基于小句关系定量分析的语篇测量方法

发布时间：2018-03-22 04:37

本文选题：语篇测量　切入点：语篇体裁　出处：《东北农业大学》2017年硕士论文　论文类型：学位论文

【摘要】：近十年来,随着计算机科学理论的发展,计算语言学在机器翻译、语音识别、人机交互等诸多方面发展迅猛。但是如今的机器翻译或语音识别等人工智能产品仍然存在着自身局限性。以机器翻译为例,它对于大篇幅的整句或者逻辑语义相对复杂的文本,翻译质量令人堪忧。其根本原因在于机器对于语篇的理解是建立在代表语言元素的二进制信息和简单的算法基础上的;而人对于语篇的理解则是建立在对于小句关系的理解之上的。另一方面,丁建新、陈安玲等人对大量语篇体裁的统计研究明确了不同语篇体裁中小句关系分布特征具有不同的特点,这为基于小句关系的语篇体裁鉴别提供了理论基础。因此本论文以韩礼德系统功能语言学中的小句关系为切入点,在广泛文献调研的基础上,充分吸收韩礼德系统功能语言学中关于小句复合体系统理论的优点,论述了该理论下小句关系分类框架存在的不足,并结合国内学者程晓堂对小句关系分类框架的改进意见,首次提出了小句关系特征矩阵和语篇相关度的概念,并在这两个概念的基础上提出了基于小句关系定量分析的语篇测量方法。小句复合体作为语篇中最高级别的语法单位,其内部各个小句之间相互作用,存在多种复杂的关系,这些关系蕴含了丰富的信息,而小句关系特征矩阵作为语篇中小句关系分布特征的直观体现,我们可以从中解读出关于该语篇的丰富的语言学意义。语篇相关度则是从统计学的角度给出了不同语篇体裁之间相关程度的量化分析方法。由语篇相关度概念引申,我们给出语篇差异这个概念,它从另一个侧面反映了不同语篇体裁之间的小句关系分布特征的差异性,并且能够直观地给出具体的差别所在。这些概念和方法的提出,使得我们可以借此对机器进行大规模的语篇数据训练,从而实现大规模语篇材料的自动体裁判别和分类的功能。本论文以定量分析为主,结合统计分析、案例分析、演绎推理、综合归纳、文献检索等诸多研究方法,以不同语篇作为样本数据,其对应的小句关系特征矩阵作为模型参数进行研究。首先对语篇中小句关系类别进行分析得到小句关系特征矩阵,进而对小句关系特征矩阵作误差校正预处理和统计学相关性检验,最后得到语义相关度、修辞相关度和投射相关度的加权平均值,即语篇相关度,可以以此定量地表示不同语篇之间相似性程度。这样便建立了一种基于小句关系定量分析的系统化的语篇测量方法。经过多个语篇案例的实际检验,验证结果与预期符合很好,充分说明了该理论的合理性、正确性和可行性。本论文所提出的基于小句关系定量分析的语篇测量方法不仅可以在微观上推断出语篇本身蕴含的丰富的语言学信息,而且可以在宏观上得出不同语篇体裁之间的相似性程度,并给出定量化的描述。该语篇测量方法在机器语篇分析中具有很强的可操作性和应用价值,为科学、客观、系统的语篇分析研究开拓了新的研究思路。
[Abstract]:In the past ten years, with the development of computer science, computational linguistics at Machine Translation, voice recognition, human-computer interaction and other aspects of the rapid development. But now the Machine Translation voice recognition or artificial intelligence products still have their own limitations. In the case of Machine Translation, it is for a large text or sentence semantic logic relatively complex that is the translation quality is worrying. The fundamental reason is that the machine is built to represent language elements of binary information and simple algorithm based on the understanding of text; and for discourse understanding is based on the understanding of the relationship between clauses. On the other hand, Ding Jianxin, Chen Anling, et al. Clear different genre and sentence distribution has different characteristics of a large number of statistical research on the genre of discourse, the clause relation of genre identification based on The theoretical basis of this paper. The clause relation Hallidy in systemic functional linguistics as the starting point, on the basis of extensive literature investigation, fully absorb Hallidy in systemic functional linguistics about the clause complex system theory discusses the advantages, shortcomings of the theory of clause relation classification framework, and combined with the suggestions for improvement the domestic scholar Cheng Xiaotang framework of clause relation classification, first proposed the clause relation feature matrix and discourse the concept of correlation degree, and puts forward the clause relation of quantitative analysis in the measurement method based on discourse on the basis of the two concepts. The clause complex as a grammatical unit of the highest level in the discourse, the each interaction between clauses, there are several complicated relationship, the relationship contains abundant information, and the clause relation characteristic matrix as a discourse and sentence distribution straight The conception, we can out of the discourse rich linguistic meaning in discourse interpretation. Correlation is to quantify the degree of correlation between different genre analysis method is given from the perspective of statistics. Composed of discourse related concept, we give the concept of discourse differences, it reflects the difference between different genres of clause relation distribution from another side, and can directly give specific difference. Put forward these concepts and methods, so that we can take the machine text data of large-scale training practice, so as to realize the automatic identification and classification of large-scale genre text materials function. This study is mainly based on the quantitative analysis, combined with statistical analysis, case analysis, deductive reasoning, summarizing, literature retrieval and other research methods, as the sample data in different texts, the corresponding clause To study the relationship between the characteristic matrix as model parameters. Firstly, analysis of the clause relation feature matrix of discourse relations of small sentence categories, clause relation feature matrix error correction preprocessing and statistical correlation test, and finally get the semantic relevance, rhetoric correlation and correlation projection weighted average value, namely, discourse this correlation, can quantitatively represent the degree of similarity between texts. It established a small sentence systematic quantitative analysis of the relation of discourse measurement method based on multiple discourse. Through actual case inspection inspection, verification and expected results are in good agreement, fully illustrates the rationality of the theory. The correctness and feasibility. This paper proposed the clause relation of quantitative analysis measurement method based on discourse can not only infer the discourse itself contains rich linguistic information in the micro level, and And that the degree of similarity between different genres at the macro level, and gives a quantitative description of the discourse. Measurement methods in machine in discourse analysis has strong maneuverability and application value, scientific, objective, discourse analysis system research has opened up new research ideas.

【学位授予单位】：东北农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：H05

【参考文献】