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

Research on Reactive Explanation of Information

信息反应性解释研究

基本信息

批准号：
10480069
负责人：
SHIMAZU Akira
金额：
$ 6.53万
依托单位：
Japan Advanced Institute of Science and Technology (JAIST)
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (B)
财政年份：
1998
资助国家：
日本
起止时间：
1998 至 2001
项目状态：
已结题

项目摘要

Aiming to clarify a method which enables computers to convey information to humans in a way of having characteristics seen in explanation by spoken language dialogue, we have conducted the following : collection of explanation dialogues in Japanese spoken language, analysis of characteristics seen in explanation dialogues and modeling of reactive explanation. The characteristics focused on in the research are small utterance units and reactive explanation according to recipient reaction.As to analysis of explanation dialogues, characteristics which seem to influence reactive explanation were studied. Mainly, utterance sequences including confirmation Utterances were shown to are different depending on who has the dialogue initiative, initiative structure in dialogues were shown to be different depending on kinds of explanation, performance and impression of dialogues were characterized according to types of confirmation utterances, three factors such as pause between minimal phrases, sentences without embedded clauses and repetition by request were characterized from the viewpoint of effective transmission of information, and explanation structures Were shown to be different depending on subjects.As to modeling of reactive explanation, we proposed a two-layered model based on a basic model that we had proposed in the previous research and incrementally and applied to a traffic guidance system. We then implemented an experimental system, which reactively makes an explanation plaft as to liver examination data. As to utterance production, we formalized sub-atomic speech acts for fine-grained utterances. As to the dialogue aspect of explanation systems, we modeled a dialogue manager for reactive explanation.

为了阐明一种使计算机能够以具有口语对话的解释特征的方式向人类传达信息的方法，我们进行了以下工作：日语解释对话的收集，解释对话的特征分析和反应式解释的建模。研究的特点是话语单位小，根据受话者的反应进行反应性解释。在解释对话分析方面，研究了似乎影响反应性解释的特征。从有效传递信息的角度分析了包含确认话语的话语序列的不同，发现对话中的主动结构随解释类型的不同而有所不同，根据确认话语的类型表征了对话的表现和印象，从有效传递信息的角度描述了最小短语间的停顿、无内嵌小句的句子和请求重复三个因素，并根据主题的不同表现出不同的解释结构。对于反应性解释的建模，我们在已有研究的基础上提出了一个两层模型，并逐步应用于交通诱导系统。然后我们实现了一个实验系统，该系统对肝脏检查数据进行了反应性的解释。在话语生成方面，我们对细粒度话语的亚原子言语行为进行了形式化描述。对于解释系统的对话方面，我们建立了一个用于反应式解释的对话管理器。