Sentence classification experiments for legal text summarisation

Abstract

Abstract. We describe experiments in building a classifier which determines the rhetorical status of sentences. The research is part of a text summarisation project for the legal domain and we use a newly compiled and annotated corpus of judgments of the UK House of Lords. Rhetorical role classification is an initial step which provides input to the sentence selection component of the system. We report results from experiments with four classifiers from the Weka package (C4.5, naïve Bayes, Winnow and SVMs). We also report results using maximum entropy models both in a standard classification framework and in a sequence labelling framework. The SVM classifier and the maximum entropy sequence tagger yield the most promising results. 1