GRS LX 700 Two hypotheses about learning Language Acquisition and Linguistic Theory ¥ LLEE (late learning early emergence) —A “commonsense” view— Things which “emerge” emerge early. Things which are learned appear later. Week 5. Optional , Unique Checking Constraint, ATOM, … Implies that parameters should be set late too (since at least the settings are learned)

Two hypotheses about learning Two-word stage? ¥ VEPS (very early parameter setting) Basic parameters are set correctly at the earliest ¥ The reason both VEPS and VEKI mention observable stages, that is, at least from the time the two-word stage is just because this is the that the child enters the two-word stage around 18 first stage where we have evidence of months of age. utterance composition. ¥ VEKI (very early knowledge of ) At the earliest observable stage (two-word stage), the child knows the grammatical and phonological properties of many important inflectional elements of their language.

Optional stage Optional Infinitive Stage

¥ Root infinitives are possible grammatical ¥ German: V2/SOV; kids put finite in sentences; second position, leave nonfinite verbs in ¥ These infinitives co-exist with finite forms; -final position ¥ The children know the relevant grammatical principles (and how they distinguish finite ¥ French: V-->I; kids move finite verbs to I, from nonfinite). leave nonfinite verbs in VP. Very Early Parameter Setting VEPS and the theory of learning

¥ As soon as you can see it, kids have: ¥ If parameters are set by the time kids are Ð VO vs. OV order set (Swedish vs. German) using multi-word utterance, no negative Ð V-->I [yes/no] (French vs. English) evidence could have played a role. Ð V2 [yes/no ] (German vs. French/English) Ð Null [yes/no] (Italian vs. Fr./E.) ¥ So, at least by the 2-word stage, they have the parameters set (maybe earlier)

Null subjects… Topic drop

¥ Null subject parameter is not initially mis-set (kids ¥ Where kids drop the subject of a finite , don’t all start off speaking Italian—contra Hyams perhaps this is “Topic-drop” 1986); rather, child null subjects are (at least in part) due to the availability of non-finite verbs (the OI stage). ¥ Proposal: ¥ Some null subjects are licensed by being the Topic-drop applies to Very Strong Topics subject of a nonfinite verb (i.e. PRO) ¥ But there are some null subjects with finite Kids sometimes take (in reality) non-VS verbs… topics to be VS topics (a pragmatic error)

Bromberg, Wexler, wh-questions, Prediction about NS and null subjects ¥ OI’s have two ways of licensing NSs: ¥ If topic drop is something which drops a Ð PRO (regular licensing of null subject) topic in SpecCP… Ð Topic drop ¥ …and if wh-words also move to SpecCP… ¥ Finite verbs have one way to license a NS: ¥ …we would not expect null subjects with Ð Topic drop non-subject (e.g., where) wh-questions where the verb is finite (so PRO is not ¥ So: We expect more null subjects with root licensed) infinitives (which we in fact see) Bromberg, Wexler, wh-questions, *Truncation and null subjects ¥ And, that’s what they found: ¥ Rizzi’s “truncation” theory predicts: ¥ No wh-questions with root infinitives Finiteness of null and pronominal subjects in Adam’s Ð wh-question ⇒ CP, but wh-questions (Bromberg & Wexler 1995) Ð CP ⇒ IP, and Finite Nonfinite Ð IP ⇒ Null 2 118 ¥ And of course we wouldn’t expect null subjects in wh-questions if null subjects are 117 131 allowed (only) in the specifier of the root.

But… *LLEE

¥ German and Dutch have extremely few root ¥ So: despite expectations of early infinitives when there is anything in SpecCP. practitioners of P&P: VEPS means *LLEE. ¥ What then do we make of the fact that kids ¥ But they are V2 languages—finite verbs are what make non-adult utterances in the face of you find in C, and when SpecCP is filled, there must be something in C. Hence, the prediction evidence that they aren’t learning the seems to be: parameters? ¥ KW: Certain (very specific, it turns out) ¥ V2 languages ⇒ no wh-question root infinitives properties of the grammar mature.

VEKI? ATOM

¥ Generally, when kids use inflection, they use it ¥ Adult clause structure: correctly. Mismatches are vanishingly rare. AgrP Ð English (Harris & Wexler 1995) Ð German (Poeppel & Wexler 1993) ′ NOMi Agr

¥ Again, this is kind of contrary to what the field Agr TP had been assuming (which was: kids are slow at, ′ bad at, learning inflection). ti T TVP ATOM ATOM

¥ Kiddie clause, missing TP (—TNS): ¥ Kiddie clause, missing AgrP (—AGR):

AgrP

′ NOMi Agr Agr TP ⇒ ′ ACC defaulti T

VP TVP

Why either missing TP or AgrP One prediction of ATOM gives us a root infinitive (DM) ¥ In English, we have the following rules for ¥ +AGR +TNS: NOM with inflected verb (-s) pronouncing this tense/agreement affix: ¥ +AGR ÐTNS: NOM with bare verb ¥ ÐAGR +TNS: default (ACC) with bare verb ¥ (V+)T is pronounced like: ¥ ÐAGR ÐTNS: GEN with bare verb /s/ if we have features [3, sg, present] (not discussed but see Schütze & Wexler 1996) /ed/ if we have the feature [past] ¯ otherwise ¥ Nothing predicts ACC with inflected verb.

EPP and missing INFL *Truncation again

¥ Prior to splitting into AgrSP and TP, the ¥ Incidentally, Rizzi’s “truncation” theory has hypothesis was that IP was missing and that the same problem—if root infinitives are IP was responsible for both NOM and tense. missing everything above tense, how come ¥ Yet, there are many cases of root infinitives so many bare forms surface with NOM with NOM subjects (atom:+Agr ÐTns) subjects? And why do the subjects raise past ¥ And, even ACC subjects seem to raise out negation, and to where? Topic?! Bah. of the VP over negation (me not go) Wait—how can you say kids are Wait—how can you say kids are UG-constrained yet drop T/Agr? UG-constrained yet drop T/Agr? ¥ So, aren’t TP and AgrSP required by UG? ¥ Perhaps what requires TP and AgrP are principles Doesn’t this mean kids don’t have UG- of interpretation… ¥ You need TP so that your sentence is “anchored” compliant trees? in the discourse. ¥ Actually, perhaps no. UG requires that all ¥ You need AgrSP … why? Well, perhaps features be checked, but it isn’t clear that something parallel…? there is a UG principle that requires a TP ¥ Regardless, kids can check all the uninterpretable and an AgrP in every clause. features even without TP or AgrSP; hence, they can still be considered to be UG-constrained.

NS/OI NS/OI

¥ Some languages appear not to undergo the ¥ What differentiates the OI and non-OI languages? “optional infinitive” stage. How can this be ¥ Agreement? Italian (non-OI) has rich agreement, consistent with a maturational view? but so does Icelandic (OI). Ð OI languages: studied to ¥ Null subjects! date (Danish, Dutch, English, Faroese, Icelandic, Norwegian, Swedish), Irish, Russian, ¥ Null Subject/OI Generalization: Brazilian Portuguese, Czech Children in a language go through an OI stage iff Ð Non-OI languages: Italian, Spanish, Catalan, the language is not an INFL-licensed null subject Tamil, Polish language.

NS/OI and Hebrew NS/OI and Hebrew (Rhee & Wexler 1995) (Rhee & Wexler 1995) ¥ Hebrew is a NS language but only in 1st and 2nd person, non-present tense. kids up to 1;11 1/2 past/fut (NS) else (non-NS) Everywhere else (3rd past, future, present) null subjects 0 (of 21) 32% (36/112) subjects are obligatory. overt subjects 0 (of 6) 0 (of 28) ¥ Hebrew-learning 2-year-olds showed optional infinitives except in 1/2-past, and all OI kids 1/2 past/fut (NS) else (non-NS) allowed null subjects elsewhere, with null subjects 0.6% (1/171) 25% (85/337) infinitives. overt subjects 1.4% (1/72) 0.6% (3/530) The UCC “Minimize Violations”

¥ Unique Checking Constraint ¥ The poor kid with the UCC is faced with a The D-feature of a DP can only check dilemma. It knows that it must obey three against one functional category things, but it can obey only any two: Ð Realize tense ¥ Adults must check the D feature of the Ð Realize agreement subject both in TP and in AgrSP. Kids only Ð Unique checking constraint get one, so they have to choose: Omit TP, ¥ So, the kid chooses which one to violate Omit AgrSP, or ignore the UCC. (satisfying two) for any given utterance.

Speculations about D-features NS/OI via UCC

¥ In adult a DP must check its D- ¥ An old idea about NS languages is that they feature against both AgrSP and TP. arise in languages where Infl is “rich” ¥ AgrSP and TP each have a D-feature of a enough to identify the subject. special sort—the kind that deletes when it is ¥ Maybe in NS languages, AgrS does not checked (an uninterpretable feature) need a D (it may in some sense be nouny ¥ Might the UCC be a consequence of kids enough to say that it is, or already has, D). mistakenly taking the D-feature of DP to be ¥ If AgrS does not need a D, the subject is uninterpretable? free to check off T’s D-feature and be done.

Is there any way to see the effects Or maybe in Korean negation… of UCC even in NS languages? ¥ Italian: Mary has laughed. ¥ Short Form Negation in Korean: ¥ Suppose that auxiliaries (like have) also Chelswu-ka pap-ul an-mek-ess-ta. have a D-feature to be checked as the Chelswu-nom rice-acc neg-eat-past-decl subject (in the adult language) passes ‘Chelswu didn’t eat rice.’ through. UCC-constrained kids will have to drop something (the auxiliary or T) ¥ Common OI-age kid error: ¥ Lyons (1997) reports that a “substantial proportion of auxiliaries are omitted in OI- na an pap mek-e age Italian.” I neg rice eat-decl ‘I don’t eat rice.’ Negation errors in child Korean Negation errors in child Korean

¥ Generalization about child errors with SFN: ¥ Can this error be made to follow from the UCC VP-internal material is privileged in its ability to (you can’t check a D-feature twice)? occur between an and the verb in child errors. ¥ Kid errors seem to involve a structure like:

neg […VP material… ] verb ¥ Subjects (except subjects of unaccusatives) never appear between an and the verb suggesting that adult negation has a movement ¥ Objects often do that kids are failing to do: ¥ often do […VP material…]i neg ti verb

One movement down… That’s two movements

¥ Adults also seem to perform a second ¥ So, the (and some of the VP-internal movement of the object; the cal material) seems to have to move twice in ‘well’ must immediately precede the verb negative sentences, once to get around cal (unlike other adverbs)—but presumably the (in any kind of sentence), and again to get object originally (at D-structure) falls around an (neg). between cal and the verb. Hence: ¥ That’s what we need to get off the ground if we want to attribute this error to the UCC. objecti … cal … ti verb

The proposal The proposal ¥ In Korean, the object moves to SpecAgrOP (step ¥ Then, AgrOP moves to an AgrNegP above one) and checks a D-feature: negation, to check a D-feature:

AgrOP AgrNegP DP AgrO′ ′ i AgrOPi AgrNeg AgrO VP [D] AgrNeg[D] NegP cal VP an Neg′ V t i Neg …ti… The proposal Predictions

¥ The kid can only do one of those ¥ Makes very specific predictions about what movements if it obeys the UCC, since each we would expect to find: one requires the same D-feature ¥ Omit AgrNegP (retaining AgrOP): (contributed by the object). ¥ So, the kid must either Ð Object moves (over cal) to SpecAgrOP Ð AgrOP (with cal and object) remain below Ð ignore the UCC, or NegP. Ð omit AgrOP, or Ð omit AgrNegP ¥ an object cal verb

Predictions Met?

¥ Omit AgrOP (retaining AgrNegP) ¥ Sadly, the experiments haven’t been done Ð Object (nearest thing with a D-feature) moves and the examples haven’t been reported in directly to SpecAgrNegP, over an and cal. the literature. Ð We need errors with transitive verbs involving ¥ object an cal verb short-form negation and the adverb cal…

¥ Note: This is totally ungrammatical in adult Korean, which requires object cal an verb.

Predictions for unaccusatives So…

¥ Unaccusative “subjects” start out in object ¥ The UCC seems to be pretty successful in position, and must presumably move through explaining why either TP or AgrSP are often many more projections (AgrOP, AgrNegP, TP, omitted for kids in languages like French, AgrSP) German. ¥ UCC kid can still just do one. ¥ The connection to the NS/OI generalization is ¥ Only one will yield a non-adult order: keep AgrOP reasonable to explain why we don’t seem to see and you get: an subject cal verb. OIs in Italian. ¥ Turns out: kids make only about 10% (detectible) ¥ The more general prediction that the UCC makes errors with unaccusatives (vs. about 30% with about double-movements to check D-features may transitives). A successful prediction? well be borne out by the facts of Korean negation. One open question… Legendre et al. (2000)

¥ The UCC says you can only use a D-feature ¥ Proposes a system to predict the proportions on a DP to check against a functional of the time kids choose the different options category once. among: ¥ This explains why sometimes TP is omitted Ð Omit TP (keeping AgrSP) and sometimes AgrSP is Ð Omit AgrSP omitted (keeping TP). Ð Omit both TP and AgrSP ¥ but if GEN infin. comes from omitting both TP and AgrSP, what could ever cause that Ð Include both TP and AgrSP (violating UCC) (particularly given Minimize Violations)?

The idea The idea

¥ Kids are subject to conflicting constraints: ¥ Sometimes Parse-T beats out *F, and then Ð Parse-T Include a projection for tense there’s a TP. Or Parse-Agr beats out *F, and Ð Parse-Agr Include a project for agreement then there’s an AgrP. Or both Parse-T and 2 Ð *F Don’t complicate your tree with Parse-Agr beat out *F , and so there’s both functional projections a TP and an AgrP. Ð*F2 Don’t complicate your tree so much as to have two functional ¥ But what does sometimes mean? projections.

Floating constraints Floating constraints

¥ The innovation in Legendre et al. (2000) *F that gets us off the ground is the idea that as kids re-rank constraints, the position of the Parse-T constraint in the hierarchy can get somewhat fuzzy, such that two positions ¥ When the kid evaluates a form in the can overlap. constraint system, the position of Parse-T is *F fixed somewhere in the range—and winds up sometimes outranking, and sometimes Parse-T outranked by, *F. Floating constraints French kid data

*F ¥ Looked at 3 French kids from CHILDES ¥ Broke development into stages based on a Parse-T modified MLU-type measure based on how long most of their utterances were (2 words, ¥ (Under certain assumptions) this predicts more than 2 words) and how many of the that we would see TP in the structure 50% of the time, and see structures without TP utterances contain verbs. the other 50% of the time. ¥ Looked at tense and agreement in each of the three stages represented in the data.

French kid data French kid data

¥ Kids start out using 3sg agreement and ¥ This means if a kid uses 3sg or present tense, present tense for practically everything we can’t tell if they are really using 3sg (correct or not). (they might be) or if they are not using ¥ We took this to be a “default” agreement at all and just pronouncing the Ð (No agreement? Pronounce it as 3sg. No tense? default. pronounce it as present. Neither? Pronounce it ¥ So, we looked at non-present tense forms as an infinitive.). and non-3sg forms only to avoid the question of the defaults.

French kids data A detail about counting

¥ We found that tense and agreement develop ¥ We counted non-3sg and non-present verbs. differently—specifically, in the first stage ¥ In order to see how close kids’ utterances were to adult’s utterances, we need to know how often we looked at, kids were using tense fine, but adults use non-3sg and non-present, and then see then in the next stage, they got worse as the how close the kids are to matching that level. agreement improved. ¥ So, adults use non-present tense around 31% of ¥ Middle stage: looks like the time—so when a kid uses 31% non-present tense, we take that to be “100% success” competition between T ¥ In the last stage we looked at, kids were basically and Agr for a single node. right at the “100% success” level for both tense and agreement. Proportion of non-present and Proportion of non-finite root non-3sg verbs forms

40% 35%

35% 30% 30% 25% 25% non-present non-3sg 20% 20% NRFs adult non-pres 15% 15% adult non-3sg 10% 10% 5% 5% 0% 0% 3b 4b 4c 3b 4b 4c

A model to predict the A model to predict the percentages percentages ¥ Stage 3b (first stage) ¥ Stage 4b (second stage) ¥ no agreement ¥ non-3sg agreement and non-present tense each about 15% (=about 40% agreeing, 50% tensed) ¥ about 1/3 NRFs, 2/3 tensed forms ¥ about 20% NRFs

2 *F *F *F2 *F ParseT ParseT ParseA ParseA

A model to predict the Predicted vs. observed—tense percentages

¥ Stage 4c (third stage) 40% ¥ everything appears to have tense and 35% 30% agreement (adult-like levels) 25% non-present 20% predicted non-pres *F2 *F 15% ParseT 10% 5% ParseA 0% 3b 4b 4c Predicted vs. observed—agr’t Predicted vs. observed—NRFs

40% 35%

35% 30% 30% 25% 25% non-3sg 20% NRFs 20% predicted non-3sg 15% predicted NRFs 15% 10% 10% 5% 5% 0% 0% 3b 4b 4c 3b 4b 4c

 For next time:

  ¥ Read Chien & Wexler (1990).  ¥ Write up a 1-2 page summary. Ð What are the primary points?   Ð What evidence supports the conclusions?   Ð Did you find the evidence convincing?  If not, why not?