Problems and some solutions in customization of natural language database front ends

Abstract

This paper is concerned with some of the issues arising in the development of a domain-independent English interface to IBM SQL-based program products. The TQA system falls into the class of multilayered natural language processing systems. As a result, there is a large number of potential points at which customization to a particular database can be done. Of these, we discuss procedures that affect the reader, the lexicon, the lowest level of grammar rules, the semantic interpreter, and the output formatter. Our tests lead us to believe that the approach we are taking will make it possible for database administrators to generate robust English interfaces to particular databases without help from linguistic experts.

Conversational querying

“…Databases: In the database field, there have been efforts of providing natural language interfaces as querying front-ends, with IBM's USL [13] and the studies of [14,15] being prominent examples. The focus of USL was on using well-formed, unambiguous natural language-like constructs in order to directly query and manipulate the database.…”

“…In [14] the authors try to address the problem of natural language database front-ends by coupling the interface with the database schema. Given this observation, the user can be presented with a richer language interface, but the approach suffers in terms of applicability to different schemata: the natural language interface has to change in order to adhere to another schema.…”

ELFS: English language from SQL

“…This bottom-up approach, when compared to the traditional top-down approach, has the advantage of producing a more application-independent front-end. 7. CONCLUSION While SQL is rapidly becoming the de facto standard for data language for relational database systems, it has many shortcomings.…”

“…In particular, Damerau [7] discussed the techniques used for translating natural language queries into SQL. Few papers, however, address the problem in the opposite direction, that is, from a query language into English.…”

Recall cues in known-item retrieval

“…In other cases, a human intermediary exists to translate more verbose, natural language expressions of user need to the input expected by the system. With the advent of some level of natural language processing (Boguraev and Sparck Jones [5], Damerau [6,7], Thompson and Thompson [g] the scope of expected user input has increased in both form and content.…”