Incremental Identification of Inflectional Types

Incremental Identification of Inflectional Types Petra Barg and James Kilbury Heinrich-Heine-Universit¨at D¨usseldorf, Seminar f¨ur Allgemeine Sprachwissenschaft, Universit¨atsstr. 1, D-40225 D¨usseldorf, Germany E-mail:{barg,kilbury}@ling.uni-duesseldorf.de Abstract We present an approach to the incremental ac- crual of lexical information for unknown words that is constraint-based and compatible with standard unification-based grammars. Although the tech- niques are language-independent and can be applied to all kinds of information, in this paper we concen- trate on the domain of German noun inflection. We show how morphological information, especially in- flectional class, is successfully acquired using a type- based HPSG-like analysis. Furthermore, we sketch an alternative strategy which makes use of finite- state transducers. 1 Introduction Systems for natural language processing must deal adequately with “unknown” words, i.e. lexemes that either have been newly coined or else have not been included in a particular lexicon (cf. Kilbury et al. (1994)). Rather than simply regarding unknown words as noise, our system instead uses their con- text as a source for the systematic accrual of lexical information that can then be utilized. Our approach differs in significant respects from those of other investigators. It is designed for unification-based grammar formalisms with typed feature stuctures as in HPSG and is not restricted to simple morphosyntactic features. In contrast to sta- tistical approaches like that of (Brent, 1991), which often do not work incrementally and are intended for the application to large corpora, ours instead aims at a detailed grammatical analysis of individual sentences with a maximal use of their information. While systems like that of (Hahn et al., 1996) deal with the general acquisition of concepts, we are con- cerned exclusively with the acquisition of structural linguistic information. Although we deal here with German noun in- flection, in a framework close to that of (Riehe- mann, 1998) and (Koenig, 1999), the techniques are language-independent and apply to other kinds of lexical information as well, as is shown in (Walther and Barg, 1998) with respect to valency information. Thus, in contrast to (Ehrlich and Rapaport, 1997), who employ tailored algorithms for the acquisition of information about nouns and verbs, we introduce an approach that is completely general with respect to the kind of structural linguistic information ac- quired. 2 German noun-inflection classes There is a vast literature on German noun inflection represented in recent studies by (Cahill and Gazdar, 1999), (Clahsen, 1999), and (Neef, 1998). Here we summarize only essential points and ignore highly irregular and archaic inflections (cf. figure 4 below). German nouns bear gender (masculine, feminine, neuter) and are inflected for number (singular, plu- ral) and case (nominative, accusative, dative, geni- tive). With the exception of class NWN (e.g. masc Bauer ‘farmer’, with gen sg Bauern), all nonfeminine nouns build genitive singular with -s. The “regular” (cf. Clahsen (1999)) but “atypi- cal” (cf. Wunderlich (1999)) nouns of class NA (e.g. Auto ‘car’) build their plural forms in -s. The plural forms of all other (i.e. “typical”) classes must end in a so-called schwa syllable -e, -el, -er, or -en (i.e. …

Content truncated to 500 words