「AIマップ-AI研究のあり方」(Vol.7, No.5, pp.796-809に掲載)についていろいろな方から御質問やコメントをいただいた.筆者は「AIマップ」という名のもとに,現在行われている多くの研究がそのなかに要素研究として位置づけられるようなAI技術の体系(の一つ)を提案することを試みたわけであるが,その意図は,現在のAI研究のあり方に多少危惧ないしは疑問を持っており,筆者なりにその対案を示すことであった.今日,AI研究の多くは要素研究として行われ,傾向としてはますます理論的な側面が強くなってきた反面,これらの研究成果は必ずしも直接的に利用されにくくなっている.応用分野の人々からのAIへの関心は大きく,応用研究の必要性がいわれているが,応用研究というとともすると「特定の」応用問題として受け取られ,論文にはなりにくいと思われている.その結果,学会誌が難しいといわれているのが現状である.実際には現在行われている要素研究を現実の応用に結びつけるためには,この間に別の「普遍的な」方法論が必要であり,それを実現する情報技術が存在すべきである.AI研究者にそのことが気づかれておらず,したがってAI研究がバランス良く行われていないように見える.「AIマップ-AI研究のあり方」(以下,本文中では本稿と表記)では,体系化を通してそのことを述べたかったのであるが,表現が不十分であり,紙面の制約もあってこの意図が必ずしも理解されなかったかもしれない.いただいた質問も,体系化を議論するうえで必要最少限に留めた個々の要素研究に対する筆者の考え方に対するものが多かった.もちろん,体系化をにらんだうえでの要素研究のあり方に関するものもあり,このような体系化の研究をいかに進めるかという研究の進め方に関する質問もあった.この意味では,問題点を全般的にカバーしているといってもよいかもしれない.それらのなかで,重複を避けた最少限のご質問・コメントを選んで回答させていただいた.

Recently there has been an increasing interest in learning systems which induce first-order logic programs from examples. However, due to the intractability of the hypothesis space, such systems which use exhaustive search, e.g., MIS, are unable to learn reasonably complex programs. Previous solutions have been proposed to overcome this difficulty : while some systems restrict their hypothesis space, others use heuristics or additional knowledge such as analogical structures or abstraction. However, existing systems still have limitations. GOLEM gives up completeness by restricting the hypothesis space to only determinate clauses. A determinate clause is a clause composed of only determinate literals. A literal is determinate if each new variable in it has only one binding. Even the commonly used generate-and-test programs generate candidate solutions for their test routine, and thus are non-determinate. FOIL avoids exhaustive search by using Gain heuristic to select a literal that greedily discriminates positive examples from negative examples. Although it is able to learn some classes of problems efficiently, the heuristic fails to capture other aspects of usefulness of a literal, i.e., it overlooks a useful literal which produces no discrimination. For instance, a literalpartition (Head, Tail, List 1, List 2) in the quick-sort program does not discriminate between positive and negative examples. We propose a new heuristic-based approach to the learning of Horn-clause logic program with list structure. Our system, CHAM, learns a class of complex programs not learned by previous systems, i.e., non-determinate programs out of the learning space of GOLEM, and programs with non-discriminating literals which pose difficulties for FOIL. Experiments on learning Prolog programs with list structure have shown that while being able to learn a large class of programs, CHAM preserves efficiency in various test problems.

Tense structure plays an important role in determining the temporal relation between the events and states mentioned in narrative text. We generally call events and states situations. This paper focuses on the tense structures of complex sentences in Japanese narrative text. The sentence investigated in this paper contains two clauses (a main clause and a sub clause) connected by a temporal/causal connective, and the sub clause is subordinate to or coordinate with the main clause. The tense structure of a clause is described by the following three entities : the time interval of the situation, denoted by "I", the time of speech, denoted by "S", and the time of reference, denoted by "R". Most of the works on tense structures are based on Reichenbach's notions of three times : the event time (E), the speech time (S), and the reference time (R). In this paper, we use the time interval (I) instead of the event time (E). This facilitates more flexible compositions of the tense structures of complex sentences. We propose a set of rules for composing the tense structure of the complex sentence which contains two clauses. In the combined tense structure of the sentence with a subordinate clause, the speech time is the speech time of the main clause and the reference time of the situation of the subordinate clause is located in the time interval of the situation of the main clause. This composition is based on the result of the linguistic work on the tense structures of Japanese complex sentences. Our idea of locating the reference time of the situation of the subordinate clause in the time interval of the situation of the main clause is compared with some other theories of the tense structures.

This paper introduces a new mechanism of PAFE which provides a flexible environment for feature extraction. This system is constructed based upon concurrent object model (multi agent model), in which many objects work co-operatively in concurrent ways. With these objects, this system performs the parallel search to extract the features from images. The features and their extraction methods are organized in a feature extraction network. In this network, features can be defined in multiple ways and can be extracted in multiple ways. Using this network, different paths (that means extraction method) can be executed in parallel. This system, therefore, provides a coarse grained parallelism which realizes the execution of multiple extraction methods for one kind of feature in parallel as well as extraction of multiple kinds of features in parallel. Also extraction combinations of top-down ways and bottom-up ways realize flexible control of the feature extraction. The efficiency was tested in some experiments in which this system was applied to some 2-dimensional objects.

This paper describes the Oil Shipment Scheduling Expert System (OSSES) to solve the oil shipment scheduling problems for berths and arms and the path selection problem. These problems have many complex constraints. Therefore, the expert system which can solve efficiently these problems is needed. OSSES can make one-day scheduling on shipping many petroleum products for many ships under the consideration of the deadlines and the line-blending and pipeline-cleaning mechanisms. OSSES can perform scheduling in response to some accident.

This paper proposes a method for dealing with a certain kind of expensive production, contriving the expression of instantiations (ordered subsets of working memory elements). Expensive productions are rules which would be required the extraordinary time and space to match (for each rule, compare the left-hand side against the current working memory). The object expensive productions are such that the left-hand side can be decomposed into plural sets of interdependent condition elements. We introduce a new simple notation expressing plural instantiations satisfying a rule's left-hand side, using direct products combining subsets of working memory elements. By this notation, computer loads of executing expensive productions can be reduced. Moreover, empirical and analytic results demonstrate that the new instantiation's expression can substantially reduce the execution time of production system programs.

This paper presents the architectural overview of the IKD system which functions as an interface between a knowledge-based system (KBS) and a relational database management system. For integrating them more flexibly and naturally at conceptual level, it is necessary to represent conceptual structure provided in both of KBS and relational database (DB). Furthermore, such mechanism are required to be built into a knowledge representation system. The IKD system is a subsystem of the knowledge representation system KBUS which is implemented by us. The IKD system is a KBS for itself, implemented in frame-based knowledge representation. Its knowledgebase includes the description of conceptual data structure of the KBS and DB, and some procedures. From structural viewpoint, this knowledgebase consists of five sub-frame-systems. Among them, two sub-frame-systems play more important role in representing the structures than others. They are data- and problem-structure description sub-frame-systems. A collection of relation schemata is represented in the former, and conceptual objects which the KBS deals with are represented in the latter. Then these sub-frame-systems are related to each other. In addition to them, the knowledgebase has a field description, cursor description, and SQL sub-frame-systems. They represent the data types of the fields in relations, cursors, and components of the database language SQL, respectively. The cursor is specified basing on some problem structure description frame, which manipulate a collection of data stored in the DB by a tuple at a time. While, operations on it are done basing on message passings among those frames. A collection of frame-based data required by the KBS is specified in terms of above described frames. A select-statement to get them is constructed by a sequence of message passings among a problem-and data-structure description frames. A collection of data stored in the DB are captured by sending a message to a cursor description frame. This system is a software tool which is designed basing on a loose-coupling approach for integrating both KBS and DB. Two mechanisms adopted in the system are distinguished as being more effective than others. The first mechanism concerns with the frame, that is, two types of frames are defined ; the one is used for representing a data structure of the DB and the other used for satisfying requirements of realization of the KBS. The second mechanism is devised so that operations on the DB are realized by using several types of message passings. These mechanisms are usable for constructing an interface between the KBS and the DB system. As an application, we tried to integrate a production system and the DB. We describe this application briefly, also. The system is useful in this application system especially for its structural flexibility.