Lexical Aspectual Classification

Abstract

This work is a first attempt at classification of Lexical Aspect. In this dissertation I describe eight lexical aspectual classes, each initially containing a few members. Using distributional analysis I generate 132 additional seeds, each of which was approved by at least seven out of nine judges. These seeds are in turn fed into a supervised machine learning system, trained on 136 lexical and syntactic features. I experiment on one 8-way classification task, one 3-way classification task, and ten binary classification tasks, and show that five of the eight classes are identified better than by a random baseline measure by a statistically significant margin. Finally, I analyze the relative contribution of each of four feature groups and conclude that the same features which are best in identifying phrasal aspect are also most informative for lexical aspect.