Вы находитесь на странице: 1из 45

The Acquisition of Copulas

Misha Becker 22 March 1999 Dissertation Prospectus

0. Introduction: What is the copula and how is it learned?

Children do not acquire the verb be uniformly. Its rate of omission varies with the particular construction in which it is used. For example, in early English be is omitted frequently in locative expressions (e.g. Mommy (is) in the kitchen) and adjectival predicatives where the adjective expresses a temporary property (she (is) sick), but it is largely overt in deictic (there is my toy), identificational (that is a dog) and existential (there are monkeys in the zoo) expressions, as well as in nominal predicatives (she is a girl). How can we explain this nonuniformity in terms of structural properties of expressions involving be? In reviewing the literature on copular constructions in adult grammar, a couple of properties of be seem particularly interesting. One is the characterization of be as a raising verb (first proposed by Stowell 1978). Be is not unique in this respect: in English we have other raising verbs like seem and appear (they can all host expletive subjects: It seems John is smart, It appears to be raining)1. These verbs, like be, carry sufficiently little semantic content that they do not assign any theta roles (Pollock 1989). If a copula is simply a raising verb that doesn't assign theta roles, then there are several different copulas, even in English. It is obvious from looking at other languages that languages can have multiple lexical items all translatable in English as 'be'. On the other hand, be in English is clearly different from other raising verbs. In particular, it is the most "bleached", or semantically empty verb of this class (even seem and appear imply a sort of evidential

1Unaccusative

verbs (like arrive) are likewise thought to raise their single, internal argument to subject position, as

they can also have an expletive subject (there arrived 3 packages). However these verbs assign a theta role to their internal argument.

interpretation)2. That is, the meaning of be does not denote properties or events, rather it is a purely structural entity. The high degree of semantic "bleaching" of be points to the next interesting property of be: it is a purely functional item. Heggie (1988) calls it a "verbal operator". The canonical predicate in a sentence is a VP. Heggie's idea is that be creates a verbal predicate out of any predicate that does not have a verbal category (i.e. that is not a VP). It is worth noting that in certain polysynthetic languages, for instance Eskimo, the copula is a suffix that "verbalizes" a nominal or adjectival root. Seen as a verbal operator, the abstractness of the copula is clear: it lacks semantic content of its own (it does not refer to an actual event (like run) or state (know)); its function is to create verbal predicates. What does a language learner have to learn about the copula? There is a good deal of crosslinguistic variation in the lexical realization of the copula: in languages like English there is a single form of the copula (be)3 and it is overt, but in some languages the copula tends to be null in present tense (e.g. Russian), while in other languages there are several different lexical items that are copulas (e.g. Chinese, Malayalam). Given this type of variation, the learner must rely on structural properties of copulas in identifying the lexical item (or items) that is a copula in the learner's particular language. For example, a learner of English might notice that be is a raising verb (it can have an expletive subject), it is semantically bleached, and it makes a nonverbal category into a verbal predicate. If the learner has an innate expectation that language will contain an item that has these structural properties, then noticing the behavior of be in these respects is what is necessary for learning that be is a copula in English. This hypothesis leads to a number of questions:

if one of the defining properties of be is its being a raising verb (i.e. involving Amovement of the subject), why are other raising verbs acquired later (seem, appear), and

2There 3Some

are also syntactic differences: She seems to be cold, *She is to be cold. people, like Den Dikken (1995) and Benveniste (1960?) would say that have is also a copula. I tend to agree

with this view, but I won't discuss the details here.

why is passive acquired so late? How do children recognize that something is a raising verb? Maybe by the presence of expletive subjects (problem of weather verbs). how does a child learn that be is semantically bleached? Perhaps there are syntactic cues available, e.g. the wide range of syntactic categories be can combine with. if be can be accounted for in a unified way (cf. Heggie's verbal operator), why is be not acquired uniformly (data (Becker 1998) to be discussed in section 2)?

It may be argued that the abstractness of be makes it difficult to acquire. It is not a semantic primitive, in the sense that to understand a sentence like John is tall, one first has to know the meaning of 'John' and the meaning of 'tall', for which there are nonlinguistic clues. But there is no nonlinguistic clue to the meaning of 'is'. On the other hand, the "verbalizer" function of be may in fact make it simple to learn: if a child notices in the input that there is an item that consistently creates a verbal predicate out of a nonverbal category, the child could hypothesize that this item universally serves to make a predicate into the right "type" (syntactically speaking: category) to be linked to (applied to) the subject. One of the more mysterious properties of copular constructions is their sensitivity to temporal boundedness. Sensitivity to temporal boundedness surfaces in different languages in different ways. For example Spanish has two copulas that are translated into English as 'be': ser and estar. Ser is associated with permanent properties (e.g. John is tall) and set membership (John is a doctor), while estar is generally associated with temporary properties and locations. (This is not exactly the right characterization of the split, as I will discuss below.) A real question is why languages make use of a distinction between temporally bounded vs. unbounded properties; nevertheless this distinction appears in a number of languages. In addition to Spanish ser/estar, Hebrew draws a distinction between essential and nonessential properties, and even in English we see syntactic differences between what are often called stage-level and individual-level predicates (e.g. There is a man sick, but *There is a man tall). What exactly is the nature of the grammatical distinction between temporally bounded and unbounded properties? It is hoped that the way

children make these grammatical distinctions will help us better understand how these distinctions work in adult grammar. This prospectus is organized into two main sections. The first section gives a brief overview of certain aspects of the behavior of copulas in adult grammar. First I provide a summary of some of the main issues that have arisen historically in the literature on the role of the copula in sentences, and then I discuss more recent analyses as developed within the generative tradition (Stowell 1978, 1981, Heggie 1988, Moro 1993, Den Dikken 1995). I will focus on topics that will become relevant for the child language data (e.g. whether different be constructions can be unified under a single analysis, the ability of be to host a predicate in the structural subject position, predicate raising in existentials). At the end of this section I will give a brief typological survey of copular constructions in a few different languages. This survey will give us a sense of the variation (and uniformity) that UG permits in the realization of the copula. The second section turns to the acquisition data. Here I present what we know so far about the distribution of the copula in children's speech (Sera 1992, Hoekstra & Hyams 1997, Becker 1998). The English data examined thus far indicate that be is mostly null in children's locative expressions, while it is mostly overt in deictic and existential expressions. I will discuss a variety of copular constructions, including locatives (a man is in the garden) and existentials (there is a man in the garden), identificational expressions (that is a dog), progressives (John is leaving), and nominal and adjectival predicatives (John is a student/tall), and I will discuss some possible similarities between child English and child Spanish copular constructions. I will also look at the question of whether instances of null be are equivalent to Root Infinitive main verbs, and whether there is cross-linguistic variation in this respect.

1. The syntax of copular constructions in adult grammar

1.2 Traditional accounts of copular/predicative constructions Throughout the history of analyses of copular constructions, there are a few recurring themes. One is whether the copula is part of the predicate (verb phrase; i.e. whether the copula is a 4

verb) or whether it is outside of it, neither subject nor predicate. Aristotle considered the copula to be the expression of tense features, equivalent to verbal inflection but not a verb itself, thus separate from the predicate. Abelard, an 11th century French philosopher and theologian also believed that the copula was separate from the predicate; in fact we owe our term "copula" (from Latin copulare 'to link') to his view that the verb be served as a liaison between the subject and predicate. Some other common themes include the ability of be to express identity4, the question of which nominal is the "real" subject when you have two definite DPs (e.g. John is the teacher; The teacher is John), whether be is lexically ambiguous in its uses as auxiliary (John is leaving) and as a "main verb," i.e. as an expression of existence or intensionality (e.g. John is being obnoxious), and how case is assigned to the postcopular NP (most accounts assume that be cannot assign case or theta roles). Here I focus on a few accounts of be as a raising verb (in particular relating to predicate raising), and an account that unifies several different types of "be" under one analysis.

1.2.1 Stowell: be is a raising verb Many current accounts of the copula make assumptions about the syntax of be that were first argued by Stowell (1978, 1981) with regard to there-insertion, namely the assumption that be is a raising verb that takes a small clause complement. Previous transformational accounts of thereinsertion held that sentences like those in (1) are derived from the corresponding sentences in (2) by rightward movement of the indefinite DP and insertion of there:

(1) (2)

There is a man in the garden. There are books on the table. A man is in the garden.

4For

instance Russell (1919) argued that be can express identity but in this case be is actually a main verb, i.e. a

predicate; Quine (1943) similarly argues that be expresses identity. Higgins (1973), in contrast, argues that be does not express identity in cases like The number of planets is nine, where [the number of planets] and [nine] are not extensionally equivalent.

Books are on the table.

Stowell's (1978) important contribution was to analyze be as a raising verb, like seem. Thus the underlying form for both (1) and (2) type sentences is as in (3). (3) [ e [ is [ a man [ in the garden]]]]

In the case of there-insertion, no rightward movement is necessary: there is simply inserted in subject position (an empty pre-auxiliary NP position). In the non there-insertion case, the NP is raised to fill the (underlyingly empty) subject position. The raising analysis is able to capture a number of facts about existential constructions. For example the fact that (4a-b) are grammatical while (5a-b) are not follows directly from the fact that the regular predicative version of these sentences are grammatical and ungrammatical, respectively (i.e. A man was sick, but *A man was tall). I give Stowell's examples here:

(4) (5)

a. There was a man sick. b. There are three pigs loose. a. *There was a man tall. b. *There are three pigs stupid.

Stowell (1978, ex.11)

The reason for the difference between (4) and (5) was originally given by Carlson (1977). Although cardinal and indefinite NPs normally denote kinds (e.g. A man is tall is okay on the generic interpretation, where a man refers to the kind, 'men'), under existential quantification they denote a stage. That is, when the existential operator binds the NP a man5, the NP denotes a stage of being a man, i.e. a "slice" of a man's life. A stage can only have accidental or temporary properties as attributes (hence the term "stage-level", e.g. sick), not permanent or "individuallevel" properties (e.g. tall). This is because it is nonsensical to imply that an inherent property

5Or:

the existential operator binds the variable introduced by the NP a man, in Heimian terms.

follows from a noninherent or temporary one (Carlson 1977: 324)6. Therefore in both existential sentences (as in (4-5)) and in their regular predicative counterparts, predicates must denote stagelevel or non-inherent properties. Additionally there is the restriction that copular sentences with NP predicates cannot become existential there sentences:

(6)

a. A friend of mine is a real jerk. b. *There is a friend of mine a real jerk.

Stowell (1978, ex. 14a-b)

Although Milsark attributes this NP restriction to the fact that NPs generally denote individual-level properties, and the few stage-level (temporally bounded) NPs are simply blocked from both regular copular sentences (like 6a) and existential sentences (like 6b), Stowell argues that there are some stage-level NPs that can occur in copular sentences, as in (7a), but are still blocked from the existential sentence (7b):

(7)

a. Two students were real nuisances in class on Friday. b. *There were two students real nuisances in class on Friday. (adapted from Stowell 1978, ex.22)

According to Stowell, the ungrammaticality of (7b) follows from the subcategorization frame of be, a transitive (now thought to be unaccusative) verb that takes an NP "object", possibly followed by an AP, PP or gerundive verb, but not another NP. Note that there is analyzed as an NP; thus it is clear that in (7a) there are only two NPs, but in (7b) there are three (thus a violation of the projection principle).
6A

simplified version of Carlson's example is: ??Alligators in the hallway are intelligent. Alligators are intelligent.

(i) (ii)

Barring the reading of (i) in which there is a particular breed of hallway-dwelling alligators, the intelligence of alligators in this sentence seems to depend on their being in the hallway, which does not make sense. (ii) is fine because the inherent property of being intelligent can be thought to follow from the inherent property of being an alligator. For discussion see Becker (1997a).

Later, Stowell (1981) analyzed the postcopular phrase as a small clause, thus drawing the analogy between expressions like I consider [John smart] and There is [a man sick]. The regular copular sentences (A man is sick) are then formed by raising the subject NP of the small clause to matrix subject position. The raising analysis of be is now taken to be standard.

1.2.2 Heggie: be as verbal operator Heggie (1988) adopts one of the main arguments of Stowell (1978, 1981), namely that of the copula as a raising verb that has a small clause complement, and argues for a unified analysis of the copula in predicative, equative/identificational, cleft and pseudo-cleft constructions.7 Heggie's analysis revolves around analyzing the copula as what she calls a "verbal operator": something which has the effect of creating a predicate out of whatever category is in the "appropriate position". "Appropriate position" is defined as the predicate position of the small clause complement of the copula. For example in (8), the XP is the predicate, and it can be any category. (8) ... bei [SC DP [ XPi ]]

The copula coindexes the predicate, which selects the subject of the small clause. At S-structure the small clause subject raises to Spec,IP. According to Heggie (among others, cf. Pollock 1989), the copula lacks both a theta-grid and the ability to assign Case. Predicate NPs, she argues, do not need Case because they are not arguments.8 She provides a number of examples in support of this position, among other things

7Heggie

considers sentences like John is the teacher to be "pseudo-equative", as she insists that the postcopular NP

is a predicate and not an argument. She considers That man is Bill to be an equative (these are often called "identificational"). Clefts are sentences like It is Mary who is my friend, and pseudoclefts, as I discuss below, are sentences like What John ate was an apple.
8She argues for the lack of Case on predicate NPs based on adjacency effects; namely material can intervene between

be and the postcopular NP, whereas it cannot intervene between a main verb and its object. (i) a. John is evidently a genius. b. *John bought evidently 3 loaves of bread.

the behavior of French "predicate" clitics. That is, unlike argument clitics (Je lafem vois "I see her"), clitics corresponding to predicates under a copula do not agree in number or gender:

(9)

a. Marie sera ma meilleure amie. Mary will be my best friend. b. Marie le/*la sera Mary will be it

Heggie's ex. (102)

Thus Heggie argues that the postcopular DP is not an argument, but a predicate. Transfering this analysis to English, she claims that postcopular definite DPs (e.g. John is the teacher, what she calls "pseudo-equatives") are likewise predicates, not arguments. As for the cases of inverted pseudo-equatives (i.e. The teacher is John), she argues that the definite DP subject does not actually refer in these cases and is a raised predicate. That is, The teacher is John comes from the same underlying structure as John is the teacher. In the inverted case, the teacher raises to a focus position (Spec,CP) and the copula undergoes subject-aux inversion.9 The case of equative constructions (That man is John) is the most difficult for Heggie to explain. She claims that the name DP, John, behaves syntactically like a predicate (evidence from clefting, see (11)), but it clearly introduces a discourse referent (evidence from himselfmodification, see (12)).

(11)

(12)

a. That man over there is Jack Jones. b. It's that man over there that is Jack Jones. c. *It's Jack Jones that that man over there is. That old woman is Peter Sellers himself.

Heggie's (141) Heggie's (148)

9A

problem for this view is the fact that in embedded wh-questions, the definite DP in fact appears to be the true John is the teacher / The teacher is John. *?I wonder who t is the teacher. I wonder who the teacher is t.

subject. Witness the following: (i) (ii) (iii)

Higgins (1973) notes that this sort of test shows that the definite DP the teacher is the underlying subject.

In (11), the name Jack Jones cannot be clefted, as it should be if it is an argument.10 In (12) himself clearly modifies the name Peter Sellers, and not the subject DP that old woman. Thus name DPs in equatives appear to be "referring predicates." As a predicate, such a DP does not need Case. As a referring DP, however, it does. Heggie discusses a number of solutions to the Case problem, in the end suggesting that the name DP gets Case through forming a chain with the deictic DP (e.g. that man over there). She argues, I believe convincingly, against Williams' (1984) type of analysis of copular constructions involving a "flat" structure. She shows that there are indeed many asymmetries between the subject DP and predicate DP (even when both are definite), such that the structure must involve strict binary branching, and the copula must select a small clause predicate. Although she does not consider there-sentences, Heggie provides a unified treatment of several other types of copular constructions. I find Heggie's characterization of be as a "verbal operator" quite appealing, in particular because the status of be as a verb or a non-verb remains unclear.11 In many ways it is verb-like: it bears tense or agreement marking as main verbs do, it can occur in nonfinite forms such as being and been, and when it is preceded by other auxiliary material it follows negation like main verbs: John will not be late; John will not leave. On the other hand, be is different from other verbs: it always inverts with the subject in interrogatives (Is John a doctor?), its morphology is more irregular than any other verb (this is true crosslinguistically), and it does not assign case (according to many accounts, though not all) or theta roles (it is semantically "bleached", as discussed in section 0). The unified analysis of be is also appealing from a learnability perspective. If be is a verbal operator, i.e. a functional element that "verbalizes" any nonverbal category in its predicate, then

10Clefting

is normally used as a test of constituency, not argumenthood, so I am not sure exactly what her tests idea is also similar to ideas expressed by Hoekstra (1992), who suggests that no verbs are really primitive

show. Also her tests involving only modification seem to give contradictory results (see pp. 99-100).
11Heggie's

but instead are derived by the incorporation of a nominal lexical item into a functional head, a kind of verbalizer.

10

perhaps what a child needs to notice is whether the ambient language has an overt lexical item that takes nonverbal elements (e.g. a girl, tall) and turns them into verbal elements (be a girl, be tall). Furthermore, if be raises predicates as well as arguments (assuming that existential there, among other things, is a predicate - I will return to this idea below), this is also something that could be noticed by learners. Predicate raising in be constructions is discussed in the literature on pseudoclefts and on predicate inversion, to which I turn now.

1.2.3 Pseudoclefts There is a sizable literature on pseudocleft constructions (sentences that host a free relative in subject position, e.g. What John ate was an apple), and I will only cover a small fraction of it here. Since Higgins (1973; based on Akmajian 1970), two types of pseudoclefts are recognized: predicational and specificational pseudoclefts. The sentence in (13) is ambiguous between the two readings, disambiguated in the (a) and (b) versions:

(13)

What John is is important. a. What John is is important to himself (=John is important to himself; specificational) b. What John is is important to him (=e.g. John's job is important to him; predicational)

Williams (1983) argues that in specificational pseudoclefts, the free relative is actually an underlying predicate, raised to subject position, while in predicational pseudoclefts the free relative is a true subject.

(14)

predicational: specificational: XP

What Sis is subject be

XP What S predicate

The XP in this case would be [important to him/himself]. So the underlying form of a specificational pseudocleft would be, for example, Important to himself is what John is.

11

Williams bases this analysis on a number of syntactic tests, such as subject-aux inversion (*Is what John is important to himself? / ok: Is what John is important to him?) and raising (*What John is seems to be important to himself. / ok: What John is seems to be important to him). He shows that only specificational pseudoclefts show "connectedness" effects (Higgins 1973). That is, the anaphor himself can corefer with the NP John even though the right c-command relation does not appear to obtain, because the subject Important to himself is "connected" to its predicate What John is. Iatridou & Varlokosta (1995) discuss some crosslinguistic variation in the availability of specificational pseudoclefts. A number of languages, including Modern Greek (MG), Italian and Bulgarian, have predicational but not specificational pseudoclefts. They show that in MG, free relatives can be used to form pseudoclefts, but these can never have a specificational reading (e.g. What John is is silly cannot mean "John is silly" in MG). Furthermore, pseudoclefts in MG do not show any connectedness effects, as the postcopular NP cannot have Accusative Case, although it is the object of the main verb in the free relative. For example:

(15)

Afto pu agapai o Kostas perisotero ston kosmo ine o skilakos tu / *ton skilako tu this which loves Kostas more in the world is his doggie-Nom/*Acc What Kostas loves most in the world is his doggie (I&V,'95p.3)

The lack of connectedness effects is expected given the lack of a specificational reading in MG pseudoclefts. Specificational pseudoclefts are not available in MG because free relatives in MG cannot function as predicates (i.e. not because the copula cannot necessarily host a predicate in its subject position), as Iatridou & Varlokosta show independently. The literature on specificational pseudoclefts suggests that be is a predicate raiser. If Williams' analysis of specificational pseudoclefts is right (i.e. the free relative in subject position is a predicate), then perhaps a child learning a language like English could use the fact that be occurs in specificational pseudoclefts to (partly) determine that be is a copular verb.

1.2.4 Predicate inversion 12

The possibility of a predicate occupying the structural subject position has also been addressed with respect to predicate inversion in regular predicative copular sentences, by Andrea Moro (1989, 1993, 1997) and some others. Moro's main point, which was adopted by several later accounts (most notably Hoekstra & Mulder 1990 and Den Dikken 1995), was to argue that the canonical subject position of a sentence (Spec,IP) can be occupied by either the underlying subject or the underlying predicate of a predicative copular sentence. One nice effect of Moro's "flexible clause structure" approach is that he is able to provide a uniform treatment of predicatives and existentials, something few other accounts manage to do (an exception is Stowell 1978, 1981). Moro's evidence for claiming that the canonical subject position can contain either a subject or a predicate in copular sentences is based on extraction effects. He notes that there are asymmetries in extraction possibilities between the following two sentences, although they appear at first blush to be equivalent.

(16) (17)

a. [A picture of Ho Chi Minh] was [the cause of the riot]. (canonical) b. [The cause of the riot] was [a picture of Ho Chi Minh]. (inverse) a. [Which riot]i do you think [IP [a picture of Ho Chi Minh] was [the cause of ti]] ? b.*[Which person]i do you think [IP [the cause of the riot]j was [SC [a picture of ti] tj]] ? c. *[Which picture]i do you think [IP [the cause of the riot]j was [SC ti tj]] ?

Extraction of or from the object position normally leads to grammaticality, as in (17a) (and this is generally true for objects of main verbs). The fact that it is impossible in (17b-c) indicates, according to Moro, that [a picture of t] is the underlying subject, while the underlying predicate, [the cause of the riot] is in Spec,IP. Example (17b) violates Subjacency, since the trace ti has moved out of a DP and an IP; (17c) violates the ECP, since the trace ti is not properly governed (the assumption is that be does not theta-govern its complement). Moro argues that existentials (there is a picture of Ho Chi Minh (on the wall)) are formed by the same predicate inversion operation that yields inverse copular sentences as in (16b). His evidence for this claim is the fact that extraction of the DP "object" in a there-sentence yields ungrammaticality (cf. 17c).

13

(18)

*[Which picture]i do you think [IP [there]j was [SC ti tj]] ?

Note however that extraction from within the subject does not yield a Subjacency violation when there is the raised predicate, in contrast to (17b) above. (19) [Which person]i do you think [IP [there]j was [SC [a picture of ti] tj]] ?

Moro claims that raising there turns the copula into a lexical head, so that it can L-mark the SC subject; this neutralizes the barrierhood of the subject DP. [I'm not sure why there can do this but an ordinary predicate DP can't, but I haven't finished reading Moro's dissertation yet.] Hoekstra & Mulder (1990) take a similar approach to existential constructions. They argue that existentials involve locative preposing, on analogy to traditional locative preposing cases like Into the room entered a man. The basis for their analogy is two-fold: the fact that they analyze regular locative preposing cases in a very similar fashion to Moro's analysis of there-constructions, namely as in (20), (20) therei/PPi [be [SC NP ti]]

and the fact that both locative preposing sentences and there-sentences are subject to an ergativity requirement: both constructions are only possible with either an ergative (unaccusative) verb (a verb that takes only one internal argument and does not assign Accusative case to it), or with an unergative verb that has been "ergativized" (e.g. as in Into the room walked a man / There walked a man into the room) (see Hoekstra & Mulder 1990 for a discussion of ergativizationbasically the presence of a locative phrase will "ergativize" an unergative V). However unlike Moro, they specifically claim that there is a locative predicate, thus a PP, not an NP. Although I think the analogy between there-sentences and locative preposing is correct (or at least on the right track), I am a bit confused about the case-related arguments. A number of Hoekstra & Mulder's explanations for properties of there-constructions, including differences in the ergativity

14

requirement between Dutch er and English there rest on differences in how Nominative case is assigned to there/er. But if there/er is a predicate and not an argument, I am not sure why the case assignment problems arise, unless I am misunderstanding the issues (possibly I am confusing case with licensing). Den Dikken (1995) adopts both Moro's claim that existentials involve predicate inversion and Hoekstra & Mulder's claim that there is a locative phrase. He adds to the picture an analysis of small clauses as AgrPs, and restrictions on movement following from the Minimal Link condition (Chomsky 199312). Specifically, Den Dikken argues that when the predicate (i.e. there) raises to subject position in an existential or locative-preposing sentence, the head of the SC (Agr0) must raise into the head of the projection where the copula is generated (he calls it F), thus forming a complex head F+Agr. His structure looks like this: (21) 4 spec LPj F' 4 Agri+FAgrP 4 spec 4 Agr ti LP tj Den Dikken's (1) Agr' FP

LP is a predicate formed by any of the lexical categories (N, V, A, P). The raising of LP to matrix subject position (Spec,FP) violates the Minimal Link Condition by bypassing Spec,AgrP as the closest landing site. The lower head Agr thus has to raise to F in order to save the derivation. Incorporation into be of the lower Agr head can have the effect of forcing an overt copula in cases

12According

to the Minimal Link Condition, an XP can only move to the next position within its minimal domain,

i.e. it's own projection. Its minimal domain can be extended however by adjunction of the head of the XP's original (underlying) projection to the next higher head.

15

where the copula would otherwise be optionally overt. This effect can be seen in the following example:

(22) (23)

a. John is the best candidate. b. The best candidate is John. a. I consider John (to be) the best candidate. b. I consider the best candidate *(to be) John.

In (23b) the predicate [the best candidate] has inverted with the subject, John, and the infinitive be in the lower clause must be overt. Note however that this analysis is not immediately compatible with the fact that the "inverted" definite DP does appear to be able to act as an underlying subject, at least as measured by embedded wh-questions: I wonder who *(is) the best candidate (is) (this was noted previously in footnote 9). Den Dikken's specific version of the existentials-as-predicate inversion analysis will become important when we look at the child data. If head incorporation into the copula drives overtness in adult grammar (cf. 23b), does it also drive overtness in child grammar? That is, if existentials are formed by predicate inversion, and predicate inversion requires head incorporation into the copula, then we can make the prediction that children's copulas in existentials will be largely overt. This does in fact seem to be the case, as I will discuss in section 2.2. Furthermore, there are direct implications in Den Dikken's analysis for the overtness of have in child grammar and its relation to be. Den Dikken not only argues that existential be contains an incorporated Agr head (F+Agr), but that the verb have is formed by the existential copula be plus an incorporated P head (hence F+Agr+P) (similar arguments for "complex" have are found in Kayne 1993, Freeze 1992 and Mahajan 1994, discussed in Becker 1997). If the extra functional material in existential be makes it overt more than non-incorporated be, might not the additional functional material in have make it overt even more than existential be? For child English this does seem to be true, since have is virtually never omitted, existential be is omitted infrequently but more than have, and certain types of predicative (unincorporated) be are omitted much more frequently than existential be. A preliminary look at Italian child data suggests that auxiliary have in past participles is omitted 16

less often than auxiliary be in past participles, but this result needs to be investigated more thoroughly. We can see, at least, that the predicate-inversion / head-incorporation approach to certain types of copular constructions makes very interesting and apparently correct predictions about patterns of be omission in child language.

1.3 Typological survey In this subsection I briefly discuss the distribution of the copula in Russian, Spanish and Hebrew. In Russian, the present tense copula is obligatorily null in certain (many) copular constructions and optionally overt in others. In Spanish there are two distinct copulas, both of which are always overt. Hebrew contains three different copulas, two of which can be null. The purpose of this section is to have a frame of reference for the types of syntactic behavior UG allows for copulas; it will be important to keep these facts in mind when evaluating the child data.

1.3.1 Russian The copula in Russian is obligatorily overt only in the past and future tenses. In present tense it shows up only under particular circumstances; otherwise it must be null. The environments in which it is null are exemplified in (24) (from Kondrashova 199x).

(24)

a. Kolja (*est') obmanut. Kolja is deceived. b. NaS utSitel' (*est') Kolja Our teacher is Kolja. c. Kolja (*est') durak / umnaja Kolja is (a) fool / smart d. Vorona (*est') ptica. (A) crow is (a) bird. e. MaSina (*est') pered domom. (The) car is in front of (the) house. f. U MaSi (*est') sinie glaza PP Masha-Gen is blue eyes-Nom "Masha has blue eyes."

(passives) (identificational expressions) (predicatives) (generic statements) (locatives) (inalienable possession)

17

The copula can be overt in existentials and alienable possession expressions, as in (25).

(25)

a. V dome (est') telefon. In house is phone-sg.Nom "There's a phone in the house." b. U Koli (est') maSina. PP Kolja-Gen is car-Nom "Kolja has a car."

The copula can be overt in the structures in (25), however its appearance is restricted. Kondrashova proposes that be is only overt (in her terms: inserted in V) when it is needed either to express tense features (i.e. in past or future tense; present tense features in Russian are "weak" and don't need to be expressed) or to support Existential Closure. That is, for Kondrashova, Existential Closure is dependent upon having lexical material in V, which she places directly under P, the locus of the existential operator, op. She points out that in each of the constructions in (24), existential quantification does not take place because the subject is either referential (so it does not introduce a free variable) or generic (in which case the variable is bound by the Generic operator). Assuming that there cannot be vacuous quantification, existential closure will not come into play where there is no free variable to be bound; if there is no need for existential quantification, then P will not be projected (under some notion of Economy) and be will not be inserted in V. Kondrashova shows that the presence of be in existentials and alienable possessives (as in (25)) depends on the subject NP (the Nominative NPs in (25)) having an indefinite interpretation. Like in English, the NP in existentials and possessives has to be indefinite.13 Inalienable possession also involves an indefinite NP (e.g. John has blue eyes), yet in Russian it cannot take overt be. The incompatibility of inalienable possessives with overt be still follows from the
13There

is a sharp definiteness effect for existentials in English (There is a/*the phone in the house). Definite NPs in

alienable possessives yield a very particular reading, that of temporary possession (John has the car). Den Dikken (1997) discusses the structural differences between temporary and permanent possession constructions.

18

existential restriction: in saying that John has blue eyes, this statement in no way means that "There are blue eyes such that John has them", rather the NP blue eyes (or having blue eyes) is a property of John, so inalienable possession may be seen as property predication (note that in cases of body parts, it is inherent property predication). Kondrashova further argues that the existential interpretation of an indefinite NP follows from the Proper Subset Principle. That is, in order for an NP to get an existential reading, the individual(s) denoted by the NP must be a proper subset of a group (specifically, a proper subset of the intersection of the set denoted by the possessor, the possessee and the restrictive adjective that modifies the possessee). This is shown by the fact that if an NP in an inalienable possessive denotes a unique individual, be cannot be overt:

(26)

a. *U Koli est' vysokij otec. "Kolja has a tall father" b. U Koli est' vysokij brat. "Kolja has a tall brother."14

The NP in (26b), brat 'brother', can be a proper subset of the set of brothers, whereas our knowledge of the world tells us that each person can only have a single father. To summarize, the present tense copula can only be overt when the subject NP can receive an existential interpretation; this in turn is only possible when the NP is indefinite and denotes an individual that is a proper subset of a restricted set.

1.3.2 Spanish Spanish has two copulas that translate as be in English. The difference between the two, ser and estar, has often been characterized in terms of individual-level and stage-level predication, respectively. However the difference is actually somewhat different. Ser is used in all nominal

14The

facts are actually slightly more complex than I make them out to be here: the adjective vysokij 'tall' is crucial:

it serves to restrict the set denoted by the NP brat 'brother'. Because of space and time limitations I will not go into these details here; see Kondrashova (199x: 184-185) for a fuller explanation.

19

predicative constructions (e.g. John is a student), even when the subject is not permanently or inherently characterized by the set denoted by the predicate nominal. For example in the sentence in (27), the verb ser is used even though there is a clear temporal restriction.

(27)

Elisa fue reina por un da. Elizabeth ser+past queen for a day "Elizabeth was queen for a day."

Sera (1992)

Sera (1992) argues that nominal predication expresses class or set membership, and that the strength of the set membership relation overrides any temporariness in the set membership. With adjectival predicatives, the use of ser vs. estar adheres to an inherent vs. accidental property distinction. Ser is used with inherent properties (roughly, individual-level), while estar (from Latin stare 'to stand') is used with accidental properties (roughly, stage-level). However in locative expressions there is an interesting difference in the uses of ser and estar. Ser is used to express the location of an event; estar is used to express the location of an object or individual (even when that object's location is permanent, e.g. Cuba est en el caribe 'Cuba is in the Caribbean'). The most plausible connection between the expression of permanent/inherent properties and the locations of events (both involve ser) seems to me to be the fact that the location of an event is an essential property of that event. If the event took place in another location, it would then be a different event. But objects and individuals are continuous in their existence across locations. Thus a location does not constitute a defining property of an object or an individual, but it is essential to an event. The distinction between essential and nonessential characteristics may turn out to be fundamental in characterizing certain properties of be. For instance, as we will see in section 2, it appears to be the case that young children distinguish copular constructions along these lines.

1.3.3 Hebrew Hebrew is an interesting language to look at in terms of copular constructions both because the copula is often nonovert, and because the lexical items that are labelled "copulas" are so labelled 20

controversially. That is, they function both as pronouns and as copulas, and some researchers (e.g. Rapoport 1987) consider them non-verbal. In her dissertation, Heggie (1988) discusses one of the Hebrew copulas, H (hu). She argues that it is a clitic in Infl. Ordering facts with respect to adverbs and negation suggest that H is in Infl; the fact that it cannot be stressed or stranded suggests that it is a clitic15. She points out that H is largely null in present tense predicative sentences (it is overt in past tense), but there are further restrictions on its overtness. Specifically, it cannot be overt in a predicative sentence if the subject is a personal pronoun (he (*is) a student). It is optionally overt in equatives where the subject is a pronoun (he (is) John). And it is obligatory in inverse equative sentences (John *(is) that man). While Rapoport (1987) argues that these facts support a nonverbal analysis of H, Heggie explains them in the following way: she analyzes H as a lexicalized form of the copula which must be overt when it needs to lexically head-govern an empty category. For example, a trace of movement of an NP that originated in the predicate under the copula must be lexically head-governed (otherwise there is an ECP violationnotice that this is very similar in spirit to Moro's proposal). Therefore in cases where the postcopular predicate has raised, H must be overt. (More must be said about why H is optionally overt in equatives with a subject pronoun, which I believe Heggie does.) Heggie's proposal that the overt copula in Hebrew is needed to lexically head-govern an empty category might also capture the Russian facts. Although Heggie's account does not say anything about restrictions on indefiniteness or existential interpretations, one might plausibly argue that existentials and alienable possessives in Russian involve predicate raising. Recall from the discussion of predicate inversion that existentials are argued to involve a raised predicate (according to Hoekstra & Mulder (1990) and Den Dikken (1995) it is a locative predicate - thus akin to the Russian PP). Den Dikken further argues that possessives, like existentials, involve predicate raising (the possessor is raised from a postcopular position). In fact Freeze (1992) shows with crosslinguistic data that in many languages existentials and possessives are syntactically equivalent; furthermore see the Appendix for evidence that some languages, e.g. Chinese, use the

15It

can be stressed as a pronoun, but not as a copula (Curtiss, p.c.).

21

same copula in existentials and possessives, but don't use that copula in other constructions. Therefore it is plausible to argue that the possessor in Russian is a raised predicate, so the copula must be overt to lexically head-govern the trace of the possessor. Heller (1999) discusses two different (pronominal) copulas in Hebrew: H (hu) and Z (ze). She examines their differences with respect to different interpretations they yield in pseudoclefts. Recall from section 1.2.3 what the predicational vs. specificational readings are (the example is repeated here):

(13)

What John is is important. a. What John is is important to himself (=John is important to himself; specificational) b. What John is is important to him (=e.g. John's job is important to him; predicational)

What Heller shows is that these two readings are disambiguated in Hebrew with the use of H (hu) vs. Z (ze). She shows that an H copula yields only a predicational reading in pseudoclefts, while the Z copula yields only a specificational reading. She also argues that there are two different Z copulas, an agreeing one and a non-agreeing one, where the nonagreeing Z copula shows connectivity effects while the agreeing Z copula does not. Recall what we said in section 1.2.3 about these two different readings of pseudoclefts, namely that (according to Williams 1983 and others) the specificational reading involves a free relative predicate in subject position, while the predicational pseudocleft involves an underlying subject in subject position. If Heller's empirical generalization is right, the conclusion we are led to is that the Z copula, but not H, allows predicates in its subject position. Let me add a word of caution here. Heller's grammaticality judgments are not shared by other native Hebrew speakers. In fact, there seems to be a good deal of variation in the syntactic and semantic judgments of these sentences. However, there does seem to be agreement that both predicational and specificational readings are available in Hebrew, and that there are differences between H and Z in this respect. Since both H and Z copulas can be null in the adult grammar, it will not be interesting if Hebrew children omit them frequently. However there are certain restrictions on the overtness of 22

these copulas that are interesting. For instance, it appears to be the case that an overt H copula can only be used with predicates that denote defining or essential properties (e.g. a table being a piece of furniture, a mountain being tall/big, or trees being green), where these are not necessarily individual-level predicates (e.g. John *H tall). Since English-speaking children tend to use an overt copula with permanent properties but a null copula with accidental properties, it would be interesting to see when Hebrew-speaking children begin producing overt H and Z copulas (e.g. whether H comes in earlier than Z), and whether they make any errors in the distribution of these elements. Finally, Hebrew contains another copula, yesh, which appears in both existential constructions and possessives (Borer, p.c.). Since yesh occurs with a dative pronoun (as possessor) and a nominal possessee, we will be able to see whether Hebrew-speaking children produce null-copular possessive constructions (i.e. to-me 0yesh NP). A few such utterances are found in the Hebrew data from CHILDES, but I don't know yet how robust their occurrence is.

1.4 Summary of section 1 In the preceeding subsections I have discussed various accounts of the syntax of copular constructions. All of the accounts I have discussed either argue for or assume the fact that be is a raising verb. More particularly, it is capable not only of raising the subject of its small clause complement to subject position, but also of raising the predicate of its SC complement. That is, unlike other verbs, be can host a predicate in its subject position. This ability is evidenced in specificational pseudoclefts, which host a free relative predicate in subject position, and in Moro's, Hoekstra & Mulder's and Den Dikken's analyses of existentials as predicate inversion structures. The crosslinguistic data suggest that predicate inversion in copular sentences, as well as inherent property-denoting predication both play a role in determining the form of be (i.e. its overtness, as in Russian and Hebrew, or its lexical form, as in Spanish). Some of the issues I will turn to in the next section are: what are the effects of predicate inversion structures on the form that be takes in child language? For instance, if existentials and possessives involve predicate raising, is

23

the child's copula in these constructions different from (e.g. overt more than) the copula in structures that don't involve predicate raising? Secondly what is the connection between the predication of essential or inherent properties vs. accidental properties and the form that the copula takes? Perhaps the predication of an essential property is somehow more fundamental to the function of be. For example, in the sentence A mountain is rocky, we can think of a mountain as a set of properties, among which we necessarily find "rocky" (i.e. if an object called "mountain" were a smooth, flat thing, it wouldn't be a mountain), so a mountain is necessarily rocky. One question that we might ask is whether this property of inherent predicates makes be conceptually easier to acquire in cases of inherent property predication, and/or what is the grammatical reflex of the connection between a (particular) form of be and inherent property predication. Finally, I would like to know whether there is any connection between these two properties of be: the ability to raise predicates and different types of predication. We already saw in section 1.2.1 that only stage-level properties can be predicated in existential sentences in English (which, if we follow Moro, are instances of predicate raising). In fact, if we look at the complete paradigm of there- vs. non-there copular constructions, some interesting patterns emerge.

(28) there-construction There is a man sick. *There is a man tall. There is a book on the table. There is a party in the garden. There are a million people in Topeka. regular predicative A man is sick. *A man is tall. (on the existential reading) A book is on the table. *A party is in the garden. ??A million people are in Topeka. (temporary)16 *There is a Swiss man a contestant on the show A Swiss man is a contestant on the show.17

16Note

that the disambiguated form for the permanent reading is Topeka has a million people. In this case, have

alternates with be.


17Adapted from Stowell (1978).

24

There are clear effects of permanent/temporary predication, and perhaps temporal boundedness more generally in the grammaticality of these constructions. The first two pairs of sentences show that temporary/accidental properties can be predicated of indefinites in both existential and non-existential predicatives, while permanent/inherent properties cannot be predicated in either (this was discussed above in section 1.2.1). The second two pairs show that indefinite nominals that denote objects or individuals can occur in both existential and nonexistential predicative constructions, while indefinite event nominals (a party) cannot be the subject of a locative predicate (events may be seen as less permanent than objects). In the fifth pair we see that the existential sentence is ambiguous between a temporary and a permanent reading (i.e. the people happen to be there, e.g. for a conference (temporary), or they live there (permanent)), while raising the indefinite to subject position forces the temporary reading. Finally, the last pair shows that while an indefinite NP subject can have a nominal predicate, two nominals cannot occur in postcopular position in an existential sentence (so it is the complement of the 'party' case). These facts are poorly understood at present. My hope is that a better understanding of the grammatical restrictions placed on copular constructions by temporal boundedness will be suggested by the acquisition data.

2. The copula in language development

Having explored some of the properties of the copula in adult grammar and some of the variation we find among languages in copular constructions, let us now take a look at what we know about the behavior of be in children's grammars. At the moment relatively little is known about the acquisition of be. In addition to my own (preliminary) investigation there is one study of Spanish-speaking children's copular sentences (Sera 1992), discussed below, and one study I know of on the copula in early African American English (Wyatt 1995). Here I will mainly report on my own findings based on a (brief) investigation of some English-speaking children. What I am interested in is the contexts in which be is overt vs. the contexts in which it is null. Some of the issues I will address are omission patterns of be in existenial, locative, predicative and progressive 25

sentences, and the question of whether a null copula is equivalent to a main verb Root Infinitive (in both cases overt tense features are missing).

2.1 Locatives and existentials Omission patterns of be in existential (there is a man in the room) and locative (a man is in the room) constructions appear to strongly support Den Dikken's (1995) account (following Hoekstra & Mulder 1990 and Moro 1989/93) of existentials as a form of predicate inversion. Recall from section 1.1.4 that according to this account of existentials, predicate raising forces the head of the post-copular SC to raise into the head occupied by be (to avoid a minimality violation; Chomsky 1993). So the underlying structure of an existential sentence is as in (1).

(1)

a. [FP Spec [F' F [AgrP XP [Agr' Agr [PP there]]]]] b. [FP e [F' be [AgrP a man in the room [Agr' Agr [PP there]]]]]18

According to Den Dikken, the PP predicate there raises to Spec,FP, the XP in Spec,AgrP being filled by the subject of the small clause. As I discussed in that section, the raising of the small clause predicate to subject position has effects on the overtness of be in the adult grammar. I repeat the relevant examples here:

(22) (23)

a. John is the best candidate. b. The best candidate is John. a. I consider John (to be) the best candidate. b. I consider the best candidate *(to be) John.

If head incorporation drives overtness of the copula in adult grammar, we might ask whether the same is true in the child grammar. That is, if existential be is the [Agr+F] complex argued for by Den Dikken, perhaps it will be overt more frequently than a simple F head. Although Den Dikken

18I

am unsure why Den Dikken places the PP [in the room], clearly an adjunct, inside the spec of the small clause.

Moro places outside of the SC, hence: ...[a man [there]] [in the room].

26

does not discuss locative constructions, I think it is safe to assume that he would derive them as regular copular constructions, i.e. raising the subject of the postcopular small clause. Thus the derivations of existentials and locatives would be as follows:

(2) (3)

existential: locative:

[FP therei [F' F+Agrk [AgrP a man in the room [Agr' tk [PP ti]]]]] [FP a mani [F' F [AgrP ti [Agr' Agr [PP in the room]]]]]

Let us now look at the child data. The subjects of this study are three children: Nina (Suppes 1973), Peter (Bloom 1970) and Naomi (Sachs 1983) (data from the CHILDES database, MacWhinney & Snow 1985). Below in Table 1 I list the files that were used for each child, and the age of the child at each data point. These files were selected because they were the earliest files containing multiple occurrences of have utterances19. At these data points the children's grammars are very much in a developing stage. That is, they produce utterances with root infinitives (nonfinite bare verb forms) and null subjects alongside utterances with finite verbs and overt subjects.

Table 1: CHILDES files examined for each child: file# (age of child) Nina 07 (2;0.24) 10 (2;1.15) 13 (2;2.6) 16 (2;3.5) Peter 06 (2;0.10) 09 (2;2.13) 10 (2;3.3) Naomi 42 (2;0.28) 47 (2;1.17) 51 (2;2.25) 56 (2;3.29) 60 (2;4.30) 62 (2;5.8)

For each file, I coded any child utterance that either contained or seemed to lack the verb be in an existential (there)20 or locative construction. In Table 2 I give the average rate of overt be in

19The

relevance of have will be discussed in the dissertation. expressions include both deictic there expressions (There is my toy) and existentials (There are monkeys in

20There

the zoo). Since the two constructions exhibit syntactic differences (definiteness effect, subject-aux inversion), the

27

existentials as compared to locatives (averaged across all files for each child). As we can see, be is overt far more frequently in existentials than in locatives.

Table 2: Rate of overt vs. null be in existentials vs. locatives, averages per child be: overt vs. null Nina overt null % overt be 2 = 25.22, p .001 Peter overt null % overt be 2 = 7.28, p .007 Naomi overt null % overt be 2 = 12.27, p .001 existentials 37 8 82% 33 15 68.7% 29 3 90.6% locatives 16 38 29.6% 10 19 34% 2 6 25%

Although Peter's data appear to be somewhat weaker than those of the other children, the results for all children are highly significant. To sum up this subsection, the omission patterns of be in existential and locative constructions seem to support the analysis of existentials as involving predicate raising, insofar as predicate raising forces head incorporation into the head of the FP/IP projection, and insofar as head incorporation can drive overtness of the copula. Another very interesting point to be made with respect to Den Dikken's analysis and the child data involves possessive constructions. Den Dikken argues that possessives, like existentials,

question is whether they should be grouped together. For the present I will keep them together, on the assumption that deictic there is pretty clearly a raised predicate (Den Dikken, p.c.).

28

involve predicate raising.21 According to his analysis, possessives are derived from the underlying structure in (4) yielding the structure in (5).

(4) (5)

[FP Spec [F' F [AgrP DP [Agr' Agr [PP Pdat DP]]]]] [FP DPi[F' F+[Agrk+Pj] [AgrP DP [Agr' tk [PP tj ti]]]]]

As we can see from the derived structure in (5), the position of the copula now contains two incorporated heads: it is [F+[Agr+P]]. The presence of the incorporated P is argued to cause it to be pronounced 'have' (Kayne 1993 and others); furthermore, if the presence of additional functional material in the copula drives overtness, and if the possessive copula contains an extra head as compared to the existential copula, then we would expect the possessive copula (have) to be overt even more than existential be. In fact this is the case for English. Looking at the same files as in Table 1 (the same ones we examined for the omission of be), we find that have is virtually never omitted:

Table 3: Occurrences of utterances containing/lacking have: n (%) (across all files) Nina overt have (%) null have (%) 108 (96%) 4 (4%) Peter 52 (98%) 1 (2%) Naomi 34 (97%) 1 (3%)

Peter and Naomi had only one case each of a null have utterance, and three of Nina's four examples were dubious cases of null have. The clear cases of omitted have are given below in (68).

21The

claim that possessives are a type of existential, or at least that the two constructions share a common

underlying structure, has been made by a number of people, perhaps most notably Freeze (1992), who shows that a number of languages form possessives in precisely the same way as existentials. Also, many languages use a single copula for both existentials and possessives (e.g. Classical Chinese and Hebrew; Spanish hay 'there is' and French il y a 'there is' both use a form of 'have' for the existential copula). The related claim that have is derived from be has also been made by a number of people (Kayne 1993, Mahajan 1994, i.a.). Den Dikken (1995, 1997) also argues in favor of this view.

29

(6) (7) (8)

Nina: that teddy bears on it (= that has teddy bears on it) Peter: lizards # Mama lizard a tongue. (= Mama lizard has a tongue) Naomi: we-'re breakfast now. (= we're having breakfast now)

Nina's three unclear cases are given in (9-11). I coded them as possible cases of null have because of contextual evidence in the conversation.

(9) (10) (11)

Nina: her big mouth (= she has a big mouth) Nina: I some more blocks (= I have some more blocks) Nina: her apple juice (= she has apple juice)

The evidence from have omission appears to lend additional support to Den Dikken's analysis of existentials and possessives, again on the assumption that these structures involve head incorporation, and on the assumption that head incorporation can drive overtness. We will now look at some cases that indicate that head incorporation is not likely to be the whole story.

2.2 Predicatives A look at predicative constructions (John is a student) reveals an apparent paradox. If be in locatives may be null because it lacks extra functional material, in contrast to existential be, then be in predicatives should likewise be omitted, since presumably predicatives do not involve predicate inversion (i.e. Johni is [ti [a student]]). However the child data show that predicative be is overt far more frequently than locative be, in fact about as often as existential be.22 In table 4 I give the average rate of overt be in predicatives (listing identificational expressions (That is a dog)separately).

Table 4 child Nina be: overt vs. null predicatives identification (that) overt 32 91 existentials locatives 38 16

22Peter's

data is strange, and I have nothing to say about it at this point.

30

null % overt be Peter overt null % overt be Naomi overt null % overt be average

16 66% 97 8 92% 18 14 56% 71.3%

21 81% 141 30 82% 24 4 82.7% 81.9%

10 79% 33 15 68.7% 29 3 90.6% 79.4%

45 26% 10 19 34% 2 6 25% 28.3%

What we see in Table 4 is that be is overt in identificationals just as often, on average, as be in existentials (both about 80%). And although there is considerable variation among the three children in their rate of overt be in regular predicatives, the rate is nowhere near the low rate of overt be in locatives (71% vs. 28%). (Even for Naomi, who has the lowest rate of overt be in predicatives, she has 56% overt be in predicatives vs. only 25% in locatives.) Since neither predicatives nor identificational expressions involve predicate raising, head incorporation is not the reason for the high rate of overt be in these constructions. However a closer look at the data reveals a different distinction. In predicate nominals be is almost always overt (94%), while adjectival predicatives, which are used with either stage-level (e.g. sick) or individual-level (e.g. big) predicates are more evenly split with respect to finiteness (and among them, individual-level predicates tend to be overt more than stage-level ones [exact figures to be filled in]). English children's use of be in predicative constructions ties in nicely with what children acquiring Spanish do. Recall from section 1.3.2 that in adult Spanish ser is used with predicate nominals, predicate adjectives that express a permanent property and locatives that express the location of an event; estar is used with predicate adjectives that express a temporary property and locatives that express the location of an object/individual. In a study on children's use of the copula in nominal, adjectival and locative predicative sentences, Sera (1992) found that children correctly used ser in cases of nominal predication and permanent-property predication, and they correctly

31

used estar with temporary-property predication. However, the children only used estar in locatives, regardless of whether the locative expressed the location of an object or an event. What is interesting about her finding is that it seems that English-speaking children use an overt copula in exactly the constructions in which Spanish-speaking children use ser, and that they use a null copula exactly where Spanish-speaking children use estar.

Table 5: production of copulas in child Spanish and English type of predication nominal predicate permanent-prop. adj. pred. temporary-prop. adj. pred. locative child Spanish (Sera 1992) ser ser estar estar child English (Becker 1998) overt be overt be null be null be

2.3 Be and have as aspect markers In English, be is used to mark progressive aspect (John is singing), while have is used to mark perfective aspect (John has left). In both German and Italian (as well as in a number of other languages) either be or have can be used in perfectives, where the choice of be vs. have (known as auxiliary selection) usually depends on whether the main verb is unaccusative (if unaccusative, then the auxiliary is be). There is a good deal of variation among English-speaking children (the 3 I have studied) in their omission patterns of be in progressives, even more variation than we found in predicatives. Their average rate (across files) of overt be in progressives is given in table 6.

Table 6: % overt be in progressives (avg.) child Nina be: overt vs. null overt null % overt be Peter overt null progressives 30 79 27% 50 7

32

% overt be Naomi overt null % overt be

87% 44 20 68%

It is interesting that we find this degree of variation, given the high degree of uniformity of the rate of overt be in other constructions, such as identificationals, existentials and locatives. Looking at each individual data point, we see that each child is gradually requiring be to be overt more and more frequently in progressives. At the moment I do not have an explanation for the high degree of variation.

2.4 Null be as a Root Infinitive A robust phenomenon found in many child languages is the appearance of children's matrix verbs in either bare or infinitive form. This stage of linguistic development is known as the Root Infinitive (or Optional Infinitive, Wexler 1994) stage. Some representative Root Infinitive (RI) utterances are given below.

(12)

a. Want more apple. b. Papa schoenen wassen. Daddy shoes wash-inf. c. Michel dormir. Michel sleep-inf. d. Nicht will duschen. not want shower-inf.

(Bloom, Lightbown & Hood 1975) (Dutch, Weverink 1989) (French, Pierce 1992) (German, Becker 1995)

Since RIs fail to exhibit finiteness features, and a null copula also lacks overt finiteness features, a logical question to ask is whether children's null copulas are syntactically equivalent to children's main verb RIs. This question is relevant because if it turns out that children's use of an overt copula with inherent property predication but a null copula with accidental or temporary property predication is robust, we might explain this pattern, in part, by saying that inherent property 33

predication, having no temporal reference, requires grammatical tense marking (through an overt copula). Accidental or temporary property predication, on the other hand, does have temporal reference (in Kratzer's (199x) terms, it has an event argument), so it might be that its temporal reference can receive a deictic interpretation, thus not requiring grammatical tense marking. The deictic marking of tense in cases of accidental/temporary property predication might be similar to children's deictic option for pronoun use (i.e. children often use pronouns without having introduced the referent of the pronoun into the discourse). One of the most important developments in child language research in recent years is the insight that not all RI phenomena are the same. In particular, in languages that have true infinitive morphology (e.g. Dutch, German, French), children's RIs behave differently than those in languages without infinitive morphology (e.g. English). One difference is the tolerance of fully specified subjects with nonfinite main verbs (English allows RIs with full DP subjects, i.e. subjects with an overt determiner, while Dutch and German do not) (Hoekstra & Hyams 1997). Another difference is the fact that in languages with infinitive morphology, children's RIs often have a modal interpretation (Wijnen 1996, Hoekstra & Hyams 1998). I will return to the issue of modal reference below. Hoekstra & Hyams (1997) show that null be utterances and (main verb) RI utterances in child English pattern together in terms of occurrence with underspecified DP subjects. That is, both finite main verbs and finite be must occur with a fully specified subject: either a pronoun or a DP with an overt determiner. Both nonfinite (bare) main verbs and null be can occur with either an unspecified subject (a null pronoun or a bare, or determinerless, N) or a fully specified subject.23

23The

ability of English RIs and null be to occur with fully specified subjects is explained in Hoekstra & Hyams

(1997), and it has to do with markedness effects. Briefly, the spec-head agreement requirement, on which H&H base their analysis, only applies to "marked" forms. In English the bare verb is the unmarked form, hence there is no spec-head agreement requirement, and the bare verb (or null be) can freely occur with either specified or unspecified subjects (while finite verbs, being marked, must agree with their subjects in finiteness, i.e. an overt determiner or pronoun). In Dutch, in contrast, it is the third person singular (3sg) form of the verb that is unmarked, so it can occur with either specified or unspecified subjects, which it does, while nonfinite verbs are marked, thus occurring only with determinerless or null subjects. A full justification of their proposal is given in H&H '97 and I will not give it here.

34

Broadly, the idea is that since RIs result from an underspecification of finiteness (or tense) features in the Infl projection, and since the copula (or auxiliary be) resides in Infl, then a null copula (or auxiliary) is likewise a case of underspecified Infl features. On the basis of this theoretical assumption and the similar behavior of null be and RIs with respect to subject properties, H&H argue that null be is nonfinite, like main verb RIs. In Table 7 I give the pattern of occurrence of finite/nonfinite main verbs and be with respect to subject overtness in English. Table 8 gives the breakdown of main verbs vs. be.

Table 7: Finiteness and subject specification Nonfinite V / null be Null determiner (bare N) subj Overt determiner/pronoun subj 91 295 Finite V / overt be 5 417

Table 8: Finiteness and subject specification: main Vs vs. be Nonfin. main V Null determiner (bare N) subj Overt determiner/pronoun subj 26 45 null be 45 250 fin. main V overt be 2 81 3 336

Table 8 shows that although we find many more occurrences of an overt determiner or pronoun subject with null or overt be than we find with main verbs, the ratio of main verbs to be in both finite and nonfinite conditions is the same (that is, 45:250 is about the same proportion as 81:336, both about 1:5), indicating that the difference in actual numbers reflects the greater overall number of occurrences of be. Most importantly, the number of finite main verbs and overt be with a bare N (null determiner) subject is equally low (2 and 3, respectively). Hoekstra & Hyams (1997) address the issue of whether null be is finite or nonfinite in response to the claim by Guasti & Rizzi (1996) that a null auxiliary be in children's wh-questions (e.g. What doing?) constitutes a finite C resulting from I-to-C movement. Instead, H&H show that English cases of null be are nonfinite: English children's wh-questions often involve a null subject altogether, and over 50% of their wh-questions with an overt subject involve a bare N, thus a null

35

determiner (based on data from Adam (Brown 1973)). In short, the behavior of be with respect to subject properties in declaratives mirrors the behavior of main verb RIs in this respect, confirming the hypothesis that null be is equivalent to a main verb RI, and hence is nonfinite.24 One important caveat I must state here is that it is not known whether null be expressions in child Dutch pattern like English null be or like Dutch main verb RIs with respect to occurrence with full DP vs. bare N subjects. If Dutch null be is like English null be, then Dutch null be declaratives should occur with both specified and unspecified subjects. If it is like Dutch main verb RIs, however, it should only occur with unspecified subjects. As far as I know, this question remains open at present (this is something to look into, also for German).25 Let us turn now to the second issue I mentioned above, namely the fact that RIs in "true infinitive" languages (e.g. Dutch and German) often have a modal interpretation, while RIs in "bare verb" languages (e.g. English) largely have a here-and-now interpretation. The availability of modal reference in Dutch/German RIs is linked to an interesting asymmetry between eventive and stative verbs in languages like Dutch and German, which I will discuss in the dissertation. Here I will just give some facts about the modal interpretation of infinitives and how null be patterns in this respect. For Dutch and German, we will see that null be does not have the same modal reference as main verb RIs. As I show in tables 3 and 4, Dutch children's infinitives largely have a modal interpretation, while English children's RIs (bare verb utterances) often have a nonmodal reading (i.e. ongoing action).

Table 9: modal reference in Dutch RIs (adapted from Wijnen 1996) past RIs finite 3% (64) 3% (21) present 10% (194) 93% (657) future/modal total 86% (1625) 1883 3% (21) 699

Table 10: modal reference in English RIs (adapted from Ud Deen 1997)
24The 25If

implication of their result is that English allows movement of a nonfinite element (null be) from I to C.

Dutch null be patterns like English null be in both declaratives and wh-questions, however, this result would be

problematic for H&H's conclusion that English allows nonfinite I-to-C movement while V2 languages like Dutch do not.

36

past RIs finite 22% (59) 37% (33)

present 65% (171) 52% (46)

future/modal total 13% (34) 26% (10) 264 89

German RIs show the same effect as in Dutch; modal reference in German RIs ranges from 57% to 100% (Becker 1999). H&H (1998) show that the ability of Dutch RIs to express a modal interpretation is directly related to their status as true infinitives, i.e. they show distinct infinitival morphology (-en), and thus carry a [-realized] feature (according to Giorgi & Pianesi 1996), meaning that they are compatible with a modal (i.e. irrealis) meaning. In contrast, English RIs are simply bare verbs, bearing no actual infinitival morphology; such infinitives carry a [+perfective] feature (Giorgi & Pianesi 1996), meaning that they are incompatible with a modal (irrealis) meaning. I should point out that child Dutch/German speakers do not produce stative RI verbs (only eventive RIs; their stative verbs are always finite). Now, be is clearly a stative verb, as it is normally incompatible with progressive aspect even when it expresses stage-level (noninherent) predication. (13) *John is being sick26

Since be is stative, and since we have independently discussed the fact that null be is nonfinite, we would expect children acquiring Dutch (or German) not to produce utterances with null be. However, this is clearly not the case, as we find many examples of null be declaratives in child Dutch and German.

(14)

a. Ah, mein strasse maputt! Oh, my street (is) broken

26Be

does occur in progressive aspect when it expresses intentionality, e.g. John is being obnoxious/gratious. For

the moment I will leave this fact aside, but note that the adjective expresses a nonpermanent or noninherent property of John, so be would appear to have an eventive interpretation in this case.

37

b. Hanna das Mama Hanna (is) the mommy c. Teddy unter dem tisch Teddy (is) under the table

(Becker 1995)

However it is significant that children almost never produce the infinitive form of the copula (e.g. @Johanna gross sein 'Johanna be big', this is also true for child English). Since a null copula fails to show infinitive morphology, just like English bare verbs, null copular expressions in child Germanic may be analogous to English RIs. If this is the case, null copular expressions should not show the modal reference effect. The data do suggest that this is the case. Neither the utterances in (4) nor those in (5-7) from Andreas (Wagner 1985) have a modal interpretation.

(15) (16) (17)

ja das nicht Gonzales yes that not Gonzales (=a toy) nein ich ein Froschknig no I a frog-king mein das mine that

Recall that the rate of modal reference for nonfinite main verbs in child German ranged from 57% to 100%, and in child Dutch it was 86% on average. So the German data are suggestive of a connection between German/Dutch null copular expressions and English bare verb RIs (namely a lack of modal interpretation). To sum up, the English data show that null be utterances pattern with regular main verb RI expressions: both null be and main verb RI utterances occur with either a bare N or a full DP subject, while finite main verbs and overt (inflected) be occur only with full DP subjects. Thus null be appears to be nonfinite, on par with main verb RIs. However it is not yet known whether the same correlation holds for null be in child Germanic. This question is highly relevant because we have seen that null be in child German patterns with English RIs in its lack of modal reference. We

38

should find out whether German/Dutch null be patterns with English RIs or German/Dutch RIs in terms of its occurrence with full DP subjects.

3. Other topics/Plan I hope to look at be-omission data in other languages as well, in particular Spanish, German, Italian and possibly Hebrew. The Spanish data are particularly suggestive and I will be looking at earlier Spanish data with John Grinstead (in particular to see about omission patterns). German and Italian are useful because they both exhibit auxiliary selection (have or be) in past participle constructions, and in both of these languages the past participle is the more common expression of past tense than the simple past form. Thus we can look at whether there are differences in the overtness/finiteness of auxiliary be vs. auxiliary have in past participles (which we cannot look at in English for obvious reasons). Furthermore, it will be interesting to compare German with Italian, since child German shows a robust effect of Root Infinitives, while Italian does not (there are almost no RIs in child Italian).

Plan Spring/Summer 1999: Review literature on passives (Borer & Wexler 1987 & 1994, Baker, Johnson & Roberts 1989) & raising to try to understand how children learn which constructions involve raising (A-movement). English data: go through adjectival predicatives more carefully & see whether kids are making a temporary/permanent distinction or an inherent/noninherent distinction, or something else. go through there- and here- constructions and separate true existentials from deictic there/here sentences Spanish data: check for early omission patterns with John Grinstead (based on English data & ser/estar distinction, I expect to find more omissions of estar)

39

read more about ser/estar distinction to try to understand the nature of the temporal restrictions on copular constructions; this should help us understand whether children's omission patterns reflect an underlying distinction between inherent & noninherent property predication. Continue background reading on acquisition: Brown, Stromswold... Review some diachronic syntax literature (e.g. Roberts) to better understand the process of grammaticalization and semantic bleaching, to try to understand how children learn which verbs/functional items are bleached (high frequency & large number of syntactic frames?) Think about possible experiments (imitation tasks?)

Summer/Fall 1999: Continue background reading: Heycock, Kratzer, typology; try to understand the nature of temporal restrictions/distinctions in copular constructions in adult grammar. Also: read McNally (others?) to understand semantics of existentials; connections between existentials and identificationals. Finish reading Moro. Look at contracted be and metrical accounts of be omission German & Italian data, aspect (tie in Antinucci & Miller?) (Hebrew?) English data check older children's data for later patterns, younger children's data for earlier patterns (pseudoclefts in older kids?) Perform experiments or collect more data (if necessary) and/or work on computational model

Fall 1999/Winter 2000: finish up any data collection/analysis continue reading, begin organizing & writing dissertation

Winter/Spring 2000: continue writing; aim to finish in spring 2000 40

Appendix: More on the linguistic typology of copular constructions A.1 Classical Chinese Classical Chinese expresses existence via the copular element you, translated as 'there is' or 'have'. Its negation is a suppletive form, wu 'there is not', 'there lacks'. While you can be used for inalienable possession (to have a property, e.g. color or shape), I do not know whether it can be used for alienable possession. For expressing predication, Chinese uses the "particle" (I'm not sure if it is a copula) yeh, as in (1):

(1)

Y jen yeh I man yeh

(Graham 1967)

The particle yeh is negated by the negative copula fei. Sentences with yeh sometimes use shih instead, and such sentences (with yeh or shih) reportedly do not express any temporal reference. Classical Chinese has a different copula for expressing identity: chi. I have only one example of an utterance with chi, wherein it occurs with yeh; so I don't know whether chi must always occur with yeh:

(2)

Tao chi hsing yeh. way chi nature yeh "The Way is human nature."

Graham 1967, p. 11

To express adjectival predication, the adjective itself acts as a verb, so no copula is used (this is similar to the verbal behavior of adjectives in Japanese). For expressing location there is a different copula, tsai 'is in.'

A.2 West Greenlandic Eskimo Eskimo, a polysynthetic language, has a number of suffixes that function as copulas (they are suffixes to the root but may be followed by other suffixes). I will only give a description of a

41

few of them. The copular suffixes are denominals; that is, they turn nouns into verbs. One of them is -u-, which seems to be a kind of inchoative, meaning 'causing to be'. (3) sikvoq 'it is ice-covered' (siko = ice) Mey 1968, p.19

There are also a few independent (i.e. non-affixal) verbs (ilivoq 'be in a state, be such' and isivoq 'become, get into a state') and one "semi-independent" verb (-)poq meaning 'to be', which can also be realized as the suffix -it-. For example:

(4)

taimitoq 'it being thus', taimailivoq 'is, becomes like that' (from taima 'thus') Mey 1968, p. 20

The application of -it- and (-)poq is more restricted than the application of -u-, since -it-/(-)poq can only apply to certain adverbs and nominals, "cases expressing quality, manner, location" (Mey 1968, p. 20). Both -u- and -it- serve to link a property to a subject. In addition there is an existential suffix, -qar-, which can be translated as 'there is' or as 'has'. (5) ateqarpunga 'I have the name...', 'my name is...' (from ateq 'name')

The suffix -qar- also seems to be used in creating passives, but the examples are too unclear. Both -u- and -qar- seem to have a much less restricted application than -it-/(-)poq. Similar to Classical Chinese wu 'there is not, lacks', Greenlandic has a negative copula suffix -it- (it has different morphophonemic properties from the -it- discussed earlier) meaning 'to be without'.

A.3 Malayalam Malayalam, a Dravidian language, has three verbs that are used as copulas. One is aan, which is used in nominal (and some adjectival) predicative sentences, to express identity and location. It can be optionally omitted in all of these contexts (Asher 1968). A second copula, unt,

42

which appears in existential, locative and possessive constructions (in possessives the possessor is marked with Dative case). Both aan and unt can be used in psych-verb constructions and locatives, and in each case the choice of the copula yields slightly different shades of meaning. (I don't understand yet exactly what the differences are.) The third copula is irikkuka, which can be glossed as 'sit'. It has more a meaning of 'staying in a place'. The negative forms of unt and ann are suppletive: ann becomes alla, and unt becomes illa. Main verbs are suffixed with -illa to the root when negated. Interestingly, the form irikkunnu (a form of irikkuka 'be/stay/sit') is added to a past tense verb to form the perfective, and unt occurs with a present tense verb to form the progressive.

(6)

a. [aa] ceriya kettitam abdulkhaadar vaatakaykku kotuttirikkunnu that small building Abdulkhadar rent/to given/has "Abdulkhadar has rented out that small building." b. ellaarum enne nookkunnunt all me look at "They are all looking at me." both exx. from Asher 1968, p.109

These facts are in line with something Benveniste (1966) noted about auxiliary verbs in both Iranian and Classical Armenian: namely, in both possessives and past participles, the verb is be, and the subject/possessor is marked with oblique case (Dative for Iranian, Genitive for Armenian). What this shows is that the connection between auxiliary verbs such as be and have, and their "main verb counterparts", i.e. verbs that express existence and possession, is in fact a robust connection that holds across languages (English, Malayalam, Iranian, Armenian), and it is probably not coincidental.

Bibliography
Akmajian, A. (1970) Aspects of the grammar of focus in English. Ph.D. dissertation, MIT. Asher, R.E. (1968) Existential, possessive, locative and copulative sentences in Malayalam. In J.W.M. Verhaar (ed.), The verb be and its synonyms. vol. 2, 88-111. Becker, Misha (1995) Acquisition of syntax in child German: Verb finiteness and verb placement. BA thesis, Wellesley College.

43

Becker, Misha (1997a) The Some Indefinites. Paper presented at Colloque de Syntaxe et Semantique de Paris, October 1997. Becker, Misha (1997b) Acquiring a grammar with covert head incorporation: The case of have as a complex verb, MA thesis, UCLA. Becker, Misha (1998) Have as Be+P: Evidence from similarities in the acquisition of have and existential be, UCLA ms. Becker, Misha (1999) The acquisition of modals and modality in child German. Paper presented at Linguistic Society of America annual meeting, Los Angeles, January 1999. Benveniste, E. (1966) Problmes de linguistique gnrale. Paris: Gallimard. Bloom, Lois (1970) Language development: Form and function in emerging grammars. Cambridge, MA: MIT Press. Brown, Roger (1973) A first language. Cambridge, MA: Harvard University Press. Chomsky, Noam (1993) A minimalist program for linguistic theory. In K. Hale & S.J. Keyser (eds), The view from Building 20. Cambridge, MA: MIT Press. 1-52. Den Dikken, Marcel (1995) Copulas. Paper presented at GLOW, Troms. Freeze, Ray (1992) Existentials and other locatives. Language, vol.68, no.3. Giorgi, Alessandra & Fabio Pianesi (1996) Tense and aspect: from semantics to morphosyntax. ms. University of Bergamo & IRST. Graham, A.C. (1967) 'Being' in Classical Chinese. In J.W.M. Verhaar (ed.), The verb be and its synonyms. vol.1, 1-39. Heggie, Lorie (1988) The syntax of copular structures. doctoral dissertation, USC. Heller, Daphne (1999) The syntax and semantics of pseudoclefts and two pronominal copulas in Hebrew. chapter 2 of MA thesis, Univ. Tel Aviv. Higgins, F. R (1973) The pseudocleft construction in English. doctoral dissertation, MIT. Hoekstra, Teun & Nina Hyams (1998a) Agreement and the finiteness of V2: Evidence from child language. Procedings of the 22nd annual BUCLD, Brookline: Cascadilla Press. Hoekstra, Teun & Nina Hyams (1998b) Aspects of Root Infinitives, to appear in Proceedings of GALA 1997 (?) Hoekstra, Teun & Rene Mulder (1990) Unergatives as copular verbs: Locational and existential predication. The Linguistic Review, vol.7, p.1-79. Iatridou, Sabine & Spyridoula Varlokosta (1995) Pseudoclefts crosslinguistically, UPenn ms. Kayne, Richard (1993) Toward a modular theory of auxiliary selection. Studia Linguistica 47.1, 3-31. Kondrashova, Natalia (199x) The Russian copula: A unified approach. source? MacWhinney, Brian & Catherine Snow (1985) The Child Language Data Exchange System. Journal of Child Language 12, 271-296. Mahajan, Anoop (1994). The Ergativity Parameter: have-be alternation, word order and split ergativity. Proceedings of NELS 24. M. Gonzlez (ed.) Amherst, MA: GLSA. Mey, Jacob (1968) On the notion of 'to be' in Eskimo. In J.W.M. Verhaar (ed.), The verb be and its synonyms. vol. 2, 1-34. Milsark, Gary (1974) Existential sentences in English, doctoral dissertation, MIT. Moro, Andrea (1997) The raising of predicates: Predicative noun phrases and the theory of clause structure. Cambridge: Cambridge University Press.

44

Rapoport, T. (1987) Copular, nominal, and small clauses: A study of Israeli Hebrew, doctoral dissertation, MIT. Sachs, J. (1983) Talking about the there and then: The emergence of displaced reference in parent-child discourse. In K.E. Nelseon (ed.), Children's Language, Vol. 4, Hillsdale, NJ: Lawrence Erlbaum. Sera, Maria. 1992. To be or to be. Journal of Memory and Language, vol.31, p.408-427. Stowell, Tim (1978) What was there before there was there? in D. Farkas et al., eds., Proceedings of the 14th Regional Meeting of the Chicago Linguistics Society. Stowell, Tim (1981) Origins of Phrase Structure, doctoral dissertation, MIT. Suppes, Patrick (1973) The semantics of childrens languages. American Psychologist, vol. 88. 103-114. Ud Deen, Kamil (1997) The interpretation of root infinitives in English: is eventivity a factor? ms. UCLA. Wagner, K. (1985) How much do children say in a day. Journal of Child Language 12, p. 475-487. Wexler, Ken (1994) Optional infinitives, head movement and the economny of deriviations. In Lightfoot & Hornstein (eds.), Verb Movement. Cambridge: CUP. Williams, Edwin (1983) Semantic vs. syntactic categories. Linguistics and Philosophy, vol.6, no.3, pp.423-446. Williams, Edwin (1984) There-Insertion. Linguistic Inquiry, vol.15, no.1, pp. 131-153. Wyatt, Toya (1995) Acquisition of the African-American English copula. Communication Development and Disorders in African-American Children.

45

Вам также может понравиться