Abstract
This handbook provides an excellent overview of the main findings in syntax and semantics over the last couple of decades, and of how the study of language acquisition has supported the generative grammar theoretical framework. There is an emphasis on English, but most authors supply insightful discussions of cross-linguistic research.
In ‘Missing Subjects and Early Child Language’, Nina Hyams examines various analyses of early null subjects, considering purely linguistic, as well as processing accounts. Hyams begins by considering her early work (Hyams, 1983, 1986) in which she suggested that young children acquiring a non-pro-drop language often leave out subjects because they have positively mis-set the pro-drop parameter, the correct setting for pro-drop languages like Italian, but not for languages like English. At some stage in their development, English-speaking children must reset the pro-drop parameter to the negative setting. Hyams’s early account seemed to violate the subset principle, but a child acquiring English can reset the pro-drop parameter to the correct value based on positive evidence alone, i.e. the occurrence of expletive pronouns and the distribution of modals. A range of syntactic accounts (Morphological Uniformity, The Topic Drop Hypothesis, Competing Grammar Hypothesis, Truncation), as well as pure processing accounts are then considered by Hyams. Various studies show that children tend to drop subjects that can be recovered from the context, i.e. arguments that express old information, and that these null subjects are replaced by overt pronouns at a later developmental stage. Processing accounts fail to explain why there is a steady increase in the production of pronominal rather than lexical subjects, and why children omit subjects more often in non-finite than in finite clauses in English and related languages. The same criticism is raised for metrical accounts. Hyams concludes that there is to date no comprehensive theoretical model that can capture all the intricate details of early null subjects.
In ‘Grammatical Computation in the Optional Infinitive Stage’ Ken Wexler considers the main features of Optional Infinitives (OIs), and evaluates different accounts, in particular Truncation and the Unique Checking Constraint (UCC). Wexler was the first to propose that such an OI stage exists and that OIs are a cross-linguistic phenomenon. A child in the OI stage knows the relevant properties of verb movement. However, in contrast to adults, children in the OI stage produce finite as well as non-finite verbs in contexts requiring finite verbs. When children produce finite verbs these generally agree with the subject, and when they produce non-finite forms these are not necessarily errors of omission as in English (e.g. talk instead of talks), but are also true infinitival forms, e.g. German sprechen ‘talk’, which consists of the root sprech and the infinitival ending –en. Different models for the OI phenomenon are discussed, but Wexler mainly concentrates on two radical omission models: UCC (Wexler, 1998) and Truncation (Rizzi, 1993). He argues that UCC is superior to Truncation because it can explain why there is a correlation between pro-drop languages and the absence of an OI stage. Most of the remainder of the chapter is dedicated to a critical evaluation of Legate and Yang’s (2007) account of OIs. These authors hypothesize that the bigger the proportion of ambiguous verb forms in the input the longer it will take the child to realize that the target language is [+Tense]. Wexler objects that the basis on which verb forms are classified as showing ambiguous or unambiguous Tense is far from clear, and that it would be unclear why the child would not simply infer from past tense –ed that English -0 in they talk must also be [+Tense]. After discussing some behavioural and genetic evidence relating to the development of finiteness Wexler concludes that ‘the UCC dies away under genetic influence, just as a first set of teeth falls out under genetic influence, and a second set develops under genetic influence’ (p. 90).
In ‘Computational Models of Language Acquisition’ Charles Yang discusses various models before introducing his variational learning model. Since technical concepts like ‘finite-state languages’ and ‘Bayesian’, for example, are invoked several times in the chapter a brief overview of these would have been welcome. One of the primary goals of research on language acquisition is to explain how a child arrives at the target grammar within such a short period of time, based on a finite number of examples in the input, although ‘a finite number of examples are consistent with infinitely many hypotheses’ (p. 125). By developing computational models, learnability research with a focus on the study of language tries to uncover how this can be achieved. Yang argues that, without prior knowledge that restricts the hypothesis space, learning a natural language would be impossible. His own learning model assumes that the child has innate knowledge of parameters and of their ‘signatures’ (comparable to triggers). The higher the proportion of signature sentences in the input the faster the child will set the related parameter. The fact that the Optional Infinitive stage ends earlier in French than in English can be explained as a direct response to the proportion of ambiguous verb forms in the input. Yang notes that computational models have gained in importance in linguistic research, but that they should be guided by findings in psycholinguistic research. He concludes that models of language learning are not yet close to matching the child’s linguistic competence.
In ‘The Acquisition of the Passive’ Kamil Ud Deen emphasizes the importance of cross-linguistic research. After a critical discussion of two approaches trying to account for the delay in the acquisition of passives in English he reviews several studies that seem to indicate that passives in other languages are acquired early, and that, even in English, there is evidence that passives are not as delayed as originally believed. English-speaking children appear to have more difficulties with so-called long passives (with an overt by-phrase) than with short passives, and also to have more difficulties with passives containing non-actional (e.g. see) as opposed to actional (e.g. chase) verbs. Both types of passives involve movement of the object to the subject position (A-movement). The A-chain deficit hypothesis (ACDH) by Borer and Wexler (1987, 1992) suggests that A-movement is delayed, and that children find actional passives easier to understand than non-actional ones because they can construe them as adjectival passives not involving A-movement, an option that is not available with non-actional passives.
A clear prediction of the ACDH is that all constructions involving A-movement should be similarly delayed. Leaving the VP-internal subject hypothesis aside, unaccusatives and raising constructions should emerge at the same time as verbal passives. Children may interpret unaccusatives as unergatives, intransitive verbs that do not involve A-movement. Experimental results on Russian partially support this option, while evidence from Japanese does not: unaccusatives are acquired early in Japanese. In the case of raising verbs, children may interpret examples like ‘John seems to be happy’ as ‘John is happy’. Some experimental data on child English support this view. According to the Theta-Transmission-Model by Fox and Grodzinsky (1998) the external theta-role is transmitted via the subject position to the nominal in the by-phrase, and it is this transmission that is problematic for children. By itself can assign a theta-role, but to an affected nominal only. The predictions are only partially supported by experimental evidence. Moreover, since passives are rare in the input to English-speaking children, it is difficult to rule out that a delay in the comprehension and production of passives is not simply due to their rare occurrence. Data from Sesotho, a southern Bantu language, where passives are frequent, evidence from priming studies with young English-speaking children, and experimental data on Cantonese, a language in which the by-phrase is obligatory, undermine the ADCH and the Theta-Transmission-Model. Dean concludes that children’s grammatical knowledge of passives may be fully intact and thus not subject to maturation.
In ‘The Acquisition Path for Wh-Questions’ Tom Roeper and Jill de Villiers discuss a wealth of data from different languages, and convincingly show that results from language acquisition research can both provide a fruitful testing ground for linguistic theories and reveal subtle properties of grammar. The authors suggest that Subject-Auxiliary Inversion (SAI) in English wh-questions lags behind SAI in yes/no questions because the child starts out with a proto-CP that does not make two positions available (Spec and Head) before they ‘split’ CP, favouring Merge over Move. The special behaviour of certain wh-words (how come you can sing) may further delay SAI in wh-questions. In German, a V2 language, inversion in wh-questions is acquired fast because finite verbs generally move to C in root contexts.
Turning to the interpretation of wh-questions, young children, unlike adults, start out by providing non-exhaustive answers, and then move from singleton reference to exhaustive reference, bypassing plural reference. In wh-questions allowing a long distance as well as a short distance construal, children prefer the long distance construal whereas adults prefer the short distance construal. Children’s preference for long distance readings, and their production of medial wh-questions – ungrammatical in English but possible in other languages – are seen as evidence for partial movement in young English-speaking children’s grammar. Roeper and de Villiers contend that ‘the child’s grammar may be constrained by an idealization of the Strong Minimalist Thesis as an interface principle that combines syntax, phonology and semantics’, and in which each Phase is treated one at a time (p. 239).
In ‘Binding and Coreference: Views from Child Language’ Cornelia Hamann reviews a large body of cross-linguistic acquisition research and critically evaluates various theoretical models. According to the standard Binding Theory, anaphors (e.g. herself) must be locally bound, i.e. c-commanded by a coindexed NP, pronouns (e.g. her) must be locally free, and R-expression (e.g. the cat, Tommy) must be free. The observation that binding theory cannot account for all observed phenomena has led researchers to assume that there may in fact be two mechanisms for establishing the reference of anaphoric elements: binding, which is constrained by syntax, and coreference, which is constrained by pragmatics. One of the early findings in acquisition research was that English-speaking children usually assign the correct meaning to anaphors, but that they often interpret the meaning of pronouns incorrectly, an error known as ‘Delay of Principle B’. However, they perform much better on sentences with quantified NPs (e.g. every bear) and pronouns. This contrast in performance led Chien and Wexler (1990) to propose that children know Principle B, but that they do not yet know when accidental coreference, governed by a pragmatic rule, is excluded. The term ‘Delay of Principle B’ was replaced by the ‘Pronoun Interpretation Problem’ (PIP). In Romance languages, in which pronouns are clitics, PIP does not surface in simple sentences but can surface in ECM (Exceptional Case Marking) constructions. By carefully controlling experimental conditions, e.g. by having lead-in sentences with a possible antecedent for the pronoun (Mama Bear says that Baby Bear washes her), even English-speaking children’s comprehension of pronouns improves considerably. Hamann considers this to show that there may be some truth in the original pragmatic explanation for PIP.
In ‘Universal Grammar and the Acquisition of Japanese Syntax’ Koji Sugisaki and Yukio Otsu review some of the major findings in Japanese. The earliest utterances by children show that they are sensitive to the basic SOV (Subject-Object-Verb) word order in Japanese. Although the children produce OV as well as VO order, they do not produce object wh-questions with VO, in agreement with the target grammar.
In Japanese, OSV order is derived by scrambling of the object to sentence-initial position. On the basis of comprehension studies, the authors demonstrate that Japanese three-year olds are aware that OSV sentences are derived by movement of the object. Based on case-marker drop and the restriction on the distribution of the formal noun koto, they provide evidence that Japanese has a VP (Verb Phrase) and that it is configurational, contrary to early assumptions in the 1980s. Production data on case drop and comprehension data on the distribution of koto show that Japanese children at the age of 3 are aware that Japanese is a configurational language. At the same age, Japanese children also display knowledge of the wh-island and adjunct-island constraints, which the authors assume to ‘directly reflect properties of UG’ (p. 303).
Based on the comprehension of constructions with quantifier float and the interpretation of questions, Japanese children are also shown to display early knowledge of structure dependency. By the age of 3, they know the main properties of zibun, a reflexive anaphor, and they understand both direct and indirect passives, provided the child’s point of view rather than the surface subject’s point of view is adopted. The authors suggest that the derivation of passives is acquired earlier than pragmatic knowledge.
To fully appreciate these Japanese findings a more cross-linguistic perspective would be useful. For instance, does knowledge of scrambling in Japanese surface earlier than in Dutch- and German-speaking children? Some information on the input, for example, how often certain constructions (case drop, VO) occur would also have been welcome.
In ‘Studying Language Acquisition Through the Prism of Isomorphism’ Julien Musolino addresses an early theory of his (Musolino, 1998), which has instigated much research into quantification. In sentences with negation and quantified NPs with every (e.g. Every horse didn’t jump over the fence) children, unlike adults, overwhelmingly tend to read off quantifier scope from surface structure, thus interpreting ‘every’ as having scope over ‘not’, an interpretation referred to as ‘isomorphic interpretation’ by Musolino. In the non-isomorphic interpretation negation takes scope over ‘every’. Under certain experimental conditions the isomorphism effect can also be elicited in adults. Given a sentence like Two frogs didn’t jump over the rock and a situation in which two frogs tried jumping over a rock but only one succeeded, adults often judge the sentence as false, although under the non-isomorphic interpretation (not two frogs jumped over the rock) it is true. On the other hand, children can be shown to access the non-isomorphic interpretation (not every horse jumped over the fence) in a sentences like Every horse jumped over the log, but every horse did not jump over the fence, given a situation in which there are three horses, and all three jump over a log, but only two manage to jump over a fence.
The observation that under certain experimental conditions the isomorphism effect can be increased in adults and reduced in children undermines Musolino’s original account. Crucially, if children, like adults, can potentially access both interpretations of an ambiguous sentence, why do they resolve scope ambiguities differently from adults? Corpus data reveal that the isomorphic intended reading is much more frequent than the non-isomorphic interpretation. In research on sentence processing the human parser is assumed to be sensitive to such distributional cues. Studies in sentence processing indicate that a commitment to surface scope interpretation, i.e. the isomorphic interpretation, is favoured, and that children are less efficient than adults at revising initial parsing commitments. This leads to garden-path effects in children. Isomorphism is thus one of several factors that play an important role in language processing. Although his early theory turned out to be inadequate, Musolino concludes that it has paved the way to a fruitful research programme, giving rise to novel and interesting research questions.
In ‘Acquiring Knowledge of Universal Quantification’ William Philip provides a pragmatic account for a common comprehension error. A large body of experimental literature shows that children as old as 6;0 still have difficulties in assessing the truth of sentences like Each dog is licking a cat when they are shown a picture of three dogs, each licking a cat, and another cat not being licked. This comprehension error goes by different names: ‘exhaustive pairing’ (EP), ‘over-exhaustive search’, ‘symmetrical interpretation’, or ‘quantifier spreading’. When asked why they judge the sentence as incorrect, children generally justify their answer by pointing out that there are not enough dogs or that one cat is not being licked, although this cat – the extra object – is irrelevant. Children also show two other types of comprehension errors labeled as ‘perfectionist response’ (‘exhaustive response’) and ‘under-exhaustive search’ respectively. The perfectionist response consists in incorrectly judging the sentence as false in the case where two dogs are each licking a cat, and a pony is either licking a cat or not licking anything. The under-exhaustive search consists in incorrectly judging the sentence as correct in the case where two dogs are each licking a cat, and another dog is licking a bird or not licking anything. These two types of errors are said to be much less common than the EP error (but the percentages of under-exhaustive search errors appear to be higher than those of ER (Exhaustive Response) errors in the French, Spanish and Norwegian studies listed in Table 2, p. 366). Philip notes that a unified explanation for all three types of errors will not be able to account for this asymmetry, and that instead he will concentrate on the EP error, since it is the most frequent. In his pragmatic account the following three factors play a role: two UG (Universal Grammar) principles – Normal World Knowledge and Salient Object Strategy – and a pattern-recognition mechanism. These two UG principles are ‘predestined to emerge as a consequence of the interaction of genetic and environmental factors’ (p. 320) and may emerge late. Philip concludes that UG governs the acquisition of quantification only indirectly, and that much of UG exists only to speed up the acquisition process, a rather provocative view.
The authors in this book not only contribute a wealth of data from English and other languages, but they also convey the excitement of experts in the field tackling frontier problems in language acquisition. I only have three minor issues with the content and presentation of the text. Firstly, although the book contains a plethora of interesting and intriguing data and provides the reader with many insightful and critical discussions, there is no discussion of bilingualism and of second language acquisition. Given the focus on first language acquisition, the title ‘Handbook of Generative Approaches to First Language Acquisition’ would have been more appropriate. Secondly, the introduction to the book by Jill de Villiers and Tom Roeper contains excellent summaries of the individual chapters but I would have appreciated a more extensive introduction to the theoretical framework. Thirdly, it would have been useful for each chapter to include a short abstract summarizing the author’s perspective on the topic.
This book should be essential reading for graduate students in language acquisition and would be a useful overview for any advanced student in syntax. This is an essential text in the library of any linguist interested in Generative Grammar.