative alignment. In Vafsi and Eshtehardi, on the other hand, the presence of ergativity on nouns makes ergative, or at least double-Oblique, alignments a viable option for SAP-pronouns. While this account cannot predict why Vafsi and Eshtehardi do not simply abandon all trace of ergativity on the pronouns (as has happened in Southern Balochi, see above), it does account for the impossibility of maintaining ergativity on the SAP-pronouns in the other az/cemen-languages. At any rate, I believe the attested correlation provides further striking support for the universal of alignment splits given in (165), and more generally, for the powerful role of animacy in mediating alignment changes. 192 Case systems in West Iranian 4.6.2.1 Cross-system harmony There is no simple answer to the question of why West Iranian has spawned such a mixed litter of hybrid case systems. Considerations of economy and the discriminatory function are undoubtedly relevant, and in the long term work together to produce certain rather typical alignment profiles. And as we have seen, animacy is a powerful factor in determining the outcome of historical change. But neither of these factors are sufficient to account for all developments attested. In this section, I will discuss what I consider to be the most important factor behind the historical changes, and I will also bring together the various themes of this chapter in an attempt to formulate a unified account of changes to the case systems. But first let me introduce the notion of cross-system harmony. The idea behind this is that ideally, clauses expressing the same basic propositional content will exhibit the same constellation of formal morphosyntactic features: word-order, case-marking, and agreement. The notion was developed from an acquisitional and processing perspective by Slobin and Bever (1982), who pointed out that cross-linguistically, there is a preference for certain “canonical” sentence types to be used in L1 acquisition across different environments, including environments in which the grammar of the language concerned actually requires a different sentence type. The general tendency is then towards uniform expression of sentences within one and the same grammar. Of course many grammars impose specific structural restrictions on sentences in certain environments. For example, in German many subordinated clauses are required to be verb-final, rather than the more frequent verb-second position of main clauses. In Turkish, many subordinate clauses take a subject in the genitive case, rather than the nominative of main clauses. This type of non-motivated grammatical variation is apparently difficult to acquire and process. Now the TSA of Iranian is also a blatant violation of cross-system harmony. It requires speakers to acquire and process two different alignment types for transitive clauses, depending on the tense of the verb. Nor are the different alignments transparently linked to any particular semantic or pragmatic content; the sole difference in meaning between the past-tense expression and its present-tense counterpart is the time reference of the verb, and this is carried by Tense/Aspect morphology on the verb anyway. The change in case forms on the nouns thus appears communicatively Towards a solution 193 redundant. We would predict that change will work towards minimizing such non-motivated deviations from the canonical sentence form. The importance of cross-system harmony with regard to alignment splits is clearly recognized by Harris and Campbell (1995: 258–264), who discuss it under the label of “Complementarity Principle”. In the present connection, the crucial aspect of this principle is that changes in case marking in different sub-systems (e.g., past vs. present tenses) will work in the direction of ensuring identical case marking for identical syntactic relations. But as they note, in the Iranian context the principle is rather difficult to verify, as there are often only two cases involved, Direct and Oblique. Thus if a syntactic relation, let us say, O, is coded with Direct in one environment and Oblique in another, any change on either will work towards levelling out the difference, hence making the claim rather vacuous. Nevertheless, the predictions can be made more rigorous in the Iranian context by the addition of two stipulations. First, we would expect change to work in the direction of levelling towards the unmarked clause type. In Iranian, this is certainly the present tense (undoubtedly more frequent, as it is also the basis for many subjunctive verb forms used in subordinate clauses). Thus we expect the past tense constructions to level towards the present, and not the other way round. The second stipulation will need a little more justification, but is nevertheless well motivated (see below): levelling out in the coding of syntactic relations across different tenses is achieved earlier for Objects than it is for Subjects. 4.7 Towards a solution Cross-system harmony is not the only factor driving alignment change, but I suggest it is the primary one. The real analytical challenge is to formulate an account that captures the complex interplay of partially independent, and indeed partially contradictory factors. In what follows, I present such an account, framed in terms of a ranking of constraints and principles, such that the highest ones must be observed by all languages, and only when they are satisfied may languages change further to comply with lower-ranking ones. Despite the nomenclature here, I am not formulating an Optimality-theoretical account, although the idea is obviously attractive, and the reader is free to 194 Case systems in West Iranian translate this account into such a formalism. My aim here is to formulate an account of observable chronological sequences and pathways, and the notion of constraints ranked in order of chronological priority offers a relatively elegant, empirically well-grounded solution. The crucial aspect of this account is that the notion of cross-system harmony, as discussed above, ranks higher than the discriminatory function. In (169), a ranked list of the constraints involved is formulated in order of decreasing priority. Below, I will exemplify and justify the formulation provided here. (169) Constraints on changes in Iranian case systems a. Leave alignment in the present tenses unchanged. b. Mark plural number on nouns. c. Alignment with SAP-pronouns may not be higher on the Ergativity Continuum (cf. 4.6.2) than alignment with nouns. d. The case of S must be identical to that of A-pres.21 e. Make the expression of O-past equivalent to that of O-pres. f. Make the expression of A-past equivalent to that of A-pres. g. Mark O more than A. Constraints (a–d) are, to my knowledge, observed by all West Iranian languages, and possibly all modern Iranian languages. They are nevertheless non-trivial, as they rule out a whole host of theoretically possible changes. For example, nowhere do we find present tenses developing non-accusative alignments to harmonize with past tenses (against Constraint a); nowhere do we find a language which retains only case-marking, but not number marking (b); nowhere do we find a language where the case marking of S differs from A-pres, although given the changes elsewhere in the system, this would be a perfectly reasonable type of change (c). Nowhere do we find a language with double-Oblique on nouns, but ergative alignment on SAP-pronouns (see discussion in Section 4.6.2). Constraints (a–d) thus represent general constraints on change, a minimal specification which all languages must satisfy.22 21 22 With the possible exception of Wakhi, see Bashir (1986). A partial exception to Constraint (b) is found in Northern Group Kurdish, where plural marking of nouns is not marked on the nouns in all contexts, but only when the noun is Towards a solution 195 Constraints (e) and (f) are the more interesting ones. Both concern crosssystem harmony, but crucially, (e) is ranked higher than (f). This ranking can be invoked to explain a host of facts regarding case-marking changes in Iranian languages, and therefore needs to be justified in some detail. First, this ranking provides a natural explanation for the fact that a special Obliquemarking of the A-past is often the last vestige of TSA which a given language retains. We even find languages that reserve a special pronoun solely for the A-past function, for instance T.ārom discussed in Section 4.4.1. In other words, non-equivalent marking of A-pres and A-past may survive long after the other differences between past and present transitive constructions have been levelled out. This suggests that identical case marking (crosssystem harmony) for the A-function should be lowest-ranked. Second, the very widespread phenomenon of double-Oblique constructions in Past Transitive Constructions receives a natural explanation. On this account, such constructions arise through the need to code the Object-function identically in all tenses. Satisfying this constraint will automatically yield a double-Oblique. Finally, this ranking accounts for the fact that we never find a consistent ‘double-Direct’ in the past, that is, we never find the expression of the Apast shifting to the Direct (thus bringing into line with the A-pres) before the Object relations have levelled out.23 In sum, all the available evidence suggests that harmonizing the expression of the O-function takes priority over harmonizing the expression of the A-function. It is instructive to look at data from the SAP-pronouns in Pamir languages, where these tendencies are also confirmed. The data come from Payne (1980), a survey of alignment in a cluster of closely-related Pamir languages (or dialects). Payne assumes an original archaic system, with the A-past in the Oblique and the O-past in the Direct. However, what commonly happens 23 Oblique. However, plural marking is still available on nouns in this language, and it is notable that in several dialects, regular plural marking of all nouns is beginning to spread; see Chapter five for discussion. An apparent exception to this is found in those languages where case marking is lost entirely. Here we vacuously find a Double-Direct, but this of course matches the present tense anyway, so that full cross-system harmony has already been reached. The important point is that if case-marking is retained anywhere, we will not find the combination of double-Direct in the past contrasting with accusative alignment in the present. 196 Case systems in West Iranian Table 28. First person singular forms in the Pamir dialect of Rošani (Payne 1980: 154–155) P RESENT PAST S A O az az az men men men Table 29. First person singular forms in the Pamir dialect of Bartangi (Payne 1980: 162–163) P RESENT PAST S A O az az az az / men men/InnObj men/InnObj Table 30. First person singular forms in the Pamir dialect of Orošori (Payne 1980: 166–167) P RESENT PAST S A O az az az az InnObj InnObj in these languages is that the O-past relation is also coded with the Oblique, giving rise to the familiar Double-Oblique. The relevant system is shown in Table 28. What is also attested is intrusion of an Innovated Object marker into the Object-function, in conjunction with a change in the form of the A-past (from men to az). This is shown in Table 29. From a system of this type, straightforward accusative alignment may arise, as shown for Orošori in Tab. 30. But what is crucially not attested are systems such as Tab. 31. Such a system would surely be a feasible outcome, given the totality of different combinations available, yet it just does not seem to occur. Note that it satisfies constraints (a–d), (f) and (g). But it fails to comply with constraint (e): the expression of the Object function is not uniform across all tenses. The extreme rarity of such patterns, despite being plausible systems in their own right, lends further support to the conclusion that cross-system harmony is ranked higher for the O-function than for the A-function, hence the ranking of constraints (e) and (f) provided above is well justified. Summary of case 197 Table 31. Unattested alignment system P RESENT PAST S A O az az az az men InnObj Consider now how the pressure to maintain the discriminatory function of case marking is handled in the present account. Although it is not explicitly stated in any of the constraints, it is inherent in constraint (g), the lowestranked constraint, according to which O should be more marked than A. On my account, it does not become operative until all the other constraints have been satisfied. And this appears to be the case. For example, the caseless Iranian languages are simply languages that comply with the first six constraints, but fail to comply with the final one. While this account is doubtless not the final word on the subject, it does cover a very large amount of the relevant data in a fairly simple fashion. It also avoids a priori judgements on the relative stability and viability of different systems. As mentioned, there is a strong tendency to view caseless systems, or double-Oblique systems, as inherently ‘unstable’, mere transitory stages in the implacable march towards regular and consistent case discrimination. On the account provided here, languages need not change their alignment at all, as long as they satisfy the first four constraints. Languages such as Zazaki, or the conservative dialects of the Northern Group of Kurdish are just such languages, and may well have been so for centuries. But if they do change, the changes occur in such a way that the highest-ranked constraints are satisfied first. Or, to put it another way, certain changes presuppose others. 4.8 Summary of case In this section, the history of case marking of core arguments is reduced to three distinct processes, and their interaction: the loss of the inherited Direct/Oblique distinction, the intrusion of Innovated Object markers into the system, and the revitalization of the old singular Oblique. As far as the first is concerned, it followed very different pathways in different domains of the 198 Case systems in West Iranian nominal lexicon: on singular nouns, complete loss of the Oblique; on plural nouns, no loss of the actual marker, but re-deployment as a general pluralmarker. And on the SAP-pronouns, the sequence of loss follows a fairly rigid sequence from plural, to second person, before finally reaching the first person singular. The intrusion of innovated Object markers also proceeds down a specific developmental sequence, beginning with the SAP-pronouns before spreading to the rest of the nominal lexicon. Finally, the revitalization of the old singular Oblique, while fairly unusual, is nevertheless a force to be reckoned with. Here, an old Oblique singular marker extends from the nouns to the SAP-pronouns, in a sense traversing the path in the opposite direction to that taken by an Innovated Object marker. Viewing the changes in the case system in terms of distinct processes, each working through a particular sequence and to some extent independent of the other processes allows us to understand why the attested systems show such an extraordinary range of case-marking patterns. Case systems are not simply mechanical, symmetical, across-the-board systems for marking syntactic relations. Nor do they change in fully synchronized steps, jumping from one alignment type to the next in a neat series of parameter shifts. Instead changes are mediated by the semantics of the nominals concerned, the most crucial factor being animacy. It was shown in Section 4.6 that this factor has been decisive in constraining the developments. The distribution of alignment types in a language, at any given point in time, appears to respect a general principle: in any given domain of the nominal lexicon, the degree of ergativity, as defined by the Ergativity Continuum in (164), will never be lower than that found in those domains lower on the Animacy Hierarchy (155). This of course reflects the well-known universal governing the types of animacy-based alignment splits in languages. What is remarkable in the Iranian case is that change respects this universal, even when the forces of morphological change would appear to militate in a different direction. A major theme of this chapter has been to assess the impact of maintaining the discriminatory function in shaping the way case systems have changed. It is certainly true that over an extended time depth, for example from Old Persian to modern Persian, the developments in the case system have indeed worked in the direction of re-instating the consistent distinction between A and O (in all tenses) that was present in Old Iranian, but lost in the Summary of case 199 intervening years. However, many of the attested sub-changes actually work in the direction of reducing the saliency of the A vs. O distinction, thus apparently moving to a functionally less ‘efficient’ system. For example, the loss of the old Oblique case, the development of double-Oblique constructions, or the wholesale abandonment of a consistent Direct/Oblique distinction on plural nouns in favour of a general marking of plurality (case marking is sacrificed to number marking, see Section 4.2.1.1) are difficult to account for if we assume that maintaining the discriminatory function is indeed the primary motor for change. I have suggested that a more important, or at least more immediately important, motivation for change is to preserve cross-system harmony. In Section 4.7, a unified account in terms of six ranked constraints on changes to case systems in Iranian was proposed, which allows a comparatively simple explanation for a considerable amount of the attested changes. On this account, maintaining the discriminatory function of case is actually ranked below the other constraints, and will thus only kick in when the other constraints have been satisfied. This would explain why, taken over a very long term (e.g., from Old to Modern Persian), it appears to be operative, but in the intervening stages, its role is minimal, or it may even be violated in order to satisfy one of the higher-ranked functional demands. Chapter 5 Kurdish (Northern Group) 5.1 Introduction The next two chapters provide a more detailed account of alignment in a single bundle of closely related West Iranian languages and dialects, the Kurdish group. The main emphasis is on the Northern Group of Kurdish, which exhibits a relatively ‘pure’ form of ergativity in its past tenses. However, even within the Northern Group, variations in past tense alignments are rife, with numerous intermediate shades being attested, including one variety with fully accusative alignment (Section 5.4.5). The Central and Southern Groups, on the other hand, are often claimed to lack ergativity completely (Bynon 1979), but they nevertheless show Tense Sensitive Alignment, as we shall see. Despite the comparatively close genetic relationship between the different varieties of Kurdish, in case and alignment patterns there is a deep rift dividing the Northern Group from the other two: The Northern Group has retained a fairly archaic case-marking system, still differentiated according to gender, but has cast off pronominal clitics completely. In the Central and Southern Groups on the other hand, case has been completely abandoned, while the pronominal clitics are deeply entrenched throughout the grammar, in fact functioning more like agreement than like anaphoric pronouns. While the Central and Southern Groups have clear predecessors in the systems attested for much of Middle Iranian, the origins of the Northern Group system are fairly obscure. Traditionally, Kurdish is considered to be a subgroup of the Northwest Iranian group of West Iranian (Blau 1989), but as noted in Section 1.2, the Northwest/Southwest distinction is not unanimously accepted. In fact, the evidence for the genetic unity of Kurdish is thin. As MacKenzie (1961b: 72) writes: In short, apart from this č- and the treatment of šm and -xm I can find no feature which is both common to all the dialects of Kurdish and unmatched outside them. To isolate Kurdish convincingly, therefore, would seem to entail comparing it with at least each West Ir. dialect, listing the common and 202 Kurdish (Northern Group) divergent features. For practical purposes, however, taking Kurdish as ‘that which is generally recognized by Iranists as Kurdish’, it is necessary only to consider for comparison its immediate neighbours, past and present. The sub-classification of the languages and dialects generally subsumed under the umbrella of ‘Kurdish’ is equally problematic. The following rough sub-grouping, “based on a combination of genetic relationship, geographic proximity, and ethnic identity” (Windfuhr 1989b: 294), represents a reasonably widely accepted consensus within Iranian studies: Northern Group comprises two major dialects: – Kurmanji (Kurm.), spoken by the Kurds of Turkey, Eastern Syria and the Caucasus, and parts of Iran. – Badı̄nānı̄ (Badı̄n.), spoken in North Iraq around the townships of Zakho, Dohuk, Amadiye, and in scattered regions of ex-Soviet Transcaucasia. Central Group spoken between Rowandiz and Suleimaniye, Iraq, with the dialect of the latter township having some prestige as a written language, and including Mukri in western Iran. Southern Group poorly and variously defined as a genetic unit, spoken in southwest Iran particularly in the province of Kermanshah.1 Residual Languages Two further languages spoken in the Kurdophone area, Gorani (under various spellings), spoken in Iran and Iraq, and Zazakı̄ (sometimes referred to as Dimili), spoken in headwaters of the Euphrates in Turkish Anatolia, are frequently subsumed under the label ‘Kurdish’. According to most scholars of Iranian languages, however, 1 The Southern Group is the weakest link in any classification of Kurdish languages. As MacKenzie (1961b: 79) notes, the languages of this group probably “differ almost as much from one another as they do from their northern kin.” Fattah (2000) presents a more recent overview of this group; the position of Lakki is controversial; Windfuhr (1989b) considers it part of the group, while Fattah (2000: 55–62) refers to it as a ‘Kurdish dialect’, though apparently not part of the Southern Group proper. Another problematic case is the position of Lor(i) (and indeed the precise demarcation of the term). Blau (1989) includes a sub-group of Lor, that of Pošt-e Kuh, in the Southern Group. Windfuhr (1989b) also refers to an idiom of this area, but as a dialect(?) of ‘Lakki’. See Anonby (2004) for recent discussion. Introduction 203 they are genetically more closely linked to each other than to the languages of the Kurdish group, and should not be included among them. In what follows, I will use Kurdish as a cover term for the Northern, Central and Southern Groups of Kurdish dialects, but excluding Zazaki and Gorani. 5.1.1 The Northern Group In terms of number of speakers, the Northern Group is the largest group of Kurdish dialects, with an estimated 15–20 million speakers. It comprises a reasonably homogenous bundle of dialects spoken in North Iraq, parts of Syria, Iran, Turkey, and the ex-Soviet Union, with the largest number of speakers being in Turkey. Many speakers have since left their ancestral homelands (estimates are as high as one third) for the large metropolitan centres of the Near East (Baghdad, Damascus, Tehran, Istanbul etc.) as well as Western Europe. Accordingly, the sources of language data are extremely varied.2 Perhaps the most important isogloss within the Northern Group separates the dialects of North Iraq, here collectively termed Badı̄nānı̄, from the dialects of Turkey, Syria and the Caucasus. In referring to the latter I will occasionally make use of the term Kurmanji (also written Kurmanjî, Kurmanci, Kurmancî). Many authors use the term Kurmanji as a cover term for the entire Northern Group. However, I adopt the terminology of MacKenzie (1961a) in referring to the entire group as the Northern Group. There is very high mutual intelligibility between Badı̄nānı̄ and neighbouring Kurmanji dialects. The dialect border is more of a continuum than a sharp break, although the political divisions that now accompany it may be leading to greater diversification. However, Badı̄nānı̄ differs from the dialects of Turkey and the Caucasus in some rather subtle facts of its syntax, features that I believe are particularly relevant to an assessment of the development of ergativity in the Northern Group; for this reason I devote Section 5.6 to just those aspects of Badı̄nānı̄ syntax. The present chapter is based on what has been termed the Hawarnorm (cf. Matras (1989) on this term, and more generally on the standardization of Northern Kurdish), a Roman-based orthographic norm. The name 2 For a general overview of sources see Haig and Matras (2002); for recent discussion of the internet as a source of linguistic data see Haig (2002b). 204 Kurdish (Northern Group) Hawar refers to a journal, published in, and on, the Kurdish language between 1932 and 1943, and recently reprinted in two volumes at the Swedish publishing house Nûdem. The Hawar-norm was developed in the journal, and was later published in slightly modified form in Bedir Khan and Lescot (1970). It is based on the Kurmanji spoken in southeastern Turkey around the areas of Cizir (Turkish Cizre) and Botan. While most written sources tend to orientate themselves towards the Hawar-norm, there are also a number of sources based on the spoken language, which show some deviations from the Hawar-norm, and in keeping with the aims of this study, I will be drawing on a number of them here (for example Makas (1897–1926 [1979]), Le Coq (1903), Kahn (1976), Ritter (1971), Ritter (1976), Haig (2006) and Haig forthc. a). In terms of alignment, the sources cited above actually show relatively little variation, although for reasons that will be discussed below, connected natural discourse contains comparatively little hard facts on case and agreement morphology in past transitive constructions. The most dramatic changes in alignment in the Northern Group are found in the Kurdish of the Gilan region in northwest Iran (Shojai 2005), while Dorleijn (1996), provides book-length treatment of alignment changes in dialects spoken in Turkey. 5.2 Overview of the morphosyntax In this section a brief overview of the most relevant aspects of the morphosyntax is given. Although there are a number of useful pedagogical and descriptive studies on the Northern Group, there is as yet no comprehensive treatment of the syntax. The following sketch draws on several of the available studies, and on my own previous work.3 The syntax of the Northern Group is neither consistently head-modifier, nor modifier-head. In the clause, it is generally predicate-final, although Indi3 Descriptions of the grammar of the Northern Group can be found in, among others, Justi (1880), Kurdoev (1957), Bedir Khan and Lescot (1970), MacKenzie (1961a) and Blau (1975) (both on Badı̄nānı̄) and Badıllı (1992). Overview articles can be found in Haig and Matras (2002) and Haig and Paul (2001). Pedagogical studies include Blau and Barak (1999), Wurzel (1997) and Rizgar (1996). Overview of the morphosyntax 205 rect Objects and Goal arguments systematically follow the predicate, unless they are introduced by an adposition (but in Badı̄nānı̄, post-predicate PPs are quite frequent). In the noun phrase, the order of constituents is head-modifier, where modifiers can be attributive adjectives or possessors, but demonstratives and quantifiers precede the head. The morphology is weakly agglutinative, both prefixing and suffixing, with some fusional properties. Nouns, verbs and adjectives can be distinguished as major lexical classes on the basis of inflectional properties. 5.2.1 Inflectional categories of the noun Nouns inflect according to gender, definiteness, number and case. Nouns have inherent gender, either masculine or feminine, reflected in distinct forms of the singular case markers, and of the Izafe-linkers. However, the system is rather more complex than it appears at first glance. Some dialects have levelled out gender distinctions in case markers, but not in the Izafe, while some have lost gender distinctions entirely. Gender will only be recorded here when it is relevant for a particular example, see Haig (2004b) for more detailed discussion. There is a two-way distinction in the nominal morphology between indefinite singular, and definite singular/generic, with the former being marked via a suffix −(e)k (both genders). Thus definite nouns are formally unmarked, and only context distinguishes them from the generic use: gundî ‘the (contextually recoverable) villager(s)’, or generic ‘villagers as a class’. In the plural, there is an ending −in occasionally used to indicate indefinite plural, but in most contexts the definite/indefinite distinction is not overtly signalled in the plural. Nouns distinguish two number values, an unmarked Singular and a marked Plural. The Northern Group has preserved the typological oddity mentioned for early Western Middle Iranian in Section 3.3.1, where Plural is only regularly marked on nouns in the Oblique case. In the Direct case, nouns not modified by an Izafe (see below) are unmarked for Plural: hesp hat ‘the horse (=hesp) came’ vs. hesp hat-in ‘the horses came’. As far as case is concerned, we find the familiar two-term case system, Direct and Oblique, 206 Kurdish (Northern Group) reflecting many of the features discussed for such systems in Chapters three and four. The Oblique case is used for: The complements of adpositions; Possessive modifiers in Izafe constructions; Post-predicate Goal arguments or Indirect Objects; Temporal adverbial expressions; O of a present tense verb; A of a past tense verb. The Direct case is used elsewhere. The forms of the cases depend on both the gender and number of the noun concerned, and on the presence or absence of a determiner (either a demonstrative, or ‘each’, or an interrogative such as kijan ‘which’, or the indefiniteness suffix −(e)k). Case marking for all feminine singular, all plural, and all determined nouns is relatively stable across the Northern Group, with only minor phonological variation (e.g., the deletion of the final −n of the plural Oblique). For masculine nouns, with no determiner, however, there is significant regional variation in the forms used. Three main patterns are found: 1. Suffix -î: hesp-î ‘horse-OBL’ 2. Raising of stem vowel (Umlaut): hêsp ‘horse(OBL)’ 3. No overt signal of case: hesp ‘horse’ DIR/OBL Umlaut (Pattern 2) only affects the short, unrounded, central front vowels (rendered orthographically as <a> and <e>). Consequently, in dialects with Umlaut, and no Oblique suffix for masculines, bare masculine nouns without the relevant vowels have a single invariant form in the Direct and Oblique singular. Consider the following examples showing the two masculine nouns gur ‘wolf’ and dîk ‘rooster’ (from Kurdish dialects spoken near the town of Midyat in Turkey, cited in Bailey 2005: 33): (170) gur dîk dî wolf(OBL) rooster see:PST(3 S) ‘The wolf saw the rooster.’ (171) dîk gur dî rooster(OBL) wolf see:PST(3 S) ‘The rooster saw the wolf.’ The distribution of the three patterns is very roughly from north to south: the dialects of North Iraq (Badı̄nānı̄) have variant 1, the dialects of southeastern Overview of the morphosyntax 207 Table 1. Case morphology in the Northern Group M ASC . F EM . PL. No determiner D IR . O BL . -0/ -î; -0; / Umlaut -0/ -ê -0/ -a(n) After Indef. Suffix -ek D IR . O BL . -0/ -î -0/ -ê / -î -0/ — With Dem. or Interrog. D IR . O BL . -0/ -î -0/ -ê -0/ -a(n) Turkey have variant 2, while the dialects of the North, e.g., that of the Erzurum district, have the variant 3. In fact, many dialects have a mix of all three. In particular, I am unaware of any dialect which consistently applies variant 2 (Umlaut) to all the relevant masculine nouns. Table 1 sums up the facts so far. Semantic cases (Instrumental, Goal, Source, Comitative etc.) are expressed through a variety of adpositions. There is a basic set of prepositions common to all the dialects of the Northern Group: ji ‘from’, li ‘in’, bi ‘through, with’. They are often combined with post-NP particles to produce additional meanings, giving rise to what are traditionally termed circumpositions, for example ji NP re/ra ‘to, for’.4 In the examples, I gloss both elements of the circumposition with ADP. As far as the personal pronouns are concerned, there is considerable dialectal variation, in particular the second and third persons (the latter essentially demonstratives), but the eight-term system retains its basic shape across all dialects. The forms were already given in Chapter four, Table 5, reproduced here for convenience as Table 2. 4 Obviously a further grammaticalization of the form noted for Early Judaeo-Persian az . . . rā ‘on behalf of, concerning’ (Paul 2002: 185). 208 Kurdish (Northern Group) Table 2. Personal pronouns in the Northern Group of Kurdish S INGULAR 1 P. 2 P. P LURAL D IRECT O BLIQUE D IRECT O BLIQUE ez tu min te em hûn me we 5.2.2 The Izafe construction Like many West Iranian languages, Kurdish languages are characterized by a particular type of complex NP, referred to here as the Izafe construction. The defining feature of the Izafe construction is a vocalic particle linking the head noun to a modifier which follows that noun. The Izafe particle itself goes back to an Old Iranian relative particle, and in the Badı̄nānı̄ dialect of the Northern Group, the Izafe particle still introduces what are, in effect, relative clauses (see MacKenzie (1961b: 80), and Section 5.6). In the dialects further to the North, however, the Izafe particle has becoming increasingly ‘suffixlike’. In fact, most standard treatments of the Northern Group discuss the Izafe particle under the morphology of the noun. However, I believe this is mistaken, mainly because a noun may be followed by more than one Izafe, in which case each modifier is preceded by an Izafe-particle: (172) Bi şeş hejmar-ên Hawar-ê ên pêşîn through six number-IZP Hawar-OBL IZP last ‘Over the last six numbers of (the journal) Hawar’ (Hawar Vol. 1: 263) In this example, both the genitive attribute Hawar-ê, and the adjectival modifier peşîn are preceded by the plural Izafe particle. If the modifier preceded by the Izafe is a noun or pronoun, it will take the Oblique case, as in Hawar-ê. Most authors draw a terminological distinction between the first and the second Izafes in examples such as (172), e.g., Blau and Barak (1999: 37), who refer to the second instance of Izafe as “l’Izafe tonique”. However, functionally, and phonologically, the two are so similar as to make the distinction very tenuous. Furthermore, the Izafe can actually precede a modifier without any overt head noun, giving the sense of ‘the one . . . ’: ya min ‘that of me’, ‘(the Overview of the morphosyntax 209 Table 3. The Izafe particle in the Northern Group Bare noun After Indef. Suffix -ek M ASC . F EM . PL. -(y)ê -î -(y)a -e -(y)ê(n) – one) which is mine’. For the purposes of this study I assume a single Izafe particle, rather than treating one as a suffix of the noun, and inventing alternative labels for the others. The Izafe is an element of a particular construction, rather than being part of the morphology of the noun.5 In the Northern Group (unlike the Central Group), the form of the Izafe particle varies according to the gender, number and definiteness of the head noun. Table 3 gives the relevant forms in the Hawar-norm, while examples of Izafe constructions are provided below. (173) pere xulam-ek-î qenc e, lê money servant-INDEF-IZM fine COP:PRES:3 S but axa-k-î xirab e master-INDEF-IZM poor COP:PRES:3 S ‘Money is a fine servant, but a poor master’ (Hawar Vol. 2: 1050) (174) şev-a tarî di evar-ê de kifş e night-IZF dark ADP evening-OBL ADP evident COP:PRES:3 S ‘The dark night is evident in the evening’ (Hawar Vol. 2: 1050) 5 It can in fact be viewed as the head of that construction, projecting to a specific type of phrase, the Izafe Phrase. The arguments in favour of such an analysis are: the Izafe is the locus for the expression of gender and number; it permits a unified account of Izafes with and without an immediately preceding head noun; it is congruent with the historical origins of the Izafe as a complementizer, arguably also the head of its phrase. However, I will not pursue the implications of this analysis here. 210 Kurdish (Northern Group) (175) Mamoste-yên ko giş tirk bû-n teacher-IZP COMP all Turk COP:PST-PL ‘The teachers, who were all Turks . . . ’ (Hawar Vol. 2: 748) (176) min ji di-ya xwe re got 1 S:OBL ADP mother-IZF REFL ADP say:PST:3 S ‘I said to my mother’ (Hawar Vol. 2: 854) (177) kûçik del-a xwe kir nav ling-ê xwe dog tail-IZF REFL do:PST:3 S between leg-IZP REFL ‘The dog put its tail between its legs’ (Hawar Vol. 2: 855) A crucial aspect of the Izafe construction is that case distinctions are neutralized on any noun qualified by an Izafe-phrase. Thus a noun followed by an Izafe particle is always in the Direct case, regardless of the function of the entire NP in the clause. Note that in another closely related language, Zazaki, Izafe constructions do vary in form according to their external case relations (Paul 1998b). One very important consequence of the invariant forms of Izafe constructions is to lower the functionality of case marking in distinguishing arguments. In transitive clauses where both A and O are Izafe constructions, only word order distinguishes A from O (likewise, the tense of the verb has no effect on case marking in such clauses): (178) a. bav-ê min di-ya min na-bîn-e father-IZM 1 S:OBL mother-IZF 1 S:OBL NEG-see:PRES-3 S ‘My father (can’t) find/see my mother’ b. di-ya min bav-ê min na-bîn-e mother-IZF 1 S:OBL father-IZM 1 S:OBL NEG-see:PRES-3 S ‘My mother (can’t) find/see my father’ (constructed example) 5.2.3 Inflectional categories of the verb Finite verbs inflect obligatorily for tense and person. Tense is primarily expressed through the basic present/past opposition of the two stems, already familiar from previous chapters. Table 4 gives the stems for some of the most frequent verbs in the Northern Group. Overview of the morphosyntax 211 Table 4. Common verbs in the Northern Group PAST STEM P RESENT STEM M EANING bûkirgotdaçûhatdîtke(w)tmirkuştgirtxwaravêt- b-, 0/ kbêjdç-, -her-ê-, werbînkevmirkujgirxwavêj- ‘be, become’ ‘do, make’ ‘say’ ‘give’ ‘go’ ‘come’ ‘see’ ‘fall’ ‘die’ ‘kill’ ‘take, hold, get’ ‘eat’ ‘throw’ Table 5. Verbal agreement suffixes in the Northern Group P RESENT STEM PAST STEM Sg. 1 2 3 -(i)m -ı̄; Imperative: -e/-0/ -e; -it(in) in Badı̄n. -(i)m -ı̄ -0/ Pl. 1 2 3 -in; -ı̄n in Badı̄n. -in -in -in; -ı̄n in Badı̄n. -in -in It will be seen that there is no straightforward rule for deriving one stem from the other, at least with these highly frequent verbs. Finite verbs obligatorily agree with one argument (there are no clitic pronouns in the Northern Group) via a suffix on the verb stem. The corresponding person agreement suffixes are given in Table 5. There are only two non-finite forms from verbs, an infinitive and a participle. Both are formed from the past stem of the verb. The infinitive adds an -n to the past stem (with an epenthetic short central vowel -i- added to 212 Kurdish (Northern Group) consonant-final stems): kir-in ‘do-INF’, bû-n ‘be-INF’. The participle adds −î: girt-î take-PTCPL‘taken, captured’. In addition to the person agreement suffixes given in Table 5, verbs may take one of several prefixes expressing Mood, Aspect and Negation. In general, only one of these prefixes is permitted with the present stem, so the Negation prefix displaces Aspect and Modality prefixes in negated verb forms. The past stem, on the other hand, can support two, for example, the combination of Negation and Progressive prefixes. The commonest prefixes are given below, with a very much simplified description of the semantics: Progressive: di- (with the present stem, the meaning is ‘Indicative’) Negation: na- (ne- in the past Indicative and Present Irrealis) Irrealis: bi- (used to express conditionals, future time, imperatives, i. e. all manner of propositions whose actual implementation the speaker is uncertain of) There is also a Perfective tense/aspect form, based on the past participle plus what appears to be a form of the clitic copula, but I am restricting this brief sketch to the most basic and dialectally uniform tense forms. Finally, there is a clitic particle, (w/d)ê which, in conjunction with an Irrealis form of the verb, indicates future time reference. This particle does not occur on the verb itself, but generally clause initially. Standard treatments of Kurdish grammar say it affixes to “the subject” (Wurzel 1997: 79) of the clause, but this account is not correct, as shown by examples such as the following: (179) Sibehê wê hakim ser-ê min jê in.the.morning FUT Prince head-IZM 1 S:OBL from.it bi-k-e IRR -do:PRES-3 S ‘In the morning the Prince will cut off (lit. make from-it) my head’ (Lescot 1940: 4) (180) “kiye?” dê bêj-e di-ya min “who is it?” FUT say:IRR:PRES-2 S mother-IZF 1 S:OBL ‘ “Who is it?” my mother would say . . . ’ (Cewerî 1986: 62–63) The canonical ergative construction 213 Thus although the Future particle often attaches to the subject, that is merely a by-product of its clause-second position. But when, as in the above examples, the subject is displaced from clause initial position, the clitic does not follow it. Having briefly sketched the most important features of the morphosyntax, we are now in a position to tackle the issue of alignment. Because this study is concerned with change over time, we are crucially interested in the attested variation, as it provides the most direct evidence for possible pathways of change. As a point of reference, however, it is useful to begin with what we can assume to be the original system, which I will term here the canonical ergative construction. 5.3 The canonical ergative construction In the Northern Group, as in all Iranian languages, alignment is accusative throughout the present tense, and with all intransitive verbs. The facts are uncontroversial; in the interests of brevity, they are illustrated with constructed examples containing pronominal arguments, rather than full NPs (in clauses with first and second person pronouns, case and agreement morphology is more transparent; see Haig (1998) for more details): (181) P RESENT, INTRANSITIVE : ez kurd-im 1 S Kurd-COP:PRES:1 S ‘I am Kurdish.’ (182) P RESENT, INTRANSITIVE : tu kurd-î 2 S Kurd-COP:PRES:2 S ‘You are Kurdish.’ (183) P RESENT, TRANSITIVE : ez te di-bîn-im 1 S 2 S:OBL IND-see:PRES-1 S ‘I see you.’ 214 Kurdish (Northern Group) (184) P RESENT, TRANSITIVE : tu min di-bîn-î 2 S 1 S:OBL IND-see:PRES-2 S ‘You see me.’ In past transitive constructions, however, a reversal of case and agreement patterns is found. The A is in the Oblique, the O is in the Direct, and the verb agrees with the O. The following examples give the past tense clauses corresponding to (183) and (184) respectively: (185) PAST, TRANSITIVE : min tu dît-î 1 S:OBL 2 S see:PST-2 S ‘I saw you.’ (186) PAST, TRANSITIVE : te ez dît-im 2 S:OBL 1 S see:PST-1 S ‘You saw me.’ As mentioned, alignment in all intransitive constructions is identical, regardless of tense. Compare (181) with the following: (187) PAST, INTRANSITIVE : ez zarok bû-m 1 S child COP:PST-1 S ‘I was (a) child.’ For the purposes of this study, I will refer to constructions such as (185) as canonical ergative constructions. Canonical ergative constructions are characterized by the following three features: 1. The A-past is in the Oblique case 2. The O-past is in the Direct case 3. The verb agrees with the O-past The canonical ergative construction is then, in terms of case marking and agreement, the mirror image of the accusative construction. The canonical ergative construction 215 5.3.1 The syntactic subject In a language with split alignment, we are unlikely to find a straightforward mapping of grammatical relations to overt cases. This does not, however, mean that grammatical relations such as subject are not relevant in such languages, but simply that the definition of such relations will need to take more factors into account than morphology. For Northern Kurdish, most of the relevant factors were already identified in Matras (1992/1993) and Matras (1997), and are largely uncontroversial. In this section I present a synopsis of the available literature supplemented by additional data and references. If we were to define grammatical relations solely in terms of case and agreement morphology, we would be obliged to conclude that a transitive construction in the present tense had a different subject to the corresponding clause in the past tense. Thus the ‘subject’ in (183) would be the A, while in the past tense version of the same clause, (185), it would be the O. However, closer examination of Kurdish syntax reveals that A-pres and A-past, despite having different morphological features, share a number of syntactic properties (and, of course, display basically the same semantic and pragmatic characteristics). Furthermore, these properties are also shared by the S. If we afford these syntactic properties due consideration, then it is possible to define a relationship of ‘subject’ in the Northern Group such that A-past, A-pres and S are all included, while the O-past is, despite its morphology, excluded. Furthermore, defining a subject relationship in this manner allows us to formulate a number of observations in a much more elegant and insightful manner than a definition based purely on morphology would permit. In the terminology of Dixon (1994), a subject relationship established largely on syntactic criteria is a ‘pivot’, but I will continue to use the term ‘subject’, or ‘syntactic subject’ here in the sense of Dixon’s ‘pivot’. Syntactic criteria for establishing subjecthood have been discussed at length in many works (see for example Van Valin 2001: 40–59 or Kroeger 2004: chs. 10–11). Here I will be briefly discussing a sub-set of those criteria, concentrating on those with direct applicability to Kurdish. 216 Kurdish (Northern Group) 5.3.1.1 Constituent order In all tenses, the pragmatically neutral order of constituents is SV, or AOV. To the extent that constituent order is relevant for defining grammatical relations, then, it appears that we can identify the clause-initial position with the grammatical relation of subject. There is some flexibility in word order, for example through the fronting of a topicalized O: (188) Ev erd hukumet-ê da-ye me this land government-OBL give:PST-PERF:3 S 1 PL:OBL ‘This land the government gave to us’ (Cewerî 1986: 13) (189) Çi şuGl me di-kir? what work 1 PL:OBL PROG-do:PST-3 S ‘What work did we do?’ (Matras 1992/1993: 147) A post-predicate (afterthought) position of an A is also possible. But crucially, there does not appear to be any significant difference in the types of word order pattern found with an A-past and an A-pres respectively. Clauseinitial position can therefore be considered a weak diagnostic for subjecthood (weak, because it can be violated). 5.3.1.2 Control of corefential deletion Kurdish has practically no non-finite syntax. Texts consist of a loose sequence of finite clauses, often with no formal indication of the relationships that hold across clauses. There is therefore no comparable structural counterpart to English constructions with raising verbs such as seem, or control verbs such as begin, promise, be able etc. The Kurdish equivalent of the latter involves two finite clauses, the second being in the subjunctive mood (glossed IRR): (190) kes-ek ni-kar-e [0/ şer-ê wê person-INDEF NEG-be.able:PRES-3 S 0/ war-IZM 3 S:OBL bi-k-e] IRR -do:PRES-3 S ‘No one is able to defeat her’ (lit. . . . (that he) might do war of her) (Lescot 1940: 40) The canonical ergative construction 217 With verbs such as ‘want’, pronoun deletion might be used as a subject diagnostic, differentiating between a construction I want [to go] from one like I want [you to go]. In fact, when the subject of the second clause is coreferential with that of the first, pronoun deletion is usually the pattern found: (191) ez di-xwaz-im [0/ bi sal-an bi te re 1 S IND-want:PRES-1 S 0/ for year-PL:OBL ADP 2 S ADP bi-jî-m] IRR -live:PRES-1 S ‘I want/have wanted for years to live with you’ (Aydogan Undated) In this example, the two clauses share a subject, thus it is deleted in the second clause. In the following example, the subject of the second clause is different to that of the first, hence there are overt pronouns in both clauses: (192) tu di-xwaz-î ez=ê jî qal bi-k-im 2 S IND-want:PRES-2 S 1 S=FUT too speech IRR-do:PRES-1 S ‘You want me to talk (about it) too’ (Bozarslan Undated) However, while coreferent pronoun deletion with xwastin ‘want’ is undoubtedly the rule in the present tense, in the past tenses, several complications arise. The verb xwastin ‘want’ (with the regional variant xwestin) is lexically transitive, hence it requires an Oblique subject in the past tenses. The clause dependent on xwastin, however, is always in the subjunctive mood, and the subjunctive mood is usually based on the present stem of the verb. Thus when xwastin is in the past tense, we have a discrepancy in tense between main and subordinate clause. In such cases, even under referential identity, the outer form of the pronouns would be different (the first Oblique, the second Direct). And under these conditions, the tendency is for the second pronoun to be retained, even under coreferentiality. Consider the following examples of pronoun retention (glosses adapted): (193) min di-xwest [ez her-im mal-ê] 1 S:OBL PROG-want:PST(3 S) 1 S go:IRR:PRES-1 S home-OBL ‘I was wanting to go home’ (Matras 1997: 627) (194) min xwest [ez otomobîl-a xwe bi-firoş-im] 1 S:OBL want:PST(3 S) 1 S car-IZF REFL IRR -sell:PRES-1 S ‘I wanted to sell my car’ (Cewerî 2001) 218 Kurdish (Northern Group) There are therefore restrictions on deleting pronouns when they do not share the same overt morphological form, most of which are discussed in Matras (1997). I will defer full discussion of these issues to 5.4.4. Of course restrictions of this nature are readily observable in expressions such as German mit mir oder ohne mich ‘with me or without me’. In English, this can be reduced to with or without me, with deletion the first pronoun. But in German, because the two pronouns have a distinct form (mir vs. mich), deletion of a pronoun is less readily acceptable. This is then a more general, and possibly universal, trait of grammar which happens in Kurdish to affect patterns of coreferential pronoun deletion. It should not, however, be considered as strong evidence against the subject relation. In Kurdish, deletion of coreferential pronouns is generally controlled by an S or an A, and therefore provides support for the existence of a subject relation. Furthermore, there is no evidence that an O argument can control coreferential deletion across clause boundaries, even when the pronouns concerned would be in the same case form (Matras 1997: 624). Thus Kurdish can be said to have a (weak) linking of S and A for coreferential deletion, both with ‘want’ and ‘begin’ types of predicate, as well as in looser types of coordinate clauses (indeed, it probably makes more sense to consider the former merely a sub-type of clause coordination). 5.3.1.3 Control of reflexives Another criterion for establishing the grammatical relation of subject is control of reflexive pronouns. For Kurdish, at least of the Northern Group, this turns out to be by far the most robust indication of syntactic subjecthood. The facts are as follows. In addition to the personal pronouns given in Table 2, there is also a non-inflecting reflexive pronoun xwe / xwa / xô. Like the other personal pronouns, it can be used both as a full NP, taking an argument position of a verb. Or it can be used as a possessive modifier in an Izafe-construction (cf. (194) above). Crucially, the rules governing the choice between reflexive pronoun and personal pronoun can only be stated with reference to the grammatical relation of ‘subject’. The rule is stated informally in (195):6 6 It is also possible to formulate these rules in terms of the Binding Principles, but there would be no measurable gain in accuracy. The canonical ergative construction 219 (195) Use xwe instead of a personal pronoun when the intended reference of xwe is identical to the syntactic subject of the first verb dominating xwe. The rules governing the reference of xwe apply in all tenses, hence are impervious to the morphological form of the syntactic subject. Consider (176), repeated here for convenience. Because the possessor of diya is coreferent with the syntactic subject min, it is only possible to use xwe: (176) min ji di-ya xwe re got 1 S:OBL ADP mother-IZF REFL ADP say:PST:3 S ‘I said to my mother’ If this clause were to be transposed into the present tense, thereby causing a change in the case form of the subject (from min to ez), it would not affect the rule requiring xwe as the possessor: (196) ez ji di-ya xwe re di-bêj-im 1 S ADP mother-IZF REFL ADP IND-say:PRES-1 S ‘I say to my mother’ In both of these clauses, the use of the personal pronoun min as a possessor (diya min ‘my mother’) would be ungrammatical. Xwe can occur as possessor in an Izafe construction, as in the above examples, or in argument function, as in the following example, this time from Badı̄nānı̄: (197) čo darva w xô hāvēta t bı̄rē-da went outside and REFL threw PREP well-LOC ‘(Shei ) went outside and threw herselfi into the well.’ (MacKenzie 1962: 314) However, the rule requiring coreference with the syntactic subject precludes xwe occurring in subject function itself (but see below). The domain within which the reference of xwe is defined is strictly local, namely the immediately dominating predicate. In the following example a main verb xwastin ‘want, request’ is followed by a subordinate clause with 220 Kurdish (Northern Group) a different subject. The constraints on the reference of xwe do not carry over from the main to the subordinate clause: (198) min jê xwast ku wer-e 1 S:OBL from.him want:PST(2 S) COMP come:IRR:PRES-3 S mal-a min / *xwe house-IZF 1 S:OBL / *REFL ‘I asked him to come to my house’ (Abdullah Incekan, p.c.) The only way of rendering the intended meaning is to use the personal pronoun min as the possessor of mal. If xwe were used here, it would be interpreted as coreferential with the subject of the immediately dominating verb, here were ‘come’, yielding ‘to his house’. The strictness with which these rules are observed is considerable, and has been underestimated even by experienced researchers. The following example is from the text collection in Lescot (1940): (199) wê nerî go kîz-a Carkê bav-êi When father-IZM DEM:OBL(fem.) see:PST COMP girl-IZF hat wîi DEM :OBL(masc.) come:PST ‘When her fatheri saw that hisi daughter had arrived . . . ’ (Lescot 1940: 8) In the notes accompanying the text, Lescot (1940: 246) corrects the use of the possessive expression kîza wî, stating “Il faudrait: kîza xwe”. But he is mistaken; the text is fully correct as it stands. The possessor must be the third person, rather than the reflexive, because it is part of the subject NP. As mentioned, the relevant domain is the immediately dominating verb, in this case the verb hat ‘come’, of which kîza wî is the subject. Coreference with the main-clause verb nerî, on the other hand, is irrelevant. Control of reflexives cannot cross clause boundaries in Kurdish. Note finally that it is the grammatical relation of subject, rather than linear precedence, which determines the reference of xwe. Even if the pronominal element occurs before the subject, if the coreference condition is met, then xwe is still required, giving rise to instances of backward control such as the following: The canonical ergative construction 221 (200) ji çîrok-an hez Di biçûki-ya xwei da mini ADP childhood-IZF REFL ADP 1 S:OBL PREP story-OBL:PL liking di-kir PROG -do:PST(3 S ) ‘In my (=self’s) childhood I was fond of stories’ (Blau and Barak 1999: 160) There is one area of uncertainty in the usage of xwe.7 When the subject NP consists of parallel NPs, the first may treated as a ‘subject’ for the purposes of control of xwe: (201) Divê ez û bira-yê xwe bi-zewic-in it.is.necessary 1 S and brother-IZM REFL IRR-marry:PRES-3 PL ‘My brother and I will have to marry’ (lit. It is necessary I and brotherof-self marry) (Zinar 1988: 102) However, it is also possible to find the personal pronoun used instead of xwe in such contexts. Finally, I should mention that in some of the peripheral northern dialects there is a tendency to overgeneralize xwe to include coreference with a topic, rather than a subject. A typical example is the following (transcription adapted): (202) Şêx Mus axa-yê eşîret-ek-î bû, pismam-ê Sheikh Mus Agha-IZM tribe-INDEF-OBL COP:PST(3 S), cousin-IZM xwe . . . baqil bû REFL . . . intelligent COP:PST(3 S) ‘Sheikh Mus was the Agha of a tribe, his nephew [. . .] was (very) intelligent’ (Le Coq 1903: 3) Here xwe refers back to the subject of the preceding clause, Sheikh Mus. The dialects of Southeast Turkey and Iraq would not use xwe in this context, but the Oblique personal pronoun wî, in accordance with the rule (195). Likewise, the Kurdish dialect of Gilan uses xwe/xa in a different manner (Shojai 2005). Despite these latter complications, for most of the Northern Group control 7 See Bedir Khan and Lescot (1970: 110–114) for the most useful discussion of variations on the usage of xwe. Dorleijn (1996: 56) found the use of xwe in her data conformed to the general rule outlined above. 222 Kurdish (Northern Group) of reflexive xwe is the single most robust diagnostic of syntactic subjecthood available in the language. 5.3.1.4 The passive Voice is crucial in defining grammatical relations because voice processes involve a shift in the mapping of gramatical relations to semantic roles. Indeed, in some theories, the presence of a passive voice is taken as criterial for the existence of a grammatical relation of ‘subject’ (Siewierska and Bakker 2004). In the Northern Group, there is a reasonably productive passive, though restricted to transitive verbs, based on the intransitive verb hatin ‘come’ coupled with the infinitive. The passive involves the following processes: The syntactic subject is deleted,8 and the O advances to the subject of the passive construction. Examples are the following (data based on examples in Rizgar (1996: 222), slightly modified; the verb ‘drink’ is based on the verb for ‘eat’, plus a preverb): (203) ACTIVE , PRESENT: em vê av-ê ve-di-xw-in 1 PL DEM:OBL water-OBL PREV-IND-eat:PRES-1 PL ‘We (can) drink this water’ (204) PASSIVE , PRESENT: t-ê ve-xwar-in av water IND-come:PRES(3 S) PREV-eat-INF ‘This water is drunk (i. e. is drinkable)’ ev DEM When we turn to the past tenses, we find that precisely the same process applies. In other words, the different morphological marking of A and O in the past tense is irrelevant for the application of the passive: 8 In the emergent language of the press, agented passives are possible, using a variety of adpositions to code the agent phrase, but they are restricted to emergent formal written genres and need not concern us here. The canonical ergative construction 223 (205) ACTIVE , PAST: me ev av ve-xwar 1 PL:OBL DEM water PREV-eat:PST(3 S) ‘We drank this water’ (206) PASSIVE , PAST: ve-xwar-in hat av water come:PST(3 S) PREV-eat-INF ‘This water was drunk (i. e. was drinkable)’ ev DEM In other words, passive manipulates the grammatical relation of subject, but is insensitive to the morphology of the NP bearing that role. It is thus further evidence for the existence of a grammatical relation of subject in the Northern Group. 5.3.2 Summary of the ergative construction The ergative construction in the Northern Group appears to be a purely morphological phenomenon, manifested in the case marking of core arguments, and the agreement markers on the verb. On these criteria, S and O-past align together, to the exclusion of A-past. The evidence from syntax (constituent order, coreferential deletion, control of reflexives and passivization) all converge on confirming the existence of the grouping of A-past and S in the syntax, in much the same way as A-pres and S pattern together. In other words, across both tenses, S and A consistently align together and control certain syntactic processes. Although not all criteria provide equally clear results, it is certainly the case that, with the exception of the complications in coreferential deletion, the ergative morphology of the past transitive constructions is irrelevant for the syntax, which remains accusative throughout. This conclusion mirrors the results for other ergative Iranian languages, as mentioned in Section 1.3: we have yet to find clear evidence for syntactic ergativity anywhere in Iranian. 224 Kurdish (Northern Group) 5.4 Deviations from canonical ergativity It will be recalled that the canonical ergative construction is defined by three properties, repeated here for convenience: 1. The A-past is in the Oblique case 2. The O-past is in the Direct case 3. The verb agrees with the O-past However, not all past transitive constructions conform to this pattern. From a diachronic perspective, the language internal variation can give valuable insights into the type of developments into, and away from, ergativity, and for this reason they are investigated in some detail in the next two sections. Unfortunately, there are practical difficulties involved in establishing the kind of variation found. The main difficulty is that in natural texts, very few constructions can be found which display overtly and unambiguously all three features of the canonical ergative construction. This is particularly true of the spoken language. For example, in one of the most reliable transcriptions of a spoken Kurdish narrative, the story recounted in Kahn (1976: 133–147), I obtained the following figures: Number of transitive past verb forms: 210 (±5, some forms are unclear) Number of clauses displaying all the features of the canonical ergative constructions: 9 (±1) Thus of the past transitive clauses which could potentially provide information on case marking and agreement, only around 5% actually supply all the necessary information. A count of the first 60 pages in Lescot (1940), based on traditional spoken narratives but in a normalized orthography, yields similar figures: of the past transitive constructions, less than 10% display the overt case and agreement features of the canonical ergative construction – see Section 5.4.1 for further discussion of these figures. In other words, the canonical ergative construction so popular in descriptive grammars, linguistic theory, and pedagogical works is, in natural discourse, a rare bird. There are several factors contributing to the paucity of canonical ergative constructions: 1. Ergative constructions are restricted to the past tense. In some narratives, particularly those collected by myself from the Erzurum dialect, the speaker uses the present tense almost throughout. 2. It will be recalled that case marking is neutralized in the presence of an Deviations from canonical ergativity 225 Izafe construction – see discussion in connection with (178). Thus the actual case marking of the arguments is often obscured. 3. In connected discourse, core arguments are often deleted if they are deemed recoverable by the speaker, again rendering the case marking invisible (see below for extensive examples of zero-anaphora). 4. When A and O are both third person singular or plural, it is impossible to determine which controls agreement. 5. Many masculine nouns cannot express case distinctions (see discussion in connection with Table 1), and in some dialects, there has been syncretism of Direct and Oblique cases for second and third singular pronouns (cf. Dorleijn (1996) for information on second singular and Matras (1997) for third singular). Despite the paucity of reliable data on case and agreement patterns, it is nevertheless quite possible to identify certain rather recurrent patterns in the deviations from the canonical ergative construction sketched above. It is also noteworthy that the dimensions of case marking, and agreement may diverge from one another in interesting ways. In the next sections, I will be concentrating on each of these areas in turn, although this should not be taken as implying that there is no interdependence between case and agreement. 5.4.1 Deviant case: The double Oblique The richest source on variant case marking in the Northern Group is Dorleijn (1996). However, there are certain drawbacks with her data. Most of it is based on elicitations of isolated sentences, and on translations of sentences from Turkish (she is centrally concerned with Turkish influence on Kurdish). But the danger that her informants (mostly Turkish/Kurdish bilinguals, including a Kurdish ‘semi-speaker’) may have been replicating Turkish structures, thereby producing utterances that differed from natural spoken Kurdish, cannot be ignored. Furthermore, the fact that isolated sentences were used leads to a quite unnatural concentration of sentences with two overt core arguments. In connected discourse, transitive clauses generally do not contain more than one full NP (see below). And patterns of pronoun deletion do have an effect on case marking and agreement, as will be shown in Section 5.4.4. 226 Kurdish (Northern Group) Finally, she concentrated on the variety of Diyarbakır, which apparently displays some important differences to other varieties of the Northern Group. For example, she concludes that constituent order has a significant effect on case marking and agreement patterns (Dorleijn 1996: 91–92), suggesting that fronting of an O “has repercussions for verbal agreement” and is generally not found with pronouns. This observation is not borne out in other texts, as in for example: (207) hûn min anî-n 2 PL 1 S bring:PST-PL ‘I brought you’ (Lescot 1940: 34) Here the O is a pronoun, it is fronted, yet the verb shows regular ergative agreement with it. For these reasons I will base my discussion on what is attested in the available texts, rather than on Dorleijn’s findings, but I will be relating them to her conclusions at several points. The most widespread type of deviant case marking is the double Oblique construction, in which both A and O are in the Oblique case. Examples are the following: (208) Gundi-yan wan bizor ji hev villager-PL:OBL 3 PL:OBL with.difficulty from each.other kir do:PST(3 S) ‘The villagers pulled them apart with difficulty’ (Baksî 1991: 31) In this example, the verb remains in the formally unmarked third person singular, although both core arguments are plural. This is generally true in such constructions: the third person singular form is the default form for past transitive verb forms, regardless of the number of the O: (209) Îşev min keç-ek-ê di xewn-a xwe da last.night 1 S:OBL girl-INDEF-OBL ADP sleep-IZF REFL ADP di-dît PROG -see:PST(2 S ) ‘Last night I saw a girl in my sleep (=dreams)’ (Zinar 1988: 103) Deviations from canonical ergativity 227 (210) Qe min wan anî-ye diny-ê? 1 S:OBL 3 PL:OBL bring:PST-PERF(3 S) world-OBL ‘Have I (not) given birth to them?’ (Baksî 1991: 33) INTERR These examples should suffice as exemplification (see Matras (1997: 620– 621), and Bulut (2000: 155–158) for further examples). Of the deviations in case marking found in the texts considered, the double Oblique, or more generally, an Oblique marked O-past (with a deleted A-past) are the only ones to reach any level of significance. Dorleijn (1996: 118) also discovered that among the deviant case marking patterns in her data, clauses with an Oblique marked O-past make up the majority. From a total of 1342 past transitive constructions considered, some 58 % had an Oblique O-past. Of these, well over half have an Oblique A-past as well, yielding a double Oblique construction, while the rest have a Direct A-past. In over one third (approx. 38 %) of Dorleijn’s entire data, A-past and O-past are in the same case form. There can be little doubt that in the Northern Group, the commonest deviation from the canonical ergative system involves putting the O-past into the Oblique case. If we assume that changes in case systems are primarily motivated by the pressure to maintain the discriminatory function of case (see Section 4.5 for a critical assessment), a change which leaves A and O morphologically indistinguishable is difficult to account for. Given the extremely widespread attestation of precisely such a change throughout Iranian, it is well worth looking at the double oblique construction in Kurdish in more detail. Let us briefly consider what the logically possible range of changes in the case system might be. Table 6 contrasts the case marking system of the past and the present tenses for transitive constructions the Northern Group, whereby I assume the canonical ergative construction for the past tense. Let us further assume that a change will, initially at least, affect just one core argument at a time. Furthermore, in compliance with the first constraint on the changes in case marking in Iranian, established in Section 4.7, we Table 6. Case marking, past and present transitive clauses P RESENT PAST A O Direct Oblique Oblique Direct 228 Kurdish (Northern Group) assume that that any changes will affect case marking in the past tense, (ignoring possible complications of Innovated Object markers (Section 4.3), because they are unknown in the Northern Group of Kurdish). Thus the set of possible changes is restricted to just two: (a) the A-past could become Direct; (b) the O-past could take the Oblique case. Notice that both of these changes would bring the ergative construction closer to the accusative construction of the present tense: Change (a) would mean that in all tenses, an A is in the Direct case, while change (b) would mean that in all tenses, an O is in the Oblique. In other words, either would achieve a higher degree of Cross-System Harmony, discussed in Section 4.6.2.1. Yet, as was pointed out in 4.7, it is overwhelmingly more frequently the case that change (b) occurs, while change (a) is virtually unattested. A comparison with the data in Dorleijn (1996: 118) supports this finding. The two changes are distributed in her data as follows:9 (a) A-past in the Direct case: approx. 11 % (b) O-past is Oblique: approx. 89 % Closer examination of Dorleijn’s (a) examples, where both A-past and Opast are in the direct case, shows that in “virtually all instances” (Dorleijn 1996: 122), one of the core arguments is in third person singular, where expression of case is irregular anyway, and a clear distinction between Direct and Oblique is not always possible.10 Thus the number of such examples is quite possibly even lower. Whatever the precise figures, there is no doubt that the vastly preferred change is (b): the O-past takes the Oblique. Given that both of the logically possible changes achieve a partial levelling of the differences between past and present tenses, the obvious preference for the double-Oblique is clearly in need of explanation. It is even more puzzling from the perspective of phonological change: all other things being equal, the 9 10 Dorleijn reports that in her total data, 43.5 % of the clauses exhibited one of the two possible changes (the rest remained unchanged ergative, or underwent changes to both A and O). The figures given show the percentage that these two changes account for within that 43.5 %. One example (Dorleijn 1996: 121) contains the A-past seg ‘dog’, treated by Dorleijn as a Direct case. But in some dialects of Turkey, this masculine noun simply does not have a distinct Oblique form. Deviations from canonical ergativity 229 loss of the Oblique on the A-past is surely a more likely type of phonological change than the the addition of an Oblique-marker to the O-past. Yet it is the latter change that actually occurs. One reason for the preference for changing the O-past is to be sought the manner that argument positions are actually realized in natural speech. As mentioned above, past transitive clauses with two overt arguments are rare in Kurdish, and indeed cross-linguistically so (cf. also Section 2.4.4). Newmeyer (2003: 686) presents a summary of the relevant figures, to which I have added a further two: – In Chamorro (Austronesian) only 10% of transitive clauses have two lexical arguments. – In Hebrew (Semitic), 93% of transitive clauses lack an overt subject. – French (Romance) only 3% of clauses contain lexical (as opposed to pronominal) subjects. – In Yimas (Papuan) clauses containing two overt core arguments are “very uncommon” (Foley 1991: 371). – The speech of two-year old children acquiring English is already characterized by the avoidance of full NP-subjects for transitive verbs, as noted by Tomasello (2003: 111). Behind these figures are two factors. First, a general avoidance of clauses containing more than one full NP (the well-known “one piece of new information per clause” constraint). Second, the cross-linguistically characteristic distribution of definite and indefinite NPs in transitive clauses: the A is the favoured role for definite reference, and is therefore usually pronominal or deleted, while the O is the favoured role for introducing new, hence indefinite referents, which are correspondingly coded as full NPs (the pattern Du Bois (1987) refers to as “Preferred Argument Structure”). Kurdish too largely complies with this cross-language tendency. The following figures were obtained from the texts in Lescot (1940), based on a count of all past transitive verb forms between page four and page 60:11 11 One problem in counting the Kurdish data is deciding on the status of the commonest transitive verb, gotin ‘speak’. It is lexically transitive, but it generally lacks a NP object. Furthermore, it often functions merely as a kind of discourse particle, with a vague sense of ‘and (they) say’. I excluded forms of the verb gotin from my counts. 230 Kurdish (Northern Group) Total number of clauses with past transitive verb forms: 437 Of these, 298 lack an overt A-past, i.e., approx 70% Thus around 70% of transitive past clauses are of the form (O)V, and in fact, I believe this is the commonest pattern in the present tenses too, although I yet to do the counting to prove this. In connected discourse, then, the overwhelmingly preferred surface form of a transitive clause in Kurdish is: (211) 0/ A Full NPO Verb What this means is that in actual usage, it is the case of the O, rather than of the A, which is most frequently overtly realized. Therefore, by bringing the O-past into line with the O-pres, we achieve the maximum levelling out of case alignments across past and present clauses, yielding a common surface form for both types of transitive clause: [O-OBL V]. Putting the A-past into the Direct case, on the other hand, would contribute far less to achieving parallelism between past and present tenses, because the A is most commonly deleted anyway. Thus one reason for the widespread preference of the double-Oblique in contrast to a perfectly possible, but scarcely attested “double-Direct”,12 lies in cross-linguistically valid patterns of argument deletion in transitive clauses. On the assumption that changes in the case system will proceed one at a time, there is no other single change that would achieve a greater convergence of case alignment across the past and the present tenses. Note that the move to a double-Oblique actually happens at the expense of sacrificing an otherwise intact system of discriminating A from O (the ergative system), leading me to suggest that maintaining cross-system harmony is the more important determining factor in change. If my explanation is on the right track, then it provides a natural account for the development of double-Oblique systems throughout the Iranian languages. 12 However, it must be mentioned that in Badı̄nānı̄, there is a ‘tendency’ to put the A-past into the Direct case (MacKenzie 1961a: 153, 194), while no examples of an Oblique O-past are mentioned by MacKenzie for this dialect. This may actually reflect a wholesale abandonment of the Oblique in this dialect, influenced by the neighbouring Central Group dialects, but this remains speculative. Deviations from canonical ergativity 231 5.4.2 Deviant agreement: A-dominated plural agreement In a canonical ergative construction, the verb agrees with the O-past in person and number. In what follows, I will concentrate on what appears to be the most widely-spread type of deviation from the canonical form of agreement: agreement with a plural A-past. As will be seen from Table 5, verbal agreement in the plural is −in for all persons (with the exception of Badı̄nānı̄, where a distinct form −ı̄n is used with the first person plural). In other words, in the plural, person distinctions are neutralized and a single suffix codes the meaning plural. Thus a plural O-past, regardless of its person, should be accompanied by the ending −in on the verb. In many instances, this is indeed the case. Typically regular examples from texts are the following: (212) Rovî qas-ek çav-ên xwe firikand-in fox moment-INDEF eye-IZP REFL rub:PST-PL ‘The fox rubbed his eyes for a moment’ (Bozarslan 1982: 45) The O can trigger agreement even if it is not actually present in the clause, as in: (213) tişt-ên ku wî di-got-in thing-IZP COMP 3 S:OBL PROG-say:PST-PL ‘The things that he used to say’ (Cewerî 1986: 53) An example contrasting a plural O-past (214a) with a singular O-past (214b) is the following: (214) a. (Gava dibê şev) bavo kinç-ên stûr li xwe (In the evening) father clothes-IZP thick on REFL di-kir-in, PROG -do:PST-PL ‘(In the evening) father would put on warm clothes,’ xwe di-girt b. çift-a shotgun-IZF REFL PROG-take:PST(3 S) ‘take his shotgun . . . ’ (Şemo 1977: 39) Despite the apparent clarity of the rules regarding verbal number agreement, all texts examined contain numerous counter-examples. The majority of them 232 Kurdish (Northern Group) occur in examples where the syntactic subject (the A-past) has been deleted due to discourse considerations. A typical example is the following: (215) geriya-n ne-geriya-n, 0/ i Siyabend Lê gundîi But villager(PL) look:PST-PL NEG-look:PST-PL 0/ i Siyabend ne-dît-in NEG -see:PST-PL ‘But however much the villagersi looked, (theyi ) did not see(PL) Siyabend’ (Zinar 1988: 44) In the second clause, the A-past has been deleted because it is coreferent with the subject of the preceding verb. Although the O-past is unambiguously singular (the proper noun Siyabend), the verb does not agree with it, in violation of the agreement rules for the canonical ergative construction. Instead, verbal agreement is with the plural A-past. Examples of this type are rife throughout the Northern Group, both written and spoken: (216) di-çû-n taştê, xwar-in, vegeriya-n li PROG -go:PST-PL breakfast eat:PST-PL return:PST-PL PREP cih-ê xwe place-IZM REFL ‘(they) went to breakfast, ate(PL), returned to their place’ (Makas 1897–1926 [1979]: 11, transcription simplified) In these examples, the pronoun is deleted through the entire sequence of clauses, a typical pattern in connected discourse. The transitive verb xwarin should, according to the rules of ergative agreement, agree with an implied singular object, ‘breakfast’, but instead it is plural. (217) hemen li wêderê suwar bû-n hat-in immediately there mounted become:PST-PL come:PST-PL kir-in got-in pirs question do:PST-PL say:PST-PL ‘There (they) immediately mounted (their horses), came, asked, said . . . ’ (Le Coq 1903: 43, transcription simplified) In this example the transitive verbs kirin and gotin are plural, although there is no obvious plural O that they are agreeing with. In the following examples, there are overt O’s (‘the girl’, ‘a room’, ‘a short distance’), obviously singular, yet the transitive verbs still agree with the (deleted plural) A-past: Deviations from canonical ergativity 233 (218) kičiḱ girt-in ô kuš’t-in girl take:PST-PL and kill:PST-PL ‘(They) took the girl and killed (her)’ (MacKenzie 1962: 358) (219) dā-n a tatô w tamô šawē manzil-ak ǰudā at.night room-INDEF separate give:PST-PL to Tato and Tamo ‘At night (they) gave Tato and Tamo a separate room’ (MacKenzie 1995: 27) (220) piçûk di-bir-in roj-a ewil mesafe-yek-e day-IZF first distance-INDEF-IZF small PROG-take:PST-PL ‘On the first day, (they) would take (travel) a short distance’ (Şemo 1977: 32) These examples should suffice as illustration. In fact, as far as I can ascertain, all dialects of the Northern Group exhibit deviations from the canonical ergative construction involving plural agreement with an A-past, albeit with varying frequency from dialect to dialect. According to MacKenzie (1961a: 193), such deviations in agreement patterns are rare in Badı̄nānı̄, the sole example cited by him being (218), from the Zakho dialect of Badı̄nānı̄. This impression is confirmed by the texts of Blau (1975), also Badı̄nānı̄, where such deviations are likewise rare.13 However, more recent sources on Badı̄nānı̄ contain numerous examples, e.g., the following: (221) Se û biçîk-êt gund-î li gurg-î kum dog and small-IZP village-OBL ADP wolf-OBL together bî-n û gurig di selk-ê da kuşt-in become:PST-PL and wolf ADP basket-OBL ADP kill:PST-PL ‘The dogs and small (children) of the village surrounded the wolf and killed(PL) the wolf in the basket’ (Şirîn 1996b: 3) Thus there seems little doubt that Badı̄nānı̄ dialects also show evidence of deviant plural number agreement, although it occurs under more restricted circumstances than in the dialects further North. 13 But not entirely absent, cf. oldaş-a got-in ‘the friends-OBL said-PL ’ (Blau 1975: 176). 234 Kurdish (Northern Group) Although absence of an overt plural A-past in the clause, as in the preceding examples, seems to be a conditioning factor,14 examples can even be found where an overt A-past is present: (222) Dîsa wan yek bi yek Orhan hembêz kir-in again 3 PL:OBL one.at.a.time Orhan embrace do:PST-PL ‘Again they embraced Orhan one after another’ (Cewerî 1986: 67) (223) Herdu heval-ên Sîdar got-in both brother-IZP Sîdar say:PST-PL ‘Both of Sîdar’s brothers said . . . ’ (Cewerî 1986: 50) (224) pîrek-a jî hirç-ik dî-n woman-OBL:PL for.their.part bear-IND see:PST-PL ‘The women for their part saw a bear.’ (Bailey 2005: 116) However, by far the more widespread pattern is that plural agreement with an A-past is restricted to clauses where there is no overt A-past. In what follows, I will refer to this type of deviant agreement as A-dominated plural agreement, because it is the A rather than the O which determines agreement on the verb. Although A-dominated plural agreement is a feature of all varieties of the Northern Group, it has received very little attention in the sources on these languages. Recent textbooks such as Barnas and Salzer (1994) and Rizgar (1996) treat agreement in the ergative construction as if it were fully in compliance with the canonical ergative construction. Yet the texts in these very books contain examples of A-dominated plural agreement: (225) wan şîn-ek-e giran girt-ibû-n 3 PL:OBL mourning-INDEF-IZF heavy take:PST-PLUP-PL ‘They had engaged in a solemn mourning’ (Barnas and Salzer 1994: 141) Blau and Barak (1999: 67–68) briefly mention that regional deviations from the canonical ergative construction can be found, citing Dorleijn (1996), but 14 According to Dorleijn (1996: 146), deletion of a plural A-past always leads to plural agreement on the verb, but this does not apply to Badı̄nānı̄. Deviations from canonical ergativity 235 they fail to note the special status of plural agreement. Kurdoev (1978: 186– 187) also notes the existence of cases of agreement with an A-past, but does not offer any explanation. In most general sources, the assumption is that the canonical ergative construction is the norm, and any deviations must be explained in terms of regional variation, or simply incorrect usage. For example, in the encyclopedic dictionary of Chyet (2003), the numerous cases of Adominated plural agreement in the example sentences receive a “[sic]” from the dictionary compiler (five instances on page 124 alone!). Yet the sheer frequency, and the broad regional distribution of such examples demand a more satisfying explanation than simply sloppy usage. It is quite possible that they may not be deviations, but a rule-based component of agreement for the ergative construction in the Northern Group. Only two traditional sources devote any type of explanation to the phenomenon. Wurzel (1997: 118, fn. 1) states that “sometimes” a plural ending is added to the verb to “avoid repetition of the (A-past) pronoun” (my translation). There is a germ of truth in this explanation, as we shall see in Section 5.4.4.1. The most extensive discussion is from Bedir Khan and Lescot (1970: 314). They note explicitly that A-dominated plural agreement is “une importante et fréquente exception” to the canonical ergative construction, and formulate two conditions under which it occurs, which can be summed up as follows: (226) a. If the A-past expresses a vague sense of ‘they’ (cf. German man, French ils, on), i.e., does not refer to a specific and fully identifiable person or group of persons, and if there is no overt A-past NP in the clause, then the verb will usually carry plural agreement. b. If there is an overt, plural A-past in the phrase, but it is ‘fairly distant’ (“assez éloigné”) from the verb, the verb may also carry plural agreement, even when the O is singular. An example illustrating (226a) is the following: (227) Osman axa hebs kir-in Osman Agha imprisonment do:PST-PL ‘(They) jailed Osman Agha’ (Bedir Khan and Lescot 1970: 314) As far as (226b) is concerned, a great deal depends on how one defines ‘fairly distant’, and how one defines clause (in the French original phrase). An ex- 236 Kurdish (Northern Group) ample provided by Bedir Khan and Lescot (1970) is the following (glosses simplified): (228) ko hînî xwendin û nivîsandin dibûn a. xortên young.people who learning.of reading and writing were b. dihatin û li qeydeyên zimanê xwe dipirsîn came and about rules.of language self enquired c. gramer-ek di-xwest-in grammar-INDEF PROG-want:PST-PL (a) ‘The young people who had learned to read and write (b) came and enquired about the rules of their language (c) (they) wanted(PL) a grammar’ (Bedir Khan and Lescot 1970: 315) This seems to me quite comparable to the examples listed above, where the A-past is discourse deleted. In earlier work (Haig 2004a) I summed up these facts in terms of a more general principle of distance, informally stated in (229): (229) The Distance Principle The likelihood of A-dominated plural agreement increases with increasing distance between the overt A-past and its verb. More recently, Bailey (2005: 105–110) has developed a more refined account of the distance factor, drawing on the work of Chafe (1994) and Lambrecht (1994) and their notions of cognitive activation and topicality. However, as she notes (Bailey 2005: 110), while A-dominated agreement is extremely widespread when an overt A is absent from the clause, even when it is present agreement patterns are not consistent, and are shaped by “subtle factors of topicality”. This is yet another area of alignment in Past Transitive Constructions where semantics and pragmatics impact on the morphosyntax. It is worth noting, however, that the ‘impersonal-they’ reading of Adominated agreement, claimed to hold in (227), is often not exactly translatable into the corresponding ‘impersonal’ of more familiar languages. What is commonly involved is the implied presence of quite a specific group of people, but whose precise number and identity remains undisclosed. In such environments, part way between a truly impersonal ‘they/one’, and merely a Deviations from canonical ergativity 237 distant, but identifiable A, agreement on the verb is in the plural – in both tenses, and an overt pronoun is omitted. A typical example of this kind of half-way context is found in the following example. The clause comes at the end of a passage in which the narrator describes a conversation he has had with an officer of the Russian army. The officer, impressed by his knowledge of languages, offers him a job as an interpreter for the army, which the narrator accepts. The narration continues as follows: (230) Fermendar emir da ku çekê eşkerî li min k-in Commander order gave that uniform army on me put-PL ‘The commander gave order that (they) put-PL an army uniform on me’ (Şemo 1977: 48), glosses simplified. The verb under consideration here is kin, a present plural form of the verb kirin ‘do, put’. Now there is no overt plural subject which can be considered to have triggered the plural agreement. Nevertheless, we can guess from the broader context what the intended referents must be: the soldiers whom we assume will be accompanying the Commander. But the soldiers are not explicitly mentioned in the preceding context. In this example, an overt pronoun (ew) in the second clause would not have been appropriate, as it would imply that the identity of the persons concerned had been unambiguously established in the preceding discourse, which is not the case here. The use of the plural marker on the verb signals exactly this kind of vague, contextually recoverable plurality intended: a group of people whose precise number and identity is unknown, or unimportant, but whose existence can be assumed from the context. Another example I recently heard is likewise typical: I was hoping to get a cup of coffee from the refreshments table at a recent conference, but found to my dismay that there was none left. When I asked a Kurdish friend where the coffee was (in Kurdish), he replied with a single word: rakirin, literally: take.away-PST:PL. What he meant was: the people responsible for the catering (an unspecified number of persons, whose precise identity was unknown) had taken the coffee away. The plural form of the verb, with no accompanying pronoun, is precisely the appropriate way to express this meaning – regardless of tense. This kind of implied reference does not fit well with the traditional descriptive categories such as definite, specific, or impersonal. But it is a very characteristic feature of Kurdish, and arguably of Iranian in general (see be- 238 Kurdish (Northern Group) low). It is regularly expressed by the use of a plural marker on the verb with a corresponding lack of an overt A in the clause. And this construction appears to be productively used in both tenses, hence to some extent cross-cutting the alignment differences between the past and the present tenses. Distance is not the only factor involved in A-dominated plural agreement. A crucial issue is person of the O-past. In general, the verb will always agree with the O-past when it is either first or second person singular (SAP, Speech Act Participant). Thus a clause such as ‘(they) saw me/you’ will, even if the A-past is deleted, still agree with the O-past: (231) van zarok-ên te ez rezîl kir-im DEM :OBL:PL children-IZP 2 S:OBL 1 S disgrace do:PST-1 S ‘Those children of yours disgraced me’ (Metê 1998: 99) A further factor is the semantics of the verbs. Several very common verbs, despite being lexically transitive, do not normally take any overt O NP. The commonest by far is gotin ‘say’. It often introduces direct speech, which takes the form of a finite clause (usually without a complementizer). Whether this clause can be considered the ‘object’ of the verb is debatable, and even if it can, it is not entirely clear what person/number value it should have (presumably third person singular). As it turns out, this verb is extremely prone to A-dominance in agreement. For example, stories often begin with a simple verb form digotin ‘(they) used to say(PST-PL)’. In fact, in many texts number agreement with gotin is generally A-dominated throughout, suggesting that in this respect, gotin is becoming more like an intransitive verb: (232) Hinek-an jî di dil-ê xwe de di-got-in some-PL:OBL too ADP heart-IZM REFL ADP PROG-say:PST-PL ‘And some would say to themselves: . . . ’ (Cewerî 1986: 16) Another verb that is prone to A-dominance is dîtin ‘see’ in the sense of ‘realize, notice’, where it takes as its complement a clause rather than an NP. The interaction with person, and with verb semantics, are suggestive of a second, competing principle at work in determining the controller of agreement. It appears that agreement with the O-past is most likely when the O is maximally salient. By ‘salient’ I mean closest to a prototypical NP in form, Deviations from canonical ergativity 239 and maximally high in animacy. The following hierarchy seems to be applicable (see also Matras (1997) for a similar set of observations regarding the interaction between object characteristics, and agreement patterns): (233) Saliency hierarchy of Objects SAP > definite Pl. non-SAP > definite Sing. non-SAP > indefinite > generic > clause / incorporated O By way of explanation, what I term “incorporated O” here is the nominal element of certain complex predicates, for example baz in the expression baz dan, lit. ‘jump give’=‘jump, run away’. Such elements have lost their argument status entirely, but the verb retains its lexical transitivity – cf. the discussion in Section 1.3.2, and in Haig (2002a). Having established the Saliency Hierarchy in (233), we may formulate the following generalization regarding agreement with past tense transitive verbs: (234) The Saliency Principle The likelihood that a verb will agree with a plural A-past increases with decreasing salience of the O of that verb, where salience is defined in (233) Thus a verb with an SAP Object is least likely to agree with a plural A, while one without an NP Object (for example, a verb introducing direct speech) is very likely to agree with a plural A. One can of course see this in terms of a decrease in transitivity along the lines of Hopper and Thompson (1980), with verbs of speech moving towards the intransitive end of the scale. Although the Distance Principle (229) and the Saliency Principle (234) are useful heuristics in predicting the probability of A-dominance in number agreement, they are not to be understood as a water-tight set of rules. Their value resides in the fact that they explicitly identify factors that are relevant in determining agreement patterns on the verb, and they permit some predictions to be made, which may well be relevant for the diachronic issues. For example, on this account, the prototypical case of A-dominated plural agreement would have a maximally distant (e.g., impersonal) plural A, and a non-NP O. The best candidates for this type of construction are verbs of speech, or thinking, used in the sense of ‘it is said/believed’ etc. As far as I am aware, this is precisely the context where all dialects of the Northern Group do indeed exhibit A-dominance in plural agreement. One might well speculate that the 240 Kurdish (Northern Group) spread of A-dominated plural agreement has advanced progressively up the scales, so that now even an A that is overtly present in the clause may trigger it, and an overt O may be overridden for agreement purposes. Some dialects, particularly those to the North, often do have A-dominated plural agreement in these environments. In fact it is almost the rule whenever the O-past is relatively low in saliency (i.e., third person singular or lower): (235) Wexta kewotk-a zîn-a nazik dît-in When pigeon-PL:OBL Zîn-IZF delicate see:PST-PL ‘When the pigeons saw the delicate Zîn’ (Džalil and Džalil 1978: 46) Why should it be agreement in plural number of all things that is so prone to A-dominance? Examination of longer stretches of connected discourse reveals that plural number agreement with an O appears to be rather labile anyway. Consider the following stretch of narrative, with a sequence of past transitive verbs (from Blau (1975: 102), transcription modified, verbs in bold type): (236) a. pîrejin ra bû, der keft, old.woman got.up went.outside b. çû sûkê. went to.market c. Çu tiştê muhtac bo xo kirî, (?) things necessary for self bought(3 S) d. li piştê hemala kir û îna mala xo. on back porters put(3 S) and brought(3 S) to.house.of self e. Da-ne ber Xace Mehmûdî placed-3 PL in.front.of Xace Mehmûd f. Alîk da dewarê wî Fodder gave horse.of him a) The old woman got up, went out, b) went to market c) bought-SG the things needed d) put-SG (them) on the back of (some) porters and brought-SG them back to her house e) placed-PL (them) in front of Xace Mehmûd f) gave fodder to his horse . . . Deviations from canonical ergativity 241 Table 7. Revised account of agreement with past transitive verbs AGREEMENT SUFFIX : -im -î -in -0/ F IRST PERSON O S ECOND PERSON O P LURAL + − − − + − − − + − − − In lines (c) and (d), the verbs kirî ‘bought’ and kir ‘put’ are singular, although the implied O (the things bought) is plural. Yet in line (e) the verb dane ‘gave’ has a plural ending, presumably to indicate that it was several things that the old woman placed in front of Xace Mehmûd. It is almost as though the speaker felt it necessary to remind the hearer at that point that the woman had bought a number of items. Thus there is obviously a weak connection between the O and the verb in terms of number agreement, allowing a certain amount of choice in applying plural agreement. The conclusion I draw from these facts is that plural ‘agreement’ in the Past Transitive Construction is not solely determined by grammatical factors. That is, it is not slavishly replicating the number and person features of the NP bearing a particular grammatical relation. There are semantic and pragmatic factors which may override grammatical relations, and any account of agreement in the past tenses must take them into account. Note that for plural agreement, there is a single unified plural ending, −in, for all persons. Thus in terms of formal distinctions, the plural is underspecified for person.15 Developing on these ideas, we can re-state the agreement facts for past tense transitive verbs in the form of a feature matrix, based on just three features. This is shown in Table 7.16 The relevant features for verbal agreement are: first person of the O, second person of the O, and plural (of either A or O). Third person singular (traditionally assigned a −0/ ‘suffix’) 15 16 With the exception of the Badı̄nānı̄ dialect, where a distinct ending −ı̄n for the first person plural is maintained. One might speculate whether the finer formal differentiation of plural marking in Badı̄nānı̄ is a reason for its greater consistency in agreeing with the O-past. See Dorleijn (1996: 130–135) for similar proposals on the possible shape of the agreement system. 242 Kurdish (Northern Group) is simply lack of agreement, or alternatively, not first/second person, and not plural. 5.4.2.1 Diachronic implications While the breakdown of agreement pattern is generally taken as part of the global demise of ergativity in Kurdish (Pirejko 1963, Dorleijn 1996), it has not been sufficiently recognized that A-dominated plural number agreement is fundamentally different. It is not merely a regionally restricted, perhaps contact-induced occasional breakdown, but a relatively systematic process that occurs, albeit with different frequencies (Badı̄nānı̄ has the least examples) in all attested varieties of the Northern Group. Therefore, explaining A-dominated plural agreement as a symptom of the breakdown of ergativity fails to account for the fact that it is so regularly encountered across geographically very diverse dialects. An alternative view would be that A-dominance in plural agreement is not evidence of a change in progress, but reflects a characteristic of the common Kurdish proto-language: A-dominated plural agreement is not a new development, but was there from the outset. Is there any evidence in favour of such a claim? Given the lack of records for the immediate ancestors of the Northern Group, this question cannot be answered with any certainty at present. Some facts are nevertheless worth recalling. First, plural agreement with verbs is often semantically rather than grammatically determined in Kurdish in both tenses.17 Examples such as the following are typical for Kurdish: (237) her yek-î cigar-ek vêxist-in each one-OBL cigarette-INDEF light:PST-PL ‘Each one lit a cigarette’ (Cewerî 1986: 4) Although her yekî is grammatically singular, it can be construed as semantically plural in that it implies more than one person, and it is the semantics that evidently trigger the plural ending on the verb. Thus semantic plurality tends to take preference over grammatical singularity generally in Kurdish. 17 See Mann (1906: CV) for Mukri (Central Group), and Hadank and Mann (1930: 159– 172) for Gorani, and more general discussion. Deviations from canonical ergativity 243 The second piece of evidence in support of seeing A-dominance in agreement as an old, common Kurdish feature, rather than an innovation hailing the breakdown of ergativity, comes from earlier stages of Iranian. Agreement based on semantics rather than syntax is discussed for earlier stages of Persian in Lazard (1963: 455–460), based on texts from the 9–11 C . AD. A typical example of the kind of “impersonal” use of plural marking is the following, here with a present-tense verb form gōy-and: (238) gōy-and ki malik i čı̄n rā sı̄ s.aD u šas.t say:PRES-3 PL that king of China I NN O BJ 3 hundred and six nāh.iyat dāraD provinces have:PRES:3 S ‘(They) say that the King of China has 306 provinces.’ (Lazard 1963: 376) Heston (1976: 164–166) notes that in Early New Persian, Pahlavi, Sogdian and Khotanese a third person plural verb without an overt subject can be used to express an impersonal ‘they’, often with the implied reference of servants or attendants. This parallels precisely the notion of ‘impersonal’ that was introduced above in connection with (230). The ‘they’ construction is reported to be particularly frequent with verbs of speech (Heston 1976: 226,fn. 10), again just as it is with Kurdish gotin. Strikingly, the construction is used in Pahlavi and Sogdian with the “passive-preterite” constructions, one of the predecessors of the ergative construction. Heston (1976: 177–178) cites a Pahlavi example with plural agreement with an A-past in precisely an environment where the A-past is not present in the clause itself, but must be recovered from the preceding context. Finally, there are striking parallels to the way in which plural agreement is used in sequences of same-subject clauses, discussed in Chapter three in connection with (120), and in Section 5.4.4 below for Kurdish. Although the data from the earlier languages is patchy, even a cursory survey reveals striking commonalities in the way plural agreement works in Past Transitive Constructions in Kurdish, and in the earlier Western Iranian languages: a plural A tends to trigger plural agreement on the verb when it is not overtly present in the clause, when it does not have specific reference, and in combination with verbs of speech and naming. In Badı̄nānı̄, it appears to be restricted to these contexts, but in the rest of the Northern Group it is 244 Kurdish (Northern Group) found in a much larger range of contexts. Although none of the languages investigated by Heston (1976) is a direct ancestor of Kurdish, the existence of plural agreement with an A-past in these languages, and in Kurdish, seems to much of a coincidence. Thus what has been claimed to be a ‘deviation’ in Kurdish actually has parallels in related languages, going back centuries. We cannot therefore rule out the possibility that Kurdish has always had Adominated plural agreement in the past tenses. If that is the case, agreement in the Past Transitive Construction may simply never really have ‘gelled’, hence remaining subject to semantic and pragmatic demands, rather than being a pure reflection of grammatical relations. 5.4.3 Summary of agreement According to standard descriptions of Kurmanji, agreement in past tenses is the mirror image of that in the present tenses: it is determined entirely by grammatical features of person and number on the O-past. Closer inspection shows that this is not the case. I proposed a simplified model of agreement for past transitive clauses, according to which just three values are relevant; first person of the O, second person of the O, and plural with either A or O. ‘Third person singular’ is simply the absence of any of these values. The issue of when a past tense verb agrees with a plural A-past was examined in some detail. I identified two principles relevant here: the Distance Principle (distance between the last overt mention of the A-past and the verb), and the Saliency of the O: highly salient Os tend to determine agreement on the verb; less salient Os can be overridden by a plural A. As we have seen above, the plural ending may reflect plurality of the A as well as the O. Thus the plural ending expresses a vaguer notion of plurality, not strictly determined by person, and crucially, not strictly determined by the grammatical relations of A and O. I believe that an adequate account of plural agreement in the Northern Group must abandon the notion of ‘agreement’ in the strict sense of ‘obligatory cross-referencing of grammatical features’, and accept that number agreement is a looser kind of reference tracking device, where pragmatic and semantic factors combine. Furthermore, there is suggestive evidence from earlier Western Iranian languages that plural agreement with Past Transitive Construction may always have worked in this manner. Deviations from canonical ergativity 245 Thus A-dominated agreement is not necessarily a symptom of the breakdown of ergativity, but part and parcel of the West Iranian brand of ergativity from the outset. 5.4.4 Clause linkage and agreement A statement of the agreement rules solely in terms of the intra-clausal grammar is inadequate. In this section I will show how factors shaping the way sequences of clauses are linked in discourse also interact to co-determine the type of agreement found in past transitive verbs. There is a strong cross-language tendency for subjects to be omitted when they are highly topical. Thus in sequences of clauses with coreferential subjects, the subject in the clauses following the first mention are generally omitted. Givón (1983: 17–20) claims that zero anaphora is the most favoured strategy for coding continuous topics (in most cases, equivalent to grammatical subjects). Kurdish is an excellent example of a language that makes copious use of zero-anaphora to code coreferential subjects – see for example the sequence of clauses with common deleted subject in (236). We can formulate this tendency as follows: (239) Delete Common Subjects If a clause has the same subject as the immediately preceding clause, delete the subject pronoun. While much of Kurdish discourse conforms with (239), there is a second principle at work which tends to counteract it. In sequences of same-subject clauses, particularly when the subjects are deleted, there is a strong tendency to avoid discordant agreement patterns on the verb. Now in the present tense, where verb agreement is always with S or A, any sequence of same-subject clauses will always have identical verb agreement, regardless of the transitivity of the verb. But in Kurdish, as soon as past tense verb forms are involved, a potential problem arises. In Section 5.3.1.2 examples were presented with a transitive past tense verb followed by a present tense verb, for example (193), repeated here for convenience: 246 Kurdish (Northern Group) (193) min di-xwest [ez her-im mal-ê] 1 S:OBL PROG-want:PST(3 S) 1 S go:IRR:PRES-1 S home-OBL ‘I was wanting to go home’ (Matras 1997: 627) In the second clause, there is an overt subject pronoun, despite referential identity with the subject of the preceding clause. Such examples clearly violate the principle of same-subject deletion formulated in (239). Examples of this nature demonstrate that ergative morphology may in fact disrupt the expected patterns of inter-clausal syntax. The most detailed investigation of the deletion of coreferential pronouns is Matras (1997), who uncovers an exceedingly complex interplay of factors.18 The most important are: the person of the pronoun, the transitivity of the verbs involved, the presence or absence of an overt complementizer (ku), and the control semantics of the matrix verb, hence the type of semantic linkage between matrix and subordinate clause. The main findings are that (a) deletion of a third person pronoun is more likely than a first person pronoun; (b) deletion is more likely in clause sequence sharing the same transitivity value (the effect of the complementizer ku, and of verb semantics are ignored here; with regard to the former, in the spoken narratives I have examined an overt complementizer is extremely rare in the relevant constructions). The first claim is certainly confirmed in the texts I have observed. There is undoubtedly a strong tendency to maintain an overt first person pronoun, whereas third person (singular) pronouns are regularly omitted. As far as the effects of transitivity are concerned, I have not investigated the matter in sufficient detail to make firm predictions. It is certainly true that sequences of same-subject intransitive clauses allow deletion of pronouns most freely, but of course in intransitive clauses, there is no clash of case and agreement patterns across the clauses anyway, regardless of the tense values. It is only when past transitive clauses are involved that restrictions can be expected. Three combinations are possible: (1) transitive-transitive; (2) transitive-intransitive; (3) intransitive-transitive. Matras (1997: 643) suggests that in general, deletion of a coreferential subject pronoun is more likely in a transitive-transitive 18 More recently, the grammar of case in same-subject clause sequences differing in alignment has been briefly discussed for Kham (Tibeto-Burman) by Watters (2002: 328). This would be an extremely interesting avenue for typological research. Deviations from canonical ergativity 247 clause sequence than in the other types, though this is also dependent on the type of clause sequence involved (coordinated clauses behave somewhat differently to complements of ‘want’-predicates etc.). It needs to be stressed, however, that most of the findings are tendencies, which allow exceptions to varying degrees. Recall the following two examples, repeated here for convenience. In both examples, a coreferential first person pronoun is retained, although one of the subordinate clauses is intransitive, while the other is transitive: (193) min di-xwest [ez her-im mal-ê] 1 S:OBL PROG-want:PST(3 S) 1 S go:IRR:PRES-1 S home-OBL ‘I was wanting to go home’ (Matras 1997: 627) (194) min xwest [ez otomobîl-a xwe bi-firoş-im] 1 S:OBL want:PST(3 S) 1 S car-IZF REFL IRR -sell:PRES-1 S ‘I wanted to sell my car’ (Cewerî 2001) The most robust finding concerns the effects of grammatical person: with third person singular subjects, pronoun deletion appears to be the rule when the matrix and dependent verbs are semantically tightly integrated, as in destpêkirin ‘begin’, karîn ‘be able’, and xwastin ‘want’ (when the two clauses are semantically more loosely connected (e.g., with zanîn ‘know’), this does not necessarily hold). Consider the following examples, with third person singular subjects and pronoun deletion in the second clause: (240) mal-ê di-xwest 0/ i her-e wîi 3 S:OBL PROG-want:PST(3 S) 0/ go:IRR:PRES-1 S home-OBL ‘He wanted to go home’ (Matras 1997: 629) (241) 0/ i ew ragir-e keçik-êi ne-kar-î girl:OBL NEG-be.able:PST(3 S) 0/ 3 S lift.up:PRES:IRR-3 S ‘The girl was unable to lift him up’ (Bozarslan Undated: 6o) In contrast, for first and second person subjects, pronoun retention in the second clause is widespread, as shown in (193). Unfortunately, retention of first/second person subjects is only a preference. Counter-examples can be found, such as the following two: 248 Kurdish (Northern Group) (242) min di-xwest 0/ av vexw-im 1 S:OBL PROG-want:PST(3 S) 0/ water drink:IRR:PRES-1 S ‘I was wanting to drink water’ (Matras 1997: 627) (243) min ne-di-zanî 0/ nav-ê xwe 1 S:OBL NEG-PROG-know:PST(3 S) 0/ name-IZM REFL bi-nivîs-im IRR -write:PRES-1 S ‘I did not know how to write my name’ (Şemo 1977: 23) Nevertheless, there is little doubt that deletion of third person is vastly commoner, as predicted by Matras (1997). There are two reasons for the greater propensity to omit third person singular pronouns. First, as Matras (1997) notes, the Oblique and the Direct forms in the third person singular have, in some spoken varieties, merged. Thus there is no conflict between the form of the main clause subject and the subject of the subordinate clause. Second, with a third person singular subject, the agreement markers on the first and second verbs are usually congruent. Agreement on the past tense of the verb xwastin is generally zero, i.e., third person singular (arguably, the verb ‘agrees’ with the second clause, or it has default third person zero agreement). Likewise, the present tense verb that follows has third person singular agreement. (240) is a typical example showing congruent agreement patterns with the third person. In such environments, deletion of a subject pronoun appears to be preferred (Matras 1997: 645). There is also good evidence from other parts of the grammar that sequences of same-subject verbs are preferred when their agreement patterns are congruent (see below). In other words, what is generally avoided is a sequence of same-subject clauses, with deletion of the second pronoun and non-equivalent agreement values on the verbs. This tendency can be formulated as follows: (244) Avoid Discordant Agreement Sequences Avoid sequences of same-subject clauses, with deleted subjects but with non-equivalent (i.e., discordant) agreement markers on the verb. There are basically two strategies employed to avoid violations of (244). The first is to insert a pronoun in the second clause. We have already seen such examples in connection with the verb xwastin ‘want’. Example (194), repeated here for convenience, demonstrates this strategy: Deviations from canonical ergativity 249 (194) min xwest ez otomobîl-a xwe bi-firoş-im 1 S:OBL want:PST(3 S) 1 S car-IZF REFL IRR -sell:PRES-1 S ‘I wanted to sell my car’ (Cewerî 2001) It will be seen that although the two verbs share a common subject, they have distinct agreement values. In such cases, the Principle of Common Subject Deletion given in (239) is overridden, and a pronoun is retained in the second clause. We can refer to this as the pronoun insertion strategy. Although it is widely used in cases such as (194), it invokes a certain cost, in that the resultant structures run counter to the strong tendency to delete common subjects. A second strategy works rather differently. Rather than insert an additional pronoun, the agreement value on one of the verbs is changed in order to bring it into line with the other verb. In order to understand this strategy, it is useful to look first at an unproblematic case of clause linkage, involving a sequence of two intransitive verbs: (245) 0/ i çû-m Ezi rabû-m nik mela-yê 1 S get.up:PST-1 S 0/ i go:PST-1 S PREP Mullah-OBL ‘Ii got up and 0/ i went to the Mullah’ (Bozarslan Undated: 58) Here the first singular pronoun is in the Direct case and the verb rabûm agrees with it. Likewise, the following verb shows an identical agreement pattern, and the pronoun is deleted. Thus this sequence satisfies both the Principle of Subject Deletion, and the Principle of Avoiding Discordant Agreement. But when the second verb is transitive, and in the past tense, there is a problem, because the two verbs would not have the same agreement pattern. Particularly with the intransitive verbs çûn ‘go’, hatin ‘come’ or rabûn ‘get up’, what often happens is that the Oblique pronoun required for the A-past of the second clause is placed at the beginning of the entire sequence, and the first verb is put into the default third person singular form. The following example is from the same text as (245): (246) rabû 0/ i ber-ê mini xwe da nik 1 S:OBL get.up:PST(3 S) 0/ i head-IZM REFL give:PST(3 S) PREP miftîyê Diyarbekir Mufti-IZM Diyarbakir ‘I got up and 0/ i headed (lit. gave my head) to the Mufti of D.’ (Bozarslan Undated: 59) 250 Kurdish (Northern Group) This strategy appears to leave an Oblique pronoun, min, as the subject of an intransitive verb, rabû, which clearly violates the clause-internal grammar of Kurdish. But it achieves a remarkable gain in terms of the grammar of clause linkage, because it leaves both verbs with third person singular agreement, and permits pronoun deletion in the second clause. This strategy is briefly mentioned by Blau and Barak (1999: 67), although they suggest that it is restricted to the verbs hatin ‘come’ and çûn ‘go’, which is not correct. They provide further examples, such as the following: (247) hat 0/ i xwarin da tei me 2 S:OBL come:PST(3 S) 0/ i food give:PST(3 S) 1 PL:OBL ‘You came and gave us food’ (Blau and Barak 1999: 67) (248) mini çû 0/ i cot-ek sol li bazar-ê kirî 1 S:OBL go:PST(3 S) 0/ i pair-INDEF shoe at market buy:PST(3 S) ‘I went and bought a pair of shoes at the market’ (Blau and Barak 1999: 67) Blau and Barak (1999) suggest that in such contexts, the verbs çûn and hatin “behave like transitive verbs”. However, the more important point appears to me not so much that they ‘take an Oblique subject’, but rather that the entire sequence of clauses is treated as a single unit in terms of case assignment. For this reason I will refer to this strategy as the verb serialization strategy, because it subordinates the case and agreement alignment of the intransitive verb to that of the transitive verb. The verb serialization strategy is predominantly used with first and second person subjects,19 where a conflict of agreement patterns would arise. With a third person singular subject, both the initial intransitive and the subsequent transitive verb usually carry third person singular agreement, thus there is no clash of agreement values. This means that the two clauses may be juxtaposed with pronoun deletion in the second clause, satisfying Delete Common Subjects, and requires no further adjustment to the verbs themselves: 19 One of the four examples cited by Blau and Barak (1999) involves a preposed third person singular Oblique subject. I have also come across such an example: Ewî jî hat, bi edeb kumê xwe rakir . . . ‘He( OBL ) too came, 0/ raised his hat politely . . . ’ (Şemo 1977: 22). However, this strategy is undoubtedly much more frequent with first and second person subjects. Deviations from canonical ergativity 251 (249) feqîr çû ji xwe re nan anî poor go:PST(3 S) ADP REFL ADP bread take:PST(3 S) ‘The poor (man) went (and) took bread for himself’ (Lescot 1940: 38) The motivation for verb serialization is to be sought in the principles of clause linkage, described above, to which clearly clause-internal grammar can be sacrificed. The serial verb strategy is also attested in other Iranian languages, for example Balochi (a simplified gloss has been added): (250) kit.agā šu, dān git, ārt grasshopper:PL:OBL went grain bought brought ‘The grasshoppers(OBL) went, bought grain and brought (it)’ (Korn forthc. b, citing Farrell 2003) Here the common subject is in the Oblique case, as required by the second and third (transitive) verbs, but not by the first. Note finally that the verb serialization strategy is not the only solution found in intransitive-transitive sequences. Probably commoner is the pronoun insertion strategy mentioned above: (251) em rabû-n û me rê şaş kir 1 PL get.up:PST-PL and 1 PL:OBL way confused do:PST(3 S) ‘We set off and we got lost’ (lit. ‘confused the way’) (Ritter 1971: 10, transcription modified) It is probably significant that in this example the two clauses are linked by the conjunction û ‘and’. The verb serialization strategy is not possible when the two clauses are separated by a conjunction, at least I have found no counterexamples in texts to this generalization in texts. This would also tie in with the observations of Matras (1997) that the use of an overt clause linker reduces the degree to which sequential clauses are fused together, and hence increases the likelihood of pronoun retention. The third strategy for adjusting clauses to comply with the principles of clause linkage is to adjust the agreement of the second clause, bringing it into line with that of the preceding one. This strategy is restricted to plural number agreement, but it is extremely common there, and in probaby most dialects it now has the status of a rule. Numerous examples of this type have 252 Kurdish (Northern Group) already been provided in Section 5.4.2, for example (221), so no further illustration is required here. 5.4.4.1 Summary of clause linkage and agreement This section has surveyed various non-canonical types of agreement found in Past Transitive Constructions. The main claim is that agreement is not strictly grammatically determined, but subject to the demands of inter-clausal grammar and semantics. The simplest account of the variation found is in terms of compromise solutions in meeting the pressures of discourse structure, and the demands of clause-internal grammar. The relevant principles for discourse are the following: (252) Delete Common Subjects If a clause has the same subject as the immediately preceding clause, delete the subject pronoun. (253) Avoid Discordant Agreement Sequences Avoid sequences of same-subject clauses where the non-initial clause has a deleted subject and an agreement marker on its verb that is distinct from the preceding verb. In some contexts, these two principles are in competition with each other, and in competition with the demands of clause-internal grammar. The three ‘compromise’ strategies attested are: 1. Pronoun retention. This strategy satisfies the demands of clause-internal syntax, and it also satisfies Avoid Discordant Agreement. However, it violates Delete Common Subjects. 2. Verb serialization. This strategy satisfies Delete Common Subjects and Avoid Discordant Agreement, but violates clause-internal principles of case-assignment and agreement. 3. Adjustment of verb agreement. This strategy satisfies Delete Common Subject and Avoid Discordant Agreement, but violates clause-internal rules of agreement. There are two main lessons to be learned from this section. First, we have seen that an adequate statement of the rules of agreement cannot be formulated without reference to the facts of clause combining: pressures from clause linkage may, under certain conditions, override the rules of clause- Deviations from canonical ergativity 253 internal grammar. These apparently deviant structures may only be understood if sufficient attention is paid to the larger discourse context. These facts bear further witness to the necessity for investigating connected discourse rather than relying solely on isolated elicited sentences. The second point that needs to be driven home is that the various types of deviant (non-canonical) case and agreement patterns described above always occur in connection with past transitive constructions. In other words, the grammar of these constructions is more vulnerable to pragmatic and semantic factors than the grammar of present-tense clauses. In a sense, one could consider the past transitive construction less strongly grammaticalized, hence its basic features (case and agreement) are more sensitive to contextual factors. 5.4.5 Full accusativity: Gilan Kurdish This chapter would not be complete without a brief description of a variety of the Northern Group in which Tense Sensitive Alignment has been completely abandoned, with all tenses showing accusative alignment in both case and agreement. This has happened in the Kurdish of Gilan, a region northwest of Tehran near the Caspian sea and described in detail in Shojai (2005).20 The history of the Kurds of Gilan is complex. In the sixteenth century the Persian ruler Shah Abbas resettled a large number of Kurdish tribes to the northeastern province of Khorasan, where a significant Kurdish population remains to this day. In the eighteenth century, around 2000 Kurdish families were subsequently displaced from Khorasan to Gilan in order to secure the border against Russian intrusions. Precise figures on the numbers of Kurdish speakers currently living in Gilan are not available, but Shojai (2005) mentions eight villages as “Kurdish speaking”. The area is characterized by multilingualism, with Talyshi and Persian currently the most important contact languages, though several other languages have impacted on the development of Gilan Kurdish over time (e.g., Azeri, Uzbek, Turkmen and Mongolian). 20 I am extremely grateful to Behrooz Shojai for drawing my attention to Gilan Kurdish, and for sharing with me his extensive knowledge of many related issues. 254 Kurdish (Northern Group) Due to their geographic isolation from the main body of Northern Group speakers, and to their Shiite beliefs, the Gilan Kurds have little contact to other speakers of Northern Group Kurdish. Nevertheless, the lexicon and morphology of Gilan Kurdish make it quite obviously a member of the Northern Group. The set of personal pronouns is essentially that shown in Table 2. It also uses the indefiniteness suffix −ek in a comparable manner to other Northern Kurdish dialects, and it has preserved an Oblique case marker on nouns, although it is no longer differentiated according to gender. The verb system also shows basically the same inventory of morphemes, and indeed, Behrooz Shojai (a native speaker of the Northern Group) had little difficulty in understanding Gilan Kurdish (Behrooz Shojai, p. c.). However, unlike the rest of the Northern Group, alignment in Gilan Kurdish is fully accusative throughout all tenses. Consider the following past-tense transitive clause: (254) tu şutur-ê biyaban-ê venda k’ir-î 2 S:DIR camel-OBL desert-OBL losing do:PST-2 S ‘You lost the camel in the desert’ (Shojai 2005: 18) Here the A is in the Direct case, the O is Oblique-marked, and the verb agrees with the A. This is essentially the canonical accusative construction, as it is found in all present tenses: (255) xudavend-ê t’ê bi-bîn-im ez 1 S:DIR God-OBL FUT IRR-see:PRES-1 S ‘I shall see God’ (Shojai 2005: 18) Full illustration of alignment in Gilan Kurdish is unnecessary; Shojai is absolutely explicit in claiming, and illustrating, that Gilan Kurdish has completely lost any trace of Tense-Sensitive Alignment. It thus appears to be the variety that has undergone the most radical changes in alignment when compared to what can be assumed for the common ancestor of the Northern Group. But what is quite remarkable is that the change to full accusative alignment has been undertaken without any additional morphology: the ‘Accusative’ marker in the past tenses is the old Oblique marker. Gilan Kurdish, unlike Persian and other Iranian languages lacking Tense Sensitive Alignment, has apparently not adopted an innovated Object marker such as Persian rā. It is extremely likely that the predecessor of the past-tense accusative construction such as (254) was a double-Oblique, with both A and O in the Summary of deviations 255 Oblique case. But Gilan took this construction a step further by dropping the Oblique case on the A, and shifting verbal agreement to be determined completely by the A. Unfortunately, we lack evidence for the intermediate stages of this development, although more detailed studies of Khorasan Kurdish might shed more light on the matter. Why did Gilan Kurdish proceed down this pathway? Was it due to contact influence (but which contact language?), isolation, a break in the normal transmission of the language leading to imperfect learning at some point, or some combination of these factors? Any answers to this question remain pure speculation. For now we must be content to note that the complete loss of non-accusative alignments without recourse to additional case morphology, is a possible outcome of alignment changes in Iranian. 5.5 Summary of deviations On most standard accounts, the canonical ergative construction is typically portrayed as the mirror image of the accusative construction: a neat reversal of case and agreement. However, closer examination of actual texts, and of regional variation, reveals a number of subtle differences. As far as case marking is concerned, the archaic feature of the A-past in the Oblique case is surprisingly stable throughout the Northern Group, Gilan Kurdish being probably the sole regular exception. Where instability occurs, it is the case marking of the O-past which is most vulnerable to change. The commonest development is one which leads to an Oblique O-past, hence a double Oblique construction. From a ‘discriminatory’ perspective on case marking, such a development is hard to explain. However, I have suggested that the prime motor of change is not preserving the discriminatory function, but maintaining (or re-establishing) a unified canonical clause structure across the past and the present tenses. The minimal change necessary to achieve a high degree of uniformity in actual usage is to put the O-past into the Oblique case. I have argued that the reasons for this change are to be sought in the characteristic distribution of noun phrases across syntactic functions in discourse: the preferred form for transitive clauses in all tenses is O-V, while the A is regularly deleted. This means that case is most visible (i. e. most frequently realized) on the O, so that bringing the case of the O-past into line with the 256 Kurdish (Northern Group) case of the O-pres achieves a maximum of cross-tense conformity in case marking. The differences between the present and past tenses are most striking in the area of verbal agreement. Agreement in the past transitive construction is not simply ‘agreement with the O-past in person and number’, in the way that agreement in the present tenses is with S and A. Rather, agreement in Past Transitive Constructions is subject to considerable variation, mediated by discourse-pragmatic and semantic considerations. The clearest area of difference is the manner in which number is reflected on the verb in the past tenses. Throughout the Northern Group, number agreement can be either with the O-past, or with the A-past, depending on two factors, Distance and Saliency of the O-past. I term this A-dominated plural agreement. This is obviously quite different to number agreement in the present tenses, which is exclusively with the A, hence determined by clause-internal morphosyntax. I have argued that agreement in the past tense is underspecified. Only three values are recognized: first person O, second person O, and plural. In this sense, it is already impoverished, and less ‘grammatical’ than that of the present tenses, where agreement is determined exclusively by the grammatical relation of Subject. In the past tenses, the tendency is for agreement to atrophy, so that for some dialects it has been claimed that past transitive verb forms show no agreement, as in the Diyarbakır dialect investigated by Dorleijn (1996). As for the chronology of these developments, I have suggested that at least as far as A-past dominant plural agreement is concerned, it may well reflect an older, common Proto-Kurdish characteristic rather than a new development, because comparable phenomena can be found in much earlier stages of Iranian. This explanation would account for the distribution of Apast dominant plural agreement in certain contexts throughout the Northern Group. If that is the case, the agreement pattern of Kurdish never was fully ergative. 5.6 Evidence from Badı̄nānı̄ In the preceding sections we have examined a number of deviations from the canonical ergative construction, and have worked on the assumption that they represent chronologically subsequent developments of the latter. Thus Evidence from Badı̄nānı̄ 257 we have mostly been concerned with the loss, or ‘decay’ (Dorleijn 1996) of ergativity in the Northern Group. But until now, relatively little has been said on how the Northern Group arrived at its particular brand of ergativity in the first place. In other words, how can we relate the alignment systems of the Northern Group to the kind of system that we have assumed for Old and Middle Iranian in Chapters two and three? On the face of it, there is relatively little common ground. The Northern Group completely lacks the pronominal clitics that played such a pronominent role in the emergence of Tense Sensitive Alignments. Likewise, we have as yet seen no evidence of non-canonical subjects, which I claimed to have been pivotal in the emergence of ergativity in past tenses. In this section I focus on a variety of the Northern Group which I suggest has preserved a number of extremely archaic features in its syntax. These features allow us to reconstruct a more detailed picture of how the ergative alignment of the Northern Group may have emerged. The dialect concerned is Badı̄nānı̄. The name (under various spellings) is loosely applied to the southernmost dialect of the Northern Group of Kurdish, spoken around the townships of Zakho, Dohuk, Amadiye and Akre. Lexically, and in the basic features of its morphology Badı̄nānı̄ is undeniably a member of the Northern Group (and mutual intelligibility with the dialects of Southeast Turkey is high). Like the other dialects of the Northern Group, Badı̄nānı̄ completely lacks the pronominal special clitics which are characteristic of most other West Iranian languages. Although spoken Badı̄nānı̄ has been admirably documented in MacKenzie (1961a), MacKenzie (1962) and Blau (1975), those who have investigated ergativity in the Northern Group, e.g., Bynon (1979) or Dorleijn (1996), have taken the descriptions of ‘Standard Kurmanji’ as representative of the Northern Group as a whole. As a result, certain aspects of Badı̄nānı̄ syntax have been ignored, and the resultant picture of the emergence of ergativity in Kurdish is seriously flawed. 5.6.1 Non-canonical subjects Like the other members of the Northern Group, Badı̄nānı̄ also exhibits an ergative construction showing the properties already discussed in Section 5.3. However, Badı̄nānı̄ also has a number of other constructions which are no longer possible in the dialects of Turkey and the Caucasus (or at best pre- 258 Kurdish (Northern Group) served in isolated idiomatic phrases). The key notion involved in these constructions is that of non-canonical subject, as it was developed in Section 1.4. Such non-canonical subjects occur in a number of different roles in Badı̄n., including External Possessor, Experiencer, and Needers/Wanters. Examples of each are provided below. 5.6.1.1 External Possessors Throughout Kurdish, simple assertions of existence are expressed with a particle ha-/he- plus a form of the copula būn (this combination is sometimes analyzed as a monomorphemic verb hebūn): (256) ha-ya mirôv-ak man-INDEF : SG existent-COP:PRES:3 S ‘There is a man’ (MacKenzie 1961a: 191) (257) l zamānakı̄ mirôv-ak ha-bô at.a.time man-INDEF : SG existent-COP:PST:3 S ‘Once upon a time there was a man’ (MacKenzie 1962: 284) In Badı̄nānı̄, such existential predicates can be extended via the addition of a fronted Oblique, yielding an External Possessor construction (see also the description in Şirîn 1996a: 30–31): (258) sē kur̄ ha-bô-n. naqlakē hāḱim-ak-ı̄ at.a.time ¯prince-INDEF : SG- OBL three son existent-COP:PST-PL ‘Once a prince had three sons’ (lit. once to-a-prince three sons existed) (MacKenzie 1962: 320) (259) ta qalam ha-ya? 2 S:OBL pen existent-be:PRES:3 S ‘Have you got a pen?’ (lit: to-you is there a pen?) (MacKenzie 1961a: 191) In Section 2.5.2 it was already noted that External Possessor Constructions are widespread in many Indo-European languages, and are considered to have been characteristic of the earliest stages of the family (cf. for example Drinka (1999: 472) and Bauer 1999: 592). I therefore assume that their presence in Badı̄n. represents a relic, rather than an innovation. The lack of such structures in the rest of the Northern Group, on the other hand, is an innovation. Evidence from Badı̄nānı̄ 259 A recurrent issue in the analysis of such constructions is the syntactic status of the fronted Oblique Possessor. In the available texts, no examples were found with a reflexive pronoun, which could have provided the most reliable indication of subject status. However, examples such as the following are at least very suggestive that the fronted Oblique is potentially a controller of coreferential deletion across coordinate clauses: (260) sē kur Ha-bı̄ hākim-ak, ¯ existent-COP:PST:3 S prince-INDEF : SG, three son ha-bı̄-n gal kičak existent-COP:PST- PL with daughter ‘There was a prince, (he) had three sons and a daughter’ (lit. A princei was (and) 0/ i three sons and a daughter existed) (MacKenzie 1962: 348) Here the subject of the second clause, semantically the Possessor, appears to have been deleted under coreference with the subject of the initial existential construction. This might of course be pragmatically-driven omission of a highly topical constituent. Or it may be nascent subjecthood. Note, however, that the EPC is not the only way to express possession. More commonly, possession is expressed adnominally by means of an Izafe construction. In many contexts, the Izafe is the only option available, as in the following: (261) bāb-ē wān mir father-IZM 3 PL:OBL die:PST ‘Their father died’ (lit. father-of them died), not *wan bāb mir (MacKenzie 1961a: 175) An EPC such as (258) is therefore only available for certain possessive expressions, but it is currently uncertain exactly what factors determine its availability. 5.6.1.2 Experiencers A very similar construction is found with certain expressions of sensory perception, generally involving a body-part term. The commonest is čav ka(f)tin, lit. ‘eyes fall’, i.e., ‘catch sight of’: 260 Kurdish (Northern Group) (262) waxt-ē min čav dôtmām-ā xô kaft-in time-OBL 1 S:OBL eye:PL cousin-IZF REFL fall:PST-3 PL ‘When I caught sight of my cousin’ (lit. When to-me eyes fell on my cousin) (MacKenzie 1962: 286) Crucially, the fronted Oblique in (262) is demonstrably a syntactic subject because it controls reflexive xô. The verb, however, continues to agree in person and number with the Theme argument, čav.21 Further examples of Experiencer-predicates based on body-part terms are the following: (263) min dil pē sot ¯ 1 S:OBL heart for.it burn:PST:3 S ‘I felt sorry for it’ (lit. to-me heart for-it burned) (MacKenzie 1962: 252) (264) kevok-ē guh lē bû [. . .] dove-OBL ear at.it be:PST(3 S) ‘The dove overheard it [. . .]’ (lit. ‘to-the-dove ear was at-it’) Although constructions with fronted Oblique Experiencers often involve a body-part term, this is not a prerequisite. The following is a typical example of an Oblique Experiencer without a body-part term: (265) 5.6.1.3 te sar =e 2 S:OBL cold =COP:3 S ‘Are you cold?’ (field-notes, Zakho, September 2006) Needers and Wanters Badı̄nānı̄ has a verb vyān, which is intransitive and basically means ‘be necessary, be desirable’. It is regularly used with a fronted Oblique ‘Needer/Wanter’ and a Direct ‘Needed/Wanted’, as in: (266) 21 ama hasp na-vē-n 1 PL:OBL horse:PL NEG-be.necessary:PRES-3 PL ‘We do not want horses.’ (lit. to-us horses are-not-necessary) (MacKenzie 1961a: 192) Plural number is not overtly signalled on bare nouns in the Direct case, cf. Section 5.2.1. Evidence from Badı̄nānı̄ 261 Often the ‘Wanted’ is not a NP but a clause, as in: (267) min t-vē-t az bi-č-im-ava 1 S:OBL IND-be.necessary:PRES-3 S 1 S IRR-go:PRES-1 S-ITERAT ‘I want to go back’ (lit. to-me is desirable I go back) (MacKenzie 1961a: 192) Note that it is the needed entity which governs agreement on the verb: (268) ta az na-vē-m 2 S:OBL 1 S NEG-be.necessary:PRES-1 S ‘You do not want me’ (MacKenzie 1961a: 192) Fronted Needers/Wanters control the reference of reflexive xô, as shown from the following example (elicited during my own field work): (269) min t-vē-t hesp-ē xô 1 S:OBL PROG-be.necessary:PRES-3 S horse-IZM REFL ‘I want/need my own horse’ (and noone else’s) As we would expect, the Fronted Oblique can also control coreferential deletion, as in the following (transcription follows the source): (270) 0/ i bi-ç-im mini d-vê-t mal-ê 1 S:OBL PROG-be.necessary:PRES-3 S 0/ i IRR-go:PRES-1 S house-OBL ‘I want/need to go home’ (Şirîn 1996a: 18) Again we have here an identical syntactic configuration to the constructions discussed above: an intransitive predicate, of which a Direct NP is the morphological subject, to which a fronted Oblique is grafted. The fronted Oblique, however, controls syntactic subject properties. 5.6.1.4 Summary of non-canonical subjects We have seen that Badı̄nānı̄ makes use of fronted Obliques in three types of construction: predicative expressions of possession, with certain expressions of sensory perception, and with the verb vyān in expressions of necessity and desire. The semantics embodied by these three constructions are absolutely typical for the semantics of similar constructions cross-linguistically. Shibatani (2002) lists the following expression types found cross-linguistically with non-canonical subjects: 262 Kurdish (Northern Group) 1. Possession/Existence 2. Psychological states 3. Visual/auditory perceptions, including the notion of ‘appearance, seeming’ 4. Necessity and wanting including the notion of obligation (‘must’) 5. Potentiality, including the ability and the notion of permission (‘may’) 6. Some other uncontrolled states of affairs (e.g., ‘finding something,’ ‘remembering,’ ‘forgetting’) With the exception of Potentiality, this list is a reasonably accurate account of the various types of non-canonical subject in Badı̄nānı̄ treated above.22 In Badı̄n., the fronted Oblique with expressions of sensory perception and expressions of need and desire is demonstrably a syntactic subject (control of reflexive xwe). Semantically, the Experiencer and the Needer/Wanter are also obligatory in the sense that an expression of perception implies a perceiver, and one of desire presupposes a desirer. Thus the fronted Obliques in these constructions appear both on semantic and syntactic grounds to be good candidates for subjecthood. With possession, matters are less clear. One could consider the possessor a merely optional extension to the existential construction. The syntactic status of the fronted Oblique in the three constructions is therefore probably not identical, so it is probably expedient to accept that there are different grades of subjecthood. Such an account would be consonant with the construction-grammar approach advocated in Croft (2001), where syntactic relations are construction-specific, rather than being valid across an entire language. Furthermore, graded optionality of non-canonical subjects is a cross-linguistically widespread phenomenon – cf. the contributions in Aikhenvald et al. (2001). In his detailed discussion of such constructions in Japanese, Shibatani (2001: 338) concludes that “many of the predicates that are said to call for non-canonical coding patterns do in fact function as intransitive predicates.” The debate on the status of such ‘non-canonical subjects’ in numerous well-studied languages testifies to the difficulties inherent in an ‘all-or-nothing’ approach to subjecthood (cf. for example the 22 In Badı̄nānı̄, expressions of ability are based on the verb şîan, which does not require a non-canonical subject (Şirîn 1996a: 51). Evidence from Badı̄nānı̄ 263 on-going debate on Japanese, Shibatani (2001) and (Kishimoto 2004)). As Comrie (1989: 110) puts it, in many instances it is simply “pointless to expect a clear cut answer to the question ‘What is the subject of this sentence?’ ”. At any rate, nothing hinges on our decision at this stage. The important point in the present context is the existence of a particular type of clausal constituent which we will group together under the label of non-canonical subjects. In the next section, we will relate the non-canonical subjects of Badı̄n. to the ergative construction.23 5.6.2 Ergativity in Badı̄nānı̄ At first glance, Badı̄nānı̄ appears to have straightforward case of ergativity in the past tenses, along the lines of the system outlined in Section 5.3: The A-past in the Oblique, verbal agreement with the O-past. However, closer examination of the data shows that there are some subtle differences. The most striking feature is the existence of past tense verb forms, formally identical to those found in the ergative construction, but used in contexts in which the precise identity of the A is not recoverable. The following examples illustrate this phenomenon: (271) az darmān kir-im 1 S medication do:PST-1 S ‘I was treated’ (lit. ‘I was medication-done’) (MacKenzie 1962: 286) (272) waxt́-ē r̄ûništ́ı̄-n-a xārē, zad ı̄nā time-OBL sit:PST-PL- DIR down, ¯food bring:PST:3 S ‘when they sat down, food was brought’ (MacKenzie 1962: 322) 23 Lazard (1984: 243) briefly comments on the striking similarities between the possessive and the ergative constructions, basing his comments on what appears to be Badı̄nānı̄ data. As far as I am aware, Lazard’s comments have not been taken up in the relevant typological literature. 264 Kurdish (Northern Group) (273) pišt́ı̄ min maḱt́ab xalās kir-ı̄, t́amām bı̄, az after 1 SOBL school finished do:PST-PTCPL, complete be:PST, 1 S t́a‘ı̄n kir-im mu‘allim appointment do:PST-1 S teacher ‘After I had finished school, it was over, I was appointed a teacher’ (lit. ‘I was appointment-done a teacher’) (MacKenzie 1962: 364) (274) aw (275) l bāpı̄r-ē=ngô sar-ā bāpı̄r-ē ma ancestor-IZP 1 S.OBL head-IZF ancestor-IZP=2 PL from bahašt-ē hāvēt-a darē ¯ paradise-OBL throw:PST-DIREC outside ‘Our ancestors were thrown out of paradise on account of your ancestors’ (MacKenzie 1962: 248) har sē diz bir-in-a lālı̄ DEM all three thief take:PST-PL -DIREC before ‘those three thieves were taken before (him)’ (MacKenzie 1962: 260) For MacKenzie (1961a: 193), these are all examples of agentless passives, and in all cases, the most natural translation is with an English agentless passive. On a visit to Zakho in 2006 I heard a striking example of this phenomenon. During a discussion on the fate of a kidnapped foreign journalist, one of the interlocutors asked another what had happened to the journalist. The answer was a single word: kuşt ‘kill:PST’. The intended meaning was: ‘She was killed’. The past tense verb form kuşt thus expressed what is essentially an agentless passive. Note that the topic of the preceding discourse was the unfortunate journalist, not her kidnappers, so it would not be appropriate to interpret the utterance as ‘(they) killed (her)’. It is important to note that the approximate identity of the Agent can usually be inferred from contextual and encyclopedic knowledge. For example, in (271) it will be the doctor that the speaker has visited. In (272) one can assume that the food is brought by servants attending the guests, although these are never explicitly mentioned in the text. But of course precisely the same can be said of most instances of agentless passives in English: an Agent is implied, and in many cases the approximate identity can be inferred. But this type of contextual construal needs to be distinguished from zero anaphora, the discourse-driven omission of previously introduced referents (although drawing the distinction between zero-anaphora and construal based on contextual Evidence from Badı̄nānı̄ 265 and encyclopedic knowledge is not always straightforward, as we shall see below). In Badı̄nānı̄, this type of usage is fairly unusual, but under certain conditions yet to be fully understood, it is obviously a possible interpretation of past tense transitive verbs. In the other dialects of the Northern Group, however, constructions such as (272) are completely absent; a past verb form of an ostensibly transitive verb will always be interpreted as active and transitive. An A-past can be omitted, but only through the normal processes of zero-anaphora. Consider the following sequence from the Erzurum dialect of Turkey: (276) a. dibê: ‘‘brayê mi jî qêmišî mi and.says: ‘‘brother-of me for.his.part pitilessness-of me nebû, not-was, b. ez ne kušt-im, I not kill:PST-1 S c. li ser vē kanîyê hîšt-ime” at this well leave:PST-1 S (a) (she) says: my brother, for his part, took pity on me (b) I not killed (c) at this well left:1 S (Haig forthc. a, glosses simplified) Although it would appear that a passive interpretation of (b) and (c) is possible (‘I was not killed, I was left here’), intensive discussion with a native speaker (the grandson of the narrator) has left me in no doubt that (b) and (c) are to be interpreted as active clauses with a discourse-deleted A, i.e., ‘My brother took pity on me, he did not kill me, he left me at this well.’ Discussion of similar examples has, up until now, always lead to the same result. Returning now to Badı̄n., the existence of past-tense verb forms expressing agentless passives raises a familiar and messy issue: that of the transitivity of the relevant verb forms. It turns out that exactly the same verb forms are used in what is ostensibly the canonical ergative construction. The following two examples illustrate the contrast. First, consider (273), repeated here for convenience: 266 Kurdish (Northern Group) (273) PAST FORM OF kirin, THE A IS NOT RECOVERABLE : pišt́ı̄ min maḱt́ab xalās kir-ı̄, t́amām bı̄, az after 1 SOBL school finished do:PST-PTCPL, complete be:PST, 1 S t́a‘ı̄n kir-im mu‘allim appointment do:PST-1 S teacher ‘After I had finished school, it was over, I was appointed a teacher’ (lit. ‘I was appointment-done a teacher’) In the same text, we find exactly the same verb form, but this time with an overt Oblique A, hukmat-ē ‘the government’: ¯ (277) PAST FORM OF kirin, OVERT A- PAST: az maǰbôr bı̄-m, hukmat-ē az ta‘ı̄n kir-im 1 S obliged be:PST-1 S, ¯government-OBL 1 S appointment do:PST-1 S . . . I was obliged (to go), the government appointed me . . . ’ (MacKenzie 1962: 364) It will be seen that past tense verb forms such as kirim are compatible both with an active, and a passive reading. The distinction is not drawn at the level of morphology, but at a constructional level: the presence of an overt, or discourse recoverable Oblique will lead to an active interpretation. Its absence leaves a passive interpretation possible. If an overt A is part of the construction, then it will have syntactic subject status and control reflexive xwe/xô, as in: (278) bālı̄l-ı̄ ḱēt́ik-ak kir-a ta barı̄ḱ-a xô-dā Bahlul-OBL cat-INDEF do:PST-DIREC PREP pocket-IZ REFL-LOC ‘Bahluli put a cat in hisi pocket.’ (MacKenzie 1962: 322) Past tense verb forms, then, are capable of both an agentless passive interpretation (273), and or a transitive verb reading, as in the ergative constructions (278) and (277). These findings are highly reminiscent of the Middle Iranian verb forms discussed in Section 3.7.1 under the heading of lingering intransitivity: an (originally) participial verb form, based on a transitive verb, can appear both with or without an overt (or textually recoverable) A-past. Again, Badı̄n. appears to have retained remarkably archaic syntactic features, largely lost in the rest of the Northern Group. Evidence from Badı̄nānı̄ 267 The nature of lingering intransitivity hinges on a distinction between two interpretations of a particular verb form. On one interpretation, the verb is construed as active, and transitive. The precise identity of the Agent is fully recoverable from the immediate linguistic context: it is either overtly present in the clause (an Oblique), yielding a fully-fledged ergative construction, or it may have been deleted through the normal processes of co-referential deletion. On the second interpretation, the precise identity of the Agent is not recoverable from the context, and we obtain an agentless-passive interpretation. While the distinction is, in principle, clear enough, there are actually quite a number of contexts where the decision depends on a subjective interpretation of the extended context. The following example is instructive (glosses simplified): (279) a. rožakē r̄ā bôn, gôt́ı̄: one.day got.up, said: l čyāy.” b. “am dē čı̄na pazā ¯ in mountain” “we FUTPRTCL will.go to.sheep c. r̄ā bôn, čôna čyāy, pazā, ¯ got.up, went to.mountain, after.sheep, d. sē paz kuštin ¯ killed (MacKenzie 1962: 320) three sheep The clause under consideration is (279d). In MacKenzie (1961a: 193), this clause is cited – with no accompanying context – and interpreted as a passive. MacKenzie’s translation for the clause in isolation is as follows: (280) Passive translation of (279d): ‘three sheep were killed’ But when one considers the foregoing context to this example, it is evident that an active translation along the following lines would be contextually more appropriate: (281) Active translation of (279d): ‘(they) killed three sheep’ And in the translations provided with the actual text, MacKenzie opts for the latter translation (in fact this passage occurs on two occasions in the text collection, and in both MacKenzie provides an active translation). In other words, when taken in isolation, the clause is interpreted as an agentless passive, exactly comparable to (272) etc. above. But in context, it is interpreted as 268 Kurdish (Northern Group) an active transitive verb, where the A has been deleted through coreferential deletion. MacKenzie does not comment on the disparity between the two interpretations, yet it neatly illustrates the difficulties of reliably distinguishing passive from active readings in actual usage. 5.6.2.1 Summary of ergativity Superficially, alignment in the Badı̄n. dialect is identical to the rest of the Northern Group. However, there are telling differences. Most important is the attested use of past tense verb forms in Badı̄n. to express agentless passives, a feature already discussed for Middle Iranian under the heading of lingering intransitivity. A further archaic feature of Badı̄n. syntax is the widespread use of non-canonical subject constructions for expressions of possession, desire etc. These show essentially the same syntactic features as the ergative construction: a fronted Oblique with syntactic subject properties, but which may be omitted in some cases. Again, the parallels bespeak of a close relationship between ergativity in Iranian and non-canonical subject constructions. But note that the link between them is not inseparable; throughout most of the Northern Group, the non-canonical subject constructions have largely disappeared, leaving the ergative construction an isolated syntactic oddity. In Badı̄n., on the other hand, the ergative construction is nested in with a family of other NCS-constructions, sharing the common feature of the fronted Oblique subject. It may be the greater critical mass of such NCSconstructions in Badı̄n. that has preserved the largely canonical nature of its ergative constructions: there are far fewer deviations from canonical ergativity in Badı̄n. than in any of the other attested dialects. Not a single example of a double-Oblique was found, and cases of A-dominated plural agreement are significantly rarer in Badı̄n. than in the dialects further north. Indeed, the general principle of non-canonical subjecthood is so firmly rooted in Badı̄n. that it has, arguably, even been extended to an additional construction, leading to what I provisionally term secondary ergativity in Badı̄n., discussed in the next section. Evidence from Badı̄nānı̄ 269 5.6.3 Secondary ergativity This discussion of alignment in Badı̄n. would not be complete without mention of a construction which appears to be a genuine Badı̄n. innovation and which illustrates with remarkable clarity the way that ergativity is linked to non-canonical subjecthood. In fact, I will suggest that the specific Badı̄n. developments here to some extent replicate what was claimed for Old Iranian. Let me first introduce the basic construction. The central element is, etymologically, an Izafe partikel (see Section 5.2.2), but in this construction, it functions not as linker between an attribute and a nominal head, but as part of a predicate. MacKenzie (1961a: 205) in fact treats it as a distinct tense/aspect value, expressing “the sense of a state or action in progress”. Some examples follow:24 (282) az-ē musāfir-im 1 S-IZM traveller-COP:1 S ‘I am (temporarily) a traveller, i. .e at this moment’ (lit. I, who am a traveller) (MacKenzie 1961a: 206) (283) Xatûn-a min ya l hîvî-ya te lady-IZF 1 S:OBL IZF at expectation-IZF 2 S:OBL ‘My lady is waiting for you (at this moment)’ (lit. my lady – who (is) at your expectation) (Blau 1975: 106) (284) du sêv wêt livirê he-in two apple:PL IZP here existent-COP:3 PL ‘There are two apples here’ (own field-work) There is little doubt that the source of the particle glossed as IZ is the Izafe particle: It is phonologically identical to the Izafe, and encompasses the same paradigm of forms, distinguishing gender and number. As already discussed in Section 5.2.2, the Izafe can occur separated from its head noun, as in ex24 A related construction also crops up in the Yezidi dialects of Tur ’Abdîn in southeastern Turkey, described in Bailey (2005). These dialects are geographically close to the Badı̄n. area, and share with the Yezidi dialects of that area close cultural ties. Interestingly, the construction described by Bailey has further grammaticalized, with the former Izafe linker now having lost gender differentiation. 270 Kurdish (Northern Group) pressions such as ya min ‘that of-me’, ‘my one’. MacKenzie (1961a: 162) discusses this type of Izafe under the heading ‘Demonstrative Izafe’, noting that one of its frequent functions is to introduce a relative clause. This is fairly obviously the source of the Izafe particle in the constructions under investigation here, as MacKenzie himself concludes. The question is: how did a nominal linker become a clausal operator, expressing a particular tense/aspect value of the predicate? I assume that this happened via the reanalysis of a fronted Topic construction.25 In the case of (283), we would have had something like my lady who/she is waiting for you, with a fronted Topic followed by a Demonstrative Izafe functioning as a resumptive pronoun. From this emerged the reading as a pragmatically neutral assertion: My lady is waiting for you. In the latter construction, the original fronted Topic (‘my lady’) has now been reanalyzed as the subject of the clause. The Izafe has been retained, but it functions as a category within the verbal domain (with a particular tense/aspect value), rather than a pronominal or demonstrative element. Cross-linguistically, this kind of development is well-attested. Trask (1996: 135), citing Li and Thompson (1977), discusses the development of copulas in Mandarin and Hebrew from erstwhile “demonstratives or pronouns used in a linking function”. A more fitting characteristic of the Izafe-particle would be hard to find. Quite apart from the cross-language evidence in support of the development, there is a striking language-internal fact that suggests an origin through some sort of grammaticalized topic-construction is very probable: the Badı̄n. construction is only possible with positive, declarative clauses. With negated clauses and questions, it is not used. Given that the latter have very different pragmatic properties and entailments, then the current restriction of the construction to affirmatives is perhaps understandable. Crucially, the above construction occurs primarily with a copula in a stative, existential or locational sense. But it is also quite common for the predicate of such expressions to be a Past Participle. The Past Participle under consideration here is a Kurdish innovation, formed by adding −ı̄/y to the past 25 In an unpublished manuscript, Unger (1994) also relates the construction to topicality of the subject, but does not go into the diachronic implications. A more detailed discussion of the theory outlined here is Haig (2007). Evidence from Badı̄nānı̄ 271 stem of the verb (MacKenzie 1961a: 187). Such participles have essentially the same properties as past participles in English (broken, made, crushed etc.) and are frequently used as adjectival modifiers. But in the construction at hand, they may occur in predicative function, as in the following: (285) . . . ‘či bı̄a?’ Gôt-ē: ‘sindôk-ā škāndı̄’ what happened? say:PST-to.him chest-IZF break:PTCPL ‘ . . . ‘what happened?’ (He) said to him: ‘The chest has been broken (into).’ ’ (MacKenzie 1962: 258) On the surface, the final phrase in (285) looks like a NP (‘the broken chest’), but in fact the participle škāndı̄ is predicative (it is the answer to the question ‘what happened?’). The phrase is, despite the Izafe particle, therefore not a NP but a clause. It has illocutionary force and expresses an assertion, just as the stative and existential clauses such as (284) above. The construction common to all the examples of this type looked at so far can be summed up as follows: NP(Theme/Patient) Izafe Stative/Locative/Resultative Predication But Badı̄n. can take these constructions a step further: when the predicate is the past participle of a transitive verb, then a fronted Oblique can be added to the construction, expressing the A. Consider the following examples, with the basic construction bracketed: (286) min [šı̄v=ā lēnāy] 1 S:OBL supper=IZF prepare:PTCPL ‘I have prepared supper’ (MacKenzie 1961a: 211) (287) . . . Xodê [rizq-ê min yê înaye mala te] ¯ God:OBL fate-IZM 1 S:OBL IZM place:PTCPL house-IZF 2 S:OBL ‘God has placed my fate in your house’ (Blau 1975: 102) (288) min [qebhet-ek ya kirî] ... ¯ 1 S:OBL bet-INDEF IZF do:PTCPL ‘I have placed (done) a bet . . . ’ (Blau 1975: 104) (289) ma [čēčik vēkr̄ā yēt kirı̄-n] 1 PL:OBL whelp together IZP do:PTCPL-PL ‘We have whelped together’ (MacKenzie 1962: 254) 272 Kurdish (Northern Group) (290) min [galak xēr yā l vē çēlē dı̄tı̄] 1 S:OBL much good IZF from DEM:OBL cow:OBL see:PTCPL ‘I have seen(=experienced) much good from this cow’ (MacKenzie 1962: 250) (291) min [kič-ā xô yā dā-ē] 1 S:OBL daughter-IZF REFL IZF give:PTCPL-to.him ‘I have given him my daughter’26 (MacKenzie 1961a: 163) The fronted Obliques also appear to be syntactic subjects, as evident from the use of reflexive xô in (291). And unsurprisingly, if the predicate is not a past participle, but an existential, a fronted Oblique can be added, which yields a Possessor reading. Compare (292) with (284) above: (292) te [du sêv wêt he-in] 2 S:OBL two apple:PL IZP existent-COP:3 PL ‘You have two apples’ (own field-work) The construction outlined in this section demonstrates quite neatly how a clause headed by a basically intransitive predicate (e.g., an existential, or a past participle) can be combined with a fronted Oblique to yield a twoargument construction. Such non-canonical subjects can code both possession, or agency. This time the result has been a new tense-aspect form in Badı̄n.27 But what is central in the present context is the recurrent theme of intransitive predicates melding with a NCS-construction to yield two-participant expressions of possession, and of agency. And the link from one to the 26 27 The clitic Indirect Object pronoun on the verb −ē has presumably obscured the expected participial suffix −ı̄. The development of this new tense form in Badı̄n. parallels the development of the periphrastic perfect based on a ‘have’-verb in English. The development for English is, very roughly, considered to have proceeded from I have a book (that is) read, where the participle read is an adnominal attribute to book. The erstwhile attributive participle was then reanalyzed as a verb form and linked to the auxiliary have: I have read a book. In Badı̄n. too, we find what was evidently once an attributively-used participle (linked to its head noun via the Izafe) now reinterpreted as a finite verb form. The function of English have – essentially a means of rendering the construction transitive, that is, accomodating a full subject and an object – is paralleled in Kurdish by the non-canonical subject construction. Summary of the Northern Group 273 other runs across participles, just as it did in Old Iranian. This time, however, we are dealing with a new layer of secondarily-created participles, but precisely the same bundle of factors continues to be operative as were claimed to be behind the developments in Old Iranian. Such a development is probably only possible in a language which already has a productive and frequent non-canonical subject construction, something that I have claimed for Old Iranian. While the developments in Badı̄n. are certainly no proof that ergativity in Iranian emerged in the manner I have depicted in Chapter three, they do provide striking support for theories that assume that the A originated not from a by-phrase of a passive construction (which, incidentally, are practically entirely absent in spoken Badı̄n., and in Northern Kurdish generally, see Section 5.3.1.4), but through the extension of a pre-existing non-canonical subject construction. 5.7 Summary of the Northern Group The Northern Group of Kurdish provides a particularly rich source of data for the investigation of alignment. There is ample material available for a wide range of dialects, and the syntax is uncluttered by the pronominal clitics which render the description of alignment in most Iranian languages so complex (see the next chapter). A further fortunate feature of the Northern Group is the existence of a robust diagnostic for syntactic subjecthood: control of reflexive xwe (cf. Section 5.3.1.3). It is thus possible to assess syntactic subjecthood with considerably more confidence than was possible for the older stages of Iranian, and indeed for many modern Iranian languages. While the Northern Group is regularly portrayed as a language with pure ergative alignment in past tenses, such a characterization turns out to be grossly oversimplified. All dialects exhibit deviations from canonical ergativity, with two main tendencies emerging: first, the tendency towards Oblique marking of the O-past, leading in many contexts to a double-Oblique construction with both A and O Oblique-marked. The theoretical question raised by the recurrent appearance of such a pattern is that double-Oblique case marking runs counter to our expectations about what case systems should be achieving, namely discriminating A from O. I have argued extensively that the discriminatory function of case marking, while an important factor 274 Kurdish (Northern Group) in shaping long-term trajectories of change, is in fact overridden by a desire towards cross-system harmony. Putting the O into the Oblique case is the single change in the morphology which achieves a maximum of case-marking parallelism between the past and the present transitive constructions in actual discourse. The second striking deviation from canonical ergativity is the tendency for agreement to realign with a plural A-past, rather than the expected agreement with the O-past (A-dominated agreement). I pointed to semantic and discourse-pragmatic factors which impact on agreement and which, under certain conditions, override the demands of clause-internal grammar. I concluded that verb agreement in Past Transitive Constructions is not simply mechanical-feature copying, with the reverse alignment to that of present tenses. Instead, it works quite differently and is in fact underspecified for certain values. It is thus more sensitive to aspects of the broader utterance context than the purely grammatical agreement that holds in present tenses. If a Past Transitive Construction regularly features double-Oblique case marking, and A-dominated verb agreement, then it has already moved a considerable way towards becoming an accusative construction. It merely requires the loss of the Oblique marking on the A, and the full realignment of verb agreement to the A. In fact, very few varieties of the Northern Group have taken these steps; most retain fairly consistent Oblique marking of the A, and indeed this seems to be the most resilient feature of the ergative construction throughout Iranian (recall the First Principle of Case Functions in Iranian, formulated in (133) in Chapter four). One dialect, however, Gilan Kurdish, has evidently gone right down the track to full accusative alignment in its past tenses, showing once again how even closely-related varieties of a single language can diverge drastically in the alignment of their Past Transitive Constructions. In the final sections of this chapter we turned to the Badı̄nānı̄ dialect. Here we find the ergative construction coexisting with a robust and textually frequent non-canonical subject construction, used for expressions of possession, sensory perception, and desire/obligation. I argued that these are archaic constructions which have survived in Badı̄n., but were lost elsewhere in the Northern Group. The key features of this construction is the optionality of the fronted Oblique (the non-canonical subject), and the lexical intransitivity of the predicate (expressions of existence, location etc.). Both these features Summary of the Northern Group 275 arguably characterize the ergative construction of Badı̄n. It transpires that the verb forms of the ergative construction here also display what I have termed lingering intransitivity, hence may be used either with a fronted Oblique to express a transitive clause (the ergative construction), or without one to express merely an agentless passive (‘the government sent me’ vs. ‘I was sent’, cf. (273) and (277) above). Badı̄n. has preserved features reminiscent of Old and Middle Iranian, and thus affords an unrivalled opportunity for demonstrating the proximity of the non-canonical subject construction and the ergative construction. Finally, Badı̄n. has developed what I have termed a secondary ergative construction based on the past participle in −ı̄, where once again the proximity of Non-Canonical subjecthood, possession, and Iranianstyle ergativity is underscored. Of course I cannot claim that developments that can be observed in modern Badı̄n. prove that similar developments must have accompanied the emergence of ergativity in Old Iranian. But they do show that non-canonical subject constructions and ergative constructions share very obvious structural and semantic commonalities, and that they may coexist in one language. Thus linking one to the other in a diachronic scenario is certainly not unreasonable. Furthermore, there is a complete lack of evidence in support of the alternative view of the emergence of ergativity, namely an origin through an agented passive construction. Periphrastic passives are occasionally used in the Northern Group (cf. Section 5.3.1.4), but passives with agent phrases are confined to the emergent written language, and the agent phrases are marked with various complex secondary adpositions bearing no relation to the case markers of the A in the ergative construction. Chapter 6 The Central group 6.1 Introduction It has been claimed that the Central Group of Kurdish is characterized by a “loss of the ergative construction”, and (at least some of the languages of this group) are now essentially nominative/accusative (Bynon 1979: 223–224, Bynon 1980: 159). However, such a view represents an oversimplification in several respects, and overstates the differences between the Northern and Central Groups. The Central Group also displays the familiar phenomenon of Tense Sensitive alignment: Alignment in present tenses is straightforward nominative/accusative, while the alignment of transitive clauses in the past tense diverges significantly from that of the present tenses. To what extent the past tense alignment in the Central Group can be considered ergative is an issue to be discussed at the end of this chapter. According to MacKenzie (1961a), the Central Group comprises the following dialects (adopting MacKenzie’s conventions for transcription, and given in approximate South-to-North order): Suleimani, Wārmāwa, Bingird, Piždar, Mukrı̄, Arbil, Rewandiz and Xōšāw. My description is based on the comparatively well-described Suleimani dialect, but is supplemented with data from neighbouring dialects. 6.2 Suleimani morphosyntax The Central Group of Kurdish shares several broad features with the Northern Group outlined in Chapter five. The most important commonalities are: 1. Izafe-construction All Kurdish dialects share the basic features of the Izafe-construction. 2. Dual-stem system in the verbs Both groups have two distinct stems for each verb. 3. Constituent order Both groups have a predominantly predicate-final constituent order in the clause, although it is quite flexible. There is regular use of post-predicate slot for Goal and Indirect Objects. 278 The Central group 4. Subordination and embedding Like the Northern Group, the Central Group lacks non-finite verbal syntax; clause linkage involves sequences of finite clauses, often with no formal indication of subordination save a non-indicative mood on the ‘subordinate’ clause. 5. Case and gender All Kurdish dialects have maximally two gender and two case distinctions. However, several members of the Central Group have lost both case and gender (see below). The major structural features distinguishing the Northern Group from the Central (and Southern) Group are the existence in the latter of a definiteness suffix (in Suleimani -aka), and a set of enclitic personal pronouns. Both features are discussed in the next section. Turning now to Suleimani, the best-described representative of the Central Group, we should note that there is no grammatical case, and no grammatical gender. Nouns appear in a uniform case (ignoring the Vocative), formally equivalent to the Direct Case of the Northern Group. There is also only one set of personal pronouns in Suleimani (see Table 1). The form of the first person singular pronoun is etymologically that of the Old oblique ((a)min), rather than of the Old Iranian Nominative. Thus while the Northern Group still has ez (1 S:DIR) vs. min (1 S:OBL), Suleimani has entirely lost any reflexes of the Direct form. However, other closely related dialects of the Central Group have a Direct/Oblique case distinction in both nouns and pronouns, and it is consequently noted in some examples. In terms of case marking, then, Suleimani is neither Accusative nor Ergative. Rather, we have an example of the typologically unusual ‘neutral’ system, with S-past, A-past and O-past in a single case form. The following sample sentences were supplied by Sorani native speakers.1 It will be noted that transitive past clauses require clitic cross-referencing of the A, which will be discussed in more detail below: 1 The speakers who kindly supplied the information are a married couple, in their late 30’s, who were born in Suleimaniye and spent most of their lives there. They have been living in Germany for the last six years. Suleimani morphosyntax 279 (293) min hāt-im bō erā 1 S come:PST-1 S to here ‘I came here.’ (294) ewā hāt-in bō erā 2 PL come:PST-2 PL to here ‘You(PL) came here.’ (295) min ewā=m bı̄nı̄ 1 S 2 PL=1 S:CLC see:PST ‘I saw you(PL).’ (296) ewā min=tān bı̄nı̄ 2 PL 1 S=2 PL:CLC see:PST ‘You(PL) saw me.’ The lack of grammatical gender means that there is a single form of Izafeparticle in Suleimani, −ı̄/-y: (297) kič-ēk-ı̄ zōr ǰwān girl-INDEF-IZ very beautiful ‘a very beautiful girl’ (MacKenzie 1962: 14) (298) dēw-ēk-ı̄ gawra demon-INDEF-IZ great ‘a great demon’ (MacKenzie 1962: 16) An important feature throughout the Central and Southern Groups is the existence of a definiteness suffix, in Suleimani −aka. Suleimani also has a plural suffix, −ān (etymologically identical to the plural oblique suffix of the Northern Group, −a(n)). The plural suffix is used mainly in combination with the definiteness suffix, yielding a composite ending −akān (<−aka+−ān). 6.2.1 Pronouns and verbal agreement The full forms of the personal pronouns in Suleimani are given in Table 1 (cf. MacKenzie (1961a: 73), the third singular forms are identical to the distal demonstratives). 280 The Central group Table 1. Personal Pronouns in Suleimani SG. 1 2 3 min tō aw/am PL. 1 2 3 (h)ēma ēwa awān Table 2. Verbal agreement suffixes in Suleimani S ET 1 S ET 2 SG. 1 2 3 im -y / -ı̄(t);-a/-0/ (Imperative) ē(t)/-ā -im -ı̄(t) -0/ PL. 1 2 3 -ı̄n -in -in -ı̄n -in -in As mentioned above, these forms are possible in all syntactic functions (e.g., S, O-pres, O-past, A-pres and A-past functions, see below). However, as the object of prepositions, a clitic form of the pronoun is more usual (see below). In the present tense, and for all intransitive verbs, verbs agree with the appropriate S and A via an agreement suffix on the verb stem. There are two paradigms of verbal agreement suffixes: Set 1 is used with verbs built on the present stem, while Set 2 affixes to past stem forms (the exact function of Set 2 agreement suffixes is complex and controversial, see below). The forms of the suffixes are given in Table 2 (cf. MacKenzie 1961a: 89,95). The past tense forms of transitive verbs may also agree with an O-past, in which case the appropriate Set 2 suffix from Table 2 is used. However, as the rules determining agreement in past tense transitive constructions differ significantly from those for the present tenses they will be treated separately in connection with the clitics in the next section. In addition to the past forms mentioned in Table 2, a Perfect Indicative is also found, based on a secondary Suleimani morphosyntax 281 participial stem ending in −uw, to which the person endings of the enclitic copula are added. In terms of the agreement constellation found, such compound tenses behave largely in the same manner as the simple past tense, i.e., verbal agreement is with S or an O-past (subject to the conditions discussed below). Examples illustrating the person/number suffixes in various functions are provided below. (299) and (300) show agreement with an Spres, (301) demonstrates agreement with an A-pres, while (302) and (303) illustrate agreement with an S-past: (299) min a-č-im 1 S PROG-go:PRES-1 S ‘I am going’ (MacKenzie 1962: 4) (300) ēme . . . a-č-ı̄n 1 PL PROG -go:PRES-1 PL ‘we (will) go’ (MacKenzie 1962: 10) (301) tō čı̄ a-ka-y lēra? 2 S what PROG-do:PRES-2 S here ‘what are you doing here?’ (MacKenzie 1962: 4) (302) wałłāhı̄, xēr hat-uw-ı̄n by.God in.peace come.PST-PTCPL-1 PL ‘By God, (we) have come in peace’ (MacKenzie 1962: 10) (303) tō la kø bu-y? 2 S where be:PST-2 S ‘Where have you been?’ (MacKenzie 1962: 18) In terms of verbal agreement, it should be evident that in the present tense, and for all intransitives, Suleimani alignment is essentially identical to that of the Northern Group, and indeed the morphemes concerned are clearly cognate with the appropriate verbal agreement markers in the Northern Group (cf. Table 5 in Chapter five). As far as the case parameter is concerned, Suleimani differs from the Northern Group through the absence of morphological case marking. However, as mentioned above, other members of the Central Group have preserved case distinctions in at least parts of the nominal lexicon, so that in general, it can be stated that alignment in the present tenses and with intransitives is identical across the Northern and Central Groups. 282 The Central group 6.2.2 Pronominal clitics Simple examples of argument cross-referencing such as those illustrated in (299)–(303) are in fact quite hard to come by. Most clauses contain, in addition to a verbal agreement marker, at least one exponent of a set of pronominal clitics. In this section, the morphosyntax of pronominal clitics will be outlined, before going on to investigate the diachronic implications. Both MacKenzie (1962) and Edmonds (1955) provide fairly detailed descriptions, which should be additionally consulted, but my mode of presentation differs from theirs in a number of respects. Table 3 illustrates the forms of the pronominal clitics. Table 3. Pronominal clitics in the Central Group SG. 1 2 3 =(i)m =(i)t; dialectal also =u(w) =ı̄, =y PL. 1 2 3 =mān =tān =yān The forms are best considered clitics rather than suffixes because (i) they co-occur with hosts of different categories (nouns, prepositions, verbs); and (ii) may stack up after inflectional suffixes and other clitics in a manner not possible for suffixes in Kurdish (but see Section 6.3.1).2 Nevertheless, they also show some properties not at all expected of clitics and in some environments, it is probably reasonable to consider them as suffixes. However, the issue of clitic vs. suffix will not directly impinge on the argumentation here, but will be taken up again in Section 6.3.1.3 A question that will be pursued below concerns the functional status of these elements, i.e., whether they are pronouns, or agreement (see Corbett (2003) for discussion). For 2 3 For example, the clitic pronouns usually follow the clitic =(ı̄)š ‘and, also’, evidently cognate with jî (same meaning) of the Northern Group. See MacKenzie (1961a: 128) for examples. See Everett (1996), who argues that the distinction is entirely superficial. Suleimani morphosyntax 283 the time being, we will assume that, despite some phonological similarities with the verbal agreement suffixes, the clitics constitute a distinct paradigm (although towards the end of this chapter I will have cause to qualify this view). Hence I will maintain a consistent terminological distinction between pronominal clitics (or clitic pronouns), abbreviated CLC in the glosses, and (verbal) agreement suffixes. Pronominal clitics are used to express a nominal constituent in a number of distinct syntactic functions, which are listed below. It will be recalled that these five functions are identical to those listed for clitic pronouns of Western Middle Iranian in Section 3.5. In general, it is a fair approximation of the functions of clitic pronouns wherever they are found in West Iranian; the only regular difference across individual languages is whether or not they use the clitics in the A-past function: 1. The Possessor of a Possessed NP 2. The object of a preposition 3. An Indirect Participant (see Section 2.5.1) 4. An O-pres 5. An A-past Of the functions listed, the first four are facultative in the sense that the NP concerned can be expressed with a full NP, in which case the clitic is not used. In other words, in these functions, speakers have a choice, presumably determined by stylistic and pragmatic factors, of whether they use the clitic form or a full pronoun. In the function of A-past, however, the clitic is obligatory: every single past transitive clause must contain a clitic referring to the A-past, regardless of whether the A-past is additionally present as a full NP or not. The first four of these functions are illustrated below, while a discussion of the final function is reserved for the following sections. Adnominal Possessor (304) ay (305) kirās-aka=t shirt-DEF=2 S:CLC ‘your shirt’ (MacKenzie 1961a: 80) šāh=im EXCL king=1 S:CLC ‘O my King’ (MacKenzie 1961a: 81) 284 The Central group (306) xušk-ı̄ dāk-ē=m sister-IZ mother-OBL=1 S:CLC ‘the sister of my mother’ (MacKenzie 1961a: 80) (307) aGa-ka=tān Agha-DEF=2 PL:CLC ‘your(PL) Agha’ (MacKenzie 1961a: 81) Object of a preposition (308) mumbārak bē lē=t for=2 S:CLC fortunate IRR:be:PRES:3 S ‘May (it) be fortunate for you’ (MacKenzie 1961a: 77) Indirect Participant (309) hata mumkı̄n=mān abē up.to possible=1 PL:CLC may.be:3 S ‘As far as (it) may be possible for us’ (MacKenzie 1961a: 78) Direct object in present tenses (O-pres) (310) bi=y-xō-n! IRR =3 S:CLC -eat:PRES -3 PL ‘eat(PL) it!’ (MacKenzie 1961a: 94) (311) da-t-ba-m FUT=2 S:CLC -take: PRES-1 S ‘I will take you’ (MacKenzie 1961a: 77) The following short text illustrates four pronominal clitics in various functions (indicated by the bracketed numbers), and provides some impression of their pervasiveness in actual discourse: (312) na-w(1)-kuž-im, ałqa-yak a-ka-m-a NEG -2 S:CLC -kill: PRES-1 S ring-INDEF PROG -do:PRES-1 S -DIREC gø=t(2) a=t(3)-ka-m ba ‘abd-ı̄ xo=m(4) “ ear=2 S:CLC PROG=2 S:CLC-do:PRES-1 S to slave-IZ REFL=1 S:CLC ‘(I) shall not kill you(1=variant of =t), (I) will put a ring in your(2) ear and make you(3) my(4) slave ’ (MacKenzie 1961a: 76) Suleimani morphosyntax 285 Clitic (1) is an O-pres clitic, attaching to the negation marker of its governing predicate. Clitic (2) is a possessor clitic, attaching to its possessed. Clitic (3) is again an O-pres clitic, this time attaching to the TAM-prefix on the verb, while clitic (4) is another possessor clitic, attaching to the reflexive pronoun. The rules determining clitic placement are discussed below. 6.2.3 Clitic placement The facts of clitic placement in Suleimani are complex; they cannot be accounted for in purely syntactic terms, or in purely prosodic ones.4 There is a clear difference to clitic placement in Old Persian, which can be described with a fair degree of reliability in terms of a Wackernagel-position. Thus the Old Persian clitics frequently affixed to a clause-initial complementizer (e.g., utā ‘and’), as they still did in Middle Persian. But this option is not available in Suleimani. In Suleimani, the general rule for clitic placement is that clitics attach to the leftmost constituent of their phrases. For Possessor and prepositional object clitics, the host is thus the possessed NP or the preposition respectively. Note that these two types of clitics are arguably simple clitics rather than special clitics; they occupy the position that the corresponding non-clitic element would. For the objects of prepositions, however, there are complications. They often detach from the prepositions and move to another element. Contrast the following two examples, apparently semantically identical (own field work): (313) a. min ı̄š a-k-am bo=tān 1 S work PROG-do:PRES-1 S for=2 PL:CLC ‘I work for you(PL)’ b. min ı̄š=tān bo a-k-am 1 S work=2 PL:CLC for PROG-do:PRES-1 S ‘I work for you(PL)’ 4 The present description of clitic placement differs from that proposed in MacKenzie (1961a: cf. esp. 76–81), but I believe accounts more economically for the same data, and captures the generalization that syntactic phrases are the crucial domains in clitic placement. 286 The Central group The clitic pronoun in (313a) is a perfectly straightforward example of a clitic hosted by its governing preposition. In (313b), on the other hand, the clitic appears to have moved leftwards from its preposition, in part a consequence of the change in word order in (313b). There are several ways of analyzing this example. One is to consider (313b) not a verb plus prepositional phrase, but an emergent ‘particle verb’, something like bo kirin ‘do for’, (historically, prepositions are one of the main sources of preverbal particles throughout Indo-European). The clitic in (313b) is then a straightforward example of the Indirect Participant usage, expressing Benefactive or Possessor, and attaching to the first constituent of the VP (see below). However, the complexities involved with these and comparable examples go beyond the scope of the present discussion (see Edmonds (1955: 494–497) for clitics with prepositions). For clitics cross-referencing an O-pres, essentially the same principle applies: the clitic attaches to the leftmost-element of its phrase, the VP. Now of course the O-pres clitic is optional; if the O-pres is overtly realized in the clause as a full NP, then no clitic is possible. In that case, the normal structure is simply (A)OV, with no clitic. But when the O is not realized as a full NP we have (A)V, leaving the predicate itself as the leftmost-element of the VP, and the O-pres clitic duly attaches to the predicate. Within the predicate complex, the following hierarchy of hosts is found: (314) Preverbal-particles etc. > Preverbal TAM/Negation > Verb stem The preferred landing sites within the predicate are preverbal particles, including the nominal components of complex predicates (cf. Haig (2002a) for discussion of such constructions). Otherwise, the clitic attaches to the preverbal TAM, as in (311), or negation markers. Several examples illustrating O-pres clitics within the predicate complex are found in (310)–(312). If none of the preceding slots are available, the clitic follows the verb stem. Although I have not come across any examples of the latter position from the Suleimani data,5 it is certainly possible in other languages with a similar system, for 5 Probably due to the fact that present tense verb forms usually have a TAM prefix which outranks the post-verbal slot as host. Suleimani morphosyntax 287 example Gorani (Residual Group, see Section 5.1), the transcription has been simplified):6 (315) bärä=š, tâ bûär-im get:IMP=3 S:CLC so IRR:eat:PRES-1 PL ‘get it, so (we) may eat (it)’ (Hadank and Mann 1930: 321) For the O-pres clitics, then, clitic placement can be described with reference to syntactic phrases. The relevant domain is the VP, and the clitic attaches to the left-most element of that domain. Turning finally to clitic Indirect Participants, a purely syntactic explanation runs into difficulties. The class of Indirect Participants encompasses a fairly heterogenous mix of Benefactives, Goals, External Possessors and so on, in fact covering a similar range of functions to the Old Persian Genitives discussed in Section 2.5.1. Etymologically the clitic pronouns of the Central Group are reflexes of the clitic Genitive pronouns of Old Persian. And they raise exactly the same issues as their ancestors with regard to syntactic status: are they arguments or adjuncts? If they are arguments, then we would expect their clitic forms to attach to the first available element of the VP, as do the O-pres clitics just discussed. In fact, this is generally the case, with the added proviso that if the verb concerned is intransitive, the S is also a potential host. Examples of clitic Indirect Participants are given below: (316) malā=yān pı̄šān a-dā mullah=3 PL:CLC showing PROG-give:PRES:3 S ‘(He) points the mullah out to them’7 (MacKenzie 1961a: 78) (317) hatā mumkı̄n=mān a-bē until possible=1 PL: CLC PROG-be:PRES:3 S ‘As far as may be possible for us’ (MacKenzie 1961a: 78) 6 7 In spoken Persian (all tenses), direct object clitics can also attach to the verb (na-didam=at NEG-see: PST -1 S=2 S:CLC ‘I didn’t see you’). Parwin Mahmutweyssi has pointed out to me that this clause could, in her dialect, receive a past tense reading, in which case it would mean: ‘they pointed the Mollah out to him’, with the first clitic obligatorily read as an A-past. 288 The Central group (318) tinu=m-a thirst=1 S:CLC-COP:PRES:3 S ‘I am thirsty’ (lit. to-me thirst is (MacKenzie 1961a: 103) (319) zōr=it čāka dagaë da-ka-m much=2 S:CLC good for FUT-do:PRES-1 S ‘(I) will do much good for/with you’ (Mukri, MacKenzie 1961a: 79) 6.3 Past transitive constructions Constructions involving a transitive verb in a past tense have a different syntax to all other kinds of clauses in Suleimani. The first point is that whereas all clitics illustrated so far are optional in the sense that they are only required if the corresponding full form is not overtly expressed in the clause, every single past transitive construction requires an A-past clitic, regardless of whether the A is realized lexically elsewhere in the clause or not. Consider (320), where the lexical A-past, duxtōr ‘doctor’ is present in the clause: (320) ōtōmbı̄l=ı̄ girt duxtōr hālan ¯ doctor immediately car=3 S:CLC take:PST ‘The doctor immediately took a car’ (MacKenzie 1962: 50) Despite the presence of the lexical A-past, a pronominal clitic is still required. The same is true even if the A-past is a personal pronoun, as in (321): (321) min šart=im kird-uwa łagal xwā 1 S bond=1 S:CLC do:PST-PTCPL with God ‘I have made a bond with God’ (MacKenzie 1962: 22) Note that there is no general constraint on A-past clitics attaching to pronouns. If the relevant constituent (see below on clitic placement) happens to be a pronoun, the A-past clitic will attach to it, as in (322): (322) min=it da ba arzā 1 S=2 S:CLC give:PST to ground ‘(if) you throw me to the ground’ (MacKenzie 1962: 22) The conclusion that must be drawn is that the A-past clitic in fact exhibits the features of an agreement marker, i.e., it obligatorily cross-references a Past transitive constructions 289 different constituent, and is prosodically dependent rather than independent. As Corbett (2003) points out, the distinction between agreement markers and pronouns is often a gradual one; the A-past clitics of Suleimani are a case in point. Below I will point out further features of the A-past clitics which bring them closer to a canonical form of agreement. The second major difference between A-past clitics and the others concerns the facts of clitic placement, and the possibilities of combining clitics and agreement suffixes (see Section 6.2.3). As far as clitic placement is concerned, it was suggested above that the general rule is that a clitic is hosted by the leftmost-element of its phrase. Now if the A-past is considered to be a ‘subject’, then presumably the relevant immediate phrase is the clause, thus we should expect the clitic to occur at the beginning of the clause. But, as already seen in (320), this is not the case. In fact, the A-past clitics follow approximately the same pattern as the O-pres clitics: they attach to the first available constituent of the VP. In (320), this is the O-past ōtōmbı̄l.8 In many cases, the first element of the VP is the verb itself. In that case, the hierarchy of landing sites given above in (314), and repeated here, is valid: (314) Preverbal-particles etc. > Preverbal TAM/Negation > Verb stem Crucially, an A-past clitic can never attach to the lexical A-past. Thus in (320) it could not attach to duxtōr. These general principles can be illustrated using the following examples: (323) kič qisa-y na-kird girl talk-3 S:CLC NEG-do:PST:3 S ‘The girl did not talk’ (MacKenzie 1962: 56) (324) sar-im haë bir̄ı̄, tamāšā-m kird head-1 S:CLC up take:PST looking-1 S:CLC do:PST:3 S ‘(I) raised my head and looked . . . ’ (MacKenzie 1962: 56) (325) šart-mān kird lagaë yaktirı̄ condition-1 PL: CLC do:PST:3 S with RECIPR ‘we made a bond with one another’ (MacKenzie 1962: 60) 8 Edmonds (1955: 501, fn. 1) suggests that the O is the overwhelmingly favoured host for the clitic, a fact that may tie in with the use of clitics as ‘case signallers’, discussed in connection with the West Iranian language Gazi in Stilo (2004a). 290 The Central group As far as the placement of the clitic within the predicate is concerned, the following examples illustrate the application of the hierarchy (314) given above (examples from MacKenzie 1961a: 79): (326) a. r̄ā=m kird running=1 S:CLC do:PST:3 S ‘I ran away’ (Clitic on preverbal particle) b. na=m-a-kird NEG =1 S:CLC -PROG -do:PST :3 S ‘I used not to do (so)’ (Clitic on negation prefix) c. a=m-kird PROG -=1 S:CLC -do:PST :3 S ‘I used to do (so)’ (Clitic on TAM prefix) d. kird=im do:PST:3 S=1 S:CLC ‘I did (so)’ (Clitic on verb stem) 6.3.1 Clitic/suffix interactions Perhaps the most remarkable feature of the A-past clitics is their interaction with agreement suffixes. It will be recalled that the verb agrees with a core argument: in present tenses, the S or A, and in past tenses, the S or the O-past (or it lacks agreement in Past Transitive Constructions, see below). The forms for the latter form of verbal agreement were given as Set 2 in Table 2 above. They are regularly used with to cross-reference the S of all intransitive verbs in the past tenses, for example hāt-im ‘I came’. The same set of verbal agreement suffixes can also occur with transitive past verbs to cross-reference the O-past. Consider the following examples, where the Opast is cross-referenced by the appropriate form of the verbal agreement suffix, while the A-past is cross-referenced through a clitic pronoun on a prior constituent: (327) bāN=yān kird-im call=3 PL:CLC do:PST-1 S ‘They called me’ (MacKenzie 1961a: 109) Past transitive constructions 291 (328) na=m-d-sand-in NEG =1 S:CLC -PROG -take:PST -3 PL ‘I was not taking them’ (Fattah 1997: 219) Based on this limited data, one might conclude that the past tenses of transitive verbs regularly agrees with the O-past via the Set 2 suffixes. However, when a transitive past predicate is the first element of a sentence, and when the predicate consists only of a verb stem without any preverbal elements, a conflict arises: as noted above, it is a rule of Suleimani syntax that all past transitive clauses must contain a clitic exponent of the A-past. At the same time, we have seen that the verb agrees with the O-past through a verbal agreement suffix. But in the situation just mentioned, only one slot is available for both markers, namely after the verb stem. On the assumption that the verbal agreement suffix is a suffix, and the A-past marker is a clitic, one could reasonably expect to find that the clitic would attach outside the suffix. Unfortunately, this is not the case. It is most often the supposedly clitic A-past marker which attaches directly to the verb stem, while the O-past agreement suffix follows it: (329) sand=im-in take:PST=1 S: CLC-3 PL ‘I took them’ (Fattah 1997: 220) The dilemma posed by this state of affairs is that, in terms of mobility (i.e., ability to attach to hosts of different categories), the A-past markers are clearly clitic-like, while the O-past marker is suffix-like (restricted to the verb stem). Yet when the two compete for the post-verbal slot, the A-past marker affixes directly to the stem in a very suffix-like manner, while the O-past marker follows it in a very un-suffix like manner. To my knowledge there has been no attempt to formulate a principled account of argument cross-referencing in past transitive constructions. Fattah (1997: 220–221) attempts to explain the positioning of the A-past marker in terms of left-most nodes in a tree-structure representing the verb, its internal arguments and additional functional categories such as Tense, Negation or Aspect. If the direct object is realized in the clause, it will host the A-past marker, and so on. While this approach accounts for a fair number of examples, it is really little more than a restatement of the rules given above, and does not advance our understanding of the construction. Furthermore, it does 292 The Central group Table 4. The order of argument cross-referencing markers when both occur on the verb O- PAST IS : A- PAST IS : SAP NON -SAP SAP NON -SAP(P L .) NON -SAP(S G .) A-O A-O or O-A O-A A-O A-O or O-A O-A not address the issue of why a supposedly clitic element should oust a suffix, as in (329). Finally, it ignores some added twists to the system, which are described by MacKenzie (1961a: 112–113). It turns out that the order of argument cross-referencing markers is not always verb-A-O, as in (329). Instead, it interacts with the person of the arguments. The relevant person categories are SAP (Speech Act Participants) vs. non-SAP, whereby non-SAP is further divided into singular and plural. Ignoring some wrinkles in the system due to phonological factors, and dialectal variation within the Central Group (cf. MacKenzie (1961a: 112–114) for details), the general rules for ordering the A-past and O-past markers when both appear on the verb stem are summed up in Table 4. The information given in the table can be expressed with the following rule: When the A-past marker refers to a SAP, it will always precede the O-past marker; otherwise, it follows the O-past marker. However, as Table 4 shows, there is an area of indeterminacy. When the A-past is third person plural, both possible orders are permitted. For example, ‘they saw me’ could be either dı̄-yān-im (A-O), or dı̄-m-yān (O-A) . The A-O alternative is the commoner of the two options (MacKenzie 1961a: 114). Thus the ordering of argument cross-referencers on the verb is not entirely dependent on the syntactic function of the arguments, but interacts with the category of Person. From a cross-linguistic perspective, such an interaction is well-attested, in particular in so-called inverse systems. Nichols (1992: 66) discusses related phenomena under the term “Hierarchical alignment”: Access to inflectional slots for subject and/or object is based on person, number, and/or animacy rather than (or no less than) on syntactic relations. Past transitive constructions 293 Thus we find in Suleimani a pocket of what is, according to Nichols (1992), a cross-linguistically rare alignment type. The agreement facts of Suleimani of course reflect a well-known typological generalization, noted for example by Croft (2001: 318–319): The typologically least marked constellation of A and O is when A is a SAP and O is a non-SAP, in Croft’s notation SAP→nonSAP (SAP acts on non-SAP).9 When these conditions are met in Suleimani, the A-past clitic appears to act as a suffix rather than a clitic. But argument cross-referencing on the verb can be further complicated by the presence of a marker cross-referencing ‘Indirect Participants’. The facts are as follows. An Indirect Participant can be expressed with a full form (NP or pronoun), in which case no additional cross-referencing is necessary. An example with a free form of a personal pronoun is the following: (330) la emā=y sand-in from 1 PL=3 S:CLC take:PST-3 PL ‘He took them from us’ (MacKenzie 1961a: 114) Cross-referencing of Indirect Participants is thus not obligatory. However, it is available as an alternative to the full pronoun of the preceding example. There are two possibilities. Either the Indirect Participant is realized as a clitic pronoun following, for example, a preposition: (331) lē=mān=ı̄ sand-in from=1 PL:CLC=3 S:CLC take:PST-3 PL ‘(He) took them from us’ (MacKenzie 1961a: 114) Or the Indirect Participant can be cross-referenced on the verb. In that case, it takes the form of a Set 2 verbal agreement suffix (cf. Table 2) rather than a clitic pronoun. Thus in (332) the Indirect Participant is cross-referenced with the appropriate form of the verbal suffix −ı̄n rather than the corresponding pronominal clitic =mān: (332) 9 awān=ı̄ lē sand-ı̄n 3 PL=1 S:CLC from take:PST-1 PL ‘He took them from us’ (MacKenzie 1961a: 114) This is of course a restatement of parts of the hierarchies of subject selection, discussed in Section 2.4.4 in connection with the choice of subjects for passive constructions. 294 The Central group The latter type of cross-referencing is also attested with External Possessors, as in the following (these examples demonstrate that the Indirect Participant category of Suleimani covers a similar range of functions as the Free Genitive of Old Persian, likewise subject to cliticization):10 (333) bačk-ak-ān=ı̄ a-xward-0-im / child-DEF-PL=3 S:CLC PROG-eat:PST-3 S-1 S ‘It used to eat my children’ (lit. ‘to me it ate the children’ (MacKenzie 1961a: 115) (334) šēt-aka das=ı̄ gazı̄-0-m / madman-DEF hand=3 S:CLC bite:PST-3 S-1 S ‘The madman bit my hand’ (lit. ‘to me the madman bit the hand’ (MacKenzie 1961a: 115) As was illustrated in connection with Table 4, both an O-past and an A-past marker may co-occur on the verb. And in addition to these two, a verb can even cross-reference the Indirect Participant as well:11 (335) dā-m-ı̄t-in=ē give:PST-1 S-2 S-3 PL=DIREC ‘I gave you to them’ (MacKenzie 1961a: 116) The Indirect Participant marker generally follows any cross-referencing markers on the verb. In other words, the usual order is (cf. the discussion in connection with Table 4): (336) a. When the A-past is a SAP: A-O-Indirect Participant (cf. ex. 335) b. When the A-past is third person singular: O-A-Indirect Participant 10 11 Example (333) is puzzling; one would have expected the verbal agreement suffix to have been plural rather than singular. However, MacKenzie (1961a: 131) notes that a plural O-past is generally reflected by the zero third person singular suffix on the verb – this point is taken up below. The final =ē in (335), glossed here as DIREC, is analyzed by MacKenzie (1961a: 123) as the ‘absolute’ form of the preposition a ‘to’. For the present purposes it suffices to note that this clitic is regularly attached to verbs of speech and giving, although its semantic contribution to the verb remains unclear. Past transitive constructions 295 MacKenzie (1961a: 116) notes that “the connexion between the Indirect Participant ending [. . .] and the verbal stem is tenuous”. For example, the directional particle (cf. fn. 11) may intervene between it and the verb stem, as in: (337) dāw-it-ı̄n=ē-n give:PST-2 S-1 PL=DIREC-3 PL ‘You have given us to them’ (MacKenzie 1961a: 116) However, In actual usage, forms with three overt argument cross-referencing markers on the verb are, as MacKenzie (1961a: 116–117) notes, very unusual.12 Two factors conspire to reduce the complexity of such forms in discourse. First, the preponderance of third person singular for the O-past, for which the corresponding suffix is -0. / Likewise, a third person singular Indirect Participant would also be expressed by a zero marker on the verb: 1. dā-0(2)-y(1)-m(3)-ē( / DIR ) ‘He(1) gave it(2) to me(3)’ 2. dā-m(1)-ı̄t(2)-0(3)-ē( / DIR ) ‘I(1) gave you(2) to him(3)’. The second factor is that there is often an overt constituent preceding the verb to which the A-past clitic can affix. In many cases, both factors combine, as in the following, where the Apast clitic (1) attaches to an overt Direct Object NP, and the O-past is third person singular: (338) søN=ı̄ dā-0-m / oath=3 S:CLC(1) give:PST-3 S(2)-1 S(3) ‘He(1) administered an oath(2) to me(3)’ (MacKenzie 1961a: 117) In sum, the apparently straightforward system of two distinct sets of crossreferencing markers, the agreement suffixes and pronominal clitics, each with their own non-overlapping functions, is obviously an oversimplification. A number of different factors interact in shaping the form and linear order of the markers concerned, leaving the resultant system probably the most complex aspect of Kurdish morphosyntax. In this section, I have merely presented some of the relevant data, following largely MacKenzie’s presentation 12 In fact, this example is considered ungrammatical by a native speaker of Sorani from Iran whom I consulted (Parwin Mahmutweyssi, p. c.). 296 The Central group and analysis. However, there are alternative ways of accounting for this data, which will be discussed in the following section. 6.3.2 Polysemy of agreement markers The view that the Set 2 agreement suffixes are primarily for cross-referencing the O-past is clearly in need of revision. We have observed that an Indirect Participant can also be reflected through such a suffix, as in (332), repeated here for convenience: (332) awān=ı̄ lē sand-ı̄n 3 PL=1 S:CLC from take:PST-1 PL ‘He took them from us’ Here the verb carries what appears to be the appropriate agreement suffix, but the function is not O-past. It is Indirect Participant. And in fact, when an Indirect Participant is cross-referenced on the verb, it always takes the form of the verbal agreement suffix rather than the (expected) pronominal clitic. Now when agreement is with the first person plural, as it is in (332), the difference between the form of an agreement suffix (Set 2) and the corresponding form of the pronominal clitic (=mān) is very obvious. However, for first person singular, the forms of the Set 2 suffixes and the pronominal clitics are phonologically identical – at least to judge by the level of phonetic detail provided in the available descriptions. Given the possibility that the Set 2 suffixes can express an Indirect Participant, demonstrated by (332), other examples are surely likewise open to a comparable interpretation. Consider (333) and (334), repeated here for convenience: (333) bačk-ak-ān=ı̄ a-xward-0-im / child-DEF-PL=3 S:CLC PROG-eat:PST-3 S-1 S ‘It used to eat my children’ (lit. ‘to me it ate the children’ (334) šēt-aka das=ı̄ gazı̄-0-im / madman-DEF hand=3 S:CLC bite:PST-3 S-1 S ‘The madman bit my hand’ (lit. ‘to me the madman bit the hand’ My glossing of these examples follows MacKenzie’s analysis, according to which these forms contain the zero exponent of the Set 2 suffixes, cross- Past transitive constructions 297 referencing the O-past, and a clitic pronoun, cross-referencing the Indirect Participant. However, it would be equally likely that there is no exponent of the O-past in these examples; rather, the Indirect Participant is expressed through a Set 2 suffix, affixing directly to the verb stem. In support of this argument, recall that that the O-past suffix in (333) should be plural −in, rather than zero. MacKenzie (1961a: 131) attributes the general failure to agree with a plural O-past to the use of the singular (zero) agreement suffix in such environments. Now there are also other conditions under which the verb does not agree with an O-past. This is the case when the O-past is realized in the clause as a personal pronoun, and the A-past clitic attaches to it. Under these conditions, the verb takes the ‘zero-marking’, as it would with a third person singular O-past:13 (339) min(2)=it(1) dı̄-wa-0/ 1 S=2 S:CLC see:PST-PTCPL-3 S ‘You(1) have seen me(2)’ (MacKenzie 1961a: 75) (The same proposition could be expressed without the free pronoun, in which case we would have, in accordance with the rules outlined above, both arguments cross-referenced on the verb.) MacKenzie (1961a) interprets all such instances as the use of the third person zero agreement marker (Set 2). But I suspect the zero is doing more work than is justified. The more likely explanation is that agreement with the O-past is eroding, and in many environments is supplanted by agreement with a different constituent, either an A-past, or an Indirect Participant. Thus I do not believe it is possible in all environments to decide whether a particular suffix ‘is’ a pronominal clitic, preceded by a zero-suffix, or whether it is an agreement suffix fulfilling the function of a pronominal clitic. Note that in the dialects of Piždar and Mukri, the first person plural forms of the pronominal clitics (in most dialects =mān) are often replaced by a form −in, clearly reminiscent of the corresponding Set 2 agreement suffix. In other words, the distinction between pronominal clitics and agreement suffixes has blurred, both functionally and phonologically. I maintain that in a number of environments it is questionable whether a zero-suffix should be postulated. Rather, it appears that the O-past is sim13 A further example of this constellation is provided by (322) above. 298 The Central group ply not cross-referenced at all, and the suffix has been co-opted to express either an Indirect Participant or an A-past. We have already discussed such a mixed-agreement system in connection with the plural agreement in the Northern Group (Section 5.4.2). Here we find the regular use of the plural verb agreement suffix −in, traditionally termed agreement with an O-past, used to cross-reference an A-past. Essentially, we appear to have a similar phenomenon here, with the difference that matters are greatly complicated by the existence of the pronominal clitics. In the original system, the O-past would have been the primary controller of agreement on the verb. But the general trend appears to be towards weakening or abandoning O-past agreement on the verb. We have already seen that the past transitive verb in several Iranian languages no longer agrees with anything in past transitive constructions. Gorani, not generally considered Kurdish but closely related and, crucially, exhibiting essentially the same system of obligatory pronominal A-past clitics in the past, shows no agreement with the O-past. Compare the following examples, where the verb remains in a single form despite different persons of the O-past (transcription simplified): (340) ämin tû=îm kûawä dî 1 S 2 S=1 S:CLC at.mountain see:PST ‘I saw you on the mountain’ (Hadank and Mann 1930: 296) (341) ämin=shân dî 1 S=3 PL:CLC see:PST ‘They saw me’ (Hadank and Mann 1930: 296) (342) shümä čísh-tân dî? 2 PL what=2 PL:CLC see:PST ‘what did you(PL) see?’ (Hadank and Mann 1930: 296) Note that in Gorani a past intransitive verb still agrees with its S. We can assume that between the stage where the agreement is with the O-past, and where agreement is with nothing, intermediate stages must have existed. During such stages, it is also reasonable to assume that changes of agreement patterns could have been mediated by factors such as Person and Animacy. For example, in the dialect of Bingird, O-past agreement with a third person plural argument depends on Animacy, as in: Past transitive constructions 299 (343) a. ćand wuëāx-ı̄ ćāk=ı̄ bō kir̄ı̄-n some horses-IZ good=3 S:CLC for buy:PST-3 PL ‘(He) bought(PL) some fine horses for (them)’ b. ćand šı̄r-ı̄ ćāk=ı̄ bō kir̄ı̄ some sword-IZ good=3 S:CLC for buy:PST ‘(He) bought(SG) some fine swords for (him)’ (MacKenzie 1961a: 131) In sum, the agreement rules for past transitive verbs are clearly complex, and cannot be formulated solely in terms of grammatical relations. While it is true that under certain conditions, the verb agrees with its O-past, on closer examination, the number of ‘exceptions’ to this rule, and the patterns of variation found in closely related languages are such that it casts doubt on whether such a rule should be the basis of the description. Suleimani has clearly moved on to a more complex mixed system of agreement, where both the A-past and an Indirect Participant may take precedence over the O-past agreement. In running texts, past transitive verb forms that appear to ‘agree’ with the A-past are in fact quite common, in particular verbs of speech: wut-ı̄ ‘(he) said’, where the A-past clitic attaches directly to the verb. A count of approximately 700 words of Gorani, with a similar system of clitic-cross referencing of the A-past, revealed the following results: Of the 55 exponents of an Apast clitic, 29 attached to the verb itself.14 In some cases, the A-past marker attaches to the verb stem even though there is an earlier constituent to which it could attach (glosses and transcription simplified): (344) î häkâyätä päî înäyä âwírd=im this story for this bring:PST=1 S: CLC ‘I brought this story (forward) for this (reason)’ (Hadank and Mann 1930: 314) Thus in actual discourse, it appears A-past agreement on the verb is actually quite common. There appears to be part of a general diachronic drift here, with verbal agreement with the O gradually giving way to verbal agreement 14 Based on a count of three short stories (VII–IX) in Hadank and Mann (1930: 311–316). 300 The Central group with the A. In the course of this development, the erstwhile A-past clitic pronoun gradually acquires the properties of a verbal suffix. In the Kermanshah dialect of the Central Group, this development has actually been taken still further. Here, past transitive verbs always carry an agreement marker with the A-past: (345) min sag-aka kušt=im 1 S dog-DEF kill:PST=1 S ‘I killed the dog.’ (Parwin Mahmutweyssi, p. c.) Note that although this now appears to be perfectly straightforward accusative alignment in agreement, the construction still differs from Modern Persian because the agreement marker is apparently – in origin at least – the pronominal clitic, rather than a verbal suffix, although I have been unable to verify this for all the relevant forms. It also differs from Persian in that there is no overt accusative marker on the direct object. Thus this system still lacks any type of nominal case marking. But from the perspective of agreement, it is a nice illustration of where the gradual shift in the status and position of Aagreement may be heading. 6.3.2.1 Changing alignment in agreement The facts of Suleimani suggest an old layer of verbal agreement with an Opast, reflecting the participial origins of the verb forms. But it is becoming increasingly overlaid with a more recent system of verb agreement with the A-past. There are several related factors involved. First, the general rightward drift of the clitics away from the clause initial Wackernagel position towards the verb. Second, the zero-form of the third person O-past verbal suffix, which in a sense ‘leaves the verb open’ to accepting A-past markers. Third, in terms of obligatoriness, cross-referencing the A-past actually outranks cross-referencing the O-past in Suleimani; there are environmnents, in which an O-past marker can be absent, yet all past transitive clauses require an A-past marker. Finally, there are discourse and person effects which favour certain combinations in transitive clauses. What we are witnessing in the Central Group is the gradual abandonment of agreement with an O-past, and the gradual grammaticalization of agreement with A-past. The latter begins in the form of an optional clitic Aligning case and agreement 301 pronoun, in the clause second position (Old and Western Middle Iranian). Over time, this clitic progressively moves rightward towards the verb phrase, and at the same time becomes obligatory, hence can co-occur with a fullpronoun A in the same clause. At this stage it already resembles agreement, rather than anaphora. Parallel to these developments, agreement with the O via a verbal suffix is increasingly weakened. One possible outcome of these developments, found for example in Persian, is regular agreement with an A, via a suffix, coupled with facultative agreement with an O, but via a clitic, in a sense the precise converse of the Old Iranian system. The precise pathways and relative chronologies of the developments in the demise of suffixal O-agreement, and the rise of A-agreement are extremely complex, and have yet to be thoroughly understood. 6.4 Aligning case and agreement According to Bynon (1979), Suleimani is not ergative, but accusative. But actually, determining the alignment of Suleimani is not that simple. We can, however, safely assume that Suleimani is syntactically accusative, as shown by Friend (1985a); I have seen no evidence that would cast doubt on that conclusion. Thus the question of alignment is solely concerned with morphological parameters of case marking and agreement. Furthermore, we only need concern ourselves with the past tenses, because the present tenses are uniformly accusative throughout Kurdish. In 6.2 we have already established that in its case system, Suleimani Kurdish is neutral: a single form does duty for S, A and O in all tenses, in both nouns and pronouns. Thus case-marking does not really offer much of interest for establishing alignment. Turning now to agreement in the past tenses, we find that there are differences between S and A, while O shows some similarities to both: 1. The S is obligatorily cross-referenced on the verb with a verbal agreement suffix. 2. The A is also obligatorily cross-referenced, but not via a suffix. Instead, a special set of pronominal clitics is used, regularly hosted by nonverbal elements. This set of clitics is never used to cross-reference an S. 302 The Central group 3. The O is only occasionally overtly cross-referenced, as in (327) and (328). However, when it is cross-referenced, then exclusively on the verb, and using the same set of suffixes that cross-reference an S. Bynon (1979) interprets this state of affairs as evidence for grouping S and A together, hence accusative alignment, because in both cases, agreement is obligatory. The similarities that group S and O together, on the other hand (the formal identity of the agreement marker), are discounted (Bynon 1979: 220). This conclusion contrasts radically with that of MacKenzie. According to him, an O-past is always manifested in a verbal ending of the appropriate [past] tense. [. . .] It is, however, an over-simplification to state that the verb ‘agrees’ with such a Direct Affectee [=O-past, G. H.], as is demonstrated by the frequent intrusion of the Agential [=A-past, G. H.] suffix between verbal stem and personal ending. (MacKenzie 1961a: 110) For MacKenzie, the factor of obligatoriness is relatively unimportant; he analyzes agreement with the O as basically equivalent to that of the S, based on identity of the forms concerned. Thus depending on the priorities different researchers attach to different criteria, the assessment of alignment in agreement can be very different. Personally I prefer to analyze agreement in past tenses as tripartite: S, A and O each determine a distinct, though partially overlapping, type of agreement. However, it must be conceded that there are clear signs of the agreement systems of S and A converging, thus bringing the system towards the accusative alignment claimed by Bynon. But at present, there are still quite obvious differences between agreement with S and A in past tenses. What does typology suggest about the interaction between alignments in different constructions and sub-systems of a particular language? An important contribution to this topic is Croft’s Subject Construction Hierarchy (Croft 2001: 155). Croft examines the distribution of accusative and ergative alignments in different constructions in a number of languages, and arrives at the following hierarchy: (346) The Subject Construction Hierarchy coordination < purposive < relativization < verb agreement < case marking Aligning case and agreement 303 The hierarchy is to be read as follows (Croft 2001: 155): The Subject Construction Hierarchy defines an implicational scale such that for any construction on the scale, if the construction patterns ergatively, then all constructions to the right of it on the scale will also pattern ergatively; and if the construction patterns accusatively, then all the constructions to the left of it on the scale also pattern accusatively. The crucial insight here is that ergative alignment is least marked in the rightmost sections of the hierarchy. At the far right, ergativity is tolerated without any further repercussions for the rest of the grammar. For example, case marking can be ergative, regardless of alignment elsewhere. But ergativity further to the left (for example in coordination) is contingent on the presence of ergativity elsewhere in the system (all positions further right). At this point we are concerned solely with the rightmost section of the hierarchy, with the relationship between case marking, and agreement. Among these two items, there are four possible combinations of case and agreement alignment. The hierarchy sanctions three of them, and rules one out. The nonsanctioned combination is ergative alignment in agreement, and accusative alignment in case marking. This can be formulated as a proposed universal in (347): (347) If a language has ergative alignment in its agreement, it cannot have accusative alignment in its case marking. Now a problem arises when we attempt to verify (347) for the Central Group, because the case alignment there is neither pure ergative nor pure accusative, but neutral. This problem can, however, be resolved by restating the proposed universal with reference to the Ergativity Continuum, introduced in Section 4.6.2 and repeated here as (348). Although in Chapter four, the Ergativity Continuum was applied to the alignment of case marking, it is also relevant to the alignment of agreement, which may also be ergative, neutral etc. Note that the terms marked/unmarked here refer to degrees of formal morphological markedness: (348) The Ergativity Continuum in Iranian a. E RGATIVE : O is unmarked, A is marked. b. N EUTRAL : O is unmarked, A is unmarked. 304 The Central group c. D OUBLE O BLIQUE /T RIPARTITE : O is marked, A is marked. d. ACCUSATIVE : O is marked, A is unmarked. On the Ergativity Continuum, the neutral and tripartite alignments are differentiated, with the latter being closer to the Accusative-pole of the scale. We now re-formulate (347) as follows: (349) For any language, the alignment found with agreement cannot be closer to the ergativity pole of the Ergativity Continuum than the alignment found in case marking. As far as I am aware, (349) captures most of the Iranian data,15 and appears to be consonant with the facts of Indo-Aryan summarized in Deo and Sharma (2006). It is also useful in the sense that it predicts the possible paths of diachronic change. For example, when a language has fully lost all casemarking, leading to a neutral system of case-marking, it cannot retain ergative agreement, or it should at least show clear signs of abandoning ergative agreement patterns. This happened in both western Middle Iranian, and is apparently happening now in the Central Group of Kurdish: a collapse of case marking leads to the breakdown of ergative agreement. Note that there is no logical necessity for this to happen; changes in nominal and verbal morphology can happen relatively independent of one another. But there are evidently constraints which the resulting systems must respect. The one under discussion here is that the grade of ergativity, as defined on the Ergativity Continuum, cannot be higher in agreement than in case. 6.5 Summary of the Central Group The syntax of the Central Group is remarkably close to that of the attested western Middle Iranian languages discussed in Chapter three: a complete lack of nominal case, and the ubiquitous use of pronominal clitics as a means of 15 Possible problems occur in connection with systems that have regular double-Oblique case-marking, and no agreement on the verb. However, in the systems known to me, there is usually some evidence of A-agreement on the verbs (for instance with a plural A), so agreement is perhaps never fully “neutral”. Desire, Obligation, Possession, and Ergativity 305 argument cross-referencing. The Central Group continues to exhibit its particular brand of Tense-Sensitive Alignment, with the morphosyntax of Past Transitive Constructions quite distinct from the rest of the grammar. The principle of obligatory cross-referencing the A-past through a clitic pronoun has been maintained, and indeed strengthened, so that the clitics in this function increasingly resemble agreement markers rather than pronouns. The clitic system may in a sense be compensating for the lack of case by providing a rich system of agreement, but this is speculative. And in quite a few languages, case markers and clitic pronouns coexist, so they are not mutually exclusive options. The most complex aspect of alignment in Central Kurdish is agreement. The system of cross-referencing the A-past with a pronominal clitic shows complex interactions with the suffixal agreement system on the verb. Broadly speaking, we can discern a weakening of suffixal agreement determined by the O, with the post-verbal slot becoming available for other agreement types: agreement with an A-past, and agreement with an Indirect Participant. While the intricacies remain to be fully understood, the broad diachronic current is reasonably apparent. Just how the Past Transitive Construction should be classified in terms of the accepted taxonomies of alignment systems remains an open question. My tentative conclusion is that case marking is neutral, while agreement is tripartite. In the final sections I briefly took up the issue of how alignment in agreement and case-marking interact, focusing on the proposals of Croft (2001). It transpires that if Croft’s proposals are reformulated in terms of the Ergativity Continuum, they yield a powerful heuristic in understanding constraints on possible paths of diachronic change. 6.6 Desire, Obligation, Possession, and Ergativity This section takes up a bundle of semantic and syntactic phenomena which co-occur with sufficient frequency in Iranian to warrant special treatment, but which rather cross-cut the organization of this book. Placing this section here is thus admittedly a little arbitrary, but it is no more so than the other available solutions. The issues at stake concern the syntax of expressions of desire, obligation and possession. Across a number of Iranian languages, expres- 306 The Central group sions of this nature show a striking affinity with the Past Transitive Construction in the languages concerned. Specifically, those languages which have a Past Transitive Construction which differs formally from the present-tense transitive construction (i. .e languages with TSA) tend to use a very similar construction for expressions of desire, obligation and possession (or at least a subset of these semantic types, see below). Crucially, the use of this construction is not restricted to the past tenses. Examples of such expressions, showing the parallels to the past tense transitive constructions, are provided in this section. For Badı̄n., we have already encountered extensive examples of expressions of possession, desire etc., and the parallels to the ergative construction of past tenses were pointed out in Section 5.6. In the rest of the Northern Group, however, expressions of possession, desire etc. generally take a different form. However, there is good reason to believe that it is Badı̄n. which preserves the older situation. The following example of standard Northern Group Kurdish is taken from the older written language and shows a very clear case of a NCS-construction with an expression of desire (Kurdistan 1898–1902: 1899, No. 6): (350) Heger we riheti-ya xwe divê bi-b-in if 2 PL:OBL peace-IZF REFL want:PRES:3 S IRR-be:PRES-3 PL File Christian ‘If you want your peace, convert to Christianity (lit. ‘become Christian’).’ Here the pronoun we is in the Oblique case, unlike the subject of an intransitive verb, yet it nevertheless controls the reflexive pronoun xwe, probably the most robust test of subjecthood in this variety of Kurdish (cf. Section 5.3.1.3). In the Central Group of Kurdish (Suleimaniye, glosses simplified for the present purposes), we also find the parallels: (351) azānı̄ min tō=m čand xōš awēt you.know I you=1 S:CLC how.much pleasure want:PRES(3 S) ‘You know how dearly I love you.’ (MacKenzie 1961a: 75) In this example, the final verbform is the third person singular present of wı̄stin ‘want’ (MacKenzie 1961a: 104–105). The syntax associated with it Desire, Obligation, Possession, and Ergativity 307 is basically identical to that of the Past Transitive Construction: the ‘Wanter’ is obligatorily cross-referenced via an enclitic pronoun (here =m), just as an A-past is also obligatorily co-referenced via an enclitic pronoun (see Section 6.3). Likewise, in both constructions the verb agrees with the O-past or the ‘Wanted’ respectively, or it shows no overt agreement. But crucially, unlike the Past Transitive Construction, (351) is in the present tense. A similar state of affairs holds for many Tatic languages. For example, in Tati (Yarshater 1970: 466) expressions of obligation or desire “may be followed by suffix pronouns (as if they were past transitive verbs – a common feature in Tati dialects)”: (352) āpāra amā mogost=emāne bešim Manjil last.year we want:PST=2 PL we.go Manjil ‘Last year we wanted/we had to go to Manjil.’ Likewise, the Tāti dialect of Kajal (Yarshater 1960: 281, fn. 3), the verb gah-/ gahast- ‘want, need, must’ follows the patterns of past transitive verbs, with an Oblique ‘wanter’: (353) avon megaha bešinde bāzār 3 PL:OBL want:PRES:3 S go:SUBJ:3 PL market ‘They want to go to the market.’ (to-them is necessary/desirable (that they) go to the market’ For Northern Talyshi, De Caro (2005) notes that the verb for ‘want’ (which I assume is cognate with the Kurdish verb for want/need vyān, see Section 5.6.1.3) is characterized by an ergative pattern, even in present tenses:16 (354) mï qâz=ïm pi-da-0/ 1 S:OBL goose=1 S:CLC want:PRES-3 S ‘I want a goose (to eat)’ The Wanter is in the Oblique, and the wanted entity, the goose, hosts the first person clitic, just as an O-past would generally host the A-past clitic in a Past Transitive Construction in this language. In the Central Plateau Languages (Le Coq 1903: 189–193) the verb for ‘want’ also uses the pronominal clitics 16 De Caro (2005) notes that these constructions may also occur with accusative alignment. 308 The Central group to express the Wanter, though with a number of complications. The parallels stretch back through all the attested stages of western Iranian. A good example from late Middle Western Iranian is the following, repeated from Section 2.5 above: (355) ā=šān ān abāyēd ka=šān gyān az tan be then=3 PL this is.necessary when=3 PL soul from body out šawēd go:PRES:3 S ‘then this is necessary for them when their souls go from their bodies.’ (Williams 1990a: 13b.3) Here again it is the clitic pronoun that expresses the Obligee, just as it does the A-past. In Early Judaeo-Persian a similar construction is also attested for expressions of obligation (with (’)b’yd ‘be necessary’, Paul 2003a: 178–179). Likewise, in Early New Persian, the Obligee could be expressed with a clitic pronoun (Heston 1976: 200), but interestingly, also with a fronted pronoun marked with rā.17 While we still await a comprehensive survey, the examples above strongly suggest that the parallels between Past Transitive Constructions, and expressions of possession, desire and obligation are too pervasive and too widespread to be mere coincidence. I suggest that they are ultimately rooted in the semantics of Indirect Participation, the notion of perceived relevance and interest, discussed in Chapter two. The core of this notion is benefaction. I have argued that in the Iranian languages, the notion of Indirect Participation came to be regularly expressed with the Genitive case, and its later offspring, the Oblique. The same notional meaning is also expressed via the clitic pronouns. Thus Indirect Participation is deeply anchored into the morphosyntax of the entire family. This family-specific linkage of form to meaning facilitated the spread of the Oblique case and the clitic pronouns to be markers of agency in past tenses. The common element combining desire, possession, obligation and agency is benefaction; in the syntax, they could be expressed with an Indirect Participant combined with an intransitive predication: 17 It is notable that in Early New Persian, the Innovated Object Marker rā assumed several of the Indirect Participant functions of the old Oblique case, but it was not used to express an A-past. Indirect Participant usage of rā is extremely restricted in Modern Persian. Desire, Obligation, Possession, and Ergativity 309 Possession: for me/in my interest it-exists Desire/Obligation: For me/in my interest it-is-necessary Agency: For me/in my interest it-is-done Many Iranian languages have preserved quite faithfully the totality of these constructions, which I argue led to the emergence of ergativity in past tenses. But it is not necessary that they do so; the Past Transitive Constructions and expressions of desire etc. have parted company in many languages, hence obscuring the original conflation. However, I believe the existing parallels provide further support that the theory sketched in Chapter two is on the right track. Chapter 7 Conclusions 7.1 A brief synopsis We are now in a position to suggest some answers to the question posed at the beginning of this book: why did the alignment associated with past transitive clauses came adrift from the rest of the grammar, and what were the actual mechanisms involved? Several ingredients were necessary in bringing about the Iranian developments. First, I assume that in the Old Iranian period, non-canonical subject constructions were available for predicative expressions of possession (‘to-me is’), and quite possibly for other semantic types as well (desire, obligation, necessity). There is sufficient evidence in the oldest attested branches of Indo-European, and in Iranian itself to warrant this conclusion. The case of the non-canonical subject was the Dative, which in Old Iranian syncretized with the Genitive. Typically, such non-canonical subjects co-occurred with intransitive stative predicates. And in the earliest stages of Indo-European, they were often combined with non-finite verb forms. The second crucial ingredient was the loss of the finite past-tense verb forms Aorist and Perfect. This left the resultative participles in −ta as the principle carriers of the semantic component ‘past tense’. Although originally verbal adjectives, these participles became increasingly integrated into the verbal paradigm of tenses (with or without copula support), where they formed the basis for a new system of past tenses. But the participles were deficient in a manner that the old finite forms were not: the participles, being basically resultative adjectives, could not assign accusative case. They were lexically intransitive, just as the cognate participles in countless IndoEuropean languages are. As a result, for the past tenses, an alternative means of accomodating direct objects was necessary. The crucial development occurred when the pre-existing non-canonical subject construction came to be combined with a participle as predicate. Initially, a possessive reading was presumably possible (‘their battle wasfought’), but an agentive reading was also available: ‘they fought battle’. In 312 Conclusions this manner, the two participants of a transitive proposition could be accommodated without assigning the Accusative case to the O. As discussed extensively in Section 2.4.4, the NCS construction already shared crucial semantic and pragmatic features with a canonical transitive clause, hence easing the extension. From the original, perhaps occasional extensions, it was presumably a case of gradual spread of the NCS to an ever greater number of participles, perhaps following in a lexeme-by-lexeme fashion the progressive loss of the finite tense forms. For West Iranian, these developments must have worked themselves out in the centuries between Old Persian and the earliest attested Middle Iranian texts, because in the latter, the rift between the present transitive construction – with straightforward accusative alignment – and the Past Transitive Construction was firmly established. Regrettably, details of this process cannot be directly observed due to gaps in attestation, but the later languages provide a rich source of data for the construction of hypotheses. Notice that in the above account, agented passives simply play no role, and there are good reasons for such a conclusion. First, agented passives in ancient stages of Indo-European were extremely rare, and only relatively weakly grammaticalized. Second, the Genitive/Dative would not have been the expected case of the Agent-phrase (Section 2.5.5). Third, agented passives serve a completely different pragmatic function to active transitive clauses, and have correspondingly different distributions of topicality and animacy in their nominal constituents (Section 2.4.4–2.5.6). Finally, in an agented passive, it is the underlying O which is the subject – both morphological and syntactic. Yet nowhere in Iranian, past or present, is there evidence from those Past Transitive Constructions that grew out of the m. k. construction that this was ever the case (Section 2.5.6). My overall conclusion is that agented passives represent a quite distinct and actually very marginal development, and can safely be ignored in formulating explanations of the emergence of non-accusative alignments in past tenses of Iranian. Given that the passiveorigins theory for both Polynesian and for Indo-Aryan are by no means unanimously accepted in the literature, the import of this conclusion for theories of diachronic syntax is considerable: uncontroversial evidence for the supposedly “well-attested [. . .] reanalysis of earlier passive constructions” (Lightfoot 1999: 140) is in fact notably absent. The non-accusative alignments of the past tenses thus grew out of a combination of developments in verbal inflectional morphology (the loss of past A brief synopsis 313 Basic 2-argument construction schema High topicality/animacy Low topicality/animacy Possessor Theme Experiencer Stimulus Agent etc. Patient SEM. ....................................................................... SYNT. NP-1 NP-2 Subject properties No subject-properties NCS-construction Accusative construction Possessor Theme Agent Patient Experiencer Stimulus etc. Experiencer Theme etc. ...................................................................................... NP-OBLIQUE Some subject properties NP-DIRECT No subject properties NP-DIRECT Subject properties NP-OBLIQUE Direct object Agent Theme Drift Experiencer etc. Patient SEM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . −→ towards AccusaSYNT. NP-OBLIQUE NP-DIRECT tivity Subject properties No subject-properties Canonical Ergative Construction Inherits morphosyntax of the NCS-construction Inherits semantics of the Accusative Construction Figure 1. Schematic overview of the development of the Past Transitive Construction in Iranian finite verb forms), and the fusion of a non-canonical subject construction with a participial predicate. The developments are shown schematically in Figure 1. 314 Conclusions At the top of the diagram, a ‘Basic two-argument construction schema’ is shown. It is important to note that it represents an abstraction, intended to capture the commonalities shared by the two, still distinct, constructions assumed to have existed in Old Iranian: the accusative construction, and the NCS construction. Both contain two arguments, with a typical distribution of topicality and animacy, and in both constructions, the first NP exhibits certain subject properties. There are also certain overlaps in the semantic roles, with the Experiencer role being possible in the NP-1 of both constructions. It was the existence of these broad syntactic and pragmatic parallels that ultimately guided the later developments. Note, however, that the two differ crucially in their morphology, hence represent two distinct constructions: NP-1 of the accusative construction is Nominative, NP-1 in the NCS construction is Oblique (Dative, Genitive, or general Oblique, depending on the language). The ergative construction, shown at the bottom of the diagram, represents the blending of the two: It inherits the semantics of the accusative construction, because it expresses clauses associated with the same set of predicates (transitive verbs), differing only in the time value (past vs. non-past). But its morphology is inherited from the NCS construction. In a sense, the ergative construction is pulled in two directions. In the subsequent history of Iranian, it has generally succumbed to the attraction of the accusative construction, hence the arrow on the bottom right of the diagram indicating the drift towards accusativity. Step by step, the formal properties of the accusative construction have been taken on board: the Oblique marking of the O, some verbal agreement with the A, and so on. The majority of languages investigated here have Past Transitive Constructions that can best be understood in this light: as ergative constructions that have adopted a sub-set of the properties of the accusative construction. The parallels between accusative construction, NCS construction and ergative construction are illustrated particularly clearly in Badı̄n. Kurdish, as discussed in Section 5.6. The three constructions are illustrated for this language below: (356) ACCUSATIVE C ONSTRUCTION : Present tenses of transitive verbs tu tišt-ek-ı̄ di-bı̄n-ı̄ 2 S:DIR thing-INDEF-OBL IND-see:PRES-2 S ‘You (can) see something.’ A brief synopsis 315 (357) E RGATIVE C ONSTRUCTION : Past tenses of transitive verbs te tišt-ek dı̄t 2 S:OBL thing-INDEF see:PST(3 S) ‘You saw something.’ (358) NCS-C ONSTRUCTION : Possession, desire, obligation (both tenses) te tišt-ek d-vē-t 2 S:OBL thing-INDEF IND-be.necessary:PRES-3 S ‘You want/need something.’ In most of the languages investigated, however, the parallels have blurred. It is a basic fact that all languages have retained the accusative construction. But the fates of the ergative construction and the NCS construction have been quite divergent. In some languages, the NCS construction has disappeared completely, leaving the ergative construction ‘stranded’, the sole construction in the language with an Oblique subject. This is the case in most dialects of Kurdish (Northern Group), as it is in neighbouring Zazaki (Paul 1998b: 144).1 In others, it is the ergative construction that disappears, while (some) NCS constructions remain. This appears to have happened in Ossetic, where the “subject” of verbs expressing obligation, desire and belief (Thordarson 1989: 470) takes the Genitive (cognate with the Oblique case of the other languages we have been considering), but otherwise case marking is accusativealigned. Middle Persian actually re-introduced external possession after the old Oblique was lost. The fronted possessor was marked with the Innovated Object marker =rā (Lazard 1963: 376). And even for Modern Persian, one can argue that an NCS construction exists in expressions such as man xoš=am miyāyad ‘me, my-pleasure comes’, I like (it)’, although the language has lost all trace of the ergative construction. The uneven fate of the NCS construction in Iranian is reminiscent of the situation in other branches of Indo-European. In Germanic, for example, English lacks an NCS construction, German has at best rudimentary traces, 1 Zazaki retains a trace of the NCS in the so-called “Pertinenzdativ” (Paul 1998b: 47), an Oblique form used for the possessor of body-part terms, though not as subjects. I believe this is the typical final niche in which NCS constructions are retained after a language has otherwise lost them. 316 Conclusions while in Icelandic they continue to thrive. Russian and Polish differ also in that the former allows a type of external possession (u menja . . . ‘at me (is X)’), which is lacking in Polish, and so on.2 Thus relatively low diachronic stability – and corresponding vulnerability to contact influence – appears to be a feature of NCS constructions. What was crucial for the Iranian scenario, as I have sketched it above, was that the presence of the NCS construction coincided chronologically with the developments in the tense morphology (the loss of finite past tense forms). Once the ergative (or non-accusative) construction had become firmly established with past-tense verbs, the NCS construction could quietly fade away or not, as the case may be. As we have seen, they have nevertheless survived in a number of languages for expressions of desire etc. (Section 6.6), and where they do, the parallels to Past Transitive Constructions remain unmistakeable. The subsequent developments in the Past Transitive Constructions involved several distinct components, one of which I will briefly take up here: the issue of the transitivity of the verb forms. As mentioned above, originally the verb forms were resultative participles, hence unable to assign the Accusative case. In this sense, I consider them intransitive. But they became the predicates regularly used to express two-participant events in past tenses, occurring in clauses that looked, syntactically, very much like transitive clauses. There thus arose a fundamental tension between a lexical specification as ‘intransitive’ on the one hand, and the distribution in clauses expressing, to all intents and purposes, straightforward transitive propositions. How was this tension resolved? My answer to that question would be: ‘slowly’! As I have repeatedly stressed, the syntactic constructions on which the Past Transitive Construction was modelled were already characterized by precisely that kind of tension. For example, the typical predicative expression of possession in Old Iranian (with an External Possessor) involved two participants, a Possessor, and a Possessed, yet the predicate was the intransitive ‘be, become’. Here then we already have the mismatch between a two-argument syntactic construction on the one hand, and an intransitive verb on the other. But the tension in the Past Transitive Construction would have been much greater. The NCS construction, unlike the ergative construction, was asso- 2 I am grateful to Szymon Słodowicz for pointing out the Slavic example to me. Areal pressure and alignment change 317 ciated with a specific semantic niche, containing just a fairly small number of predicates of possession and physical/mental perception etc. The ergative construction on the other hand could not be readily associated with any welldefined semantic domain. It was used with all transitive verbs, regardless of the roles involved. Furthermore, the same verbs had a different alignment in the present tenses, so it was not possible to link the ergative construction to any clearly definable lexical semantics. Ultimately the pressure to bring the expressions based on the same verb lexeme into line with each other (what I refer to as cross-system harmony in Section 4.5) led to the abandonment of distinct Past Transitive Constructions. In Section 3.7.1, I introduced the term lingering intransitivity to refer to the gradual passage from participle, a verbal adjective, to finite verb form displaying normal transitivity values. In most of the Middle Iranian texts examined the effects of lingering intransitivity can be observed: Past tense verb forms are capable both of occurring in the Past Transitive Construction, with an A, or of being used in their resultative, agentless senses. The same feature crops up in some modern Iranian languages, as we have seen in Chapter five. The available facts are not amenable to a neat and tidy solution. Until someone arrives at the theoretically more elegant solution, I am content to point out that the verbs involved in the Past Transitive Construction cannot readily be classified in terms of a binary transitive vs. intransitive feature. Rather, they appear to be positionable along a gradual diachronic cline from intransitive, to transitive. 7.2 Areal pressure and alignment change It is now widely recognized that language contact has a profound affect not only on the lexicon of a language, but also its syntax (Harris and Campbell (1995), Heine and Kuteva (2005), Stilo (2005), among many others). What effect has areal influence had on the development of alignment in Iranian? First, we should note that Tense-Sensitive Alignment (TSA) of the Iranian brand is very much a family affair, restricted to Iranian and its closest relatives, the Indo-Aryan languages. From this fact alone we can conclude that the main triggering factor is shared genetic origins, not geographic contiguity. However, there is one very striking example of non-Iranian languages which have developed a similar system of non-accusative alignments in their 318 Conclusions past tenses: Eastern Neo-Aramaic, encompassing several languages spoken in southeastern Turkey, North Iraq and parts of Iran. These languages formed “larger or smaller language islands” (Jastrow 1997: 347) among a largely Kurdish-speaking populace, and were thus subject to long and intense contact with the Northern Group of Kurdish. Eastern Neo-Aramaic has developed a brand of TSA showing uncanny structural parallels with the Iranian construction: the past-tense verb forms are based on past participles, and the A-past is cross-referenced via a suffix which is identical to the suffix used for a possessor. Consider the forms from the T.uroyo dialect, where the suffix −li expresses both past A, and Possessor (from Jastrow (1997: 362, 374), stress is omitted in my transcription): grı̆š-li ‘I pulled him’; kit-li ‘I have (it)’, lit. ‘there.is-to.me’. Furthermore, TSA is notably lacking in the closelyrelated Western Neo-Aramaic dialects, which is strongly suggestive that its presence in the Eastern group is linked to language contact. Thus TSA of the Iranian brand can also spread beyond the family to the neighbours. But given the overall extent of Iranian language contacts, this example of contactinduced alignment shift stands out mainly for its comparative isolation; other plausible cases of TSA spreading to contact languages are not known to me (although Armenian is a possible candidate). Nichols (1992) claims that alignment is a diachronically stable genetic trait, hence likely to characterize all members of a particular family, while less likely to cross genetic borders into neighbouring languages. Taken in its absolute form, this claim is too strong. For one thing, we have seen that Persian, for example, changed its past-tense alignments completely over about 1000 years, while related languages retained the old alignments. And if the East Aramaic example is accepted as contact-induced, then it is clear that alignment patterns can, under exceedingly favourable conditions, cross genetic borders. But Nichols nevertheless correctly identifies the powerful link between alignment, and genetic groupings. As noted, once Tense-Sensitive Alignment entered the Iranian languages, it was duly inherited, in one form or another, by all the offspring. This is in stark contrast to many other features of phonology or grammar, for example clicks, basic word order, techniques of clause linkage and complementation (Matras 2002), adpositional order (Stilo 2005), or NCS constructions (see preceding section), which readily adapt to an areal profile, often cross-cutting genetic borders (Haig 2001). Drinka (1999: 493)’s more cautious conclusion on the diachronic stability of align- Areal pressure and alignment change 319 ment is a fair approximation of the facts: “alignment systems do appear to be more conservative than some other features of the grammar”. Tense-Sensitive Alignment is presumably less likely to diffuse because it is not a single, salient feature of a grammar, but a bundle of co-occurring features, linked to both verbal and nominal morphology. Furthermore, I have suggested that the roots of Tense-Sensitive Alignment in Iranian lie in changes to the tense morphology of finite verbs. Verb morphology is, as Johanson (2002: 152) notes, perhaps the most impenetrable aspect of a grammar, hence most likely to remain constant. It is a striking fact that the basic two-stem system of Iranian verbs characterizes most of the languages since the Middle Persian stage, although case-morphology, or person marking, have developed in so many different ways. And one necessary condition for Iranian-style Tense-Sensitive Alignment is a highly specific constellation of verb stems – something which is invariably inherited, but rarely borrowed. Nevertheless, sub-aspects of alignment systems are certainly borrowable. In particular, Innovated Object markers (see Section 4.3) may be adopted from neighbouring languages; Persian appears to exert strong pressure on its neighbours in this respect. And once such Innovated Object markers have entered the system, the old non-accusative alignments generally begin to die out. More generally, several authors have suggested that intense contact to languages with uniform accusative alignment generally leads to a breakdown of Tense-Sensitive Alignment. Conversely, Farrell (2003) suggests that contact to other languages with Tense-Sensitive Alignment may strengthen its presence. Thus Eastern Balochi, in contact to Indo-Aryan languages, generally has retained more ergativity than western Balochi. A very general trend is the following: outlier languages, cut off from other Iranian languages with TSA, have a higher probability of losing TSA completely. Examples are Gilan Kurdish (Section 5.4.5), Turkmen Balochi (Axonov 2006), and Ossetic (Thordarson 1989). All three languages have entirely lost TSA, even though close relatives have retained it, and all three are isolated from their relatives. There thus appears to be a ‘supportive-nest’ effect in maintaining TSA, although this tendency remains to be confirmed. 320 Conclusions 7.3 Alignment in Indo-European Over the years, a number of suggestions have been put forward regarding the alignment of Proto-Indo-European (PIE). One view is that accusative alignment was predominant, as articulated by, for example, Kurzová (1999: 502). The opposing view, however, that PIE was ergative, has since lost favour, mainly because the assumed system would have run counter to the wellestablished universals of animacy-based alignment splits (see Section 4.6.1). More recently, several scholars have argued for an active (i. e. split or fluid-S) system in at least part of the PIE grammar. This theory has become quite influential, with some scholars even apparently accepting it as an established fact (Hewson and Bubenik 2006: 288). The Iranian data considered here actually casts some interesting light on these theories, which I will briefly comment on here. The main proponent of the active-alignment theory has been Winfred Lehmann, developing ideas of Klimov and others; my discussion is based on more recent work by Bauer (2000). In a language with an idealized active alignment, there is no distinction between transitive and intransitive verbs, and no corresponding distinction in the syntax between two fundamentally different clause types. Instead, verbs are distinguished semantically into active and stative verbs (under various labels). An important consequence is that in such a language, there is no necessity to postulate three categories S, A and O. Instead, only two are necessary, corresponding roughly to the subject of the active and stative verbs respectively (cf. the terminology of Dixon (1994) “split, or fluid-S” ). Active alignments are claimed for several native American Indian languages. Their most central typological trait is, according to Bauer (2000: 88), “the absence of transitivity as a grammatical feature.” Other features that commonly cooccur are the centrality of an animate/inanimate distinction on nouns, weak word classes, a paucity of nominal case marking, and rich morphology on the verb. Bauer (2000) argues that the earliest stages of PIE were also characterized by a significantly weaker transitive/intransitive distinction than the later languages. This, she claims, is supportive evidence for reconstructing an active alignment for earliest PIE. Later stages of the language were characterized by a successive “spreading of transitivity” (Bauer 2000: 350). Lack of transitivity is apparently associated with “nominal” rather than “verbal” syntax, and Alignment in Indo-European 321 in the “history of Indo-European a consistent process took place by which nominal syntax gave way to verbal syntax.” There are several sources of evidence that can be brought to bear on the PIE-active-theory (see especially Drinka (1999) and Comrie (2001) for discussion). Bauer (2000) chooses to focus on what she calls “impersonal constructions”. Impersonal constructions cover three main categories: meteorological verbs, verbs of emotion and experience, and modal verbs (obligation etc.). In addition, Bauer also discusses at length the mihi est-type of possession in Latin. It should be evident that, with the exception of meteorological verbs, these are the same types of predicate that occur again and again in NCS constructions cross-linguistically. Bauer concludes that (a) constructions of this type are an archaic feature of Indo-European, and (b) they are related to numerous types of “absolute constructions” (ablative absolutes, genitive absolutes etc.) with nominal verb forms (e. g. participles). Both of these conclusions are entirely in accord with my claims regarding the emergence of ergativity in Iranian. However, for Bauer, their main relevance lies elsewhere: she considers them to be relics of the “nominal syntax” that accompanied an assumed active alignment in the earliest stages of PIE (just how early is not actually specified). Throughout the subsequent history of Romance, the main data source for Bauer (2000), these traces of nominal syntax were gradually replaced by the verbal syntax of finite verb forms, and canonical subjects. I have several reservations regarding the relevance of “impersonal constructions” for determining the alignment of a language, or its ancestor. In Section 7.1 we noted considerable differences across the Iranian languages with regard to the extent that they deploy impersonal constructions of the type Bauer is referring to. Other branches of Indo-European are likewise quite heterogenous in this respect. The general trend that Bauer notes, then, which involves a steady decline of such constructions, may be true for the developments in Romance from Latin through to French, but in other branches, these constructions continue to thrive, or may have been reintroduced. Thus to view them as “relics” of a putative early PIE active stage is not fully justified; my impression is that a language can lose or gain these constructions relatively swiftly. The Northern Group of Kurdish is a case in point: In Badı̄n., mihi est-type possessives, and similar types of expressions for desire, obligation and physical perception abound. In the mutually intelligible dialects a few kilometers further north, they are almost completely lacking. 322 Conclusions More generally, I fail to see how the “impersonal constructions” are indicative of a “lack of transitivity”. If that were the case, then we should expect Icelandic, for example, to show evidence of such a lack of transitivity, or an active alignment, but to my knowledge this has never been suggested. I believe there is a basic confusion of levels here; most of the verbs involved in the “impersonal construction” are simply intransitive verbs. The fact that they enter into a particular construction, with an oblique first argument, tells us that the language concerned licenses such constructions as part of its inventory of syntactic constructions. It does not, however, mean that there is no basic distinction between transitive and intransitive verbs in the language concerned. Very different evidence, primarily from verbal inflection and voice morphology, would be necessary to back up the far-reaching claims regarding a “lack of transitivity”. Now I did of course note that the past-tense verbs of Iranian were characterized over centuries by what I termed lingering intransitivity, basically a tension between lexical intransitivity, and deployment in a twoargument construction. But this was quite a specific development, affecting erstwhile participles. It did not characterize the lexicon of verbs generally, which continued to be dividable into two relatively well-defined classes of transitive and intransitive, generally applicable in the Iranian languages to this day. It is necessary to dwell on these issues here because otherwise, the emphasis I have put on the NCS construction in the emergence of ergativity in Iranian might be misconstrued as support for the active-alignment view of PIE alignment. Let me clarify that point: I do claim that the presence of NCS constructions in Old Iranian was crucial in triggering the development of ergativity. But the presence of such constructions is simply not relevant to the issue of what the overall alignment in pre-Old Iranian was, and certainly not in PIE. In fact, I believe their presence is orthogonal to determining the alignment of a particular language, or its ancestors. My impression is that alignment in Old Iranian was quite straightforwardly accusative, as far back as the historical records allow us to project, and, for this branch at least, there is no compelling reason to assume that things were ever any different. Much could be said on the relevance of the Iranian developments for alignment in the other branches of Indo-European, but this must remain a future challenge. Most pressing is a comparison with the developments in Indo-Aryan. Whether the mechanisms proposed here provide a plausible ac- On explanations for change 323 count remains to be seen. There are also striking similarities in the Slavic branch. In Slavic, finite past verb forms became defunct, and a non-finite form of the verb became the basis for past tenses. What emerged in Russian at least could actually be called a version of Tense-Sensitive Alignment: agreement with past tenses works quite differently to agreement with present tenses (the former reflecting gender and number, the latter person). However, alignment in case marking is uniform in all tenses. This suggests that the nonfinite forms that were pressed into service in Slavic past tenses were capable of assigning Accusative case. Presumably, they were more like infinitives, as opposed to the resultative participles which were the source of the modern Iranian past tense verbs, but these issues remain to be resolved. However, particularly in earlier stages of Lithuanian, very striking parallels to the Iranian developments have long been noted (Lavine 1999; Ambrazas 2004), but these constructions remained relatively marginal. In fact, what makes the Iranian developments unusual in the Indo-European context is not so much the nature of the individual developments (loss of finite past tenses, presence of an NCS construction etc.), which have numerous parallels throughout IndoEuropean. For example, loss of the old finite past tenses characterizes almost the entire family. What was exceptional in Iranian was the combination of these factors at a particular time, leading to the massive extension of NCS constructions to become the sole means of expressing past transitive clauses. These developments must have been the result of a contingent combination of contributing factors, unique to Iranian, and yet to be fully understood. 7.4 On explanations for change The type of explanations proposed for syntactic change are intimately dependent on the model of syntax behind a particular analysis. Consider for example the following quote, illustrating nicely how some of the basic claims of a particular model, here what Culicover and Jackendoff (2005) refer to as Mainstream Generative Grammar (MGG), can dictate the way historical syntax is practised: Language change is by definition a failure in the transmission across time of linguistic features. Such failures, in principle, could occur within groups of 324 Conclusions adult native speakers of language, who for some reason substitute one feature for another in their usage, as happens when new words are coined and substituted for old ones; but in the case of syntactic and other grammatical features, such innovation by monolingual adults is largely unattested. Instead failures of transmission seem to occur in the course of language acquisition; that is, they are failures of learning. (Kroch 2001: 699, emphasis added.) If one assumes that the syntax and the lexicon are fundamentally distinct, then Kroch’s assumption that change in one component will be fundamentally different to change in the other follows quite naturally. From a Construction Grammar perspective, however, where the organizational principles obtaining in the syntax and the lexicon are not viewed as radically distinct, there is no necessity to assume that syntactic change is quite distinct from change in the lexicon. In fact, as I have argued in earlier sections, it is precisely the polysemy of constructions that provides the impetus for change, just as polysemy of lexical items is a prerequisite for change in the lexicon (Evans and Wilkins 2000). Furthermore, there is no necessity to assume that syntactic change cannot occur among “monolingual adults”. But perhaps the most crucial consequence of the Mainstream Generative Grammar approach to syntactic change concerns the locus, and the speed of change: if one considers the syntax to be an autonomous, modular system, then it is not likely to undergo significant change once it has matured. Therefore, the locus of language change must be first language acquisition: “children arrive at grammars generating data quite different from grammars of an earlier generation.” (Lightfoot 1999: 150) Thus we would predict syntactic change to be, on the whole, quite rapid, occurring during language transmission from one generation to the next. Most of the evidence from this study, however, does not support this view. On the contrary, the alignment changes considered are, as far as we can discern from the available data, quite gradual, taking many centuries to work themselves through. Furthermore, over the long term, that is over several centuries, change can follow a clear trajectory: a drift in a particular direction. Smaller sub-changes may briefly run counter to the overall drift, but they will tend to be levelled out when a sufficient time-depth is considered. For many historical linguists, this conclusion is self-evident, and has been accepted for more than a century. But from a MGG perspective, the existence On explanations for change 325 of such long-term cross-generational drift poses a problem, simply because there is no place within the theory for accomodating such mechanisms. Consider the following comments of Lightfoot (1999: 209): “[. . .] the fallacy is in requiring a principled explanation for such large-scale change taking place over such a long period. What’s wrong with a series of coincidences, or a series of independent events?” Lightfoot goes on to ridicule “mystical” explanations for large-scale changes, such as the shift from one word order type to another: This raises the question of how a child attains a language which is exactly half-way between the subject–verb–object and the subject–object–verb types; how does she know whether this is a subject–verb–object language changing into a subject–object–verb, or vice versa? How does she know that her generation must push the language a little towards, say, an[sic] subject–verb–object type? It seems that the only conceivable answer is to postulate a racial memory of some kind, such that the child knows that this is a subject–verb–object language changing towards subject–object–verb. For Lightfoot, for whom language change, like language itself, is something primarily seated in the brain of the individual, the phenomenon of large-scale change following a common path across hundreds of generations is indeed a paradox. But it nevertheless exists. Consider for a moment the Iranian languages. At some point during, or prior to, the Old Iranian stage, the verb system began to break down, leaving participles as the major carriers of past tense meanings. That particular change triggered off a series of changes in the syntax of all the descendent languages, although over the course of time the speakers became separated by thousands of miles, and came to speak mutually unintelligible languages. Now at any given point in that time span, speakers had no conception of how the language had been centuries before, nor how it would be centuries later. Yet the cumulative effect of dozens of generations were, despite the many differences of details, remarkably similar: non-accusative alignments throughout the past tenses. And the later development was always in a similar direction: a progressive levelling out of differences between the Past Transitive Construction and the corresponding present-tense constructions. While the languages differ as to how far down this pathway they go, and also differ with regard to the details of their routes, the general direction 326 Conclusions remains unmistakeable. No language changed its present tenses; no language took its past tenses further away from its present tenses. It goes without saying that the “racial memory” solution mentioned by Lightfoot is a straw-man, which no serious historical linguist has ever even considered. The only reasonable explanation for the undeniable existence of long-term directed change (drift) in related languages is that language acquisition does not involve parameter setting. Instead, language acquisition is primarily input-driven, sensitive to finer-grained irregularities, semantic effects, and differences in token frequency. The grammar acquired by a child will differ from that of its parents, but not radically. But functional pressure can for example work in favour of increasing the frequency of certain forms over its rivals. And over time, the incremental effects of such tiny differences leads to certain constructions being exclusively favoured in certain environments. Syntactic change over long periods is thus best understood in terms of the existence of constructional variants for particular meanings at any given point in time, with their relative frequencies and meanings changing gradually over time. This is certainly the only type of account that is consonant with the data presented in this book. The reluctance to accept this type of scenario is, I suspect, ultimately rooted in the conviction that the only meaningful sense of the term grammar is in the sense of Chomsky’s I-language (Chomsky 1986: 27), the biological endowment of an individual. Indeed, in the same passage Chomsky explicitly denies that the conventional notion of “a language” has any theoretical status in a scientific theory of language. On this view of grammar, the notion of changes spanning dozens of centuries simply makes no sense; where are the supposed regularities of this type of change to be located – hardly in the minds of speakers. But this view of grammar is unnecessarily reductionist. There is also a sense of grammar at a higher order, the regularities that obtain across a speech community, consisting among other things in statistically valid patterns of usage that emerge through the interaction of many individuals sharing – more or less – the same grammar. And at this level, that of interaction, language change happens in the manner outlined above. This is not simply a re-statement of the distinction between competence and performance. Rather, it is the reaffirmation of the view that language communities have grammars, and these too are systems, albeit of a somewhat different nature to the mental grammars of individual speakers. On explanations for change 327 The view that syntactic change occurs during first language acquisition is closely linked to the view of syntax as Parameter Settings: Universal Grammar comes pre-installed with certain options, parameters, which guide speakers in the acquisition process by reducing the number of possible grammars available. Crucially, parameters involve a bundling of certain grammatical features together under a single parameter, so that when the value for one of the relevant features under a particular parameter is correctly established, acquisition of the others follows automatically.3 Consider an example from Baker (2001), the so-called Polysynthesis Parameter. This parameter subsumes, among other features, obligatory agreement with objects and subjects, and the lack of independent reflexive pronouns like English herself. The parameter-theory makes interesting predictions regarding syntactic change: if one sub-aspect of a parameter changes its setting, then this model predicts a cluster-effect, with the other sub-aspects shifting their setting quite rapidly in accordance with the pre-set combination in the parameter. It is thus in principle an attractive model, because it makes explicit claims, and is integrated into a more general theory of grammar. However, parameter setting as a model for language acquisition is rejected in Tomasello (2003: 182–195), and its application to diachronic syntax has also been the object of considerable criticism (Harris and Campbell 1995: 34–45). The broader issues go beyond the scope of this book; here I will briefly assess the viability of a parameter-setting account for the alignment changes we have been surveying. Both Baker (2001) and Lightfoot (1999) assume the existence of an “Ergativity Parameter”. The central aspect of this parameter is, according to Lightfoot (1999: 139), a rule of case assignment. The rule basically states that S and O receive case when they move from the VP to the specifier of the immediately dominating category, “perhaps an agreement element”, while an A must move further, “to the specifier of the next functional category, perhaps Tense”. Lightfoot suggests that a parameter re-setting could occur when the morphology originally associated with an accusative system became opaque or eroded, leading children to reanalyze the system, hence to a new parameter 3 The more recent view of parameters within the Minimalist Program locates them in the lexicon, thus confining basically all differences between individual languages to the lexicon. This potentially interesting development is not taken up here. 328 Conclusions setting. No particularly clear evidence is provided for such a process, except a brief re-statement of the Polynesian case (the passive-origin analysis). But as we have seen, a simple rule of case-assignment will not account for most of the Iranian data. Very few languages have pure ergativity. They show a variety of intermediate stages. Nor is there good evidence that these systems change abruptly in one direction or another. The claim that the parameter is re-set following erosion of old accusative morphology is also not particularly convincing. In fact, one and the same morphological system can still host two quite different alignments, for example in the Northern Group of Kurdish, where Gilan Kurdish has full accusative alignment, while the (mutually-intelligible) other dialects of the group have various forms of ergative or near-ergative alignment in past tenses (Section 5.4.5). I am unaware of any obvious change in the inflectional morphology of Gilan Kurdish that would have somehow triggered the change to accusative alignment. Finally, note that in many Iranian languages the expected correlation between nominal case and agreement is not maintained; verbs may agree with NPs in the Oblique case, depending on factors such as person and number. In fact, it is hard to imagine how one would define the syntactic rules to be bundled together in a putative Ergativity Parameter in Universal Grammar, nor how the existence of such a parameter would benefit a child acquiring the particular brand of non-accusative alignment in Pashto, Kurdish or Middle Persian. The child must still learn, on the basis of quite specific input, the particular caseassignment and agreement patterns prevalent in her language community, and must also learn that it is associated only with certain verb-forms. This is, incidentally, precisely what a construction-based account of language acquisition, such as Tomasello (2003), proposes. In short, parameter re-setting has little to offer as an explanatory model for alignment change in Iranian (see also Newmeyer (2005: 76–103) for a more general critique of the parametric approach to grammar, and to syntactic change). An alternative to the formalist explanations for syntactic change are socalled functionalist approaches. The basic idea is that change is not so much determined by the pre-given settings of an autonomous Universal Grammar, but by the demands imposed on speakers and listeners in the actual course of language use. Such demands can be motivated in physical properties of the human perceptual and articulatory system, short-term memory, but also more general principles governing the way human beings interact with one On explanations for change 329 another, both linguistically and extra-linguistically: empathy, power relations, cooperation as a means of achieving shared goals, and so on. One fairly typical application of functional explanations to grammar concerns case systems, discussed in Section 4.5. The explanation runs as follows: a central communicative function of language is to distinguish who is acting upon whom in a given event. Case marking should, therefore, ideally fulfil this function by ensuring that A and O are formally distinguished from each other. Furthermore, grammatical systems should avoid redundancy. Therefore, the functionally ideal case systems are the accusative and the ergative respectively: both distinguish A from O, but do so with the minimum expense in terms of formal markers. And indeed, there is good evidence for a cross-linguistic preference for these two types of systems: globally, they are statistically the most frequent of the several theoretically possible types. Change, one might expect, should also move in the direction of either of these two systems. And in the long-term, it actually does. But in the short term, it does not necessarily work that way. Intermediate steps are attested which actually produce ‘less efficient’ case systems, such as the double-Oblique, or the complete lack of case marking (through most of Middle Persian, for example). Another common change was to re-deploy the old Oblique Plural marker of early western Middle Iranian so that it became a general marker of plural number – thereby losing its ability to indicate case distinctions. There is thus obviously a paradox here: on the one hand, functional explanations for change are supposedly rooted in real-world demands on successful communication between speakers, yet the changes that best illustrate the impact of functional principles often take many generations to work themselves out. How can language change be driven by functional considerations, yet nevertheless produce ‘dysfunctional’ intermediate stages, of no apparent benefit for generations of speakers? Again, the only reasonable solution to this paradox is the assumption of competing constructional variants, present in differing frequencies at any given time point in any given language. Their realization is dependent on a variety of conditioning factors (subtle semantics, register, prestige, regional variation etc.). A complex interplay of functional, phonological, contact-induced and other factors works on the system, pulling it in different directions and hence yielding, over the short term, seemingly unpredictable results. But over extended periods, certain factors may ultimately tip the balance in favour of a particular constellation (for 330 Conclusions example, typological harmony). Such a view of historical change is perfectly consonant with that held in other disciplines, as in for example the following quote from Diamond (1998: 424): Thus, the difficulties historians face in establishing cause-and-effect relations in the history of human societies are broadly similar to the difficulties facing astronomers, climatologists, ecologists, evolutionary biologists, geologists and paleontologists. To varying degrees, each of these fields is plagued by the impossibility of performing replicated, controlled experimental interventions, the complexity arising from enormous numbers of variables, the resulting uniqueness of each system, the consequent impossibility of formulating universal laws, and the difficulty of predicting emergent properties and future behaviour. Prediction in history, as in other historical sciences, is most feasible on large spatial scales and over long times, when the unique features of millions of small-scale brief events become averaged out. The fascination of the Iranian languages from the perspective of historical syntax lies in the potential for actually observing the regularities in such truly long-term developments. Let me close with some comments on the relevance of language typology for understanding alignment changes. I have presented the alignment changes in Iranian in terms of sub-developments: the history of the case-system, the developments of pronominal clitics, and agreement. One drawback of this approach is that it is difficult to relate to the traditional typological literature on alignment, which treats alignment as a taxonomy of discrete alignment types, and formulates its theories in these terms. Likewise, much of the alignment literature in formal syntax is formulated in terms of just a small number of idealized alignment types (see above on parameter setting). My reason for decomposing alignment to its component parts is motivated by the data at hand: in the Iranian Past Transitive Constructions, case-alignment, and agreement develop to some extent independently of one another. Standard accounts of agreement in languages with alignment splits postulate a link between case and agreement, such that agreement is uniformly with the unmarked NP (Nominative, or Absolutive, Mohanan 1994: 105). But as we have seen, the past transitive constructions of Iranian provide numerous exceptions to this rule. In fact, the Iranian data rather suggest that the existing taxonomy of alignment types is merely a construct, a crude heuristic that captures some very broad tendencies, but little more. On explanations for change 331 Does this imply that each morphological subsystem is free to pursue its own pathways of change, constrained only by phonology and paradigminternal considerations? To some extent, it does, and this is actually why we find such a proliferation of mixed alignment types. The changes in case morphology, for example, are on the whole of the commonplace type that can be observed in most Indo-European languages, irrespective of their alignments. But there are nevertheless some more subtle tendencies in the observed changes which underscore the validity of typological generalizations. In particular, many of the complex systems outlined in Chapter four demonstrate with great clarity the relevance of the Animacy Hierarchy in guiding morphosyntactic change. Likewise, the Subject Construction Hierarchy given in Section 6.4 provides a principled explanation for the lack of certain combinations of case and agreement. And finally, a fairly articulated model of changes in case systems was made possible by introducing an Ergativity Continuum (Section 4.6.2), which situates different alignments on a scale of relative markedness of the O. Thus while morphological change in various subsystems is ultimately just that and no more, the potentially unlimited outcomes that arise through the combination of individual diachronic developments are demonstrably constrained by universal principles, uncovered through synchronic typological research on the limits of variation across the languages of the world. Appendix A.1 Case in Old Persian An overview of the Old Persian case system is provided in Table 1 (based on Brandenstein and Mayrhofer 1964). It gives the case forms for five different types of nominal: (i) Nouns continuing the Indo-European -u- and -ūstems (apparently only sparingly attested in plural forms; the a-stems (continuation of Indo-European o-stems) are more richly represented in the plural); (ii) nouns with -ā stems; (iii) the case forms of the demonstrative iyam ‘this’ (in three genders); (iv) the case forms of the first person pronoun; (v) the case forms of the third person pronoun (only the full forms of the pronouns are given here, see Table 2 for the clitics). It will be seen from Table 1 that case marking of the -u-/-ū stems come close to what can be assumed for Indo-European, apart from the loss of the Dative. However, in the other types of noun shown, the number of formal distinctions drawn is notably fewer. For example, there does not appear to be any other part of the grammar where a formal distinction between Ablative Singular and Instrumental Singular is maintained (although of course the Table 1. The Old Persian case system ‘ THIS ’ Idg.-u-,-ū- -ā M. F. N. 1.P ERS . 3.P ERS . SG. Nom. Acc. Inst. Abl. Gen. Loc. -uš -um -uvā -auš -auv -auš -auv -avā -ā -ām -āyā -āyā -āyā -āyā iyam imam anā – – – iyam imām – – – ahyāyā ima ima – – – – adam mām – – manā – hauv – – – – – PL. – – – -ūnām – -ā -ā – -ānām -āuvā imaiy imaiy – – – imā imā – amāxam – – imā imaibiš – – yayam – – – – – – – – – Nom. Acc. Inst. Gen. Loc. 334 Appendix relevant forms may well have been present, but simply not attested). Ignoring for a moment comparative evidence from other ancient Indo-European languages, the justification for assuming two distinct cases (Ablative and Instrumental) in Old Persian is actually very thin, resting solely on forms from one declensional class. The forms of the -ā stems show the most extensive syncretism, with a single form for Instrumental, Ablative, Genitive and Locative. The table shows further that these cases are not distinguished (or at least not attested) in the available forms of the proximal demonstrative either.4 The constellation found in this declensional class foreshadowed the later developments in the case system towards Middle Persian, where a single Oblique case is attested.5 Schmitt (1999: 99–104) interprets some of the forms found in Late Achaemenian texts as evidence for the collapse of the case system well before the Middle Persian period. In other words, the case system sketched in Table 1 can be considered to be a fairly conservative one, and in the contemporary vernaculars, the breakdown of the system may well already have progressed much further. A.2 Changing rules of clitic placement There are two interacting tendencies observable in the syntax of clitics throughout the history of western Iranian: Rightward drift, and head attraction. The main facts are summed up in Table 2 overleaf, which charts the possibility of different types of elements to act as clitic hosts across different chronological stages. The data has been collated from a number of sources, and must be read together with the accompanying notes. 4 5 These distinctions are also not attested for the Old Persian relative article hya. The distal demonstrative (masculine) ava, however, did distinguish an Ablative avanā from an Genitive avahyā. In the plural no Ablative is recorded, and the form given in the table as Instrumental covered functions normally associated with the Ablative – cf. Kent (1953: 82). Changing rules of clitic placement 335 Table 2. Clitic hosts insome West Iranian languages H OST Co-/subordinator Governing Preposition Possessed NP Finite verb Past transitive verb PARTHIAN M ID .P ERS . E.N EW P ERS . N EW P ERS . yes nob yes/noc yesd yese yes yes/nob, f yes/noc, f yesd, f yese, f yes/noa yes/nob yes yesd yese no yes yes yes/nod yese Key: ‘yes’=well attested, few restrictions; ‘yes/no’=attested, but subject to restrictions and/or infrequent; ‘no’=not attested. Notes to Table 2 a Early New Persian allowed the relativizer and the ‘if’-subordinator to host clitics, but not the ‘and’-coordinator (Heston 1976: 89–90). b This criterion constitutes a major syntactic distinction between Parthian, where clitics are virtually never attested hosted by a governing preposition (Brunner 1977: 110–113), and Middle Persian, where the three simple prepositions required their complements, if pronominal clitics, to attach to them. For Early New Persian see Heston (1976: 90, esp. fn. 33). c Parthian and Middle Persian: a possessed noun is a possible host, but there are two apparent qualifications: first, only a noun, not a postposed adjective modifying that noun, is a possible host (for Middle Persian (Pahlavi): Heston (1976: 91), Brunner (1977: 96); for Parthian: Brunner (1977: 98–99), though one example there is unclear). A second qualification is that although a noun may be the phonological host of a clitic, it is (in Middle Persian at least) not usually the case that the clitic is the Possessor of its host; rather, it stands in a relationship to another element of the clause (Brunner 1977: 98). Whether this also holds for Parthian is not evident in the sources, but it is certainly the case that nouns regularly host clitic pronouns that are not their Possessors (and I suggest that several of the examples of “Possessors” quoted in the sources are better interpreted as Indirect Participants). In Early New Persian, it is clearly the NP as a phrase that hosts a clitic Possessor (cf. the examples in (Heston 1976: 88)), and this state continues through to Modern Persian. d Parthian and Middle Persian allow clitics on finite verbs, typically imperatives (exclusively so in Parthian, Brunner 1977: 105), because these occur towards the beginning of the phrase. The clitics express generally Direct Object or Indirect Participants. In New Persian, these functions are also expressed through clitics on the verb, although the verb is most commonly towards the end of the clause. However, in other modern West Iranian languages, finite verbs may be impossible, or at best a ‘last resort’ for clitic placement (e.g. Kurdish, Central Group, cf. the discussion in Section 6.2.3.) e Possible in Parthian, where the clitic always cross-referenced an A-past (Brunner 1977: 105); in Middle Persian attachment to a past tense form was possible for clitics crossreferencing an A-past (Brunner 1977: 104–105), and apparently “in late texts”, also as an S with an intransitive past form. On Middle Persian, see Note (6) below. In Early New Persian as Object and Indirect Participant cross referencer (cf. .tlb krd-nd=š searching do: PST - 336 Appendix 3 PL =OBL .3 S‘(they) searched for him’, unfortunately no data on the possibility of A-past clitics on such forms (Heston 1976: 89). If it was possible, it did not survive into New Persian. Lazard (1963: 257) notes “des vestiges de cette construction” in Firdausı̄, and Don Stilo (p. c.) informs me that in colloquial Tehrani Persian, forms such as goft-aš ‘said=3 S’, ‘he said’, with a pronominal clitic indicating the A, are possible (cf. Lazard (1963: 257– 258) for discussion). f For Middle Persian, Heston (1976: 143, fn. 33) states that in her materials, 186 pronominal clitics are attached to the conjunction AP ‘and’, while the next most frequent host is the subordinator AMT [ka], which accounted for 25 occurrences. Although she does not give the total number of clitics attested in her materials, these two figures strongly suggest that in Middle Persian Pahlavi, the vast majority of hosts are clause-initial coordinators and subordinators, with other hosts playing a marginal role. 7.2.1 Rightward drift As noted, in terms of their placement, Middle Iranian clitics are rather similar to Old Iranian clitics: both tended towards the Wackernagel (clause-second) position. Nevertheless, this tendency was already showing some signs of weakening in the Middle Iranian period, and in the later languages, it has been almost entirely abandoned. This phenomenon is referred to here as rightward drift; it appears to be a common West Iranian phenomenon, with only very few exceptions (see Chapter three, fn. 10). In this sense, the general drift parallels that postulated for Romance (Vincent 2001: 30–31), from clause-initial to verb-adjacent (see also Franks and King (2000) for Slavic data on the distinction between clause-initial and verb-adjacent clitics). 7.2.2 Head attraction In Western Middle Iranian, there were already two interesting sets of exceptions to the second-position rule. First, in Middle Persian, but not in Parthian, clitics could attach to the prepositions that govern them, for example pt=š ‘with/to=him’ (Heston 1976: 91). In fact, according to Sundermann (1989a: 157), the three primary prepositions of Middle Persian (pad, az, ō ‘to/for etc., from, to’) always required a pronominal complement to attach to them, regardless of position in the clause (see Brunner (1977: 110–113) for extensive examples and discussion). The second group of exceptions to the second- Changing rules of clitic placement 337 position rule cited by Heston (1976: 90) involve clitic pronouns attaching to nouns, where they can be construed as expressing the possessors of those nouns. Both of these tendencies can be described in terms of a single, overriding principle, which I will term head attraction: rather then being dependent on prosodic factors, which leave clitics in clause-second position, clitics gravitate towards their governing heads, for example an adposition, or a possessed noun. Note that head attraction almost invariably accompanies a rightward drift of the clitics, because the heads concerned are seldom in clause-initial position, but occur later in the clause. Thus head attraction coincides with rightward drift, already familiar from the history of Romance languages, where in Latin, pronominal clitics were clause-second, but their reflexes in the modern Romance languages now cluster within the verb phrase (Vincent 2001: 30). Head attraction is most obviously at work when clitics are (i) governed by an adposition; (ii) when they express a possessor, or (iii) a direct object. Thus in modern Persian we have: (359) a. dast=am ‘my hand’ b. be=m ‘to me’ c. didam=aš ‘I saw him/her’ P OSSESSED =P OSSESSOR P REPOSITION =C OMPLEMENT V ERB =O BJECT In these examples head attraction has the effect of making the clitics more closely resemble some form of agreement suffix, rather than a clitic. We can observe thus fairly clearly the grammaticalization of the pronominal clitics. However, the notion of head attraction has rather interesting implications for clitics expressing a past tense A, or an Indirect Participant. In connection with these two types of clitics the question arises as to what, if anything, can be considered their governing head? There is no obvious answer, or at least none as obvious as there is in the case of clitics governed by a preposition, for example. For a past-tense A, the answer will depend on how these elements are to be analyzed syntactically, that is, as full subjects, or just syntactic subjects, or something in between. But it will also depend on one’s syntax theory. If one considers “subject” to be to be a structurally defined position in a particular type of representation, for example the Specifier position of the IP (Faarlund 2001: 100), this suggests that the subject does not really have any governing head (the Specifier position being largely neutral with regard to headhood). If on the other hand one follows some version of the VP-internal subject theory 338 Appendix it might be argued that the verb is the relevant head. For Indirect Participants (e.g., Benefactives), similar problems apply, because their status with regard to the adjunct/complement distinction is controversial. References Aikhenvald, Alexandra, R.M.W. Dixon, and Masayuki Onishi, ed. 2001 Non-canonical marking of subjects and objects. Amsterdam: Benjamins. – Cited on p. 19, 262 Allen, Cynthia 1995 Case marking and reanalysis. Grammatical relations from Old to Early Modern English. Oxford: Oxford University Press. – Cited on p. 175 Allen, W. Sidney 1964 Transitivity and possession. Language 40:337–343. – Cited on p. 29 Ambrazas, Vytautas 2004 On the genitive with neuter participles and verbal nouns in Lithuanian. In Studies in Baltic and Indo-European linguistics: in honour of William R. Schmalstieg, ed. Philip Baldi and Dini Pietro, 1–6. Amsterdam: Benjamins. – Cited on p. 87, 323 Anderson, Stephen 1977 On mechanisms by which languages become ergative. In Mechanisms of syntactic change, ed. Charles Li, 317–363. Texas: University of Texas. – Cited on p. 30 Anderson, Stephen 1992 A-morphous morphology. Cambridge: Cambridge University Press. – Cited on p. 9 Anonby, Erik 2004 Kurdish or Luri? Laki’s disputed identity in the Luristan province of Iran. Kurdische Studien 4:5–20. – Cited on p. 202 Arkadiev, Peter 2004 “Two-term case systems in the Indo-Iranian languages: a typological perspective.” Paper held at LENCA 2, 11–14.05.2004, State University of Kazan, Tataristan. – Cited on p. 177 Axonov, Serge 2006 The Balochi language of Turkmenistan. a corpus-based grammatical description. Uppsala Universitet: Acta Universitatis Upsaliensis. – Cited on p. 139, 167, 319 Aydogan, Ibrahîm Seydo. undated Sorê Gulê û romaneke nîcomayî. www.nefel.com, part of an electronic corpus supplied by Abdullah Incekan, p.c. – Cited on p. 217 Azami, Cheragh Ali, and Gernot Windfuhr 1972 A dictiorary of Sangesari with a grammatical outline. Tehran: Sherkat-e Sahami-ye Katabhaye Jibi. – Cited on p. 153 Back, Michael 1978 Die sassanidischen Staatsinschriften. Leiden: Brill. – Cited on p. 98 Badıllı, Kemal 1992 Türkçe izahlı kürtçe grameri. İstanbul: MED. – Cited on p. 204 340 References Bailey, Denise 2005 A comparative study of grammatical relations in Northern Kurdish. Unpublished MA-Thesis: Seminar für Iranistik, University of Göttingen. – Cited on p. 206, 234, 236, 269 Baker, Mark 2001 The atoms of language: The mind’s hidden rules of grammar. New York: Basic Books. – Cited on p. 14, 327 Baksî, M 1991 Gundikî Dono. İstanbul: Gökyüzü Sanat Ürünleri. – Cited on p. 226, 227 Baldi, Philip 2002 The foundations of Latin. Berlin: de Gruyter. – Cited on p. 64 BardDal, J., and T. Eythórsson 2003 The change that never happened: the story of oblique subjects. Journal of linguistics 39:439–472. – Cited on p. 33 Barnas, Uso, and Johanna Salzer 1994 Lehrbuch der kurdischen Sprache. Berlin: Înstîtuya Kurdî. – Cited on p. 234 Bashir, Elena 1986 Beyond split-ergativity. Subject marking in Wakhi. Proceedings of the 22nd regional meeting of the Chicago Linguistics Society 15–35. – Cited on p. 7, 97, 194 Bauer, Brigitte 1999 Impersonal Habet constructions in Latin: At the cross-roads of Indo-European innovation. In Language change and typological variation: In Honor of Winfred P. Lehmann on the occasion of his 83rd birthday. Vol. II: Grammatical universals and typology, ed. Carol Justus and Edgar Polomé, 590–612. Washington: Institute for the Study of Man. – Cited on p. 64, 258 2000 Archaic syntax in Indo-European. The spread of transitivity in Latin and French. Berlin: Mouton. – Cited on p. 86, 320, 321 Bedir Khan, Emir, and Roger Lescot 1970 Grammaire kurde. Paris: Adrien Maisonneuve. – Cited on p. 204, 221, 235, 236 Beekes, Robert 1988 A grammar of Gatha-Avestan. Leiden: Brill. – Cited on p. 4 Benveniste, Emile. 1952/1966. La construction passive du parfait transitif. In Problèmes de linguistique generale, ed. Emile Benveniste, 176–186. Paris: Gallimard [all quotes are from this 1966 reprint]. – Cited on p. 27, 28, 31, 76 Besnier, Niko 2000 Tuvaluan: A Polynesian language of the Central Pacific. London: Routledge. – Cited on p. 40 Bhaskararao, Peri, and K. V. Subbarao, ed 2004 Non-nominative subjects, Vols. 1 and 2. Amsterdam: Benjamins. – Cited on p. 19 Blake, Barry 1994 Case. Cambridge: Cambridge University Press. – Cited on p. 45 References 341 Blau, Joyce 1975 Le Kurde de Amādiya et de Djabal Sindjār. Analyse linguistique, textes folkloriques, glossaires. Paris: Klincksieck. – Cited on p. 204, 233, 240, 257, 269, 271 1989 Le kurde. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 327– 335. Wiesbaden: Reichert. – Cited on p. 201, 202 Blau, Joyce, and Veysi Barak 1999 Manuel de kurde kurmanji. Paris: L’Harmattan. – Cited on p. 204, 208, 221, 234, 250 Bossong, Georg 1985 Empirische Universalienforschung. Differentielle Objektmarkierung in den neuiranischen Sprachen. Tübingen: Narr. – Cited on p. 2, 13, 96, 152, 153, 158, 167, 168, 177, 179, 189 Boyce, Mary 1975 A reader in Manichaean Middle Persian and Parthian. Leiden: Brill. – Cited on p. 91, 93, 109, 110, 111, 112, 115, 118, 119, 122 Bozarslan, M. Emîn 1982 Gurê bilûrvan. Borås: Invandrarförlaget [Collection of short stories]. – Cited on p. 231 Undated Meyro. Borås: Invandrarförlaget. – Cited on p. 217, 247, 249 Brandenstein, Wilhelm, and Manfred Mayrhofer 1964 Handbuch des Altpersischen. Wiesbaden: Harrassowitz. – Cited on p. 24, 46, 64, 333 Brunner, Christopher 1977 A syntax of Western Middle Iranian. Delmar, New York: Caravan. – Cited on p. 93, 106, 108, 109, 335, 336 Bubenik, Vit 1989 An interpretation of split ergativity in Indo-Iranian languages. Diachronica VI:181–212. – Cited on p. 14, 29 1994 On Wackernagel’s law in the history of Persian. Indogermanische Forschungen 99:105–122. – Cited on p. 2, 113 2001 On the actualization of the passive-to-ergative shift in Pre-Islamic India. In Actualization. Linguistic change in progress, ed. Henning Andersen, 95–118. Amsterdam: Benjamins. – Cited on p. 2 Bulut, Christiane 2000 Indirectivity in Kurmanji. In Evidentials. Turkic, Iranian and neighbouring languages, ed. Lars Johanson and Bo Utas, 147–184. Berlin: Mouton. – Cited on p. 227 Butt, Miriam 2001 A reexamination of the accusative to ergative shift in Indo-Aryan. In Time over matter. Diachronic perspectives on morphosyntax, ed. Miriam Butt and Tracy Holloway King, 105–141. Stanford: CSLI. – Cited on p. 2 342 References Bynon, Theodora 1979 The ergative construction in Kurdish. Bulletin of the School of Oriental and African Studies 42:211–224. – Cited on p. 29, 201, 257, 277, 301, 302 1980 From passive to active in Kurdish via the ergative construction. In Papers from the 4th International Conference on Historical Linguistics, ed. Elisabeth C. Traugott, Rebecca Labrum, and Susan Shepherd, 151–163. Amsterdam: Benjamins. – Cited on p. 29, 44, 79, 277 2005 Evidential, raised possessor, and the historical source of the ergative construction in Indo-Iranian. Transactions of the Philological Society 103:1–72. – Cited on p. 30, 81 Cardona, George 1970 The Indo-Iranian construction mana (mama) kr.tam. Language 46:1–12. – Cited on p. 29, 30, 76, 77, 79, 94 Cewerî, Firat 1986 Girtî. Stockholm: Nûdem. – Cited on p. 212, 216, 231, 234, 238, 242 2001 Çîroka min û Nûdemê. Nûdem 39–40. – Cited on p. 217, 247, 249 Chafe, Wallace 1994 Discourse, consciousness, and time. Chicago: Chicago University Press. – Cited on p. 236 Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris. – Cited on p. 16 1986 Knowledge of language. Its nature, origin, and use. Westport, Connecticut: Praeger. – Cited on p. 326 Christensen, Arthur 1935 Contributions á la dialectologie iranienne II. Dialectes de la région de Sèmnān: Sourkhéı̄, Lāsguerdı̄, Sängesärı̄ et Chämerzādi.. København: Levin & Munksgaard. – Cited on p. 153, 154, 160, 183 Chyet, Michael 2003 Kurdish-English dictionary. New Haven: Yale University Press. – Cited on p. 235 Cole, Peter, Wayne Harbert, Gabriella Hermon, and S.N. Sridhar 1980 The acquisition of subjecthood. Language 56:719–743. – Cited on p. 33 Comrie, Bernard 1978 Ergativity. In Syntactic typology. Studies in the phenomenology of language, ed. Winfred Lehmann, 329–324. Hassocks, Sussex: Harvester. – Cited on p. 6, 33 1988 Passive and voice. In Passive and voice, ed. Masayoshi Shibatani, 9–23. Amsterdam: Benjamins. – Cited on p. 36, 37, 38, 47 1989 Language typology and language universals. Chicago: University of Chicago [2nd Ed.]. – Cited on p. 50, 75, 173, 174, 176, 179, 180, 263 2001 Typology and the history of language. In Aspects of typology and universals, ed. Walter Bisang, 21–35. Berlin: Akademie. – Cited on p. 321 Corbett, Greville 2000 Number. Cambridge: Cambridge University Press. – Cited on p. 99, 180 References 343 2003 Agreement: The range of the phenomenon and the principles of the Surrey Database of Agreement. Transactions of the Philological Society 101:155– 202. – Cited on p. 282, 289 Croft, William 2001 Radical construction grammar. Syntactic theory in typological perspective. Oxford: Oxford University Press. – Cited on p. 6, 20, 21, 40, 262, 293, 302, 303, 305 Şemo, Ereb 1977 Şivanê kurd. İstanbul: Özgürlük. – Cited on p. 231, 233, 237, 248, 250 Şirîn (ed.) 1996a An explanation of Kurmanji grammar, Behdînî dialect. Nî Roj (no address). – Cited on p. 258, 261, 262 1996b Stories told in Kurdistan, Behdînî Kurmanji, as told by Buşra. Nî Roj (no address). – Cited on p. 233 Culicover, Peter, and Ray Jackendoff 2005 Simpler syntax. Oxford: Oxford University Press. – Cited on p. 15, 323 Dabir-Moghaddam, Mohammad 2007 “Split cross-referencing and typological variation in the Iranian languages.” Paper held at the Second International Conference on Iranian Linguistics, Hamburg, 17–19.08.2007. – Cited on p. 113, 116 Dahl, Östen, and Maria Koptjevskaja-Tamm 2006 The resilient dative and other remarkable cases in Scandinavian vernaculars. Sprachtypologie und Universalienforschung 59:56–75. – Cited on p. 99, 103 De Caro, Gerardo 2005 “A text-based account of argument-marking in Northern Talysh.” Unpublished MA Thesis, SOAS, London. – Cited on p. 138, 148, 149, 307 2006 “Patterns of pronominal alignment in Northern Talyshi.” Paper held at the Leipzig Students’ Conference in Linguistics, Max Planck Institute for Evolutionary Anthropology, Leipzig, March 25–26, 2006. – Cited on p. 149 Deo, Ashwini, and Devyani Sharma 2006 Typological variation in the ergative morphology of Indo-Aryan languages. Linguistic typology 10:369–418. – Cited on p. 2, 153, 304 Diamond, Jared 1998 Guns, germs and steel. London: Vintage. – Cited on p. 330 Dik, Simon 1978 Functional Grammar. Amsterdam: North Holland. – Cited on p. 189 Dixon, R.M.W 1972 The Dyirbal language of North Queensland. Cambridge: Cambridge University Press. – Cited on p. 38, 103, 182 1977 The syntactic development of Australian languages. In Mechanisms of syntactic change, ed. Charles Li, 365–415. Texas: University of Texas. – Cited on p. 38, 39 1994 Ergativity. Cambridge: Cambridge University Press. – Cited on p. 6, 7, 8, 9, 33, 87, 215, 320 344 References Dorleijn, Margreet 1996 The decay of ergativity in Kurdish. Tilburg: Tilburg University Press. – Cited on p. 13, 204, 221, 225, 226, 227, 228, 234, 241, 242, 256, 257 Drinka, Bridget 1999 Alignment in early Proto-Indo-European. In Language change and typological variation: In Honor of Winfred P. Lehmann on the occasion of his 83rd birthday. Vol. II: Grammatical universals and typology, ed. Carol Justus and Edgar Polomé, 464–520. Washington: Institute for the Study of Man. – Cited on p. 63, 258, 318, 321 2003 The Perfect in Indo-European. Stratigraphic evidence of prehistoric areal influence. In Language contacts in prehistory. Studies in stratigraphy, ed. Henning Andersen, 77–105. Amsterdam: Benjamins. – Cited on p. 42 Du Bois, John 1985 Competing motivations. In Iconicity in syntax, ed. John Haiman, 343–365. Philadelphia: Benjamins. – Cited on p. 7 1987 The discourse basis of ergativity. Language 63:805–855. – Cited on p. 50, 51, 73, 181, 229 Du Bois, John, Lorraine Kumpf, and William Ashby, ed. 2003 Preferred argument structure. Amsterdam: Benjamins. – Cited on p. 51 Durkin-Meisterernst, Desmond 2006 The Hymns to the Living Soul. Middle Persian and Parthian texts in the Turfan collection. Turnhout, Belgium: Brepols. – Cited on p. 91, 101, 115, 119 Džalil, O., and Dž. Džalil 1978 Kurdskii folklor, zargotina K’urda (Vol. I). Moskva: Izdatels’tvo Nauka. – Cited on p. 240 Edmonds, C.J 1955 Prepositions and personal affixes in Southern Kurdish. Bulletin of the School of Oriental and African Studies XVII:490–502. – Cited on p. 282, 286, 289 Estival, Dominique, and John Myhill 1988 Formal and functional aspects of the development from passive to ergative systems. In Passive and voice, ed. Masayoshi Shibatani, 441–492. Amsterdam: Benjamins. – Cited on p. 33, 34, 87 Evans, Nicholas, and David Wilkins 2000 In the mind’s ear: the semantic extensions of perception verbs in Australian languages. Language 76:546–592. – Cited on p. 22, 324 Everett, Daniel 1996 Why there are no clitics. An alternative perspective on pronominal allomorphy. University of Texas, Arlington: SIL. – Cited on p. 282 Eythórsson, T., and J. BardDal 2003 Oblique subjects: A Germanic inheritance! Working Papers in Scandinavian Syntax 71:145–202. – Cited on p. 33 References 345 Faarlund, Jan Terje 2001 The notion of oblique subject and its status in the history of Icelandic. In Grammatical relations in change, ed. Jan Terje Faarlund, 99–135. Amsterdam: Benjamins. – Cited on p. 75, 337 Farrell, Tim 1995 Fading ergativity? A study of ergativity in Balochi. In Subject, voice and ergativity. Selected essays, ed. David Bennett, Theodora Bynon, and B. George Hewitt, 218–243. London: School of Oriental and African Studies. – Cited on p. 12, 185, 186 2003 Linguistic influences on the Balochi spoken in Karachi. In The Balochi and their neighbours. Ethnic and linguistic contact in Balochistan in historical and modern times, ed. Carina Jahani and Agnes Korn, 167–209. Wiesbaden: Reichert. – Cited on p. 11, 251, 319 Fattah, Ismaïl Kamandâr 2000 Les dialectes kurdes méridionaux. étude linguistique et dialectologique. Leuven: Peeters. – Cited on p. 142, 176, 202 Fattah, Muhammad Maruf 1997 A generative grammar of Kurdish. University of Amsterdam: Unpublished PhD Thesis. – Cited on p. 291 Filiminova, Elena 2005 The noun phrase hierarchy and relational marking: problems and counterevidence. Linguistic Typology 9:77–113. – Cited on p. 180 Foley, William 1991 The Yimas language of New Guinea. Stanford: Stanford University Press. – Cited on p. 229 Franks, Steven, and Tracy Holloway King 2000 A handbook of Slavic clitics. Oxford University Press: Oxford. – Cited on p. 336 Fried, Mirjam 1999 From interest to ownership. A constructional view of external possession. In External possession, ed. Doris Payne and Immanuel Barshi, 473–504. Amsterdam: Benjamins. – Cited on p. 58 Friend, Robyn. 1985a “Ergativity in Suleimaniye Kurdish.” Unpublished paper presented at the Middle East Studies Association Annual General Meeting, November 1985. – Cited on p. 301 1985b Some syntactic and morphological features of Suleimaniye Kurdish. Unpublished PhD thesis: UCLA. – Cited on p. 111 Givón, Talmy 1983 Topic continuity in discourse: an introduction. In Topic continuity in discourse: a quantitative cross-language study, ed. T. Givón, 1–31. Amsterdam: Benjamins. – Cited on p. 245 346 References Goldberg, Adele 1995 Constructions. A construction grammar approach to argument structure. Chicago: University of Chicago Press. – Cited on p. 15, 16, 17, 22 2006 Constructions at work. The nature of generalization in language. Oxford: Oxford University Press. – Cited on p. 15, 21 Greenberg, Joseph 1966 Some universals of grammar with particular reference to the order of meaningful elements. In Universals of grammar, ed. Joseph Greenberg, 73–113. Cambridge, Mass.: MIT. – Cited on p. 174 Hadank, Karl, and Oskar Mann 1930 Kurdisch-persische Forschungen, Abteilung III, Band II: Mundarten der Gûrân, besonders das Kändûläî, Auramânî und Bâdschaälânî, bearbeitet von Karl Hadank. Berlin: De Gruyter. – Cited on p. 106, 143, 146, 147, 148, 150, 184, 242, 287, 298, 299 Haig, Geoffrey 1998 On the interaction of morphological and syntactic ergativity: Lessons from Kurdish. Lingua 105:149–173. – Cited on p. 32, 213 2001 Linguistic diffusion in present-day East Anatolia: From top to bottom. In Areal diffusion and genetic inheritance: problems in comparative linguistics, ed. Alexandra Aikhenvald and R.M.W. Dixon, 195–224. Oxford: Oxford University Press. – Cited on p. 318 2002a Complex predicates in Kurdish: Argument sharing, incorporation, or what? Sprachtypologie und Universalienforschung / Language typology and universals 55:15–48. – Cited on p. 11, 12, 239, 286 2002b The Corpus of Contemporary Kurdish Newspaper Texts (CCKNT): A pilot project in corpus linguistics for Kurdish. Kurdische Studien 1:109–122. – Cited on p. 203 2004a Alignment in Kurdish: A diachronic perspective. Unpublished Habilitationsschrift: Philosophische Fakultät der Christian-Albrechts-Universität zu Kiel. – Cited on p. v, 236 2004b Das Genussystem in der kurdischen Sprache: Strukturelle und soziolinguistische Aspekte. In Gender in Kurdistan und der Diaspora, ed. Siamend Hajo, Carsten Borck, Eva Savelsberg, and Dogan Şukriye, 33–58. Münster: Lit. – Cited on p. 205 2006 Turkish influence on Kurmanjî Kurdish. Evidence from the Tunceli dialect. In Turkic-Iranian contact areas. Historical and linguistic aspects, ed. Lars Johanson and Christiane Bulut, 283–299. Wiesbaden: Harrassowitz. – Cited on p. 204 2007 The emergence of the Tense Ezafe in Badînî Kurdish. Paper held at the Second International Conference on Iranian Linguistics, Hamburg, 17–19.08.2007. – Cited on p. 270 forthc. a Spoken Kurdish from the Erzurum and Muš regions. Wiesbaden: Reichert. – Cited on p. 204, 265 References 347 forthc. b Tempering typology with diachrony: explaining animacy-based alignment splits in iranian. In Studies in the typology of Iranian languages, ed. Behrooz Mahmoodi Bakhtiari. München: Lincom. – Cited on p. 167 Haig, Geoffrey, and Yaron Matras 2002 Kurdish linguistics: a brief overview. Sprachtypologie und Universalienforschung / Language typology and universals 55:3–14. – Cited on p. 203, 204 Haig, Geoffrey, and Ludwig Paul 2001 Kurmanjî kurdish. In Facts about the World’s languages: An encyclopedia of the World’s major languages, past and present, ed. Jane Garry and Carl Rubino, 398–403. New York/Dublin: Wilson. – Cited on p. 204 Hale, Mark 1988 Old Persian word order. Indo-Iranian Journal 31:27–40. – Cited on p. 47, 64, 65, 66 Harris, Alice, and Lyle Campbell 1995 Historical syntax in cross-linguistic perspective. Cambridge: Cambridge University Press. – Cited on p. 2, 6, 29, 34, 35, 51, 87, 193, 317, 327 Haspelmath, Martin 1993 More on the typology of inchoative/causative verb alternations. In Causatives and transitivity, ed. Bernard Comrie and Maria Polinsky, 87–120. Amsterdam: Benjamins. – Cited on p. 121 1999 External possession in a European areal perspective. In External possession, ed. Doris Payne and Immanuel Barshi, 109–135. Amsterdam: Benjamins. – Cited on p. 61, 62 2002 Explaining the ditransitive person/role constraint. A usage-based account. Ms. available at the author’s website. – Cited on p. 74 Hawar Vol. 1. Hawar. Reprint of the journal, published in 1998 by Nûdem, Stockholm. Vol. 1 contains the numbers 1–23 (1932–1933). – Cited on p. 208 Vol. 2. Hawar. Reprint of the journal, published in 1998 by Nûdem, Stockholm. Vol. 2 contains the numbers 24–57 (1934–1943). – Cited on p. 209, 210 Heine, Bernd, and Tania Kuteva 2005 Language contact and grammatical change. Cambridge: Cambridge University Press. – Cited on p. 317 Heny, J.L 1984 Enclitics in Pahlavi and Early Classical Persian: A theoretical analysis. In Middle Iranian studies. Proceedings of the International Symposium at the Katholieke Universiteit Leuven, 17–20.05.1982, ed. Wojciech Skalmowski and Alois van Tongerloo, 83–94. Leuven: Peeters. – Cited on p. 124 Heston, Wilma Louise 1976 Selected problems in fifth to tenth century Iranian syntax. University of Pennsylvania: Unpublished PhD Thesis. – Cited on p. 30, 70, 89, 91, 106, 107, 348 References 108, 109, 112, 115, 119, 120, 123, 124, 126, 127, 243, 244, 308, 335, 336, 337 Hettrich, Heinrich 1990 Der Agens in passivischen Sätzen altindogermanischer Sprachen. Göttingen: Vandenhoeck and Ruprecht [Nachrichten der Akademie der Wissenschaften in Göttingen aus dem Jahre 1990, Philologisch-Historische Klasse, pp.55–108]. – Cited on p. 71, 72, 78, 86 Hewson, John, and Vit Bubenik 2006 From case to adposition. The development of configurational syntax in IndoEuropean languages. Amsterdam: Benjamins. – Cited on p. 320 Hole, Daniel 2005 Reconciling “possessor” datives and “beneficiary” datives. Towards a unified account of dative binding in German. In Event arguments: Foundations and applications, ed. Claudia Maienborn and Angelika Wöllstein, 213–142. Tübingen: Niemeyer. – Cited on p. 63 Hopper, Paul, and Sandra Thompson 1980 Transitivity in grammar and discourse. Language 56:251–299. – Cited on p. 11, 239 Jahani, Carina 2003 The case system in Iranian Balochi in a contact linguistic perspective. In The Balochi and their neighbours. Ethnic and linguistic contact in Balochistan in historical and modern times, ed. Carina Jahani and Agnes Korn, 113–132. Wiesbaden: Reichert. – Cited on p. 96 Jastrow, Otto 1997 The Neo-Aramaic languages. In The Semitic languages, ed. Robert Hetzron, 334–377. London: Routledge. – Cited on p. 318 Johanson, Lars 2002 Structural factors in Turkic language contacts. Richmond: Curzon. – Cited on p. 319 Justi, Ferdinand 1880 Kurdische Grammatik. St. Petersburg [quoted from the reprint: Vaduz 1976]. – Cited on p. 204 Kahn, M 1976 Borrowing and regional variation in a phonological description of Kurdish. Ann Arbor, Michigan: Phonetics Laboratory of the University of Michigan. – Cited on p. 204, 224 Kazenin, Konstantin 2001 The passive voice. In Language typology and language universals, Vol. 2, ed. Martin Haspelmath, Ekkehard König, Wolfgang Raible, and Wulf Oesterreicher, 899–904. Berlin: De Gruyter. – Cited on p. 52 Keenan, Edward 1976 Towards a universal definition of “subject”. In Subject and topic, ed. Charles Li, 303–333. New York: Academic. – Cited on p. 32 References 349 Kent, Roland 1953 Old Persian. New Haven, Connecticut: American Oriental Society [second, revised edition]. – Cited on p. 24, 25, 26, 27, 28, 30, 31, 44, 47, 48, 52, 53, 55, 56, 57, 58, 62, 63, 64, 65, 66, 67, 68, 69, 73, 77, 78, 84, 93, 95, 144, 334 Kieffer, Charles 2003 Grammaire de l’ōrmur.ı̄ de baraki-barak (Lōgar, Afghanistan). Wiesbaden: Reichert. – Cited on p. 14 Kishimoto, Hideki 2004 Transitivity of ergative-case marking predicates in Japanese. Studies in Language 28:105–136. – Cited on p. 19, 263 Klaiman, M. H 1987 Mechanisms of ergativity in South Asia. Lingua 71:61–102. – Cited on p. 2, 9 König, Ekkehard 2001 Internal and external possession. In Language typology and language universals, Vol. 2, ed. Martin Haspelmath, Ekkehard König, Wolfgang Raible, and Wulf Oesterreicher, 970–978. Berlin: De Gruyter. – Cited on p. 75 Korn, Agnes 2003 Balochi and the concept of Northwestern Iranian. In The Baloch and their neighbours. Ethnic and linguistic contact in Balochistan in historical and modern times, ed. Carina Jahani and Agnes Korn, 49–60. Wiesbaden: Reichert. – Cited on p. 4 forthc. a “Das Nominalsystem des Balochi, mitteliranisch betrachtet.” Manuscript, to appear in Festschrift für Gert Klingenschmitt, edited by Günther Schweiger. – Cited on p. 96, 98, 167 forthc. b The ergative system in Balochi from a typological perspective. In Studies in the typology of Iranian languages, ed. Behrooz Mahmoodi Bakhtiari. München: Lincom. – Cited on p. 251 Kroch, Anthony 2001 Syntactic change. In The handbook of contemporary syntactic theory, ed. Mark Baltin and Chris Collins, 699–729. Oxford: Blackwell. – Cited on p. 22, 324 Kroeger, Paul 2004 Analyzing syntax. a lexical-functional approach. Cambridge: Cambridge University Press. – Cited on p. 37, 215 Kurdistan. 1898–1902. Kurdistan. Rojnama kurdî ya pêşîn (‘The first Kurdish journal’). [Reprinted with transliteration, Turkish translation and editing by M. Emîn Bozarslan. Uppsala: Weşanxana Deng, 1991]. – Cited on p. 306 Kurdoev, K. K 1957 Grammatika kurdskogo jazyka (kurmandži). Fonetika. Morfologija. Moskva: Akademii Nauk. – Cited on p. 204 1978 Grammatika kurdskogo jazyka na materiale dialektov kurmandži i sorani. Moskva: Akademii Nauk. – Cited on p. 235 350 References Kurzová, Helena 1999 Syntax in the Indo-European morphosyntactic type. In Language change and typological variation: In Honor of Winfred P. Lehmann on the occasion of his 83rd birthday. Vol. II: Grammatical universals and typology, ed. Carol Justus and Edgar Polomé, 501–520. Washington: Institute for the Study of Man. – Cited on p. 320 Lambrecht, Knud 1994 Information structure and sentence form. Cambridge: Cambridge University Press. – Cited on p. 50, 236 Lavine, James 1999 Subject properties and ergativity in north Russian and Lithuanian. In Linguistic convergence and areal diffusion. Case studies from Iranian, Semitic and Turkic, ed. Katarzyna Dziwirek, Herbert Coats, and Cynthia Vakareliyska, 307–328. Michigan Slavic Publications: Ann Arbor. – Cited on p. 323 Lazard, Gilbert 1963 La langue des plus anciens monuments de la prose persane. Paris: Klincksieck. – Cited on p. 91, 243, 315, 336 1984 Deux questions de linguistique iranienne: La construction passive du parfait transitif. La versification du moyen-iranien occidental. In E. Benveniste aujourd’hui. Actes du Colloque international du C.N.R.S, Université François Rabelais, Tours, 28–30.09.1983, ed. Jean Taillardat, Gilbert Lazard, and Guy Serbat, 239–248. Paris: Peeters. – Cited on p. 44, 77, 263 1998 Actancy. Berlin: Mouton. – Cited on p. 6, 7 1999 Review of Dorleijn 1996. Linguistic Typology 3:369–373. – Cited on p. 14 Le Coq, Albert 1903 Kurdische Texte. Kurmānǧí-Erzählungen und -Lieder. [quoted from the undated reprint, Amsterdam: Philo Press.]. – Cited on p. 184, 204, 221, 232, 307 Lecoq, Pierre 2002 Recherches sur les dialectes kermaniens (Iran central). Grammaire, textes, traductions et glossaires. Leuven: Peeters. – Cited on p. 137, 140, 142 Leech, Geoffrey, Paul Rayson, and Andrew Wilson 2001 Word frequencies in written and spoken English: based on the British National Corpus. London: Longman. – Cited on p. 181 Lehmann, Christian, Yong-Min Shin, and Elisabeth Verhoeven 2004 Person prominence and relational prominence. On the typology of syntactic relations with particular reference to Yucatec Maya [Second revised version]. Arbeitspapiere des Seminars für Sprachwissenschaft der Universität Erfurt No. 12, available online from the Institute’s website. – Cited on p. 59, 60 Lescot, Roger 1940 Textes kurdes. Première partie: Contes, proverbes et énigmes. Paris: Libraire Orientaliste Paul Geuthner. – Cited on p. 212, 216, 220, 224, 226, 229, 251 References 351 Li, Charles, and Sandra Thompson 1977 A mechanism for the development of copula morphemes. In Mechanisms of syntactic change, ed. Charles Li, 419–444. Texas: University of Texas. – Cited on p. 270 Lightfoot, David 1999 The development of language. Acquisition, change and evolution. Blackwell: Malden, Mass. – Cited on p. 20, 175, 312, 324, 325, 327 Luraghi, Silvia 2003 On the meaning of prepositions and cases. The expression of semantic roles in Ancient Greek. Amsterdam: Benjamins. – Cited on p. 71 MacKenzie, D. N. 1961a Kurdish dialect studies, Vol. I. London: Oxford University Press. – Cited on p. 203, 204, 230, 233, 257, 258, 259, 260, 261, 264, 267, 269, 270, 271, 272, 277, 279, 280, 282, 283, 284, 285, 287, 288, 290, 292, 293, 294, 295, 297, 299, 302, 306 1961b The origins of Kurdish. Transactions of the Philological Society 68–86. – Cited on p. 201, 202, 208 1962 Kurdish dialect studies, Vol. II. London: Oxford University Press. – Cited on p. 219, 233, 257, 258, 259, 260, 263, 264, 266, 267, 271, 272, 279, 281, 282, 288, 289 1995 The Kurdish of Mullā Sat’ı̄d Šamdı̄nānı̄. The Journal of Kurdish Studies I:1– 29. – Cited on p. 233 1966 The dialect of Awroman (Hawrāmānı̄ luhōn). København: Det Kongelige Danske Videnskabernes Selskab. – Cited on p. 10, 143, 145, 146, 147, 150, 151, 184, 185 1999a The ‘Indirect Affectee’ in Pahlavi. In Iranica Diversa, Vol. I, 3–6. Roma: Instituto Italiano per L’Africa e l’Oriente. [Orig. in: Dr. J.M. Unvala memorial volume, Bombay, 1964. Page numbers quoted refer to reprint]. – Cited on p. 126 1999b Kerdir’s inscription. In Iranica Diversa, Vol. I, 217–273. Roma: Instituto Italiano per L’Africa e l’Oriente. [Orig. in: G. Herrmann (ed.) The Sasanian rock reliefs at Naqsh-i Rustam. Naqsh-i Rustam 6. Iranische Denkmäler, Reihe II: Iranische Felsreliefs, I, Berlin 1989, 35–72; page numbers quoted refer to reprint]. – Cited on p. 114 1999c Mani’s šāBUHRAGāN. In Iranica Diversa, Vol. I, 83–117. Roma: Instituto Italiano per L’Africa e l’Oriente. [Orig. in BSOAS 42(3) (1979), pp. 500–534; page numbers quoted refer to reprint]. – Cited on p. 91, 99, 124 Makas, Hugo. 1897–1926 [1979]. Kurdische Texte. St. Petersburg/Leningrad: [Reprinted as the first part of a book entitled Kurdische Texte und Kurdische Studien. (1979). Amsterdam: Philo Press. Page no’s as in original]. – Cited on p. 204, 232 352 References Mann, Oskar 1906 Kurdisch-persische Forschungen, Abteilung IV, Band III, Teil I: Die Mundart der Mukri-Kurden. Grammatische Skizze, Texte in phonetischer und persischer Umschrift. Berlin: Reimer. – Cited on p. 242 Manning, Christopher 1996 Ergativity. Argument structure and grammatical relations. Stanford: CSLI. – Cited on p. 6, 8, 36, 38, 39, 87 Matras, Yaron 1989 Probleme der Sprachstandardisierung am Beispiel der Orthographie des Kurdischen. Unpublished MA-Thesis, Univ. of Hamburg. – Cited on p. 203 1992/1993 Ergativity in Kurmanji (Kurdish). Notes on its use and distribution. Orientalia Suecana 41–42:139–154. – Cited on p. 11, 215, 216 1997 Clause combining, ergativity, and coreferent deletion in Kurmanji. Studies in Language 21:613–653. – Cited on p. 215, 217, 218, 225, 227, 239, 246, 247, 248, 251 2002 Kurmanjî complementation: Semantic-typological aspects in an areal perspective. Sprachtypologie und Universalienforschung / Language typology and universals 55:49–63. – Cited on p. 318 Mayrhofer, Manfred 1989 Vorgeschichte der iranischen Sprachen; Uriranisch. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 4–24. Wiesbaden: Reichert. – Cited on p. 4 Metê, Hesenê 1998 Epîlog. Stockholm: Nûdem. – Cited on p. 238 Mithun, Marianne 2004 “How stable are grammatical relations?” Paper held at LENCA 2, 11– 14.05.2004, State University of Kazan, Tataristan. – Cited on p. 50 Mithun, Marianne, and Wallace Chafe 1999 What are S, A and O? Studies in Language 23:579–606. – Cited on p. 7 Mohanan, Tara 1994 Argument structure in Hindi. Stanford: CSLI. – Cited on p. 330 Moore, John, and David Perlmutter 2000 What does it take to be a dative subject? Natural language and linguistic theory 18:373–416. – Cited on p. 19 Mosel, Ulrike 1987 Subject in Samoan. In A world of language: papers presented to Professor S. A. Wurm on his 65th birthday, ed. Werner Winter and Donald Laycock, 455–479. Canberra: Pacific Linguistics. – Cited on p. 40 1991 Reflexivity and transitivity in Samoan. Australian Journal of Linguistics 11:175–194. – Cited on p. 40 Mosel, Ulrike, and Even Hovdhaugen 1992 Samoan reference grammar. Oslo: Scandinavian University Press. – Cited on p. 39, 40 References 353 Næss, Åshild 2007 Prototypical transitivity. Amsterdam: Benjamins. – Cited on p. 12, 60, 177 Nedjalkov, Vladimir 2001 Resultative constructions. In Language typology and language universals, Vol. 2, ed. Martin Haspelmath, Ekkehard König, Wolfgang Raible, and Wulf Oesterreicher, 928–940. Berlin: De Gruyter. – Cited on p. 41 Newmeyer, Frederick 2003 Grammar is grammar and usage is usage. Language 79:682–707. – Cited on p. 229 2005 Possible and probable languages. A generative perspective on linguistic typology. Oxford: Oxford University Press. – Cited on p. 114, 328 Nichols, Johanna 1992 Linguistic diversity in space and time. Chicago: University of Chicago Press. – Cited on p. 6, 292, 293, 318 Nyberg, Henrik 1964 A manual of Pahlavi, Part I: Texts. Wiesbaden: Harrassowitz. – Cited on p. 124 1974 A manual of Pahlavi, Part II: Glossary etc. Wiesbaden: Harrassowitz. – Cited on p. 97, 99 Palancar, Enrique 2002 The origin of agent markers. Berlin: Akademie Verlag. – Cited on p. 12, 70, 79, 82 Paul, Ludwig 1998a The position of Zazaki among West Iranian languages. In Proceedings of the Third European Conference of Iranian Studies. Part 1: Old and Middle Iranian studies, ed. Nicholas Sims-Williams, 163–177. Wiesbaden: Reichert. – Cited on p. 4 1998b Zazaki. Grammatik und Versuch einer Dialektologie. Wiesbaden: Reichert. – Cited on p. 7, 140, 150, 210, 315 2002 Grammatical and philological studies on the Early Judaeo-Persian texts from the Cairo Geniza. Universität Goettingen. – Cited on p. 91, 120, 121, 207 2003a Early Judaeo-Persian in a historical perspective: the case of the prepositions be, u, pa(d), and the suffix rā. In Persian origins – Early Judaeo-Persian and the emergence of New Persian, ed. Ludwig Paul, 178–194. Wiesbaden: Harrassowitz. – Cited on p. 308 2003b The position of Balochi among Western Iranian languages: The verbal system. In The Baloch and their neighbours. Ethnic and linguistic contact in Balochistan in historical and modern times, ed. Carina Jahani and Agnes Korn, 61–71. Wiesbaden: Reichert. – Cited on p. 4 Payne, Doris, and Immanuel Barshi, ed. 1999a External possession. Amsterdam: Benjamins. – Cited on p. 58 354 References Payne, Doris, and Immanuel Barshi 1999b External possession. what, where, how, and why? In External possession, ed. Doris Payne and Immanuel Barshi, 3–29. Amsterdam: Benjamins. – Cited on p. 75 Payne, John 1980 The decay of ergativity in Pamir languages. Lingua 51:147–186. – Cited on p. 14, 29, 117, 137, 195, 196 1998 Ergative construction. In Encyclopædia Iranica, Vol VIII, 555–558. Costa Mesa: Mazda. – Cited on p. 29, 35 Payne, Thomas 1997 Describing morphosyntax. A guide for field linguists. Cambridge: Cambridge University Press. – Cited on p. 49, 173, 180, 181 Peterson, John 1998 Grammatical relations in Pāli and the emergence of ergativity in Indo-Aryan. München: Lincom. – Cited on p. 2 Phillott, D. C 1919 Higher Persian grammar. Calcutta: Babtist Mission Press. – Cited on p. 120 Pirejko, L.A 1963 Ergativnaja konstrukcija v kurdskom jazyke. In Iranskij sbornik: k semidesjatipjatiletiju Prof. I.I. Zarubina., ed. V.I. Abaev, 150–158. Moskva: Akademii Nauk. – Cited on p. 242 Ritter, Helmut 1971 Kurmânci-Texte aus dem t.ûr’abdîn. Oriens 21-22:1–135. – Cited on p. 204, 251 Ritter, Helmut 1976 Kurmânci-Texte aus dem t.ûr’abdîn, Teil II. Oriens 25-26:1–37. – Cited on p. 204 Rizgar, Baran 1996 Learn Kurdish. A multi-level course in Kurmanji. London: Lithosphere. – Cited on p. 204, 222, 234 Schmeja, H 1982 ‘Ging ein zum Throne der Götter’. In Monumentum George Morgenstierne II, ed. J. Duchesne-Guillemin, 185–188. Leiden: Brill. – Cited on p. 68 Schmitt, Rüdiger 1989 Mitteliranische Sprachen im Überblick. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 95–105. Wiesbaden: Reichert. – Cited on p. 90 1990 Epigraphisch-exegetische Noten zu Dareios’ Bı̄sutūn-Inschriften. Wien: Österreichische Akademie der Wissenschaften. – Cited on p. 24, 57 1999 Beiträge zu altpersischen Inschriften. Wiesbaden: Reichert. – Cited on p. 24, 26, 334 2000 Die iranischen Sprachen in Geschichte und Gegenwart. Wiesbaden: Reichert. – Cited on p. 4, 24 Schulze, Wolfgang 2000 Northern Talysh. München: Lincom. – Cited on p. 138, 141, 148, 149, 177 References 355 Sedighi, Anousha 2005 Subject-predicate agreement restrictions in Persian. Unpublished PhD thesis: University of Ottawa. – Cited on p. 108 Shibatani, Masayoshi 1995 A. A. Xolodovič on Japanese passives. In Subject, voice and ergativity. Selected essays, ed. David Bennett, Theodora Bynon, and B. George Hewitt, 7–19. London: School of Oriental and African Studies. – Cited on p. 58 2001 Non-canonical constructions in Japanese. In Non-canonical marking of subjects and objects, ed. Alexandra Aikhenvald, R.M.W. Dixon, and Masayuki Onishi, 307–354. Amsterdam: Benjamins. – Cited on p. 19, 262, 263 2002 “The non-case for ‘Experiencer’ and ‘Possessor’.” Paper held at the Syntactic Roles Conference, leipzig, 05–07.12.2002. – Cited on p. 261 2004 Voice. In Morphologie. Ein internationales Handbuch zur Flexion und Wortbildung, ed. Geert Booij, Christian Lehmann, and Joachim Mugdan et al., 1145–1165. Berlin: De Gruyter. – Cited on p. 37 Shojai, Behrooz 2005 “The Kurdish of Gilan.” Unpublished manuscript, based on an MA-thesis at the Department of Asian and African Languages, Uppsala University. – Cited on p. 204, 221, 253, 254 Siewierska, Anna 2004 Person. Cambridge: Cambridge University Press. – Cited on p. 6, 135 Siewierska, Anna, and Dik Bakker 2004 “Three takes on grammatical relations: a view from the languages of Europe and North and Central Asia.” Paper held at LENCA 2, 11–14.05.2004, State University of Kazan, Tataristan. – Cited on p. 222 SigurDsson, Halldór 2002 To be an oblique subject: Russian vs. Icelandic. Natural language and linguistic theory 20:691–724. – Cited on p. 19 2004 Icelandic non-nominative subjects: facts and implications. In Non-nominative subjects, ed. Peri Bhaskararao and K. V. Subbarao, volume 2, 137–159. Amsterdam: Benjamins. – Cited on p. 19, 20 Silverstein, Michael 1976 Hierarchy of features and ergativity. In Grammatical categories in Australian languages, ed. R.M.W. Dixon, 112–171. Canberra: Australian Institute of Aboriginal Studies. – Cited on p. 180 Sims-Williams, Nicholas 1981 Notes on Manichaean Middle Persian morphology. Studia Iranica 10:165– 176. – Cited on p. 90, 97, 106 1989 Sogdian. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 173– 192. Wiesbaden: Reichert. – Cited on p. 42 1998 The Iranian languages. In The Indo-European languages, ed. Anna Giacolone Ramat and Paolo Ramat, 125–153. London: Routledge. – Cited on p. 4 356 References Skjærvø, Prods Octor 1985 Remarks on the Old Persian verbal system. Münchner Studien zur Sprachwissenschaft 45:211–227. – Cited on p. 29, 30, 31, 42, 43, 44, 78 1983 Case in Inscriptional Middle Persian, Inscriptional Parthian, and the Pahlavi Psalter. Studia Iranica 12:47–62, 151–181. – Cited on p. 96, 97, 99 2003 “An introduction to Young Avestan.” manuscript available online at: http:// www.fas.harvard.ed/~iranian./avesta/avestancomplete.pdf. – Cited on p. 23, 71 Slobin, Dan, and T.G. Bever 1982 Children use canonical sentence schemas: A crosslinguistic study of word order and inflections. Cognition 12:229–265. – Cited on p. 192 Statha-Halikas, Hariklia 1979 How not to tell a passive: the case of Old Persian manā krtam reconsidered. Berkeley Linguistic Society 5:350–361. – Cited on p. 29, 30, 31 Stiebels, Barbara 2002 Typologie des Argumentlinkings. Ökonomie und Expressivität. Berlin: Akademie. – Cited on p. 6 Stilo, Donald 1981 The Tati language group in the sociolinguistic context of Northwestern Iran. Iranian Studies XIV:137–187. – Cited on p. 160 2004a “Double-duty pronominal clitics in Gazi.” Paper held at the conference Syntax of the World’s languages, Max Planck Institute for Evolutionary Anthropology, Leipzig, August 7, 2004. – Cited on p. 289 2004b Vafsi folk tales. Transcribed, translated and annotated by Donald L. Stilo. Supplied with folklorist notes and edited by Ulrich Marzolph. Wiesbaden: Reichert. – Cited on p. 14, 155, 156, 161 2005 Iranian as buffer zone between the universal typologies of Turkic and Semitic. In Linguistic convergence and areal diffusion. Case studies from Iranian, Semitic and Turkic, ed. E.A. Csato, B. Isaksson, and Carina Jahani, 35–63. London: Curzon. – Cited on p. 70, 317, 318 forthc. a “Gazi.” To appear in the Encyclopædia Iranica. – Cited on p. 161 forthc. b A grammar of Vafsi. Berlin: Mouton. – Cited on p. 11, 161 Sundermann, Werner 1973 Mittelpersische und parthische kosmogenische und Parabeltexte der Manichäer. Berlin: Akademie-Verlag. – Cited on p. 91, 119 1981 Mitteliranische Manichäische Texte kirchengeschichtlichen Inhaltes. Berlin: Akademie-Verlag. – Cited on p. 91, 119 1989a Mittelpersisch. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 138–164. Wiesbaden: Reichert. – Cited on p. 336 1989b Parthisch. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 114– 137. Wiesbaden: Reichert. – Cited on p. 90, 91, 92, 101, 118 1989c Westmitteliranische sprachen. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 106–113. Wiesbaden: Reichert. – Cited on p. 90, 93 References 357 Sundquist, John 2006 Syntactic variation in the history of Norwegian and the decline of XV word order. Diachronica 23:105–141. – Cited on p. 175 Tafazzoli, Ahmad 1986 The ‘Indirect Affectee’ in Pahlavi and in a Central Dialect of Iran. In Studia grammatica iranica. Festschrift fr̈ Hubert Humbach, ed. Rüdiger Schmitt and Prods Oktor Skærvø, 483–487. München: Kitzinger. – Cited on p. 126 Thordarson, Fridrik 1989 Ossetic. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 456– 479. Wiesbaden: Reichert. – Cited on p. 1, 315, 319 Tomasello, Michael 2003 Constructing a language. a usage-based theory of language acquisition. Cambridge, Mass.: Harvard University Press. – Cited on p. 181, 229, 327, 328 Trask, Robert Lawrence 1979 On the origins of ergativity. In Ergativity. Towards a theory of grammatical relations, ed. Frans Plank, 385–404. London: Academic. – Cited on p. 30 1996 Historical linguistics. London: Arnold. – Cited on p. 270 Unger, Christoph 1994 “Aspects of the dialect of Behdinan.” Unpublished paper presented at the Kurdistan Forum, SOAS, London 05.05.1994. – Cited on p. 270 Utas, Bo 1974 Utas, Bo 1976 Verbal forms and ideograms in the Middle Persian inscriptions. Acta Orientalia 36:83–112. – Cited on p. 119 Verbs and preverbs in the Ayyātkār ı̄ Zarērān. Acta Orientalia 37:75–110. – Cited on p. 125 Van Langendonck, Willy 1998 The dative in Latin and the indirect object in Dutch. In The dative, Vol. II: Theoretical and contrastive studies, ed. Willy Van Langendonck and William Van Belle, 211–259. Amsterdam: Benjamins. – Cited on p. 72 Van Valin, Robert 2001 An introduction to syntax. Cambridge University Press: Cambridge. – Cited on p. 59, 215 Vincent, Nigel 2001 LFG as a model of syntactic change. In Time over matter. Diachronic perspectives on morphosyntax, ed. Miriam Butt and Tracy Holloway King, 1–42. Stanford: CSLI. – Cited on p. 336, 337 Watters, David 2002 A grammar of Kham. Cambridge: Cambridge University Press. – Cited on p. 246 358 References Williams, A. V. 1990a The Pahlavi Rivāyat accompanying the Dādestān ı̄ Dēnı̄g. Part I: Transliteration, transcription and glossary. København: Munksgaard. [=Historiskfilosofiske Meddelelser 60:1, Det Kongelige Danske Videnskabernes Selskab]. – Cited on p. 91, 95, 100, 108, 109, 110, 124, 308 1990b The Pahlavi Rivāyat accompanying the Dādestān ı̄ Dēnı̄g. Part II: Translation, commentary and Pahlavi text. København: Munksgaard. [=Historiskfilosofiske Meddelelser 60:2, Det Kongelige Danske Videnskabernes Selskab]. – Cited on p. 91, 110 Windfuhr, Gernot 1989b Western Iranian dialects. In Compendium Linguarum Iranicarum, ed. Rüdiger Schmitt, 294–295. Wiesbaden: Reichert. – Cited on p. 202 1992 Case. In Encyclopædia Iranica, Vol V, 25–37. California: Mazda. – Cited on p. 92, 133 Wurzel, Petra 1997 Rojbaş. Einführung in die kurdische Sprache. Wiesbaden: Reichert. – Cited on p. 204, 212, 235 Yar-Shater, Ehsan 1969 A grammar of Southern Tati dialects. The Hague: Mouton. – Cited on p. 98, 149, 150, 154, 159, 160, 161, 162, 167, 168, 169, 190, 191 Yarshater, Ehsan 1959 The dialect of Shāhrud (Khalkhāl). Bulletin of the School of Oriental and African Studies 22:52–68. – Cited on p. 140, 159, 164 1960 The tāti dialect of Kajal. Bulletin of the School of Oriental and African Studies 23:275–286. – Cited on p. 140, 159, 160, 162, 307 1970 The Tati dialects of T.ārom. In W. B. Henning Memorial Volume, ed. Mary Boyce and Ilya Gershevitch, 451–467. London: Lund Humphries. – Cited on p. 30, 160, 169, 170, 171, 307 2003 The Xo’ini dialect. In Persica. Annual of the Dutch-Iranian Society, XIX, ed. G.R. et al. van den Berg, 165–182. Leuve: Peeters. – Cited on p. 141, 159, 161, 162, 164 Zinar, Zeynelabidîn 1988 Siyabend û Xecê. Stockholm: Pencînar Weşanxaneya Çanda Kurdî. – Cited on p. 221, 226, 232 Zwicky, Arnold 1977 On clitics. Bloomington: Indiana University Linguistics Club. – Cited on p. 46 Subject index A 6–8, 11, 12, 25, 32, 34, 38–41, 49–54, 80–83, 92–96, 105, 118, 122, 123, 125–127, 131, 173–178, 181–183, 187, 197–199, 214–216, 222–224, 245, 254, 271–274, 278, 286, 301, 314, 318, 329 A-past 97, 105–107, 111, 112, 113, 115–117, 147, 148, 157–160, 164–169, 171, 183, 190, 193–198, 206, 214–216, 222–224, 226–239, 232–240, 255, 256, 263–268, 274, 278, 280, 283, 288–302, 305, 335, 337 A-pres 97, 105, 128, 157–159, 164–169, 281 Ablative 46, 47, 78, 94, 106, 321, 334 Absolutive 40, 47, 78, 94 Accusative (alignment) 1, 2, 6, 7, 9, 10, 12–15, 21, 25, 34, 85, 86, 117, 127, 129, 132, 133, 148, 158, 161, 168, 174, 182, 184, 185–188, 194, 195, 196, 201, 213, 223, 228, 253–255, 274, 277, 278, 300–304, 307, 312, 313, 315, 317, 319, 320, 322, 327–329 Accusative (case) 7, 13, 25, 35, 42, 46–48, 54, 65, 69, 85, 94, 96, 102, 104, 106, 112, 118, 128, 132, 149, 161, 182, 254, 300, 312, 316, 323 Achaemenian Dynasty 24 Active (alignment) 6, 7, 320–322 Active (voice) 24, 28, 31, 32, 35, 37, 40, 41–44, 49–51, 53, 55, 73, 89, 118, 120, 123, 222, 223, 265–268, 312 Addressee 62, 69, 78, 135 Adjunct 108, 338 Adposition 45, 74, 77, 96, 101, 105, 111, 116, 152, 155, 205–207, 222, 275, 318, 337, 346 Adversative 63 Affectedness 12, 58–61, 63 Agented passive 3, 31, 32, 34, 35, 43, 45, 51, 73, 75, 82, 83, 87, 275, 312 Agentless passive 31, 118, 119, 121–123, 128, 264, 266, 267, 275 Agglutinative 145, 146, 150, 151, 175, 205 Agreement 1, 3, 6, 13–15, 21, 25, 26, 34, 35, 80, 92, 106, 114, 116, 117, 123–128, 131, 178, 187, 192, 201, 204, 211, 212–215, 223–226, 231–256, 263, 268, 274, 279–283, 289–305, 307, 314, 323, 327, 330, 337 Agreement, A-dominated 231, 234, 235, 236, 238–240, 242, 245, 256, 268, 274 Alignment 6, see Active (alignment); Accusative (alignment); Ergative (alignment); Hierarchical alignment; Tripartite alignment Anaphora 50–53, 225, 245, 264, 301, see also Zero anaphora Animacy 38, 41, 49–51, 70, 73–75, 360 Subject index 82, 87, 103, 173, 179–186, 188, 189–192, 198, 239, 292, 298, 312, 313, 320 Animacy-based alignment split 103, 179–181, 182–186, 189, 320 Animacy hierarchy 180, 183, 184, 188, 198, 331 Aorist 24, 42, 43, 54, 84, 311 Aramaic script 90 Areal 4, 70, 86, 317–319 Argument (syntactic) 6, 16–19, 21, 37, 39-42, 45, 54, 75, 76, 96, 108, 121, 127, 178, 179, 211, 218, 227, 229, 239, 282, 290, 292, 295, 305, 313, 314, see also Core argument Aspect 9, 29, 34, 43, 192, 212, 269, 270, 272, 291 Auxiliary 85, 92, 93, 119, 123–125, 272 Benefactive 55–62, 65, 68, 70, 72, 76, 86, 109, 111, 126, 286 Binding 218 Canonical sentence 192, 193 Case marking 1, 6, 11, 12, 20, 25, 35, 60, 96, 97, 101, 104, 131–133, 143, 158, 171, 172–179, 183, 187, 192–195, 197, 201, 206, 214, 223, 224–228, 255, 274, 278, 281, 300–305, 315, 320, 329, 333, see also Ablative; Absolutive; Accusative (case); Dative; Genitive; Instrumental Case, discriminatory function of 132, 172–179, 192, 194, 197–199, 227, 255, 273 Case, semantic 45, 47, 49, 207 Case, structural 13, 54, 86 Caucasus 202, 203, 257 Causative 60, 71 Caused Motion Construction 17 Central Asia 89, 91, 353 Circumposition 207 Clause linkage 126, 245–253, 278, 318, see also Subordinate clause Cliticization 27, 47–49 (in Old Persian), 66, 69, 294 Clitic placement 47, 107, 111, 116, 285–289, 334–338 Clitic pronoun 15, 26–28, 46, 62, 66, 69, 89, 106–108 (in Western Middle Iranian), 110, 112, 115, 122, 161, 211, 278, 283, 286, 287, 289 (and agreement), 290, 297, 300, 301, 305, 308, 335–337 Comitative 207 Complementarity Principle 193 Complex predicate 11 Construction Grammar 15–22, 121, 262, 324, 342, 344 Control of reflexives 6, 34, 218–222, 262, 273 Coordinate clause 52–54, 218, 259, 302, 303 Copula 26, 31, 34, 42, 43, 67, 82, 85, 92, 120, 125, 212, 258, 270, 281, 311, 349 Core argument 6, 13, 37–39, 41, 45, 96, 105, 112, 173, 178, 186, 197, 223, 225–229, 290 Coreferential deletion 34, 39, 52–54, 122, 218, 223, 259, 261, 268 Cross-system harmony 3, 172, 192–198, 228, 230, 274, 317 Cyrillic 5 Dative 18, 35, 45, 49, 56–59 (in Old Iranian), 61–64 (with EPCs), 71, 72, 77, 82, 84, 87, 94, 99, Subject index 361 102, 106–112, 128, 144, 311, 312, 314, 333 Dative continuum 72 Dative Subject 19, see also Non-canonical subject Declensional class 144, 334 Definiteness 145, 151, 181, 190, 205, 209, 278, 279 Demonstrative 74, 101, 102, 135, 147, 153, 205–207, 270, 279, 333, 334 Derivational approaches to syntax 20, 21 Desire (expressions of) 9, 20, 261, 262, 268, 274, 305–309, 311, 315, 316, 321 Deverbal adjective 34 Differential Object Marking (DOM) 2, 14, 132, 157, 158, 159, 161, 179, 185, 189 Direct (case) 6, 13, 90, 96–102, 104, 113, 116, 132–134, 136–138 (with plural), 139–143 (on SAP pronouns), 154–172, 177, 183, 185, 187, 190, 193, 195, 197, 206, 208, 210, 214, 225, 227–230, 238, 248, 254, 260, 278, 313 Discordant agreement 245, 248–249, 252 Distance Principle 236, 239, 244 Double Oblique 165, 167, 177–179, 187, 191, 194, 197, 225–230, 254, 268, 273, 304, 329 Drift 299, 313, 324–326 Equi-NP deletion 6 Ergative (alignment) 2, 6–11, 13, 21, 26, 30, 34–36, 51, 79, 84, 86, 87, 113, 120, 122, 128, 136, 148, 166, 174, 178, 182–191, 223, 224, 227, 256, 257, 273, 274, 277, 278, 301–306, 314, 316, 320, 328, 329 Ergative (case) 11–13, 79, 182, 183 Ergative (construction, canonical) 213–224, 227, 235, 255, 256, 265, 313 Ergativity Continuum 187–189, 194, 198, 303–305, 331 Ergativity, morphological 8 Ergativity, secondary 269, 272, 273 Ergativity, syntactic 8, 32, 223 Exceptional Case Marking (ECM) 21 Experiencer 18, 57–63, 69, 86, 108, 258, 259–262, 313 External possessor (construction, EPC) 58, 61–65, 69, 70, 73–76, 81–85, 258, 259, 287, 294, 315, First language acquisition 324, 327 Fluid-S 7, 320 Gender 93, 99, 136, 144, 150, 151, 161, 171, 201, 205, 206, 209, 254, 269, 278, 279, 323, 333, 345 Genitive (case) 26, 29, 35, 45, 46–49 (Old Persian), 54, 56–79 (semantics of), 82, 94, 106, 112, 128, 144, 192, 308, 311, 314, 334 Genitive, adnominal 55, 66, 69 Genitive, clausal 63, 67–69, 72, 294 Gerundivum 71 Goal 13, 173, 205–207, 278, 287 Grammaticalization 145, 170, 187, 207, 300, 337 Head attraction 107, 334, 336, 337 Hierarchical alignment 292 Imperfect 10, 24, 43, 84, 94 362 Subject index Inclusive/Exclusive 139 Indefiniteness 144, 145, 150, 168, 173, 206, 254 Indirect object 6, 45, 48, 58, 66, 69, 86, 205, 278 Indirect Participation 58–66, 72, 74, 81, 86, 105, 107–117, 126–128, 283–287, 293–299, 305, 308, 335, 337, 338 Inherited case 96, 132, 135–137, 140–143, 149, 158, 167, 184, 189 Innovated case 96, 104, 127, 134–137, 143, 152–160, 163, 167–172, 177, 179, 184, 190, 196, 228, 254, 308, 315, 319, see also Object marker Instrumental 12, 207, 333 Islam 89, 341 Izafe 69, 70, 145, 146, 205, 206, 208–210, 218, 219, 225, 259, 269–272, 277, 279 Kinship 51, 97, 98, 136, 149, 150 LFG 16, 355 Lingering intransitivity 117, 119–121, 123, 266–268, 275 317, 322 Mainstream Generative Grammar (MGG) 15, 323, 324 Manā kartam (m. k.) construction 26–36, 41–54, 67, 70, 73–87, 312 Manichaeism 91 Mihi est (construction) 28, 64, 321 Necessity (expressions of) 71, 261, 262, 311 Neutral alignment 183 Non-Canonical Subject (NCS) 3, 19–22, 75, 87, 257–263, 268, 272–275, 306, 311–316, 323 Non-finite verb forms 71, 77, 84, 92, 211, 216, 278, 311, 323 O 6–8, 12, 25, 32, 34, 38–40, 49–51, 80 (topicalization of), 95, 113, 118, 123, 127 (agreeement with), 173–182, 184, 187, 194 O-past 92, 112, 147, 157–170, 183, 191, 194–196, 214, 227–232, 240 (saliency of), 255, 263, 273, 280, 289–292, 295–301 O-pres 112, 116, 128, 147, 157, 164, 170, 194, 230, 280, 283–287 Object, direct see O Object marker -rā 7, 101, 104, 128, 152, 156, 168, 308 Obligation, expressions of 9, 71, 262, 274, 305–309 Oblique case 8, 11–13, 35, 95–102, 113, 116, 132, 133–199, 206, 214, 225, 227–230, 248, 254, 259–262 (Possessor), 266, 274, 304, 307, 308, 313–315, 322, 328, 334 Parameter setting 14, 326–330 Participle 10, 26, 28, 30, 41–44, 54, 71, 72, 76, 82, 83, 85, 92–94, 117–123, 127, 128, 211, 266, 270–273, 275, 300, 311–313, 316–318, 322–325 Passive 3, 21, 27–36, 37–45, 47, 49–52, 54, 55, 71, 73, 75–83, 87, 93, 94, 118–123, 128, 222, 223, 243, 264–268, 273, 275, Subject index 363 293, 312, 328, 340–343, 347, 348, 352, 354 Past Transitive Construction (PTC) 12, 14, 15, 21, 84, 86, 89, 92–95 (Middle Iranian), 117–123, 161, 167, 183, 187, 188, 195, 204, 313 Person-based split see Animacy Polysemy 12, 22, 76, 144, 296, 324 Possessor Raising 63, 81 Pragmatics 47, 49, 77, 181, 236 Preferred Argument Structure 229, 343 Preposition 46, 76, 111, 117, 152, 162, 207, 280, 283–286, 293, 335, 336 Projection Principle 16, 17, 76 Pronoun, personal 18, 46–51 (and animacy), 74, 90, 93, 98, 101, 102 (Parthian), 103–122, 135, 152, 159, 161, 162–165, 201, 207, 213, 217, 232, 245–253 (pronoun deletion), 270, 278, 279–283, 288, 333, see also Clitic pronoun; Reflexive pronoun; SAP pronoun Psych-verbs 87 Quirky Subject 19, see Non-canonical subject Reanalysis 35, 36, 86, 87, 270, 272, 312, 327 Recipient 12, 49, 55, 58–62, 74, 86 Reflexive pronoun 6, 72, 218–222, 285, 327 Relative clause 26, 44, 80, 122, 208, 270 Relativization 132, 134, 150, 152, 197, 198 Resultative 26, 34, 41, 42, 76, 85, 117, 121, 271, 311, 316, 317, 323 Revitalization 132, 134, 146, 148–152, 197 Rightward drift 107, 300, 334–337 Saliency Hierarchy of Objects 239 Speech Act Participant (SAP) pronoun 74, 101, 102, 135, 139–144, 146–148, 150–157, 161, 162, 164, 169–171, 180, 182–186, 188–191, 194, 195, 198, 238, 239, 291–294, 315 Subject Construction Hierarchy 302–303, 331 Subject see S; A Subject properties 32–37 (transfer of), 75, 86, 87, 261–268 (syntactic), 313 Subjecthood, graded 20, 32, 33, 75, 83, 87, 88, 215–218 (tests of), 259, 262, 273, 306 Subordinate clause 192, 217, 219, 220, 246–250, 278 Suppletion 141, 172 (emergence of) Surface Generalization Hypothesis 21 Syncretism 92, 95, 131, 225, 334 Tense-Sensitive Alignment (TSA) 8, 9–11, 89, 105, 127, 153, 158, 166–168, 254, 317–320, 323 Tense/Aspect split see Tense-Sensitive Alignment Topic 27, 47, 51, 55, 74, 75, 80, 90, 126, 181, 221, 245, 264, 270, 302, 344, 347 Topicality 41, 49, 51, 73, 74, 82, 83, 87, 181, 236, 245, 259, 270, 312–314 364 Subject index Topicalization 34, 66, 80, 81, 176, 216 Transitivity 11–12 (lexical vs. syntactic), 42, 121, 239, 245, 265, 316, 317, 320–322 (lack of), see also Lingering intransitivity Tripartite alignment 173, 187, 188, 303, 305, 306 Turkey 1, 7, 202–204, 206, 207, 221, 228, 257, 265, 269, 218 Typology 3, 4, 30, 302, 330 Universal Grammar 327, 328 Verb Phrase (VP) 64–66 (in Old Persian), 286–289, 301, 327, 337 Verb serialization 250–252 Voice 37–51, 54, 82, 222, 322 Wackernagel’s Law 46, 285, 300, 336 Word order 8, 65, 80, 81, 90–93, 133, 166, 174–178 (change), 187, 192, 210, 216, 286, 318, 325 Zero-anaphora 50–53, 225, 245, 264, 265 Index of Iranian languages Abyānei, 140, 142, 143, 182–184, 188, 189 Bactrian, 89 Badı̄nānı̄, 10, 119, 202–206, 208, 211, 219, 230, 231, 233, 234, 241–243, 256–275, 306, 314, 321 Balochi, 5, 11, 12, 70, 96, 134, 151, 167, 185, 186, 188, 191, 251, 319 Balochi, Turkmen, 98, 139, 319 Bartangi, 196 Bayray, 176 Central Plateau Languages, 137, 307 Chali, Southern Tati, 163 Choresmian, 89 Dānesfāni, Southern Tati, 163, 191 East Iranian, 4, 7, 114, 117, 137 Ebrāhim ābādi, Southern Tati, 163 Esfarvarini, Southern Tati, 163, 169 Eshtehardi, Southern Tati, 150, 163, 166–168, 171, 190, 191 Gatha Avestan, 4 Gazi, 161, 289 Gorani, 10, 106, 143, 145–152, 184–186, 188, 202, 203, 242, 287, 298, 299 Hawrami, 184–186, 188 Judaeo Persian, 120 Kajal, 140, 159, 160, 162–164, 307 Kandulai, 146, 147, 150, 151, 184 Khotanese, 5, 89, 243 Kurdish, Central Group, 1, 106, 111, 177, 202, 209, 230, 242, 277–306 Kurdish, Northern Group, 13, 98, 118, 125, 136, 139, 144, 194, 197, 201–275, 277, 278, 281, 298, 306, 315, 318, 321, 328 Kurdish, Southern Group, 113, 134, 142, 176, 201–203, 279 Kurmanji, see Kurdish, Northern Group, Lakki, 202 Lāsguerdı̄, 183, 184 Median, 23 Middle Iranian, 30, 42, 70, 80, 81, 83, 84, 87, 89–129 Middle Persian, 25, 70, 89, 90, 93–95, 99, 106–108, 113, 115, 122, 123, 128, 132, 134, 137, 155, 176, 285, 315, 319, 328, 329, 334, 336 Mukri, 202, 242, 288, 297 Old Avestan, 4, 23, 25, 56–58, 74 Old Persian, 3, 4, 23–87, 93–95, 98, 107, 109, 127, 144, 198, 285, 287, 294, 312, 333–334 Ōrmur.ı̄, 14 Ossetic, 1, 5, 315, 319 Pahlavi, 70, 89, 99, 123, 243 Pamir Group, 5, 117, 137, 152, 188, 195, 196 366 Index of Iranian languages Parthian, 5, 89–92, 101, 102, 108, 109, 113, 118, 127, 132, 139, 141 Pashto, 2, 5, 113, 114, 328 Persian, 1, 5, 7, 101, 104, 107–109, 112, 113, 120, 127, 131, 134, 135, 152, 155, 156, 158–160, 176, 188, 198, 253, 254, 287, 300, 301, 308, 315, 319, 337 Piždar, 277, 297 Rošani, 117, 188, 196 Talyshi, 10, 70, 134, 138, 141, 148, 149, 151, 152, 253, 307 T.ārom, 160, 163, 165, 169–171, 188, 195 T.āromi, 163 Tati, 2, 13, 30, 98, 104, 113, 114, 132, 136, 137, 149, 160–172, 190, 191, 307 Vafsi, 8, 11, 14, 151–153, 155, 156, 158, 161–163, 166–168, 171, 184, 187, 188, 190, 191 Sagz-ābādi, Southern Tati, 163 Sangesari, 152–156, 159, 160, 173 Sarmatian, 89 Sarykoli, 137, 141 Scythian, 23 Shāhrud, 140, 159, 164 Shāhrudi, 163 Sogdian, 5, 42, 89, 243 Sourkhéi, 182 Wakhi, 7, 97, 194 West Iranian, 4, 11, 69, 70, 89–91, 131–197, 201, 208, 245, 257, 283, 289, 312, 336 Tākestāni, Southern Tati, 161, 163, 169 Zazaki, 7, 132, 134, 140, 150, 152, 157, 158, 197, 203, 210, 315 Xiāraji, Southern Tati, 163 Xo’ini, 141, 159, 161–164 Xoznini, Southern Tati, 163 Young Avestan, 23, 71 Index of non-Iranian languages Aramaic, 24, 90, 318 Czech, 58 Dyirbal, 38, 39, 103, 182 English, 17, 18, 21, 49, 50, 102, 103, 114, 121, 122, 141, 175, 181–183, 218, 229, 272, 315, 327 French, 183, 229, 235, 321 German, 18, 62, 63, 65, 72, 79, 102, 192, 218, 235, 315 Germanic, 33, 175, 315 Greek, 71 Icelandic, 19, 20, 84, 175, 316, 322 Indo-Aryan, 2, 4, 9, 81, 84, 86, 153, 304, 312, 317, 319, 322 Indo-European, 2–4, 28, 31, 41, 42, 49, 62, 63, 65, 71–73, 77, 78, 82, 84–87, 101, 118, 144, 258, 286, 311, 312, 315, 320–323, 331, 333, 334 Japanese, 19, 88, 262, 263 Kham (Tibeto-Burman, Nepal), 246 Latin, 28, 64, 71, 321, 337 Lithuanian, 84, 323 Norwegian, 175 Old English, 20 Old Irish, 72 Russian, 19, 75, 88, 237, 253, 316, 323 Samoan (Polynesian, Samoa), 39, 40 Southern Tiwa (Tanoan, Mexico), 50 Turkish, 71, 176, 192, 202, 225 Tuvaluan (Polynesian, Tuvalu), 40 Yana (Hokan stock, California), 50

Log In

Alignment change in Iranian languages: A Construction Grammar approach. 2008

Alignment change in Iranian languages: A Construction Grammar approach. 2008

Related Papers

RELATED PAPERS