Towards the Automatic Acquisition of Lexical Data

Towards the Automatic Acquisition of Lexical Data H.Trost, E.Buchberger Department of Medical Cybernetics and Artificial University of Vienna, Austria Intelligence Abstract Creating a knowledge base has always been a bottleneck in the implementation of AI systems. This is also true for Natural I,anguage Understanding (NhU) systems, particularly [or data-driven ones. While a perfect system for automatic acquisition of all sorts of knowledge is still far from being realized, partial solutions are possib]e. This holds especial] y for ]exica] data. Nevertheless, the task is not trivial, in particular when deal :i ng with ] anguages rich in inflectional forms like German. Out system is to be used by persons with no specJ<]c linguistic knowledge, thus linguistic expertise ]]as been put int:o the system to ascertain correct clas!;ifJcation o[ words. C]assJ lication is done by means of ~I small rule ba~ed sy'stem wJ th i ex] ca] know] edge and ] anguage -speci fic houri sties. 'Phe key idea is the Jdent:i fJcatJon e~ three sorts o[ knowledge which are processed distinctly and the optLma] use el knowledge already contained in the existing lexicon. 1 . Introduction In this paper we introduce a system For the semi-automatic enlargement of a morphological lexicon. ]7t forms part of VIE-I,ANG, a Geru~an language dialogue system (Buchbetgor et all. 1982). ViE-hANG serves not only as an oh)oct but as a meta system as we]]: its knowledge has.; is to be en]argnd, and its [acJ] [Lies are used to support that process: the parser serves to anal.yze tile input to the acqu] s] t] on system, the generator ]s used to provide examples. IZn contrast Lo English the morphological analysis of German words is no trivia] task, due to two causes: - ]"Jrst:, there is a rich ]nl]ectiona] system, connJsting of about 60 d~ [ferent endings (where most endings have various different interpretations), some prefixes ('re-' [or PPP, 'zu' Coy some infinitive forms), 'umlautung', and Jrregu].ar Terms, - second, ] emma] J zation has to be complemented by interpretation, because the functional structure of a sentence Js not bound to constituent order but to the case system (expressed by inflection) instead. To build up a lexicon, one needs a classification of (;erman words on the basis of their graphemic real ization. There exist several works on ]hi s sub~ect, e.g. Bergmann (1982), Knopik (1984), Schott (]978), Schu]ze and Heinze (1982), WJ]lee (1979). For VIE-LANG we developed our own classification scheme (T~ost and Doff]nor 1986), based on Kunze and Ruediger (1968). For all those schemeR Jt takes an expert to classify new words correctly. Our acquisition system contains ]]ngulstic expertise in the fornl of different typos of rules, which allows for semi-automatic acquis(tJon of ]exica] knowledge in an interaction with a user who need not have spec] f J c i ingu] st i c knowledge. Whereas different approaches for knowledge acquisition for N],U systems have been proposed (e.g. Ba] lard (1984), Haas and HendrJx (1982)), we concentrate on the acquisition oC [exical data /[or the German language …

Content truncated to 500 words