Week 5b. Transfer and the “initial state” for L2a

CAS LX 500A1Topics in

Linguistics: Language

Acquisition

“UG in L2A” so far UG principles

(Subjacency, Binding Theory) UG parameters of variation

(Subjacency bounding nodes, Binding domains, null subject, VT)

Justified in large part on the basis of L1. the complexity of language the paucity of useful data the uniform success and speed of L1’ers

acquiring language.

“UG in L2A” so far To what extent is UG still involved in L2A? Speaker’s “interlanguage” shows a lot of

systematicity, complexity which also seems to be more than the linguistic input could motivate.

The question then: Is this systematicity “left over” (transferred) from the existing L1, where we know the systematicity exists already? Or is L2A also building up a new system like L1A?

We’ve seen that universal principles which operated in L1 seem to still operate in L2 (e.g., ECP and Japanese case markers).

Initial state: 3 options The L1 (parameter settings)

Schwartz & Sprouse (1996) “Full Transfer/Full Access”

Parts of the L1 (certain parameter settings) Eubank (1993/4) “Valueless Features

Hypothesis” Vainikka & Young-Scholten (1994) “Minimal

trees” Clean slate (UG defaults)

Epstein et. al (1996) Platzack (1996) “Initial Hypothesis of Syntax”

Vainikka & Young-Scholten

V&YS propose that phrase structure is built up from just a VP all the way up to a full clause.

Similar to Radford’s L1 proposal except that there is an order of acquisition even past the VP (i.e., IP before CP). Also similar to Rizzi’s L1 “truncation” proposal. And of course, basically the same as Vainikka’s L1 tree building proposal.

V&YS propose that both L1A and L2A involve this sort of “tree building.”

Vainikka (1993/4), L1A An adult clause,

where kids end up. The subject

pronoun is in nominative case (I, he, they), a case form reserved for SpecAgrP in finite clauses (cf. me, him, them or my, his, …).

Agr

Agr

AgrPC

C

CP

that

she

DP

T

T

TP

V DP

V

VPwill

eatlunch

Vainikka (1993/4), L1A Very early on, kids are

observed to use non-nominative subjects almost all the time (90%) like:

My make a house Nina (2;0)

The fact that the subject is non-nominative can be taken as an indication that it isn’t in SpecAgrP.

Agr

Agr

AgrPC

C

CP

that

she

DP

T

T

TP

V DP

V

VPwill

eatlunch

Vainikka (1993/4), L1A Vainikka’s

proposal was that children who do this are in a VP stage, where their entire syntactic representation of a sentence consists of a verb phrase.my

DP

V DP

V

VP

makea house

Vainikka (1993/4), L1A As children get

older, they start using nominative subjects

I color me Nina (2;1)

But interestingly, they do not use nominative subjects in wh-questions

Know what my making? Nina (2;4)

Agr

Agr

AgrP

I

DP

T

T

TP

V DP

V

VP

colorme

Vainikka (1993/4), L1A I color me

Nina (2;1) The nominative

subject tells us that the kid has at least AgrP in their structure.

Know what my making? Nina (2;4)

Normally wh-movement implies a CP (wh-words are supposed to move into SpecCP).

Agr

Agr

AgrP

I

DP

T

T

TP

V DP

V

VP

colorme

Vainikka (1993/4), L1A Know what my

making? Nina (2;4)

However, if there is no CP, Vainikka hypothesizes that the wh-word goes to the highest specifier it can go to—SpecAgrP. Which means that the subject can’t be there, and hence can’t be nominative.

Agr

Agr

AgrP

my

DP

T

T

TP

V

DPi

V

VP

making

what

ti

Vainikka (1993/4), L1A Finally, kids reach

a stage where the whole tree is there and they use all nominative subjects, even in wh-questions.

Agr

Agr

AgrPC

C

CP

that

she

DP

T

T

TP

V DP

V

VPwill

eatlunch

Vainikka (1993/4)

So, to summarize the L1A proposal: Acquisition goes in (syntactically identifiable stages). Those stages correspond to ever-greater articulation of the tree. VP stage:

No nominative subjects, no wh-questions. AgrP stage:

Nominative subjects except in wh-questions. CP stage:

Nominative subjects and wh-questions.

Vainikka & Young-Scholten’s primary claims about L2A

Vainikka & Young-Scholten take this idea and propose that it also characterizes L2A… That is…

L2A takes place in stages, grammars which successively replace each other (perhaps after a period of competition).

The stages correspond to the “height” of the clausal structure.

Vainikka & Young-Scholten

V&YS claim that L2 phrase structure initially has no functional projections, and so as a consequence the only information that can be transferred from L1 at the initial state is that information associated with lexical categories (specifically, headedness). No parameters tied to functional projections (e.g., V->T) are transferred.

V&YS—headedness transfer

Cross-sectional: 6 Korean, 6 Spanish, 11 Turkish. Longitudinal: 1 Spanish, 4 Italian.

In the VP stage, speakers seem to produce sentences in which the headedness matches their L1 and not German.L1 L1 head % head-final VPs in L2Korean/Turkish final 98

Italian/Spanish (I) initial 19Italian/Spanish (II) initial 64

V&YS—headedness transfer

VP-i: L1 value transferred for head-parameter, trees truncated at VP.VP-ii: L2 value adopted for head-parameter, trees still truncated at VP

NL VPs V-initial V-finalBongiovanni I 20 13 (65%) 7Salvatore I 44 35 (80%) 9Jose S 20 15 (75%) 5Rosalinda S 24 24 (100%) 0Antonio S 68 20 48 (71%)Jose S 37 23 14 (38%)Lina I 24 7 17 (71%)Salvatore I 25 6 19 (76%)

Predictions Different parts of

the tree have different properties associated with them, and we want to think about what we would predict we’d see (if Vainikka & Young-Scholten are right) at the various stages.

Agr

Agr

AgrPC

C

CP

DP

T

T

TP

V DP

V

VP

Predictions T/Agr (=INFL):

Modals and auxiliaries appear there

Verbs, when they raise, raise to there.

Subject agreement is controlled there

C Complementizers

(that, if) appear there

Wh-questions involve movement to CP

Agr

Agr

AgrPC

C

CP

DP

T

T

TP

V DP

V

VP

Predictions So, if there is just

a VP, we expect to find: No evidence of verb

raising. No consistent

agreement with the subject.

No modals or auxiliaries.

No complementizers.

No complex sentences (embedded sentences)

No wh-movement.

Agr

Agr

AgrPC

C

CP

DP

T

T

TP

V DP

V

VP

V&YS L2A—VP stage

At the VP stage, we find lack of verb raising (INFL

and/or CP) auxiliaries and

modals (generated in INFL)

an agreement paradigm (INFL)

complementizers (CP)

wh-movement (CP)

stage L1 Aux Mod DefaultVP Kor 1 1 68VP Tur 0 1 75VP-i It 0 0 65VP-ii It 0 0 82VP-i Sp 8 5 74VP-ii Sp 1 1 57

All came from Rosalinda (Sp.); three instances of wolle ‘want’ and five with is(t) ‘is’—evidence seems to be that she doesn’t control IP yet.

V&YS L2A—VP stage At the VP stage, we find lack of

verb raising (INFL and/or CP) auxiliaries and modals (generated in INFL) an agreement paradigm (INFL) complementizers (CP) wh-movement (CP)

Antonio (Sp): 7 of 9 sentences with temporal adverbs show adverb–verb order (no raising); 9 of 10 with negation showed neg–verb order.

Turkish/Korean (visible) verb-raising only 14%.

V&YS L2A—VP stage At the VP stage, we find lack of

verb raising (INFL and/or CP) auxiliaries and modals (generated in INFL) an agreement paradigm (INFL) complementizers (CP) wh-movement (CP)

No embedded clauses with complementizers.

No wh-questions with a fronted wh-phrase (at least, not that requires a CP analysis).

No yes-no questions with a fronted verb.

V&YS L2A—TP stage After the VP stage, L2 learners move

to a single functional projection, which appears to be TP.

Modals and auxiliaries can start there.

Verb raising can take place to there. Note: the TL TP is head-final, however.

Agreement seems still to be lacking (TP only, and not yet AgrP is acquired).

V&YS L2A—TP stage Characteristics of the TP stage:

optional verb raising (to T) some auxiliaries and modals (to T) lack of an agreement paradigm (not up to AgrP yet) lack of complementizers (CP) lack of wh-movement (CP)

stage L1 Aux Mod DefaultTP Sp 21 9 41TP Tur [0] 5 68–75

Now, Korean/Turkish speakers raise the verb around 46% of the time.

V&YS L2A—AgrP stage

After the TP stage, there seems to be an AgrP stage (where AgrP is head-initial—different from the eventual L2 grammar, where AgrP should be head-final)

Properties of the AgrP stage: verb raising frequent auxiliaries and modals common agreement paradigm acquired some embedded clauses with

complementizers complex wh-questions attested.

V&YS L2A—AgrP Properties of the AgrP stage:

verb raising frequent auxiliaries and modals common agreement paradigm acquired some embedded clauses with complementizers complex wh-questions attested

Turkish/Korean speakers raising the verb 76% of the time.

CP structure? Seems to be “on its way in”, but V&YS don’t really have much to say about this.

Vainikka & Young-Scholten

Summary of the proposed stages

Top XP

V-mmt

aux/modals

obligsubjs

S–Vagrt

embedded w/ C

question formation

VP no no no no no noFP opt some no no no noAgrP yes yes yes yes no no

Stages So, L2’ers go through VP, TP, AgrP, (CP) stages… An important point about this is that this does not

mean that a L2 learner at a given point in time is necessarily in exactly one stage, producing exactly one kind of structure. (My response on V&YS’s behalf to an objection raised by

Epstein et al. 1996; V&YS’s endorsement should not be inferred.)

The way to think of this is that there is a progression of stages, but that adjacent stages often co-exist for a time—so, “between” the VP and TP stages, some utterances are VPs, some are TPs.

This might be perhaps comparable to knowledge of register in one’s L1, except that there is a definite progression.

V&YS summary So, Vainikka & Young-Scholten propose that

L2A is acquired by “building up” the syntactic tree—that beginner L2’ers have syntactic representations of their utterances which are lacking the functional projections which appear in the adult L1’s representations, but that they gradually acquire the full structure.

V&YS also propose that the information about the VP is borrowed wholesale from the L1, that there is no stage prior to having just a VP.

Lastly, V&YS consider this L2A to be just like L1A in course of acquisition (though they leave open the question of speed/success/etc.)

Problems with Minimal Trees

White (2003) reviews a number of difficulties that the Minimal Trees account has.

Data seems to be not very consistent. Evidence for DP and NegP from V&YS’s own data. E->F kids manage to get V left of pas (Grondin &

White 1996) but cf. Hawkins et al. next week. Also, these are kids who

might have benefited from earlier exposure to French. V&YS also propose at one point that V->T is the default

value. Some examples of early embedded clauses and SAI

(evidence of CP) but V&YS’s criteria would also lead to the conclusion of no IP at the same point. (Gavruseva & Lardiere 1996).

Problems with Minimal Trees

Criteria for stages are rather arbitrary. V&YS count something as acquired if it appears

more than 60% of the time. Why 60%? For kids, the arbitrary cutoff is often set at 90%.

Is morphology really the best indicator of knowledge? Prévost & White, discussed a couple of weeks hence,

say “no”— better is to look at the properties like word order that the functional categories are supposed to be responsible for.

To account for apparent V2 without CP, V&YS need a weird German story in which TP/AgrP starts out head-initial but is later returned to its proper head-final status.

Paradis et al. (1998) Paradis et al. (1998) looked at 15 English-speaking

children in Québec, learning French (since kindergarten, interviewed at the end of grade one), and sought to look for evidence for (or against) this kind of “tree building” in their syntax.

They looked at morphology to determine when the children “controlled” it (vs. producing a default) and whether there was a difference between the onset of tense and the onset of agreement.

On one interpretation of V&YS, they predict that tense should be controlled before agreement, since TP is lower in the tree that AgrP.

Agr before T

T before Agr

Both T and Agr at outset

3pl before tense

3pl after tense

Both 3pl and tense at outset

8 0 7 0 12 3Past before Fut

Fut before Past

Both Fut and Past at outset

6 2 7

Paradis et al. (1998)

Agr reliably before T 3pl late (of

agreements). Future late (of

tenses).

Paradis et al. (1998) So, the interpretation of this information might

be that:

(Child) L2A does seem to progress in stages.

This isn’t strictly compatible with the tree building approach, however, if TP is lower than AgrP. It would require slight revisions to make this work out (not necessarily drastic revisions).

Eubank: Valueless Features

Hypothesis Another contender for the title of Theory of the Initial State is the “Valueless Features Hypothesis” of Eubank (1993/4).

Like Minimal Trees, the VFH posits essentially that functional parameters are not initially set (not transferred from the L1).

Unlike Minimal Trees, the VFH does assume that the entire functional structure is there. But, e.g., for V->T, the parameter/feature value that determines whether V moves to T is “undefined”.

VFH The interpretation of a “valueless” feature

is the crucial point here. It’s not clear really what this should mean, but Eubank takes it to mean something like “not consistently on or off”. Hence, again using V->T as an example, the verb is predicted to sometimes raise (V->T on) and sometimes not (V->T off). E.g., either is fine in L2 English of: Pat eats often apples. Pat often eats apples.

VFH and V->T In fact (as we’ll discuss more carefully in a

couple of weeks), White did a well-known series of experiments on F>L2E learners that did show that the learners accepted both. Pat eats often apples. Pat often eats apples.

Eubank takes this as evidence for VFH, but White (1992, 2003) notes that it’s unexpected for the VFH that they don’t also allow verb raising past negation. *Pat eats not apples. Pat does not eat apples.

Yuan (2001) and {F,E}>L2C

Yuan (2001) looked at E>L2C and F>L2C learners’ responses to alternative verb-adverb orders in Chinese. L1 Chinese allows only Adv-V order (no raising). Zhangsan changchang kan dianshi. *Zhangsan kan changchang dianshi.

But neither group (and notably not even F>L2C) ever produced/accepted the V-Adv order. *VFH, but also possibly *FTFA (to be discussed soon).

One further note: Yuan’s subjects were adults, White’s were children. This might have mattered.

Eubank’s own experiments

Eubank & Grace (1998) tried an interesting methodology in an experiment to test for grammaticality of raised-verb structures in IL grammars. Something like a “lexical decision task” but with sentences (“are these the same or different?”), recording the reaction time, and based on the finding that native speakers are slower to react to ungrammatical sentences.

Eubank & Grace (1998) E&G tested C>L2E speakers, divided

them into two groups based on a pretest of their production of subject-verb agreement (idea: “no-agreement” subjects would have not valued their features yet, “agreement” subjects have at least valued some of them).

Finding: No-agreement subjects acted like native speakers, agreement subjects didn’t differentiate between grammatical and ungrammatical verb-adverb orders.

Hmm.

Eubank et al. (1997) Same basic premises, different tasks:

Tom draws slowly jumping monkeys. For a V-raiser, this should be ambiguous

(is the jumping slow or is the drawing slow?). Eubank et al. (1997) used a kind of TVJ task to test this.

Even prior to looking at the results, one problem here is that this is fine in L1 English if slowly is taken as a parenthetical (“Tom draws— slowly— jumping monkeys”). But that’s the crucial interpretation that is supposed to show verb raising is grammatical. What could we conclude, no matter what the results are?

Eubank et al. (1997) The actual results didn’t go along very well

with the predictions either. Pretty low acceptance rate of raised-V interpretations if they’re really supposed to be grammatical in the IL. And the agreement group wasn’t acting native-speaker-like either, even though they should have valued the feature.

Eubank et al. actually go further with the VFH, hypothesizing that this is not only the initial state, but also the inescapable final state—L2 features cannot be valued (hence the lack of serious improvement among the agreement group—”Local Impairment”, for next week).

Schwartz 1998 Promotes the idea that L2 patterns

come about from full transfer and full access. The entire L1 grammar (not just short

trees) is the starting point. Nothing stops parameters from being

reset in the IL.

Erdem (Haznedar 1995) An initial

SOV stage (transfer from Turkish) is evident, followed by a switch to SVO.

N-Adj orderParodi et al. (1997)

jene drei interessanten Bücherthose three interesting.pl books

ku se-kwon-uy caemiissnun chaek-tulthat three-cl-gen interesting book-pl

ben-im pekçok inginç kitab-Im1sg-gen many interesting book-1sg

quei tre libri interessantithose three books interesting.pl

esos tres libros interesantes those three books interesting.pl

N-Adj in Romance The standard way of looking

at N-Adj order in Romance (in terms of native speaker adult syntax) is like this: Adj N is the base order

German, Korean, Turkish N moves over Adj in Romance

Spanish, Italian

What did the L2’ers do learning German?

D

D

DP

N

adjective N

NP

…

Parodis 1997—N-Adj orderNL N-Adj (error)

Bongiovanni I 3/81/5

37.5%20.0%

Lina I 3/230/81/11

13.0%0.0%9.1%

Bruno I 9/3217/640/12

28.1%26.6%0.0%

Ana S 7/280/10

25.0%0.0%

Koreans K 1/102 1.0%Turks T 0/103 0.0%

So… So, movement seems to be initially

transferred, and has to be unlearned. The evidence for the tree building

approach doesn’t seem all that strong anymore. No nice Case results like in L1. Higher parameters seem to transfer (*VFH,

*Minimal Trees) Morphology and finiteness somewhat separate

(to be discussed in two weeks).

No transfer/Full access Epstein, Flynn, and Martohardjono

(1996) wrote a well-known BBS article endorsing the view that L2A is not only UG-constrained, but that it basically “starts over” with UG like L1A does.

Editorial comment: It’s worth reading, but the responses are at least as important as the article.

New parameter settings Japanese vs. English = SOV vs. SVO. EFM make a mysterious statement:

“Left-headed C° correlates with right-branching adjunction and right-headed C° with left-branching adjunction”

…followed by an example of how English allows both left and right adjunction.

What EFM must mean is that SVO language-speakers prefer postposed adverbial clauses. The worker called the owner [when the engineer finished the

plans]. [When the actor finished the book] the woman called the

professor.

New parameter settings And then EFM proceed to report that

Japanese speakers (J>L2E) don’t significantly prefer preverbal adverbial clauses (purported SOV preference), and even eventually prefer postverbal adverbial clauses (purported SVO preference).

But preferences are not parameter settings in any obvious way. Nothing is ruled out in any event—this is not a very useful result (see also Schwartz’s response).

Martohardjono 1993 Interesting test of relative

judgments. It is generally agreed that ECP

violations… Which waiter did the man leave the

table after spilled the soup? are worse than Subjacency

violations Which patient did Max explain how the

poison killed? Do L2’ers get these kinds of

judgments?

Martohardjono 1993 Turns out, yeah, they seem to. But it turns out that speakers of languages

without overt wh-movement had lower accuracy on judging the violations overall.

So: L1 has some effect (although EFM don’t really talk about this much, something which occupies much of the peer reviewers’ time).

EFM suggest that these judgments cannot be coming from the L1 alone, but of course this also relies on the view that L1 is significantly impoverished by “instantiation” (not the common view, not even in 1996).

EFM’s experiment Elicited imitation, Japanese speakers

learning English (33 kids, 18 adults).

Trying to elicit sentences with things associated with functional categories (tense marking, modals, do-support for IP; topicalization, relative clauses, wh-questions for CP).

The point was actually more to refute the idea that adults have UG “turned off” after a “critical period” than anything else (a discussion we’ll return to)

EFM’s experiment Kids did equally well in this repetition task as adults. Kids seemed to get around 70% success on IP-related

things, around 50% success on CP-related things. The deeper topicalizations are harder than shallower topicalizations.

EFM would have you believe: Based on their data collapsing over all kids and over all

adults, there are no stages. CP is there just as much as IP is there, despite the higher

success with IP, just because CP-related structures are intrinsically harder/more complex.

It could be true, but it’s certainly not a knock-down argument against V&YS or any of the other alternatives.

Also, as White (2003) notes, none of these sentences were ungrammatical (which we might have expected to be “repaired” under repetition)… if this is even a reliable task to begin with.