Identification of negated regulation events in the literature: exploring the feature space

Background. Regulation events are of critical importance to researchers trying to understand processes in living beings. These events are naturally complex and can involve both individual molecular entities and other biomedical events. Of equal importance is the ability to capture statements that refer to regulation events that do not take place. In this paper we explore the identification of negated regulation events in the literature using a number of features. Results. We construe the problem as a classification task and apply support vector machines that use lexical, syntactic and semantic features associated with sentences that represent events. Lexical features include negation cues, part-of-speech tagging and surface distances, whereas syntactic features are engineered from constituency parse trees, the command relation between constituents and parse-tree distances. Semantic features include event subtype and participant types. On a test dataset, best precision has been achieved by combing all features, while ignoring surface-level distances resulted in best recall. Overall, the best Fmeasure was 54%. Conclusions. Syntactic features proved to be useful for improving recall, whereas semantic features proved useful for improving precision, demonstrating the potential and limits of taskspecific feature engineering to negation detection. Contrasting statements are used frequently to express negated events and many false negatives were due to not capturing those events.