Semantic Role Labelling With Chunk Sequences

We describea statisticalapproachto semantic role labellingthatemploys only shallow information. We usea Maximum Entropy learner , augmentedby EM-basedclusteringto model the fit betweena verb and its argumentcandidate. The instancesto be classifiedare sequencesof chunksthatoccurfrequentlyasargumentsin thetrainingcorpus.Ourbestmodel obtainsanF scoreof 51.70on… CONTINUE READING