Comparative Study on the Methodologies of Analyzing Textual Genres and Styles by Means of Multivariate Analysis

多元分析，如因子分析，长期以来一直被用于分析语料库数据。B.Biber(1988)等人是语篇类型学中运用语域理论的典型代表，并似乎成功地解释了语域的差异。然而，其他多变量方法，如主成分分析(PCA)和列联表量化(QCT)也被用于或多或少相同的目的。伯罗斯(等人)将主成分分析用于简·奥斯汀小说的文体研究，西格利(1998年)利用主成分分析提出了不同文本类型的形式指数。本研究的研究者之一Tabata(1995)等人也将其用于狄更斯文本的文体学研究。这些研究似乎在分析文本数据方面也取得了成功，Nakamura(2002)等人也是如此，他们利用QCT来根据各种语言f…的分布来确定语料库的结构用QCT对程度副词、频率副词和方式副词在不同语篇类别中的分布进行的初步研究表明，方式副词的表现与其他语类不同：程度副词和频率副词的分布主要归因于口语和书面语篇的二分法；想象语篇是解释方式副词行为的主要因素，语域的二分法被证明是次要因素。在本研究中，我们将进一步使用主成分分析和定量关联分析来考察BNC世界版中不同语篇类别中的方式副词的行为。基于列的主成分分析在文本领域的分类中是有效的，但副词的分布不能有效地将它们分成有意义的组。行分析不能提取因素来产生有意义的解释。Tabata还以狄更斯的小说和小品小说为研究对象，进行了以下几个方面的研究：1)语篇类别分布的对应分析(Anacor：原则上与QCT相同)；2)狄更斯小说对话中最常见的30-60个词类的主成分分析；3)1246种副词的Anacor分析。这些研究揭示了狄更斯文本在不同文本类别之间的差异、随时间的变化、由于社会动态而导致的小说话语中意识形态的分化以及正式和口语化风格的差异。本研究的主要目的是对方式副词进行进一步的分析，发现Anacor的效果要好于PCA。综上所述，为了选择一个适合进行多变量分析的变量，应该考虑变量的类型，如变量的类型，即定量的还是定性的，原始分数是否归一化，变量的数量等。较少

Multivariate analysis such as Factor Analysis has long been used in analyzing corpus data. B. Biber (1988 and others) are the typical examples of its use in text typology and seem to have been successful in explaining the differences of registers. However, other multivariate methods like Principal Component Analysis (PCA) and Quantification of Contingency Table (QCT) have also been employed for more or less the same purposes. Burrows (1989 and others) used PCA for stylistic investigations of Jane Austin's novels, and Sigley (1998) used it in proposing a formality index of different text types. One of the investigators of the present research, Tabata (1995 and others), also used it in investigating stylistics of Dickens' texts. These studies also seem to have been successful in their ways of analyzing textual data The same can be said of Nakamura (2002 and others) which made use of QCT in determining the structures of corpora based upon the distributions of various kinds of linguistic f … More eatures or items.Pilot studies by means of QCT concerning the distributions of degree, frequency and manner adverbs across different text categories of the BNC Sampler revealed that manner adverbs behaved differently from the others : the distributions of degree and frequency adverbs were mainly ascribed to the dichotomy of spoken vs. written texts ; imaginative texts played the main role in explaining the behaviors of manner adverbs and the dichotomy of registers turned out to be a secondary factor. In the present research, the behaviors of ?ly manner adverbs across different text categories in the BNC World Edition will be further examined both by PCA and QCT. Column-wise Analysis of PCA turned out to be effective in classifying text domains but the distribution of adverbs are not so effective in separating them into meaningful groups. Row-wise analysis could not extract factors to produce meaningful interpretations. In contrast, QCT provided quite reasonable interpretation for both the distribution of textual domains and that of adverbs with the primary factor being narrative vs. expository style and the secondary factor being informal vs. formal style.Tabata also used Dickens's novels and sketches and conducted several studies : 1) correspondence analysis (ANACOR : same as QCT in principle) of word-class distributions across texts, 2) PCA of 30-60 most common word-types in the dialogue of Dickens's fiction and 3) ANACOR of 1246 types of ' ?ly' adverbs. These studies reveal various aspects of Dickens's texts variation across text categories, variation over time, differentiation of ideolects in fictional discourse due to social dynamics and formal vs. colloquial styles. Further analysis of manner adverbs, the main interest of the present research, revealed that ANACOR generally works better than PCA.In conclusion, in order to choose an appropriate for multivariate analysis, the types of data in question such as types of variable, I.e., quantitative or qualitative, whether raw scores are normalized or not, the number of variables, etc should be taken into consideration. Less