1. Children's early speech includes formulae

Let us assume that the claim about formulae is true: there are at least some formulae in early child speech, although there is no evidence that even very early child speech consists entirely of formulae. Formulae could be either rote expressions or expressions that allow some limited productivity, as in Where's the X? or See X (Braine, Reference Braine1963; Pine & Lieven, Reference Pine and Lieven1997). Does the existence of formulae bear on whether there is innate syntactic content? No. Adults use formulae, too, and, at least with respect to Determiners, parents use formulae to the same extent that two-year-olds do (Valian et al., Reference Valian, Solt and Stewart2009). If formulae are a characteristic of fully competent speakers' productions, their existence in the speech of young children is neither here nor there. Further, Valian et al., found that the least advanced two-year-olds actually use formulae involving Determiners less than adults do. Even when formulae were removed from consideration, children and adults showed the same productive flexibility in using Determiners with Nouns.

2. There are effects of frequency

Expressions that are more frequent in adult speech also tend to be more frequent expressions in the speech of the children that the adults are talking to. But frequency effects are ubiquitous in the speech of fully competent adults as well as neophyte children. Frequency effects are also inconsistent and limited. Children produce structures they have never heard.

One likely reason for frequency effects is priming. Adults as well as children show syntactic priming effects in which they reuse a syntactic structure to which they have been exposed (as first shown for adults by Bock, Reference Bock1986; Levelt & Kelter, Reference Levelt and Kelter1982; Weiner & Labov, Reference Weiner and Labov1983; for review, see Branigan, 2007). All speakers, experienced or tyro, use structures to which they have recently been exposed – if they can syntactically represent the structure. Syntactic priming is used by researchers to detect whether a speaker's grammar has abstract representations of infrequently used structures like the passive. Even three-year-olds, who do not produce and seldom hear passives, produce them when primed (Bencini & Valian, Reference Bencini and Valian2009). Thus, frequency effects, like the existence of formulae, are neither here nor there.

3. The input is ‘rich’

I put ‘rich’ in quotes because ‘richness’ is in the ear of the beholder. For organisms with a grammar, the input is incredibly rich: it contains examples of all syntactic features and categories, it reveals connections between word order and semantic roles; it at least indirectly illustrates various principles of grammar such as those that determine when it is grammatical to use a reflexive pronoun. But for organisms that will never acquire a grammar, like dogs, the input is not rich at all. The presence of rich input would only be relevant if it could compensate for lack of innate content. Infrahuman primates like bonobos can make use of some sound−meaning correspondences for lexical items but they never develop syntax.

4. Children's productivity is limited in ways that adults' productivity is not

Children's productivity is largely limited to their being able to fill in low-level formulaic slots with fillers (e.g. Pine & Lieven, Reference Pine and Lieven1997). This, I will argue, is factually incorrect with respect to Determiners, the category that has been most intensively studied. Even if children's productivity is limited, however, that is insufficient to demonstrate that empiricism is correct. Since the child's syntactic system interacts with, and is limited by, other components of nascent language and cognition, it would be surprising if children's productivity were not limited compared to adults'.

5. There is no known gene (or set of genes) for language or any component of language

The absence of a clear and direct connection between genes and language is irrelevant to judging nativist claims. Consider first the KE family, which has a mutation in just one gene, FOX2P, involving one nucleotide change (for reviews, see Marcus & Fisher, Reference Marcus and Fisher2003; Fisher & Marcus, Reference Fisher and Marcus2006). The KE family shows deficits in tests of syntax comprehension and production, as well as having difficulties distinguishing words from non-words. Indeed, the latter difference alone can distinguish affected and unaffected family members (Watkins, Dronkers & Vargha-Khadem, Reference Watkins, Dronkers and Vargha-Khadem2002). Affected individuals also have some cognitive and motor difficulties. Finally, the FOX2P gene is found in a number of species and, even in humans, is related to lung and other organ functions as well as cognitive function.

But let us imagine that, counter-factually, there is a gene that is associated only with the accurate production of the past tense. That would still not be an argument for innateness of tense. That gene might control something else that makes it impossible to acquire tense. Nor does the lack of a 1:1 correspondence between the expression of a particular gene and the expression of a particular aspect of syntax have a bearing on whether that aspect of syntax is innate.

Complex traits are not controlled by single genes (Risch, Reference Risch2000). “At least 40 loci have been associated with human height, a classic complex trait with an estimated heritability of about 80%, yet they explain only about 5% of phenotypic variance despite studies of tens of thousands of people” (Manolio et al., Reference Manolio, Collins, Cox, Goldstein, Hindorff, Hunter and Visscher2009). But when 295,000 single-nucleotide polymorphisms (SNPs) are considered simultaneously, they explain about 45% of the variance in height (Yang et al., Reference Yang, Benyamin, McEvoy, Gordon, Henders, Nyholt and Visscher2010). Maybe height is more complicated than the past tense, but the search for a gene, or a simple set of genes, that controls different aspects of syntax is not a fruitful enterprise.

6. The ability to speak and understand language is species-specific

If no special innate endowment were required for acquiring language, then any two species with identical abilities to learn and remember information and with identical repertories of cognitive concepts should be able to acquire language on the basis of the input provided. It is necessary to ensure that both species have the same concepts because some aspects of language may be inaccessible without certain concepts. For example, although syntactic tense is not equivalent to time (the present tense can be used to talk about the past, about ongoing events, about the future, and about timeless events), a concept of time may be necessary for acquiring tense. If one of the two species is nevertheless unable to acquire tense, we have an argument for innate content (if only for the notion of tense). If we are genuinely able to keep everything else constant, the remaining causal factor to explain the absence of syntax in infrahumans must be the presence (in humans) or absence (in infrahumans) of innate syntactic content.

This argument is inconclusive because it is impossible to be certain that we have kept everything else constant. Bonobos and humans, for example, could have highly similar learning abilities and similar cognition but differ in subtle ways that are relevant to language. Because arguments for innate content based on cross-species differences crucially rely on the assumption of cross-species similarity of the non-linguistic systems and of learning mechanisms, the arguments are at best suggestive. Even the smartest non-human primates are very different from humans and are unable to learn many things that humans learn. We can never be certain that we are dealing with different innate syntactic content rather than different concepts in other domains or different learning mechanisms.

Notice further that the species-specificity argument only works in one direction. Imagine that we discover a species that is indistinguishable from humans in its ability to acquire language (Abe & Watanabe, Reference Abe and Watanabe2011, for example, have claimed that passerine birds are capable of responding to hierarchical structure). Such a finding would be compatible with nativism. We would only have discovered another species that had the same innate linguistic content as humans. Thus, species-specificity is suggestive but not definitive.

7. Language acquisition has a cut-off point

Although the data are not decisive, it seems that, sometime during puberty the ability to acquire a language as a native language wanes dramatically. Let us assume for the sake of argument that the data in humans are as decisive as the data in some bird species (Brainard & Doupe, Reference Brainard and Doupe2013) and that there is a ‘critical period’, or at least a ‘sensitive period’, for language acquisition. If a child is not exposed to a language before the critical period ends, she or he will not fully acquire a natural language (as was the case for Genie, for example; Curtiss, Fromkin, Krashen, Rigler & Rigler, Reference Curtiss, Fromkin, Krashen, Rigler and Rigler1974). If a child is exposed to a second language after the critical period ends, she or he will never fully acquire that language.

Facts about critical or sensitive periods are neither here nor there for innate syntax. A critical period demonstrates that there is some innately specified schedule to learning within a particular domain. It does not demonstrate that there are innate concepts for that domain. Consider the fact that there is a critical period for bonobos as well (Rumbaugh & Savage-Rumbaugh, Reference Rumbaugh, Savage-Rumbaugh, Velichkovsky and Rumbaugh1996; Savage-Rumbaugh, Murphy, Sevcik, Brakke, Williams, & Rumbaugh, Reference Savage-Rumbaugh, Murphy, Sevcik, Brakke, Williams and Rumbaugh1993). Apes introduced to language before age 2;6 seemed to learn at least some sound−meaning pairings almost effortlessly, while those introduced to symbols after that age required specific tuition and never reached the level of those apes who were exposed early in life.

If we assume that bonobos do not have the hallmarks of language (Valian, Reference Valian, Lewis, Karimi, Harley and Farrar2009b), we can see the irrelevance of age-limited learning for nativism. Processes may be time-limited without entailing that the concepts those processes draw on are innate. Chess (Gobet & Campitelli, Reference Gobet and Campitelli2007) and go masters, to take another example, almost uniformly start playing at a very young age and starting age is important independent of amount of practice (Gobet & Camptelli, Reference Gobet and Campitelli2007). Ballet dancers cannot start dancing at age eighteen and hope to be principal dancers. All of those activities are ‘special’, but nothing about them entails any innate ideas.

Now consider the converse. If language learning continued at a high level throughout the lifespan, that would not imply the absence of innate concepts. It could simply mean that innate concepts were indefinitely available as a basis for learning. Critical periods are irrelevant to nativism.

8. Language learning is largely independent from cognition

Children with high scores on intelligence tests do not appear to learn basic syntax more quickly than children with average scores. Further, variables such as socioeconomic status (SES), which are related to performance on intelligence tests, do not appear relevant to syntax acquisition (although SES does correlate with scores on some more advanced tests of language skill, Farah et al., Reference Farah, Shera, Savage, Betancourt, Giannetta, Brodsky and Hurt2006, and with vocabulary, Hart & Risley, Reference Hart and Risley1995). One interpretation of such purported facts is that all children are born with the same abstract syntactic concepts. Children's rich stock of innate linguistic universals allows them to learn language in the same way and at the same rate despite enormous differences in their life circumstances, the type of input they receive, and even despite having various cognitive deficits. On a gross level, children seem to acquire language at about the same speed, although their ability to access their knowledge shows more variation. But even a cursory examination of different children's transcripts shows that some children rarely show errors of a particular type whereas others have frequent errors of that same type; some children seem to zip through acquisition whereas others show a more protracted course of development (e.g. Adam vs. Sarah; Brown, Reference Brown1973).

Language development in children who have different forms of retardation, such as Down syndrome or Williams syndrome, is an interesting test case. Children with Williams syndrome are profoundly retarded conceptually. Although their language is imperfect, it is relatively intact compared to their conceptual repertoire (Grant, Valian & Karmiloff-Smith, Reference Grant, Valian and Karmiloff-Smith2002; Thomas & Karmiloff-Smith, Reference Thomas and Karmiloff-Smith2005). Such data suggest that the language-learning function can to some degree operate independently of cognition. On the nativist interpretation of such data, the child has innate linguistic concepts that are spared by the cognitive deficit, though those concepts may not be accessible as early in development in individuals with Williams syndrome as in typically developing children (Thomas & Karmiloff-Smith, Reference Thomas and Karmiloff-Smith2005). Those innate concepts allow syntax acquisition to take place relatively normally.

But the independence of language from cognition is susceptible to empiricist interpretations. One possibility is that the various cognitive concepts that are necessary and sufficient for language learning are freely accessible for language but not for other cognitive domains. Although this seems unlikely, it is possible. An analogy would be a city that normally has several roads into it. Due to a disaster, all but one road is destroyed. The city is still there, intact, but only vehicles traveling down the remaining road – the language road – are able to reach it. Another possibility is that the child has special learning mechanisms for language that are not shared with general cognition. That, too, seems unlikely, but possible.

Now consider the reverse: language acquisition is correlated with intelligence test performance and socioeconomic status. We could not conclude from that that children lack innate linguistic concepts. Rather, such correlations might reflect the fact that higher general intelligence, and greater possession of factors that contribute to it, make access to innate syntax easier.

9. Language is typically localized in particular brain areas

Although this fact has not been recruited to argue that language is innate, it has been used to argue that language is special. It is obvious that localization of function does not imply the existence of innate linguistic concepts; any concept might be localized without being innate. Conversely, if language function were spread out across the brain, that would carry no implications for the presence of innate concepts.

1. Children create language

The creation of home sign systems by deaf children of hearing parents (Goldin-Meadow, Reference Goldin-Meadow2005), and the development of Nicaraguan Sign Language (NSL; Senghas, Senghas & Pyers, Reference Senghas, Senghas, Pyers, Langer, Parker and Milbrath2005), unlike the other examples I have discussed, do demonstrate syntactic content in the absence of experience. In both cases, children have not been exposed to a natural language in a natural way. In the case of home sign, there is some suggestion that at least one child has hierarchical structure and combines signs representing demonstratives with signs representing Nouns into a complex phrase (Hunsicker & Goldin-Meadow, Reference Hunsicker and Goldin-Meadow2012). Similarly, in NSL, over the years since the development of a deaf community, speakers have created devices that are similar to those used in sign languages, such as separating the gestures used to express path and manner, unlike hearing individuals, who gesture by combining path and manner.

Unlike the previous facts adduced to argue for or against nativism, the evidence from created languages is bidirectional. Most of the empirical evidence marshaled in favor of nativism is only suggestive, but the creation of language by those deprived of input is solid evidence in favor of nativism. The fact that children do create language (or in the case of home sign, a communication system with some signal hallmarks of language) in the absence of experience argues for innate content. And if the communicative devices that children created in the absence of input did not resemble natural languages, that would be evidence against nativism.

We know something about how language develops from the research on NSL. Most importantly, we know that children cannot leap from no language to a full language. Successive cohorts of NSL users developed more and more sophisticated syntactic devices (Senghas et al., Reference Senghas, Senghas, Pyers, Langer, Parker and Milbrath2005), and the development of the language may still be ongoing. The increasing richness of syntactic devices is not, however, due to exposure to a fully developed language but due to exposure to a partially developed language. Learners continue to build the language on their own but need language-like input to do so.

2. Acquisition of the Determiner category

One limitation of the investigation of home sign and developing sign languages is that the target is undefined. We have no way of specifying what the endpoint will be because there is as yet no endpoint. In home sign and NSL, the role of input cannot easily be assessed. In normal acquisition, there is an endpoint. It is possible to compare child learners with fully competent adult speakers to see how they differ. It is possible to assess the effects of input from the target. For example, we know that young children make two errors with wh-questions: they fail to supply an auxiliary when needed and they fail to invert the auxiliary when they do supply the auxiliary. Estimates of errors rates vary widely, but those two error types are recognized. Children make those errors despite perfect input: adults never produce uninverted wh-questions. It is difficult even to characterize what an error is in home sign or NSL. Similarly, the basic category system of English, for example, is clear; the basic category system of home signs and NSL is not clear.

My focus here is on the form of the empirical argument for innateness of categories, Determiners in particular, rather than the specifics of our data (Valian et al., Reference Valian, Solt and Stewart2009) or challenges to our analyses (Pine, Freudenthal, Krajewski & Gobet, Reference Pine, Freudenthal, Krajewski and Gobet2013). By studying the acquisition of syntactic categories that researchers widely agree is a feature of adult grammars, it is possible to map out a developmental trajectory for children and determine whether, at any point, the child appears to be using limited-scope formulae or whether her language behavior appears to continuously make use of abstract categories.

Work in my laboratory has examined the corpora of twenty-one English-speaking two-year-olds to establish whether children's early use of Determiners is better characterized as abstract from the outset of combinatorial speech or as based on low-level formulae (Valian et al., Reference Valian, Solt and Stewart2009). Using a variety of tests, we discovered that children used their Determiners with a variety of Nouns to the same extent that their parents did. Other than using Determiners less often than their parents did, and knowing fewer Determiners than their parents knew, their behavior with the Determiners they did have was as flexible as their parents'.

We concluded that previous failures to find similarities between child and parent in flexibility and productivity were due to small sample size. The more opportunity children had to display flexibility – by using a Noun with a Determiner multiple times – the more flexibility and productivity children displayed. The limitations were in the researchers' data, not in the children. The production data suggest that children have the category Determiner as soon as combinatorial speech begins. Even before combinatorial speech, infants use Determiners as markers for Nouns, have an equivalence class for Determiners, and do not initially mark details about Determiners (for fuller presentation, see Valian, Reference Valian and Bavin2009a, Reference Valian, Sanz, Laka and Tanenhaus2013). Children appear to continuously treat Determiners as abstract categories as early as we can measure and never look as if what they know are local details or limited scope formulae.

The argument thus has the following form. Choose an area of syntax where there is widespread agreement and where the involvement of semantics and reference is minimal: Determiners meet both criteria. Demonstrate that the child has flexible productive control in that area: assume that Valian et al. (Reference Valian, Solt and Stewart2009) have successfully met that criterion. Examine the precursors of that category for evidence of early under-specification (because only abstract knowledge is present, hence an absence of details) and early evidence of an equivalence class: the infancy work demonstrates both properties. Examine all the data for any evidence of lexically specific or details-only representations: assume there is no such evidence. If all of those criteria are met, then we have good evidence for the innateness of Determiners.

We do not have a proof, because there are no proofs in empirical work. Instead, we have an inference to the best explanation. Given the assumed facts, the best explanation is that Determiners – in their skeletal, abstract form as functional categories that take NPs as their complement – are innate. What is learned is what counts as a Determiner in the child's target language and local details about the features of Determiners. Some, in English, can be used only with plural count Nouns and mass Nouns; a can only be used with singular Nouns; the can be used with any Noun. Those are facts that children have to learn.