Dialogue act classification using Inductive Logic Programming

摘要

Dialogue act classification is an important problem for natural language processing. Correctly identifying an act can help interpreting it and, in dialogue systems, responding appropriately. Unfortunately, automatically classifying acts is difficult. Inductive Logic Programming (ILP) can be used to learn rules, including decision lists, from examples. Decision lists are well suited to learn exceptional cases. Given the nature of dialogue acts, where a few generic classes contain many acts and many exceptional classes contain a few acts, this feature of ILP seems very promising for the classification task. In this paper we report on our use of ILP to learn rules to classify dialogue acts. We present our ILP system and how we adapted it to deal with the complexity of the task. We also present results that are in the same range as other classifying methods. The resulting rules are relatively few and often easier to explain than results from other methods.