HeidelTime: Tuning English and Developing Spanish Resources for TempEval-3

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 15–19, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics HeidelTime: Tuning English and Developing Spanish Resources for TempEval-3 Jannik Str¨otgen Julian Zell Michael Gertz Institute of Computer Science, Heidelberg University Im Neuenheimer Feld 348, 69120 Heidelberg, Germany {stroetgen,gertz}@uni-hd.de, j.zell@stud.uni-heidelberg.de Abstract In this paper, we describe our participation in the TempEval-3 challenge. With our multi- lingual temporal tagger HeidelTime, we ad- dressed task A, the extraction and normaliza- tion of temporal expressions for English and Spanish. Exploiting HeidelTime’s strict sep- aration between source code and language- dependent parts, we tuned HeidelTime’s ex- isting English resources and developed new Spanish resources. For both languages, we achieved the best results among all partici- pants for task A, the combination of extraction and normalization. Both the improved English and the new Spanish resources are publicly available with HeidelTime. 1 Introduction The task of temporal annotation, which is addressed in the TempEval-3 challenge, consists of three sub- tasks: (A) the extraction and normalization of tem- poral expressions, (B) event extraction, and (C) the annotation of temporal relations (UzZaman et al., 2012). This makes sub-task A, i.e., temporal tag- ging, a prerequisite for the full task of temporal an- notating documents. In addition, temporal tagging is important for many further natural language pro- cessing and understanding tasks, and can also be ex- ploited for search and exploration scenarios in infor- mation retrieval (Alonso et al., 2011). In the context of the TempEval-2 challenge (Ver- hagen et al., 2010), we developed our temporal tag- ger HeidelTime (Str¨otgen and Gertz, 2010), which achieved the best results for the extraction and nor- malization of temporal expressions for English doc- uments. For our work on multilingual information retrieval (e.g., Str¨otgen et al. (2011)), we extended HeidelTime with a focus on supporting the simple integration of further languages (Str¨otgen and Gertz, 2012a). For TempEval-3, we now tuned Heidel- Time’s English resources and developed new Span- ish resources to address both languages that are part of TempEval-3. As the evaluation results demon- strate, HeidelTime outperforms the systems of all other participants for the full task of temporal tag- ging by achieving high quality results for the extrac- tion and normalization for English and Spanish. The remainder of the paper is structured as fol- lows: We explain HeidelTime’s system architecture in Section 2. Section 3 covers the tuning of Heidel- Time’s English and the development of the Spanish resources. Finally, we discuss the evaluation results in Section 4, and conclude the paper in Section 5. 2 HeidelTime HeidelTime is a multilingual, cross-domain tempo- ral tagger. So far, it can process English, Ger- man, and Dutch text. In previous work, we an- alyzed domain-dependent challenges and demon- strated that domain-sensitive strategies for normal- izing temporal expressions result in significant nor- malization improvements when switching between news- and narrative-style documents (Str¨otgen and Gertz, 2012b). Although TempEval-3 only ad- dresses news documents, the tuned …

Content truncated to 500 words