Null Subjects in Child Wh-Questions*

Hilary Sara Bromberg, Harvard University Kenneth Wexler, Massachusetts Institute oJ Technology

I. Introduction

Is there a deep structural relationship between the Optional Infinitive (OI) and Null Subject (NS) stages? We will provide evidence to support tile suggestion of Wexler (1992, 1994 footnote 44) that nonfinite verbs in the Oi stage of early child Englisll grammatically license a null subject, a suggestion further developed in a theory of the O! stage presented in Wexler 1995. The evidence we shall display emerges from tile particular pattern of null subjects that exists in declarative senlences and wh-quesiions.

Hyams (1986) proposed thai the omission of subjects in child English represented the misseliing of the null-subject parameter. In Italian and other null-subject (pro-drop) languages, subjects may generally be omiued, subject to certain pragmatic constraints. In English and other non-null-subject languages, subjects may generally not be omitted. Work in generative grammar (e.g., Rizzi 1982) has assumed that a property of the INFL system licenses null subjects in null-subject languages, whereas the corresponding property of the INFL system in English does not license such subjects. For example, one ,night claim thai agreement, perhaps represented by an AGR node, licenses null subjects in Italian but not in English.

The possibility thai there is a misseuing of the null-subject parameter has led Io much speculation about why such misseliing should occur, speculation •_ which has not received a satisfactory answer. Moreover, it would be quite an important property of the learning of parametric systems if in fact such misset- ling occurred. One purpose of this paper is to argue that there is no n|issetting of the null-subject parameter in early English and other non-null-subject languages. Rather, null subjects are largely a result of Ihe O! stage. I If this conclusion turns out to be true, and there is actually no misseuing of the null-subject

* We thank the Research Training Group in Language Acquisition and Computation (funded by NSF grant #DIR9113607) for providing an excellent research atmosphere. Thanks also to Carson Schiitze, Kevin Broihier, and Jenny Ganger for helpful comments and suggestions. This research wasconducted while HSB was a student at MIT. I The suggestion Ihal null subjects are relaled to the OI stage rather than to a misset par- ,meter was lirst made in Wexler 1992, 1994. Since then there have been a number of papers which agree with this conclusion, in various ways, including Rizzi 1994b, Roeper and Rohrbacher 1995,and Sano and llyams 1994.

MIT W_,rking Papers in Ltngui._tU_ 26, 221-247 Paperx on Language Pno_ essing and Acqut._ition. _ 1995 ttdary Sara Bromberg & Kenneth We,_ler tlilary Sara Bromberg and Kenneth Wexler Wh Null Subjecls

subject in English, and the sentence is ungrammatical. (The standard assumption parameter, then the null-subject parameter will pattern with other basic inflec- is Ihat in "null-subject" hmguages finite TENSE licenses a particular kind of tional/clausal parameters--that is, it will be set correctly from at least the very empty subject: pro; English is not a null-subject language, so pro is not licensed asbeginningproposedofbytheWeproductixler (1995),on of unmuhi-wordder the assumptiutteranceson of(i.e.,Veraroundy Earl), 18Paramonths),meter. by finite TENSE in English.) Whatever the exact grammatical explanation, it is Setting. generally assumed that nonfinite TENSE licenses a null subject: PRO. "[bus, we might look to the existence of root infinitives to explain the presence of null subjects in early child language, specifically during the Ol stage, when such nonli- A nmjor discovery concerning early child language is that children in nile TENSE is rampant. many languages go through un Optional Infinitive (OI) stage (Wexler 1992, 1994). In this stage, the child produces nonfinite root verbs along with finite Wexler (1992, 1994) argues that the characterization of tbe Optional ones. Crucially, the finite and nonfinite verbs occur in different distributional Infinitive stage should be extended to English, despite the fact that morphologi- contexts; it has been shown that finite verbs and only finite verbs move in child cally one cannot distinguish the infinitive from the verb st_ma'itn English. Thus, sentences to. positions to which finite verbs move in the adult language. Thus, he claims that when children produce forms like (2) they are showing the key Wexler concluded that children in the Oi stage actually "know" the difference characteristic of the O! stage--they are omitting TENSE. between finite and nonfinite morphology, despite the fact that they are willing to omit TENSE in some sentences demanding it. (2) a she go h is an open question exactly in which sense TENSE is ouritted from b. she not go nonfinite root sentences. Certainly the surface tense marking is omitted The c. she going question is, though, exactly what is missing from the representation, the structural description, of the nonfmite root sentence. As Wexler (1992, 1994) argued, In (2a), the (3rd singular) present TENS'E morpheme -s has been omitted. In some representation of TENSE other than the surface phonology had to be omit- (2b), TENSE has been omitted so that do-insertion, which is allowed onlg when ted; this is clear because the nonlinite root verb appears in the position of nonji- there is an unbound TENSE morpheme, does not occur. In (2c), TENSE has nite verbs. Thus, it is not a simple phonological "error." But exactly what is been omitted so that be is not required. (The assumpiion is that be is semanti- cally empty and only inserted to bind TENSE.) In addition, Harris and Wexler missing? is it the TENSE node? The projection headed by TENSE? Or simply a TENSI- li:ature? (in press) provide much empirical justification for the assumption that early En- glish represents the O1 stage; we will not review that evidence here (for further evidence, see Rice, Wexler, and Cleave in press). Wexler (1995) (see also lturris and Wexler in press) suggests that in the nonfinite sentences the entire TENSE projection, including the head TENSE, is Note that we are not attempting here to argue for any particular gram- missing in the nonfinite root (O!) sentences. One advantage to such an assump- matical characterization of the empty subject of non finite verbs. One possibility lion is that it explains why the infinitival morpheme to, which is often taken to is a characterization in terms of null-case (Chomsky & Lasnik 1991). Alterna- exist in the head of nonfinite TENSE, is missing in the OI sentences. Probably the results that we attain here can be carried out with a number of different as- lively, Wexler (1995) proposes that the TENSE projection is omitted from the Oi structures (for reasons that we will not delve into here) and that the lack of sumptions about what aspect of TENSE is missing in OI sentences. For example, TENSE means that in fact no case has to be assigned to the subject position. On •,, it is possible that in OI sentences, tt|e TENSE projection exists, but that the his analysis, the empty subject does not in fact need case of any kind--the case imsl/mmpast features which make the projection finite are missing In this paper filter is not part of the Minimalist framework That is, there is no UG principle we will not go into the question of whether Ibis assumption, or the assumption of that requires that all DPs have structural case, so long as there are no unchecked Wexler (1995) that the entire TENSE projection is missing, is best. features.

Wexler (1992, 1994) suggested that null subjects in child English (and Let us try to specify these considerations a little more precisely, men- in child versions of other non-null-subject languages) might be related to the OI tioning alternative ways ot working them out. First, we can divide the different stage, rather titan to a misset parameter. In particular, it seems natural to suggest possible TENSE features in a sentence into 3 kinds: TENSEl+finitel, that root nonfinite verbs license null subjects, since in general nonfinite verbs It- TENSEl-finitel and -TENSE. The third, -TENSE, corresponds to the sintple tense empty subjects. For example, consider control constructions such as ( I ): lack of a TENSE projection at all, what we are assuming for the O! sentences. The first (TENSE[+finitel) is the standard, Nominative-assigning finite TI-NSE. (I) a Mary expects le to leavel The second (TENSEl-finiteJ) is the TENSE found in the standard "intinitival" b. *Mary expects [e leaves] constructions in adults, for ex.'unple in control sentence, it comes in English with a to morpheme, which we assume, again following standard assumptions, fills The empty subject e of(la) (traditionally called PRO) is licensed by the nonfi- the head of TENSE. nile TI-NSE of to le_zve. But the finite TENSE in (Ib) dt)es not license :m empty

223 222 Iiilary Sara Bromberg and Kenneth Wexler W/t Null Subjects

easy to lind null subjecls ol finite verbs. There have been a number of ploposals Following Minimalist assumptions, TENSF.[-finile] assigns null ease. Io the effect thai child null subjects may be sortie khtd of "Topic-Drop" (de Ilaan This is Ihe case that is borne by only one kind of noun phrase, an entpty one. and "l'uijnman 1988; llyants & Wexler 1993; Wexler 1992, 1994) or "l)iary- Titus, Ihe slandard assumption is that the only kind of NP thai can be drawn Drop" (H:tegeman 199(); Rizzi 1994a), and we will assume that these proposals from Ihe lexicon with null case is the empty NP. can account for the null subjects of finite verbs. In particular, we will assume

Since OI sentences do not contain to, we have suggesled, IOIIowing thai there is a kind of Topic-Drop, which accounls fl_r finite null subjects. Wexler (1995), that the T projeclion is entirely missing. Titus, no case features-- That is, we suggest Ihat Ihere are two kinds of null subjects in Ihe lan- neither NOM nor NULL--are assigned in Ol representations to the subject post- • guage of children in the OI slage who are speaking non-null-subject la,lguages. lion. There is no structural case. A default morphological case is assigned to a Firsl, there is the kind of null subject that is licensed by a nonfinite verb. Second, visible NP which is the subject of an Oi (i.e., a senlence with no TENSE projec- there is the kind that represents stone kind of Topic-Drop, and may occur wilh lion). As Wexler (1995) points out, the subject of a small clause has default case finite verbs and (possibly) also with nonfinite verbs. assigned to it, at least in English. That is, subjects of small clauses are non- NOM. If small ehtuses are like OIs in that there is no TENSE projection (a fairly if finite verbs only allow null subjects which are a result of Topic-Drop, standard assumption), then it is natural in a Minimalist framework to assign de- but nonfinite verbs allow both Topic-Drop and nonfinite-tense licensed null sub- fault morphological case (i.e., not structural case) to the subject of a small jects, then we expect a higher proporlion of null subjects with nonfinite verbs clause2 than with finite verbs, This is exactly what we find in other non-null-subject

What kind of case, if any, is assigned to an empty subject of an O1? languages. Since there is no TENSE projeclion in an OI--not even a nonfinite TENSE pro- To take the example of German, Poeppel and Wexler (1993) studied jection--there is no TENSE-assigning element. Thus, the empty subject does nol one child, Andreas, in the Ol stage. As predicted by the properties of Ibis stage, have any structural case assigned to it. A more general question is, what kinds of it is virtually always the case Ihat finite verbs occur in second (V2) position and phi features, if any, does the empty subject of an OI have? At first sight it would nonfinite verbs occur in final position. Poeppel and Wexler found Ihat Ior finite appear as if the empty subject (let us call it PRO) of an O! must have phi lea- verbs Andreas produced about 9% null subjects (17 of 197 utterances), whereas lures, because it is interpreted correctly--lhat is, the etnply subject refers to a fi_r nonfinite verbs he produced about 35% ( 13 of 37 utterances) null subjects. particular person and a particular nuntber. Thus, one possibility is that the PRO subject of ;lit O1 has pJti features, but these leatures do not have to be checked in in Dutch, Haegeman (in press) shows that one child (ilein, 28-37 any way (see Chomsky 1995). months) has 85% null subjects with OIs but 32% null subjects with tensed verbs. Clearly, in German and Dutch, nonfinite verbs (and we rele:r here to loot verbs, Anolher possibilily is that in fact the PRO subject of an Ol does not products of the OI stage) are much more likely Io allow a null subject than are have phi te:llures. Rather, PRO is inlerpreted through the discourse--that is, it is finite verbs. We suggest thai the small nmnber of null subjects with finite verbs co-indexed with somelbing in the discourse and picks up its interpretation reflects a certain amount of Topic-Drop. through Ihis co-indexation. This manner of interpretation is familiar, for example, from the case of certain reflexives. If this PRO does not have phi features, In OI I+rench, it is well-known that nonfinite verbs allow many ,norc ,_, then there is nothing to check, and no case needs to be assigned, null subjects than do finite verbs. Pierce (1989, 1992) was the first to examine such data, and showed that while Ihere are plenty of pronominal subjects with fi- We will not take a position here on the appropriate mechanistn. We nile verbs, there are almost none with nonfinite verbs. We reproduce her table simply wish to point out that it is natural to assume Ihat a representation without Ior four subjects here: "i'IiNSI:. will be one in which the subject can grammatically have a null subject.

If all null subjects in child English were only a result of being licensed by nonlinite verbs, then we would expect to see no null subjects with finite verbs in child corpora. This is probably too strong a conclusion. Although we have not performed a precise count of null subjects given finite and nonfinite verbs, it is

2 Alternatively, it might be Ihat tile subjects of small clauses check off Accusative Case againsl tile higher verb. If some version of this proposal works out, it mighl explain why subjccls of sin;ill clauses cannol be enlpty: enlply NPs do not have Accusative Case. A proposal such us Ihis would be supported by observalions that subjecls of small clauses are Accusative even in languages in which defaull case is Nominative.

225 224 l lilary Sara Brombcr g and Kcnnelh Wcxlcr Wh Null Subjccls

ask why nonlinite TENSE alh)ws null subjects, in other words, we need tt_ a_;- Table I sunte thai nonfinite TENSE licenses null subjects in any case. Number of Finite and Nonfinite Pronominal- and Null-Subject Utterances in Four French Subjects Turning to wh-questions, note thai we expect only one kind _f null , subject to appear with them. Suppose thai some wh-phrase appears in Spec,C Finite Nonfin,ite This implies thai 11oolher phrase can appear in Spec,C. In English, Topics must appear in Spec,C. Therelore, if the wh-phrase is already in Spec,C, the sentence I)aniel I;8, I-I ;1 I, I can contain 11oTopic. Thus, no phrase can be deleted by Topic-Drop. 3 Thus, Ihe kind of null subject which is a result of Topic-Drop will not be possible with wh- Null Subjects 247 273 questions. On the other hand, if a wh -question has a nonfinite verb, then there is i'ronominal Subjects 1114 14 no reason for this nonfinile verb not to license a null subject. So, we do expect null subjects with wh-questions, but only those of the kind that are licensed by Gr6goire I;9,2-2;3,(1 nonfinite verbs. Thus, all null subjects in wt,-questions should have nonfinile Null Subjects 158 108 predicates.

Pronominal Subjects 143 3 Thus far, we have predicted that wh-questions can contain null subjects. Moreover, since we assume thai null subjects in declaratives can be tile result of Nathalie I;9,3-2;3,2 either Topic-Drop or licensing by a nonfinite verb, whereas null subjects in wh- Null Subjects 90 131 questions can only be the result of licensing by a nonfinite verb, we predict that there will be more null subjects in declatatives than in wh-questions. Pronominal Subjects 162 7 A further prediction, somewhat more subtle, concerns the expected out- Philippe 2; 1,3-2;3,0 come of a change in one of the factors responsible for null subjects, l,t particular, Null Subjects 90 34 we know that there is a large reduction over time in the possibility of nonfinite verbs in root sentences. Since we predict that nonfinite verbs license re:my of the Pronomimd Subjects 196 3 null subjects in declaratives, and all of them in wh.questions, we expect that as nonfinite verbs decrease, so will null subjects in both declaratives and wh-ques- (Adapted fro,n Pierce 1992: I I(1) lions. In this study we did not count proportions of finiteness for declarative verbs. However, we predict that proportions of null subjects should follow the As can be seen, there .are abundant pronominal subjects with finite same temporal patterning]orboth declarative sentences atul wh-questions. That verbs- for some of the children, there are even afore pronominal subjects than is, if there is a significant plummeting of null subjects in declarative sentences, null subjects--and crucially, there are ahnosl no pronominal subjects with non- tltere should be a synchronous descent in wh-questions. 4 finite verbs. This is an extremely strong effect. (The reason Ihat the appropriate compa)ison to null subjects is pronominal subjects is that null subjects are as ,,. sunted to appear in the sa,ne interpretive contexts as prono,ninal subjects (e.g., ttyams a,td Wexler 1993). If it is crucial in discourse to specify a lexical subject, then neither ;, null nor a pronominal subjecl would be expected. At any rate, i,t 3 For a suggestion that there are topic-drop null subjects in English children, see ttyams and Wexler (1993). Carson $chillze (p. c.) poinls out Io us Ihat a wh-phtase is usually our results later in this paper, we show comparisons of rates of null subjects to considered to be a Jocus, not a topic, since a topic is old information and a wh-phrase is bath pr_mominal subjects and to pronomi,t;,I and lexical subjects combined.) usually not old information. Thus, we would not expect the wh-phrase itself to delele by Tnpic-Drop. At any _ate, there would be no way for a phrase other than the wh-phrase It) II may be somewhat more difficult to evaluate in French our claim Ill;it delete, since the existence of the wh-phrase in Spec,C prevents there being a lupic in the there are more null subjects will) nonlinile than with finite verbs because French sentence. st,hject pronouns are citrics. It may be, as Pierce proposes, that subject pronouns 4 We rre ignoring in this prediction any &)ssible change in the proportion of topic-drop in F, ench only occur together with agreement, and titus with "tENSE. On the null-subjects; thai is, Ihe prediction is based upon the drop-offin null-subjectslicensed by other baud, French cbihlren, as we have seeu, use a large number of null subjects infinitives, given that OIs drop-oil. One could expect a somewhat mote complicated pat- wilh nonlinite verbs, and sotnething utust alh)w these. If we accepl that the null tern if there weue a drop-oil in lopic-drop null-subjects. At this point we do nol know il subjects a)c licensed by nonfinite TF.NSI-, then we c,m understand how children children use more Iopic-drop null-subjects than do adults. At any rate, we expect that c:m produce sucll utterances If we are I(1_.iSSulneihill the parameter has been thereate far more null-subjeclsof inlinitives than of finite sentences, so thai the maj_t effect over-all will be due to the drop-off in null-subjects licensed by inlinilives. Thus we misscl so dial finite TENSE licenses null subjects in early French, we still must expect the temporal parallelismof declarative and wh null subjects mentioned m the texl. I

226 227 I 1

i lilary Sara Bromberg and Kennelh Wexler Wh Null Subjecls

sent the hick ol TI-NSE. 6 They used this result to argue against "Fru,calio, We Rizzi (1994b) has proposed thai the Optional Infinitive stage is the re- shall replicate this result here will,,additional children. We shall also provide suitof 7"runcatir.w,hich. trteantshatthechildcan starta derivationatsoute comparisonsoftheplummetingsofnullsubjectsinwh-questionisn,ddecliualive point in the functional hierarchy below CP. in particular. Rizzi assumes thai the sentences. 7 Roeper and Rohrbacher's work first led us to tabuhtte the finiteness child is missing only the assuntption thi,I Root = CP -she can start Ibe deriva- ol the verbs of u,ltqueslio,ls with null subjects that we had counted. lion any where it..the hierarchy but once a category is used, all categories which it dominales must also be used Rizzi assumes that null subjects for childre,i in (3) Summary ()f Predictions: non mdl subject languages may occur when the subject is the specifier of a root (as in Diary-Drop; Haegeman 1990). Wh-questions (non-subject) have the speci- it Null subjects will exist in wh-questions. tier (d their roots filled with something other than a subject, namely the wh- phrase. Thus, Truncation predicts that wh-queslions may not have null subjects, " b. There will be proportionally fewer null subjects in wh-questions and preliminary support fi_rthis prediction is presented in Rizzi 1994b. -_ (Of than in decbuatives. _' course, in situ wh-questions would allow null subjects; we will not discuss this possihilily here.) c. The proportion of null subjecls in w/i-questions will pluinmet con-

Alternatively, let us suppose thai OIs in children are not necessarily currently with that in declaratives. ,t,issing a higher CP level. Perhaps just TENSE is missing, as suggested by d. Wh-questions with null subjects will not be tensed (as in Roeper Wexler ( 1992, 1994). Then we could not construe all null subjects in child En- and Rohrbacher 1995). glish ((u other non-null-subject languages) as the (empty) specifier of a root. Thus, we would notconclude that mill suhjects in child English would be impos- ?_ Melhod sible in wit-questions. That is, there should be null subjects in wh-questions. Analyses of the utterances of Adam, Eve, Sarah (lirown 1973), and Thus, we decided to count null subjects of wh-questions in child En- Peter (L. Bloom 1970) from CLIILDES (see MacWhinney & Snow 1985) were glish In see which prediction--Truncution or TF.NSE omission--received fur- perfornled in order to test the above predictions. ther support. Quite simply, in the first case, wh-questions would not have null subjccls; it.. the second case, tbey would. Moreover, if TENSE omission were For the entire Eve corpus ( I ;5,12-2;2,2 I) and for the first 20 files of tbe corlect, wllatever process was responsible fi_Inull subjects should be manilest ill Adaul corpus (2;3;3-3;I);I I), all wh-questions were isolated and examined in or- broth declarative sentences and wh-questions. We know thai null subjects uhi- der to detect the presence of null subjects and of +/-TENSE. In addition, the null ma,ely dintinish In a very low proportion of adult utterances in F.nglish. Thus, subjects in all declarative phrases were examined. We also analyzed the corpora wc decided to count null subjects in declaratives as well. so we could determine of Sarah (2;3.7-4;2,1) and Peter (I;I 1,17-2;8,12) only with respect to Ibe pro- if the pluuHt..eting in declarative null subjects was paralleled by a plummeting m portions of null subjects in wh-questions. 8 wh-question null subjects. The counts for declaratives also allowed us a baseline in _rder I_._examine whether the proportion of null subjects in declarative sen- 2.1 Wh null-subject analysis tenets was similar to that in wh-questions. '' For each file, the fraction of valid w/i-questions that had a null subject Rt_eper and Rohrbacher (1995) counted null subjects in wh-questions was calculated. The number of "possible mlls"--wh-questions with pronominal Ior Atlant's C|IILI)ES data (Brown 1973; see MacWhinney & S,iow 1985), and subjects (e.g., where he goes?)--was counted and added to the number of wh- Iouud that there were many null subjects in these questions, but that they ap- questions with null subjects. This total was used as the denominator in the above pcared only with nonfinite verbs, not with finite verbs, accepting Wexler's

( 1992. 1994) proposal Iha! certain forms in children's I--nglish ulter;tnces repre- 6 Roeper and Rohrbacher provide a theoJybased on Speas 1994, whichwe will lint dis cuss here. 7 Roeper and Rohrbacher did no! docounts of null subjects in declaratives, so they could not compare drop-offs of null suhjecls in wh.questions to those in declaralives or the 5 R_eper (1991) displayed some u,h-qucs,ions with null subjects in English I'rom the over-all nulnber of null suhjec,s in declaratives versus wh-questions. (-'IIII.DES da,abase. This was a suggestive result, bill there were no systematic counts, 8 Tile reason ,hal we did no, do a full analysis of Sarah and Peter is that they had so low and thus n_)way of knowing whclher these represented a granonatical possibility for,lie wh.questions ,hal it wouldhave been impossible In apportiontheir data meaningfullyinto children, or were merely perlnrmance errors Valian (1991) first claimed that there were Imile versus uonlinile veJbs and In compare ,o the proportions in declarativeutterances ;lllUllS,i|(_null subjccls of non-subject wh-questions. We tin not know why there should Wo decided to count ,heir null subjects in wh-questions in order ,o see if they existed at he a tlillcrcllcc between Valian's counts and ores. PcHlaps her subjecls' wh-questions oc - all. ttufc0 mt)s,ly al a la,ea age. when Ihecllildten were pasl the OI stage.

229 228 Itilat-y Sara Brontberg and Kenneth Wexler Wh Null Subjects

3. Results fraction. The assumption is that only a pronominal subject mighl be the prag- malic equivalent of a null subject. 9 3.1 Some telling examples

Excluded from tile analyses were subject wh-questions (e.g., who goes Without further ado, it must be averred that null subjects are abundantly there?), w/t-questions with lexical subjects, and incompletely transcribed, unin- present in wh-questions. Before presenting the numbers and figures which make terpretable, directly repetitive, or ambiguous utterances. All utterance tokens the case clear, we shall put forth a few examples which demonstrate that null rather than merely singular types were included, subjects in wh-questions seem to represent a genuine grammatical option fog young children, rather than a mere perlormance fluke or the result of a few fixed 2.2 Wh finiteness analysis phrases.

As h_r the null-subject analysis, the fraction of wh-questions with in (4), we see two utterances of Adam, one following directly after tile +TENSE verbs was determined for each file. The "possible finite" denominator other in the actual transcript: II included all finite (+TENSE) utterances plus all utterances which should have been tensed but unambiguously were not. These primarily included verbs used in 3rd singular contexts without the 3rd singular marker -s. (e.g., where go?) and (4) a where go? missing copula verbs, (e.g., where going?). (See Wexler 1992, 1994 tot an anal- b. where dis goes? ysis which treats these as missing TENSE.) Utterances were counted as It is clear From this example that there is a null dis in (4a), which is fully ex- +TENSI- if they contained a modal verb or a verb with an overt tense inflection; pressed in (4b), a likely result of tile inflection present in (4b). 12 these included -s, -ed, and irregular finite verbs (finite forms of be and do; also possessive has). Except for the later transcripts, there were not many modal or Other examples demonstrating the same point can be seen in Adam's past-lense furms, corpus. One rather extreme example demonstrating the sheer robustness of null 2.3 Wh null.subject and -TENSE analysis subjects in wh-queslions can be seen in (5):

In order to determine directly the co-occurrence of null subjects and (5) ADAM: How(singing)__ wonderTwinkleyoulittleare.star. -TENSI- in wh-questions, utterances (aheady coded as finite or "possible finite," MOTHER: h's how I wonder what you are. null _t "possible null") were recounted into a 2×2 malrix representing the ADAM: How wonder Mommy are. nullhnm-null and finite/nonfinite di,nensions. Even when singing a ,nemorized ditty and presented with a helpful dose of 2.4 Declarative analysis negative evidence, Adain insists upon omitting the requisite I. 13Is there a gram- In order to make a strong claim about the gramtnaticality of null sub jecis in wh-questions, a direct file-by-file comparison was made between the _' Iraction oi wh-questions with null subjects and the fraction of declarative utter- I I The explanation of (4a) as (ollowing I'roma "performance" limitation, rather than antes with null subiects. Null subjects were counted for declaratives in the same from a state of Adam's grammar,would have to assume that there was some kindot ran- manner its for wh- questions (i.e., as a fraction of null plus pronominal subjects), dora Iluctuation in Adam's performance abilities, so that he could not produce the subject wilh tile addilional codicil that imperatives (e.g., bring me cookie) were ex - in (4a), but immediately ailerward could. The assumption underlying the use of pairs like eluded from the null counts. 10 TENSE/-TENSE was not counted for the (4a, b) to argueagainst performanceaccounts is that such statisticalfluctuationis itnplau sible, ht the particularcase of (4a, b) we also point out that, in addition to the subject be- declaratives in this study, ing added t<_(4b). TENSE is also added; this correlation is exactly whatour account predicts, and that a performance account cannot accommodate. 12 These examples are analogous to the "replace,nent sequences" of Ilyams (1986) demonstrating theexistence of null subjects in declarative sentences. 13 Of course, it is not clear that an utterance like (i) is a wh-question, rather, probably flot. 9 We also coun,ed wh-questions wi,h lexical subjects anti computed ra,es of null subjects (i) how I wonder what you are i_ _.'hquestions out of all wit-questions, including lexical subjects. But it is possible tha,/tow is in Spec,C, which is the relevant point The question here is I1) If the |ctuowd o1the subject o[ a declarative senlence would generate a ph.tse that analogous to the question of whelher topics (e.g., (iia)) in English are in Spec,C; usually cl)uhl he i,_terpletedas being imperative(eg, he bring me]'ood _ bring mefi_od). Ihat they are takento be so. We suggest that how in(lib) is in Spec,C, analogous Io (iia). sculence WilSuo, counted its a "possible null."

231 230 Hilary Sara Brontberg and Kenneth Wexler Wh Null Subjects

In other words, over a period of fewer than 75 days there is a large, malical basis underlying this juvenile recalcitrance? Let us proceed to some systematic drop in the proportion of null subjects in declaratives, trom a point at numbers, in order to tell tbe story more compellingly, which tile majority of the utterances have null subjects to a point at which vely few of them do. This rather quick decrease, after 7 months of recording, shows a 3.2 Null subjects lit wit-questions and declaratives rather remarkable shift, and seenls likely to be indicative of a change in Adam's

Figure I shows two graphs for Adam.14 First, it shows the proportion granunar. We will return to a discussion of this change. of null subjects out of all non-lexical subjects (i.e., null and pronominal subjects) Now let us consider the null subjects in Adam's wh-questions, also iu dechlrative utterances as a function of age in days. Second, it shows the pro- shown in Figure I. First, note that there are null subjects with wh-questions. The txlrtion (it null subjects in wh-questions as a function of age in days. proportions of wh-questions witil null subjects are distinctly higher than zero for Ioo°/o ...... many files. Second, there are somewhat fewer null subjects for wh-questions aoo/. . _ "x_- _x.. a jt .,,ll ..a..." than for declaratives (the non-wh forms); this is evident beyond the e,'uly files ...... (where ceiling effects and sniall numbers of wh-questio'n's conspire to inask the a°°/* . differences; that is, the small number of wh-questions in the early files masks

_ 5OO/.o6q °% I ,,'t ...... number {)vet time. Fourth, note the striking observation that the proportions of ,lo°,...... i i " ...... l...... Third,null subjectsthe proportionover timeofinnulwhl -questionssubjects inseemwh-questionsto paralleldecreasesthe proportotionsa veofry nulllow subjects in declaratives. Although it seems that the drop in the proportion of null I 30% I ft Oaclai, alives It _l k subjects in wh-questions occurs slightly, before that for declaraiives (the shift ap- 20*/. I .__ pears to begin by File 12), the proportions calculated from Files 12 and 13 suffer -,>'- wtl-Oueslio.s I "_"4,, ,it,; _l-_ from a paucity of"possible nulls" (only 7 and II. respectively, while other Files lO°/o l il = x.d" _.,_ .._ • vary from 16 to 55), rendering them unreliable. For this reason, we will consider 0% , , j "_ the shift from frequent to infrequent null subjects to occur, conservatively, trout 800 asp 900 950 1000 105011001 1150 File I I (975 days) to File 19 (I,090 days), a period of 115 days, for both declara- Agoin days lives and wh-questions. In Table 2, we show the fraction of null subjects in Files I I and 19for declaratives and wh-questions. As we will discuss, the fact that the Figure I large plummeting in proportion of null subjects occurs si multaneously for wh- Fr;tciiotl ol Adam's w/t-questions and declarative sentences that have null queslions and declaratives suggests thai there is a common underlying mecha- subjects, calculated with respect to pronominal-subject utterances nism licensing (at least some) null subjects in declaratives and wh-questions. (excluding lexical-subject utterances) "Fable2 As is well known (see, ior exatnple, tlyams and Wexler 1993), Adam has a Fraction of Null Subjects in Declarative Sentences and Wh- profusion I)1null subjects. One striking result is/hat until about 1,025 days (File Questions (Excluding Lexical Subjects) in Files I I and 19 for 14), the proportion ol null subjects in declaralive sentences is quite high and Adam roughly constant, varying nlostly between 80 and 95%. At any rate, there is no ' ' systematic decrease. Between Files 14 and 19 (I,{)25 to 1,090 days), the pro- File Age Declaratives Wh-Questions portion of null subjects in declarative sentences shows a marked and steady decrease, winding up at about 15%. (This fraction stays roughly constant through I I 2;7,26 91% 0891208) 94% (15/16) File 24, tile last file we counted. Because of this, data from Files 21 and higher 19 2;11,21 16% (28/180) 2% (1/54) lilt not reported here.)

in Figure 2, for completeness, we show the proportion of null subjects in declaratives calculated as a fraction of all valid declaratives, that is, including null. pronominal, and lexical subjects. Similarly we show the proportion of null {ii) a. Beans, I like ' subjects in wh-questions as a fraction of all valid wh-questions, including null, b Ih)w youeat! pronominal, and lexical subjects. The overall pattern of results is the same as The point we make ill 15)dellCnds only Oilwhether how is in Spec.C. when we only counted prononiinal subjects, although the fractions are lower 14Ilyams and Wexler (1993) counted every lltilerfile trotAdam. On the files which both when Icxical subjecls are considered (and the problem of low denominators Ior we and Ilyams and Wexler COLimed, we obtainedcomparable figures forthe dcclaralives.

232 233 tlilary Sara Bromberg and Kennelh Wexler Wh Null Subjects

10o*/. Table 3

L l-faction ol Null Subjects in Declarative Sentences and 90°/.. _ oectarativos I Wh-Questions (Excluding Lexical Subjects) for Eve 70%...... File Age Declaratives Wh-Queslions 600/. I I;5,12 88% (15117) -- (0/0) 50%- '_ \_ _ fl ;-"_ " 2 I;5,26 75% (18124) -- (0/0) E 40% "_ i _ ;'.d _/ ! K '

_o. 30% ; ' _'_ _r" I, N_ 3 I;6,9 80% (33141) --(0/0) i i \ 4 I;6,23 89% (40145) -- (0/0)

10o/. "_-_j,-J ...... "_,f 5 I;7,6 80% (24130) 83% (516) o%° _ _ i i _ _'l 6 I;7,27 87% (27/31) 100% (1/I) 8oo 850 9oo 9so Iooo 1o50. 11oo 115o Auoindays 7 I;8,10 64% (30147) -- (0/0) 8 I;8,24 61.% (38162) 100% (515) Figure 2 9 I;9,7 50% (19/38) 09'o (0/2) Fraction ifl' Adaln's wh-questions and declarative sentences that have null subjects, calculaled with respect Io pronominal- and lexical-subject utterances I0 I;9,21 55% (28151) 100% (3/3)

Files 12 and 13 no longer exists---note the coincidence of the drop in declarative I I I; 10,4 17% (9_52) 100% (lid and w/i-question null subjects, as well as the consistently larger proportion of 12 I;10,18 13% (12/93) 0% (0/3) null subjects in declaratives Ih:m in wh-queslions), and we will not discuss these dala;my furlher. 13 I;11,8 19% (25/131) 6.7% (1/15)

I.el us now luru to the analyses of Eve's data. Eve has many fewer wh- 14 I;I 1,22 15% (10166) 14% (2/|4) sentences than does Adam, and a file-by-lile graph of null subjects in wh-sen- 15 2;0,12 12% (20/161) 0% (0/24) tences woultl not be meaningful, because there would be a large number of en- tries based on only one data point. Therefore we will present Eve's data in tabu- 16 2;0,26 6.4% (6/94) 0% (0112) lar lotto. In Table 3 we present the data for declarative and wh-question null 17 2;I,9 I1% (181159) 0% (0/20) subjects. As is well known (see for example tlyauns and Wexler 1993), Eve has many null subjects in declaralives. 15These proportions decrease over time and 18 2;I,23 7.9% (131164) 0% (0/10) ,, seem to show a particularly large drop from File I0 (55% nulls) to File I I (17% 19 2;2,1 10% (131127) 0% (0_6) nulls), alllmugh the drop does not appear as dramatic as Adam's does. 20 2;2,21 10% (5151) 14% (I/7)

As lot wh-questions, in her early stages, before the significant drop in declarative null subjects, Eve does not tnave a large number of wh-questions. Nevertheless, there are significant numbers of null subjects in her early wh-ques lions. Moreover, the null subjects in wh-questions decline dramatically at the same point that null subjects in declaratives plummet to their residual figure, that is, after File I0. The proportion of null subjects in File I through I0 compared It Files I I on are shown in Table 4.

I 5 As lot Adam, Ilyaans aa_d Wexlcr 11993) counted every other file for Eve On Ihe files which both we nnd Ilyaunsand Wcxlercounted, we obtained comparable figures fl)rthe tlt'ci;Ir;lll vex

235 234 itilary Sara Bromberg and Kenneth Wexler Wh Null Subjects

Table 5 Table 4 Fraction of Null Subjects in Declarative Sentences and Fraction of Null Subjects in Declarative Sentences and Wh- Wh-Questions (Including Lexical Subjects) for Eve Questions (Excluding Lexical Subjects) ill Files I-I0 and 11-20 for Eve File Age Declaratives Wh-Questions Declaralives Wh-Questions I I;5,12 44% (15/34) -- (0/0) Files 1-10 70% (272/386) 82% (14117) 2 I;5,26 67% (18/27) -- (0/0) Files 11-20 12% (13111098) 4% (5/112) 3 I;6,9 61% (33/54) -- (0/0) 4 I;6,23 61% (40/66) --.,, (0/0) There-are only 17 wh-questions through File 10, but the fact that 14 of them have null subjects suggests strongly that these are grammatical for Eve. 5 I;7,6 32% (24/75) 83% (5/6) Moreover, tile large decrement (to 4%) from File I I on indicates that there has 6 I;7,27 34% (27/79) 33% (1/3) been a change in Eve's grammar, and this corresponds to the decrement in declarative null subjects. Again, these parallel plummetings for null subjects in 7 I;8,10 43% (30/70) -- (0/0) declaratives and wh-questions suggest that there is a single mechanisnl for (at 8 I;8,24 29% (38/133) 100% (5/5) least some ol) the null subjects in these two classes. Finally, as in Adam, comparison of null-subject proportions in wh-questions and declaratives (examining 9 I;9,7 23% (19181) 0% (0/2) only the later liles, when the numbers for wh-questions are large enough to pro- I0 !;9,21 31% (28191) 75% (3/4) vide reliable proportions) reveals more null subjects tot declaratives than for wh- questions. II I;10,4 12% (9/77) 100% (1/I) 12 I;10,18 9% (121138) 0% (0/8) Agaiu, for completeness, in Table 5 we show the proportion of null sub- jccts m declaratives calculated as a fraction of all valid declaratives, that is, in- 13 I;I 1,8 14% (251182) 5% (1/19) eluding null, pronominal, and le_ical subjects. Similarly we show the proportion 14 I;11,22 12% (10/84) 13% (2/15) of null subjects in wh-questions as a fraction of all valid wh-questions, including n.ll, pronominal, and lexical subjects. Again tile overall pattern of results is the 15 2;0,12 9% (20/222) 0% (0/27) same as when we counted only pronominal subjects, although the fractions are lower when lexical subjects are considered, and again we will not discuss these 16 2;0,26 4% (6/136) 0% (0/23) data:my further. 17 2;I,9 9% (18/194) 0% (0/27) 18 2;I,23 6% (131214) 0% (0/14) * 19 2;2,1 9% (131143) 0% (0/11) 20 2;2,21 9% (5/57) 14% (I/7) II

Let us now turn to Sarah's data. We broke up the analysis of null subjects in wh-questions for Sarah into 2 time periods (Files 1-37, age 2;3,7 to 3;0,18 and Files 38-90, age 3;0,27 to 4;2,1). We selected File 37 because that is where there seemed to be a large drop-off in null subjects, and File 90 is the oc- casion of Sarah's last null subject (she produced no null subjects in her 166 wh- sentences from File 91 through File 135 (age 4;2,9 through 5;I ,6)). For Files 1- 37 Sarah had 12 null subjects out of 29 total wh-questions (lenical subjects were included in additio,i to pronominal subjects) (41%) and for Files 38-90, she had 8 null subjects out of 96 wh-questions (8%).

236 237 Hilary Sara Bromberg and Kenneth Wexler W// Null Subjects

Wh-questions in the early stage are thus very inlrequeut for Sarah. The i00%- .3 j .: j .: z

resultthe s=nashllowingnumbethrs,at ntheot rconce arleusive.null subjects16 at this stage is suggestive, but given 90%- ""_a"'J_i f"q/ .--.fl- Tensed% ]1 80%- _31 Ii -J'- Null % I Turning to Peter, we also separated his data on null subjects into two 70%- tI time periods at the point where there appeared to be a decrement in null subjects 6o./.. t ...... ill wh-questions. The first period consists of Files I-5 (I;I 1,17-2;2,13) and the i " second of Files 6-13 (2;3,3-2;8,12). in the first period, Peter had 7 null subjects 5o*,,.- . _¢/A 'J out of 22 wh-questions (including lexical subjects) (32%) and in the second, he 40°/° had 3 null subjects out of 84 wh-questions (4%). Thus, for Peter, as for tile other 3o./. children, ,lull subjects do exist in wh-questions, and there is a decrement over i time, which may he rather sharp. For Peter, as for Sarah, there are very few wh- 2°%-

data,qUestino°tns's°thatthedata'wlby itsel,' definitive.)ilesuggestiveandc°nsistentwith17 Adan'andEve'Sis ,0-/.0./. , .._//_ ..... I ', ..... _ _a"b","a 800 8so 9oo 95o tooo toso t too 1t50 We can conclude that all told, the data show that there are a significant Ageindays ,umber of null subjects in w/t-questions and that they plummet in a parallel lash=on with null subjects in declaratives (with residual null subjects higher for declarlJtives than for wh-questions). Figure 3 Fraction of Adam's wh-questions that are tensed and have null subjects. 3.3 Finiteness a,d null subjects in wh-questions calculated with respect to pronominal subject (excluding lexical-subject utterances). We now turn to the analysis of data from Adam and Eve concerning whether the null subjects in wh-questions essentially appear only in nonfinite number in untensed senleuces (I 18/249, or 47%). Essentially, tensed sentences contexts, which is our hypotbesis. Let us start with Adam. never have a null subject and untensed sentences do a significant portion of the time. First, in Figure 3 we see that the proportion of finite (tensed) wh-queslions increases over time, 18 its wl_expect in the O! stage (Wexler 1994). At the Table 6 same time, the overall proportion of null subjects decreases. There is a strong Finiteness of Null and Pronominal Subjects in negative correlation (r = -.60. p < .004) between the proportion of finite and Adam's Wh-Questions null-subject whquestions over time. Finite Nonfinite .,, In order to deter=nine whether there is an exact causal relation between Null 2 ! 18 finiteness and null subjects, we need to look more closely at the relationship of these two variables within sentences. Table 6 shows the numbers of null and Pronominal 117 131 pronomiual subjects for fiuite and nonfinite verbs. Note the extremely small nu=uber of null subjects in teased se=,ences (2/I 19. or 2%) compared to the We wanted to ensure that this result was not an artifact of tensed verbs ...... increasing in later files as null subjects decrease, but rather reflected an actual 16 As we mentioned, given the small numbers el wh questions in the early period, we did causal relationship between these factors. Thus, we cleaved Adam's data into not undertake the large effortof analyzing Sar:lh'sdeclarative sentences for null subjects, two time periods between Files 14and 15, in the middle of the drop from high to To supporta claim that null subjects existed fordeclaratives but not for wh-questious, one low null-subject frequency. In the first period he had extremely few tensed wh- would have to show thai from lile 38 (3;0.27) ou a large proportion of declaratives had questions. These data are displayed in Table 7. Crucially, even in the early (2;3 null sul)jecfs, which seems unlikely, to 2;9) stage, when Adam hardly used tense, the tensed wh-questions ahnost 17As lOTSarilh. we did not calculate =henull subjects in declaratives tOTPeter. To make never had null subjects, while almost all the null subjects occurred in untensed ;Lcase =11;|=he= reare no null subjects in wll-queslions loTPeter. hut that the_e are 1o= utterances. Similarly for the secoud (2;10--3;0) period, when tense was used tlcclar_dives, olle w{mld h;ive to show a I;,ge proportion el declarative uull subjects in the more frequently, all of the finite utterances had a non-null subject, while all of 'iccolvl pcril)dI_l I'c=cr. the null subjects occurred in nonfinite utterances. The data are quite striking; fi 18The t;irgc50%Iigurc h)r=helira=file represe,ts 4 u=tcvaucesand thelehu-c slmuld lit)= nile tense prevents a null subject in a wh-question for Adam, but nonfinite teu,,ie I)C I;|kcl| ;is lnc;itlil|glui.

239 2 38 tlilary Sara Brombcrg and Kennelh Wexler Wh Null Subjects

allows null subjecls. The corresponding data for Eve are presented ill Tables 8 subiects out of 3 tensed wh-queslions, but ,'ill of her 14 untensed wh-queslions and 9. have null subjects. Although the absolule numbers of wh-questious are small, the almost categorical results strongly support the conclusion thai TENSF. prevenls Table 7 null subjects and nonfinite sentences allow them. The same result holds for Files Finiteness of Null- and Pronominal-Subject I I to 20 ( I null subject of 41 finite utterances and 4 null subjects of 63 nonfinite Wh-Questions in Files 1-14 and 15-20 of Adam utterances), although the numbers of null subjects are very small. Crucially, we still have a difference: 2% of her tensed wh-questions have null subjects, and 6% Finite Nonfinite null subjects for unlensed senlences--i.e., 3 times as high a proporlion of untensed wh-questions are null as of lensed wh-questions--and 80% of her null- Files I- 14 subject utterances are nonfinite. 19 Null 2 90 Clearly, lensed wh-queslions have a far smaller proportion of null sub- Prononzinal 21 9 jects than do untensed wh-queslions. Moreover, Ibis result is nol due Io an arli- Files 15-211 Iacl from Iwo v;triables changing as a result of increasing age. Ralher, we have to conclude lor Adam and Eve that tensed wh-questions allow null subjecls Null 0 28 much less frequently titan do untensed questions. Pronominal 96 122 , , Recall that for Peter and Sarah we did not tabulate the wh-questions with non-null subjects, so we cannot calculate a base rate of tensing with non- Table 8 null subjects. Thus, we will just look at their null-subject wh-questions to see Finiteness of Null- and Pronominal Subject how many of lbem were tensed, recalling that we expect the proportion to be WhQuestions in Eve close to zero. II

Finite Nor_nite Of Peter's I0 wh-queslions wilh null subjects, i¢is unclear phonelically Null I 18 for one of them whether it is lensed. The other 9 are all untensed. Clearly, Peler's data (100% untensed)conforms Io prediclion (3d). No mailer how much Pronominal 43 59 Peler is tensing wh-queslions wilh non-null subjects, it is unambiguous Ihal Ihe rate lot null subjecls is smaller, since it is zero. Table 9 I"ioileness of Null- and Pronominal-Subject Sarah had 20 wh-questions with null subjects. 15of these had untensed Wh-Queslions in Files I-I0 and 11-20 of Eve verbs, 3 had tensed verbs and 2 had the triodal should, which normally would be ...... considered finite. The 3 finite examples a_e given in (6): Finite Non, nile -' Files 1-10 Null 0 14 Prortomirlol 3 0 Files I 1-20 19It is unclear why there is such a small percentageof null subjects for nonfinite verbs in wh-questions for Files 11-20 of Eve. If nonfinile verbs license null subjects to se.lch= Null I 4 I_rgeextent i. the earlier files, whydo they license them to such a smaller extent in latc_ files? Perhapsthe kinds of verbsthat we considered to be nonGnite in Eve included somt I_ronontirtr# 40 59 that were finite in some basic sense; thai is, perhapssome kinds of verbsthatappear to b_ nonlinile are some kind of surluceomission that we should not count as nonfinile. I)cr- Overall, Eve produces a null subjecl only once out of 44 (2%) lensed haps this type of verb increases in the later files. Another possibility is that, ahhougl wh =sentences, but she produces 18 null subjects out of 77 (23%) unlensed wit- grammatically licensed, the null subjects of nonfinileroot verbs containa pragmatic vi.- latiou. The older child may have matured wilh respect to this pragmatic contpetc.ce sentences. 'I'ENSI'."clearly preve.is null subjects, whereas nonfinite sentences al - Note that the percentage of null subjectsof nonfinite verbsalso decreases in the later lilt.* low Ihem. Ahln)ugll the numbers a,e small wbe,i we break Eve down into two of Adam, though not to tie:tOyso low a level as Eve. We will leave the issue here, awaJt- I)criotls, we see rba( the eflecl still holds, hi Files I-I0. Eve produces no null tug lurtherlesearch

240 241 Hilary Sara Bromberg anti Kenneth Wexler Wit Null Subjects

gtanlmalically licenses a null subject. That is, there exists all iml)t>lta,ll type t_l (6) a when rains null subject which is not licensed because it is in a Specifier ol Root position, _l h. where sees because it is a Topic, but liecause the verb is nonfinite. As we discussed in Sec- t. when saw Chantilly 21) lion I, such an assumption predicts tile results in (7).

For Sarah, as for Peter, we do not have counts of TENSE in wh-ques- All important issue that remains is the status of null subjects of finite titus with non-null subjects, so we cannot compare the proportion of TENSE in verbs. They only occur, as we stated, in declaratives. We assume that they repre- wh-questions with null subjects'with the proportion of TENSE in declaratives, sent a kind of Topic-Drop, as suggested by Hyams and Wexler (1993) lot all However, the proportion in wh-questions is small enough, and the overall num- null subjects (but here we suggest this in particular for the null subjects of finite bet of examples small enough, such that Sarah's data are in general consistent verbs). Since we assume that a Topic ,nust be in Spec,CP position, if the sen- with the picture drawn from the analysis of the nmch more informative files of fence is a wh-question, there can be no other topic in Spec,CP. That is, by dell,It- Adam and Eve. Certainly Sarah's files do not provide ally evidence against the lion, we cannot have a sentence with the same position occupied by both a wh- prediction; the dala are simply too scanty to be conclusive, form and another topic thai can be dropped. Thus, we predict that even if children have a Topic-Drop process resulting in null subjdets with finite verbs, this 4. Discussion process will not operate in wh-questions. Thus, there will be no null subjects of The data from the 4 children confirm the predictions we made in (3): finite wh-questions. Indeed, much evidence exists to support the claim that there is a Topic- (7) Sumnmry of Results: Drop process in young children in the OI stage, in languages like Dutch and German, which are not null-subject languages, children drop Topics, but not a There are many null subjects ill wh-quesnons (true for all 4 chil- subjects which are not in Topic position. See Wexler 1994 for a review of sonic dren), of this evidence (in particular de Haan and Tuijnman 1988). Of course, the aduh languages in those cases also have a Topic-Drop process (of somewhat diflerent h. There are more null subjects in declaratives than in wh-questions strengths in the two languages). Does English have such a process? Not one st) (data examined for only 2 children, clearly true of those), clearly manifest, though one could perhaps find examples to support a weak process. That is, it is well-known that even English can drop its subjects when c. Null subjects in wh-questions plununet concurrently with null sub- they are topics, as in the (Sb) rejoinder to (Sa), or in diary contexts exemplified jects in declaratives (data examined for only 2 children, clearly in (9). true of those).

d Wh-questions with null subjects almost always have nonfinite (8) it what happened It Mary? verbs (clearly and strongly true for Adam and Eve, completely b. _ went away for a while. true for Peter and inostly true for Sarah, despite scanty data). (9) a __ pulled a piece of skin off my lip last night. Note that the notion that truncation leads It the OI stage is not sup- b. _ felt a joy yesterday, _ soon clouded. ported by these data, since there are many null subjects in wh-questions (Ta). S. Plath, Journals (McCullough 1982) " Moreover, tile theory of truncation does not predict an interaction between It may be the case that children in English are perfor,ning Topic-Drop TENSE and null subjects in wh-questions (Td). Something would have to be with finite sentences in a way which the adult language allows, except that the _,dded to that theory to derive the correct prediction. The same is true of prediclion (7c); tile notion of truncation--or in fact, the notion that null subjects in children may be performing it in a wider range of contexts, including those infe- chihhe,I speaking non-null-subject languages are in specifiers of roots--will not licitous for adults. If indeed this is true, we will be able to reduce the differences predict a,ly relation between the pattern of null subjects in declaratives and wh- between English-speaking children in the OI stage and adults to the following: questio,ls.

The predictions we made (3), supported in (7), Iollow from a theory ill which TI'NSE is o|niUed in OI sentences and ill which the nonfinite sentence

21)Ex;unl)lc (6c) is Irom whenS;_ralwl asalrc;,dy4 yearsohl;it mostlikely represenlsa perl(_rmanccthor.

2,13 242 ltihtry Sara Brombcrg and Kenneth Wexler Wh Null Subjects

Even more strikingly, note that OOM (or ally model that assumes that (1(1) I:urlher Predictions: subjects are dropped under memory pressure from other morphemes) predicts

a The OI stage allows nonfiuite verbs where the adult requires finite " Thusthat tile, whenmoreTENSEelementsis athereddedareto ina vea rVP,b, itIheshouldmoreinakelikelyititmoreis to likelyomit thethatsubject.a null verbs (and thus null subjects of these nonfinile verbs are allowed subject appears than if TENSE is omitted We have demonstraled fl_r wh-qucs- as a matterof grammaticalprinciple), lions that exactly the opposile result holds; null subjects are impossible wilh b. Young children are less discriminating than adults as to the variety lensed wh-questions, and quite likely with untensed ones, directly contradicting the predictions of OOM. As Hyams and Wexler (1993) pointed out, the of pragmatic contexts in which they will allow null topics. "grammatical" account of null subjects makes a number of correct empirical lu short, most of Ihe properties of null subjects are simply a reflection predictions which are impossible for the "performance" account to match; we of the OI stage; the residue is inleresting with regard to pragmatic considera- have added to these. lions. Clearly, further analysis of these residual cases would be useful. 5, Conclusion Why do null subjects plummet at the same point of time in declaratives and wh-questions (7c)? We attribute this, as we have stated, to the withering Much of the null-subject phenomenon in early child language (when away of the OI stage. When root nonfinite verbs become very infrequent, the the adult language is not a null-subject language) is due to the existence of the null subjects licensed by nonfinite verbs will be eliminated, both in declaratives O! stage, to ils untensed sentences. These distinctive properties of early child and wh-questions. Thus, we expect a synchronicity in the slunrpings of null sub- becomeslanguage moremay thusunderstbeainlegrndableatedwheninto wea unifiedknow thatsystem.infinitivesThe NulllicenseSubjectnull stagesub- jecls in declaratives and wh-questions, jects. Such integration from the study of child language may provide important The question that remains is: why does the child ever exit the OI stage? insights for syntactic Iheory in general, a welcome result. That is, why does TENSE become obligalory in root contexts? The only answer thai has been proposed Io dale is the maturatioual one, as suggested by Wexler Moreover, we have evidence that the null-subject parameter has tuft ( 1992, 1994) and in nmch of tile literature that follows. Thus, the decline of null been misset in early English (or German, French, Dutch, and other non-null- subjects will be the result of the maturation of the requirement that TENSE is subject hmguages). English-speaking children in the OI stage know that their llbligatory. 21 censedlanguageby isAGR.not aWenull-subjectwill not reviewlanguage,the evidencethat is, onehere,wherebut nullit appearssubjectsthatateItal-li- Refore we conclude we should point out the ramifications of our data Jan-speaking children know that their language is a null-subject language (for and analyses Ior what Ityams and Wexler (1993) called Output Omission Mod- example, wh-queslions with finite verbs and null subjects are allowed (Rizzi els (()OM) lif null subjects in early English. For example, P. Bloom (1990), fol- 1994b)); also, there are more null subjects in child Italian than in child English lowing k. Bloom (1970), proposed that subjects are omitted in early speech in (Valian 1991)). Thus, there is evidence that at quite an early age, Italian chihlren English because children's memory was limited and they could not produce all have set the null-subject parameter to "yes" and English-speaking children have the elements _fftheir representations. Hyams and Wexler (1993) presented much set it to "no." ,. evidence against this model. Our data adds yet further arguments against OOM. First, as ityams and Wexler pointed out, the existence of null subjects with wh- The mall-subject parameter thus interpreted is consistent with Wexler questions argues against the assumplion of OOM that it is elements in]irst post- (1995), who proposes that Very Early Parameter-Setting (VEPS) holds, arguing lion in a sentence that are mostly omitted, because of processing difficulties with that by the very earliest observed stages, (that is, by the beginning of the produc- first position. We have demonstrated the existence of a large number of null sub- clauselion of structuremulti-wordparametersutterances,havearoundbeen correctl18 months),y set. bTheseasic inflectionalinclude theandV/orll.i jeers with wh-questions; thus, the assumption of OOM is undermined. TENSE parameter, the V to C parameter, word order parameters, and, as we argue in this paper, the null-subject licensing (AGR) parameter.

.... Recent work on language acquisition reveals a very different picture ol 21 N_liCeIhal wherelense is uou-linile in adult I:.nglish,we can observe residual cases dt_l tile Null Subject stage than had previously been put forth. The discovery ol Ihe wh-qucstillns wilh null-stilljects: existence of the Optional Infinitive stage has led to a broader perspective on tile (i) a Whale to d_)? interrelationships between grammatical processes. Young children know a greal b *What e am It)do? deal not only about Universal Grammar but also about particular parameler-sct- c What am I to dt)'? Ihllh linilc _uldnon-linile verbs _lreall(iwcd, bul only Ihe nonfinite case zllows {in Ihis t-;t_t.,rctluire_,)a null-suhjecl. ()theirlanguages have thisprocess to an eveu I_ilgcrexlctlt.

245 2,14 ltilary Sara Brombctg and Kcnne(h Wexler Wh Null Subjects

tings. 22 They are also hosts to other distinctive characteristics, such as the prop- -- (199(ed4),b). L.ngl-arlyuagenullucsubjectsquirilion andstudrooties nullin genesubjecratitvse hrgraT.mmIloekst_aar, Amsterdam:& II SchwBmcn-tz erties of the OI stage. Elaboration and extension of these characteristics will be jamins, 249-272. crucial to subsequent research. Roeper. T. (1991). Why a thenry of triggers supports the PRO-drop analysis. Ms.. Lhti- versily ol Massachusetts, Amherst. References . Roeper, T. & B. Rohrbacher (1995). Null subjects in early child English and the theory of economy of projection. Ms., University of Massachusetts, Amherst, a,ld Uni- Ilh)om, l.. (1970). lxmgnage det,eh,l.nent: Form and function in emerging grammars versity of Pennsylvania. Cambridge, MA: MIT Press. Sand, T. & N. |lyams (1994). Agreement, finiteness, and the development of t_ull argu- Bloom, P. (1990). Subjectless sentences in child language. Linguistic Inq,riry 21(4), 491 - ments. In Merc_: Gonz_lez ted.), NELS 24. Proceedings o[rhe Narth i'.'a_tLm- 504 guistir" Societ),, GLSA: University of Massachusetts, Amherst, 543-558. Ih_ow., R. (1973). A first language: The early stages. Cambridge, MA: Harvard Uniter- Spcas. M. (1994). Null arguments in a theory of economy of projection. University oJ sity Press. Ma._sachusens Occasional Pal,err i,r l.inguistic$ 17. Chonrsky, N. (1995). Chapter 4. Ms., MIT. Valian, V (1991). Syntactic subjects in the early speech of Amd_ican and Italian children. Chomsky. N. & II. Lasnik (1991). Principles a,ld parameters theory. To appear in J. Cognition 40, 21-81. lacobs, A. van Stechnw, W. Sternefeld & T. Vennemann reds.), Syntax: An in- Wexler. K. (1992). Optional infinitives, head movement and the economy of derivation in lernatmnal handbook of contemporary research, Berlin: Walter de Gruylcr. child grammar. Occasional paper #45, Center for Cognitive Science, MIT. de Ilaan, G. J. & K. Tuij,nnao (1988). Missing subjects and objects in child grammar, lit -- (1994). Optional infinitives, head movement and the economy of derivations, lit P Jordcns & J. Lalleman reds.), Language Development, Dordrecht: Forts, D. Lightfoot & N. Ilornstein reds.), Verb Movement, Cambridge: Cambridge 11)1-122. University Press, 305-350. Ilacgeman, I.. (1990). Underst(_ subjects itl English diaries. Mid#lingua 9(2), 157-199. -- (1995). Feature-interpretability a,rd optionality in early child grammar. Paper pre- -- (in press). Root inlinitives, tense and tluncated structures in Dutch. LangluJge Ac- seated at the Workshop on Oplionality, Utrecht. quisition. Ilams, A. & K. Wcxlcr (in press). The Optional Infinitive stage in child English: Evi- (Bromberg) dence from ,legation. In II. Clahsen ted.), St,rdies in the acquisition oJinflec- Department of Psychology, Harvard University tr,,n. Willia,n James Hall Ilya,ns, N. (1980). Language acq,asitmn and the theory of parameters. Dordrecht: Rei- 33 Kirkland Street del. Cantbridge, MA 02138 tlyams. N & K. Wexler (1993). On the grammatical basis of null subjects in child lan- USA gauge i.iug,,stic h_quiry 24(3), 412--459. MacWhiimcy. B. & C. Snow (1985). The child language data exchange system. Journal of ('hihl l_mguage 12, 271-296. hilary@ isr. harvard, edu McCulltaJgh, F. (1982). Thejournuls o[Svlvia Plath. New York: Dial Press, Pierce, A (1989). On the emergence of syntax: A crosslinguistic study. Ph.D. disserta- (Wexler) lion, Department of Brain & Cognitive Sciences, MIT. Department of Brain and Cognitive Sciences -- (1992). lxingnage acquisitio, and syntactic theory: A ('onlparative analysis o.I Department of Linguistics and Pllilosophy Frent'h und Engli*h child grummars. Dordrccht: Kluwer. E 10-020, MIT ,'4' i't_el_pcl, D. & K Wexler (1993). The full co,npetence hypothesis of clause structure in Cambridge, MA 02139 child German. Language 69, 1-33. USA Rice, M. K. Wexler, & L. I. Cleave (in press). Specific language itnpairment as a period ol extended optional infioitive. Journal of Speech and Ilearing Research wexler@ps)'che.nlit.edu Rizzi, 1. (1982). Issue._ in Italian synta.r. Dordrechl: Forts. 1994a). Some notes on linguistic theory attd language development: The case of ro_l i,dinitives. Language Acquisition 3(4), 371-393.

22 A,t uuportant question is the extent to which Vi:.PS is true If it is generally true, it migh! have consequences for the Ihet_ry ol piuauleter-setling, its WexleJ- (1995) points ,t_ul lu p_nticulaf, to the extent that VEPS is tt-ue, the theory of parameter-setting inust be such Ill;it the parameter-setting nlechimisln does its work before a child actually produces the tlttcrances which rellcct Ihe parameter; the pltramelers ate learned "perceptually"; and the child's (produced) errors are not part ol the paraMeter-setting process.

247 246