Understanding Temporal Expressions in Emails

Proceedings of the Human Language Technology Conference of the North American Chapter of the ACL, pages 136–143, New York, June 2006. c⃝2006 Association for Computational Linguistics Understanding Temporal Expressions in Emails Benjamin Han, Donna Gates and Lori Levin Language Technologies Institute Carnegie Mellon University 5000 Forbes Ave, Pittsburgh PA 15213 {benhdj|dmg|lsl}@cs.cmu.edu Abstract Recent years have seen increasing re- search on extracting and using temporal information in natural language applica- tions. However most of the works found in the literature have focused on identi- fying and understanding temporal expres- sions in newswire texts. In this paper we report our work on anchoring tempo- ral expressions in a novel genre, emails. The highly under-specified nature of these expressions fits well with our constraint- based representation of time, Time Cal- culus for Natural Language (TCNL). We have developed and evaluated a Tempo- ral Expression Anchoror (TEA), and the result shows that it performs significantly better than the baseline, and compares fa- vorably with some of the closely related work. 1 Introduction With increasing demand from ever more sophisti- cated NLP applications, interest in extracting and understanding temporal information from texts has seen much growth in recent years. Several works have addressed the problems of representing tem- poral information in natural language (Setzer, 2001; Hobbs and Pan, 2004; Saur´ı et al., 2006), extracting and/or anchoring (normalizing) temporal and event related expressions (Wiebe et al., 1998; Mani and Wilson, 2000; Schilder and Habel, 2001; Vazov, 2001; Filatova and Hovy, 2001), and discovering the ordering of events (Mani et al., 2003). Most of these works have focused on capturing temporal informa- tion contained in newswire texts, and whenever both recognition and normalization tasks of temporal ex- pressions were attempted, the latter almost always fell far behind from the former in terms of perfor- mance. In this paper we will focus on a different combi- nation of the problems: anchoring temporal expres- sions in scheduling-related emails. In our project work of building personal agents capable of schedul- ing meetings among different users1, understand- ing temporal expressions is a crucial step. We have therefore developed and evaluated our system Tem- poral Expression Anchorer (TEA) that is capable of normalizing such expressions in texts. As input TEA takes English text with temporal expressions al- ready identified, and transduces the expressions into their representations using Time Calculus for Nat- ural Language (TCNL) (Han and Kohlhase, 2003). These representations, or TCNL formulae, are then evaluated by incorporating the contextual informa- tion to give the final normalized output. TCNL has the following characteristics: (1) a human calendar (e.g., the Gregorian calendar) is explicitly modeled as a constraint system to deal with the highly under- specified nature of many temporal expressions, and it allows easy extension to include new temporal primitives; (2) a set of NL-motivated operators with a granularity-enriched type system facilitates the representation of the intensional meaning of a tem- 1Project RADAR, http://www.radar.cs.cmu.edu/external.asp 136 poral expression in a compositional way; and (3) the use of temporal references such as “focus” in …

Content truncated to 500 words