Abstract

Aspectual classification maps verbs to a small set of primitive categories in order to reason about time. This classification is necessary for interpreting temporal modifiers and assessing temporal relationships, and is therefore a required component for many natural language applications.

A verb's aspectual category can be predicted by co-occurrence frequencies between the verb and certain linguistic modifiers. These frequency measures, called linguistic indicators, are chosen by linguistic insights. However, linguistic indicators used in isolation are predictively incomplete, and are therefore insufficient when used individually.

In this article, we compare three supervised machine learning methods for combining multiple linguistic indicators for aspectual classification: decision trees, genetic programming, and logistic regression. A set of 14 indicators are combined for classification according to two aspectual distinctions. This approach improves the classification performance for both distinctions, as evaluated over unrestricted sets of verbs occurring across two corpora. This demonstrates the effectiveness of the linguistic indicators and provides a much-needed full-scale method for automatic aspectual classification. Moreover, the models resulting from learning reveal several linguistic insights that are relevant to aspectual classification. We also compare supervised learning methods with an unsupervised method for this task.

This content is only available as a PDF.