Inferring Sentence-internal Temporal Relations

Inferring Sentence-internal Temporal Relations Mirella Lapata Department of Computer Science University of Sheffield Regent Court, 211 Portobello Street Sheffield S1 4DP, UK mlap@dcs.shef.ac.uk Alex Lascarides School of Informatics University of Edinburgh 2 Buccleuch Place Edinburgh EH8 9LW, UK alex@inf.ed.ac.uk Abstract In this paper we propose a data intensive ap- proach for inferring sentence-internal tempo- ral relations, which relies on a simple prob- abilistic model and assumes no manual cod- ing. We explore various combinations of fea- tures, and evaluate performance against a gold- standard corpus and human subjects perform- ing the same task. The best model achieves 70.7% accuracy in inferring the temporal rela- tion between two clauses and 97.4% accuracy in ordering them, assuming that the temporal relation is known. 1 Introduction The ability to identify and analyse temporal information is crucial for a variety of practical NLP applications such as information extraction, question answering, and sum- marisation. In multidocument summarisation, informa- tion must be extracted, potentially fused, and synthesised into a meaningful text. Knowledge about the temporal or- der of events is important for determining what content should be communicated (interpretation) but also for cor- rectly merging and presenting information (generation). In question answering one would like to find out when a particular event occurred (e.g., When did X resign?) but also to obtain information about how events relate to each other (e.g., Did X resign before Y?). Although temporal relations and their interaction with discourse relations (e.g., Parallel, Result) have re- ceived much attention in linguistics (Kamp and Reyle, 1993; Webber, 1991; Asher and Lascarides, 2003), the automatic interpretation of events and their temporal re- lations is beyond the capabilities of current open-domain NLP systems. While corpus-based methods have acceler- ated progress in other areas of NLP, they have yet to make a substantial impact on the processing of temporal infor- mation. This is partly due to the absence of readily avail- able corpora annotated with temporal information, al- though efforts are underway to develop treebanks marked with temporal relations (Katz and Arosio, 2001) and de- vise annotation schemes that are suitable for coding tem- poral relations (Ferro et al., 2000; Setzer and Gaizauskas, 2001). Absolute temporal information has received some attention (Wilson et al., 2001; Schilder and Habel, 2001; Wiebe et al., 1998) and systems have been developed for identifying and assigning referents to time expressions. Although the treatment of time expressions is an im- portant first step towards the automatic handling of tem- poral phenomena, much temporal information is not ab- solute but relative and not overtly expressed but implicit. Consider the examples in (1) taken from Katz and Arosio (2001). Native speakers can infer that John first met and then kissed the girl and that he first left the party and then walked home, even though there are no overt markers sig- nalling the temporal order of the described events. (1) a. John kissed the girl he met at a party. b. Leaving the party, John walked home. c. He remembered talking to her …

Content truncated to 500 words