Week 14. Markedness, the Subset Principle, and the Overall Wrap-up GRS LX 700 Language Acquisition and Linguistic Theory.

Presentation on theme: "Week 14. Markedness, the Subset Principle, and the Overall Wrap-up GRS LX 700 Language Acquisition and Linguistic Theory."— Presentation transcript:

1 Week 14. Markedness, the Subset Principle, and the Overall Wrap-up GRS LX 700 Language Acquisition and Linguistic Theory

2 Markedness “Markedness” actually has been used in a couple of different ways, although they share a common core. Marked: More unlikely, in some sense. Unmarked: More likely, in some sense. You have to “mark” something marked; unmarked is what you get if you don’t say anything extra.

3 “Likeliness” Typological/crosslinguistic infrequency. –VOS word order is marked. More complex constructions. –[ts] is more marked than [t]. The non-default setting of a parameter. –Non-null subjects? Language-specific/idiosyncratic features. –Vs. UG/universal features…?

4 Implicational universals SU > DO > IO > OPREP > GEN > OCOMP Any language which can relativize on DO can also relativize on SU. A language which can relativize on SU may or may not be able to relativize on DO. So: SU you get “for free”, saying nothing further. DO you only get if a language is marked. This is basically frequency.

5 Typological facts If the typological generalizations are somehow part of what we know about language (and how “likely” things are), we might expect to find more typologically common things to be easier to acquire— perhaps they will appear first in acquisition.

6 Mazurkewich (1984) –John gave a book to Mary“unmarked” –John gave Mary a book.“marked” –To whom did John give a book?“unmarked”? –Who did John give a book to?“marked” Assuming that the second of each pair is marked, Mazurkewich asked about timing of each in L2A. But although maybe more languages allow the first of each pair than the second, the pied-piping example should make us suspicious.

7 Mazurkewich (1984) French-->English and Inuktitut-->English French lacks pied-piping and double-object constructions. Inuktitut is different enough that it is hard to find an analog to either the marked or unmarked constructions. Did the L2’ers prefer the unmarked structures? Did they acquire them first?

8 Mazurkewich (1984) French-L1 beginners do appear to prefer the unmarked structures (2-to-1), and the marked structures gain ground as L2’ers become more advanced. But French lacks the marked structure; did they “start with the unmarked structure” or did they “start with the structure of their L1”? As for Inuktitut, they weakly preferred the unmarked structures (beginners 81% to 96%). Not very dramatic, not very convincing.

9 Mazurkewich (1984) Worse, on a different task (“question italicized phrase”), although the French speakers showed a moderate preference for “unmarked” (pied-piping) structures, the Inuktitut speakers showed a preference for the marked structure. However, it could be that the whole experiment isn’t getting at what we want. The controls preferred the marked structure 3 or 4-to-1, so these “unmarked” structures seem to be marked from a language-internal perspective. Plus, this gives the learner a lot of evidence.

10 Problems so far If L1 has an “unmarked” value for something and L2 has a “marked” value, if the L2’er prefers (or, better, learns more quickly) the “unmarked” value, it could be either transfer or reverting to an unmarked value. The actual marked/unmarked set must be convincingly chosen—means nothing if we aren’t actually looking at marked/unmarked.

11 Best test would be… Find a convincingly marked vs. unmarked pair, … Find an L2 which allows only the marked option, … Test speakers of an L1 which also only allows the marked option, … …and see if L2’ers use/accept the unmarked option early on.

12 Unmarked vs. transfer At the start of L2, do we start with unmarked (default) values for the parameters? Do we transfer the parameter settings from L1? Is transfer affected by markedness? (“Unmarked properties of the L1 more likely to transfer than marked properties of L1”) Wait, how is that different from the first one?

13 Liceras (1985, 1986) Another potential marked/unmarked pair: –Allows Ø comp. (marked; English) –Disallows Ø comp. (unmarked; Spanish) English->Spanish –Beginners: 49% acceptance of Ø comp. –Intermediate: 25% acceptance. –Advanced: 9% acceptance. Looks like transfer (not initial unmarkedness) (contra Liceras’ hypothesis)

14 Subset principle Reminder: The Subset Principle is that learners are conservative—they only assume a grammar sufficient to generate the sentences they hear, allowing positive evidence to serve to move them to a different parameter setting. Given a choice, the L2’er assumes a grammatical option that generates a subset of the what the alternative generates.

15 Subset principle and markedness Based on the Subset principle, we’d expect the unmarked values (in a UG where languages are learnable) to be the ones which produce the “smallest” grammars. Given that in L1A we don’t seem to see any “misset” parameters, we have at least indirect evidence that the Subset principle is at work. Is there any evidence for it in L2A?

16 Subset options One possibility is that L2A starts with all of the defaults, the maximally conservative grammar. Another, mutually exclusive possibility (parameter by parameter, anyway) is that L2A starts with the L1 setting. –This means that for certain pairs of L1 and L2, where the L1 has the marked (superset) value and L2 has the unmarked (subset) value, only negative evidence could move the L2’er to the right setting. Or, some mixture of the two in different areas.

17 (Non-)Resiliency of results White (1989) covers a discussion of case adjacency parameter in French vs. English –*Mary does quickly her homework. –Marie fait lentement ses devoirs. …and the “configurationality parameter” (flat structure vs. hierarchical structure) –Japanese –English But nobody believes that these are the right characterizations of dimensions along which languages vary. For example, Case adjacency falls out of something else (verb movement) which isn’t in a subset relation.

18 Governing categories Binding Theory. Recall: –Principle A: Anaphors must be bound in their governing category. –Principle B: Pronouns must be free (unbound) in their governing category. BT seems to be universal. But maybe the governing category differs from language to language… based on the following facts:

19 Governing categories English: same clause, roughly. –John said Mary saw herself/*himself. –John wanted Mary to see herself/*himself. Russian: same finite clause, roughly. –John said Mary saw herself/*himself. –John wanted Mary to see herself/himself. Japanese: unbounded, roughly. –John said Mary saw herself/himself.

20 Governing categories So, these look like they are in a superset/subset relationship. –Smallest: English (clause) –Larger: Russian (finite clause) –Largest: Japanese (unbounded) Subset Principle: Adopt smallest first. Transfer: Adopt L1 GC first. UG: Adopt at least a possible value. *UG: Adopt an “unnatural” value.

21 Finer & Broselow (1986) Korean (unbounded) -> English (clause) Looks like they didn’t transfer the “unbounded” setting for the finite clause (adopted the Subset value). But there is a distinction between finite and nonfinite (not from L1, not from L2, but like Russian) LocalNon- local Either Finite9280 Nonfinite58384

22 Finer & Broselow (1986) Yet, wait… Why were there so few either responses? The Russian grammar and the Korean grammar both allow long-distance reflexives, but they don’t prohibit short-distance reflexives. Are we really seeing a (UG- )grammatical option here? LocalNon- local Either Finite9280 Nonfinite58384

23 Hirakawa (1990) Japanese->English Similar; still a finite/nonfinite distinction. (MacLaughlin 1995 analyzed the data more closely and confirmed that this is representative of some individual subjects as well). Yet… why so many non-local finite examples, if they are “speaking Russian? LocalNon- local Either Finite77176 Nonfinite54378

24 Thomas (1989) Spanish/Chinese->English Amazingly, the Spanish-speakers seemed to be accepting a lot of “non-local” or “either” antecedents. Can’t be coming from the Subset principle, nor can it be coming from the L1. And it isn’t Russian either. all finiteLocalNon- local Either Spanish601921 Chinese69724

25 Several mysteries… Why did so many subjects go only for LD reflexives (rather than either), when this isn’t what either Korean or Russian does? (H90, F&B86) Why did subjects, even while making a finite/non- finite distinction, seem to allow some LD reflexives out of finite clauses (like, yet unlike Russian)? (H90) Why do the Spanish speakers allow so many LD reflexives when it isn’t in either L1 or L2, and disobeys the Subset Principle? (T89)

26 Why we don’t know anything about the L2A Subset Principle First of all, the idea that there is a global setting for “governing category” that covers these cases is almost certainly wrong. Icelandic, Dutch, and, according to recent information, probably Thai, and in fact probably even Russian, have both LD and strictly local reflexives. –(See MacLaughlin 1995 for some related commentary as well)

27 Binding possibilities General principles Where the reflexive element is morphologically simple (carrying no information about person), it can be bound long-distance. Where the reflexive element carries person information (himself, myself), it can only be bound locally. Languages can have both kinds, no one GC parameter.

28 Binding possibilities So, if GC isn’t a parameter except in the morphological form of the verb, the whole Superset/Subset question disappears. Where L2’ers treat reflexives as able to be bound long-distance, it might be because they’ve morphologically mis-analyzed it. The source of the finite/non-finite distinction, say, for Russian svoj ‘self’s’ may submit to a similar kind of re-thinking.

29 Some thoughts on the mysteries… Why did so many subjects go only for LD reflexives (rather than either), when this isn’t what either Korean or Russian does? (H90, F&B86) Probably, the tasks are not successfully testing knowledge (that is, possibilities), but rather preferences (despite perhaps explicit instructions). Best test would be to do whatever possible to bias the toward each reading, so that the reading is obvious. Then see if the L2’ers get them.

30 Speaking of bias and preferences John said that Mary likes ____ –herself –himself –either –neither John said that Bill likes himself. –John –Bill –either –neither

31 Speaking of bias and preferences Turns out: NSs tend to mess these up, seeing only one reading where two are grammatically possible. Mary showed Susan a picture of herself. Preferences are obscuring the competence (if for NSs, surely also for L2’ers).

32 Speaking of bias and preferences White et al. (1997) tried L1 techniques on L2’ers (focus on desired reading rather than asking subjects to come up with it themselves) Story which sets up a context, and then asks for if test sentence is true or false. Picture about which sentence can be true or false. Turned out: Story task worked well for NSs, picture task highly biased (guess was: subjects read sentence first, got an interpretation which might not match picture, and were quick to say “false”).

33 White et al 1997 The results basically were that Japanese and French L1-speakers learning English were nearly as accurate as NSs with respect to binding anaphors… Except LD binding (in both French and Japanese groups, although higher in the Japanese group) out of nonfinite clauses was allowed significantly more often than a) NSs allow it, b) out of a finite clause.

34 Back to the the mysteries… Why did subjects, even while making a finite/non- finite distinction, seem to allow some LD reflexives out of finite clauses (like, yet unlike Russian)? (H90) This remains somewhat mysterious, but might revolve around why Russian allows svoj ‘self’s’ to get out of finite clauses in the first place. Plus, this kind of data is hard to interpret without a “biasing” experiment to give us a better idea of what’s really possible.

35 More thoughts on the mysteries… Why do the Spanish speakers allow so many LD reflexives when it isn’t in either L1 or L2, and disobeys the Subset Principle? (T89) Perhaps a “morphological misanalysis” explanation; quite possibly the Subset Principle doesn’t govern this part of grammar acquisition, yet UG provides information about the binding possibilities of morphologically simple anaphors. Could the people who allowed LD be people who didn’t properly analyze himself?

36 End of the semester Wrap up & Take-Home Points

37 Some major views on L1A/syntax Radford: kids lack functional elements initially, have only lexical elements. Wexler: kids have access to all the same grammatical elements that adults do. Rizzi: kids have “truncated trees” Vainikka: kids “grow trees”

38 L1A: Case errors Kids will sometimes make case errors with the subject (until around 2). –Me got bean. –In English, accusative (me) is the “default.” –Very often taken to indicate a subject not in SpecIP (a.k.a. SpecAgrSP). No IP? (Radford) Sometimes IP and above (Rizzi, Vainikka)? No AgrSP? (Wexler)

39 L1A: Null subjects Kids will also often drop out subjects, even in languages where null subjects are not allowed. –Hyams (1986): Mis-set parameter; they’re speaking Italian initially. –Kids who are learning null subject languages drop more subjects than kids who are learning non-null subject languages. –Bloom: Long sentences are harder, drop what you can. –Wexler/Hyams: Kids drop more subjects with nonfinite verbs.

40 L1A: Optional Infinitives In non-null subject languages, kids will allow infinitive verbs in root clauses sometimes, early on (up to a little after 2). –Rizzi: until maturation of Root=CP, trees truncated sometimes below tense. –Wexler/Schütze: Syntax intact, but something prohibits the same (subject) DP from licensing both TP (finite tense) and AgrP (Nom case).

41 L1A: Finite vs. nonfinite During Optional Infinitive stage, kids with OI’s treat finite verbs like finite verbs and nonfinite verbs like nonfinite verbs. German (Poeppel & Wexler): V2 for finite verbs, final V for nonfinite verbs. French (Pierce): Verb before pas for finite verbs, verb after pas for nonfinite verbs.

42 L1A: Principle B and Principle P Even older kids seem to allow co-reference in apparent violation of Principle B: Mary saw her. Chien & Wexler show that when quantifier binding is available and requires coindexation, Principle B is respected. Principle P is slow in coming (matures?), which says coreference --> coindexation.

43 L1A: A-chains, passives Kids are also slow to master passives and unaccusatives. Borer & Wexler (1987): This is maturation of the ability to represent “A-chains”— more specifically, the ability to move an object-type thing into a subject-type position (non-local assignment of  -roles).

44 L1A: Negation outside of IP Kids for a while seem to have trouble with negation outside the IP, and repair their utterances so that it remains inside (usually in an adult-ungrammatical way). –What kind of bread do you don’t like? –Where he couldn’t eat the raisins?

45 L1A: Syntax In general, the errors kids are making seem to be very systematic. They seem to know many aspects of the grammatical system, allowing us to pinpoint (if we look closely enough and ask the right questions) what parts don’t seem to be working. –A-chains. –Using a [D] feature twice to check functional features. –Allowing negation in C. –Requiring coreference to imply coindexation.

46 L2A: What can we say? Certain things are required to explain L1A. Kids don’t get negative evidence –or if they do, it is inconsistent, it is noisy, and moreover sometimes when we try to give them negative evidence, they ignore it. The kids must be able to learn a system that assign * to some sentences, based only on positive evidence. Conclusion: Universal Grammar constrains the kinds of languages there can be, those languages cannot generate certain kinds of sentences (hence: *).

47 L2A: What can we say? L1A: Languages differ from one another. Something needs to be learned from the environment. Yet much of the grammatical system seems common across languages. Languages can be thought of as varying not in the system (the principles) but in the parameters. The kids, who learn their native language so fast, must have some help setting the parameters. A Language Acquisition Device (LAD) designed to choose among the options made available by UG.

48 L2A: What can we say? L2A is generally much harder, more conscious, slower, less successful. What’s different about L2A? Did UG disappear? Did the LAD disappear? Question: What is the state of the L2’ers knowledge about the L2? Does this conform to what UG would allow?

49 L2A: UG-accessibility In general, it seems that the evidence points to the interlanguages being allowable human languages. This could either be influence from UG (constraining possible languages) or because the IL is a variation on L1. Can we tell? Look at parameter settings: Does IL represent a different option from L1?

50 L2A: Transfer If the IL is UG-constrained, what is the initial starting assumption? Is it some kind of general default setting for all the parameters (likely to be a “subset” grammar from which all other grammars can be learned via position evidence alone)? Is it just carrying over the parameter settings from L1? Some combination of these?

51 L2A: Tricks In order to look properly at parameters, we need to know what they are. And what a “default” setting might be. This turns out to be hard. Pro-drop parameter. Default: Drop subjects? Subset learnable? Correlated with anything else? Binding Theory Governing Category? Default? Language-wide? Strictly predictable from morphology?

52 L2A: Interlanguage = L1+prescriptive rules? Is the IL just L1 plus some prescriptive rules (LLK)? (Fundamental Difference) Or does the IL actually show resetting of parameters? Resetting should entail: cluster of properties comes with new value (again requires that we know what the parameters, values, clusters are) If we can find a non-L1, non-L2, but UG-available option in the IL, that also suggests parameter setting.

53 L2A: Is there a difference between kids and adults? L2A is harder as you get older. L1A is quite possibility bounded in time. Evidence for CPs seem to point to different CPs for different subsystems (phonology 6, morphosyntax 15, comprehension 10) CPs exist in vision, maybe we can find a brain correlate? Yet some people may manage to overcome this and become indistinguishable from a native speaker. Some plasticity remains? What disappears/deteriorates? UG? LAD?

54 L2A: Negative evidence useful? L1A doesn’t use negative evidence. If there is parameter transfer into IL from L1, logical subset relations might require negative evidence to reach correct parameter setting. Providing people with negative evidence seems to help—but only in the short term (without prolonged practicing), it may not yield any permanent “parameter resetting.”

55 L2A: Markedness? Are “unmarked” things easier/quicker to learn than “marked” things? Does teaching the “marked” things give you the “unmarked” things for free? What are the marked and unmarked things? Why do we see generalization beyond the marked (e.g., in Doughty’s NPAH experiment)

56 Bottom line: Especially with respect to L2A, there are a lot of things left to discover because careful and theoretically informed experiments still need to be done. Many of the experiments that are in the literature rely on misleading simplistic notions (a monolithic UG subsuming the LAD, a single once-and-for-all CPH, a one-stage-at-a time view of acquisition, a subset relation for adverb placement or binding domain definitions)…

57           