| Corpus,composed of a great deal of actual used language information,is a language database,specialized for language research,analysis and description.Corpus is built on the basis of randomly collecting people's actual used and representative real language material.Corpus,which is not only the comfortable tool to do quantitative analysis and research on language but also the technical development of linguistic research method,signifies the significant change in language research thought,making language research transit from traditional intuitive experience method to quantitative statistical approach and improving language research efficiency(Tan,2005:61-63).Multidimensional Analysis Model(Biber,1988:72),MDA in short,is initially applied to various research of spoken and written language.The method takes advantages of latent model reflected by massive language characteristics to discover and interpret systematic style variance and believes language characteristics co-occurrence model from the sense of statistics can reveal the deep interpersonal function of text.Subsequent research on language variance based on MDA method has proved its general applicability(Cao&Xiao,2015:5-9).MDA does not focus on single language feature but investigates co-occurring language characteristic which has certain statistical frequency,that is,the feature whose co-occurring frequency is higher than accidental frequency.Each co-occurring feature group constitutes a"dimension",in other words,a textual characteristic change continuum.Each dimension reflects the similar distribution of co-occurring feature in text,constructing text's partial shared meaning and function,therefore,analysis and interpretation on each dimension will reveal text global characteristics(Biber,2003;Csomay,2015).The author first extracted 250 English argumentative writings from SWECCL 2.0(Wen,2008)as a corpus,divided the corpus data into high-and low-score groups with 125 writings in each group according to the means of human rating scores and Juku online evaluating grades,then respectively input argumentative writings of the two groups into natural language processing software Coh-Metrix 3.0 and analyzed argumentative writings'local cohesion(e.g.,connectives),readability and syntactic features(e.g.,syntactic pattern density).On the basis of the data provided by Coh-Metrix 3.0 and SPSS 21.0,the study is aimed to explore the difference between high-and low-proficiency writings in terms of connectives,syntactic pattern density and readability by t-test,investigate the relationship between these textual features and writing score through correlation analysis,and the predictive powers of these textual features for writing level by means of regression analysis.The research protrudes the value of three textual features,i.e.connectives,readability and syntactic pattern density,in English argumentative writings,which is of some certain practical significance.For one thing,according to research result and data analysis,students can adjust and perfect their writing strategies,such as,using more temporal connectives,increasing the frequency of adverbial phrase,and enhancing writing's readability so as to achieve higher writing quality;for another thing,based on aforementioned aspects,teachers,engaged in teaching writing can provide students with more targeted instructions and assist them in improving argumentative writing proficiency.In the light of the statistical information in the present thesis,researchers who work on English writing can make further betterment of the present research and conduct research on other textual features of argumentative writings by means of Coh-Metrix. |
PDF Full Text Request