I still have Time(s): Extending HeidelTime for German Texts

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4723–4728 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4723 I still have Time(s): Extending HeidelTime for German Texts Andy L¨ucking, Manuel Stoeckel, Giuseppe Abrami, Alexander Mehler Goethe University Frankfurt, Text Technology Lab Robert-Mayer-Str. 10, D-60325 Frankfurt am Main, Germany {luecking, m.stoeckel, abrami, mehler}@em.uni-frankfurt.de Abstract HeidelTime is one of the most widespread and successful tools for detecting temporal expressions in texts. Since HeidelTime’s pattern matching system is based on regular expression, it can be extended in a convenient way. We present such an extension for the German resources of HeidelTime: HeidelTimeext. The extension has been brought about by means of observing false negatives within real world texts and various time banks. The gain in coverage is 2.7 % or 8.5 %, depending on the admitted degree of potential overgeneralization. We describe the development of HeidelTimeext, its evaluation on text samples from various genres, and share some linguistic observations. HeidelTimeext can be obtained from https://github.com/texttechnologylab/heideltime. Keywords: HeidelTime, German, TimeBanks, BIOfid 1. Motivation Biodiversity literature exhibits frequent references to times. Hence, processing biodiversity texts, as is done in the context of specialized information services (Driller et al., 2020), involves recognizing temponyms, that is, temporal expressions which have – possibly aided by context infor- mation – a unique interpretation on a time line, like to- ponyms do in terms of geo-spatial entities (Kuzey et al., 2016a). To this end, HeidelTime (Kuzey et al., 2016b; Str¨otgen and Gertz, 2013) is used within the BIOfid project (www.biofid.de) to detect mentions of time-denoting expressions in mainly German biological texts from the 19th century until today – see (L¨ucking et al., 2021) for an overview. These texts, however, contain temponyms that are outside of the extension of HeidelTime’s rule system and include spelling variants of mundane tem- ponyms like Herbste ‘fall’1, duration-forming construc- tions such as in den letzten beiden Jahren ‘in the past two years’2, and set-forming constructions such as in einem zweij¨ahrigen Turnus ‘on a biennial basis’3. Based on such missing patterns of temporal expressions, among others, we extend the German component of HeidelTime (sec- tion 2.). We want to emphasize that HeidelTime is de- signed in such a way that such extensions can be im- plemented in a comparatively easy manner. The gain in coverage in assessed on various kinds of corpora (sec- tion 3.). A manual inspection of the novel instances found by HeidelTimeext brought two overgeneralizing rules to the light, however. Depending on how much potential overgeneralization is admitted, the relative coverage im- provement of HeidelTimeext is 2.7 % (excluding one source of overgeneralization) or 8.5 % (including the source of overgeneralization). The HeidelTime extension – called HeidelTimeext – is available at https://github.com/ 1E.g., https://sammlungen.ub.uni-frankfurt. de/3728500 and https://sammlungen.ub. uni-frankfurt.de/4497114 2E.g., https://sammlungen.ub.uni-frankfurt. de/3721623 3E.g., https://sammlungen.ub.uni-frankfurt. de/3748527 texttechnologylab/heideltime. 2. Procedure HeidelTime is developed to detect a certain kind of temponyms, namely temponyms that are categorized as TIMEX3 expressions (ISO, 2012). TIMEX3 expressions comprise dates (e.g., 10 …

Content truncated to 500 words