【作者】 杨春雷；

【导师】 许余龙；

【作者基本信息】 上海外国语大学 ， 外国语言学及应用语言学， 2008， 博士

【摘要】 本文的主要目标是在中心词驱动的短语结构语法理论框架内建立一个汉语中反身代词“自己”和量化名词短语的语用解释的约束条件体系。约束条件是指能够对某些语言现象进行歧义消解的语言规律;约束条件体系通过这些约束条件的共同作用来达到这个目的。首先,论文先分别定义了“自己”和量化名词短语的语用解释的种类:“自己”有2种(特指与泛指);量化名词短语有6种(定指、特指、分指、全指、互指和泛指)。我们先从自然语料中选出含有研究对象的部分分别建立针对两个研究对象的专门语料库,通过语料分析发现并验证可能的约束条件,如:表达形式及约束条件值的分布、特殊词汇、修饰词的数量、句法位置、时态、体、研究对象和先行词或同指成分的距离、文本类别等。如果约束条件证明有效,那么将对其进行评估和量权。研究建立了两个约束条件体系,分别包含8个关于“自己”的约束条件和23个关于量化名词短语的约束条件。然后根据每个约束条件对语用解释影响力的大小,分别给它们施加一定的点数。把在具体的研究对象上发现的约束条件对应的点数相加得到一个点数和。这个点数和是确定语用解释的最终依据。完整的约束体系将应用于自然语料分析,验证其对研究对象在语篇中的语用解释的预测力,并最终确定具有最佳预测结果的点数和作为判断标准。研究表明两个约束条件体系都能成功预测出绝大部分的语用解释,其中关于“自己”的判断成功率为91%,量化名词短语为80%。本文还分别研究了应用中心词驱动的短语结构语法理论分析“自己”和量化名词短语的相关文献,并把两个约束体系嵌入到该理论框架内。之所以采用这个框架,主要是基于三个原因:首先,该理论把各个层次的信息统一纳入到一个特征结构中描写,这符合约束条件体系的性质;其次,汉语中词汇包含了大量决定语法和语义结构的信息,该理论的词汇主义特色也比较符合这个特点;最后,该理论的模块性结构为约束体系的计算实现提供了便利。本文最后还分别提出了语言学研究中约束体系的产生和它的计算实现的流程图。更多还原

【Abstract】 This dissertation aims at establishing expert systems for pragmatic interpretations of Chinese reflexive ziji and quantified noun phrases (QNPs) in discourse respectively. An expert is a constraint on various levels: morphological, syntactic, semantic, or pragmatic, which influences the pragmatic interpretations of certain linguistic phenomena. An ES is a combination of such constraints for the decision-making on pragmatic interpretations.In this research, pragmatic interpretations are first defined for each phenomenon, 2 (i.e. specific and generic) for ziji and 6 (definite, specific, distributive, collective, reciprocal and generic) for QNPs.Then more valid experts are explored via observation about phenomena in question in corresponding databases containing natural language (NL) materials, for example, the distribution of linguistic forms and expert values, features of antecedent, particular lexical items, distance between ziji or QNPs and their antecedents or co-referring expressions, syntactic position, tense, aspect, voice, number of modifier, co-indexation and text style, to name just a few.Next, proved to be effective in actual analyses, experts are to be evaluated and certain vote(s) are attached to them. Votes on each expert are added up to make a tally. In total, 8 experts of ziji and 23 of QNPs are identified. By abstracting required information by each expert, tallies are calculated to predict the optimal pragmatic interpretation in discourse. It turns out that our ESs can successfully predict the pragmatic interpretations of ziji (91%) and QNPs (80%) respectively in discourse.The ESs are embedded into HPSG framework at the end of each study, which is especially suitable mainly for three reasons. First, it incorporates information on different levels in one feature structure, which is consistent with the nature of an ES. Second, the lexicalist characteristic conforms to the fact that lexical items in Chinese include rich information which to a great extent determines syntactic and semantic structures (陆俭明2006). Third, the modularity of HPSG is highly compatible with its computational realization.At the end of this dissertation, we’ll propose a life cycle of linguistic expert system and a model of its computational implementation based on the two studies we conducted in this dissertation.更多还原