Combining Different Summarization Techniques for Legal Text

Abstract

Summarization, like other natural language processing tasks, is tackled with a range of different techniques - particularly machine learning approaches, where human intuition goes into attribute selection and the choice and tuning of the learning algorithm. Such techniques tend to apply differently in different contexts, so in this paper we describe a hybrid approach in which a number of different summarization techniques are combined in a rule-based system using manual knowledge acquisition, where human intuition, supported by data, specifies not only attributes and algorithms, but the contexts where these are best used. We apply this approach to automatic summarization of legal case reports. We show how a preliminary knowledge base, composed of only 23 rules, already outperforms competitive baselines.