12h00
UPS Pouchet, salle 159
A generative model of learning tested on a complex construction
Anat Ninio
The Hebrew University of Jerusalem
Syntactic structure is taken to be constructed by a binary Merger or Dependency operation, as assumed by current generative grammar (the Minimalist Program, Chomsky, 1995) and by Dependency Theory (Hudson, 1990). In such a grammar, longer structures are constructed from two-word long building blocks by applying in a recursive manner the Merge or Head-Dependent relation to a set of words.
We are assuming that the same generative process characterizes children’s learning. Our hypothesis is that in order to be able construct a multiword structure, children need to learn to produce its atomic constituents, which are two-word long units from which the structure is build up in a modular fashion.
We tested this learning model on controlled-subject constructions. Control constructions were defined as chained structures with a finite verb getting as a complement a nonfinite verb or predicate whose understood subject is identical to the finite verb’s syntactic subject or object. Most such sentences in adult speech involve an auxiliary or a copula as the finite verb. A generative model of learning predicts that producing combinations with the auxiliary verb omitted, namely, “telegraphic speech”, would assist learning control as the telegraphic two-word combination is also a building-block of the control sentence. In a large sample of 439 English-speaking children, mean age 2;2.30, 88% already produced a considerable number of sentences with an adult pattern of control, mostly in constructions with copulae, tense/aspect auxiliaries, modal auxiliaries and do-support. Producing telegraphic sentences significantly increased the probability of a child producing also grammatical control sentences with expressed auxiliaries. Learning the components of complex constructions helps with their mastery, even if it involves producing ungrammatical sentences.
