Aspectual Type and Temporal Relation Classification

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 266–275, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics Aspectual Type and Temporal Relation Classification Francisco Costa Universidade de Lisboa fcosta@di.fc.ul.pt Ant´onio Branco Universidade de Lisboa Antonio.Branco@di.fc.ul.pt Abstract In this paper we investigate the relevance of aspectual type for the problem of temporal information processing, i.e. the problems of the recent TempEval challenges. For a large list of verbs, we obtain sev- eral indicators about their lexical aspect by querying the web for expressions where these verbs occur in contexts associated with specific aspectual types. We then proceed to extend existing solu- tions for the problem of temporal informa- tion processing with the information ex- tracted this way. The improved perfor- mance of the resulting models shows that (i) aspectual type can be data-mined with unsupervised methods with a level of noise that does not prevent this information from being useful and that (ii) temporal informa- tion processing can profit from information about aspectual type. 1 Introduction Extracting the temporal information present in a text is relevant to many natural language process- ing applications, including question-answering, information extraction, and even document sum- marization, as summaries may be more readable if they follow a chronological order. Recent evaluation campaigns have focused on the extraction of temporal information from writ- ten text. TempEval (Verhagen et al., 2007), in 2007, and more recently TempEval-2 (Verhagen et al., 2010), in 2010, were concerned with this problem. Additionally, they provided data that can be used to develop and evaluate systems that can automatically temporally tag natural language text. These data are annotated according to the TimeML (Pustejovsky et al., 2003) scheme. Figure 1 shows a small and slightly simpli- fied fragment of the data from TempEval, with TimeML annotations. There, event terms, such as the term referring to the event of releasing the tapes, are annotated using EVENT tags. States (such as the situations denoted by verbs like want or love) are also considered events. Temporal ex- pressions, such as today, are enclosed in TIMEX3 tags. The attribute value of time expressions holds a normalized representation of the date or time they refer to (e.g. the word today denotes the date 1998-01-14 in this example). The TLINK elements at the end describe temporal relations between events and temporal expressions. For in- stance, the event of the plane going down is anno- tated as temporally preceding the date denoted by the temporal expression today. The major tasks of these two TempEval evalu- ation challenges were about guessing the type of temporal relations, i.e. the value of the relType attribute of the TLINK elements in Figure 1, all other annotations being given. Temporal relation classification is also the most interesting problem in temporal information processing. The other relevant tasks (identifying and normalizing tem- poral expressions and events) have a longer re- search history and show better evaluation results. TempEval was organized in three tasks (TempEval-2 has four …

Content truncated to 500 words