Abstract

The research reported here is a model of language acquisition that addresses how the child maps the input data into output. The project is presented as a response to certain learnability constraints on theories in developmental psycholinguistics. One of these constraints is the continuity of categories across development. Where other learnability models assume innate perception of syntactic categories , this simulation assumes that they are learned. ;The subjects were 11 mother-child dyads, who were videotaped in a laboratory playroom once a month from the time the children were 8 months old, until they were about 28 months old. The input data were obtained from mother's speech during three observations which were milestones in the children's language acquisition: the first word, vocabulary spurt , and the point at which the mean length of utterance surpassed 1.5 words. The output data were obtained from the child's speech during two observations: one was the third milestone and the second was the month immediately following. ;The child's primary data were coded and then analyzed by a computer program. The input consisted of language paired with situational representations. The categories for the situation representation assume a nominalist ontology and are, therefore, "concrete" categories. The categories derive from world, rather than linguistic, knowledge and are a subset of Jackendoff's conceptual semantic categories. The subsetting was motivated by concerns within the casual theory of meaning. ;The output of the simulation consisted of verb general and verb specific sentence frames. Both types of frames in the children's speech were subsets of the mothers' input, meaning that what the child learned was available in the mother's speech. These frames fulfill the function of canonical sentences, which, in several learnability-theoretic accounts, are presumed innate. However, they do not, themselves, account for the presence of abstract objects in the child's later syntactic projections. The verb specific argument structures are shown to be adequate to this task. Harrisian transformations on sets of sentence frames , were invoked to account for the abstract objects