Disambiguating lexical markers of cause and effect using actantial structures and actant classes

The aim of this project was to develop and evaluate a method for the disambiguation of English lexical knowledge patterns that can be used to identify cause-effect relations. Recently, terminologists and computer scientists have investigated various techniques for locating informative contexts in specialized corpora automatically or semiautomatically. Some of these methods rely partly or exclusively on linguistic patterns that express a specific relationship (genericspecific, part-whole, entity-function, cause-effect, etc.). Some patterns are efficient, while others are highly ambiguous, leading to poorer results. The method we developed focuses on the actantial structures of lexical units in patterns and the semantic classes of terms linked by these patterns (with the attribution of classes relying partly on external resources). The experiment was carried out on a corpus of English medical texts. Our results show that this approach is promising, but also that other strategies need to be implemented in order to disambiguate patterns completely. 1 The authors wish to thank Sylvie Vandaele and Alain Polguère for their valuable insight and assistance, and the Social Sciences and Humanities Research Council of Canada and the Université de Montréal for their generous support. We would also like to thank the colleagues who commented on various aspects of the project as presented at the LSP conference and an OLST-RALI seminar.