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

Building resources and a model for computing paraphrase based on lexical semantics

构建基于词汇语义的释义计算资源和模型

基本信息

批准号：
17300047
负责人：
INUI Kentaro
金额：
$ 10.15万
依托单位：
Nara Institute of Science and Technology
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (B)
财政年份：
2005
资助国家：
日本
起止时间：
2005 至 2007
项目状态：
已结题

项目摘要

Aiming at building a computational model and computational recourses for computing paraphrase at the level of predicate-argument structure, this research project gained the following results:(i) For paraphrase knowledge, a large-scale hierarchical lexicon of predicate-argument structure was built. The lexicon organizes about 4,000 Japanese basic verbs (about 7,000 senses in total) with predicate-argument structure information in a fine-grained semantic hierarchy so that lexical entries in a semantic class can be regarded as near synonyms. For augmenting this knowledge base, additional knowledge about event relations are extracted from glosses found in a human-use dictionary of Japanese. Over 35,000 relations are extracted and classified into 8 relation types, all of which are considered useful for recognizing paraphrase or textual entailment.(ii) For scaling the basic paraphrase knowledge above, automatic acquisition of semantic relations between events from a large corpus was also exp … More lored. We proposed several extensions to a state-of-the-art method originally designed for entity relation extraction, reporting on the present results of our experiments on a Japanese Web corpus. The results show that (a) there are indeed specific cooccurrence patterns useful for event relation acquisition, (b) the use of cooccurrence samples involving verbal nouns has positive impacts on both re-call and precision, and (c) over five thousand relation instances are acquired from a 500M-sentence Web corpus with a precision of about 66% for action-effect relations.(iii) For building a computational model of paraphrase, we explore the regularity underlying these classes of paraphrases, focusing on the paraphrasing of Japanese light-verb constructions (LVCs). We propose a paraphrasing model for LVCs that is based on transforming the Lexical Conceptual Structures (LCSs) of verbal elements. We also propose a refinement of an existing LCS dictionary. Experimental results show that our LCS-based paraphrasing model characterizes some of the semantic features of those verbs required for generating paraphrases, such as the direction of an action and the relationship between arguments and surface cases. Less

本研究旨在建立一个在谓词-论元结构层次上计算释义的计算模型和计算资源，取得了以下成果：（1）对于释义知识，建立了一个大规模的谓词-论元结构层次词典。该词典将约4，000个日语基本动词（总共约7，000个意义）与谓词-论元结构信息组织在细粒度的语义层次结构中，以便语义类中的词条可以被视为近义词。为了增强这个知识库，额外的知识提取事件关系的注释中发现的人类使用的日语词典。超过35，000个关系被提取并分类为8种关系类型，所有这些关系类型都被认为对识别释义或文本蕴涵有用。(ii)为了扩展上述基本的释义知识，本文还研究了从大型语料库中自动获取事件间语义关系的方法 ...更多信息爱我们提出了几个扩展到一个国家的最先进的方法，最初设计的实体关系提取，报告我们的实验在日本的Web语料库的目前结果。结果表明：（a）确实存在有助于事件关系获取的特定共现模式;（B）使用包含动词名词的共现样本对重访率和准确率都有积极影响;（c）从500 M句子的Web语料库中获取了超过5000个关系实例，动作-效果关系的准确率约为66%。(iii)为了建立一个释义的计算模型，我们探讨了这些类别的释义背后的规律性，重点是日语轻动词结构（LVC）的释义。我们提出了一个基于动词成分词汇概念结构转换的LVC释义模型。我们还提出了一个改进现有的LCS字典。实验结果表明，我们的基于LCS的释义模型特征的一些语义特征所需的生成释义，如动作的方向和参数之间的关系和表面情况。少