Verb argument structure Shanley

representation of these structures (see Tomasello (2000) and references therein, and Sections 13.3.1 and 13.3.2). In theweird word order paradigm, children hear an experimenter describing events using a weird word order (e.g. Ernie Bert pushing or pushing Ernie Bert to describe Ernie pushing Bert) and are then asked to describe similar events in their own words. When the verb is frequent (e.g. push), English-, French-, and German-speaking children as young as age 2 typically ‘correct’ the weird word order to Subject–Verb–Object. When the verb is infrequent (e.g. dab) or novel (e.g. meek), however, only children over 3;6 consistently correct the word order. Younger children very often use the weird order or avoided using the verb altogether (Abbot-Smith, Lieven & Tomasello 2001, Akhtar 1999, Matthews, Lieven, Theakston & Tomasello 2005, 2007). 278 S H A N L E Y E . M . A L L E N C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 279 [271--297] 17.7.2015 5:00PM The third type of elicited production study involves training children in use of either intransitive–transitive (e.g. This tiger is bouncing / This tortoise is bouncing this tiger) or passive–transitive alternations (e.g. The tiger’s gonna get bounced / The tortoise is gonna bounce the tiger) for a set of familiar verbs, then presenting them with a novel verb in either intransitive or passive and encouraging them to use it in the transitive. Trained children aged 2;6 generalized the transitive structure with novel verbs more than twice as often as a group of control children who did not receive training, indicating that developing a representation of the transitive structure is influenced by input frequency (Abbot-Smith, Lieven & Tomasello 2004, Childers & Tomasello 2001). Overall, the results of both spontaneous speech and elicited production studies suggest that 2-year-old children typically restrict their use of a verb to the syntactic frame in which it is learned and do not easily generalize to other frames as would be predicted if they had innate knowledge of categories such as Agent and Subject (Tomasello 2000). Several of the studies cited in Section 13.2.1.1 that challenge the semantic bootstrapping view also constitute evidence for the usage-based view that lexical factors like individual verb and construction frequency play a crucial role in learning argument structure. In addition, a recent computational modelling study supports this general approach, showing that a grammar formed only through item-based information from child-directed speech can accurately predict children’s productions at age 2;0 (Bannard, Lieven & Tomasello 2009). However, adding information about lexical categories (N and V) substantially improves the model for predicting children’s utterances at 3;0. 13.2.3 ‘Weak’ or ‘graded’ abstract representations Fisher (2002a) critiqued the strong version of the usage-based approach while acknowledging its valuable contributions to understanding the role of lexical learning in acquisition. She argued that many of the findings interpreted by usage-based theorists as lack of abstraction could rather be interpreted as evidence of syntactic priming (Bock 1986) – children persist in using the learned syntactic frames because they have just heard them so they have been primed to that frame – or as evidence of appropriate conservatism – children know that not all English intransitive verbs can be used transitively (e.g. sleep, giggle). More importantly, she pointed out that a non-trivial number of 2-year-olds in studies claimed to support a strong usage-based theory in fact generalized novel verbs to new sentence frames (Abbot-Smith et al. 2001, Brooks & Tomasello 1999; see also Kline & Demuth 2014) or changed ungrammatical to grammatical word orders (Akhtar 1999). Finally, she noted that virtually all of the evidence for the usage-based approach derives from production studies, which require active behavioural decision-making and thus relatively strong syntactic Verb argument structure 279 C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 280 [271--297] 17.7.2015 5:00PM representations. In contrast, comprehension studies using the preferential looking paradigm, which places fewer performance demands on children and thus is more sensitive to weak syntactic representations, have provided evidence for abstract representations of argument structures as young as 2;1 (see Section 13.2.1.2). Supported by these three types of evidence, Fisher suggested that 2-year-old children do have abstract representations of the syntactic frames in question although they are weaker than those of older children and adults. In their reply to Fisher’s critique, Tomasello and Abbot-Smith (2002: 210) conceded that young childrenmay ‘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’. A connectionist network simulation by Chang et al. (2006) provides support for the claim that both early success in preferential looking tasks and later success in production tasks can be accounted for within one model. More recent research has paid specific attention to better understanding the interplay between abstract representations and lexical learning. This research has twomain foci: uncovering effects of performance and processing demands that reveal the differential strength of abstract representations, and applying a new testing method (structural priming) that is particularly sensitive to representation strength. 13.2.3.1 Evidence from effects of performance and processing demands As noted by Fisher (2002a), it has become increasingly evident that children are affected by performance and processing demands in being able to use their abstract representations, and further investigation of these effects is likely to provide insight into the strength and developmental trajectory of the representations. Explorations in this domain have focused on the relative demands of different task types, the demands of materials and procedures within the task, child-specific factors, and language-specific factors. Different task types clearly entail different processing and performance demands, which could lead to differences in performance across studies. Two studies have made controlled direct comparisons of children’s performance on the same materials using two different tasks. Dittmar et al. (2008b) found that German-speaking children aged 2;7 did not show comprehension of semantic roles in transitive sentences using novel verbs in an act-out task (relatively heavy task demands), but a different group of same-aged children did show comprehension of the same sentences in a forced choice pointing task (relatively lighter task demands). In the second study, Chan et al. (2010) compared the comprehension of semantic roles in transitive sentences in an act-out task (heavy demand) vs a preferential looking task (light demand), using the same English-speaking children as 280 S H A N L E Y E . M . A L L E N C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 281 [271--297] 17.7.2015 5:00PM participants in each. While the oldest two groups (aged 2;9 and 3;5) did well in both tasks, the youngest group (aged 2;0) showed comprehension only in the preferential looking task. Forced-choice pointing is less demanding than act-out or elicited production tasks, but more revealing of children’s ability to actively use their abstract representations than preferential looking since it requires only a simple point to select one picture over another as the best match for a sentence, but indicates an active choice on the part of the child. Fernandes et al. (2006) successfully used this method to show that children aged 2;4 and 2;10 can map semantic roles to syntactic roles in transitive sentences with novel verbs; Noble et al. (2011) extended this finding to 1;11, and Dittmar et al. (2011) to 1;9. Another such task, structural priming, is discussed in detail in Section 13.2.3.2. The procedures and materials within tasks also present differing demands that test the strength of children’s abstract representations. The amount of training received during the familiarization phase of a given task is one element of the procedure that can strongly affect children’s performance. A preferential looking study by Dittmar et al. (2008a) tested this explicitly. They found that German-speaking children aged 1;9 only succeeded in identifying the scene where the semantic roles of the argumentsmatched the transitive promptwhen theywere exposed during the familiarization phase to numerous exemplars of the same transitive syntactic structure they had to comprehend in the test phase. When the syntactic structure in the familiarization phase was neutral (e.g. Look, eating), the children performed at chance during the test phase. A difference in amount of training could also explain the apparently contradictory results for English-speaking children’s comprehension of semantic roles in preferential looking tasks: the children tested by Gertner et al. (2006) got substantial training and showed above-chance performance at 1;9, while the children tested by Chan et al. (2010) got little training and showed only chance performance at 2;0. Other examples of the effect of amount of training are discussed in Section 13.2.3.2; children show different results on priming tasks depending on howmany primes are presented before the test item. Linguistic complexity of thematerials can also affect children’s ability to use their abstract representations. For example, children younger than 2;1 perform at chance in the original syntactic bootstrapping paradigm contrasting conjoined-Subject intransitives such as The bunny and the duck are gorping with single-Subject transitives such as The bunny is gorping the duck (Bavin & Growcutt 1999, Hirsh-Pasek & Golinkoff 1996), even though they show abstract representations for intransitive and transitive structures using other materials. Yuan, Fisher and Snedeker (2012) suggested that this arises from managing the linguistic complexity of the conjoinedSubject intransitive, and showed that children as young as 1;7 can derive verb meaning from syntactic frames as long as the processing demands of Verb argument structure 281 C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 282 [271--297] 17.7.2015 5:00PM the original paradigm are reduced in two ways: (a) the number of nouns children hear in a sentence matches the number of argument roles (He is gorping vs He is gorping him), and (b) the caused-motion and non-causative events are presented in two separate clips in the initial exposure phase instead of together in one clip. Gertner and Fisher (2012) have also acknowledged this linguistic complexity, and showed that it leads children aged 1;9 to focus simply on the number of nouns present in the sentence rather than their syntactic structure. Their participants interpreted the first noun in a sentence as the Agent and the second as the Patient, regardless of the placement of the two nouns in a sentence with respect to the verb, in sentences such as The boy and the girl are gorping, The girl and the boy are gorping, and The boy is gorping the girl. Even in a condition where children received prior dialogue information that a given novel verb can be intransitive (e.g. Tom is gonna gorp. Really? He’s gonna gorp?), they still interpret the first noun of a conjoined Subject as Agent and the second as Patient. However, Arunachalam, Escovar, Hansen and Waxman (2013) showed that children aged 1;9 can in fact interpret two nouns before the verb as a conjoined Subject when the syntax ismade very explicit to aid in dealing with the linguistic complexity (e.g. The man and the lady are gonna moop. They’re gonna moop.). Another type of linguistic complexity affecting task performance arises from the use of novel verbs. Novel verbs are ideal for examining argument structure knowledge because they have not previously been heard by the child, allowing for careful control of both frequency and semantic class information. However, they also present a higher processing demand than familiar verbs because their meaning must be worked out during the task. They are thus useful as a tool for exploring representation strength because children show earlier evidence of abstract representations with familiar verbs than with novel verbs, and several studies have exploited this. Abbot-Smith, Lieven and Tomasello (2008) found that children aged 2;4 mostly corrected incorrect Agent–Patient assignment in transitive sentences with familiar verbs, but tended to avoid producing the structure in trials with novel verbs. Similarly, Dittmar et al. (2008b) found that children aged 2;7 could act out transitive sentences with familiar verbs but not with novel verbs, while children aged 4;10 correctly acted out both familiar and novel transitives. Chan et al. (2010) found that children aged 2;0 could comprehend transitives with familiar but not novel verbs in a preferential looking task; children aged 2;9 showed comprehension with both verb types in a preferential looking task but performed better with familiar than novel verbs on an act-out task. Dittmar, Abbot-Smith, Lieven and Tomasello (2013) explored this question from the opposite side by looking at a situation where information from verb frequency is a disadvantage. They reasoned that since the passive is not the preferred structure for familiar verbs, children should disprefer a passive interpretation for familiar verbs, but they should have no entrenched preference for novel 282 S H A N L E Y E . M . A L L E N C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 283 [271--297] 17.7.2015 5:00PM verbs. Children indeed showed differential performance by verb type at 2;7: they dispreferred a passive interpretation for familiar verbs and showed chance performance for novel verbs. Child-related factors are a third way that processing demands can affect the ability of children to use their abstract representations. For example, Dittmar et al. (2011) found that both fatigue and language ability affected children’s ability to use their abstract representations in a forced pointing task with novel transitives. Children aged 1;9 performed above chance when only the first two (of three) trials per participant were analysed, but only at chance when all three trials were analysed, arguably due to fatigue by the third trial. Similarly, children with ‘low’ CDI vocabulary scores performed at chance (regardless of age), but those with ‘high’ scores performed above chance. Kidd (2012b) also found effects of vocabulary knowledge, grammatical knowledge, and nonverbal reasoning ability on performance in priming tasks, as detailed in Section 13.2.3.2. Finally, several studies show an effect of language-specific factors: children learning different languages become productive with the same argument structure at different ages, which is unexpected if innate linking of semantic and syntactic roles is sufficient on its own to guide children to abstract representations. The most convincing evidence comes from a study comparing same-aged children learning three different languages – Cantonese, English and German – on the same act-out task differentiating argument roles in reversible transitive sentences (Chan, Lieven & Tomasello 2009). While English-speaking children used the first-noun-asAgent strategy, children learning Cantonese and German crucially also depended on language-specific cues such as case and animacy. Indeed, numerous studies support the idea that several cues interact in determining how and when children show their knowledge of abstract representations (English: Ibbotson, Theakston, Lieven & Tomasello 2011, Naigles, Reynolds & Küntay 2011; German: Dittmar et al. 2008b; Italian: AbbotSmith & Serratrice 2014; Turkish: Göksun et al. 2008), mostly using the framework of the Competition Model (Bates & MacWhinney 1987). 13.2.3.2 Evidence from priming studies Structural priming, a paradigm that is relatively new to child research but well-established in adult psycholinguistic, is particularly sensitive to revealing representation strength (Bock 1986). In a typical structural priming study, participants are presented with a ‘prime’ sentence that uses one of the two options in an argument structure alternation, such as the passive The ball was hit by the boy in the active–passive alternation. The participant is then asked to produce or comprehend a ‘target’ sentence that can be expressed using the same alternation. Adults show a tendency to repeat the structural option heard in the prime (here, the passive) rather than the other possible option (here, the active). The general concensus is Verb argument structure 283 C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 284 [271--297] 17.7.2015 5:00PM that this happens because the prime strengthens the participant’s activation of the abstract mental representation of the relevant structure, independent from effects of overlap with lexical items, prosodic structures and the like (see Branigan (2007) for a review). Crucial for our discussion of child acquisition of argument structure, a child should only show structural priming if she already has at least a weak abstract representation for the structure in question, and thus can generalize that representation to a new verb. A strong nativist view would predict that young children evidence priming effects similarly to adults, a strong lexicalist view would predict that young children only show priming if lexical items themselves (i.e. verbs and/or arguments) are repeated across prime and target, and a graded representations view would predict that priming effects are weak at the youngest ages (perhaps lexically driven and susceptible to task effects) and then gradually increase in strength with age. Research hasmainly focused on whether children show a priming effect in general or only when lexical items (verbs or arguments) are identical, and what other conditions need to hold for priming to be evidenced. The earliest studies seemed to indicate that children pass through a stage where they only show priming when lexical items are shared between prime and target. Savage, Lieven, Theakston and Tomasello (2003) showed that 6-year-old English-speaking children could easily generalize the passive structure to new verbs after being primedwith five passive utterances, but 3and 4-year-olds could only generalize if all the primes were identical except for the verb (e.g. It got pushed by it and It got caught by it rather than The brick got pushed by the digger and The ball got caught by the net) and thus had high lexical overlap with the potential target utterance. A follow-up study with 4-year-olds revealed that the effect of priming was stronger when varied verbs were used in the priming phase than when a single verb was used; further, the effect of varied primes persisted for up to a month suggesting that learning occurred during the study (Savage, Lieven, Theakston & Tomasello 2006). Huttenlocher, Vasilyeva and Shimpi (2004) found productive generalization of both active/passive and dative structures with 5-year-olds using methods similar to those used by Savage et al. (2003). However,most language acquisition researchers now support the graded representations view: 3-year-olds show weak representations of argument structures (usually passive or dative) that can be accessed under favourable experimental conditions, and this representation gradually gets more robust through the age of 5 or 6. Shimpi, Gámez, Huttenlocher and Vasilyeva (2007) found weak but productive generalization of both active/ passive and dative structures with 4-year-olds when ten examples were modelled in the priming phase (i.e. double the number used by Savage et al. 2003), and with 3-year-olds when children’s responses were interspersed with the primes rather than the primes and responses occurring in separate blocks. Thothathiri and Snedeker (2008b) found stronger 284 S H A N L E Y E . M . A L L E N C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 285 [271--297] 17.7.2015 5:00PM priming for the dative alternation in 4-year-olds than in 3-year-olds, using an eye-tracking method where children’s eye gaze was analysed to determine whether they expected the test utterance to have the same structure as the prime utterance. Bencini and Valian (2008) found that 3-year-olds showed significant priming effects using the active/passive alternation even when the prime and target did not share either a verb or arguments, as did Kidd (2012b) for children aged 4–6 years. Rowland, Chang, Ambridge, Pine and Lieven (2012) and Peter, Rowland, Chang and Blything (2015) showed similar results for children aged both 3–4 years and 5–6 years using the dative alternation, when only one instance of the prime was presented. Messenger and colleagues showed that 3and 4-year-olds are not simply repeatingword order or semantic role structure from the prime, but that they indeed have an abstract representation: Messenger, Branigan and McLean (2011) showed productive priming from a truncated passive prime to a full passive target, while Messenger, Branigan, McLean and Sorace (2012) showed productive priming from Experiencer–Theme (The girl was seen by the boy) and Theme-Experiencer (The girl was scared by the boy) passives to Patient–Agent (The girl was kissed by the boy) passives. (See Chapter 19 for more on priming in passives.) Two additional points are predicted to distinguish the theories (Rowland et al. 2012). Under the graded representations view, children should show larger priming effects than adults, consistent with findings that individuals with fewer or less well represented abstract structures (e.g. aphasics, L2 learners) are more susceptible to priming (Flett 2006, Hartsuiker & Kolk 1998). By the same logic, children should show a larger priming effect when the prime and target share a verb than when they do not – termed ‘lexical boost’ – and this boost should be larger in children than adults. A nativist view, however, would predict that children show a similar magnitude of priming effect and lexical boost to adults since these two groups have equivalent abstract representations. It is difficult to compare the size of priming effect between adults and children because of substantial methodological differences across studies. However, those studies that tested children and adults using the same methodology reported no significant differences in priming effect between 3-year-olds and adults despite a numerically higher priming effect in children that may reflect their greater susceptibility to priming (Messenger et al. 2011, 2012, Rowland et al. 2012). Several studies have also found no lexical boost in children compared with a clear lexical boost in adults, supporting neither view (Peter, Chang, Pine, Blything & Rowland 2015, Rowland et al. 2012, Thothathiri & Snedeker 2008a). Further studies will need to clarify these findings. Despite these many priming studies, it is still unclear whether priming tests the samemechanisms in adults and children, and thus whether it can really be used as an indicator that children have abstract structural representations (Kidd 2012b). As Kidd (2012b) notes, child priming results Verb argument structure 285 C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 286 [271--297] 17.7.2015 5:00PM typically exhibit high variability, including a substantial proportion of children who do not exhibit any priming. In his own study of passive priming, he found that each of vocabulary knowledge, grammatical knowledge, and nonverbal reasoning ability significantly predicted whether children were primed, the magnitude of the priming effect, and the type of passives used. In contrast, age was not a significant predictor of any of these outcomes. Such results strongly suggest that it is not justified to interpret a given performance on priming tasks as an indicator of developmental level for a particular age group (Kidd 2012b). While it is clear that the priming methodology offers important insights into the question of children’s early abstract knowledge of argument structure, further research is needed to interpret these insights in a more accurate and nuanced way. 13.3 Trajectory and patterns in the acquisition of argument structure alternations Looking at argument structure acquisition from the point of view of theory often focuses on abstract questions for which data are secondary. In the next section, we focus on argument structure acquisition from the point of view of the structures themselves – specifically passive and dative – in order to get a sense of the trajectory and patterns of development within one structure. 13.3.1 Passives Typical passives are shown in the (b) sentences in (2) and (3), with their active transitive counterparts in the (a) sentences. (2) a. Elizabeth chased Gisela. b. Gisela was chased by Elizabeth. (3) a. Marion climbed the big tree. b. The big tree was climbed by Marion. In the active transitive sentences the Agent appears in Subject position while the Patient appears in Object position. In passive sentences the linking between syntactic roles and semantic roles changes: the Patient appears in Subject position while the Agent optionally appears in an adjunct phrase introduced by the preposition by in English. This entails that the standard word order also differs for the two structures: Agent– Verb–Patient for active transitives and Patient–Verb–Agent for passives. The passive is permitted crosslinguistically for virtually all transitive action verbs like those in (2) and (3) with Agent Subjects and Patient Objects. However, only a subset of non-action verb classes for which 286 S H A N L E Y E . M . A L L E N C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 287 [271--297] 17.7.2015 5:00PM the Subject and Object are linked to other semantic roles permit the passive, and those classes differ across languages (Pinker et al. 1987). For instance, psychological verbs permit passivization in English (e.g. The paintings were admired by the tourists) but verbs of pure possession do not (e.g. *A new game was had by the brothers). From the discourse perspective, the passive focuses attention on the Patient of the transitive action and defocuses the Agent. The passive is the most frequently studied argument structure in the acquisition literature. It is an ideal test of whether children’s apparent comprehension and production of argument structures derive from a true understanding of the argument structure of a particular verb, or reflects knowledge of the real-world context or the most frequent or default pattern that occurs with that verb in the input. In addition, the differential distribution of the passive across semantic classes of verbs predicts that children will quickly learn and easily overgeneralize passives using verbs with Agent Subjects and Patient Objects, but will learn the passive of other classes of verbs more slowly (Pinker et al. 1987). Four main findings concerning passives have been central to the literature on argument structure acquisition. First, children’s early comprehension and production of passives is strongly influenced by their reliance on the word order and linking patterns of the much more frequent active structure and by their knowledge of real-world context (see O’Grady (1997) and references therein). Both comprehension and production errors involve reversing the roles of the arguments – treating the Patient Subject as if it were the Agent and the Agent in the adjunct by phrase as if it were the Patient. Children also comprehend and produce passives earlier when the Agent and Patient are non-reversible as in (3) (i.e. one cannot say Marion was climbed by the tree) than when they are reversible as in (2), reflecting the influence of real-world plausibility of events. Indeed, English-speaking children may not fully understand the argument structure of passives until age 6 or later even though they produce passive structures earlier than that. A second finding is that children are sensitive to semantic classes of verbs in learning the passive. Two semantic classes are typically distinguished: action verbs and non-action verbs. Sudhalter and Braine (1985) tested the ability of children aged 3–11 to identify the Agent (for action passives) or Experiencer (for non-action passives). Children aged 3–6 performed almost twice as well on passives containing action verbs (54–58%) as compared to non-action verbs (26–29%). Even for 11-yearolds there was a clear difference between the two types of verbs (85% for action; 70% to 77% for different types of non-action). Similar results have been found in other studies using both real verbs (e.g. Gordon & Chafetz 1990, Maratsos, Fox, Becker & Chalkley 1985) and novel verbs (Pinker et al. 1987). However, Messenger et al. (2012) point out that this Verb argument structure 287 C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 288 [271--297] 17.7.2015 5:00PM apparent late mastery of non-action passives may in fact be a task effect: although 3to 4-year-old children in their study performed poorly with non-action verbs on a picture-sentence matching task like that used in most previous studies, they performed well on a relatively less demanding priming task. Third, the timing of passive acquisition is affected by the frequency of use of passives in the input. The passive is acquired relatively late – usually sometime between 3 and 4 years of age – in languages such as English and Germanwhere it is used infrequently in child-directed speech (0.4–1.3 times per hour; Abbot-Smith & Behrens 2006, Pinker et al. 1987). However, the passive is acquired and productively used as early as 2;0 in Inuktitut, K’iche’ Mayan, Kigiriama, Kiswahili, Sesotho and Zulu, languages in which passives occur as much as forty times more frequently in the input than in English (Alcock, Rimba & Newton 2012, Allen & Crago 1996, Demuth 1989, Demuth & Kline 2006, Pye & Quixtan Poz 1988). English-speaking children’s production and comprehension of the passive improves when frequency of passive input is increased as part of experimental conditions (Baker & Nelson 1984, Brooks & Tomasello 1999, Pinker et al. 1987, Vasilyeva, Huttenlocher & Waterfall 2006). Input frequency may also explain the difference in time of acquisition between action and non-action passives since action passives are much more frequent in child-directed speech (Gordon & Chafetz 1990). In addition, AbbotSmith and Behrens (2006) showed that the timing of acquisition of one German-speaking child’s acquisition of the adjectival (sein) vs verbal (werden) passive was strongly related to the differential frequency of his mother’s use of these two forms, as well as to the child’s and his mother’s frequency of use of other related constructions (e.g. sein copula construction, werden future). Finally, children show evidence of abstract representations for the passive structure later than that for the active transitive, and the strength of these representations develops with age. For example, a forced choice pointing task reveals that German-speaking children interpret active transitive sentences with both familiar and novel verbs reliably by 2;3 but show much later proficiency with passives: they are at chance in interpreting passives at age 2;3, interpret them as transitives at age 2;6, correctly interpret passives with familiar verbs at 3;7, and correctly interpret passives with novel verbs only at 4;7 (Dittmar et al. 2013). Priming studies with passives reveal a similar trajectory of growth of representation strength (see Section 13.2.3.2). Overgeneralization errors are another source of evidence for weak early abstract representations for passives: such errors are virtually non-existent before 2;6, and typically do not start appearing until well after age 3 in English (Pinker et al. 1987). Examples shown here are from English (4a), German (4b) and Inuktitut (4c) (Allen & Crago 1996, Pinker et al. 1987). 288 S H A N L E Y E . M . A L L E N C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 289 [271--297] 17.7.2015 5:00PM (4) a. Until I’m four I don’t have to be gone (= taken to the dentist). (3;6) b. Der Löffel ist besuppt. ‘The spoon is souped.’ (3;6) c. Siaqri-tau-vuq. slide-PASS IVE IND ICAT IVE .3S INGULAR .SUB JECT ‘It was slidden.’ (3;3, child’s foot slid on a slippery surface) Elicited production studies also show evidence of the limited early generalizability of the passive (see Section 13.2.2). Twoand 3-year-olds trained on novel verbs with the active transitive structure rarely generalize these verbs to the passive during the testing phase, even when prompted to do so with questions focused on the Patient such as What happened to the car? for an action in which Big Birdmeeked the car (Brooks & Tomasello 1999b, Wittek & Tomasello 2005). 13.3.2 Dative alternation Verbs that permit the dative alternation may appear in either the prepositional dative construction (5a) or the double object dative construction (5b). (5) a. Elizabeth gave the book to Wolfgang. b. Elizabeth gave Wolfgang the book. Most verbs of transfer that take Patient and Recipient Objects allow the alternation such as give, bequeath, take, send, slide, throw, sell, build, prepare and tell. However, the double object construction is not permittedwith Latinate verbs (e.g. *Melba donated the library the book), in situations where the Recipient cannot reasonably be construed as a possessor of the Patient (e.g. *Melba sent China the book), or with a variety of semantic classes of verbs including verbs of saying (e.g. *Melba confessed Roy the secret), verbs of manner of speaking (e.g. *Melba barked Roy an order), and verbs of selection (e.g. *Melba selected Roy a new tie). The alternation is also restricted by the form in which the arguments are realized: the double object dative is atypical when the Patient is a pronoun (e.g. ?Melba gave Roy it), and much more common than the prepositional dative when the Recipient is a pronoun (e.g. Melba gave him the book). From a discourse perspective, the prepositional dative highlights the transfer event while the double object construction highlights the endstate of transfer (usually possession of the Patient by the Recipient). Bresnan, Cueni, Nikitina and Baayen (2007) offer a statistical model to determine the complex interaction of several discourse and syntactic factors in predicting when the prepositional vs double object datives are likely to be used. Children comprehend and spontaneously produce both forms of the dative alternation from a relatively early age. The first spontaneous forms appear in children’s speech in English when their utterances have Verb argument structure 289 C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 290 [271--297] 17.7.2015 5:00PM a mean length of two words; this corresponds to ages between 1;6 and 3;4 depending on the child (Campbell & Tomasello 2001, Gropen et al. 1989, Snyder & Stromswold 1997, Viau 2006). Several different verbs appear in children’s earliest dative constructions although give is one of the first verbs produced and the most frequently used. In spontaneous speech transcripts of seven children aged 1;3–5;1, Campbell and Tomasello (2001) found that the majority of children used some dative alternation verbs in both possible constructions (give, get, make, show, bring, read), some in only the double object construction (tell, feed, hand, pay), and some in only the prepositional dative construction (fix, leave, open, take); this differentiation occurs in adult speech as well (Thothathiri & Snedeker 2008a). Campbell and Tomasello (2001) also found that most verbs that appeared in one or other of the dative constructions had first appeared in the child’s data in a simple transitive construction where the child was the implied recipient (Read story, Give that) or where the child specified the recipient in later conversation (Make a cake, I may give some). Findings from the above-mentioned studies all show that the first use of the double object construction in English typically precedes or occurs at the same time as the first use of the prepositional dative; Viau (2006) shows an average temporal gap of 3.3 months between the two in his transcript study of twenty-two children. This difference may be influenced by input frequency since the double object construction occurs more often in speech to English-speaking children, even though both dative constructions are used frequently in the input with multiple verbs (Campbell & Tomasello 2001, Snyder & Stromswold 1997). It may also be influenced by semantic differences between the two constructions such as the emphasis on motion of the Patient (prepositional dative) vs eventual possession of the Patient (double object construction) proposed by Gropen et al. (1989). Viau provides evidence that linguistic elements containing the semantic primitive HAVE, assumed to underlie possession in the double object construction, are acquired earlier than those containing the semantic primitive GO which is assumed to underlie the motion component of the prepositional data. Givenness also plays a role in children’s use of the two options: both Englishand Norwegian-speaking children aged 4–6 tend to produce the prepositional dative when the Theme is given, and either produce double object datives or omit the Recipient when the Recipient is given (Anderssen, Rodina, Mykhaylyk & Fikkert 2014, Gropen et al. 1989, Stephens 2010). De Marneffe, Grimm, Arnon, Kirby and Bresnan (2012) used a statistical model to analyse the complex interaction of several discourse and syntactic factors in determining children’s choice between the double object dative and prepositional dative structures. Evidence pertinent to the development of verb-general representations of the dative alternation comes from the same three sources as for the passive: overgeneralizations, elicited production studies, and priming 290 S H A N L E Y E . M . A L L E N C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 291 [271--297] 17.7.2015 5:00PM studies. Gropen et al. (1989) summarized literature showing that children rarely overgeneralize the dative alternation, that dative overgeneralizations begin appearing younger than passive overgeneralizations and continue for several years, and that they appear only after children have begun using the dative forms correctly. Their own study of transcripts from five children showed that overgeneralizations account for about 5 per cent of the double object constructions produced, and occur only once in every 4,000 or so utterances. Some examples compiled by Gropen et al. are given in (6). (6) a. I’ll brush him his hair. (2;3) b. How come you’re putting me that kind of juice? (2;4) c. I said her no. (3;1) White (1987) showed that children age 3–5 can interpret and act out overgeneralized double object constructions, while Mazurkewich and White (1984) found that 9-year-olds judged as grammatical almost half of the erroneous overgeneralized double object constructions on a grammaticality judgment test. Ambridge et al. (2012, 2014) also showed that children aged 5–10 accept a broad range of dative overgeneralizations with novel and familiar verbs, constrained by effects of frequency and semantic class (see Section 13.2.1.1). These results suggest that children have some abstract representation of the dative structures from relatively early in acquisition but take a long time to work out the limits of the pattern. Gropen et al. (1989) conducted an elicited production study to test the strength of children’s verb-general representations of the dative alternation. They taught four novel verbs each denoting a novel event (e.g. sliding a ball through a tunnel to a mouse at the other end), two with the double object dative and two with the prepositional dative. After each verb was taught, they asked the child to describe the event with questions eliciting both the double object construction (e.g. Can you tell me what I’m doing with the mouse?) and the prepositional dative (e.g. Can you tell me what I’m doing with the ball?). The children in their study, aged 6–8 years, easily generalized the novel verb to the non-modelled structure. In a similar study with 3-year-olds in which the children were simply asked to describe the novel event to their caregiver, Conwell and Demuth (2007) found that virtually all child descriptions used the modelled construction. However, children in a follow-up study who heard one action described with the double object dative and the other with the prepositional dative used the non-modelled construction in 31 per cent of their own descriptions. This suggests that children have an understanding of the dative alternation that they can use productively in at least some circumstances, consistent with the ‘weak representation’ hypothesis (see Section 13.2.3). Rowland and Noble (2011) found similar evidence from a forced-choice pointing task using novel verbs. Threeand 4-year-olds correctly pointed to the transfer event (as opposed to a caused action event) for both prepositional object Verb argument structure 291 C:/ITOOLS/WMS/CUP-NEW/6465742/WORKINGFOLDER/BAVIN/9781107087323C13.3D 292 [271--297] 17.7.2015 5:00PM and double object dative prompts, although they were successful with the double object datives only when the theme and recipient nouns were distinctively marked (here, one as a proper noun and one as a common noun). Several priming studies have tested the strength of children’s abstract representations of the dative, concluding that they are relatively weak at age 3 and strengthen with age (see Section 13.2.3.2). A computa-representations of the dative, concluding that they are relatively weak at age 3 and strengthen with age (see Section 13.2.3.2). A computationalmodelling study focused on the dative alternation provides evidence for different levels of generalization emerging at different stages of learning (Barak, Fazly & Stevenson 2014). 13.4 Argument realization Much of the literature on argument structure acquisition assumes that arguments are always present in caregiver speech to provide the full input necessary to child learners, and that children produce all the arguments that a verb requires. However, this is not always the case. The omission of arguments is common in many languages. (7) a. Habl-o con mi abuel-a cada dia. speak-1S ING . SUB J ECT with my grandparent-FEM every day ‘(I) speak with my grandmother every day.’ b. Bei1. carry ‘(The child) carried (the puppy to Grandma).’ (Lee & Naigles