tailieunhanh - Báo cáo khoa học: "DATR AS A LEXICAL COMPONENT FOR PATR"

One advantage of this bundled information is its reusability, which allows redundancy to be reduced. The representation of lexical information should enable us to express a further kind of generalization, namely the relations between regularity, subregularity, and irregularity. Furthermore, the representation has to be computationaUy tractable and -- possibly with the addition of"syntactic sugar" -- more or less readable for human users. | DATR AS A LEXICAL COMPONENT FOR PATR James Kỉlbury Peưa Naerger Ingrid Renz Seminar ftir Allgemeine Sprachwissenschaft Heinrich-Heine-ưniversităt Diisseldorf UniversitătssttaBe 1 D-4000 Dusseldorf 1 Federal Republic of Germany e-mail kilbury@ naerger@dd0rud81 .bimet renz@dd0rud81 .bitnet ABSTRACT The representation of lexical entries requires special means which basic PATR systems do not include. The language DATR however can be used to define an inheritance network serving as the lexical component. The integration of such a module into an existing PATR system leads to various problems which are discussed together with possible solutions in this paper. 1 MOTIVATION In the project Simulation of Lexical Acquisition SIMLEX unification is used to create new lexical entries through the monotonic accumulation of contextual grammatical information during parsing. The system which we implemented for this purpose is a variant of PATR as described in Shieber 1986 . Besides collecting the appropriate information for an unknown word . a lexeme not already specified in the given lexicon the creation of its new lexical entry is a major goal. In this context questions about the nature of lexical information the structuring and the representation of this information must be answered. The present paper is mainly concerned with the structuring and representation of information in lexical entries. 2 REPRESENTATION OF LEXICAL INFORMATION We assume that certain conditions must be met by an adequate representation of lexical information. The most important of these is that it captures linguistic generalizations which means that associated information is represented together or bundled. One advantage of this bundled information is its reusability which allows redundancy to be reduced. The representation of lexical information should enable US to express a further kind of generalization namely the relations between regularity subregularity and irregularity. Furthermore .