Linking the Nativist and Empiricist Views of Grammar Learning

Everyone seems to agree that language learning is a slow and incremental process, but what exactly is being learned? On the one hand, nativists would suggest that children are learning the peculiarities of their mother tongue and "tailoring" their innate grammars to those peculiarities. According to this view, language is interpreted according to pre-existing abstract grammatical categories for things like "noun phrase" and "past participle." On the other hand, empiricists would say that the slow and incremental learning process reflects the case-by-case learning of statistical regularities that only get turned into abstract grammatical rules - or something that guides rule-like behavior - at the end-point of learning.

So, what evidence supports the empiricist view that grammatical rules are an end-product of learning, rather than the nativist view of them as pre-existing and underlying competencies? In his 2000 article in Trends in Cognitive Sciences, Tomasello uses a a few major points to support the empiricist view:

Some observational studies that suggest children's speech errors reflect abstract grammar knowledge can actually be explained as resulting from more item-specific knowledge (e.g., the phrase allgone sticky need not indicate linguistic knowledge any more abstract than "certain kinds of things can be allgone").

Tomasello's own observations of his daughter further support the idea that grammatical knowledge is acquired piecewise, such that some verbs are used in
very limited grammatical contexts (e.g., "cut") while others are used in a variety of constructions (e.g., "draw on" "draw for" " "draw ___ on", etc). One might not expect such relatively limited uses of the word "cut" if the child possessed comprehensive knowledge of grammar and knew "cut"'s part of speech. Tomasello reviews other supporting evidence, such that the nouns with which children use the determiner "a" are completely orthogonal from those with which they use the determiner "the." Children only develop the ability to use new, recently-encountered words in novel but grammatically correct ways by the age of 3 or 4 - younger children are capable of using those words only in constructions which they have already experienced. This result has been confirmed in both production and comprehension measures, with a variety of different grammatical properties, and in a variety of different languages. Again, it's unclear why this would be the case if children possess complete knowledge of grammar all along.

As Tomasello points out, it is still possible to reconcile these data with one version of the nativist position: what if children possess an innate universal grammar all along, but require item-based incremental learning in order to "link up" the pecularities of their mother tongue with that universal grammar? Thus, the abrupt changes in linguistic competence around 4 years of age might be when the "linking up" mechanism is put into action. Tomasello rejects this because "it is very difficult to envision how an innate universal grammar could be biologically prepared ahead of time to link up its specific categories and schemas to the particular syntactic conventions of the many different languages of the world (e.g. ergative–absolutive versus nominative–accusative systems)" and because the only such nativist theory of a "linking" mechanism is at odds with empirical data.

Instead, Tomasello advocates the "word island" empiricist perspective, in which constructions are learned on an item-by-item basis before the regularities between them are extracted into more abstract grammatical rules (Tomasello uses the phrase "linguistic gestalts," but this seems like a empiricist-euphemism for rules). Notably, the extraction process occurs more quickly for those constructions and parts of speech that are more salient or more frequently encountered. Thus, the "concrete-noun" category develops early, whereas more complicated verb constructions may come only later.

The irony here is that "linking" mechanisms are central to both the nativist theory that Tomasello discounts and to his own "word island" theory. It is critical to specify what kind of mechanisms support usage generalization in a way that matches empirical data - only then will we know whether such a mechanism is likely to be grammar-specific, or whether it is at work in other domains as well.