首页|汉语让转复句的预期类型与常识发现研究

汉语让转复句的预期类型与常识发现研究

扫码查看
自然语言有关表达中潜藏了丰富的预期和常识,让转复句"虽然A,但是B"格式是其中的代表.从语言学知识应用于人工智能的角度来看,挖掘语言中的常识有助于构建常识库,支持常识推理.研究过程是通过构建"取反→简化→补全→泛化→联结"的形式化操作流程,提取让转复句中蕴涵的预期信息,证明语用信息也能通过形式化方式提取.根据格式中"A"和"(-)B(非B)"是否存在因果关联,让转复句中的预期可分为基于直觉推理的预期、基于经验的预期和基于百科知识的预期三类.根据"A"和"(-)B"的语义关联性及推理方式,基于直觉推理的预期可进一步分为属性类推、属性对比、量级类推、属性继承、属性值保持和属性值趋同六类;按照时间和空间两个维度,基于经验的预期可分为基于预测类经验的预期和基于共现类经验的预期两类.其中,基于经验的预期和基于百科知识的预期可以归纳为常识,收集并构建常识库.最后举例介绍了怎样利用本文抽取的小型常识库进行智能问答和溯因推理.
Research on Expectation Types and Common Sense Discovery in Chinese Concessive and Adversative Compound Sentences
Some expressions in natural language harbor rich implicit expectations and common sense,with constructions of concessive and adversative compound sentence like"Suiran(although)A,Danshi(but)B"serving as a prime example.From the perspective of applying linguistic knowl-edge to artificial intelligence,mining common sense from language aids in constructing a common sense knowledge base to support common sense reasoning.The research process involves a formalized operational flow of"negation → simplification → completion → generalization → connection,"extrac-ting the implicit expectations inherent in concessive constructions,demonstrating that pragmatic in-formation can also be extracted through formalized means.Depending on the causal relationship be-tween"A"and"(-)B(not B)"in the format,the expectations in concessive constructions can be cate-gorized into three types:intuitive reasoning-based expectations,experience-based expectations,and encyclopedia-based expectations.Based on the causal relationship and reasoning mode of"A"and"(-)B,"intuitive reasoning-based expectations can be further classified into six types:attribute analogy,attribute contrast,magnitude analogy,attribute inheritance,attribute value preservation,and attrib-ute value convergence.Along the dimensions of time and space,experience-based expectations can be divided into sequence-based expectations and co-occurrence-based expectations.Both experience-based and encyclopedia-based expectations can be generalized as common sense,which can be collected and constructed into a common sense knowledge base.Finally,examples are provided to illustrate how to use the extracted small-scale common sense knowledge base for Question Answering(QA)and Ab-ductive Inference.

expectationcounter-expectationcommon senseconcessive and adversative com-pound sentence

卢达威、袁毓林

展开 >

中国人民大学文学院/数字人文研究院 北京 100872

澳门大学人文学院 澳门 999078

北京大学中文系/中国语言学研究中心 北京 100871

预期 反预期 常识挖掘 让转复句

国家社会科学基金项目

18CYY030

2024

语言科学
徐州师范大学语言研究所

语言科学

CSSCICHSSCD北大核心
影响因子:0.583
ISSN:1671-9484
年,卷(期):2024.23(4)
  • 5