Classifying Temporal Relations with Simple Features

Approaching temporal link labelling as a classification task has already been explored in several works. However, choosing the right feature vectors to build the classification model is still an open issue, especially for event-event classification, whose accuracy is still under 50%. We find that using a simple feature set results in a better performance than using more sophisticated features based on semantic role labelling and deep semantic parsing. We also investigate the impact of extracting new training instances using inverse relations and transitive closure, and gain insight into the impact of this bootstrapping methodology on classifying the full set of TempEval-3 relations.