AN EFFICIENT KNOWLEDGE REPRESENTA TION AND ACQUISITION TECHNIQUE FOR INTELLIGENT SYSTEMS

Abstract

Expert Systems, also called as Intelligent Systems or Knowledge Based Computer Systems, are the best known manifestations of Artificial Intelligence. It is envisaged that they will be able to solve problems in areas where computers have previously failed, or indeed never been tried. An Expert System is a computer system that encapsulates specialist knowledge about a particular domain of expertise and is capable of making intelligent decisions within that domain. Thus the development of an Expert System in any domain requires the study of that domain, as well as the expert's knowledge on a level independent of implementation details; a level referred to as the "knowledge level". Using the "knowledge level" analysis, the designer selects a mechanism for the representation of the expert's knowledge. The problem essentially is that it can be more difficult to reason correctly which Knowledge Representation should be chosen as compared to others. Moreover, this difficulty increases as the expressive power of the knowledge representational techniques increase. There is a tradeoff between the expressiveness of a representational technique and its computational tractability. This tradeoff underlies the difference among the number of representational techniques. We propose a new representaion technique which provides an integrated solution to the development, maintenance, and operation in diagnostic as well as non-diagnostic environments. This representation technique not only speeds up the inference, but also requires less storage space. It has a proven record of success in the development of PRESCRIBER -XT: An Expert Homoeopathic Consultation System; MEALMAKER: An Expert System for Diet Prescription; and even

NLSEMT: A Natural Language processing system for Symptom Extraction from Medical Text. The three Expert Systems mentioned above have been integrated using Blackboard Architecture, to form PRESCRIBER-AT: An Expert System for Homoeopathic Consultation and Diet Prescription. The Knowledge Acquisition module developed for this system gives the flexibility for further refinement or addition/modification of the knowledge in the Knowledge Base. The Automated Learning Module constructs the Heuristic Knowledge Base by assigning a possibility value to the medicine for a set of symptoms, after getting feedback from the patient. This module dynamically changes the priority of prescription of a particular medicine. While constructing the NLSEMT, we define sublanguage as a particular language used in the body of texts dealing with Homoeopathic Medicines. The Bit-coded representation technique is utilized for representing lexical information and grammar rules. The lexicon is developed using the trie structure. Thus the traversal in the dictionary and Knowledge Base becomes very fast and efficient. The Explanation Module developed for the system explores the system's behaviour by answering the questions posed by the user in an interactive environment. Based on this module and the proposed Intelligent Strategic Learning model, we have developed IHT: An Intelligent Homoeopathic Tutoring system, which teaches case-taking and how to prescribe medicines. A scheme for Single Understanding Multiple Translater (SUMT) system is proposed, in which we choose a language to develop an understanding

system and interface other languages to this system by developing translators. An attempt has been made to design SLIP: An Integrated Parser for Sanskrit Language, in which we represent the Sanskrit lexicon and syntactic/semantic grammar rules in the proposed representation technique, TAG and FTAG based on the using Paninian model. The entire developmental work has been done on IBM-PC/AT in PC Scheme, which is a dialect of LISP and C language.