Conceptual Schema Approach to Natural Language Database Access

Conceptual Schema Approach to Natural Language Database Access In-Su Kang, Seung-Hoon Na, Jong-Hyeok Lee Div. of Electrical and Computer Engineering Pohang University of Science and Technology (POSTECH) Advanced Information Technology Research Center (AlTrc) San 31, Hyoja-dong, Nam-gu, Pohang, 790-784, R. of KOREA {dbaisk, nsh, jhlee}@postech.ac.kr fax: +82-54-279-5699 Abstract Natural language database interfaces re- quire translation knowledge to convert user questions into formal database que- ries. Previously, translation knowledge acquisition heavily depends on human specialties such as NLP, DBMS and do- main engineering, consequently under- mining domain portability. This paper attempts to semi-automatically construct translation knowledge by introducing a physically-derived conceptual database schema, and by simplifying translation knowledge into two structures – class- referring documents and class- constraining selection restrictions. Based on these two structures, this paper pro- poses a noun translation method that em- ploys an information retrieval framework. 1 Introduction A natural language database interface (NLDBI) allows users to access database data in a natural language (Androutsopoulos et al., 1995). In a typi- cal NLDBI system, a natural language question is analyzed into an internal representation using lin- guistic knowledge that normally includes domain knowledge to reduce analysis ambiguities. The in- ternal representation is then translated into a target database query by applying mapping information (Androutsopoulos et al. 1995) that associates the analysis results with target database structures. This paper uses translation knowledge to refer to both domain knowledge and mapping information, because both are required to translate a natural lan- guage question into a database query. Previous approaches can be classified according to the extent that translation knowledge is inte- grated into linguistic knowledge. Tightly-coupled approaches hard-wire translation knowledge into linguistic knowledge in the form of semantic grammars (Hendrix et al., 1978; Waltz 1978; Templeton and Burger 1983). This approach shows a good performance for a particular domain. How- ever, in adapting to other domains, new semantic grammars should be created with a considerable effort (Allen 1995). In order to improve domain portability, many re- searchers have concentrated on isolating transla- tion knowledge from linguistic knowledge through loosely-coupled approaches. These approaches can be further classified according to the extent that question analysis is performed. Syntax-oriented systems (Ballard et al., 1984; Damerau 1985; Lee and Park 2002) analyze questions up to a syntactic level, after which translation knowledge is applied to generate a database query. Logical form systems (Warren and Pereira 1982; Grosz et al., 1987; Al- shawi et al., 1992; Androutsopoulos 1993; Klein et al., 1998) interpret a user question into a domain- independent literal meaning level. Thus, in loosely-coupled approaches, transport- ing to a new database domain does not need to change linguistic knowledge at all, only tailoring translation knowledge to new domains. Even in this case, however, translation knowledge is diffi- cult to describe. For example, syntax-oriented sys- tems have to devise conversion rules that transform parse trees into database query expressions (An- droutsopoulos et al. 1995), and logical form sys- tems should define database relations for logical predicates. In addition, creating these translation knowledge demands …

Content truncated to 500 words