Knowledge Information Processing System Based on Analogical Reasoning

本项目旨在开发一个基于类比推理（AR）坚实理论基础的知识信息处理系统（简称KIPS）。在这个项目中，我们用一组Horm子句来表示知识，以充分利用逻辑规划的理论成果。首先，我们建立了由分句实例化、分句变换和现式三个推理规则组成的反义推理理论，从而我们的反义推理可以在与演绎推理和归纳推理相同的框架中进行处理。其次，我们利用Prolog在Sun-3系统上实现了AR系统，大大提高了系统的效率；新版本的运行速度是旧版本的6到12倍。我们还在AR系统和日本用户之间划分了一个新的简单的日语界面。通过使用一个非常小的字典，界面与用户沟通，分析日语文本，并解决其中的语法歧义，然后将其翻译成霍恩类。反向翻译，即利用与使用者交流过程中获得的语法信息，完全自动地将霍恩分句翻译成日语。因此，我们的界面作为用户和系统之间的协调器。这个简单的想法确实减轻了界面的负担。在进行AR之前，我们必须从知识库中为AR选择一个源域。为了这个目的，我们发展了一种抽象知识的技术。通过使用抽象的知识，KIPS可以检索候选的源域，也可以排除无用的源域。基于上述自然语言接口和抽象技术，我们完成了基于AR的KIPS的开发。对于AR的其他理论基础，我们从罗素确定、层次域和归纳推理等角度对AR理论进行了概括。在知识信息处理（包括AR. Less）中，我们也获得了平行推理复杂性的重要结果

This project aimed at developing a knowledge information processing system(KIPS,for short)based on firm theoretical foundations of analogical reasoning(AR,for short). In this project we represent knowledge by sets of Horm clauses so as to take full advantage of theoretical results of logic programming.First we have established a theory of AR which consists of three inference rules, instantiation of clauses, transformation of clauses and modus ponens,and thus our AR can be dealt with in the same framework as decluctive and inductive inferences. Secondly we have realized an AR system in Sun-3 system using Prolog, and made it much more efficient;the new version works 6 to 12 times as fast as the old one.We have also divided a new simple Japanese language interface between the AR system and Japanese users. By using a very small dictionary, the interface communicates with the users to analyze Japanese texts and to resolve syntactical ambiguities in them before translating them into Horn cla … More uses. The converse translation, that is, translations from Horn clauses into Japanese are carried out fully automatically by using the grammatical information obtained in the process of the communication with the users. Thus our interface works as a coordinator between users and the system. This simple idea has really lightened burden of the interface.Prior to AR, we must select a source domain for the AR from the knowledge base. For this purpose we have developed a technique for abstraction of knowledge. By using the abstracted knowledge the KIPS can retrieve a candidate of the source domain and also can exclude helpless source domains.Installing the above mentioned natural language interface and abstraction technique, we have completed the development of the KIPS based on AR.As to the other theoretical foundations of AR we have generalized the theory of AR from such viewpoints as Russell's determination, hierarchical domain and inductive inference. We also have obtained important results on the parallel complexity of inferences in knowledge information processing including AR. Less