DErivBase: Inducing and Evaluating a Derivational Morphology Resource for German

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 1201–1211, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics DERIVBASE: Inducing and Evaluating a Derivational Morphology Resource for German Britta Zeller∗ Jan Šnajder† Sebastian Padó∗ ∗Heidelberg University, Institut für Computerlinguistik 69120 Heidelberg, Germany †University of Zagreb, Faculty of Electrical Engineering and Computing Unska 3, 10000 Zagreb, Croatia {zeller, pado}@cl.uni-heidelberg.de jan.snajder@fer.hr Abstract Derivational models are still an under- researched area in computational morphol- ogy. Even for German, a rather resource- rich language, there is a lack of large- coverage derivational knowledge. This pa- per describes a rule-based framework for inducing derivational families (i.e., clus- ters of lemmas in derivational relation- ships) and its application to create a high- coverage German resource, DERIVBASE, mapping over 280k lemmas into more than 17k non-singleton clusters. We focus on the rule component and a qualitative and quan- titative evaluation. Our approach achieves up to 93% precision and 71% recall. We attribute the high precision to the fact that our rules are based on information from grammar books. 1 Introduction Morphological processing is generally recognized as an important step for many NLP tasks. Morpho- logical analyzers such as lemmatizers and part of speech (POS) taggers are commonly the first NLP tools developed for any language (Koskenniemi, 1983; Brill, 1992). They are also applied in NLP applications where little other linguistic analysis is performed, such as linguistic annotation of corpora or terminology acquisition; see Daille et al. (2002) for an informative summary. Most work on computational morphology has focused on inflectional morphology, that is, the handling of grammatically determined variation of form (Bickel and Nichols, 2001), which can be understood, overimplifying somewhat, as a normal- ization step. Derivational morphology, which is concerned with the formation of new words from existing ones, has received less attention. Exam- ples are nominalization (to understand → the un- derstanding), verbalization (the shelf → to shelve), and adjectivization (the size → sizable). Part of the reason for the relative lack of attention lies in the morphological properties of English, such as the presence of many zero derivations (the fish → to fish), the dominance of suffixation, and the rel- ative absence of stem changes in derivation. For these reasons, simple stemming algorithms (Porter, 1980) provide a cheap and accurate approximation to English derivation. Two major NLP resources deal with derivation. WordNet lists so-called “morphosemantic” rela- tions (Fellbaum et al., 2009) for English, and a number of proposals exist for extending WordNets in other languages with derivational relations (Bil- gin et al., 2004; Pala and Hlaváˇcková, 2007). Cat- Var, the “Categorial Variation Database of English” (Habash and Dorr, 2003), is a lexicon aimed specif- ically at derivation. It groups English nouns, verbs, adjectives, and adverbs into derivational equiva- lence classes or derivational families such as askV askerN askingN askingA Derivational families are commonly understood as groups of derivationally related lemmas (Daille et al., 2002; Milin et al., 2009). The lemmas in CatVar come from various open word classes, …

Content truncated to 500 words