A rudimentary analogy-based model of how syntactic constructions emerge from linguistic experience

This paper highlights theoretical issues in construction grammar and presents a simple computational language model as a preliminary solution to these issues. The specific issues dealt with are the lack of explicit definition of syntactic categories and the lack of explicit proposals regarding how constructions can be learned from linguistic experience. The proposed language model, called the "analogical path model", learns two-word syntactic patterns from sentences by finding distributional analogies between word pairs. The theoretical relevance and implications of the analogical path model are discussed at the end of the paper.