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