Syntactic bootstrapping as a procedure for verb learning

It is often supposed that word-meanings are acquired by observing the real-world contingencies for their use. Thus rabbit or run or rhubarb would mean what they do owing to the fact that these words are uttered in the presence of rabbits, running, and rhubarb. In this dissertation, it is argued that such observational information is insufficient to account fully for the child's vocabulary acquisition. To redress this insufficiency, it is proposed that the child uses a second source of information: the varying syntactic structures in which the meaningfully varying words occur. This kind of acquisition of meaning from form is called "syntactic bootstrapping."

To document the position, studies were designed to elicit children's and adult's interpretations of certain common English verbs (Experiments 1 and 2) and novel, possible English words (Experiments 3 and 4). In Experiments 1 and 2, the subjects are asked to act out (with toys) grammatical and ungrammatical sentences containing familiar verbs. It is shown that children as young as thirty months of age reliably take the novel ("ungrammatical") structures to imply novel semantic properties of those verbs. The new forms lead regularly to new interpretations of old items. This relationship is sensitive to three factors: the age of the subject, the perceived grammaticality of the stimulus sentence, and the structure (transitive vs. intransitive) of the stimulus sentence.

In Experiments 3 and 4 the preferential looking paradigm of Golinkoff and Hirsh-Pasek (1981) was used to present subjects with pairs of novel actions: one causal and one noncausal, accompanied by novel verbs in either transitive or intransitive sentences. It was found that the children (MA = 25 months) selected the action referent of the novel verb according to the sentence type in which it was presented. This effect was obtained for one transitive and two intransitive structures.

It is concluded that vocabulary learning is successful partly because learners presuppose that there will be stable correlations between word meanings and syntactic formats. They therefore use knowledge of syntactic structures derived from linguistic observation to make conjectures about word meanings. The implications of this theory for language learning and language representation are discussed.