【作者】 叶枫；

【导师】 胡曙中；

【作者基本信息】 上海外国语大学 ， 外国语言学及应用语言学， 2011， 博士

【摘要】 意义是语篇交际的主要内容,语义是语篇的本质属性。语篇语义学是对超越句子的语言复合体的意义的研究。但是,语篇语义学理论不充分、不完善,是当前制约自然语言处理的瓶颈之一。语篇意义的研究有较长的历史,但是作为一个研究领域“语篇语义学”是在1980年代才正式被提出,而且目前的几种学派各有侧重,研究理论和方法尚未形成统一的、完整的体系。甚至语篇语义的实体概念都没有清晰地提出过,基本都是把语篇意义作为心理学或话语分析的一个话题展开,偏重把语篇的意义看作是心理的属性,而语篇语义的语言学范畴的系统研究不多。本文以语言项的形式-意义制约关系为主线,把制约的视角拓展到语篇的层次,探索语篇的语义内容和结构的来源、成因,以及处理的方法,构建一个完整的语篇语义分析和处理的体系。本文的目的是在小句之下、句子层、句际关系、语篇层等四个层次上,讨论知识整合、信息操作、形成推理、语义表征等方面语义操作,构建一个综合的、整体性的语篇语义分析和处理的模型,对语篇意义的来源或基础、语篇整体的语义信息的处理过程、语义内容的组织方式、语篇语义结构的形式化表征等方面做出解释和说明。国外语篇语义研究主要有四大学派。欧洲大陆学派的语篇语义研究,侧重于从语言学和心理学的角度研究语篇的意义内容。van Dijk(1980)讨论了语篇语义的宏观结构,分析如何通过宏观规则的操作形成宏观命题;后来又在话语分析的框架下提出了话语的语义分析的七大特征或原则(Van Dijk, 1985a)。Seuren(1985)的《语篇语义学》是第一本以此为题的专著。它以模型-理论范式,整合了简化的命题(reduced propositions)之间的语篇计算和真值条件计算,从而展现了句子的意义是如何受到连贯语篇中的条件制约的。但是,他的着眼点最终还是以句子为主,对语篇的语用功能、语篇整体的特征等都没有涉及。悉尼学派的语篇语义研究,是在系统功能语法框架下,从社会学的角度研究作为社会符号的语篇在人际交往中的意义或功能,主要是结合系统理论和功能理论,研究语篇的概念意义、人际意义和谋篇意义,并在系统功能的基础上提出了语言评价理论和语篇格律理论。语篇语义学是系统功能语言学的一部分,不是一个独立的分支。美国的形式语义学派的语篇语义研究,叫做动态语义学,代表理论是话语表征理论。语篇语义的基础是句子的真值条件,依靠符号和公式做严格的演算和表征,来计算、推导小句之间的意义关系。美国的功能主义学派的语篇语义研究,主要是在功能语法和话语分析领域,在信息结构、话题结构、指示关系等课题的研究成果深化了对语篇语义的认识。在语篇结构方面,修辞结构理论可以说明语篇段之间功能上的关系和整个语篇的层级性结构。在心理学的语篇理解的研究中,Kintsch(1988)的构造-整合模型,解释语篇中的语言知识是如何操作和整合,以形成语篇的心理表征的。Gernsbacher(1990)的结构建造框架,认为语篇理解是通过对语篇提供的信息的心理表征的强化或抑制,实现语篇的心理表征的构建。国内的语篇语义学研究刚起步,只有少数几篇论文,都是系统功能语言学框架下的语篇意义的研究。心理语言学的语篇语义研究侧重依照语言理解的心理学机制解释语篇语义的处理过程,但是在语言学上的观察和实验却是离散的,缺乏对语言现象的全面的把握和处理。形式语义学通过真值条件的逻辑公式来描写语篇语义关系,虽然准确、简洁、无歧义,但是对真值条件不确定的句子,如涉及语境和语用因素的,则无法处理。功能主义的语篇语义研究,较好地分析和掌握了语言的形式、结构和功能之间的关系,但是没有形成对这个分析过程的形式化的即表征,需要一个形式化的体系来全面而系统地记录语义分析和处理的过程。本文有两个根本问题。一个核心问题是:语篇语义学在多大程度上可以落实到语言的实体上?换句话说,语篇的语义在多大程度上是语言为中心的?因为语篇的是“使用中的话语”,必然涉及社会、心理等方面的因素。语篇语义学在多大程度是语言学的,多大程度上是社会学和心理学的?这个问题又可以分解为几个子问题。首先,语篇语义的基础来源是什么?是词语?句子?还是语篇?其次,语篇的语义是怎么形成的?如何从词语、句子,以及句子和语篇的组织获得语篇整体的意义?第三,语篇的语义具有怎样的结构?另一个核心问题是:语篇的语义如何描写或表征?即,如何把语篇的语义分析的结果整合在一个统一的框架中描写出来。本文的语义的基础是语言哲学上的整体论,即一个元素只有在整体中才有意义。语篇语义学既尊重语篇语义的组合性,从词、句的语义实体出发,进行语篇语义的构建;又重视语篇意义的整体性,充分考虑语篇整体的规约结构和意义。而且,整体论的语言学体系是非模块性,本研究中语篇语义分析的四个层次的区分是为分析的方便,它们并不是自然的、自治的,而是有内在的联系的。本文的理论基础是语篇语言学中的语义理论和构式语法中形式-意义对应的观点。在语义构建方面,本文认为语言项体现了形式-意义的制约关系,把语篇语言学中语义特征的分析和语篇语义构建中的各成分之间的关系,都分析为形式-意义之间的制约条件。这些制约条件标记为属性-值矩阵,以盒装图的形式表征出来,实现对语篇语义的描写。本文的研究方法是分析和比较语篇语义研究的理论,以制约条件为主导,构建起本研究的理论框架,然后通过语言实例来分析和论证,完善和改进该模型。在本研究的语篇语义学模型中,对语篇语义的处理包括三个方面:(1)内容,(2)结构,(3)表征。首先是对语篇语义的基础来源和基本内容的分析,进而是对这些内容的组织结构的分析,最后是把前两者予以形式化的表征。语言项是意义与形式的结合体,二者是一种互动的制约关系,这种制约是语篇的语义信息处理的指向标,指导着语篇的信息单位以何种形式来操作:建立新信息,强化或者抑制已有信息。本文在四个层次上来分析语篇语义和结构方面的制约关系。1.小句之下的信息和特征,主要包括:(1)独立的程序语的语义信息和特征,(2)短语的可及性与信息处理的关系,(3)关键词对于语篇意义框架的激活作用三个方面。2.句子层的信息和特征。小句是语篇语义的基本单位,语义分析包括三个步骤:(1)先从小句的题元结构分析获得基本的概念意义或逻辑意义;(2)然后通过信息结构的分析,明确语义内容在小句内的分布所传递话题、焦点等意义;(3)最后,还要考虑句子整体所具有的构式意义。3.小句关系的信息和特征。明确句际关系的本质是语义关系,在表达上分为两种情况:一是有语言标识的句际关系,有联系语和语法结构两个类别;二是无语言标识的句际关系,这主要是推理。4.语篇整体的信息和特征。把语篇看作是体现了功能和形式的对应的规约化的构式,相对应地本文从体裁和语篇类型分析语篇的交际功能和形式结构,重视语篇整体结构所传达的信息。这四个层次中,1和2属于语篇语义内容的分析,3和4属于语义结构的分析。二者又可归为一个范畴,即对语篇语义的实体的分析。语篇语义的另一个范畴是表征,即对语篇语义的形式化描写。本文提出以嵌套的特征结构盒装图为主要手段,表征语篇段的语义关系和结构,实现对语篇语义内容分析和结构构建的过程的描写。本研究试图证明,以制约条件为主线,通过构建一个整体性的分析和表征的体系,一方面,我们可以实现自下而上的从词语到语篇的语义分析,也可以实现自上而下的语篇整体对其成分的意义和结构的制约因素的分析,另一方面,也可以通过特征结构的盒装图实现对整个语篇语义分析和处理做全程的描写和表征。本研究也存在一些局限性。首先,研究方法依然属于内省法。理论的完整性和实践的可行性还需要在自然语篇的意义处理中进一步检验和完善。在语料方面还不够丰富,中英文语料也没有区分,而且对语料的分析可以更详细和深入。语篇语义分析的盒状图的形式表征还要进一步细化。更多还原

【Abstract】 Discourse is“the totality of codified linguistic usages attached to a given type of social practice”. The chief function of discourse is to transfer meaning in communication, and semantics is the core in discourse studies. Discourse Semantics (DS) is the study of the meaning of any language complex beyond the scale of sentence.Discourse semantics is just at the beginning both in theoretical and empirical studies. By far researchers have not yet made consistent and systematic description or analysis on the phenomena and facts concerning discourse meaning. In a sense, discourse semantics has become the bottle-neck to restrain full processing of natural discourse. Consequently, there is an emergent call for comprehensive and sufficient studies in this area.Studies on discourse meaning have a long history, but Discourse Semantics as a branch of linguistics was not acknowledged until by the 1980s, when several research schools address to the topic from different perspectives. As the approaches adapted by these schools are so diversified, it cannot be truly said that DS has developed into a full-fledged research paradigm, in particular that there still lack a widely acknowledged theoretical framework and practical methodology in the field. Even the notion of“discourse semantics as an entity”has not been clearly raised and defined. In fact, researchers have been treating DS as a subtopic or a subdivision of psychology and/or Discourse Analysis, putting premium on the psychological aspects of DS, with the linguistic aspects largely less attended.This dissertation is a tentative study on Discourse Semantics which attempts, first, to build up a comprehensive and holistic framework that incorporates the processes of discourse semantic analysis in terms of knowledge integration, information management and inference forming, and then, to develop a representative mechanism to formalize such an analytical and constructive procedure. In general, the model of discourse semantics here includes three parts: the analysis of semantic sources and contents of a discourse, the analysis of the semantic structure, and the representation of discourse meaning.For the convenience to analyze the contents and structure of discourse meaning, the paper breaks down the analytical process into four levels, i.e. the under-clause level, the clause level, the inter-clause relations, and the discourse level. It should be clearly noted in advance that such four levels are an arbitrary classification for the sake of technical processing, which is by no means to say that they are natural and autonomous boundaries or levels of discourse semantics. In fact, it is rather common to find that, in a discourse, the processing of some language items simultaneously involve more than one levels of semantic analysis.Language items in discourse are taken as“constructions”, a notion borrowed from Construction Grammar, which displays constraints between form and meaning. The notion of“form-meaning constraint”is the main thread running through the process of discourse semantic analysis and representation. From the constraint perspective, it first analyzes the sources of discourse meaning from words to clauses, till to the discourse level, and then incorporates the contents of such sources into an integrated system.Studies in DS can be classified into four major schools. First, the European Continental School took initiatives in the area in the 1980s. From the psycholinguistic perspective, researches usually follow the pattern of language comprehension in mental mechanism while analyzing discourse semantics. Chief researches and achievements in this school are van Dijk’s Macrostructure theory. Seuren gave a different picture in Discourse Semantics, in which DS is reduced to incremental values of sentence meanings, determined not only be truth-conditional computation but also by discourse computation. However, Seuren’s model overlooks the function of discourse patterns.Second, the Sydney School studies DS within the research paradigm of Systemic and Functional Grammar (SFG). From the sociological view, discourse is taken as social semiotics which fulfils the communication of ideational meaning, interpersonal meaning and textual meaning. Chief achievements of DS in this school are Appraisal Theory and Periodicity Theory. As a result, DS is not an independent branch of linguistic study, for it is just a level or a division of language analysis in SFG.The third school is with the study of formal semantics in the U.S., which is known as Dynamic Semantics. Influential theory in this respect is Discourse Representation Theory (DRT), and its revision of Standard Discourse Representation Theory (SDRT). DS in this approach is based on truth values of sentences, which are computed and represented by rigid signs and formulae.The fourth school is the American Functional School, mainly in the fields of Functional Grammar and Discourse Analysis. Research findings in information structure, topic and focus, reference identification, etc. have enhanced understandings in discourse meaning. As a result studies about theories and methodology in discourse semantics are brought to a new stage.Besides, in researches of discourse comprehension in psychology, Kintsch’s Construction-Integration Model and Gernsbacher’s Structure-Building Model provide different but comprehensive framework for the processing of discourse segments in the mental mechanism, which are highly helpful in guiding linguistically semantic processing of discourse meaning.DS in China is just at the beginning, with only a few papers in the area, all of them in the Systemic Functional Grammar approach.DS from the psycholinguistic approach gives an account of semantic processing in light of psychological mechanism, but the manner of observation and analysis on language data is discrete, which means that it may neglect some crucial linguistic aspects concerning discourse meaning. DS in formal semantics is accurate, concise, and nonambiguous, but it fails to account for context-sensitive sentences whose truth values are not clear or even without truth values. The functional approach gives comprehensive and systematic analysis for DS, but it requires a representative formalism to record the analytical procedure and the construction of discourse semantic structure.The research addresses two fundamental questions. One is to what extent DS can be reduced to linguistic entities. In other words, how can we say DS is derived from language facts? As discourse is“language in use”, discourse semantics naturally involves social and psychological factors, among other things. Then, how much can we say that DS is linguistic? And how much social and/or psychological?In this research, discourse semantics is treated as linguistic entities. This question is further addressed in several subquestions. First, what are the sources of discourse semantics? Is DS from words, sentences, or discourse? Second, can we obtain meaning from the structure of a discourse by analyzing text organization? Third, does DS have a semantic structure? What is it like?The other question is how to describe and represent DS. That is, how can we represent the analytical results of semantic contents and structure of a discourse in a consistent and systematic formal mechanism?The semantic view of this research is holism in language philosophy. Hence compositionality and holistic nature of discourse are treated on a par in this model. This model starts from building up semantic contents to form a structure of discourse meaning in the bottom-up fashion, and in the meantime, it also takes into serious account of the holistic features of discourse as conventional patterns in the top-down fashion. What’s more, according to holism, linguistic system is not modular. Therefore, the distinction of four levels of semantic analysis in this dissertation is simply for analytical convenience. It is by no means to say that any level is a natural and self-contained part of DS. In fact, there are interconnections and cross-references among them.In theory, the research is based on two foundations: theories in Discourse Analysis (or Text Linguistics) and that in Construction Grammar. Following the principle of constraints between form and meaning, it first carries out thorough analysis on semantic contents and semantic structure of discourse, which are treated as linguistic entities of DS. These semantic entities are incorporated in a representational formalism in terms of attribute-value matrix, displayed in the shape of box diagrams.The research takes a theoretic-model approach to DS, but further reduces semantics to manageable proportions. By an analytical and critical review of previous theories in related areas, the dissertation first establish for itself a scaffolding of discourse semantic analysis, and then set up a framework of representation formalism. Then it supplies with language data to test the operational feasibility of such model.In general, the research consists of three parts, which are contents, structure, and representation of discourse semantics respectively. It first accounts for the sources of discourse meaning, i.e. words and sentences; then it discusses semantic structure of discourse meaning in terms of organizational relations and discourse patterns; in the end it provides a representative formalism for the former two parts.A language item stands for a correspondence between form and meaning, with inevitable constraints between the two sides. Such constraints give indication to the direction and the manner of how to process such an item in a discourse: either to enhance the previous information unit by projecting the new information on it if the constraints are closely related, or to suppress the previous information if the constraints are opposite, or to establish a new information unit if the constraints are irrelevant. Constraints in discourse semantic contents and structure are analyzed in four levels.1. Information and properties beneath the clause level. Information and properties which contribute to discourse semantics in this level are in three categories: 1) the independent procedural information, 2) forms of phrases and their accessibility, and 3) functions of key words in facilitating interpretation of other language items and in forming inference by activating a situational frame.2. Information and properties at the clause level. Clauses constitutes the basic units of discourse meaning. Semantic analyses of clauses in discourse are carried out in three stages: 1) argument structure analysis, to get the basic information of its conceptual meaning and logical structure, 2) information structure analysis, to capture properties of information distribution within the clauses, and 3) construction analysis, to include the constructional meaning of the clause as a whole.3. Information and properties of inter-clausal relations. Clause relations in nature are semantics relations, which are classified into two categories according to the surface linguistic expressions: 1) clause relations with linguistic signals and 2) those without. For the former, there are two types: i) those connected with connectives and ii) those connected by grammatical structures. For the latter, clauses are connected by forming inference.4. Information and properties of discourse as a whole. Discourse patterns are conventional constructions, also with correspondence between the function and form of discourse, which are analyzed in terms of genres and text types respectively. This section puts particular emphasis on the holistic features and functions of discourse. Among the four levels, the first two belong to the contents of discourse semantic analysis, while the latter two are the structure of discourse semantic analysis. Those two categories constitute the linguistic entities of discourse analysis, thus forming one unified respect of DS in this model.The other respect in the model is the representational mechanism, i.e. formalism of DS. The model describes semantic entities and relations in DS in terms of feature structures, which take the shape of (embedded) box diagrams.The research proves that we can establish an integrated and holistic framework to account for the analysis and representation of DS according to constraints between form and meaning. On one hand, it not only simulates the procedure of building up meaning from words, sentences, till to discourse in the bottom-up way, but also accounts for the functions and properties of discourse against its components in the top-down way. On the other hand, it incorporates the analytical procedure into a representational formalism, thus giving a full description to discourse semantic processing.The research also has some limitations. The approach is fundamentally a retrospective one, in that it first sets up a theoretical model and then support with language facts to prove it. Naturally, it calls for more language data to test the feasibility of such a model in natural language processing and then to improve or amend it. Besides, during analysis the language data are not distinguished between English and Chinese, but they may be very different in nature. As a tentative attempt to develop a representative formalism for DS, much of it is rather crude, and in need of fine elaboration in future research.更多还原