Lexical semantic typologies from bilingual corpora — A framework

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 90–94, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics Lexical semantic typologies from bilingual corpora — A framework Steffen Eger Department of Computer Science / Carnegie Mellon University 5404 Gates Hillman Complex / Pittsburgh, PA 15213, USA seger@cs.cmu.edu Abstract We present a framework, based on Sejane and Eger (2012), for inducing lexical semantic ty- pologies for groups of languages. Our frame- work rests on lexical semantic association net- works derived from encoding, via bilingual corpora, each language in a common reference language, the tertium comparationis, so that distances between languages can easily be de- termined. 1 Introduction Typologocial classifications have a long tradition in linguistics. For example, typologies based on syn- tactic categories have been proposed e.g. by Green- berg (1961), leading a.o. to ‘word order’ catego- rizations of natural languages as belonging to SVO, VSO, etc. types. Relatedly, genealogical classifica- tion systems based on phonological and morpholog- ical similarities date back at least to the compara- tists of the nineteenth centuries, among them Jacob Grimm (1785-1863), Rasmus Rask (1787-1832), and Karl Verner (1846-1896). Typological investi- gations into (lexical) semantic relations across lan- guages have, in contrast, attracted little attention. Still, some results have been established such as classifications based upon treatment of animal con- cepts and corresponding meat concepts (see the ex- cellent introduction to lexical typologies by Koch, 2001). As further exceptions, based on computa- tional principles, may be considered Mehler et al. (2011), who analyze conceptual networks derived from the Wikipedia topic classification systems for different languages; Gaume et al. (2008), who pro- pose (but do not realize, to the best of our knowl- edge) to compare distances between selected word pairs such as meat/animal, child/fruit, door/mouth across language-specific monolingual dictionaries in order to categorize the associated languages and, partly, Cooper (2008), who computes semantic dis- tances between languages based on the curvature of translation histograms in bilingual dictionaries. Recently, Sejane and Eger (2012) have outlined a novel approach to establishing semantic typologies based upon the language-specific polysemy relation of lexical units which entails language-dependent ‘lexical semantic association networks’. To illus- trate, French bœuf has two meanings, which we may gloss as ‘cow’ and ‘beef’ in English. Similarly, French langue and Spanish lingua mean both ‘lan- guage’ and ‘tongue’, whereas Chinese hu`a means both ‘language’ and ‘picture’. Sejane and Eger’s (2012) key idea is then that this language-specific polysemy can be made observable via the trans- lation relation implied e.g. by a bilingual dictio- nary. For instance, using a Chinese-English dictio- nary, one might be able to uncover the polysemy of hu`a by assessing its two English translations, as given above. More formally, one might create a link (in a network) between two English words if they have a common translation in Chinese (cf. Eger and Sejane, 2010); doing the same with a Spanish- English and French-English dictionary, one would obtain three different lexical semantic association networks, all encoded in the English language, the tertium …

Content truncated to 500 words