Research On The Working Model For Computational Stylistics

This paper works to propose and testify a working model for Computational Stylistics, not to construct the whole discipline completely, but expecting to reveal that it is the time to recognize the discipline status of Computational Stylistics at present through operation of the proposed model. Computational Stylistics has not been given an academic position as a discipline, although it has brought out many successful style-computation case studies.Based on these case studies, this paper tries to draw out a general working model. Furthermore, to testify the operationality of this model, the paper takes Katherine Mansfield's short stories for a specific case study: computation of affect flow through texts. The study adopts the methodologies of Computational Linguistics, Corpus Linguistics and Statistics.Computational Stylistics has been considered a branch of Computational Linguistics, and still has not had its independent discipline status. But this research goes through massive literary style computation case studies, and discovers that the Computational Stylistic studies have already surmounted the traditional definition of Computational Stylistics formerly recognized, and has already had its own computation model; therefore we hold that Computational Stylistics should win its independence from Computational Linguistics. For discipline construction, in terms of the object of study and goal, Computational Stylistics maintains consistence with the modern Stylistics, and in working pattern, it has the aid of Computational Linguistics, Corpus Linguistics and Statistics, and has formed a unique system of its own.The object of Computational Stylistic study is no longer marginal linguistic phenomena, but those at all levels of language; that is, it can work completely in accordance with the definition of Modern Stylistics to carry on language phenomenon computation. Its research goal is no longer to merely render special service in machine automatic reduction, namely Artificial Intelligence's realization, but to calculate the significance of the theme of text, with data collected during the operation process of computation.To examine the Computational Stylistics working model, this research takes affect flow in Mansfield's works as the object of study. The research uses the working pattern of computation to realize emotion computation, thus confirming the feasibility of the working pattern proposed.Emotion flow is not something new in the field of literary criticism and stylistic analysis. Affect, emotion, attitude are generally considered as synonyms and used for the interpretation of the theme and aesthetic effect of the text. But up till now, affect flow has not been regarded as a kind of grammatical structure, let alone being explored. Yet, just as L vi-Strauss takes structuralism into the analysis of myths and even takes the term of'mytheme', and his pupils take an effort to construct the grammar of myth, this paper tries to touch upon the structure of affect flow in literary texts.Affective computing is a difficult process of exploration, mainly because emotion is considered purely subjective awareness and response, and its computation—the objective data indicating emotion is very difficult to find echo in the heart of the reader, and when applied to the literary text, affect computing is more difficult to be accepted. But this study holds that since emotion is shown in its performance of objective discourse at the language level, certainly it has its objective foundation. To affirm these objective language phenomena will guarantee the feasibility of affective computing.Inspired by the Appraisal System, this study proposes that a text emotion is mainly decided by the emotion lexis, and the affect flow mainly formed by the emotion lexis distribution in the text, thus we want to explore: How is the literature emotion flow formed, whether it has much connection with the emotion lexis?Based on studies on Chinese commendatory and derogatory terms, on flow of emotion in the field of Stylistics, on semantic prosody by Corpus Linguistics, esp. on relevant studies concerning the attitude of text by Appraisal System, this research has gradually formed the hypothesis upon the mechanism of the literary language emotion flow, namely: The affect lexicon forms its own domain in discourse, and produces radiation to the neutral lexis within the domain. The emotion lexicon takes along in one's own domain the grading methods (graduation system, by Appraisal System), forms the domain of which the size varies, which fluctuates forward like waves, resulting in the emotional flow of discourse. The continuous emotion words of smiliar appeal combine to form the chain, the bigger emotional domain. The stop marker of the emotion domain is two adjacent emotion lexis with similar appeal.Based on this assumption, this research separates out some factors: affect lexis, stop marker of domain, the graduation system, and proposes a further hypothesis from this: we can conduct a discourse emotion computation by using these factors. It is discovered in this research that the emotion lexis may be divided into the positive lexis, the negative lexis and the neutral lexis. In neutral lexis some are especially active, and are easily infected with the appeal of emotional lexis. Matched with positive lexis, these active neutral words present the positive appeal, and matched with negative lexis, they present the negative appeal, but they sometimes maintain their neutrality, and present no appeal. These words in this research are named as the Neutral Sensitive Lexis. In the text, lexis with the sentimental appeal is called the neutral positive lexis or the neutral negative vocabulary. When collecting the neutral sensitive lexis, we discover that many words have a dual identity, which poses disturbance to the computation accuracy, and can only be revised at the manual work stage. Moreover some words may not explicitly determine alone their local emotional color, which can only be determined with the sentence as the unit. Such language phenomenon is unable at present to undergo machine's automatic identification, which can only be amended in the manual work phase.The various grading methods, under present technological conditions, we find, can not be subjected to the computing completely. The graduation is a supplementary means to emotion flow which only influences the emotion intensity, and which does not affect the emotion main key, so we have decided to leave the grading method for later study. Thus, in this research affect computation is a computation of the basic emotion flow or the emotion main key.The entire computational process is designed in line with the principle of from lexis to text, experiencing the stages of judgment and collection of affective lexis -- the affective lexis corpus building– affect flow schema chart drawing. Meanwhile we have designed two tests: one of which tests what differences there are between the researcher's individual judgment and similar readers'judgments, namely whether the researcher's judgment is representative among the reader community she stands in; and one of which tests the accuracy of emotion lexis corpus in its practical application.The first chapter is an introduction of the thesis. It introduces the reasons and operating processes of this research, the research goal and scope. Based on Corpus Linguistics and Statistics, it defines and limits the scope and capacity of the corpus taken in the study.The second chapter is the literature review (1). It introduces the development of Computational Stylistics, unfolding the computer software's processing from raw data to tagged data, revealing that the human being's unceasing demand for computer technology in carrying on stylistic analysis, from sole reliance on the statistical techniques for the literary style marker tagging to symbolization of human qualitative analysis and recognition for the annotation of raw data, thereby improving the precision of computation and its theory. Finally it brings out a working model for Computational Stylistics, and proposes the emotion flow computation model to testify the feasibility of Computational Stylistics model.The third chapter is the literature review (2). It talks about emotion flow-related linguistics research. First it points out that in the field of Stylistics, literary style study has not conducted an exhaustive research and given an explicit definition to emotion flow. This research's significance regarding the stylistic analysis is: it adds a new angle to stylistic analysis, and adopts the Corpus, Computational Linguistics and Statistics method. The Corpus Linguistics'research into the semantic prosody indicates that the emotion lexicon has an emotion radiation function upon its collocated neutral lexis. Systemic Functional Grammar studies the Interpersonal function of language, and the Appraisal System notes the emotion lexis'central function, proposes definitions concerning the domain of prosody, graduation system, stop marker of domain, but its research mainly carries on the style of news reporting; moreover some definitions of, for example, the domain stop marker, is not explained clear. This chapter has laid the foundation for the theoretical assumption of Affect Flow Structure theory.The fourth chapter is an elaboration on the hypothesis of Affect Flow Structure theory. Several factors which the hypothesis involves, like the neutral sensitive lexis, domain of prosody and stop marker of domain, graduation system and so on, are further studied.The fifth chapter is mainly the emotion lexis research. Discussions are centered upon classification of emotion, classification criterion of emotion lexis as well as emotion lexis collection method conducted in this study. According to Psychology, Chinese lexics, Appraisal System's research, the emotion is divided into positive and negative emotion, the emotion lexis into positive lexis, negative lexis, neutral sensitive lexis (including neutral positive lexis and neutral negative lexis). The confirmation standard is: all expressions that show recommendations, affection, respect and those which bring happiness, security and satisfaction, can be regarded as positive lexis. Those expressions that show detests, dislikes, despises and those which bring unhappiness, insecurity and dissatisfaction, are negative lexis. The confirmation basis is the dictionary meaning. The gathering process is divided into three phases. 1) According to dictionary explanation, judge lexis in Mansfield work one by one, identify and select the emotion lexis. 2) Let computer carry on the emotion lexis identification in texts, tag relevant words, and then give manual repair—according to the context, patch out inappropriate tags. 3) Collect finally the annotated lexis, and obtain Mansfield's emotion lexis corpus.The sixth chapter is an experimental study. It carries on analysis upon the relevance between the reader's subjectivity and the emotion lexis confirmation. In order to examine whether the researcher's confirmation is representative, it is designed that three similar readers read the same texts with the researcher. Then, the study contrasts the confirmed emotion lexis tagged by fellow readers in the reading process. The result shows that these readers'difference is not very great, which indicates that the researcher's confirmation is representative.Chapter 7 mainly describes drawing process of the emotional flow curve. Based on the emotion lexis confirmation, the study inspects the lexis in the emotion flow displayed at the discourse level. Through analysis of two examples, this study draws up the emotion flow schema chart. According to this research pattern, the computer software is designed to automatically take up the statistical work and computation of the text's emotion lexis and draw up the corresponding emotion flow schema chart.The eighth chapter studies emotion lexis distribution to forecast the text attitude. Based on affect flow schema charts, the study carries on the classification to the Mansfield's 42 novels, which shows that these schema charts can reveal the significance of the text theme and support the analysis and disclosure of the text theme, and proves that the emotion computation is helpful to the explanation of theme to a certain extent. Further, in accordance with the distribution value of the emotion lexis, the classification of a text emotion can be conducted and numerical standards set can be used to predict the Mansfield's fiction emotion category.The ninth chapter is an experimental study, examining the accuracy of Mansfield emotion lexis corpus in the actual utilization obtained. Materials used here are the random-selected novels in Mansfield's other two books, and Irving's and Lawrence's novel fragments. After machine automatic annotation of the data and manual modification, the study carries on statistics of the selected emotion lexis. The ANOVA and T-test indicate that the Mansfield emotion lexis corpus has the extremely high accuracy in her own works, and it has some errors in other two writers'works, which is not remarkable. This shows that there is some uniformity in different writers'use of emotion lexis. Perhaps the unanimous understanding and the lexis use may prove that the tentative plan to construct a literature emotion lexis corpus is to a certain extent applicable. Meanwhile the scalability of this method is proved proper: it may be worth a try to identify a text's emotion automatically, after annotating massive texts and building up a general emotion lexis corpus.Chapter 10 is the conclusion on the research. It concludes the stylistic features of Mansfield's, based on the working model proposed and meanwhile states the constructive implecation of this model. And it summaries the achievements made by the study, and points out the remaining issues and further research projects.If this article has contributed to related research, it is shown in the following aspects:It has proposed the discipline status of Computational Stylistics explicitly. At present domestic researches limit studies in the literary style form extraction done by Corpus Linguistics, and there is no explicit conclusion on the discipline definition and academic status of Computational Stylistics. Overseas researches regarding this are on a case verification stage, and have not given Computational Stylistics an independent discipline status. This research, on the basis of massive case studies of style, has put forward the discipline definition and its working model, and points out that Computational Stylistics accepts Stylistics theoretical guidance, and for the working model, absorbs factors from Computational Linguistics, Corpus Linguistics and Statistics.To testify the feasibility of Computational Stylistics working model, this thesis takes emotion computation as a case study. Based on the working model, this thesis proposes an emotion computation model. In the computing process, it makes clear the definition of emotion lexis and its method of classification, annotation and collection, and designs the computer software to carry on the description of emotion lexis distribution, and forms an emotion flow schema chart. The study adds a new angle for the stylistic analysis—the emotion lexis as well as the stylistic features reflected in their distribution at the discourse level, and has discovered in the actual computation process the author's'writing fingerprint'and the text's'emotion fingerprint'.In the computing process, by ushering in statistical methods, including those from experimental design to data collection, statistical modeling, data analysis, the researcher has examined relevant assumptions, and has realized that the theory and method of mathematical probability and the method are of great help to the proof of subjective assumptions.