The intelligent computer-assisted instruction (CAI) system has been developed for plant operators to study emergency procedure guidelines by themselves. The system has two types of tutoring courses : teaching text knowledge by using a question and answer and teaching actual procedures and process behavior by using a simulator. The system instructs operators according to their understanding levels evaluated by the student model, which is common to both courses. The main characteristics of the system are as follows : The student model is combined with a hierarchical function model which is considered as the plant operator's mental model, and The understanding level of each node (element) is evaluated by personal history conditions calculated from both the tutoring record of the node and the understanding level of the connecting nodes. The student model and tutoring strategy function were evaluated using seven students of different understanding levels. It was confirmed that the developed system could successfully determine their understanding levels and recommend proper courses according to them.

A legal case consists of a number of temporally entangled affairs. To formalize this temporal structure, we need to write down relations between events and states with much effort, while most of them may be useless. We propose a method to generate these temporal relations, automatically from temporal features of each affair. We first distinguish static affair, processive affair, and events that culminate to the target state. Thereafter we formalize the classification by a unique temporal ontology, that consists of several intervals and a time point, and the notion of perspective given upon the ontology. According to the perspective, that is also some temporal domain in an affair, time intervals are introduced for each affair. We give default rules for aligned sequence of affairs, to relate intervals and affairs. All these temporal relations are formailized in situation thoery ; a situation that is a temporal domain introdued by a preceding affair supports a culmirative point of the succeeding affair. These temporal relations are represented in a formal typed language, and we give a logical model for that language in which we can evaluate truth values of supporting relations of situation theory. A Prolog program is developed to illustrate the feasibility.

A logic-based hypothetical reasoning system, which employs the mechanism of abduction, is a useful knowledge processing framework applicable to many practical problems including diagnosis, design, etc. However, it has a crucial problem of slow inference-speed due to its non-monotonic inference nature. Knowledge-base reformation, which has been executed so far before the real usage of the knowledge-base, is one way to overcome this problem. This paper presents a run-time knowledge reformation method, which works as a prepossessing connected to networked methods of hypothetical reasoning. This reformation method consists of three phases. The first phase is the partial compilation of extracted knowledge as relevant one to the proof of a given goal. The second phase is to reform the knowledge structure for removing its redundancy. The third phase is the transformation of the compiled knowledge into multi-level logical knowledge so as to remove further redundancy of the knowledge structure. The speed of two hypothetical reasoning methods has been improved, i. e., the total cpu-time including both the reforming process and their own reasoning computation has been remarkably reduced to a large extent particularly when original knowledge structure is complex for hypothetical reasoning task.

This paper proposes a new approach to discover knowledge hidden in a database. The approach is based on theories of concept formation, which is a type of concept learning that learns classification of data without a priori knowledge about which data is classified to which class. However, conventional theories of concept formation proposed so far have some crucial problems to be applied to a large amount of data in a database. Therefore, we introduces a new type of concept formation that can be applied to databases. Our approach is featured in the following aspects of learning ; 1) non-incremental learning, 2) learning with background knowledge about attributes of the database, and 3) learning by data both with nominal and with numerical attributes. The conventional theories of concept formation, on the other hand, employ incremental learning without any background knowledge through data with either nominal or numerical attributes. Since they assume that data comes one by one, they must incorporate, create, merge, or split concepts whenever a data comes. It is inefficient, however, when all data are given at once. So, our approach generates concepts in top down direction in the hierarchy using the background knowledge about attributes of data in the database. This becomes possible, because it assumes that all data are given at once. Both the background knowledge and the top down strategy realize efficient concept formation from a database. Employment of background knowledge also gives the capability to express the generated concepts in labels used in the knowledge. Then, concept formations including our approach employ a criterion to measure the quality of each generated concept. Hierarchical concepts are generated using this criterion. However, conventional approaches use a criterion which is defined only for nominal attributes or for numerical ones. Such restriction is too unrealistic to apply to ordinary databases both with nominal and numerical attributes. Therefore, our approach introduces a new criterion that can deal with the both attributes. Finally, the proposed approach is applied to a real estate database, and its effectiveness is shown.

This paper describes a new route planning model with three layers (two symbolic layers and one subsumption layer) for a mobile robot and the result of an experiment in computer simulation. Each layer consists of a map-generator and a route-planner, and a plan monitor coordinates three layers. These three layers are classified according to the term of generated plans in a top-down design, in which a lower planner refines a part of the route generated by the higher in sequence. The lowest planner decides a direction on the spot to get closer to a goal, while the others generate or refine to a goal based on A algorithm. Though higher layers take control of lowers by giving representation (A route on a map), each module is asynchronous and drives replanning by comparing the representation with situation. For example, a map-generator will suggest re-planning, when the environment change of its field is more than a given parameter. By giving two layers to symbolic reasoning, the load of symbolic reasoning in dynamic planning and wasteful planning cost in repairing plans can be reduced. In addition, the plan monitor takes a role of small repairing plans without driving symbolic reasoning. The model is implemented on UNIX workstations.

Term logic is a logic system where the meanings and the derivation relations of sentences can be directly described as relations of terms (which correspond to nouns of natural language). It has a unique character different from predicate logic where expressions are represented based on relations among individuals. Previously, we proposed the term logic system LVPM which can deal with compound terms corresponding to noun phrases modified by adjectives or relative clauses. In this system, a compound term is directly represented by an expression composed of terms and verb symbols. But the term-quantifiers usable in LVPM are only universal ones, corresponding to "all" or "every". Thus, the meanings of noun phrases or sentences including existential quantifiers, such as "man that has a car" or "Every man loves a woman", can't be represented in LVPM. In this paper, we introduce the existential term-quantifier into the term logic system. We construct this extended term logic as an axiomatic system, and give a set theoretical semantics for it. Then we prove the completeness of the axiomatic system under the semantics. Furthermore it is shown that the relation of equality between terms can be treated in this system.

In this paper, we deal with the database proposed by Gabbay, et al., which realizes negation as inconsistency, containing negation as failure, to focus our attention on its double negation problem. The database consists of a clausal form program as a positive declaration, and of a set as a negative declaration. A formula as a goal is expected to be transformed and added to a clausal form program and the negative declaration set. The negative declaration comes up with the idea that the inconsistency of the negative declaration with the program is expected to infer the negation of the goal. In the database, if the negative declaration is defined to be a set of formulas finitely failing in SLDNF resolutions, the negation as inconsistency is equivalent to the negation as failure. However, the double negation law does not necesserily hold. We give a class of databases in which the double negation law may hold. The database in the class consists of a general logic program as a positive declaration, and of a set, as a negative declaration, whose member is not in a stable model of the general logic program.

Daily conversation contains a variety of query expressions that can not be classified into general query expressions. For example, the request form may be used as a query, or additional information may be attached to a query expression. This paper proposes an utterance content planner (UCP) that decides the appropriate utterance content for generating a conversational query expression. As input for the UCP, we define five parameters that characterize the topic - which type of information is being requested, how unpredictable the query is, and how special is the knowledge needed to answer the query, etc. The UCP decides the utterance content according to the values of the parameters. The UCP outputs combinations of two kinds of utterance content units (UC unit), which are Kernel UC unit such as Yes/No interrogative or WH interrogative and Auxiliary UC unit which supplies additional information like "Explanation about topic" or "Situation explanation". We define six Kernel UC units and six Auxiliary UC units. In order to select the appropriate UC units, some heuristic rules and statistical models are used. The selected UC units are combined using combination rules and combination constraints. An evaluation of the UCP showed that it was able to predict real conversational query expressions quite well.

To describe knowledge in Prolog, programmers have to represent it only by terms, which might not only impose complicated programming on them, but also result in inefficient programs in some applications. To improve the above points in Prolog, many extended languages with more expressive data structures are proposed so far. Infinite trees in Prolog II and ψ terms in LOGIN are such extensions of Prolog terms. Each extended language, however, only provides specific and fixed data structures. Still, it is not well suited to various applications. This paper reports a knowledge representation language called UL/α. It provides facilities for programmers to define logical objects and unification rules, where logical objects are extensions of Prolog terms and work as "logical" data structures. Using UL/α, programmers can easily define many logical objects such as class objects, constraints, sets, strings, typed objects and function objects. In order to justify computations logically, we should have theoretical basis for newly-introduced data structures. It is, however, not so easy to construct a new theory of logic each time a programmer defines data structures for some application domains. This difficulty is reduced in the case of UL/α, since it is designed on a general axiomatic theory of extended logic programs, which provides programmers with safety conditions of newly-defined objects and their unification rules so that the programs consisting of the objects have logical meanings. Thus, UL/α is unique as a knowledge representation language in that (1) it provides programmer-defined logical objects, (2) it permits programmers to define both syntactic forms and unification rules for logical objects, and (3) it is based on an axiomatic theory which enables programmers to justify their unification rules.

In natural language processing, various kinds of information are represented as constraints. To deal with such constraints, the logic-based constraint solver based on the Prolog program transformation is one of the most potent schemes. Constraints in the scheme are written as programs of pure Prolog. Although this kind of constraint programming scheme gives us a good way to express some constraints, we cannot use negative expressions, which are usually represented by the 'negation as failure (NAF)' in Prolog programs. It is due to the fact that the Prolog program transformation cannot process programs which contain cut operators. However, it is pointed out that negative expressions are not only useful for making representations be concise, but also essential in some cases. Fortunately, we have already had the method, called 'Negation Technique', to derive a program whose meaning is the negation of the given program without NAF. Even if we use the method, we still have to deal with universally quantified 'not-equal' goals, which the output programs of Negation Technique usually include. One of naive ways to cope with the problem is to adopt the lazy evaluation, in which the evaluation of not-equal goals will be suspended until all variables will be bound. But it is not efficient and not suitable for the scheme of the Prolog program transformation. Accordingly, we propose a method to decompose a not-equal goal with universal quantifiers into the not-equal goals which hold the essential information of the original not-equal goal in terms of the Prolog program transformation. In the method, the program corresponding to a certain not-equal goal is dynamically calculated by using the unifier of two terms in the not-equal relation. With this method and 'Negation Technique', we can represent negative constraint to a great extent. Our method can also treat feature structures, which are widely used in recent natural language processing systems.