A Baseline Temporal Tagger for all Languages

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 541–547, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. A Baseline Temporal Tagger for all Languages Jannik Str¨otgen and Michael Gertz Institute of Computer Science, Heidelberg University Im Neuenheimer Feld 348, 69120 Heidelberg, Germany {stroetgen,gertz}@informatik.uni-heidelberg.de Abstract Temporal taggers are usually developed for a certain language. Besides English, only few languages have been addressed, and only the temporal tagger HeidelTime covers several languages. While this tool was manually extended to these lan- guages, there have been earlier approaches for automatic extensions to a single tar- get language. In this paper, we present an approach to extend HeidelTime to all lan- guages in the world. Our evaluation shows promising results, in particular consider- ing that our approach neither requires lan- guage skills nor training data, but results in a baseline tagger for 200+ languages. 1 Introduction & Related Work Temporal tagging (the extraction and normaliza- tion of temporal expressions) is a crucial task in NLP, and many applications can benefit from tem- poral information, e.g., in the context of question answering and information retrieval (Campos et al., 2014). Thus, there has been a lot of research during the last years addressing temporal infor- mation extraction, as reflected by manually an- notated corpora, approaches to temporal tagging, and research challenges such as the TempEval se- ries (Verhagen et al., 2010; UzZaman et al., 2013). While, in the meantime, the range of anno- tated corpora covers several languages different from English, e.g., the French, Portuguese, Ital- ian, and Romanian TimeBank corpora (Bittar et al., 2011; Costa and Branco, 2012; Caselli et al., 2011; Forascu and Tufis, 2012), most approaches to temporal tagging focus on processing English text, e.g., DANTE (Mazur and Dale, 2009; Mazur, 2012) and SUTime (Chang and Manning, 2012). Of course, exceptions exist, e.g., TipSem for En- glish and Spanish (Llorens et al., 2010), and Hei- delTime (Str¨otgen and Gertz, 2013) even covers 13 languages. However, it required a lot of man- ual effort to extend HeidelTime, and researchers of different institutes and countries have been in- volved for German (Str¨otgen and Gertz, 2011), Dutch (van de Camp and Christiansen, 2012), Spanish (Str¨otgen et al., 2013), French (Moriceau and Tannier, 2014), Croatian (Skukan et al., 2014), Vietnamese and Arabic (Str¨otgen et al., 2014), Italian (Manfredi et al., 2014), Chinese (Li et al., 2014), Russian, Estonian, and Portuguese. Obviously, it is possible to manually extend HeidelTime and thus a temporal tagger in general. However, each language is typically addressed separately, and the extension process is time- and labor-intensive, so that a method to automate this process is desirable. While there have been earlier approaches for automatic extensions of temporal taggers to further languages (Saquete et al., 2004; Negri et al., 2006; Spreyer and Frank, 2008), these were limited to a few languages and the results were considered less successful, in particular for the normalization subtask. In contrast, Angeli and Uszkoreit (2013) presented an approach to language-independent parsing of …

Content truncated to 500 words