Access to Multiple Underlying Systems in Janus
Abstract
The job of the back-end of any natural language interface is to translate a logical description of what the user wants (a request) into an efficient plan for fulfilling that request. Typically the request is to produce data from some underlying system; that is, the database, applications program, or other system with which the user is communicating by means of the interface. There has been a fair amount of work on the problem of natural language interfaces to single underlying systems. As computer systems become more complex, there is more opportunity for combining the strengths of more than one system in order to perform a task. For example, one might imagine combining several resources: a database for storing relational information with an applications program to perform calculations based on that information, an expert system to perform inferences, and a display system to present data in a useful way. In such an environment a seamless natural language interface can become a very effective tool, allowing the user to retrieve and manipulate information without needing to pay attention to the details of any particular resource. The back-end of such an interface, however, is necessarily more complex: not only must it be able to translate the user's request into executable code, but it must also be capable of organizing the various resources at its disposal, choosing which combination of resources to use, and supervising the transfer of data among them. We call this the multiple underlying systems (MUS) problem. This document describes one approach to the MUS problem, a MUS component implemented as part of the back end of the Janus natural language interface.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1989
- Accession Number
- ADA214585
Entities
People
- Philip Resnik
Organizations
- BBN Technologies