Multiple Underlying Systems: Translating User Requests into Programs to Produce Answers

Multiple Underlying Systems: Translating User Requests into Programs to Produce Answers Robert J. Bobrow, Philip Resnik, Ralph M. Weischedel BBN Systems and Technologies Corporation 10 Moulton Street Cambridge, MA 02138 ABSTRACT A user may typically need to combine the strengths of more than one system in order to perform a task. In this paper, we describe a component of the Janus natural language interface that translates inten- sional logic expressions representing the meaning of a request into executable code for each application program, chooses which combination of application systems to use, and designs the transfer of data among them in order to provide an answer. The com- plete Janus natural language system has been ported to two large command and control decision support aids. 1. Introduction The norm in the next generation of user en- vironments will be distributed, networked applications. Many problems will be solvable only by use of a corn- bination of applications. If natural language technol- ogy is to be applicable in such environments, we must continue to enable the user to talk to computers about his/her problem, not about which application(s) to use. Most current natural language (NL) systems, whether accepting spoken or typed input, are designed to interface to a single homogeneous under- lying system; they have a component geared to producing code for that single class of application sys- tems, such as a single relational database[12]. Providing an English interface to the user's data base, a separate English interface to the same user's plan- ning system, and a third interface to a simulation package, for instance, will neither be attractive nor cost-effective. By contrast, a seamless, multi-modal, natural language interface will make use of a heterogeneous environment feasible and, ff done well, transparent; this can be accomplished by enabling the user to state information needs without specifying how to decompose those needs into a program calling the various underlying systems required to meet those needs. We believe users who see that NL technology does insulate them from the underlying impleman- tation idiosyncrasies of one application will expect that our models of language and understanding will extend to simultaneous access of several applications. Consider an example. In DARPA's Fleet Com- mand Center Battle Management Program (FCCBMP), several applications (call them underlying systems) are involved, including a relational data base (IDB), two expert systems (CASES and FRESH), and a decision support system (OSGP). The hardware platforms include workstations, conventional time- sharing machines, and parallel mainframes. Suppose the user asks Which of those submarines has the greatest probability of locating A within 10 hours? Answering that question involves subproblems from several underlying applications: the display facility, to determine what "those submarines" refers to; FRESH, to calculate how long each submarine would take to get to A's vicinity; CASES, for an intensive, paral- lelizable numerical calculation estimating the probabilities; and the display facility again, to present the response. While acoustic and linguistic processing can determine what the user wants, the problem of trans- lating that into an effective …

Content truncated to 500 words