Enhancing Temporal Relation Classication by Features Extracted from a Syntactic Parser
Natsuda Laokulrat, Makoto Miwa, Yoshimasa Tsuruoka, Takashi Chikayama · 2013
Temporal relation classification [1] refers to the task of identifying temporal relationship between a pair of events. It is one of the keys to deep language understanding and could help advance other NLP applications such as textual entailment, document summarization, and question answering. Previous research on temporal relations used hand-coded rules. With the emerging of the annotated corpora, the Timebank corpus, machine learning approaches with different sets of features have been proposed. This paper introduces new types of features extracted from a syntactic parser for classifying temporal relations between events in newswire documents. We propose to use paths between event words in syntactic trees and path lengths as features for temporal relation classification. We first describe temporal relations and the corpora in Section 2. Section 3 describes briefly the related work. The features used for classification are presented in Section 4. Then, the evaluation and results are followed in Section 5. Lastly, we give the discussion and conclusion in Section 6 and 7.