There is a range of views on how children acquire language. The generative approach assumes that children are biologically fitted with language-specific principles and parameters, i.e. a Universal Grammar (UG). The theory of UG is motivated by poverty-of-stimulus considerations – evidence that children know too much, at too early an age, to have learned certain aspects of their linguistic competence from experience. Therefore, in addition to more general capacities for formulating and testing generalizations, UG postulates that every child without cognitive or other nonspecifically linguistic deficits comes equipped with a set of innate principles and parameters. These establish the boundary conditions for language development; the child is free to navigate within these boundaries but cannot venture beyond them.

Empirical support for UG comes in three forms. First, there are studies demonstrating that children do not overstep the boundary conditions imposed by the principles of UG. Children form generalizations based on the primary linguistic data, but these generalizations are held in check by UG principles, so they are not the kinds of generalizations one would expect if children were solely employing domain-general learning mechanisms. A second source of evidence that children adhere to a UG provides the foundation for the Continuity Hypothesis. This hypothesis states that child and adult languages can differ, but only in ways that human languages can differ from each other (cf. Crain & Pietroski Reference Crain and Pietroski2001). One way for this to happen is for children to adopt parameter values that are attested in human languages but not characteristic of the target language. Evidence of parameter mis-setting is probably the most compelling evidence for UG and against experience-based approaches to language acquisition. The third source of evidence that UG is operative in child language is the observation that children command seemingly unrelated facts that are tied together by ‘deep’ properties of UG, sometimes encoded in a parameter. In his book, William Snyder focuses on this third aspect of the UG-based account of parameters. Interestingly, though, the approach Snyder takes to the assessment of children's linguistic competence compels him to reject the Continuity Hypothesis.

Of course, advocates of experience-based approaches reject all three of these nativist claims about the course of language acquisition, including the Continuity Hypothesis. These researchers view children's developing grammars as closer and closer approximations to those of adult language users. Let us call this conception of language development ‘grammatical conservatism’. In one sense, grammatical conservatism serves a similar function for experience-based approaches as innate constraints do for advocates of UG, namely to prevent children from committing overgeneralizations. But grammatical conservatism is incompatible with the observation that children's grammars differ in qualitative ways from those of adult speakers of the local language. Whereas children's utterances may not be a perfect match to those of adults, children's utterances are expected to be ‘pared-down’ versions of them, according to experience-based accounts. In short, if grammatical conservatism is right, the Continuity Hypothesis is wrong.

Snyder's ‘parametric approach’ straddles the fence. While Snyder maintains that children come equipped with the principles and parameters of UG, he also embraces grammatical conservatism, to which he is led due to his insistence that children's spontaneous speech is the optimal source of evidence about children's grammars, as seen in the following quote.

The privileged status that Snyder grants to children's spontaneous productions brings him in line with the experience-based approach in the sense that children are seen as attempting to match their input, rather than as free to try out grammatical hypotheses based on the parameters made available by UG. If Snyder is right, and the Continuity Hypothesis is wrong, then we have one fewer arguments to adduce in favour of UG.

Grammatical conservatism leads Snyder to consider certain theoretical modifications to the parameters of UG. The proposal that is most responsive to grammatical conservatism is to restructure parameters so that their (binary) values form a subset/superset relation, although Snyder points out that this may be problematic for some parameters. Assuming the absence of negative evidence in the input to children, the subset value is the initial setting. This ensures that children's spontaneous speech will always match the language that is spoken by adults, regardless of the value assigned by adult language users. If at first children assign a different value than adults, they will simply undergenerate, with their productions constituting a subset of those spoken by adults. In that case, children are compelled to extend their grammars in response to positive evidence that the superset value of the parameter is operative in the local language.

As for the empirical evidence for the Continuity Hypothesis, there are two types: (i) errors of omission and (ii) errors of commission. The parade case of an error of omission resulting from parameter mis-setting appeared in Hyams (Reference Hyams1986), who proposed that English-speaking children initially adopted the ‘wrong’ setting of the pro-drop parameter. But Snyder notes that such errors of omission are not compelling evidence of children's linguistic competence, since they could just as well be the product of the child's immature performance system (see Bloom Reference Bloom, Barss, Nicol and Conway1990), and that errors of commission would constitute better evidence of mis-set parameters. Most of Snyder's discussion of errors of commission centres on the experimental tasks that have evoked children's non-adult behaviour. According to Snyder, the findings from experimental tasks tend to induce errors that do not accurately represent children's grammatical competence. I have more to say on this below.

By necessity, Snyder's attribution of a principles-and-parameters-based grammar to children takes a different form from the one in the classic nativist approach. To investigate a given parameter of UG, Snyder adopts the following research strategy. The first step is to survey the typological landscape across languages, identifying the linguistic properties (manifest in different linguistic constructions) that are associated with the alternative values of the parameter under investigation. The second step is to investigate the time course of acquisition of these properties, to see if the constructions associated with a particular value emerge at the same time in children's spontaneous productions. Since two constructions could emerge at the same time for other reasons, such as their computational complexity, such non-linguistic explanations of simultaneous emergence must be factored out. This is done, statistically, by taking Mean Length of Utterance (MLU) to be a measure of computational complexity, and assessing its contribution to the correlations that obtain among the relevant constructions. Provided the contribution of MLU to the correlation among constructions is not significant, the inference is licensed that the two constructions are related in the child's grammar, in virtue of being tied to the same parameter.

Snyder's book is an exposition of research using and developing the logic of this research strategy. The introductory chapter (‘A brief introduction’) explains how longitudinal studies of children's spontaneous speech are a valuable resource for understanding cross-linguistic variation in the language faculty. This chapter also introduces grammatical conservatism as a major theme of the book. Chapter 2, ‘The view from syntactic theory’, introduces the generative model of syntax, expounding the role that parameters played in the pre-Minimalist Principles and Parameters theory, and their role in the recent Minimalist Program, where cross-linguistic variation, in the form of parameters, is reduced to variation residing in lexical items. The chapter contains an excellent discussion of three parameters that have been proposed within the Minimalist Program (Longobardi's (Reference Longobardi, Mark and Collins2001) Referentiality parameter, Bobaljik & Thráinsson's (Reference Bobaljik and Thráinsson1998) Split-IP parameter and Bošković's (Reference Bošković2004) DP parameter), and of whether or not they can be reduced to the Minimalist desideratum of lexical variation. Chapter 3, ‘The view from phonological theory’, is concerned with phonology and introduces the broad underpinnings of Optimality Theory and Government Phonology. Here, the discussion centres on whether one or the other framework is more likely to generate output from the child that is grammatically conservative. The result is inconclusive, leaving open the question of whether conservatism extends to children's phonological output.

Chapter 4, ‘The view from children's spontaneous speech’, and chapter 5, ‘Statistical methods for longitudinal studies’, will be of most interest to researchers who may be planning to use spontaneous speech from the Child Language Data Exchange System (CHILDES) to test linguistic hypotheses. In chapter 4, Snyder draws on his considerable experience with database research to provide an extremely useful step-by-step introduction to the testing of hypotheses and the organization and analysis of naturalistic data, using an Excel spreadsheet. A case study of the verb–particle construction serves as a working example. Chapter 5 contains a tutorial on the kinds of statistics that are appropriate (or not) for evaluating the concurrent or sequential emergence of constructions hypothesized to be related to a particular parameter. Using once again the verb–particle construction for illustration, Snyder observes that English-type languages with verb–particle constructions also have productive compounding. Therefore, the experimental hypothesis is that no child should acquire the verb–particle construction significantly earlier than root compounds. This careful presentation of empirical predictions and the statistical means for evaluating them will make Snyder's book the standard companion for researchers using this methodology to study children's spontaneous productions.

Chapter 6, ‘Experimental and statistical methods for cross-sectional studies’, discusses other experimental methods that are frequently used to assess children's linguistic knowledge. The discussion centres on the utility of these methods for investigating parameters. Most of the discussion concerns two tasks, the elicited production task and the truth-value judgement task, but the picture-selection task is also briefly discussed. There is also discussion of whether the statistical methods prescribed by Snyder for longitudinal studies are appropriate for cross-sectional studies. Chapter 7, ‘Case studies in the parametric approach’, presents three additional case studies that follow Snyder's research strategy, one from phonology (syllable structure in Dutch), and two concerned with syntax (Spanish-speaking children's acquisition of noun-drop and English-speaking children's acquisition of questions with stranded prepositions). Chapter 8, ‘Conclusions: Grammatical conservatism and cross-linguistic variation’, summarizes the findings of the book.

The most contentious conclusions are reached in chapter 6, where Snyder argues for the superiority of spontaneous production data, and against tasks such as elicited production and truth-value judgement that have yielded findings that challenge grammatical conservatism. Snyder's conclusion that elicited production data do not accurately depict children's linguistic competence is reached on the basis of a representative study by Thornton & Gavruseva (Reference Thornton and Gavruseva1996, published as Gavruseva & Thornton Reference Gavruseva and Thornton2001). As Snyder reports, the study found that eleven of the twelve child subjects produced possessive long-distance questions like those in (1). In the experiment, the child subjects and a puppet, Dino, watched an appropriate story, which ended with a summary sentence and the lead-in question ‘Ask Dino whose he thinks’. The finding was that the children often failed to reproduce the possessive wh-phrase whose, which they had just heard. Instead, children split whose into two parts, with the expression that bears the wh-features (who) being moved, and the possessive marker ('s) being left behind and attached to the verb (think) in the matrix' clause.

1. (1)

   1. (a) Who do you think's cat came up on the building?

      ‘Whose cat do you think came up on the building?’

   2. (b) Who do you think's food the baboon tried?

      ‘Whose food do you think the baboon tried?’

As Snyder remarks, these data look, at first glance, like ‘a clear indication that elicited production lacks the property of grammatical conservatism’ (103). But rather than concede that grammatical conservatism is untenable, Snyder contends that ‘elicitation can yield utterance types that are ungrammatical even for the child who produces them’ (103). Snyder supports this conclusion with two observations. First, he notes that equivalent data are not attested in children's spontaneous speech. Second, he discusses data from another similar elicitation study, by Chen, Yamane & Snyder (Reference Chen, Yamane and Snyder1998), which did not elicit the same coherent pattern of non-adult responses from children. The children in the Chen et al. study produced a wide variety of errors, including ‘clear dysfluencies’. The question to raise is whether the elicited production task can be discounted on the basis of one experiment that evoked errors that would normally be analysed as performance errors. Naturally, safeguards, in the form of controls, must be in place in any experimental study. In the Thornton & Gavruseva study discussed by Snyder, a range of questions (matrix vs. long-distance, subject gap vs. object gap, bare wh-word vs. complex wh-phrase vs. whose-phrase) were elicited. In the final event, the controls that were in place led Gavruseva & Thornton (Reference Gavruseva and Thornton2001) not to interpret children's ‘erroneous’ questions in (1) as instances of the left-branch extraction found in Slavic languages, but as a reflex of computational economy. While failure to replicate may be worrisome for the researcher who reported the original data, the data do not, in themselves, suffice to disqualify an entire task outright.

To conclude the discussion, Snyder advises that use of the elicited production task be limited to verifying children's knowledge of grammatical forms that are consistent with those of adult speakers of the local language. As he states, ‘[t]he following one-way implication is still plausible: if a given form is in fact grammatically available to the child, then it can be elicited in a well-designed elicited production study’ (103). But this advice is predicated on the assumption that one already embraces grammatical conservatism. If one is agnostic, then children's consistent non-adult productions, elicited in a well-designed elicited production study, arguably are the products of their grammars, and, if so, they constitute circumstantial evidence against grammatical conservatism.

Snyder is also concerned with the asymmetry between the findings from studies of children's spontaneous production data and those of the truth-value judgements task. He remarks that

This assessment is based largely on the findings from studies of the acquisition of Principle B of the Binding Theory. Using the truth-value judgement task, many researchers have found that (some) children permit a pronoun to be anaphorically linked to a local referential antecedent, as in The _jprincess dried her _j, but not to a quantificational antecedent, as in Every _jprincess dried her _i/*j. Snyder points out that a study of first person subjects in children's spontaneous production reveals that errors with I as the subject and me as the direct object are extremely rare (Bloom, Barss, Nicol & Conway Reference Bloom, Barss, Nicol and Conway1994). Of course, findings from the truth-value judgement task must be interpreted with care, and the jury is still out on children's knowledge of Principle B, but the observation of asymmetries between production and comprehension does not itself cast doubt on the comprehension technique. For one thing, it may be more challenging for children to decide on the intended interpretation of an ambiguous utterance made by an interlocutor, as compared to deciding on the most direct way to express their own intended message, when their grammars afford alternatives. It certainly does not follow, as Snyder suggests, that ‘the child sometimes gives responses in truth-value judgement that violate her own grammar’ (108).

Both the elicited production task and the truth-value judgement task are valuable tasks precisely because they can be used to encourage children to access linguistic structures that rarely appear in their spontaneous speech. Turning Snyder's argument on its head, spontaneous production data may seriously underestimate children's linguistic competence in several ways. First, in spontaneous production, children (and adults) may tend to favour, in virtue of parsing priorities, the selection of structures that impose the fewest demands on computational resources. Obviously, young children (and adults, for that matter) will only use object gap relative clauses, passives, clefts, parasitic gap sentences, and so forth, if the experimental task encourages them to. These constructions are all rare in the transcripts of spontaneous production. One would not conclude, however, that knowledge of such structures is absent from the grammars of adults; the same goes for children. The second point is that databases of children's spontaneous productions may be too sparse, due to sampling procedures, to include robust samples of infrequent or complex linguistic structures, which may emerge suddenly and then vanish in a brief period of time.

Overall, Snyder's book is a clearly written and well-presented exposition of how to use spontaneous speech to investigate children's linguistic knowledge of parameters. By covering a number of studies in considerable detail, Snyder presents relevant aspects of experimental design and statistical method. The careful coverage in chapters 4 and 5 will be extremely useful to students and researchers, and these chapters will no doubt be added to many class reading lists. The arguments for the claim that children adhere to grammatical conservatism, as measured by their spontaneous production output, are not convincing, however. As a final comment, by taking children's performance to be the most direct measure of their linguistic competence, Snyder is in danger of blurring the competence/performance distinction, which is at the heart of generative grammar. The worry is that, by adopting Snyder's approach, we may be less able to examine the role played by UG in language development. Hence, following Snyder's research strategy may leave us less able to see whether UG or a usage-based approach to language development is closer to the truth.