权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

An Inference Method to Obtain Approximate Solutions under Incomplete Knowledge

不完全知识下求近似解的推理方法

基本信息

批准号：
15560351
负责人：
ISHIKAWA Tsutomu
金额：
$ 2.43万
依托单位：
Takushoku University
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
2003
资助国家：
日本
起止时间：
2003 至 2006
项目状态：
已结题

项目摘要

We have studied an inference scheme to obtain approximate solutions under incomplete knowledge and the common knowledge bases used in the inference, aiming at realization of the same flexible processing as human.Concerning the former, we have improved the inference method developed before, called ARSK (Approximate Reasoning with Similar Knowledge), which derives approximate solutions using similar knowledge, and have proposed a new knowledge representation scheme for it, called WPL (Word-based Predicate Logic). WPL is a kind of extended order-sorted logic. Both a simple sentence and a complex sentence are transformed to one atomic formula with structured sort symbols using the existent natural language processing tools. The inference processes are executed by semantically interpreting sort symbols, which consist of multiple ordinary words, word by word. We have proposed the new inference rules which can deal with such a complex sort symbols.We have also developed and evaluated a tool measuring the semantic similarity between words by using both the knowledge base of words, which was constructed based on Japanese dictionaries and an existent thesaurus. The evaluation result shows that this tool can obtain nearly the same performance in the accuracy as human judgment.

本文研究了在不完备知识下求近似解的推理方案和推理中所使用的公共知识库，目的是实现与人一样的柔性处理，在前者方面，我们改进了已有的推理方法ARSK（具有相似知识的近似推理），其使用相似知识导出近似解，并提出了一种新的知识表示方案WPL（Word-based Predicate Logic）。WPL是一种扩展的序排序逻辑。利用现有的自然语言处理工具，将简单句和复杂句转化为带有结构化排序符号的原子公式。推理过程是通过语义解释排序符号来执行的，排序符号由多个普通单词组成。我们提出了新的推理规则，可以处理这样一个复杂的排序符号，我们还开发和评估了一个工具来衡量词之间的语义相似性，使用的知识库，这是建立在日语词典和现有的叙词表。评估结果表明，该工具可以获得与人类判断的准确性几乎相同的性能。