Evens, M.W.: Natural language interface for an expert system (2002)
0.00
0.002269176 = product of:
0.004538352 = sum of:
0.004538352 = product of:
0.009076704 = sum of:
0.009076704 = weight(_text_:a in 3719) [ClassicSimilarity], result of:
0.009076704 = score(doc=3719,freq=40.0), product of:
0.053105544 = queryWeight, product of:
1.153047 = idf(docFreq=37942, maxDocs=44218)
0.046056706 = queryNorm
0.1709182 = fieldWeight in 3719, product of:
6.3245554 = tf(freq=40.0), with freq of:
40.0 = termFreq=40.0
1.153047 = idf(docFreq=37942, maxDocs=44218)
0.0234375 = fieldNorm(doc=3719)
0.5 = coord(1/2)
0.5 = coord(1/2)
- Abstract
- A natural language interface to an expert system is a program that enables the user to communicate with the system in English or some other human language. It is designed to spare the user from learning some special programming language or command input language. Today this input is almost always typed at a keyboard or assembled with a mouse. Only a few research systems understand spoken input and produce spoken output. The precise definition of an expert system is a matter of argument. For the purposes of this article an expert system is a computer system that is capable of providing expert advice or otherwise performing at an expert level, usually in a rather narrow area. An excellent discussion of the controversy surrounding this term is given in Ref. 1. A typical expert system has at least three different kinds of interfaces. Some have four. One interface is designed to understand user queries and commands, another to generate answers and explanations. The knowledge-engineering interface provides a way for a human expert to endow the system with the expertise it needs to function. This may be a natural language interface as well. Some expert systems also produce documents, such as medical case reports or legal wills or petitions for divorce. The first paradigm expert system, Shortliffe's MYCIN system (2), provided natural language interfaces for both the end user and the engineer. The first widely used expert system that Shortliffe developed, ONCOCIN (3, 4), not only provided natural language interfaces for the end user and the knowledge engineer, it also generated the lengthy patient reports required by complex drug trials. In this article we will concentrate mainly an the natural language understanding and generation required to communicate with the end user, but we will also discuss interfaces for the knowledge engineer. We will describe some document generation techniques briefly.
The explanation facility, the ability to display its reasoning to the user, has been a key component of the expert system from the very beginning. Even though this facility may not be used very often, its presence gives users some crucial reassurance that they can explore the system's decision-making processes and themselves make a reasoned decision about whether or not to accept the advice given by the system. Elaine Rich (S) was the first to enunciate a fundamental principle of explanation generation in expert systems: It is essential that the explanation generated be derived from the actual decision-making process used by the system so that as that process changes, the explanations change with it. If the system relies an previously stored "canned explanations," then changes in the rules or the inference processes will leave the system providing explanations that are no longer valid. She argues also that the system can give deeper explanations if it operates off the internal reasoning process. From the very beginning, expert systems were thought of as vehicles for learning, particularly through the text that the system provides to explain its reasoning. When William Clancey (6) set out to produce a tutoring system based an the MYCIN system, people thought that this was going to be a quick and easy thesis, but Clancey soon realized that MYCIN's rules, written by experts for other practicing physicians, were not an appropriate way to teach diagnosis to medical students. He spent 10 years building and rebuilding the NEOMYCIN/GUIDON system as an effective tutoring system for medical students. Because of the historic connection between expert sytems and tutoring systems, we add a discussion of natural language interfaces for tutoring systems at the end of this article. Dialogue issues are becoming important as hardware systems speed up and software systems become sophisticated enough to carry an an actual dialogue with the user. This is particularly true in tutoring systems that teach languages. We will conclude with a brief mention of some systems of the future that are still in the research stage.
- Type
- a
