Hypothesis testing processes have been studied for several decades in the psychology of science. Previous studies suggent that a strategy to collect data which conforms to one's current hypothesis (i.e., positive strategy) is useful for successful hypothesis testing in many situations that people face in scientific learning. It was also suggested that entertaining alternative hypotheses is useful for successful hypothesis testing. In order to inventigate the effectiveness of these strategies, we used a production system model that can solve the Wason's 2-4-6 task, and executed many simulations in which the parameters of the model were systematically varied. It is assumed that this approach compensates for the limitations of psychological experimentation. The summary of the simulations is as follows : The positive test strategy works well when the terget concept is specific. On the other hand, it prevents subjects from finding a correct target when the target concept is general. Diagnostic test strategy (i.e., entertaining alternative hypotheses and producing instances which distinguish the competing hypotheses) is always effective for finding a correct concept.

Cooperative use of diagrams and natural language text is quite effective for sophisticated and flexible information processing. Therefore, it is important to investigate the integration of pattern information and natural language information (especially in multi-media). For this purpose, we propose a new method for diagram understanding by using the integration of those information. We examined our method and the correct recognition was 95%. Furthermore, we combine this diagram information and meta-knowledge which was extracted from explanation of pictorial book, and obtained more precise meta-knowledge.

In this paper, we discuss an effective method to retrieve pieces of information from a different viewpoint. The main purpose of retrieving such information is to support human divergent thinking. Brainstorming is a well-known method for supporting human divergent thinking and experience tells us that it is effective that an outsider attends a brainstorming session being held by only experts of the same domain. The goal of our research is to construct a virtual outsider system. The method of information retrieval discussed in this paper is one of the important elements to realize the system. We proposed "An outsider model" to retrieve such information and constructed a prototype system based on the model. Moreover, we discussed what kind of information effectively stimulates human divergent thinking and proposed an subjective evaluation method whether a piece of information is effective for supporting human divergent thinking or not. Using the prototype system, we conducted subjective experiments and evaluated results based on the evaluation method. By these experiments, we got the following results : Using the outsider model, (1) medium relevant as well as medium unexpected information can effectively be gotten, (2) Furthermore, there is possibility to obtain highly relevant as well as highly unexpected information, too. Such information is expected to effectively stimulate human divergent thinking and is hard to be gotten without any mechanism like the outsider model to obtain hidden relevance along with relevance.

In this paper, we propose Motion Sketch as a representation to show the interaction between a vision-based learning agent and its environment. By obtaining this representation, a one-eyed mobile robot can learn several behaviors such as obstacle avoidance and target pursuit. The Motion Sketch represents tight coupling between visual and motor behaviors. Visual behaviors consist of optical flow detection and visual tracking by a real-time visual tracking routines. Motor behaviors consist of sets of uninterpreted action sequences. It is obtained by statistical technique and Q-learning, a most widely used reinforcement learning method based on the visual motion cues. Finally, the experimental results of real robot implementation with real-time visual tracking routines are shown.

Many processes concurrently behave and timing constraints are strict in communication protocols. Formal specification and timing verification of communication protocols are important. As concerns formal specification and verification of communication protocol, SDL and LOTOS, Estelle have been standardized and studied. But timing conditions can not be specified and verified in these methods. In this paper, I propose formal specification and timing verification method. I have developed CASE tool supported for this method. The main feature of this method and CASE tool is as follows. (1) formal specification method: Concurrent behavior between processes is specified by process algebra, and internal behavior in processes is specified by automaton. System specification is automatically generated from both concurrent behavior specification and internal behavior specification. (2) formal timing verification method: Verification property specification is specified by timed automaton, and timing verification is realized by language inclusion algorithm and geometric region. I show that this method and CASE tool are effective by the example of automobile control system.

A dynamic model that permits human beings to respond to changes in their environments by modifying their opinions is proposed. The goal is to reproduce the decision making ability of humans ; The model can be structured to express the personality characteristics of adaptability and hysteresis. Adaptability indicates the ability to readily change opinion while hysteresis suggests person resistance to change. Simulations confirm : (1) a person weighted toward Public Self-Consciousness responds to changes in his environment by readily changing his opinions, on the other hand, weighting for Private Self-Consciousness stiffens his resistance to changing his opinion ; and (2) the strength with which an opinion is held becomes most unstable immediately before it is altered.

A distributed constraint satisfaction problem (Distributed CSP) is a CSP in which variables and constraints are distributed among multiple automated agents, and various application problems in Distributed Artificial Intelligence can be formalized as Distributed CSPs. We develop a new algorithm for solving Distributed CSPs called asynchronous weak-commitment search, which is inspired by the weak-commitment search algorithm for solving CSPs. This algorithm can revise a bad decision without an exhaustive search by changing the priority order of agents dynamically. Furthermore, agents can act asynchronously and concurrently based on their local knowledge without any global control, while guaranteeing the completeness of the algorithm. The experimental results on various example problems show that this algorithm is by far more efficient than the asynchronous backtracking algorithm, in which the priority order is static. The priority order represents a hierarchy of agent authority, i.e., the priority of decision making. Therefore, these results imply that a flexible agent Organization, in which the hierarchical order is changed dynamically, actually performs better than an Organization in which the hierarchical order is static and rigid.

Discourse processing has been widely recognized as a key technology for improving the accuracy of text analysis, but it has also been considered a high cost procedure that requires an enormous amount of background knowledge and deep inference mechanisms. However, without constructing a precise model of the discourse through deep semantic analysis, rich information for resolving ambiguities in sentence analysis, including various discourse-dependent problems, can be obtained by analyzing a simple set of parsed trees of each sentence in a text. For example, if it is assumed that morphologically identical words within a discourse have the same word sense and modify or are modified by similar words, the results of word sense and attachment disambiguation applied in one sentence can be shared with all other morphologically identical words within the discourse. Besides facilitating information on word sense and attachment disambiguation, processing a whole text at one time makes it possible to refer to other information in the discourse, such as word frequency and the position of each word, which can be used for resolving pronoun reference and the focus of focusing subjuncts, such as also and only, as well as for adding supplementary phrases in some elliptical sentences. We have developed a method of sentence analysis based on a simple discourse model that improves the accuracy of a natural language processing system, in particular, a machine translation system. Our framework is highly practical, since it does not require any knowledge resources that have been specially hand-coded for discourse processing, or a deep inference mechanism; instead, it uses syntactic information on all the other words in the discourse, such as modifiee-modifier relationships and position in the text. Moreover, our approach is fundamentally different from previous approaches to discourse processing, in that it does not consider any discourse structure and is aimed at improving the accuracy of natural language processing rather than obtaining a perfect analysis. In this paper, we describe our robust discourse processing method, focusing on its effect in a machine translation system.

The nearest neighbor algorithm has been one of the most basic class of techniques in the field of Pattern classification. It is also the most basic and important technique in the fields such as case-based reasoning (CBR), memory-based reasoning (MBR), and example-based natural language Processing (EBNLP). In the nearest neighbor algorithm, the computational cost of example retrieval is one of the most important issues, especially when the number of examples in the database becomes large. In the field of pattern classification, there exist several techniques for reducing the computational cost of the nearest neighbor algorithm, while in other fields such as CBR, MBR, and EBNLP, there has been no technique except for the one using massively parallel computers. This paper proposes a novel technique for optimizing the nearest neighbor algorithm, especially for the use in CBR, MBR, and EBNLP. The basic idea is to use similarity calculation template, a data structure that enumerates all the possible patterns of calculating similarity between two examples. In the method, the nearest neighbor retrieval process is optimized by generating retrieval queries from an input and similarity calculation templates in a certain order. Its major advantages are as follows : 1) it can be implemented without any expensive hardwave such as massively parallel computers, 2)it is easy to add new examples to the example database. Experimental results show that nearly constant time nearest neighbor retrieval is achieved, independently of the number of examples in the database.

This paper describes a new method of discovering theorems in a plane geometry domain. In order to discover useful theorems from a figure, the following subjects should be discussed : the contents of acquired data by observing the figure, the techniques for avoiding combinatorial explosion of expressions, and the criteria for choosing useful theorems from generated expressions. Our discovery method is based on the sides in a given figure. All the geometric relations among sides are observed from a figure. Deduced expression which shows the relation of closely located sides in the figure are regarded as a useful theorem. It is because a human often considers such an expression to be an important theorem. Observed geometric relations among sides are used for both choosing useful theorems and avoiding combinatorial explosion of generated expressions. Although most of the methods of previous discovery systems such as AM, KEKADA and IDS require much initial knowledge for evaluating the usefulness of generated knowledge, our method needs little initial knowledge about plane geometry. Our discovery method has been implemented in a system called PLANET, a discovery system for plane geometry theorems. PLANET has succeeded in discovering many useful geometry theorems and trigonometric formulas, including Menelaus' theorem, Ceva's theorem, and addition theorems of trigonometric functions, through trial and error.