* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Investigating the abstractness of children`s early knowledge of
Modern Greek grammar wikipedia , lookup
French grammar wikipedia , lookup
Kannada grammar wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
English clause syntax wikipedia , lookup
Chinese grammar wikipedia , lookup
Udmurt grammar wikipedia , lookup
Polish grammar wikipedia , lookup
Ojibwe grammar wikipedia , lookup
Portuguese grammar wikipedia , lookup
Macedonian grammar wikipedia , lookup
Navajo grammar wikipedia , lookup
Old Norse morphology wikipedia , lookup
Old Irish grammar wikipedia , lookup
Proto-Indo-European verbs wikipedia , lookup
Ukrainian grammar wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Spanish grammar wikipedia , lookup
Swedish grammar wikipedia , lookup
Japanese grammar wikipedia , lookup
Latin syntax wikipedia , lookup
Modern Hebrew grammar wikipedia , lookup
Ancient Greek verbs wikipedia , lookup
Germanic strong verb wikipedia , lookup
Icelandic grammar wikipedia , lookup
Russian grammar wikipedia , lookup
Germanic weak verb wikipedia , lookup
Yiddish grammar wikipedia , lookup
Georgian grammar wikipedia , lookup
Hungarian verbs wikipedia , lookup
Serbo-Croatian grammar wikipedia , lookup
Old English grammar wikipedia , lookup
Italian grammar wikipedia , lookup
Sotho verbs wikipedia , lookup
German verbs wikipedia , lookup
Lexical semantics wikipedia , lookup
Journal of Child Language http://journals.cambridge.org/JCL Additional services for Journal of Child Language: Email alerts: Click here Subscriptions: Click here Commercial reprints: Click here Terms of use : Click here Investigating the abstractness of children's early knowledge of argument structure KATHLEEN MCCLURE, JULIAN M. PINE and ELENA V. M. LIEVEN Journal of Child Language / Volume 33 / Issue 04 / November 2006, pp 693 - 720 DOI: 10.1017/S0305000906007525, Published online: 08 November 2006 Link to this article: http://journals.cambridge.org/abstract_S0305000906007525 How to cite this article: KATHLEEN MCCLURE, JULIAN M. PINE and ELENA V. M. LIEVEN (2006). Investigating the abstractness of children's early knowledge of argument structure. Journal of Child Language, 33, pp 693-720 doi:10.1017/S0305000906007525 Request Permissions : Click here Downloaded from http://journals.cambridge.org/JCL, IP address: 194.94.96.194 on 25 Nov 2015 J. Child Lang. 33 (2006), 693–720. f 2006 Cambridge University Press doi:10.1017/S0305000906007525 Printed in the United Kingdom Investigating the abstractness of children’s early knowledge of argument structure* KATHLEEN MCCLURE Lehman College, City University of New York J U L I A N M. P I N E University of Liverpool AND E L E N A V. M. L I E V E N Max Planck Institute for Evolutionary Anthropology, Leipzig (Received 15 March 2004. Revised 23 August 2005) ABSTRACT In the current debate about the abstractness of children’s early grammatical knowledge, Tomasello & Abbott-Smith (2002) have suggested that children might first develop ‘weak ’ or ‘partial’ representations of abstract syntactic structures. This paper attempts to characterize these structures by comparing the development of constructions around verbs in Tomasello’s (1992) case study of Travis, with those of 10 children (Stage I–II) in a year-length, longitudinal study. The results show some evidence that children’s early knowledge of argument structure is verb-specific, but also some evidence that children can generalize knowledge about argument structure across verbs. One way to explain these findings is to argue that children are learning limited scope formulae around high frequency subjects and objects, which serve as building blocks for more abstract structures such as S+V and V+O. The implication is that children may have some verb-general knowledge of the transitive construction as early as Stage I, but that this knowledge is still far from being fully abstract knowledge. [*] Address for correspondence : Kathleen McClure, Lehman College, CUNY, Department of Speech-Language-Hearing Sciences, 250 Bedford Park Boulevard West, Bronx, NY 10468, USA. tel : 1-718-960-8460; e-mail : [email protected] 693 M C C L U R E E T A L. INTRODUCTION In recent years there has been a debate about how abstract children’s knowledge is at the beginning of multi-word speech. On one side is Tomasello (2000) and his constructivist account of early grammatical development ; on the other side, is Fisher (2002 a, b) and a syntactic bootstrapping proposal. For Tomasello, children do not start out with abstract syntactic categories ; rather, grammatical structure builds around individual lexical items and these item-specific constructions only gradually become more general. Tomasello outlines a position in stark contrast to that proposed in generative accounts of grammatical development, which ascribe adultlike syntactic categories to the child at the beginning of word combination. Although Fisher does not specify just how much syntactic competence children start out with, she cites results from comprehension studies that point to children having more abstract knowledge and earlier than Tomasello proposes. In response, Tomasello & Abbot-Smith (2002) have suggested that in the process of moving towards the adult system from low-scope representations, children might develop ‘weak ’ or ‘ partial’ representations of more abstract linguistic structures. However, they do not specify in detail what these representations might consist of. This paper attempts to give them a preliminary characterization by looking at the development of constructions around verbs at the earliest stages of syntactic development. The recent Tomasello–Fisher debate concerns two main questions : (1) how lexically specific is children’s early grammatical knowledge and (2) at what point do children possess abstract syntactic categories. According to Tomasello (2000) young children do not come to the task of combining words with ‘ abstract categories and schemas ’. Instead their earliest combinations ‘ revolve around concrete items and structures ’ (Tomasello, 2000 : 215). These specific words and phrases develop in their own way depending upon the individual child’s experience. For example, children do not start out with an abstract ‘ verb’ category or an abstract ‘transitive ’ or ‘ intransitive ’ construction with which to assimilate new verbs. Rather they learn the argument structure of individual verbs on a case-by-case basis. They learn, for example, that the verb hit can have a hitter, a thing hit, and a thing hit with argument. This lexically specific view of early syntactic development derived from observational as well as experimental data. Tomasello (1992) conducted a detailed diary study of all the different uses of his daughter Travis’s early verbs and word combinations from age 1;0 to 2;0. He found that Travis was conservative in how she used her verbs. Almost half of the 162 verbs and relational words she produced were used in only one verb– argument construction type (e.g. ‘ Mommy break ’ is one type, ‘break cup ’ is another). Furthermore, he found that the verbs developed grammatical 694 INVESTIGATING ABSTRACTNESS structure in quite individual ways. For semantically similar verbs like cut and draw, where one would expect to find the child using them in the same construction type (cut with, draw with), he found that cut was used in only one, whereas draw was used in many. He also found unevenness in how certain arguments were used across verbs. For example, at the same stage of development one verb might be marked with an instrument whereas another would not. These findings led Tomasello to conclude that, with the exception of some kind of ‘ noun’ category, early language learners did not possess syntactic categories. Instead their earliest categories were ‘ lexically specific categories such as ‘ kisser’, ‘ kissee’, ‘seer ’, ‘thing seen ’, ’ etc. (Tomasello, 2000 : 215). This lexically specific view has been supported by other observational studies (Pizzuto & Caselli, 1992 ; Pine & Lieven, 1993 ; Pine & Martindale, 1996 ; Lieven, Pine & Baldwin, 1997; Pine, Lieven & Rowland, 1998 ; Rubino & Pine, 1998). Although Tomasello’s case study was highly informative about what one particular child did and did not produce, it could not control for what the child heard and thus address the issue of how productive her language was. To do this, experimental studies were needed. Tomasello and his colleagues conducted a series of experiments (see Tomasello, 2000, for a review) in which they taught children new verbs in one construction frame and investigated whether they could use them productively in another. If they could, then this would imply the presence of syntactic categories. If, on the other hand, the children used the new verbs conservatively, that is, only in the ways they had heard them, then this would imply the lack of an abstract syntactic system, and that children were learning the verbs individually. The experiments involved presenting children with new verbs in different constructions (e.g. intransitive, passive, and nonSVO word order) and then seeing whether they could produce them in correct transitive constructions. For example, Tomasello & Brooks (1998) presented 16 children at 2 ; 0 and 16 children at 2;6 with a new verb modelled as an intransitive (e.g. The ball is dacking) and another as a transitive ( Jim is tamming the car). The experimenter then tried to elicit transitive uses by asking, ‘ What’s X doing? ’ with X being the agent. If the child was able to say, ‘ He’s dacking the ball ’, this would mean he was using the verb productively because he had switched from intransitive to transitive, and if he said, ‘ He’s tamming the car ’, this would mean he was using the verb in the way it had been modelled. Results showed that although 11 out of 16 of the two-year-olds and all of the older children could produce a new transitive sentence, only one two-year-old and three of the older children could switch from the intransitive model to the transitive. This developmental pattern was found in numerous experiments using different constructions, though most involved the transitive (see Tomasello, 2000, Figure 1). The youngest children (under age 3;0) were found least 695 M C C L U R E E T A L. able to produce a transitive utterance when they had heard a new verb modelled in another construction, but older children (above 3; 0) were able to do so. Tomasello (2000) interpreted this reluctance of children to innovate early on as an indication that they lacked verb-general knowledge and an abstract transitive construction. Fisher (2002a), however, challenges this interpretation. She argues that Tomasello’s experiments rest on faulty assumptions. First, she claims that Tomasello’s experiments rest on the assumption that children will infer ‘ that any new verb could be used transitively ’, which is why when children fail to produce transitives in elicitation tasks, this is taken as evidence that they do not have an abstract transitive construction. However, she argues, children’s failure may reflect the fact that they have not yet learned that a verb used in one particular context could be used in another. This takes time and experience and ‘ only after learning a great deal about how a particular language organizes its verbs could a child guess that a new verb can be extended into a new syntactic structure ’ (Fisher, 2002 a : 264). This is what one would expect, she says, and is a better explanation of the gradual trajectory found in Tomasello’s (2000) Figure 1 summary of his results. Here Fisher seems to agree with Tomasello’s characterization of children’s early verb knowledge as lexically specific, but disagree with him that the interpretation of his results show that children lack abstract syntactic knowledge at the very beginning of word combination. Tomasello & Abbot-Smith (2002) reply that the verbs used in the relevant experiments (except for those in the earliest studies) were all carefully chosen so that they would ‘ fall into the appropriate semantic classes, as outlined, for example in Levin (1993), thus ensuring that they were of a type that adults use both as transitives and in other constructions depending upon the study (e.g. active–passive or transitive–intransitive alternation) ’ (Tomasello & Abbot-Smith, 2002 : 208). Second, Fisher claims that Tomasello does not consider any other reasons besides the lack of syntactic categories to explain children’s reluctance to innovate with new verbs. Performance factors such as syntactic priming may explain the conservative findings. Again Tomasello & Abbot-Smith (2002) respond that Fisher has not accurately represented Tomasello (2000) and that three alternative explanations and possible controls that seemed to rule them out were discussed. Third, Fisher claims that children may not interpret new verbs in the ways the experimenters expect. For example, if children do not interpret the new verbs as causative, they would not then switch to the transitive. Tomasello & Abbot-Smith (2002) acknowledge that some children in the intransitive–transitive studies might have done this. However, they argue that this criticism does not apply to all the studies (e.g. studies of the active–passive constructions), because it focuses on differences in the 696 INVESTIGATING ABSTRACTNESS number of arguments around verbs in the transitive and intransitive studies, which is not the case in the active–passive studies. Tomasello is quite clear in his claim that children’s early grammatical knowledge is lexically specific ; whereas Fisher’s (2002 a) position seems to be that children’s early grammatical knowledge is a mixture of abstract and item-based knowledge. Thus, she argues that children must have some kind of relatively abstract ‘ noun’ category, but also ‘abstract representations of sentence structure ’ (Fisher, 2002 b : 56) that allow them to interpret sentences and ‘ structure-matching biases’ (Fisher, 2002 a : 274) that allow them to map structures found in the sentence to referents in the scene being observed. For example, children use the number of nouns in an utterance in order to interpret the meanings of sentences ‘by mapping an utterance containing two known nouns onto a salient conceptual relation between their referents, and an utterance containing one noun phrase onto the most salient conceptual predicate involving the referent of that noun ’. The number of nouns gives a ‘ probabilistic estimate of the number of syntactic arguments of the sentence’s verb’ (Fisher, 2002 a : 273). Thus, the structure of a sentence helps the child interpret it. If there are two nouns, the child matches this structure to a conceptual representation of a particular event with two participants. If there is only one, it is matched to an event with only one participant (Fisher, 1996). The second major issue concerns when children begin to have abstract grammatical knowledge. Tomasello’s (2000) position is that children do not have innate abstract grammatical categories. Instead these grammatical categories grow gradually as the child has more and more experience with individual lexical items. For example, an abstract ‘verb ’ category develops as children become more familiar with different verbs and their argument structure. Although Fisher (2002 a) agrees with Tomasello that children do not start out with ‘ adult syntactic competence ’ (Fisher, 2002 a : 272), she argues that children have abstract knowledge earlier than Tomasello claims (i.e. earlier than 3 ; 0). As evidence she cites Brooks & Tomasello’s (1999), study 2, in which children aged 2 ; 10 were presented with new verbs in both an active and a passive sentence and then encouraged to produce a sentence in a new frame. Brooks & Tomasello found that 40% of the children who were presented with the active sentence were able to switch to the passive and 35% of those presented with the passive could produce the active. These were higher rates than those found in study 1, and Brooks & Tomasello attribute the difference to priming effects. The children in study 2 were trained on a nonce verb in both active and passive utterances. Fisher notes the high percentages and concludes that these children (who were younger than 3; 0) also had some ‘ notion of both active transitive and passive sentence constructions that was already abstract enough to be primed across different newly-learned verbs ’ (Fisher, 2002 a : 270). 697 M C C L U R E E T A L. Furthermore, Fisher (2002 a) cites a study by Akhtar (1999) in which children at 2 ; 8, 3 ; 6, and 4; 4 were trained with new verbs in three different word orders : SVO, SOV and VSO. Akhtar found that the younger children (2 ; 8, 3; 6) used the non-SVO word orders as often as they corrected them to SVO word orders ; whereas the older children (4;4) corrected their utterances to SVO almost all of the time. These findings support Tomasello’s claim that the process of acquiring an abstract grammatical structure is a gradual one. However, they also show more verb-general knowledge in younger children than had previously been thought. When presented with a new verb in an odd word order and encouraged to respond, about half of the two-year-olds switched to the SVO word order, and when presented with a new verb in the SVO order, all of them responded with an SVO order. This demonstrates some knowledge of the transitive construction. Akhtar concludes that between the ages of 2 ;0 and 4; 0 children are developing an understanding of what word order is in English : ‘ Whereas the younger children seem to be in the process of constructing a truly general understanding of the syntactic significance of word order (that all English sentences must employ SVO order), the four-year-olds were simply not willing to use the non-SVO structures ’ (Akhtar, 1999 : 354). Finally, Fisher (2002 a) presents evidence from comprehension studies (Naigles, 1990 ; Naigles & Kako, 1993) that children have more abstract knowledge of the transitive construction than Tomasello’s (2000) production data demonstrate. She cites a preferential looking study by Naigles (1990) in which children with a mean age of 2;1 looked longer at a videorecorded scene showing a transitive action (a duck bending a rabbit over) when they heard a novel verb used in a transitive frame and longer at a scene showing an intransitive action (a duck and a rabbit making circles with one of their arms) when they heard a novel verb used in an intransitive frame. This is taken to mean that the children used sentence structure (either transitive or intransitive) to help them learn the meanings of new verbs, and hence that they did have some abstract knowledge of these structures. In response to Fisher, Tomasello & Abbot-Smith (2002) concede that children under 3 ; 0 may have more abstract syntactic knowledge than had previously been demonstrated, acknowledging Akhtar’s (1999) results and citing similar results in Abbot-Smith, Lieven & Tomasello (2001). The new finding in Abbot-Smith et al.’s study is that the youngest children (2 ; 4) corrected the non-canonical word orders 21 % of the time, which is less than half that of the children aged 2; 8 in Akhtar’s study. However, this is what one would expect if the children were acquiring a transitive schema gradually. The children also demonstrated some familiarity with the transitive construction – they used the new verb they heard in the 698 INVESTIGATING ABSTRACTNESS SV order more often than the new verb in the VS order. To explain these findings Tomasello & Abbot-Smith suggest that perhaps these very young children have a ‘ weak transitive schema – one that enables certain kinds of linguistic operations but not others – whereas older children have a stronger and more robust schema based on a wider range of stored linguistic experience’ (Tomasello & Abbot-Smith, 2002 : 210). Tomasello & Abbot-Smith (2002) also use this notion of a ‘weak verb-general representation of the transitive construction ’ (Tomasello & Abbot-Smith, 2002 : 212) to explain the results from the preferential looking studies that Fisher (2002a) cites. A weak verb-general representation would mean that the children have partial knowledge of what is involved in the transitive construction. It might explain why two-year-olds in comprehension tasks were found to be sensitive to a difference between transitive and intransitive but similar aged children in production tasks (Akhtar & Tomasello, 1997) did not do well in acting out transitive sentences with novel words. The performance tasks of the older children may have required a stronger cognitive representation of the category. In summary, there is evidence from numerous observational (Tomasello, 1992 ; Pine & Lieven, 1993; Pine & Martindale, 1996 ; Lieven, Pine & Baldwin, 1997 ; Pine, Lieven & Rowland, 1998) as well as experimental studies (cited above) that children’s early grammatical knowledge is lexically specific or limited in scope. For example, Tomasello’s (1992) findings from the case study of his daughter indicated that syntactic structure developed around individual verbs in a lexically specific way. In what became known as the VERB ISLAND HYPOTHESIS, verbs had a special role to play. They acted like ‘ islands’ around which children attached arguments and syntactic markings ‘ in an otherwise unorganized grammatical system ’ (Tomasello, 1992 : 23). However, despite this evidence of verb-specific knowledge, there is also evidence of some more verb-general knowledge building up in children between 2 ; 0 and 3; 0 (Akhtar, 1999; Brooks & Tomasello, 1999 ; Abbot-Smith et al., 2001; Tomasello & Abbot-Smith, 2002). To investigate just how verb-specific early grammatical knowledge is, this study seeks to compare Tomasello’s (1992) findings of one child’s development with those of a larger group of children. The aim was to investigate : ’ ’ ’ To what extent Tomasello’s description of Travis generalizes to a larger group of children To what extent children’s knowledge seems to be growing gradually around specific lexical items To what extent children show evidence of verb-general knowledge in the early stages. 699 M C C L U R E E T A L. It was also hoped that, by identifying instances of verb-general knowledge at the early stages, it would be possible to develop a more precise description of what ‘ partial’ knowledge of syntactic constructions might actually consist of. METHOD Participants Ten children participated in the study (6 girls, 4 boys). They were part of a year-length longitudinal study of early language development of 12 first-born children. Two children were excluded from the present study because they did not have Stage I data. The children were recruited from newspaper advertisements, local nurseries, and doctors ’ surgeries. They were all from monolingual English-speaking families and were, for the most part, from middle-class homes. Their ages ranged between 1;10.7 and 2 ;0.7 at the beginning of the study (M=23.12 months) and 2;03 and 3 ;0.10 at the end (M=27.2 months). The children’s MLUs were between 1.06 and 2.01 at the beginning (M=1.46) and between 2.03 and 3.261 at the end (M=2.43). Table 1 presents the age, MLU, and number of audio-recorded sessions at Stage I and Stage II for all the children in the study. Note that the MLU and age ranges refer to the first and last recording for each stage. Design of study In his case study, Tomasello (1992) observed his daughter Travis from age 1 ; 0 to 2 ; 0 and kept a detailed diary record of all the verbs and verb combinations she produced by age of acquisition. Beginning when she was 1; 5, Travis was video-recorded at the start of each month for an hour until the end of the study. Her MLU at the first recording was 1.56 and her MLU at the final recording was 2.58. MLUs for the recordings in between are not provided. However, since it is clear that Tomasello’s study focuses primarily on data from Stage I and Stage II, only data from Stage I (MLU, 1.0–1.99 [Brown, 1973]) and Stage II (MLU, 2.0–2.49) were used in the present study, and all of the developmental analyses involved comparisons between Stage I and Stage II data. This design obviously prevents us from looking for developmental changes in children’s language within [1] To make a transition from one stage to another, a child had to have three consecutive sessions (hours of data) in which the MLUs were above the relevant borderline. The MLU of 3.26 reflects a particularly high MLU in one session for a child who failed to meet the criterion for transition to Stage III (because her MLU subsequently dropped below 2.5). 700 INVESTIGATING ABSTRACTNESS TABLE 1. Age, MLU, and number of audio-recorded sessions for the Manchester children at Stage I and Stage II Children Aran Age MLU Tape Session Becky Age MLU Tape Session Dominic Age MLU Tape Session Anne Age MLU Tape Session Ruth Age MLU Tape Session Gail Age MLU Tape Session Joel Age MLU Tape Session Liz Age MLU Tape Session Nicole Age MLU Tape Session Warren Age MLU Tape Session Stage I (No.) 1; 11.12–2; 0.2 1.41–1.83 1–3 Stage II 2; 0.9–2; 1.28 2.22–2.27 4–8 (No.) 2; 0.7–2; 2.22 1.46–1.97 1–8 2; 2.30–2; 3.20 2.06–2.41 9–11 (No.) 1; 10.24–2; 1.26 1.20–1.78 1–10 2; 2.9–2; 5.22 2.12–2.48 11–21 (No.) 1; 10.7–1; 11.20 1.61–1.92 1–6 2; 0.15–2; 1.18 2.27–2.21 7–10 (No.) 1; 11.15–2; 3.6 1.41–1.97 1–12 2; 3.25–2; 7.24 2.04–2.03 13–25 (No.) 1; 11.27–2; 0.19 1.76–1.88 1–3 2; 0.25–2; 2.12 2.04–2.42 4–8 (No.) 1; 11.1–2; 1.10 1.33–1.87 1–8 2; 1.23–2; 4.2 2.00–2.48 9–16 (No.) 1 ;11.9–2; 1.21 1.35–1.88 1–5 2; 0.28–2; 2.30 2.02–2.42 6–12 (No.) 2; 0.25–2; 6.16 1.06–1.71 1–17 2; 6.23–3; 0.10 2.04–3.26 18–34 (No.) 1 ;10.6–1; 10.15 2.01–1.95 1–2 1; 10.29–2; 0.3 2.36–2.33 3–6 stages (Ingram, 1989). However, it does increase the reliability of those comparisons that are made by ensuring that they are based on relatively large amounts of data for each child for each developmental stage. Three predictions that derive from the Verb Island hypothesis were tested. The first concerns argument structure and the way that it builds around individual verbs. Tomasello (1992) hypothesized that syntactic 701 M C C L U R E E T A L. structure builds gradually from simple to complex; therefore, one can predict that : 1. Few verbs will first appear in multi-argument structures. This was true for Travis, but would it generalize to other children as well? The second prediction concerns how lexically specific children’s early multi-word speech is. If verbs show individual developmental trajectories, as Tomasello proposes, then structure should be more complex around verbs that the child has known for a longer period of time. Using Stage II data one can test this by comparing verbs that had been acquired at Stage I and then reappeared at Stage II (called Old Verbs), with verbs that had just been acquired at Stage II (called New Verbs). One can therefore predict that : 2. Utterances with Old Verbs at Stage II will have more complex structures than utterances with New Verbs at Stage II. Complexity was measured by calculating the MLUs and the number of arguments per utterance. MLUs for both tokens and first uses of verb types were calculated. One would expect that the MLUs for utterances with Old Verbs would be greater than the MLUs of utterances with New Verbs. However, as MLU measures more than just argument structure, a more precise test of the hypothesis is to measure the number of arguments per utterance for Old and New Verbs. One would predict that utterances with Old Verbs would have more arguments than utterances with New Verbs. A third prediction concerns whether children are acquiring any verbgeneral knowledge. According to Tomasello, knowledge about structure from verbs already learned should not generalize to new verbs. To test this prediction New Verbs at Stage I were compared to New Verbs at Stage II. The specific prediction was that : 3. Utterances with New Verbs at Stage II should be no more complex than utterances with New Verbs at Stage I. If children were generalizing some of their knowledge about the verbs they learned at Stage I to the verbs they were learning at Stage II, then utterances with New Verbs at Stage II would have more complex structures than utterances with New Verbs at Stage I. Procedure The children were audio-recorded in their homes for one hour twice every three weeks for a year. They were recorded interacting with their mothers, half the time playing with their own toys in a free play session, and half the time with toys brought by the investigator in a structured play session. 702 INVESTIGATING ABSTRACTNESS In the structured session the investigator introduced a number of additional toys to stimulate the child’s interest in play ; however, the child was not restricted to playing with these toys in any way. The investigator did not play with the child, but did respond to talk. The recordings were transcribed orthographically using the CHAT system from the CHILDES project (MacWhinney, 2000) and can be accessed from the CHILDES archive as the Manchester Corpus (Theakston, Lieven, Pine, & Rowland, 2001). Children’s speech corpora. The corpus for each child was searched for utterances that included verbs at Stage I and II, and these were extracted from the transcripts and collected into separate files. Excluded were incomplete or unclear utterances, imitations, self-repetitions, and routines (songs, counting sequences). A verb was defined in the adult sense of the word. This definition departs slightly from Tomasello’s (1992) definition, which also included predicates such as off, on, over, up, and down. Once these files had been organized, they were further divided into three categories of verbs per child : (1) Stage I New Verbs : all verbs acquired at Stage I; (2) Stage II Old Verbs : verbs that had first appeared at Stage I and then reappeared at Stage II; and (3) Stage II New Verbs: verbs that had only first appeared at Stage II. Excluded from these categories were copula constructions, double-verb constructions (e.g. want to go) and formulaic expressions (e.g. thanks, thank you, excuse me, bless you, pardon me). Three additional categories were then organized per child, but this time they only included first uses of Old and New Verbs at Stage I and II. Note that since the Manchester corpus does not include any maternal diary data, first use simply refers to the first utterance including a given verb in the child’s spontaneous speech data from a particular MLU stage. Coding. In order to evaluate the predictions concerning the number of arguments per verb, all first uses of Old and New Verb types were coded for the number of arguments they appeared with. Argument was defined, as per Tomasello (1992) to mean constructions developing around the verb. Because the Verb Island hypothesis proposes that the child does not start out with abstract argument structure but has to construct it, each verb has its own argument types and by rights should have its own terminology. However, developing verb-specific terminology that is not overly cumbersome (e.g. ‘ the one who hits’ and ‘thing hit ’) is not easy, and Tomasello compromises and uses terms such as Actor (to include agent and experiencer), Object (to include patient and theme), Instrumental, Locative, and Recipient. As the purpose of this study was to compare Travis to a larger group of children, these categories were also used and refer to 703 M C C L U R E E T A L. TABLE 2. Argument frames 1-Argument frames ___OBJ ___LOC ___REC __ACT ACT___ INST___ OBJ___ 2-Argument frames ‘ Climb it’. (Aran) ‘ Crawl in dark ditch’. (Aran) ‘ Give Caroline’. (Gail) ‘ Read rabbit’. (Warren) ‘ Lady repair’. (Aran) ‘ The key start’. (Aran) ‘ Cake drop’. (Aran) ACT___OBJ ACT___REC ACT___LOC OBJ___LOC REC___OBJ OBJ__REC ___OBJ LOC ___OBJ ACT ___REC OBJ ___OBJ REC ‘ I shot a dog ’. (Aran) ‘ I show lady’. (Aran) ‘ I sit there’. (Warren) ‘ Tractor lives here’. (Aran) ‘ That called Diesel’. (Dominic) ‘ They fit me, no ’. (Ruth) ‘ Put him in train’. (Gail) ‘ See it Anne’. (Anne) ‘ Feed duck bread’. (Dominic) ‘ Make tea a baby please’. (Ruth) ACT OBJ___ ACT LOC___ OBJ LOC___ OBJ ACT__ LOC OBJ___ ‘ Mama me stop’. (Ruth) ‘ Me on there knock’. (Ruth) ‘ Me there roll too’. (Ruth) ‘ What do you think?’ (Gail) ‘ There it goes ’. (Anne) 3-Argument frames ACT___OBJ LOC ACT___OBJ REC ACT___REC OBJ ‘ I load one there’. (Aran) ‘ Nicola give that me’. (Aran) ‘ I’ll tell you what’. (Anne) NPs required by the verb, NPs in prepositional phrases not necessarily required by the verb, and the locative adjuncts there and here. Each argument was also categorized as to whether it contained one or more words. The verbs were then organized into three tables for each child (New Verbs at Stage I, New Verbs at Stage II, and Old Verbs at Stage II) according to the number of arguments they first appeared with : zero, one, two, or three and their argument frames. Table 2 shows all the argument frames that were found along with an example utterance from the data. The reliability of the argument frame coding was assessed by having a second independent coder re-code 400 of the 725 utterances to which argument frames were assigned. Agreement was a satisfactory 94.8% (Kappa=0.93). Analyses. Once files for Old and New Verbs for each child were set up, the MLUs for tokens and the MLUs for first uses of Old and New Verb types were calculated. The average number of arguments per utterance was then calculated for each child. A larger figure for Old Verbs meant 704 INVESTIGATING ABSTRACTNESS that these verbs had more surrounding argument structure. Because an argument can be more than one word long, the differences in MLU and in the number of arguments were examined to see to what extent any differences were due to differences in the number of arguments and differences in the complexity of the arguments. Finally, the percentage of multi-argument utterances was calculated. According to the hypothesis, few verbs should first appear with two or more arguments. Two additional types of analysis were conducted to investigate the extent to which the results generalized across different types of verbs. The first of these involved calculating the number of transitive, mixed and intransitive verb types that first appeared with more than one argument in order to see whether transitive verbs were more likely to appear in multi-argument structures than mixed and intransitive verbs. All verbs both for Travis and for the children in the present study were divided into fixed transitive, mixed, or fixed intransitive verbs. ‘ Fixed transitive ’ or ‘fixed intransitive ’ verbs were verbs that were categorized as only transitive or only intransitive in Levin (1993). ‘ Mixed’ verbs were those listed as both transitive and intransitive in Levin. If the verbs were not included in Levin, then the same procedure was followed using the Oxford English Dictionary, but ignoring rare, idiomatic and literary uses. The reliability of this procedure was assessed by recoding the verbs found in Levin (1993) using the Oxford English Dictionary. Agreement was a satisfactory 93.9 % (Kappa=0.85). The second type of analysis involved repeating the Old versus New Verbs at Stage II and New Verbs at Stage I and Stage II analyses on transitive verbs only. Since, in practice, the category of mixed verbs includes many verbs that are unlikely to be mixed from the child’s point of view, for the purposes of this analysis, transitive verbs were defined as verbs for which the utterance being examined either included a direct object argument or lacked an obligatory direct object argument. This allowed us to include in the analysis mixed verbs that were clearly being used transitively by the child rather than restricting the analysis to an artificially small group of transitive verbs that could never be used intransitively even in the adult language. Note that if we had not used this more realistic child-centred definition of transitivity, there would have been so few fixed transitive verbs in the relevant subsections of some of the children’s data that meaningful comparison of the number of arguments with which children used Old and New transitive verbs during Stage II and transitive verbs at Stage I and Stage II would not have been possible. RESULTS Table 3 presents the number of New Verbs (types) at Stage I and New and Old Verbs (types) at Stage II for the children in this study. If one adds 705 M C C L U R E E T A L. TABLE 3. Number of verb types for Stage I New Verbs and Stage II New and Old Verbs for each child Children Stage I New Verbs Stage II New Verbs Stage II Old Verbs Aran Becky Dominic Anne Ruth Gail Joel Liz Nicole Warren 56 89 49 87 15 65 65 48 99 35 52 24 50 26 61 60 65 63 70 35 40 48 41 58 12 43 46 38 83 28 the number of Stage I New Verbs and Stage II New Verbs, one gets the total number of verb types per child. The table shows that, contrary to what one might expect, the number of Old Verbs in Stage II is always lower than New Verbs in Stage I. This is because not all the verbs that first appeared in Stage I reappear in the Stage II data. Multi-argument analysis The central claim of the Verb Island hypothesis is that verbs first appear in simple structures and show individual developmental trajectories. The Verb Island hypothesis therefore predicts that few verbs should first appear with two or more arguments. The results for the multi-argument analysis are presented in Table 4. The children in this study are compared with Travis in terms of how many verbs first appear in the data with two or more arguments and the results show that the first prediction was confirmed. The table shows that in Tomasello’s (1992) study, only 16 of Travis’s 115 verbs, or 13.9 %, first appeared with two or more arguments. The total number of Travis’s verb types was 162. This number, however, included 22 relational words, 12 holophrases, and several instances of different forms of the same verb. Since the present study did not include relational words, holophrases, or duplicate forms and the purpose was to compare the children in this study as closely as possible with Travis, these words were excluded from Tomasello’s total. The table also shows that the children have a mean of 19.8% (range, 8.8 %–37.3%) of their verbs first appearing with more than one argument. From the table, which combines Stage I New Verbs and Stage II New Verbs to make it easier to compare with Travis, it is clear that most of the children do in fact look like Travis. For example, Becky has 9% of her verbs first appearing with two or more 706 INVESTIGATING ABSTRACTNESS T A B L E 4. Comparison of Travis with Manchester children for number and percentage of first uses of verb types with more than one argument in Stage I & II combined Children No. ( %) 16/115 (13.9) 18/108 (16.7) 10/113 (8.8) 27/99 (27.3) 21/112 (18.8) 28/75 (37.3) 17/125 (13.6) 25/130 (19.2) 16/108 (14.8) 38/169 (22.5) 13/69 (18.8) (19.8) Travis Aran Becky Dominic Anne Ruth Gail Joel Liz Nicole Warren Mean arguments; Gail, 14% ; Liz, 15% ; Aran 17% ; and Anne, Joel, and Warren all have 19%. There are only two exceptions : Dominic and Ruth, and only Ruth, with 37% of her verbs first appearing with more than one argument, looks radically different from Travis. These results suggest that, like Travis, the children in this study do initially tend to use their verbs in single argument structures. This finding is consistent with the Verb Island hypothesis and a lexically specific view of early grammatical development, and suggests that Travis is not an isolated case. However, the results presented in Table 4 do not address the issue of whether certain verb types (e.g. transitive verbs) may be more likely to first appear in multi-argument structures than others. To investigate this possibility, all verb types for Travis and the Manchester children were categorized as either fixed transitive, mixed, or fixed intransitive verbs (following Levin’s, 1993, categorization scheme or the OED). The results are presented in Table 5. It is clear from Table 5 that the data for Travis and the Manchester children are still broadly comparable (33.3% versus a mean of 37.1% for fixed transitive verbs ; 10.5% versus a mean of 17.0% for mixed verbs and 0% versus a mean of 8.4% for fixed intransitive verbs). However, it is also clear that a higher percentage of fixed transitive verbs first occurred in multi-argument structures than mixed verbs or fixed intransitive verbs. These differences were analysed by conducting a one way repeated measures analysis of variance on the data from the Manchester children. This analysis revealed a significant effect of verb-type (F(2, 9)=14.87, p<0.001). Additional post hoc analyses using Scheffe tests revealed a significant difference between fixed transitive verbs and mixed verbs 707 M C C L U R E E T A L. 5. Comparison of Travis with Manchester children for percentage of fixed transitive, mixed and fixed intransitive verb types with more than one argument TABLE Children Travis Aran Becky Dominic Anne Ruth Gail Joel Liz Nicole Warren Mean Fixed transitive 7/21 9/22 3/20 9/20 4/21 8/16 6/19 4/16 7/17 10/30 7/10 33.3% 40.9% 15.0% 45.0% 19.0% 50.0% 31.6% 25.0% 41.2% 33.3% 70.0% 37.1% Fixed intransitive Mixed 9/86 9/81 7/87 18/73 15/84 19/51 10/97 20/104 8/82 27/128 6/55 10.5 % 11.1 % 8.0 % 24.7 % 17.9 % 37.3 % 10.3 % 19.2 % 9.8 % 21.1 % 10.9 % 17.0 % 0/8 0% 0/5 0/6 0/6 2/7 1/8 1/8 1/10 1/9 1/11 0/4 0% 0% 0% 28.6% 12.5% 12.5% 10.0% 11.1% 9.1% 0% 8.4% (F=13.79, p<0.01), a significant difference between fixed transitive verbs and fixed intransitive verbs (F=28.24, p<0.001, but no significant difference between mixed verbs and fixed intransitive verbs (F=2.56, NS). When taken as a whole, these results suggest that Travis’s data are fairly typical of children learning English. However, they also suggest that the low percentage of children’s early verbs that first occur in complex structures is at least partly a reflection of the kind of verbs that children are learning during the early stages. The implication is that the percentage of verbs that first occur in multi-argument frames may not be a particularly informative measure, and hence that a better way to evaluate the Verb Island hypothesis may be to focus on differences in the complexity of the structures in which children use new and old verbs at the same stage of development. Stage II Old Verb vs. Stage II New Verb analysis Results for Old versus New Verbs at Stage II are presented in Table 6. They show that the second prediction was confirmed : Old Verbs at Stage II do tend to occur in more complex structures than New Verbs at Stage II. MLU analyses. The table indicates that the MLUs for tokens of Old Verbs are greater than the MLUs for tokens of New Verbs at Stage II. The mean MLU for Old Verbs is 3.26, whereas that for New Verbs is 3.02. A t-test indicated a significant difference between Old and New Verbs for tokens : t (9)=4.88, p<0.001, one-tailed. A similar result was found for first uses of verb types. The mean MLU for Old Verbs is 3.10 and for 708 INVESTIGATING ABSTRACTNESS TABLE 6. Means (and S.D.s) for MLU and number of arguments/utterance for Old vs. New Verbs at Stage II Analyses MLU tokens MLU types (first uses at St. II) Arguments/Utterance (first uses at St. II) Arguments/Transitive utterance (first uses at St. II) Stage II New Verbs Stage II Old Verbs t-test 3.02 (0.31) 2.81 (0.41) 3.26 (0.22) 3.10 (0.32) *** ** 1.0 (0.21) 1.14 (0.21) * 1.25 (0.15) 1.41 (0.17) * * p<0.05, ** p<0.01, *** p<0.001. New Verbs, 2.81. A t-test indicated a significant difference between the two : t (9)=3.47, p<0.01, one-tailed. This result indicates that there is more complexity surrounding Old Verbs than New Verbs at Stage II. However, as it is unclear whether this difference in complexity reflects differences in the average number of arguments per utterance, a separate analysis was performed. Argument analyses. When one looks at the analysis for the number of arguments per utterance, Table 6 shows that utterances with Old Verbs have more arguments than utterances with New Verbs at Stage II, which is what the Verb Island hypothesis predicts. Old Verbs have a mean of 1.14 arguments per utterance, whereas New Verbs have a mean of 1.0. A significant difference was found between Old and New Verbs (t (9)=2.26, p<0.05, one-tailed), thereby giving further support to the Verb Island hypothesis. When one looks at the difference in the size of the mean differences in the MLU for types and the number of arguments, it appears that the children are both using more arguments and extending the length of their utterances in other ways. The mean difference for MLU types is 0.29 and the mean difference for number of arguments is 0.14, indicating that half of the effect for MLU is due to the difference in number of arguments and the rest to other differences in complexity, including differences in the complexity of the verb forms produced (e.g. the inclusion of modals or auxiliaries) and differences in the complexity of the arguments produced (e.g. the inclusion of determiners or adjectives). Transitive analyses. The difference in the number of arguments for Old and New Verbs suggests that children’s knowledge about argument structure is not fully verb-general. However, this result could also reflect differences in the type of verbs that first appear at Stage I and II. For example, it could be that a higher proportion of the children’s Old Verbs at Stage II is transitive, which would explain why Old Verbs at Stage II tend 709 M C C L U R E E T A L. TABLE 7. Number and percentage of transitive Old vs. New Verbs at Stage II Children Stage II Old Verbs no. (%) Stage II New Verbs no. (%) Aran Becky Dominic Anne Ruth Gail Joel Liz Nicole Warren 17/40 17/48 20/41 35/58 3/12 22/43 28/46 15/38 51/83 16/28 38/52 17/24 36/50 13/26 43/61 36/60 34/65 39/63 47/70 16/35 (43) (35) (49) (60) (25) (51) (61) (39) (61) (57) (73) (71) (72) (50) (70) (60) (52) (62) (67) (46) to have more arguments. To evaluate this possibility, an analysis of the percentages of Old and New Verbs that were transitive was conducted. Table 7 shows that Old Verbs at Stage II do not have a higher percentage of transitive verbs than New Verbs at Stage II. For 7 out of 10 children, New Verbs have a higher percentage of transitive verbs. Furthermore, when the analysis for arguments was restricted to just transitive verbs (Table 6), a significant effect for arguments between Old and New Transitive Verbs at Stage II was still found (t (9)=2.41, p<0.05, onetailed). This result indicates that the difference found between all Old and New Verbs at Stage II (Table 6) was not simply due to the category of Old Verbs including more transitive verbs, and suggests that the children’s ability to use verbs in more complex structures is related to the length of time that they have known the relevant verb, which is consistent with a lexically specific view of early grammatical knowledge. Stage I New Verb vs. Stage II New Verb analysis The Verb Island hypothesis states that children’s knowledge about verbargument structure is entirely verb-specific. Therefore, it not only predicts that Old Verbs at Stage II will occur in more complex structures than New Verbs at Stage II as seen above ; it also predicts that there will be no difference in children’s ability to manipulate New Verbs at Stage I and New Verbs at Stage II. The results indicate, however, that this is not the case. Table 8 shows that MLUs and the number of arguments per utterance for New Verbs at Stage II are greater than those for New Verbs at Stage I. This suggests that children’s knowledge about argument structure is not entirely verb-specific. 710 INVESTIGATING ABSTRACTNESS TABLE 8. Means (and S.D.s) for MLU and number of arguments/utterance for New Verbs at Stage I and Stage II Analyses MLU tokens MLU types (first uses) Arguments/Utterance (first uses) Arguments/Transitive utterance (first uses) Stage I New Verbs Stage II New Verbs t -test 2.53 (0.19) 2.27 (0.22) 3.02 (0.31) 2.81 (0.41) *** ** 0.77 (0.15) 1.0 (0.2) * 1.13 (0.16) 1.25 (0.15) * * p<0.05, ** p<0.01, *** p<0.001, one-tailed. MLU analyses. Table 8 shows the mean MLU for New Verb tokens at Stage II was 3.02, and that of New Verb tokens at Stage I, 2.53. A significant difference between Stage I and II was found for tokens : t (9)=6.19, p<0.001, one-tailed. The mean MLU for New Verb first uses of verb types was 2.81 at Stage II and 2.27 at Stage I. This difference was also significant : t (9)=4.59, p<0.01, one-tailed. Argument analyses. The analysis of the number of arguments per utterance shows that utterances with New Verbs at Stage II have more arguments than New Verbs at Stage I. New Verbs at Stage II have one argument per utterance ; those at Stage I have 0.77. Again, the results for first uses are significant : t (9)=2.72, p<0.05, one-tailed. The difference in size of the means of the MLU types (0.54) and the number of arguments (0.23), indicates that there are more, and more complex, arguments around New Verbs at Stage II. Transitive analyses. As with the analysis between Old and New Verbs at Stage II, this effect for arguments could be an artifact of difference in the type of verbs learned during Stage I and Stage II. For example, if a higher proportion of the children’s Stage II verbs was transitive than their Stage I verbs, this might have the effect of increasing the number of arguments per verb in Stage II. Table 9 shows the proportions of New Verbs at Stage I and II that were transitive. It can be seen that a greater proportion of each child’s verbs at Stage II was transitive. However, when the analysis for arguments was restricted to transitive verbs (Table 8), a significant difference was still found (t (9)=1.84, p<0.05, one-tailed). This suggests that the increase in the number of arguments for New Verbs at Stage II was not just an artifact of changes in the proportion of the children’s verbs that was transitive. The implication is that although there is some evidence from the analysis of Old versus New Verbs at Stage II that the children’s knowledge about 711 M C C L U R E E T A L. TABLE 9. Number and percentage of transitive New Verbs at Stage I vs. transitive New Verbs at Stage II Children Stage I New Verbs no. (%) Stage II New Verbs no. (%) Aran Becky Dominic Anne Ruth Gail Joel Liz Nicole Warren 17/56 17/89 20/49 35/87 3/15 22/65 28/65 15/48 51/99 16/35 38/52 17/24 36/50 13/26 43/61 36/60 34/65 39/63 47/70 16/35 (30) (19) (41) (40) (20) (34) (43) (31) (52) (46) (73) (71) (72) (50) (70) (60) (52) (62) (67) (46) argument structure is not fully verb-general, there is also evidence from the analysis of New Verbs at Stage I versus Stage II that it is not fully verb-specific either. Productive positional patterns One way of reconciling these findings is to propose that children are not restricted to learning lexically specific patterns around verbs, as Tomasello (1992) suggests, but can also learn lexically specific patterns around other high frequency items (Braine, 1976; Pine et al., 1998). In order to investigate this issue each of the children’s corpora was searched for evidence of lexical patterns of the form : ‘ high frequency subject+X’ or ‘ X+high frequency object ’, where X is a verb. The following procedure was used. First, each child’s corpus was searched for verb types (at Stage I and Stage II) that occurred with subjects and then these were examined to see whether any of the subjects occurred with 6 or more different verb types. Thus, ‘ high frequency subject ’ was defined as any subject that occurred with 6 or more different verb types. The same procedure was followed for verb types with objects. The consistency of the patterns was evaluated using Braine’s (1976) statistical criterion for the attribution of a productive positional pattern based on the binomial theorem (i.e. six different instances in the same order and no instances in a different order, or eight instances in the same order and one instance in a different order, etc.). For example, ‘X+it ’ was considered a productive pattern if there were 6 correct examples of it used as an object and no examples of it used as an object in the opposite pattern, ‘ It+X’. 712 INVESTIGATING ABSTRACTNESS Evidence of 10 productive subject patterns was found: ‘I+X’, ‘me+X ’, ‘ the child’s name+X’, ‘ Mummy/Mama+X ’, ‘you+X’, ‘the cat’s name+X ’, ‘ the investigator’s name+X’, ‘ Daddy+X’, ‘that+X ’, and ‘ it+X ’ (Table 10). In addition, evidence of seven productive object patterns was found : ‘ X+it ’, ‘ X+that’, ‘ X+them ’, ‘X+this ’, ‘ X+me ’, ‘ X+baby ’, and ‘ X+fly’ (Table 10). All 10 children had the ‘ I+X ’ and the ‘ X+it ’ patterns. Table 10 presents the patterns that were productive for each child by the end of each stage. For example, Aran had 12 verb types that showed the pattern : ‘ I+X’, where X was got, want hat, bite it, kick, need book, pull, fall, ride, throw, kiss, walk, wipe it by the end of Stage I. There were no counter-examples. By the end of Stage II he had acquired 19 additional verb types in this pattern for a total of 31 verb types. When the number of verb types becomes productive for a pattern, it is highlighted in bold. For example, Aran’s pattern ‘ it+X’ did not become productive in Stage I because there was one counter-example, but it did become productive at Stage II with 8 verb types and only one counter-example. The most important finding was that all of the 10 children had at least one productive pattern by the end of Stage I and at least 3 by the end of Stage II. Although lexically specific, such patterns are, by definition, verb-general. The implication is that although children’s ability to produce syntactic constructions can be seen as lexically specific (in the sense that it does not automatically generalize across lexical items), it is not completely verb-specific (in the sense that children appear to have lexically specific constructions which accept a range of different verbs as slot-fillers). This may explain how children who are producing Old Verbs in more complex structures than New Verbs at Stage II are nevertheless able to generalize at least some of the knowledge of argument structure that they have built up during Stage I to New Verbs at Stage II. Although this evidence of verb-general structure is not consistent with a strict version of the Verb Island hypothesis, it is consistent with the kind of limited scope formulae account of early grammatical development proposed by Braine (1976). According to Braine, children’s early systems are considerably less abstract than those of adults. For example, young children do not have an adultlike subject category because their knowledge of how to order some Subject+VP sequences (e.g. Agent+Action sequences such as ‘ Doggie bark’) does not transfer to other Subject+VP sequences (e.g. Located+Locative such as ‘ Doggie in basket ’) which are often initially ordered incorrectly (e.g. ‘ In basket doggie ’). Children’s systems are therefore best seen not as generative grammars, but as inventories of limited scope formulae that map the components of semantic representations into positions in the surface structure. One advantage of this view over a strict version of the Verb Island hypothesis is that it allows for the acquisition of lexically specific patterns 713 M C C L U R E E T A L. TABLE Children Aran Becky Dominic Anne Ruth Gail 10. All productive S+V and V+O patterns by end of Stage I and II Patterns (counter-patterns) No. verb types Stage I (counter-examples) No. verb types Stage I & II (counter-examples 12* (0) 7 (1) 6 (3) 2 (0) 3 (0) 24 (0) 21 (0) 4 (0) 14 (0) 7 (0) 4 (0) 10 (0) 14 (0) 19 (0) 21 (1) 32 (0) 14 (0) 5 (0) 8 (1) 5 (0) 5 (1) 31 (0) 8 (1) 11 (3) 9 (0) 6 (0) 30 (0) 27 (0) 6 (0) 32 (0) 11 (0) 8 (0) 11 (0) 31 (0) 21 (0) 24 (1) 39 (0) 16 (0) 6 (0) 40 (1) 24 (0) 19 (1) 6 (0) 10 (1) 8 (0) 13 (0) 9 I+X (X+I) It+X (X+it) X+it (It+X) X+that (That+X) X+fly (Fly+X) I+X (X+I) X+it (It+X) X+this (This+X) I+X (X+I) Mummy+X (X+Mummy) Daddy+X (X+Daddy) That+X (X+that) X+it (It+X) I+X (X+I) Anne+X (X+Anne) X+it (It+X) X+that (That+X) X+baby (Baby+X) Me+X (X+me) I+X (X+I) Mummy+X (X+Mummy) X+it (It+X) X+that (That+X) X+baby (That+X) Gail+X (X+Gail) I+X 5 (0) 4 714 INVESTIGATING ABSTRACTNESS TABLE Children Joel Liz Nicole Warren 10. (Cont.) Patterns (counter-patterns) No. verb types Stage I (counter-examples) No. verb types Stage I & II (counter-examples (X+I) Caroline+X (X+Caroline) You+X (X+you) X+it (It+X) I+X (X+I) Mummy/Mama+X (X+Mummy/Mama) You+X (X+you) X+it (It+X) Liz+X (X+Liz) I+X (X+I) Mummy+X (X+Mummy) X+it (It+X) I+X (X+I) Nicole+X (X+Nicole) Mummy/Mum+X (X+Mummy) You+X (X+you) Molly+X (X+Molly) X+it (It+X) X+that (That+X) X+them (Them+X) I+X (X+I) Warren+X (X+Warren) Mummy+X (X+Mummy) X+It (It+X) (0) 3 (1) 6 (0) 21 (0) 25 (0) 7 (0) 1 (0) 15 (0) 13 (0) 1 (0) 4 (0) 8 (0) 35 (0) 31 (2) 12 (0) 6 (0) 3 (0) 22 (0) 11 (0) 6 (0) 12 (0) 7 (0) 6 (0) 9 (1) (0) 8 (1) 7 (0) 41 (1) 41 (0) 8 (0) 7 (0) 32 (0) 23 (0) 18 (0) 8 (0) 25 (1) 50 (0) 43 (2) 14 (1) 8 (0) 6 (0) 44 (0) 14 (1) 8 (0) 20 (0) 19 (0) 12 (0) 15 (1) * Numbers in bold indicate when the pattern becomes productive. 715 M C C L U R E E T A L. based around items other than verbs. Thus, an English child might have learned a formula that allows him to talk about himself as agent (or as agent and experiencer) by placing the nominative pronoun ‘I ’ in front of the word that denotes the action or experience. Such a formula is inconsistent with a strict version of the Verb Island hypothesis since it is, by definition, verb-general. However, it is consistent with more recent lexical constructivist formulations, which accept that grammatical structure does not revolve exclusively around verbs (Pine et al., 1998 ; Akhtar, 1999 ; Childers & Tomasello, 2001). Another advantage of Braine’s limited scope formula account is that it allows for the possibility that children’s early formulae can be represented at a number of different levels of abstraction. Thus, although for Braine some of children’s early rules were clearly lexically specific formulae, Braine was also prepared to attribute more abstract patterns to children who used a sufficiently wide range of lexical items to express similar semantic relations. For example, he credited Bowerman’s (1973) subject Kendall with an Agent+Action pattern on the grounds that, unlike some of the other children for whom corpora were available, she produced a relatively large number of lexically diverse and correctly ordered Agent+Action sequences. This interpretation of the data is clearly inconsistent with the claim that children’s early systems consist entirely of inventories of lexically specific patterns. However, it is consistent with the claim that children are gradually building adultlike representations by a process of functionallybased distributional analysis of the input (Tomasello, 1992, 2003). It is also consistent with the idea that, although children do not have adultlike knowledge of the transitive construction, they may have weak or partial representations that are sufficiently abstract to support performance in syntactic bootstrapping experiments (Tomasello & Abbot-Smith, 2002). DISCUSSION The main purpose of the present study was to use data from 10 children to test the claim that children’s early knowledge of argument structure builds up gradually around particular verbs. The results provide evidence that children build grammatical structure in a lexically specific way; however, they also show that this structure is not necessarily verb-specific. Among the findings consistent with the view that children’s knowledge builds up gradually around verbs was the fact that most of the children were like Travis in having few of their verbs first appearing with two or more arguments. The implication is that, rather than reflecting a somewhat extreme approach to early word combination, the lexical specificity of Travis’s early speech is actually quite typical of English-speaking children during the early stages. 716 INVESTIGATING ABSTRACTNESS Also consistent with the view that children’s early knowledge of argument structure is verb-specific was the evidence that children were able to use some verbs in more complex structures than other verbs with similar argument structures. Thus, the children in the study appeared to know significantly more about Old Verbs at Stage II than New Verbs at Stage II, even when the analysis was restricted to transitive verbs. Old Verbs had more argument structure and higher MLUs for both tokens and first uses of verb types. These findings suggest that the children were not able to automatically generalize what they knew about one verb to another verb with the same argument structure and hence that their knowledge of argument structure was not fully verb-general in the sense that it did not allow them to use all verbs of a particular type (e.g. transitive verbs) in the same range of sentence types (e.g. (S) V+O and S+V+O). However, not consistent with the view that children’s early knowledge of argument structure is verb-specific was the evidence that the children were able to generalize some of their knowledge about Verbs at Stage I to New Verbs at Stage II. Thus, there was a significant difference between the number of arguments surrounding New Verbs at Stage II and New Verbs at Stage I. This suggests that the children were learning something about verbs at Stage I that they were able to apply spontaneously to New Verbs at Stage II. The Verb Island hypothesis cannot explain this finding. However, one way in which it could be explained is in terms of the learning of limited scope formulae such as ‘Mummy+X ’, or ‘I+X ’, ‘ me+X’, ‘ the child’s name+X’, or ‘ X+it ’. Structures like these, which are composed of high frequency subjects and objects, and accept verbs as slot-fillers, may allow the child to use their New Verbs at Stage II in slightly more complex structures than they were able to use their New Verbs at Stage I. They may also serve as the building blocks for more abstract structures such as S+V and V+O. Although this evidence of verb-general structure is not consistent with a strict version of the Verb Island hypothesis, it is consistent with Tomasello’s more recent formulation that grammatical structure does not have to revolve exclusively around verbs, but can include other lexical items, particularly pronouns (Childers & Tomasello, 2001). Childers & Tomasello found that for English-speaking children, pronouns play a very important role in the acquisition of the transitive construction. They found that when children (M=2; 6) were trained with utterances that had pronoun frames, such as he’s [verb]-ing it and closely associated noun frames, this helped them to use novel verbs productively. Childers & Tomasello concluded that as children are learning individual words, they are also learning constructions (e.g. the transitive) and this may be facilitated by low-scope representations such as I’m [verb]-ing it or he’s [verb]-ing it or [verb]-it. 717 M C C L U R E E T A L. Other evidence that children may be using pronoun frames comes from Akhtar (1999). In the cases where children heard a novel verb in a non-English word order and then were able to use it productively in an utterance in SVO word order, they did so, in all but one case, using pronouns. This suggests they may be learning the privileges of occurrence of certain high-frequency pronouns (he, him, it) and that this knowledge helps them produce utterances in SVO word order. A pronoun frame in SVO order may therefore constitute partial knowledge of what a full abstract transitive construction might consist of. In other words, children learn which pronouns frequently occur in preverbal position and which pronouns occur frequently in post-verbal position and they use this knowledge to structure sentences with new verbs. In fact, the youngest children in Akhtar’s study were 2;8, though similar results have also been found for slightly younger children (Abbott-Smith et al., 2001). The results of the present study suggest that children may have some verb-general knowledge of transitive structure as early as Stage I – in the sense that they appear able to use new verbs with a variety of high frequency subjects and objects. However, they also underline the dangers of assuming that such knowledge necessarily represents fully abstract knowledge of the transitive construction. To have such knowledge would presumably mean to have knowledge that was not tied to any specific lexical items, whether they are particular verbs or particular high frequency subjects and objects. It is not clear to us that there is currently any strong evidence that children have such knowledge. However, as Tomasello & Akhtar (2003) point out, one way of showing that they do, would be to demonstrate syntactic priming effects for transitive sentences that show little or no lexical overlap with the transitive sentences that they primed. Interestingly, in a recent study that adopted this kind of approach Savage, Lieven, Theakston & Tomasello (2003) found that while older children did show syntactic priming effects for sentences with both high and low levels of lexical overlap, younger children only showed priming effects for sentences in which the level of lexical overlap was high. These results, like those of the present study, are consistent with the view that although children’s knowledge of argument structure is not completely verb-specific, nor is it fully abstract during the early stages. The question for future research is therefore how do young children construct such abstract syntactic representations in the course of development. One way of addressing this issue is to use comprehension and production studies to investigate the precise nature of children’s early grammatical representations at different points in development (e.g. Fisher, 2002 b; Savage et al., 2003 ; Matthews, Lieven, Theakston & Tomasello, 2005). Another is to build models of language learning that can provide an integrated account of the results of comprehension and production studies, and hence shed light 718 INVESTIGATING ABSTRACTNESS on the mechanisms that underlie this process (e.g. Chang, 2002, 2004 ; Alishahi, 2004). REFERENCES Abbot-Smith, K., Lieven, E. & Tomasello, M. (2001). What preschool children do and do not do with ungrammatical word orders. Cognitive Development 16, 679–92. Akhtar, N. (1999). Acquiring basic word order : evidence for data-driven learning of syntactic structure. Journal of Child Language 26, 339–56. Akhtar, N. & Tomasello, M. (1997). Young children’s productivity with word order and verb morphology. Developmental Psychology 33, 952–65. Alishahi, A. (2004). A computational model for early argument structure acquisition. Unpublished doctoral dissertation. University of Toronto. Bowerman, M. (1973). Early syntactic development : a crosslinguistic study with special reference to Finnish. Cambridge : CUP. Braine, M. (1976). Children’s first word combinations. Monographs of the Society for Research in Child Development 41(1) (serial no. 164). Brooks, P. J. & Tomasello, M. (1999). Young children learn to produce passives with nonce verbs. Developmental Psychology 35(1), 29–44. Brown, R. (1973). A first language : the early stages. Cambridge MA : Harvard University Press. Chang, F. (2002). Symbolically speaking : a connectionist model of sentence production. Cognitive Science 26, 609–51. Chang, F. (2004). The development of the transitive construction : a connectionist account. Paper presented at the 29th Annual Boston University Conference on Language Development. Childers, J. B. & Tomasello, M. (2001). The role of pronouns in young children’s acquisition of the English transitive construction. Developmental Psychology 37, 739–48. Fisher, C. (1996). Structural limits on verb mapping : the role of analogy in children’s interpretations of sentences. Cognitive Psychology 31, 41–81. Fisher, C. (2002a). The role of abstract syntactic knowledge in language acquisition : a reply to Tomasello (2000). Cognition 82, 259–78. Fisher, C. (2002 b). Structural limits on verb mapping : the role of abstract structure in 2.5-year-olds’ interpretations of novel verbs. Developmental Science 5(1), 55–64. Ingram, D. (1989). First language acquisition : method, description and explanation. Cambridge : C.U.P. Levin, B. (1993). English verb classes and alternations. Chicago : University of Chicago Press. Lieven, E. V. M., Pine, J. M. & Baldwin, G. (1997). Lexically-based learning and early grammatical development. Journal of Child Language 24, 187–219. MacWhinney, B. (2000). The CHILDES project : tools for analyzing talk (3rd ed.). Mahwah, NJ : Erlbaum. Matthews, D., Lieven, E. V. M., Theakston, A. L. & Tomasello, M. (2005). The role of frequency in the acquisition of English word order. Cognitive Development 20, 121–36. Naigles, L. (1990). Children use syntax to learn verb meanings. Journal of Child Language 17, 357–74. Naigles, L. & Kako, E. (1993). First contact in verb acquisition : defining a role for syntax. Child Development 64, 1665–87. Pine, J. M. & Lieven, E. V. M. (1993). Reanalysing rote-learned phrases : individual differences in the transition to multi-word speech. Journal of Child Language 20, 551–71. Pine, J. M. & Martindale, H. (1996). Syntactic categories in the speech of young children : the case of the determiner. Journal of Child Language 23, 369–95. Pine, J. M., Lieven, E. V. M. & Rowland, C. F. (1998). Comparing different models of the development of the English verb category. Linguistics 36, 807–30. 719 M C C L U R E E T A L. Pizzuto, E. & Caselli M. C. (1992). The acquisition of Italian morphology : implications for models of language development. Journal of Child Language 19, 491–557. Rubino, R. B. & Pine, J. M. (1998). Subject–verb agreement in Brazilian Portuguese : what low error rates hide. Journal of Child Language 25, 35–59. Savage, C., Lieven, E., Theakston, A. & Tomasello, M. (2003). Testing the abstractness of children’s linguistic representations : lexical and structural priming of syntactic constructions in young children. Developmental Science 6, 557–67. Theakston, A. L., Lieven, E. V. M., Pine, J. M. & Rowland, C. (2001). The role of performance limitations in the acquisition of verb–argument structure. Journal of Child Language 28, 127–52. Tomasello, M. (1992). First verbs : a case study of early grammatical development. Cambridge : C.U.P. Tomasello, M. (2000). Do young children have adult syntactic competence? Cognition 74, 209–53. Tomasello, M. (2003). Constructing a language : a usage-based theory of language acquisition. Cambridge, MA : Harvard University Press. Tomasello, M. & Abbot-Smith, K. (2002). A tale of two theories : response to Fisher. Cognition 83, 207–14. Tomasello, M. & Akhtar, N. (2003). What paradox? A response to Naigles (2002). Cognition 88, 317–23. Tomasello, M. & Brooks, P. J. (1998). Young children’s earliest transitive and intransitive constructions. Cognitive Linguistics 9, 379–95. 720