【作者】 张丹；

【导师】 瞿云华；

【作者基本信息】 浙江大学 ， 外国语言学及应用语言学， 2022， 博士

【摘要】 歧义是语言中普遍存在的现象。歧义现象是汉语语言处理面临的重大难题和障碍。歧义结构的特点和歧义消解问题一直以来备受关注。已有不少研究对歧义结构进行了结构分析、语义分析和消歧研究。但是前人文献大多缺少系统的理论指导,忽视歧义结构句法和语义之间的相互关系,鲜有研究为具体歧义结构研究提供统计数据支持,缺乏面向计算机处理的形式化规则研究。因此,本研究目标形成一套有理论指导的、把握歧义结构特征及其消歧条件的、重视句法和语义之间互动关系的结构消歧形式化研究。本研究基于“潜在歧义论”、“模式语法”和“知识本体”三者构成的理论框架,构建了适用于计算机处理的汉语潜在歧义模式消歧模型。汉语歧义模式消歧模型分为:“潜在歧义模式的语义组合-歧义类型对应”、“潜在歧义模式的语义组合-句法关系对应”和“潜在歧义模式概率消歧”三大模块。基于此模型,归纳潜在歧义模式的语义组合与歧义类型的对应关系、分析语义组合与句法关系的对应关系、以及为潜在歧义模式进行概率消歧,为汉语潜在歧义模式归纳出形式化消歧规则。基于汉语潜在歧义模式消歧模型的研究步骤如下:(1)构建语料库。本研究以浙江大学汉语书面语和口语平衡语料库为基础;增加部分新浪新闻语料,以此反映互联网语言特征。(2)汉语语料句法分词。首先利用Stanford句法分词系统对汉语语料进行句法分词,并将Stanford分词系统中原有的标注替换为ICTCLAS词性标注。(3)提取汉语模式。利用Python编写自动提取汉语模式的程序,利用自动提取程序提取汉语模式。(4)人工分类判断汉语潜在歧义模式的歧义类型,将汉语潜在歧义模式分类为歧义消除实例,准歧义实例及真歧义实例。(5)对汉语潜在歧义模式进行语义标注。利用《新编同义词词林》对提取出的汉语模式进行知识本体语义标注。(6)归纳各类汉语潜在歧义模式的语义和句法对应特征。(7)对潜在歧义模式进行概率消歧。(8)归纳和统计汉语潜在歧义模式的语义组合特征,总结汉语潜在歧义模式消歧规则。构建潜在歧义模式消歧模型的过程中主要有以下发现:1.潜在歧义模式与语义类型有互动关系,提出“语义-歧义互动关系”体系,发现语义类型限定歧义模式的歧义类型,语义类型组合构成的句法关系限定歧义模式的歧义类型,语义类型与歧义模式的句法结构具有概率关系;2.某些模式具有歧义性;3.知识本体概念搭配对句法结构有制约作用;4.认知概念图式可以解释模式产生歧义的原因。本研究的方法创新体现在以下三方面:1.从模式的语义类型和句法结构的关系视角研究歧义现象;2.将传统的质性歧义研究提升到了定量研究阶段;3.基于模式语法理论的歧义研究体系优于传统的短语结构语法歧义研究体系。本研究的主要贡献体现在:1.构建了潜在歧义模式的本体研究平台;2.构建了汉语潜在歧义模式研究的理论框架;3.构建了一套适用于计算机处理的汉语潜在歧义模式消歧模型;4.归纳出基于句法和语义互动关系的形式化消歧规则;5.提出了“语义-歧义互动关系”体系。更多还原

【Abstract】 Ambiguity,as a universal phenomenon in language,is a major issue as well as an obstacle in Chinese language processing.The characteristics and disambiguation of ambiguous structures have long attracted widespread attention.Previous studies have generally been carried out from three perspectives: syntactic analysis,semantic analysis,and disambiguation strategies.The existing literature,however,has seldom been guided by systematic theories,rarely focused on the relationship between the syntax and the semantics,and rarely provided the statistical data for the specific ambiguous structures,leading to the lack of the disambiguation rules for the computer language processing.In the light of the shortcomings of the previous researches on Chinese ambiguous structures,guided by the systematic theories,this thesis is aimed at conducting research on Chinese potential ambiguous patterns by emphasizing the correlation among syntax and semantics in which the formal rules for disambiguation can be inducted.On the basis of the theoretical framework consisting of "Potential Ambiguity Theory","Pattern Grammar",and Ontology,a disambiguation model for Chinese potential ambiguous patterns is constructed and applied in the research.The disambiguation model is classified into three modules: "Correspondence between the semantic collocation and the ambiguity types","Correspondence between the semantic collocation and the syntactic relation",and "Probabilistic disambiguation for the potential ambiguous patterns".In accordance with this disambiguation model,firstly,the corresponding relation between the semantic collocation and the ambiguity types is summarized;next,the correspondence between the semantic collocation and the syntactic relation is analyzed;and finally,the probabilistic disambiguation for the potential ambiguous patterns is carried out.In the process,the formal disambiguation rules for Chinese potential ambiguous patterns are generalized.The following are the specific research techniques used to develop the disambiguation model for the potential ambiguous pattern:(1)Construct a corpus.This research begins with selecting a balanced Chinese written and spoken corpus by Zhejiang University;secondly,in order to reflect the characteristics of the Internet language,this research chooses part of the Sina News Corpus.(2)Word segmentation.The Chinese corpus is initially segmented using the Stanford word segmentation system.The original tags in the Stanford word segmentation system are replaced by ICTCLAS.(3)Extract the patterns.Python is used for programming for the purpose of extracting the Chinese patterns automatically.(4)Classify the potential ambiguous patterns.The potential ambiguous patterns are categorised into the ambiguity elimination examples,the quasi-ambiguity examples,as well as the true ambiguity examples.(5)Label the potential ambiguous patterns with the ontology knowledge."Xinbian Synonyms Cilin" is utilised to label the extracted Chinese patterns.(6)Generalize the correspondence between the semantic and the syntactic features of the potential ambiguous patterns.Features in the three types of potential ambiguous patterns are investigated.(7)Make use of the probabilistic method to disambiguate the potential ambiguous patterns.(8)Summarize the disambiguation rules for the potential ambiguous patterns.The disambiguation rules for Chinese potential ambiguous patterns are summarized on the basis of the statistics of the semantic collocation.The main findings are as follows: 1.Semantic collocation rules are consistent with the semantic characteristics of the majority of potential ambiguous instances;2.Semantic types limit the ambiguity types and a particular type of the potential ambiguous patterns is composed of specific semantic types;3.Semantic types limit the syntactic structures and the specific syntactic relation corresponds to the particular semantic types;4.The semantic collocation can be employed in the probabilistic disambiguation,providing the maximum probability for the syntactic analysis.The "semantics-ambiguity interaction" system is proposed as a result of these findings,which reveals the close relationship between ambiguous patterns and the semantic types.The methodological innovations in this research consist of: 1.focusing on the relationship between the semantic types and the syntactic structures in order to study the ambiguity phenomenon;2.investigating the ambiguity with the quantitative method which is uncommon in earlier studies;3.based on Pattern Grammar,there are more advantages in analysing the relation between the form and the meaning in comparison to the traditional phrase structure grammar.The primary contributions of this research are: 1.a research platform based on Ontology is constructed for the potential ambiguous patterns;2.a theoretical framework is developed for the studies on Chinese potential ambiguous patterns;3.a disambiguation model is created for Chinese potential ambiguous patterns in computer language processing;4.the formal disambiguation rules are generalized for the potential ambiguous patterns;5.the system of "semantics-ambiguity interaction" is proposed.更多还原