Machine Learning of Temporal Relations

Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL, pages 753–760, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Machine Learning of Temporal Relations Inderjeet Mani¥§, Marc Verhagen¶, Ben Wellner¥¶ Chong Min Lee§ and James Pustejovsky¶ ¥The MITRE Corporation 202 Burlington Road, Bedford, MA 01730, USA §Department of Linguistics, Georgetown University 37th and O Streets, Washington, DC 20036, USA ¶Department of Computer Science, Brandeis University 415 South St., Waltham, MA 02254, USA {imani, wellner}@mitre.org, {marc, jamesp}@cs.brandeis.edu, cml54@georgetown.edu Abstract This paper investigates a machine learn- ing approach for temporally ordering and anchoring events in natural language texts. To address data sparseness, we used temporal reasoning as an over- sampling method to dramatically expand the amount of training data, resulting in predictive accuracy on link labeling as high as 93% using a Maximum Entropy classifier on human annotated data. This method compared favorably against a se- ries of increasingly sophisticated base- lines involving expansion of rules de- rived from human intuitions. 1 Introduction The growing interest in practical NLP applica- tions such as question-answering and text sum- marization places increasing demands on the processing of temporal information. In multi- document summarization of news articles, it can be useful to know the relative order of events so as to merge and present information from multi- ple news sources correctly. In question- answering, one would like to be able to ask when an event occurs, or what events occurred prior to a particular event. A wealth of prior research by (Passoneau 1988), (Webber 1988), (Hwang and Schubert 1992), (Kamp and Reyle 1993), (Lascarides and Asher 1993), (Hitzeman et al. 1995), (Kehler 2000) and others, has explored the different knowledge sources used in inferring the temporal ordering of events, including temporal adver- bials, tense, aspect, rhetorical relations, prag- matic conventions, and background knowledge. For example, the narrative convention of events being described in the order in which they occur is followed in (1), but overridden by means of a discourse relation, Explanation in (2). (1) Max stood up. John greeted him. (2) Max fell. John pushed him. In addition to discourse relations, which often require inferences based on world knowledge, the ordering decisions humans carry out appear to involve a variety of knowledge sources, in- cluding tense and grammatical aspect (3a), lexi- cal aspect (3b), and temporal adverbials (3c): (3a) Max entered the room. He had drunk a lot of wine. (3b) Max entered the room. Mary was seated behind the desk. (3c) The company announced Tuesday that third-quarter sales had fallen. Clearly, substantial linguistic processing may be required for a system to make these infer- ences, and world knowledge is hard to make available to a domain-independent program. An important strategy in this area is of course the development of annotated corpora than can fa- cilitate the machine learning of such ordering inferences. This paper 1 investigates a machine learning approach for temporally ordering events in natu- ral language texts. In Section 2, we describe the annotation …

Content truncated to 500 words