Automatic definition extraction using evolutionary algorithms

Abstract

Learning texts contain implicit knowledge such as definitions which provide an explanation of a particular term, or how it relates to other terms. Students assimilate new knowledge about a new topic by referring to such definitions to help them understand and conceptualise new ideas. To help the learning process, definitions could be presented in the form of a glossary which could be queried when new terms are encountered. Tutors could identify definitions present in their learning material manually, and place them in a glossary for easy reference. However, it is a laborious task to create such glossaries. In this thesis we look at automatic definition extraction from eLearning texts using machine learning techniques. We carry out two main experiments. The first uses Genetic Algorithms to learn weights of a fixed set of features used to identify definitions. These weights give an indication of the level of importance to the respective features, and when used in a definition extraction tool, they can also be used to rank definitions according to the level of confidence. In the second experiment, Genetic Programming is used on a training corpus of definitions and non-definitions, and attempts to learn rules which could be used for automatic classification of sentences in these two classes. The results achieved are promising, and we show that it is possible for a Genetic Program to automatically learn similar rules derived by a human linguistic expert and for a Genetic Algorithm to then give a weighted score to those rules so as to rank extracted definitions in order of confidence.