SAPHIR + RESEDA: A New Approach to Intelligent Data Base Access

This paper describes a transportable natural language interface to databases, augmented with a knowledge base and inference techniques. The inference mechanism, based on a classical expert system's type of approach, allows, when needed, to automatically convert an Input query into another one which 1s "semantically close". According to RESEDA' S theory, "semantically close" means that the answer to the transformed query Implies what could have been the answer to the original question. The presented system Integrates natural language processing, expert sys­ tem and knowledge representation technology to provide a cooperative database access. I. I NTRODUCTI ON Most of the existing natural language interfaces (NLIs) to databases too numerous to be cited free the user from learning the cumbersome and difficult syntax of for­ mal database query languages. Still they impose a very Strict semantic model, which corresponds to the way a real world domain Is represented 1n a particular database (DB). It is therefore difficult to produce a "valid question" without being aware of the structure of the database concerned. This issue is the main concern of the SAPHI R+RESEDA (Euzenat et al.. 1984a. 1984b) research pro­ ject at ERLI.* SAPH1R (Normier, 1984, Normier et al.. 1984) our transportable, domain independent NLl to data­ bases, is in the course of being provided with a relatively general knowledge base (KB) and associated inference mechanisms. Before going into the technical details of this new system, let us show a typical problem that SAPHI R+RESEDA will be able to handle. Suppose that you want to ask a question such as "who has ever been in the US", and that youn staff management database does not contain any Information concerning the employees' excursions. Suppose further that the DB knows instead where each one of them was born and what their degrees are, Using simple common sense knowledge, without accessing the database, we can tell the user that although we do not explicitly know who has been in the USA, we can neverthe­ less look for people born there and/or having an American degree. If the user accepts this proposal then we trans­ mit the modified query to the DBMS. • This research was supported in part by the Informatique" research contract ADI-84-068 "Agence de Figure I contains a simplified flowchart of our sys­ tem capable of such rational behaviour + * | natural language query I I I I | | GENERAL PARSFR | | | | (domain independent but language dependent )| | I I I | deep predicate representation | |+ > . » | II I I || / does this have ° | || /a representation "__ yes Query the DB | || ° in the particular / | | | D database r? / | | | c / I II | no | || | TRANSFORMATI ON MECHANI SM | | I I I in the "RESEDA style" | I || | (uses a "general" knowledge base)| \ I I I I || a "different" deep predicate representation | I I I !