Abstract
The principles and parameters model (Chomsky, 1981) introduced a precise and flexible approach to the uniformity and variation of language : human languages are systems governed by universal principles, but involving binary choice points, the parameters. This approach has created a precise technical language for comparative syntax, which has developed considerably on this basis.
According to this view, the acquisition of syntax is fundamentally an operation of setting parameters on the basis of experience. Here we find another manifestation of " learning by forgetting " : setting parameters means selecting certain values on the basis of experience from a set of possibilities given a priori, and thus " forgetting " the other possible options. This framework thus opens up the question of the temporality of parametric fixation : when are parameters fixed by the learner ? Synthesizing a decade of corpus work, Wexler (1996) hypothesized that parameters are set early by the learner, indeed before the production of complex structures begins. Let's take the example of a fundamental word-order parameter : the order verb - object(read book, as in French or English), or object - verb(book read, as in Japanese or Turkish). The first productions at two words, around 18 months, respect the fundamental order of the language : the French-speaking child will therefore say lire livre, and the Japanese-speaking child will say (the equivalent of) livre lire, which is expected on the basis of Wexler's hypothesis. Nevertheless, the question arises of the abstract nature of grammatical knowledge at this age. Does the one-and-a-half-year-old already have the abstract and general knowledge " my language is a VO (or OV) language ", generalizable to cases of new combinations, never heard ? Or is the child simply repeating what he or she has heard, without (in these early phases) the ability to generalize ? Tomasello's (2003) " neo-constructivist " (or item-based) approach would predict such an inability to generalize.
It is possible to test experimentally whether children are able to generalize their grammatical knowledge to new cases at an early stage : we can use possible verbs that do not exist in the language, forms of " jabberwocky " (in the sense of Lewis Carroll's famous poem) such as " daser " or " pouner " in French. Indeed, Franck et al (2013) have shown that the French-speaking child of around one and a half years interprets a sequence like " le cheval dase le chien " as a transitive sentence where the horse performs an action on the dog, whereas he assigns no interpretation to a sequence like " le cheval le chien dase ". Even though he has never heard the verb " daser ", the child knows that his language is a " Subject - verb - object " order language, and applies this knowledge to a new sentence, a result replicated in Chinese, also an S V O language. Conversely, the child exposed to a " Subject - object - verb " (or S O V) language like Hindi assigns a phrasal interpretation to noun - noun - verb sequences like the equivalent of " the horse the dog dase ", while he assigns no phrasal interpretation to sequences impossible in that language (Gavarró et al, 2015; see also Zhu et al., 2022). These results suggest that the fixation of certain fundamental parametric properties is an early operation, carried out by the child before the onset of the production of complex structures.
