Temporal Anchoring of Events for the TimeBank Corpus

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 2195–2204, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Temporal Anchoring of Events for the TimeBank Corpus Nils Reimers†‡, Nazanin Dehghani†‡ ∗, Iryna Gurevych†‡ †Ubiquitous Knowledge Processing Lab (UKP-TUDA) Department of Computer Science, Technische Universit¨at Darmstadt ‡ Research Training Group AIPHES Technische Universit¨at Darmstadt http://www.ukp.tu-darmstadt.de Abstract Today’s extraction of temporal informa- tion for events heavily depends on an- notated temporal links. These so called TLINKs capture the relation between pairs of event mentions and time expressions. One problem is that the number of possible TLINKs grows quadratic with the num- ber of event mentions, therefore most an- notation studies concentrate on links for mentions in the same or in adjacent sen- tences. However, as our annotation study shows, this restriction results for 58% of the event mentions in a less precise infor- mation when the event took place. This paper proposes a new annotation scheme to anchor events in time. Not only is the annotation effort much lower as it scales linear with the number of events, it also gives a more precise anchoring when the events have happened as the complete document can be taken into account. Us- ing this scheme, we annotated a subset of the TimeBank Corpus and compare our re- sults to other annotation schemes. Addi- tionally, we present some baseline exper- iments to automatically anchor events in time. Our annotation scheme, the auto- mated system and the annotated corpus are publicly available.1 1 Introduction In automatic text analysis, it is often important to precisely know when an event occurred. A user ∗ Guest researcher from the School of Electrical and Computer Engineering, University of Tehran. 1https://www.ukp.tu-darmstadt.de/data /timeline-generation/temporal-anchoring -of-events/ might be interested in retrieving news articles that discuss certain events which happened in a given time period, for example articles discussing car bombings in the 1990s. The user might not only be interested in articles from that time period, but also in more recent articles that cover events from that period. Knowing when an event happened is also essential for time aware summarization, au- tomated timeline generation as well as automatic knowledge base creation. In many cases, time plays a crucial role for facts stored in a knowledge base, for example for the facts when a person was born or died. Also, some facts are only true for a certain time period, like being the president of a country. Event extraction can be used to automat- ically infer many facts for knowledge bases, how- ever, to be useful, it is crucial that the date when the event happened can precisely be extracted. The TimeBank Corpus (Pustejovsky et al., 2003) is a widely used corpus using the TimeML specifications (Saur´ı et al., 2004) for the anno- tations of event mentions and temporal expres- sions. In order to anchor events in time, the Time- Bank Corpus uses the concept of temporal links (TLINKs) that were introduced by Setzer (2001). A TLINK states the …

Content truncated to 500 words