MucLex: A German Lexicon for Surface Realisation

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4653–4657 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4653 MucLex: A German Lexicon for Surface Realisation Kira Klimt1, Daniel Braun1, Daniela Schneider2, Florian Matthes1 1Technical University of Munich, 2Allianz SE Munich, Germany {kira.klimt, daniel.braun, matthes}@tum.de daniela.schneider1@allianz.de Abstract Language resources for languages other than English are often scarce. Rule-based surface realisers need elaborate lexica in order to be able to generate correct language, especially in languages like German, which include many irregular word forms. In this paper, we present MucLex, a German lexicon for the Natural Language Generation task of surface realisation, based on the crowd-sourced online lexicon Wiktionary. MucLex contains more than 100,000 lemmata and more than 670,000 different word forms in a well-structured XML file and is available under the Creative Commons BY-SA 3.0 license. Keywords: Lexicon, Surface Realisation, Natural Language Generation 1. Introduction A rule-based surface realiser can only be as good as the lex- icon it is based on. This is especially the case for languages like German, which have a relatively high grade of irregular inflection, compared to e.g. English. The German language is characterised by various irregular word inflection forms. Adjectives, nouns, and verbs can require a change or mutation of their stem vocal in their inflected forms. Moreover, separable prefix verbs are pre- ceded by a prefix, which has to be separated from the stem and put behind the verb in most cases. Therefore, a com- prehensive German lexicon, enriching inflection rules with irregular forms, is needed for German surface realisation. Unlike lexica for other tasks, a lexicon for NLG and specif- ically for surface realisation does not need to contain in- formation about the semantics of a word, unlike e.g. Ger- maNet (Hamp and Feldweg, 1997). However, it is also not sufficient for such a lexicon to contain the different lem- mata, it also has to include the inflections of each lemma. In German, a single noun lemma can have six additional inflections: genitive singular, genitive plural, dative singu- lar, dative plural, accusative singular, and accusative plural form. For irregular verbs like “sein” (to be), this number can easily double. So far, no well-structured and openly available lexicon of this kind exists for the German language. Creating such a resource from scratch would require a huge effort. How- ever, with Wiktionary (cf. Section 3.), there exists a crowd- sourced resource which can form the base of such a lexicon, as we will show in this paper. 2. Other Lexica SimpleNLG (Gatt and Reiter, 2009) is arguably the most popular rule-based open source surface realiser. Version 4.4.8 of SimpleNLG1 comes with a default lexicon con- taining more than 6,000 lemmata and only very little in- formation about inflection. Even irregular verbs like “be” only contain information about the past participle and sim- ple participle. 1https://github.com/simplenlg/simplenlg The old German version of SimpleNLG (Bollmann, 2011) comes with what the author calls a “toy lexicon”. This lex- icon …

Content truncated to 500 words