Machine Recognition of Rule-Based Phonological Patterns

Abstract

This thesis is project-oriented, with the goal of the research being to create a computer program that will take a set of words in a language written in the International Phonetic Alphabet (IPA) and two sets of phonemes where one undergoes a phonological change to the other and return a phonological rule describing this data. The program was written in Python. It included several functions as well as a new class termed "Phoneme" to facilitate the storage of phonemic features. A reference was made of Phoneme objects for the program to refer to for feature information while analyzing a data set. The program's parameters were that it be able to return a rule given that the data that it was analyzing contained only phonemes found in the reference and that the change be caused by a phoneme directly adjacent to the phoneme undergoing the change. Five datasets were passed through the program and, with the exception of one, were found to return a passable phonological rule. The most notable improvement to be made is the elimination of feature redundancies within the rule.