Developing Database Semantics as a Computational Model

DEVELOPING DATABASE SEMANTICS AS A COMPUTATIONAL MODEL* Kiyong Lee Department of Linguistics, Korea University, Seoul 136-701, KOREA Email : klee@mail.korea.ac.kr ABSTRACT Both Hausser [1] and Lee [2][3] proposed Database Semantics as a computational model for natural language semantics that makes use of a database management system, DBMS. As an extension of these efforts, this paper aims at dealing with ways of representing linguistic descriptions in table forms because all the data in a relational model of DBMS is conceived of being stored in table forms. It is claimed here that, if an algorithm can be developed for converting linguistic representations like trees, logical formulas, and attribute-value matrices into table forms, many available tools for natural language processing can be efficiently utilized as part of interface or application programs for a relational database management system, RDBMS. 1. INTRODUCTION For business transactions or academic administration, a commercial database management system (DBMS) like DB2, Oracle or Informix is widely used. Database Semantics is an attempt to adopt such a system for doing semantics for ordinary language. Since an interpretation model or background is necessary for processing linguistic information, Database Semantics can use as its model a database in a DBMS that provides both lexical meaning and world knowledge information. Furthermore, a DBMS constantly updates its database with new data, as fragments of a natural language like Korean or English are processed through a linguistic processing system LIPS. Hence, Database Semantics can consistently process even a larger fragment of discourse in natural language. (1) A Model of Database Semantics But, before developing Database Semantics as an application program into an RDBMS, the issue of representation must be resolved at least in the logical or conceptual level. Unlike ordinary semantics, Database Semantics as a computational model ultimately aims at a computational implementation. Hence, it should be able to build a successful interface between the two different levels of linguistic description and computational processing. As a result, representation comes to play an important role of bridging the * This work was partially supported by the 1999 research grant from Korea Research Foundation. Here I would like to thank anonymous referees for the PACLIC14 and all my colleagues who helped me to com- plete this paper, especially, Suk-Jin Chang, Roland Hausser, Jae-Woong Choe, Masatoshi Kawamori and my graduate assistants Jungha Hong and Seungchul Choe. —231--- and computational processing. As a result, representation comes to play an important role of bridging the gap between human and computer interactions. Linguistic descriptions are often represented in trees, logical forms, or attribute-value matrix (AVM) structures.' On the other hand, a computer system like RDBMS, which only knows a relational language like SQL, can conceptually recognize data stored in a table form only. This paper will thus focus on ways of representing linguistic information in a table form so that it can be recognized by an RDBMS. The main task of this paper is then to discuss ways of converting linguistic representations like trees, logical forms, and matrices into table forms. If …

Content truncated to 500 words