Children - Architects or Brickies?

Andrew Radford, University of Essex, 1994

1. Introduction

There are two broad approaches to the study of the acquisition of syntactic structure which are found in the acquisition literature. One approach (the structural continuity model) assumes that UG provides the child with a 'template' which specifies the (universal) structure of phrases and clauses. The core assumption underlying this approach is that adult phrases and sentences and their child counterparts have the same syntactic structure, and differ only in respect of their phonetic form, in that certain morphemes/lexemes which are overtly realized in adult sentences have a null realization in child grammars. On this view, we might say that UG provides the child with the basic scaffolding which underpins syntactic structure, and children are mere brickies (i.e. brick-layers) who have to determine which (morpholexical) bricks to position in which parts of the structure.

The second approach (the structure-building model) posits that although UG principles determine how they are built up, syntactic structures are projections of the lexical items they contain, and lexical items vary from one language to another (e.g. French has clitic pronouns which might be argued to project into Clitic Phrases, whereas English lack clitic pronouns and so has no Clitic Phrase constituents). If we make the traditional assumption that children acquire some types of lexical item before others (e.g. verbs before auxiliaries, nouns before determiners), it follows from the lexical projection view that children will gradually build up syntactic structures 'one layer at a time' (in that acquiring a new type of item will lead to the formation of a new type of projection): in other words, the lexical projection view presupposes a structure-building account of acquisition. On this view, children are more than mere brickies: they are also architects, in that they have to design the structures that they build (while ensuring that the relevant structures conform to UG building regulations). In this paper, I shall adopt a specific version of the structure-building approach, and posit that syntactic structures are minimal lexical projections (MLPs) – i.e. are the minimal syntactic projections of the lexical items they contain.

To put the discussion on a concrete footing, let's consider the nature of the earliest clause structures produced by one-year-old children acquiring English as their L1 (for ease of reference, I will refer to these as Children's Initial Clauses, or CICs). At the stage in question (which children typically go through somewhere between the ages of 1;6 and 2;0), the verbal clauses produced by English-acquiring children are headed by a nonfinite verb, as in (1) below (the name and age in years and months of the child producing each utterance is given in parentheses):

(1) Mummy doing dinner (Daniel 1;10)

Wayne taken bubble (Daniel 1;9)

Machine make noise (Kathryn 1;9, from Bloom 1970)

Under the analysis proposed here, clauses like those in (1) are Small Clause (= SC) constituents which are simple projections of a head nonfinite lexical V constituent, and thus have the structure (2) below:

(2) VP

N V'

V N

Mummy doing dinner

Wayne taken bubble

Machine make noise

An analysis along essentially these lines is suggested in Radford 1986, 1990; Lebeaux 1987; Guilfoyle and Noonan 1988; Kazman 1988; and Vainikka 1993.

Under the MLP analysis, children's initial clauses are VPs which (for the most part) are direct projections of argument structure. It follows from this assumption that early child clauses have no functional architecture, and thus lack IP and CP projections. Evidence for the nonprojection of IP comes from the fact that children make no use of infinitival to or auxiliaries at this stage; for example, in contexts where adults require an IP headed by infinitival to (with an overt or covert subject) children use a simple VP, as illustrated by the following examples taken from a longitudinal study of a boy called Daniel:

(3) Want [have money] (1;7) Want [open door] (1;8)

Want [Dolly talk] (1;9) Want [Teddy drink] (1;7)

Likewise, correponding to adult negative clauses containing the finite auxiliary do, children typically produce auxiliariless negative structures such as those in (4) below (from the Nina files on the CHILDES data base):

(4) No my play my puppet (2;0;2)

No lamb have it (2;0;3)

No dog stay in the room (2;1;2)

No Leila have a turn (2;1;3)

Evidence for the nonprojection of a CP constituent in CICs comes from the fact that children's earliest complement clauses such as those bracketed in (5) below lack complementizers:

(5) Want [Baby talking] (Hayley 1;8)

Want [Mummy come] (Jem 1;9)

Want [lady open it] (Daniel 1;10)

and from the fact that they omit complementizers on sentence-repetition tasks (cf. Phinney 1981). Further evidence comes from the fact that they do not produce structures containing inverted auxiliaries; for example, the earliest yes-no questions produced by Claire (a child studied longitudinally by Hill 1983) at age 2;0-2;1 were of the form (6) below:

(6) Chair go? Kitty go? Car go? Jane go home? Mummy gone?

The adult counterpart of Jane go home? would be Did Jane go home?, with the preposed auxiliary did occupying the head C position of CP. The fact that children's clauses do not contain complementizers or inverted auxiliaries at this stage is consistent with the hypothesis they are producing Small Clause (VP) structures which lack CP and IP projections.

Since it first appeared, the Small Clause Hypothesis (SCH) has been attacked by a number of linguists on three main grounds – viz. (i) lack of descriptive adequacy (in that SCH fails to provide an adequate account of the morphosyntax of particular phenomena in early child grammars – e.g. interrogatives, negatives, null arguments, case, etc.); (ii) lack of universality (in that SCH is incompatible with known facts about the acquisition of languages with a more complex morphosyntax than English), and (iii) lack of explanatory adequacy (in that SCH presupposes an essential discontinuity between adult and child grammars, and fails to explain why early child clauses should be 'smaller' than adult clauses, or how children's 'small clause' VPs develop into adult 'full clause' CPs. In each of the sections below, I answer specific criticisms which have been levelled at SCH. For the most part, the discussion here relates to the L1 acquisition of English, though section 6 looks at acquisition from a wider perspective.

2. Null subjects

A well-known characteristic of early grammars is that between a third and a half of CICs have null subjects (as noted e.g. in Hyams 1986 and much subsequent work). Typical null subject CICs are illustrated in (7) below:

(7) Want one. Gone out. Coming to rubbish (Bethan 1;8)

Want mummy come. Pee in potty (Jem 1;9)

Find Mommy. Taste cereal. Close...door. Go house. Sit lap (Kendall 1;10, from

Bowerman 1973)

How can the VP analysis account for the occurrence of null subjects in CICs?

Luigi Rizzi (1994a) draws an important distinction between two different types of null subject, namely pro and nc (= null constant). Pro is the kind of null subject found in languages like Spanish, Italian, Romanian etc. and is morposyntactically licensed and identified (viz. licensed through case and identified via a rich set of agreement inflections carried by finite verbs). It would seem implausible to suggest that the null subject in child sentences such as (7) is pro, since verbs in CICs are not inflected for agreement. A second type of null subject discussed by Rizzi is the null constant (nc). Rizzi argues that null constants represent a type of null definite description which must be A-bar bound by a non-quantificational operator, but that binding requirements hold only if they are satisfiable in principle. It then follows that nc will be exempt from the binding requirement only where it occupies a position where it cannot be so bound (i.e. where it cannot in principle have a c-commanding identifier) – i.e. only in a root specifier position.

If (following Rizzi), we argue that null subjects in children's clauses like those in (7) are null constants, then it follows that a child null subject sentence such as Want one will have a structure along the lines of (8) below (assuming SCH):

(8) VP

nc V'

V N

want one

If (as here) the overall clause is analysed as a VP, the null subject (= nc) will be in spec-VP and thus occupy a root specifier position. It therefore follows that it can be discourse-identified (by virtue of the fact that it has no c-commanding identifier). Thus, the VP analysis provides a relatively straightforward account of the syntax of null subjects in CICs.

3. Case

Subjects in CICs typically surface with objective case as noted e.g. by Radford 1986, pp. 20-22, Kazman 1988, pp. 11-12, and Aldridge 1989, pp. 82-5. Typical examples (from Radford 1990, pp.175-6) are given below:

(9) Me got bean (Stefan 1;5)

Me want it (Bethan 1;8)

Him swimming (Bethan 1;9)

Her gone in there (Angharad 1;10)

How are we to account for the case-marking of subjects under SCH? Although it has been suggested in the literature that the subjects of CICs are assigned structural case (cf. Vainikka 1993) or inherent case (cf. Budwig 1989), the analysis I shall suggest here is that the subjects of CICs receive default case: more particularly, we might suppose that by virtue of being arguments they must be made visible through case, and that when they occupy a position which is not inherently case-marked, they receive default (objective) case (cf. the proposal to this effect made by Roeper and de Villiers 1991, pp. 8-10). It might be argued that a default case account involves no developmental discontinuity in that objective case also serves a default function in adult grammars: e.g. in contexts where a (pro)nominal receives neither nominative case, nor genitive case, nor null case, objective case is assigned by default (so accounting for the objective case carried by pronominal sentence fragments (cf. 'Who did it? – Me') and by the subjects of 'Mad Magazine' sentences (cf. 'What? Me cheat on you? Never!').

4. Negation

The earliest type of (non-anaphoric) negative structures produced by young children are typically presubject negatives, as illustrated by the examples in (10) below (from Pierce 1989 pp. 93-4, and Déprez and Pierce 1994, p. 61):

(10) Not Fraser read it (Eve 1;9)

No the sun shining (Adam 2;4)

No Mommy doing. David turn (Nina 2;0;2)

No lamb have it (Nina 2;0;3)

No dog stay in the room (Nina 2;1;2)

No Leila have a turn (Nina 2;1;3)

Déprez and Pierce (1994, p. 61) report that of the earliest negative sentences produced by Eve at ages 18-21 months, Peter at 23-25 months and Nina at age 23-25 months, 96% (71/74) contained sentence-initial negatives.

Nina Hyams (1992, p. 378) claims that 'There does not seem to be any easy way of accommodating external negation into the small clause analysis.' On the contrary, however, the SC analysis permits us to arrive at a straightforward analysis under which the syntax of early negatives is determined by UG principles. Let us suppose (following Kazman 1988, p. 17, Guilfoyle and Noonan 1988, p.37, Lebeaux 1988, p. 39 and Radford 1990 p. 154) that negatives are generated as VP-adjuncts, so that a sentence such as No Fraser sharpen it will have the structure (11) below:

(11) VP

NEG VP

No N V'

Fraser V PRN

sharpen it

One of the UG principles which 'guides' the child to analyse negatives as clausal adjuncts is the Isomorphism Principle posited by Hyams (1986, p. 162) which specifies that grammars maximize the structural isomorphism between different levels of representation. Hyams herself (ibid.) notes that this principle will account for the preclausal position of negation in CICs: if we assume that the child knows that negatives must have scope over (and hence c-command) the entire clause at LF, then it follows that the Isomorphism Principle will determine that negatives occupy preclausal position in the overt syntax as well, in order to maximize the isomorphism between LF and PF.

However, a second UG principle which plays a central role in determining the locus of negation is the Minimal Projection Principle (= MPP), which might be characterized informally as in (12) below:

(12) Minimal Projection Principle (MPP)

Syntactic representations are the minimal projections of the lexical items they contain

which are consistent with grammatical and lexical requirements

The MPP (12) is related to a number of conditions proposed in the literature. These include Chomsky's (1989) Economy Principle, Grimshaw's (1993) Minimal Projection Principle, Safir's (1993) Structural Economy Principle, Boskovic's (1993) Minimal Structure Principle, and the principle of Economy of Projection proposed in Speas 1994. One effect of this principle is to minimize extended projections (IP and CP are analyzed as extended projections of V by Grimshaw 1993). This would mean (e.g.) that in negating a simple (VP) projection such as Fraser sharpen it, the child will not initially develop an extended projection in which no(t) is generated as the specifier of an abstract functional projection FP, as in (13) below:

(13) [FP No [F ] [VP Fraser [V sharpen] it]]

The reason is that the FP structure (13) will violate MPP (12), since the VP structure (11) is more economical by virtue of being a simple projection (i.e. (11) is a single-headed projection of V), whereas the alternative functional structure (13) is a more complex extended projection (in that it is a double-headed projection of F and V). In terms of MPP (which leads children to prefer single-headed structures to multiple-headed structures), we can explain the observation made by Solan and Roeper (1978) and Hoekstra and Jordens (1994) that children's initial strategy for incorporating new material into a clause is to adjoin it to the overall clause.

An interesting consequence of the adjunction analysis is that it enables us to provide a principled account of children's negative null subject sentences such as (14) below (produced by Kathryn at 1:10, from Bloom 1970):

(14) No fit. No fit here. No stand up. No go in. No go first. No go in there. No lock ¶ door.

No find ¶ tank. No have ¶ this. No like celery, Mommy

If we posit that no is a VP-adjunct in such cases, then a clause such as No like celery will have a structure along the lines of (15) below:

(15) VP

NEG VP

No nc V'

V N

like celery

Since the null subject nc is in a root specifier position, it can be discourse-identified. Thus, the adjunction analysis of early child negative clauses correctly predicts that we should find null subject negative sentences. By contrast, if negatives are analysed as root specifiers (as in (13) above), we wrongly predict that children will not be able to produce negative null subject sentences (since the null subject will not be in a root specifier position, and so cannot be discourse-identified).

A final point to note in connection with the VP-adjunct analysis of early child negative sentences is that we maximize continuity with adult grammars, if we follow Zanuttini (1989) in positing that not is a VP-adjunct in adult English.

5. Wh-Questions

CICs typically contain a very limited range of wh-questions. Leaving aside potentially formulaic copula questions such as What(s) dat?, the earliest verbal wh-questions produced by young children are typically wh-complement questions (i.e. questions in which the wh-word is the complement of a verb) of the form What N do(ing)? or Where N go(ing)? For example, Klima and Bellugi (1966) report sentences such as What cowboy doing? and Where horse go? as the earliest wh-question types produced by Adam, Eve and Sarah. Bowerman's (1973) transcripts of Kendall's speech at 23 months includes the questions 'Where doggie go?' and 'Where pillow go?' The corpus included in Hill (1983, pp. 119-141) contains the following examples of verbal questions with overt wh-complements and overt subjects produced by Claire at age 2;0:

(16) Where girl go? Where pencil go? Where cow go? (x2) Where Daddy go? Where bathtub

go? What kitty doing? What the dog doing? What squirrel doing? What lizard doing?

Once wh-structures become more productive, we find a wider range of wh-complement questions, as illustrated by the following set of examples produced by Adam at age 2;4 (from Vainikka 1993, p. 34):

(17) Who me tickle? What say? What dat tell her? Where dat come from? Where find plier?

Where go drop it?

As noted earlier in relation to the examples in (6) above, a well-known characteristic of early wh-questions is that children at this stage do not make use of the presubject auxiliary found in adult questions – and the obvious question to ask is 'Why?' The traditional answer is that auxiliaries in wh-questions in colloquial speech tend to be 'weak' monosegmental clitic forms, or even null forms: cf.

(18) Where'd Daddy go? [= did]

Where's Pixie live? [= does]

What ø you doing? [= are]

Where ø you been? [= have]

Given the hypothesis (put forward by Gleitman and Wanner 1982) that items which lack acoustic salience fail to be parsed by young children, it follows that the child's 'intake' (in the sense of White 1982) for questions like (18) will be as in (19) below:

(19) Where Daddy go? Where Pixie live? What you doing? Where you been?

Thus, the absence of auxiliaries in early wh-questions is not difficult to account for. Moreover, if we assume that the child 'knows' that the wh-word is the complement of the verb, and that the canonical position of complements is postverbal, it follows that the child concludes that structures like (19) are CSV [= Complement+Subject+Verb] structures which involve movement of a wh-complement. But how does the child determine the landing-site of the preposed wh-word?

The answer is that this is determined by the interaction of two principles of UG. One is the principle that wh-phrases must have scope over (i.e. must c-command) all the other constituents of the clause containing them (cf. Penner 1992, p. 252). The second is the Minimal Projection Principle (12). This will interact with the Scope Principle to determine that wh-expressions move into the minimal A-bar position in which they have scope over the other constituents in their containing clause. In the case of a typical child wh-question such as What kitty doing?, the minimal A-bar position into which the wh-word what can move is VP-adjunct position. Thus, if we assume that what is adjoined to VP, the resulting structure will be (20) below:

(20) VP

What_i VP

N V'

kitty V e_i

doing

(where e is an empty category bound by the wh-pronoun what). The resulting structure (20) is a simple projection of V, and hence in keeping with the assumption that children do not form extended (functional) projections until forced to do so. As in the case of negative operators, the

S-structure location of wh-operators is determined for the child by UG principles relating to scope and minimal projection.

An interesting prediction of the wh-adjunction analysis is that we should expect to find null subject wh-questions. In spite of the fact that Nina Hyams (1994, fn.13, p.48) claims that such sentences are 'rare', null-subject wh-questions are widely reported in the acquisition literature. For example, Klima and Bellugi (1966, p. 200) report What doing? as a typical stage I question: Plunkett (1992, p. 58) reports that one of the earliest wh-questions produced by her son was Where go?; and Vainikka (1993) and Roeper and Rohrbacher (1994) report null subject wh-questions such as the following produced by Adam and Eve:

(21) Where put? (Eve 1;9) Where go? (Adam 2;3) What say? (Adam 2;4) Where find plier?

(Adam 2;4) Where go drop it? (Adam 2;4)

Moreover, if we look at the corpus of utterances produced by Claire at age 2;0 (in Hill 1983), we find her producing null-subject utterances such as What doing? and Where go? In actual fact, 39% [11/28] of Claire's questions containing overt wh-words have null subjects. Thus, Hyams' claim that such sentences are 'rare' seems a doubtful one.

But how does the VP analysis account for null subject wh-questions? If we suppose that null subjects are discourse-identified null constants and that wh-expressions in early child grammars are adjoined to VP, it follows that a null subject wh-question such as What say? will have the structure (22) below:

(20) VP

What_i VP

nc V'

V e_i

say

(where e is an empty category bound by what). Since the null constant subject nc is in a root specifier position in (22), our grammar correctly predicts that the null subject can be discourse-identified, and hence that children will produce null subject wh-questions.

6. Universality

Nina Hyams (1994, p. 22) dismisses the SCH as a 'historical accident', arising out of the fact that proponents of SCH looked at the acquisition of English; she suggests that it is only the fact that English has minimal and nonuniform finite verb morphology and relatively rigid SVC

[= Subject+Verb+Complement] word order which enables the SCH to account for CICs in English. However, she argues, if we look at the acquisition of (e.g. Romance or Germanic) languages with a richer verb morphology and verb syntax, 'We see that children acquire certain inflectional elements at a very early age, from the beginning of their multiword utterances. Moreover, they control syntactic operations such as verb raising and verb second (V2), which are dependent on the presence of functional heads' (Hyams, ibid.). Hyams' claims would appear to be borne out by numerous empirical studies arguing that in the initial stages of the acquisition of other languages such as French (cf. e.g. Pierce 1989), German (cf. e.g. Poeppel and Wexler 1993) and Italian (cf. e.g. Guasti 1992), children already differentiate finite from nonfinite verbs both in respect of their morphology and in respect of their syntax (in that e.g. finite verbs are positioned before negatives and nonfinite verbs after negatives, and clitics attach to the left of finite verbs but to the right of nonfinite verbs). Moreover, finite verbs carry (at least some) agreement inflections: for example, Guasti (1992: 147) reports that Martina before age 2;0 had produced not only inflected Italian infinitives such as aprire 'open', vedere 'see' and mangiare 'eat', but also agreement-inflected finite present tense forms such as chiudo '(I) close', apri '(you) open', lava '(he) washes' and fanno '(they) make'.

Thus, morphosyntactic evidence would suggest that CICs in languages like Italian with a richer verbal morphology than English project up (at least) as far as IP (with finite verbs raising from V to I). But if this is so, how can we reconcile our claim that CICs are VPs in English with the claim that their counterparts in languages like Italian are IPs? One answer which we might suggest is that the differences between English CICs and their Italian counterparts directly reflect differences between the morphosyntax of verbs in adult English and adult Italian. Following Lasnik 1994, we might suppose that verbs in languages like Italian are fully inflected in the lexicon (so that there are no bare verb forms in Italian), and must raise to INFL to check their inflectional (e.g. agreement) features. By contrast, English verbs are bare in the lexicon, and acquire their inflections by merging with an affix in INFL (through a PF process of affixation). Given this typological difference between the two languages, we might posit that CICs in both types of language are minimal lexical projections – i.e. minimal projections of verb forms in the lexicon. In Italian, verbs are fully inflected in the lexicon, and have to raise to INFL to check their inflectional features: consequently, CICs in Italian will be IPs. By contrast, English verbs are bare in the lexicon, and carry no inflections: consequently, they do not raise (either overtly or covertly) to INFL. Hence, a child who projects a 'bare' verb form into the minimal structure required to contain the verb and its arguments need not (and therefore, in consequence of MPP, will not) initially project the clause beyond VP. Only when they acquire auxiliaries and the finite affixes +s/+d (at around the age of 2;0) will English children come to project clauses beyond VP into IP.

If the arguments presented in this section are along the right lines, it follows that only children acquiring languages (like English) in which verbs are bare in the lexicon will be expected to show evidence of a Small Clause stage; for children acquiring languages like Italian in which verbs are inflected in the lexicon, there will be no such stage. Thus, the conclusion we reach is exactly the opposite of that reached by Nina Hyams – namely that it is languages like English (not those like Italian) which tell us most about CICs, since they provide us with crucial counter-evidence to the strong continuity theory.

7. Explanatory Adequacy

Given that the ultimate goal of any theory is explanatory adequacy, a crucial question which the SCH has to answer is why the earliest clausal structures developed by English children are lexical VPs, and why children do not acquire IPs and CPs at the same time as they acquire VPs.

A number of different answers have been given to the 'Why VPs?' question. One is a lexical learning account, which holds that (in consequence of the Minimal Projection Principle) children project only those lexical items which they have acquired at any given stage of development, and that they acquire contentives before functors. It follows from these assumptions that their earliest clauses are projections of the four major categories of contentive – noun, verb, adjective and preposition. Of course, this account raises the question of why contentives are acquired before functors: typical answers given to this question are that functors are late acquired because of their lack of acoustic salience (Gleitman and Wanner 1982, p. 17), or their greater cognitive/ semantic complexity (Hyams 1986, p. 82), or their greater grammatical complexity (Radford 1990, pp. 264-266), or the fact that they are subject to substantial parametric variation across languages (cf. the functional parametrization hypothesis of Chomsky 1989).

A second type of explanation (cf. Radford 1990, pp. 266-268) is a teleological one. We might argue that it is in the nature of the grammatical structure being acquired that some parts of the structure must be 'in place' before others can develop. For example, if we adopt the view of Grimshaw (1993) that all clauses share a common VP 'core', and that IP and CP are extended projections of VP, then it follows that children cannot in principle develop IP or CP projections until they have developed VP.

A third type of explanation which might be offered is a maturational one (cf. e.g. Cinque 1988). That is, we might suppose that the principles which enable the child to project argument structures onto lexical syntactic structures (i.e. structures which are projections of lexical categories) come 'on line' at an early age (e.g. 1;6), whereas the principles which enable the child to form extended (functional) projections of lexical structures come on line somewhat later (e.g. 2;0). The onset of the functional period would coincide with dendritic development in Broca's area, and with an increase in 'neural connectivity' (cf. Simonds and Scheibel 1989, Greenfield 1991).

8. Developmental Adequacy

A second problem posed by the VP analysis of CICs in English is that of 'explaining how the functional categories are acquired if they are initially absent' (Hyams 1994, p. 21): providing a satisfactory answer to this How? question is crucial to attaining developmental adequacy. Given lack of space, the account of development given below is of necessity partial, and focuses mainly on the development of wh-question structures. Following Guilfoyle and Noonan 1988 and Vainikka 1993, I shall suppose that subsequent to the VP stage, children (typically at around age 2;0) pass into an IP stage, and thence (a few weeks or months later) into a CP stage. Thus, CICs might be said to show a VP>IP>CP pattern of development. Since wh-VP questions were discussed in section 5, no more will be said about the VP stage here.

The earliest type of INFL constituent developed by young children is typically an auxiliary of some kind: for example, Radford and Aldridge (1987) report on a group of children whose first INFL constituents were a subset of modal auxiliaries (and infinitival to). Pierce (1989, pp. 85-89) suggests that copula/auxiliary be is the first INFL constituent to be acquired by Naomi. Vainikka (1993) argues that the first INFL constituents produced by Nina are modals at age 2;1 (file 10) with past tense +d appearing (and being overgeneralized) at the same time (file 10), and the dummy auxiliary do appearing shortly afterwards (file 12). Nominative case is acquired at the same time as modals/tense (Nina produces only one I subject in file 9, but 56 in file 10), suggesting a direct correlation between the acquisition of tense and nominative case (cf. the claim by Watanabe 1993 that tensed verbs license nominative subjects).

An interesting type of wh-structure produced by children at the IP stage is illustrated the examples below (from Bellugi and Klima 1966, p. 205):

(23) Where the other Joe will drive? Where I should put it when I make up? What he can ride

in? Why he don't know how to pretend? Why kitty can't stand up? How he can be a

doctor? How they can't talk? Which way they should go?

The fact that sentences like those in (23) contain finite auxiliaries and nominative subjects makes it clear that they project at least as far as IP. However, it seems unlikely that they are CPs, since if they were we should expect that (in consequence of the wh-criterion of Rizzi 1991) the auxiliary would raise to C and hence be positioned in front of the subject. Rather (as suggested by Guilfoyle and Noonan 1988, p. 40), it seems more likely that the wh-expression is adjoined to IP. This would be consistent with the view expressed earlier that children initially project the minimal structure which will ensure that the wh-expression has scope over the rest of the clause.

There seems every reason to suppose that an IP-adjunction analysis would be consistent with principles of UG. After all, Jane Grimshaw (1993) argues that in French interrogative structures such as (24) below:

(24) Qui elle a rencontré?

Who she did meet? (= 'Who did she meet?')

the interrogative pronoun qui 'who' is adjoined to IP (and Rudin 1988 similarly argues that wh-adjunction to IP is licensed in multiple wh-questions in languages like Polish, when spec-CP is already filled by a wh-operator).

A few weeks or months after they enter the IP stage, children begin to acquire a COMP projection – as is suggested by the fact that they begin to use overt complementizers: cf.

(25) See if swimming water's there (Jem 2;3)

You know that the flute is in there (Hannah 2;7)

Leave a little space for them to get out (Helen 2;7)

At this point, we start to find other items seemingly miscategorized as complementizers: data presented by Akmajian and Heny 1975 and Davis 1987 suggest that some children initially miscategorize inverted auxiliaries like is/are in root questions as yes-no question complementizers, so resulting in questions such as 'Is I can do that?' or 'Are this is broke?'

At the point where we have evidence of the acquisition of complementizers, it seems reasonable to suppose that children enter a third (CP) stage in their acquisition of clause structure – a stage at which an additional potential landing-site for wh-phrases becomes available – namely spec-CP. If wh-phrases move to spec-CP, we expect to find that auxiliaries move from INFL to COMP (to satisfy the wh-criterion), so giving rise to adult-like structures such as:

(26) What's he doing? What's he do? (Eve 2;0)

How did he get out? Why can't we open this piano? (Nina 2;9)

What's she doing? What's he got? Why was he gone? (Heather 2;2)

Jill de Villiers (1991) argues that there is a significant correlation between the point at which children acquire embedded questions containing a given wh-item and the point at which they acquire root inversion questions with the same wh-item (e.g. a correlation between the emergence of embedded what? questions and root what+auxiliary+subject questions). Why should this be?

One likely answer is that embedded wh-questions 'force' the child to posit an additional functional projection. The reason is that UG principles (cf. Chomsky 1986) prohibit adjunction to arguments (i.e. to selected complements of lexical heads), so that the child 'knows' that embedded wh-questions cannot involve wh-adjunction to IP. Once the child has acquired CP (and comes to 'realize' that all finite complement clauses are CPs), the child concludes that wh-expressions in embedded clauses move to spec-CP. It seems reasonable to suppose that the child further assumes that if spec-CP is the landing-site for preposed wh-phrases in embedded clauses, spec-CP will also be available as a landing-site for preposed wh-expressions in root clauses as well. In the case of a root wh-question, spec-CP will then be filled by a preposed wh-constituent and the head COMP position will be filled by moving an auxiliary from INFL to COMP (in order to satisfy the wh-criterion), so resulting in 'wh-inversion' structures like (26).

9. Truncation

The VP>IP>CP model of the development of clause structure presented here might seem to suggest that children pass through (at least) three clearly distinct structural stages, and that (e.g.) when they reach stage II, their clauses are always and only IPs, never VPs. However, Luigi Rizzi (1994b) has suggested that children under age 2;6 optionally project functional categories, with the result that their clauses may sometimes project only as far as VP, sometimes as far as IP, and sometimes as far as CP. In consequence, from an adult perspective, early child clauses seem to be truncated structures in which part of the functional superstructure of the clause is 'missing'. If we graft Rizzi's truncation hypothesis onto the model proposed here, we might suggest that in the earliest phase of the CP stage, children will sometimes project clauses up to CP, sometimes to IP, and sometimes only to VP.

For some children at least, this truncation phase seems to continue long beyond age 2;6 (the point at which Rizzi suggests that it ceases). By way of illustration, I shall briefly discuss data from a longitudinal study of a girl called Iris.

As we would expect, there is abundant evidence that by age 2;9, Iris has developed I- and C-projections. Typical IP structures she produces (with finite verbs, auxiliaries, preverbal negatives, infinitival to, nominatives, etc.) are:

(27) Mum, I'll help Fraser too. We are going now. I might see Pauline in there. You may like

one soup, Mummy. We want him. I play piano (2;9) I saw a rabbit. I want hankies.

You're tired (3;0) I'm going to sit on these. We had a nice one. He lay on that one (3;3).

Typical CP structures which she produces (with inverted auxiliaries and preposed wh-expressions) include:

(28) May I hang some up? What is that? What's this (2;9). Can I have some tea, Mummy?

Will you put my gloves on? Why doesn't he have slippers like that? (3;0). Can I have a

drink of juice? Can you? What am I going to do? (3;3).

Yet in spite of the fact that there is clear evidence that Iris from age 2;9 can project operator clauses into CP and other clauses into IP, she also produces clauses which seem to be simple VPs. For example, we find her producing root nonfinite clauses with objective or null subjects: cf.

(29) Me going haircut Fraser. Me being play piano. Me pull up my trousers. Me put this on

table? Us do again. Me want see outside. Mummy, done it! Go see Mummy? Got bit

cheese, Fraser? (2;9). Me going to use this one. Me sit down here. Me take microphone

off. Me put them on? Me get in bath? Want sit here. Going to have a go on the swing

(3;3).

Likewise, she often produces to-less infinitive complements: cf.

(30) Me going haircut Fraser. Me want see outside. Want sit in my chair (2;9). Want go in

the car (3;0). Want sit here (3;3)

Significantly, she also produces nonanaphoric presubject negatives such as (31) below:

(31) No Fraser play with me doll house (2;9). No me got one. No me got him (3;3)

The relevant clauses in (29-31) seem to be VPs, even though their adult counterparts are IPs.

We find a similar picture in relation to the syntax of questions. Alongside adultlike yes-no CP structures (with an inverted auxiliary in C) we find simple VP questions (with null or objective subjects): cf.

(32) Me put this on table? Got bit cheese, Fraser? Go see Mummy? (2;9) Me put them on?

Me get in bath? Want to go to bed? (3;3)

Similarly, alongside adultlike CP wh-structures with a wh-pronoun in spec-CP, we find Iris producing wh-VPs like (33) below, with objective or null subjects:

(33) Where candle go? Why me turning? Why turning? What get in shops? What put on top?

(2;9) How Fraser have his dinner? Where Fraser put his plate? Where Fraser gone? How

me going to get on? How me going to have a bath? How me get in that? How going put it

on? Why no gonna use that one? (3;3)

The fact that Iris produces (what appear to be) VP structures such as (29-33) at the very same stage that she is producing CP/IP structures such as (27-28) suggests that in the earliest phase of the CP stage (and hence throughout the earlier IP stage), children like Iris only optionally project I and/or C.

An important question which is raised by the truncation analysis is why children initially optionally project I and C, and how they later come to obligatorily project them (where required). Luigi Rizzi (1994b) offers a maturational answer to this question, suggesting that the axiom that clauses are CPs matures at age 2;6. However, maturation seems an intuitively implausible answer and is hard to square with the fact that even at age 3;6, Iris still sporadically produces VP structures such as those in (34) below, alongside IP and CP structures:

(34) Me fit one. Me have my apron off. Me tired. Fraser playing with my hooter. Mummy

get clean pants for you. Where put it? Why Fraser done jobbies in his pants again? (3;6).

So if maturation isn't the answer, what is?

An alternative which we might explore is a morphological learning account. Let's suppose that children like Iris have problems is mastering the morphosyntax of (finite) inflectional affixes in English. This may well be attributable to the fact that the present tense affix in English is null (save in the case of the 3SG/default form +s), leaving the child uncertain as to whether to analyse sentences such as 'Bears love honey' as IPs headed by a null affix which merges with the verb love, or as VPs which are headed by a bare V. The child may (for some time) entertain both hypotheses, thereby alternating between root IPs in which the verb carries a finite (+s/+ø) affix and has a nominative subject, and root VPs in which the verb carries no affix and its subject is assigned (default) objective case.

Two aspects of morphological learning may ultimately lead the child to abandon the bare (VP) analysis in favour of the affix (IP) analysis. One is eventual mastery of overt affixes such as present tense +s and past tense +d, leading the child to hypothesize that all verb forms carry an (overt or covert) affix, and to conclude that apparently bare forms carry a null affix. The other is the realization that root clauses always have nominative subjects in English (save e.g. in Mad Magazine sentences), forcing the child to posit an IP projection headed by a null inflectional affix in structures such as (e.g.) 'They love honey'. At this point, all root clauses project up to IP and have nominative subjects (and operator clauses further project to CP). Of course, questions of detail need to be resolved before our story can have a happy ending...

10. References

Akmajian, A. and Heny, F.W. 1987. An Introduction to the Principles of Transformational

Syntax. Cambridge Mass.: MIT Press

Aldridge, M. 1989. The Acquisition of INFL. Bloomington: Indiana University Linguistics

Club Publications.

Bloom, L. 1970. Language Development. Cambridge, Mass.: MIT Press.

Boskovic, Z. 1993. Selection and categorial status of infinitival complements. Ms, Univ. of

Connecticut

Bowerman, M. 1973. Early Syntactic Development. London: Cambridge University Press.

Budwig, N. 1989. The linguistic marking of agentivity and control in child language. Journal of

Child Language 16: 263-284.

Chomsky, N. 1986. Barriers. Cambridge, Mass.: MIT Press.

Chomsky, N. 1989. Some Notes on Economy of Derivation and Representation. MIT Working

Papers in Linguistics 10: 43-74.

Cinque, G. 1988. Comments on D. Lightfoot. Behavioural and Brain Sciences.

Davis, H. 1987 The Acquisition of the English Auxiliary System and Its Relation to Linguistic

Theory. PhD diss., University of British Columbia.

Déprez, V. and Pierce, A. 1994. Crosslinguistic Evidence for Functional Projections in Early

Child Grammar. In Hoekstra and Schwartz, pp. 57-84.

de Villiers, J. 1991. Why Questions? In Maxfield and Plunkett, pp. 155-173.

Gleitman, L. and Wanner, E. 1982. Language Acquisition: The state of the state of the art. In

Wanner and Gleitman (eds), pp. 3-48.

Greenfield, P. 1991. Language, tools and the brain. Behavioural and Brain Sciences 14: 531-95.

Grimshaw, J. 1993. Minimal Projection, Heads, and Optimality. Ms, Rutgers University.

Guasti, M. T. 1992. Verb Syntax in Italian Child Grammar. Geneva Generative Papers 0:2, pp.

145-162.

Guilfoyle, E. and Noonan, M. 1988. Functional Categories and Language Acquisition. Paper

presented to 13th Boston University Conference on Language Development.

Hill, J.A.C. 1983. A Computational Model of Language Acquisition in the Two-Year-Old.

Bloomington: IULC

Hoekstra, T. and Jordens, P. 1994. From Adjunct to Head. In Hoekstra and Schwartz (eds), pp.

119-149.

Hoekstra, T. and Schwartz, B.D. (eds). 1994. Language Acquisition Studies in Generative

Grammar. Amsterdam: Benjamins.

Hyams, N. 1986. Language Acquisition and the Theory of Parameters. Dordrecht: Reidel.

Hyams, N. 1992. The Genesis of Clausal Structure. In Meisel (ed), pp. 371-400.

Hyams, N. 1994. V2, Null Arguments and COMP Projections. In Hoekstra and Schwartz, pp.

21-55.

Kazman, R. 1988. Null Arguments and the Acquisition of Case and Infl. Ms, Carnegie Mellon

University.

Klima, E.S. and Bellugi, U. 1966. Syntactic Regularities in the Speech of Children. In Lyons, J.

and Wales, R. (eds) Psycholinguistic Papers Edinburgh: Edinburgh University Press, pp. 183-

207.

Lasnik, H. 1994. Syntactic Structures meets the Minimalist Program. Ms, University of

Connecticut.

Lebeaux, D. 1988 Language Acquisition and the Form of the Grammar. PhD diss, UMass.

Maxfield, T.L and Plunkett, B. (eds) 1991. Papers in the Acquisition of WH. Amherst: GLSA

Publications.

Meisel, J. (ed) 1992. The Acquisition of Verb Placement: Functional Categories and V2

Phenomena in Language Acquisition. Dordrecht: Kluwer

Penner, Z. 1992. The Ban on parameter resetting, default mechanisms, and the acquisition of V2

in Bernese Swiss German. In Meisel (ed), pp. 245-281.

Phinney, M. 1981. Syntactic Constraints and the Acquisition of Embedded Sentential

Complements. PhD diss, UMass.

Pierce, A. 1989. On the Emergence of Syntax: A Crosslinguistic Study. MIT PhD diss.

Plunkett, B. 1992. Continuity and the Landing Site for Wh Movement. Bangor Research Papers

in Linguistics 4: 53-77.

Poeppel, D. and Wexler, K. 1993. The Full Competence Hypothesis of Clause Structure in Early

German. Language 69:1, pp. 1-33.

Radford, A. 1986. Small Children's Small Clauses. Bangor Research Papers in Linguistics

1: 1-38.

Radford, A. 1990. Syntactic Theory and the Acquisition of English Syntax. Oxford: Blackwell.

Radford, A. and Aldridge, M. 1987. The Acquisition of the Inflection System. In Perspectives on

Language in Performance, ed. Lörscher, W. and Schulze, R. vol 2, pp. 1289-1309. Tübingen:

Narr.

Rizzi, L. 1991. Residual Verb Second and the Wh Criterion. Technical Reports no. 3, Geneva.

Rizzi, L. 1994a. Early Null Subjects and Root Null Subjects. In Hoekstra and Schwartz (eds),

151-176

Rizzi, L. 1994b. Some notes on linguistic theory and language development. Ms, Geneva.

Roeper, T. and de Villiers, J. 1991. Introduction. In Maxfield & Plunkett, pp. 1-18.

Roeper, T. and Rohrbacher, B. 1994. Null subjects in Early Child English and the Theory of

Economy of Projection. Ms, UMass and UPenn.

Rudin, C. 1988. On multiple questions and multiple Wh fronting. Natural Language and Linguistic Theory 6: 445-501.

Safir, K. 1993. Perception, Selection and Structural Economy. Ms, Rutgers University.

Simonds, R, and Scheibel, A.D. 1989. The postnatal development of the motor speech area: a preliminary study. Brain and Language 37: 42-58.

Solan, L. & Roeper, T. 1978. Children's use of Syntactic Structure in Interpreting Relative Clauses. In Papers in the Structure and Development of Child Language. UMass. Occasional Papers in Linguistics, vol 4.

Speas, M. 1994. Null arguments in a theory of economy of projection. Paper presented to LSA.

Vainikka, A. 1993. Case in the Development of English Syntax. Ms, UPenn.

Wanner, E. and Gleitman, L. (eds) 1982. Language Acquisition: the state of the art. London: CUP

Watanabe, A. 1993. The Notion of Finite Clauses in AGR- based Case Theory. MITWPL 18: 281-296

White, L. 1982. Grammatical Theory and Language Acquisition. Hingham Mass.: Kluwer.

Zanuttini, R. 1989. Two types of negative markers. In Proceedings of NELS 20, UMass: GLSA Publications.