The generative power of human language, our ability to create an infinite variety of new words, phrases and sentences, depend critically on our ability to form implicit linguistic categories, both phonological and syntactic. For example, an adult having heard the sentence """"""""A snerg zugged"""""""" is immediately capable of generating """"""""is the snerg zugging?"""""""" without benefit of understanding the semantic content of either utterance. Creating the new utterance depends on implicitly treating """"""""snerg"""""""" and """"""""zug"""""""" as members of different lexical categories (i.e., noun and verb). The ability to group together into categories superficially distinct acoustic, lexical and phrasal tokens is key to language development in both symbolic and connectionist approaches (e.g., Elman, 1990; Guenther and Gjaja, 1996; Maye and Gerken, 1999; Pinker, 194; Valian and Coulson, 1988). Nevertheless, our understanding of the nature of these categories and how we form them is murky. Focusing, as this project will, on lexical categories, such as noun and verb, there are several questions that recur in the literature. These questions cluster along two dimensions: First, can abstract linguistic categories be induced from the input, or must the learner be born with some expectations about the category structure? Second, is the distribution of words across sentences sufficient for determining their category, or is referential information required in category formation? The proposed research locates itself at the intersection of these two dimensions. The goal of this research is to examine the limits on distributionally based category formation in an artificial language by adults and 12- to 18-month-old infants. The studies all ask the question: Under what conditions will learners generalize between training stimuli and test stimuli on the basis of distributional evidence?
Showing the most recent 10 out of 20 publications
