A computational model for the testing of linguistic hypotheses concerning language change

Abstract

This dissertation describes a computational program designed for the modeling of language change. It differs from most other work in computational linguistics in that its goal is the examination of scientific hypotheses rather than their efficient implementation. At the heart of the program (called FLRN) is an implementation of the lexicase (Starosta 1988) syntactic framework. In the discussion of this implementation, a formalization of regular morphology is provided for the theory. A simple parser is set up using a combination of lexicase filters and a performance (phrase structure) grammar learned through experience. The language learning portion of the program includes a minimum number of general learning abilities: classification, simplification, and analogical reasoning or generalization. Hypotheses concerning language change may be placed in either a FOCUS module (for claims about perceptual strategies, and limitations), a pragmatic module (for claims related to discourse and situation), or an all powerful GUIDE module (for claims concerning universals of language or learning). A comparison of the output of the program and observed linguistic structures provides a measure of the effectiveness of the claims in accounting for language change. The study includes examples of FLRN's use to model a simple language learner and a change from postpositions to prepositions in the Chinese languages.