Natural Language for Database Queries: A Laboratory Study

Technical feasibility and promise of practical use for querying databases in natural 1 anguage (for example, English) has been demonstrated by a large number of experimental systems, and the commercial availability of at 1 east one such system. Yet natural 1 anguage continues to be the most controversi al among the 1 anguage interfaces that have been proposed for direct interaction with databases. Most Natural Language Query Systems (NLGS) have focused on a certal n cl ass of users appl icati on speci al 1 sts not requi red to possess technical skills and have emphasized easy transportability to a variety of application domains. Based on these principles, and considering the limitations of state-of-the-art natural language processing, these NLGS have adopted particul ar design structures and goal s. Are these query systems meeti ng thei r design goal s? More importantly, are these the "appropriate" goals? These seem to be the major questions for which no concl usive answers have yet been given. Most experimental research in the area has addressed the first question. Fiel d studies alone are often hampered by implementation limitations, and of course, by the lack of a controlled environment. Thus, a negative answer to the first question, as is usually the case with prototype systems, makes the determination of an answer for the second question very difficult. A recently compl eted study at New York University constitutes a step toward resolving some of the issues pertaining to the use of natural language for database queries. The overall approach involves a. combi nation of expl oratory field evaluations with controlled laboratory studies to examine these issues by comparing performance between subjects using *This work was carried out as part of a joint study being conducted with the IBM Information Systems Group in White Plains, N.Y. This paper is forthcoming in the MIS Quarterly, Vol ume 7, Number 4, December 1983.