The outcome of scientific research depends on how a phenomenon is viewed and how the questions are phrased. This applies also to the nativist view of language acquisition. As a complement to MacWhinney's discussion of nativism from the viewpoint of cognitive psychology, I would like to devote this commentary to the question of the title from the viewpoint of computational linguistics.

In his review of Crain & Thornton's (C&T) (1998) Investigations in Universal Grammar (IUG), Drozd raises many substantial concerns that call into question the correctness and usefulness of the Modularity Matching Model, and the adequacy of the experimental designs put forth to support it. My comments will focus on Modularity Matching. First, I elaborate on Drozd's suggestion that C&T's comparisons between Modularity Matching and other (potential) models do not provide convincing support for their model. Second, I suggest that Modularity Matching, while a strong claim, considers such a narrow range of phenomena in child language that its usefulness in explaining language development in general is likewise constrained.

According to the AGREEMENT/TENSE (Agr/Tns) OMISSION MODEL, children's failure to produce finite verb morphemes represents the selection of an optional infinitive form, in which tense and/or agreement is not specified. When agreement is specified, nominative case is licensed. Following the assumptions of this model, a child's utterance such as She run reflects a failure to specify tense only, given that the subject pronoun shows nominative case. We tested this assumption in two studies through the analysis of spontaneous speech samples from young typically-developing (TD) children and children with specific language impairment (SLI). In Study 1, 15 children were included (TD aged 2;1–3;11, SLI aged 4;0–6;2); in Study 2, 33 children were included (TD aged 2;5–3;11, SLI aged 3;6–6;9). We determined whether there was a relationship between the children's use of past tense -ed and their use of third person singular -s and copula is when nominative case was also used. Because nominative case was used, any failures to produce third person singular -s and copula is should be attributable to tense and not agreement. Such use should therefore be related to the children's use of -ed which presumably hinges on tense only. However, a relationship was not found in the speech of either group of children. This was true both for the children in each group who were consistent in using nominative case pronouns and for those who were not. Possible reasons for these findings are discussed.

It is not unusual for developmental psychologists to become frustrated with the theory of universal grammar (UG), whose proponents have tended to dismiss most research on children's language production and comprehension as irrelevant to explaining how human languages are acquired. This is because children's actual linguistic behaviour is presumed to reflect factors besides their grammatical competence, rendering most methods of sampling linguistic behaviour unsuitable for evaluating UG theory. This means, in practice, that UG proponents do not view performance errors as evidence against their hypothesis that grammatical knowledge is largely innate. When children perform at ceiling on a given task, this is usually taken as proof of their adultlike grammatical competence, while poor performance is dismissed as due to research design flaws or limitations in information processing capacities (e.g. working memory). Crain & Thornton (1998) attempt to eliminate what they consider to be post hoc processing accounts of children's linguistic behaviour by arguing, counter to Chomsky (1965) and many others, that children and adults share identical language processing mechanisms, and that linguistic performance directly reflects grammatical competence. Therefore, if UG principles are available from an early age, child and adult performance should be the same when tasks are properly constructed to avoid extra-linguistic demand characteristics (excepting adult–child differences predicted by parameter-setting or maturational models). It should not be surprising then that some psycholinguists, such as Drozd (target article), would find C&T to be misguided with respect to these issues, because children's linguistic behaviour surely differs from adults' in seemingly unpredictable ways.

MacWhinney argues that the logical problem of language acquisition is no longer. It was a problem back in the 1970s when the guiding premises of the field were that language structure is highly abstract, that the child's input is a poor source of information about grammar, and that the child's learning mechanisms are inadequate to bridge the gap between the input the child receives and the grammar the child acquires. Now, however, we have new linguistic theories which show that language structure is not abstract but arises from cognition, and we have evidence that input is a rich source of information, and we have reason to believe that the child has a powerful array of learning procedures at his or her disposal. The gap between input and acquisition thought to exist in the 1970s has been closed, and thus, MacWhinney argues, there is no need to attribute innate linguistic knowledge to the child.

For over 35 years, beliefs about the learnability of natural language have acted as roadblocks in the way of further development in linguistics. Rigorous and useful formal work has led to the unwarranted adoption of extreme positions which in turn have tended to stifle collaboration and polarize debate. In his article, MacWhinney suggests a way around such roadblocks – or rather many ways around. He demonstrates that we should look for multiple mechanisms to help understand how it is that children learn language from the evidence to which they are exposed.

In this commentary I will take up the topic of the Crain & Thornton's (C&T) Modularity Matching Model vs. the approach C&T dub the Competing Factors Model, arguing that there is reason to adopt a model that includes elements of both those approaches, as I understand them.

It would seem to be incomplete for any introduction to language acquisition to leave the ‘logical problem of language acquisition’ untouched – in fact, it would seem to be logical to just start the introduction with the logical problem. MacWhinney lays out a proposal here that attempts to dismantle the logical structure of the logical problem: there is no logical problem, if we consider carefully (a) the actual input that children receive from parents, and (b) the mechanisms that children use to handle possible pitfalls in extracting grammar from the input.

Drodz in his review of Crain & Thornton's Investigations in Universal Grammar (IUG) raises a number of important issues which, I think, deserve further discussion. In what follows I will concentrate on two aspects that might be the source of children's non-adultlike linguistic behaviour, namely memory and methodological weaknesses in experimental work. I will briefly comment on the use of statistics.

Drozd's critique of Crain & Thornton's (C&T) (1998) book Investigations in Universal Grammar (IUG) raises many issues concerning theory and experimental design within generative approaches to language acquisition. I focus here on one of the strongest theoretical claims of the Modularity Matching Model (MMM): continuity of processing. For reasons different to Drozd, I argue that the assumption is tenuous. Furthermore, I argue that the focus of the MMM and the methodological prescriptions contained in IUG are too narrow to capture language acquisition.

MacWhinney's list of proposed mechanisms for language acquisition seems so all-encompassing that it might appear churlish to dwell on a mechanism that's left out, but on his ultimate list of critical mechanisms there is an indeed an important (and intended) omission: innate, linguistically-specialized constraints. Such constraints are indeed mentioned, but if I understand the point of the article correctly, it is to argue that seven cognitive mechanisms, ranging from competition to cue construction, indirect negative evidence and monitoring, collectively obviate the need for any innate machinery that it specialized for language. Why not include room for innate, domain-specific constraints, too?

C&T deserve credit for their concern for attaining empirical evidence, in particular evidence from language acquisition, for linguistic theory. They also deserve credit for bringing to the attention of the linguistic community the fact that variations in methodology do indeed produce variations in child behaviours, a fact well appreciated in behavioural sciences such as psychology, but often not fully appreciated outside of it; in particular, not in linguistics, a field which is fundamentally and rightfully ambivalent about its standing as a behavioural science, considering its relation to mathematics and formal theory as well as to sociolinguistics (Lust, Flynn, Foley & Chien, 1999). C&T also deserve credit for their appreciation of the importance of what children do NOT do (constraints) in language, and attempts to empirically verify such.

MacWhinney (2004) has provided a clear and welcome synthesis of many strands of the recent research addressing the logical problem of first language acquisition from a non-nativist or non-generative grammar framework. The strand that I will comment on is the one MacWhinney calls the ‘pivot’ of his proposal, namely, that acquiring a grammar is primarily a function of learning ITEM-BASEDPATTERNS (e.g. pp. 23–29, 41, passim). These item-based patterns serve a number of dominant roles within MacWhinney's proposal, including enforcing children's conservatism (thereby reducing greatly their overgeneralizations and need to recover from the same), supporting the probabilistic nature of grammar, and enabling the competition that promotes recovery from the overgeneralizations that do occur. My concern here is primarily with the first role, that of enforcing children's conservatism, and especially with the exclusive use of language PRODUCTION as the demonstrated support of this conservatism.

The central question posed by the so called ‘logical problem of language acquisition’ is how it comes to be that children are able to GENERALIZE from a finite set of linguistic data to acquire (learn, develop, grow) a computational system (grammar) that applies to novel examples not encountered before. The difficulty of this generalization problem was first posed cogently by Gold and while Macwhinney discusses the Gold framework and the linguistic literature on this matter, it is worth noting that the Gold framework is not the only one. There are at least two important new sources of insight from computational learning theory in the decades following Gold that need to be kept in mind. First, there is the development of empirical process theory that forms the basis of any analysis of statistical learning (see summary in Vapnik, 1998). Applying this approach to language (see Niyogi, 1998 for a treatment), one concludes that the family of learnable grammars must have a finite Vapnik Chervonenkis (VC) dimension. The VC dimension is a combinatorial measure of the complexity of a class of functions. Grammars may be viewed as functions mapping sentences to their grammaticality value. In this more sophisticated sense of the VC dimension, the class of grammars must be constrained. Second, there is the development of the theory of computational complexity suggesting that while a learning algorithm might exist, it may not be efficient, i.e. run in polynomial time. These two developments come together in the influential Probably Approximately Correct (PAC) model (Valiant, 1984).

Our commentary concerns the Competing Factors Model (CFM). As Drozd says, the CFM is fictitious and without proponents. (See also McDaniel, 2001.) But we will argue that something like it is necessary to our field. Everyone recognizes multiple influences on language performance, even C&T. Most important, recognizing these factors does not require a denial of the grammar's existence or even of its dominance.

MacWhinney's paper focuses on the role of input and learning strategies in explaining how language is acquired, essentially denying the poverty of stimulus assumption on which the so-called ‘logical problem of language acquisition’ is based.

Drozd notes that Crain & Thornton's view of children's processing capacities (namely that they are the same as those of adults) is inconsistent both with the more widely held view that there are at least quantitative differences between the performance systems of children and adults, and with what is known about the acquisition of many grammatical phenomena. There is, I believe, an additional consideration that must be brought to bear here: there are highly suggestive parallels among children acquiring their first language, adults acquiring a second language, and Broca's (agrammatic) aphasics with respect to the comprehension of patterns that involve comparatively heavy demands on the processor. These parallels not only seem to confirm that children's processing resources are initially compromised, they point to the possibility of a greater role for processing in understanding grammatical development.

MacWhinney is to be commended for reopening questions about the logical problem of language acquisition in the light of new data and models. Unfortunately his discussion is marred by technical errors, false dichotomies, and inadequate attention to detail.