Towards Semantic Validation of a Derivational Lexicon

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1728–1739, Dublin, Ireland, August 23-29 2014. Towards Semantic Validation of a Derivational Lexicon Britta D. Zeller∗ Sebastian Pad´o† Jan ˇSnajder§ ∗Heidelberg University, Institut f¨ur Computerlinguistik 69120 Heidelberg, Germany †Stuttgart University, Institut f¨ur maschinelle Sprachverarbeitung 70569 Stuttgart, Germany §University of Zagreb, Faculty of Electrical Engineering and Computing Unska 3, 10000 Zagreb, Croatia zeller@cl.uni-heidelberg.de pado@ims.uni-stuttgart.de jan.snajder@fer.hr Abstract Derivationally related lemmas like friendN – friendlyA – friendshipN are derived from a common stem. Frequently, their meanings are also systematically related. However, there are also many examples of derivationally related lemma pairs whose meanings differ substantially, e.g., objectN – objectiveN . Most broad-coverage derivational lexicons do not reflect this distinction, mixing up semantically related and unrelated word pairs. In this paper, we investigate strategies to recover the above distinction by recognizing semantically related lemma pairs, a process we call semantic validation. We make two main contributions: First, we perform a detailed data analysis on the basis of a large German derivational lexicon. It reveals two promising sources of information (distributional semantics and structural information about derivational rules), but also systematic problems with these sources. Second, we develop a classification model for the task that reflects the noisy nature of the data. It achieves an improvement of 13.6% in precision and 5.8% in F1-score over a strong majority class baseline. Our experiments confirm that both information sources contribute to semantic validation, and that they are complementary enough that the best results are obtained from a combined model. 1 Introduction Morphological processing forms the first step of virtually all linguistic processing toolchains in natural language processing (NLP) and precedes other analyses such as part of speech tagging, parsing, or named entity recognition. There are three major types of morphological processes: (a) Inflection modifies word forms according to the grammatical context; (b) derivation constructs new words from individual existing words, typically through affixation; (c) composition combines multiple words into new lexical items. Computational treatment of morphology is often restricted to normalization, such as lemmatization (covering inflection only) or stemming (covering inflection and derivation heuristically, Porter (1980)). An important reason is that English is morphologically a relatively simple language. Composition is not marked morphologically (zoo gate) and an important derivational pattern is zero derivation where the input and output terms are identical surface forms (a fish / to fish). Thus, lemmatization or stemming go a long way towards treating the aspects of English morphology relevant for NLP. The situation is different for languages with a complex morphology that calls for explicit treatment. In fact, recent years have seen a growing body of computational work in particular on derivation, which is a very productive process of word formation in Slavic languages but also in languages more closely related to English, like German (ˇStekauer and Lieber, 2005). Derivation comprises a large number of distinct patterns, many of which cross part of speech boundaries (nominalization, verbalization, adjectivization), but some of which do not (gender indicators like master …

Content truncated to 500 words