A Preliminary Study Of Text Features And Evaluation Model Of Chinese L₂ Speech

The combination of computer technology and language testing is an irresistible trend for future researches as well as the emerging direction in the domain of second language acquisition. Currently, automatic scoring software based on computer system has been developed and applied abroad, but it’s still not often seen at home especially when it comes to oral proficiency scoring. In recent years, Chinese learning becomes more and more popular and there are more and more people participating in Chinese tests. Traditional manual scoring which is not only inefficient but also strongly subjective is still used in large-scale oral Chinese tests in China such as the HSK (advanced) oral test.This study aims to explore the core link in the automatic scoring system of oral Chinese proficiency, namely, to build an evaluation model. Complete evaluation model of oral proficiency involves two aspects, the voice and the transferred text, but due to the author’s limited ability and time, the paper only discusses the evaluation model from the text angle. The paper aims to solve mainly two problems. Firstly, the paper extracts a series of text features that can effectively measure oral Chinese proficiency and probes into the predictive ability of various text features for oral Chinese scores. Secondly, the paper preliminarily constructs a scoring model for oral Chinese proficiency based on the settlement of the first problem.The first chapter reviews domestic and foreign literatures on current oral proficiency scoring methods, the "Accuracy, Complexity and Fluency Analysis" of linguistic performance, automatic oral scoring system and Chinese scoring model. It focuses on SpeechRater, the oral proficiency automatic scoring system that has already been operated in some foreign countries. At present, the "Accuracy, Complexity and Fluency Analysis" is the main method to measure the linguistic performance of learners in the second language research field. Due to the characteristics of the spoken language text, and the limited time and energy of this study, the paper only studies the fluency and accuracy of the language.The second and third chapters are the extraction of text features and the indicator system to establish the oral proficiency evaluation system. The second chapter analyzes the accuracy of the sample from two perspectives, namely vocabulary and grammar. According to the classification made by Lu Jianji (1987,1994), the paper divides vocabulary errors into improper use of vocabulary, vocabulary missing, vocabulary redundancy and non-Chinese vocabulary; divides grammatical mistakes into omission, addition, misuse and disorder and then analyzes those errors and mistakes with texts combined. On that basis, this study extracts 20 quantitative indexes to measure accuracy. The third chapter defines language fluency and establishes 13 text features which reflect text fluency from two aspects, the time index and the language expression index respectively.The fourth chapter is the data analysis part of the paper. First of all, three Chinese teachers were invited to give marks to 40 texts from five aspects, namely voice, vocabulary, grammar, writing and the overall impression and then calculated the average score of each text. Then with the average scores as the dependent variables and the 33 text features as the independent variables, a regression equation containing 6 independent variables was concluded using the SPSS Multiple Stepwise Regression Method as the evaluation model of oral proficiency of this paper.The study believes the oral proficiency score of the samples can be well predicted by the 6 indexes entered in the regression equation. They are Average Length of Pause, Mean Length of Runs, Times of Repetition, Effectiveness of Correction, Total Errors of Function Words and Number of Inerrant Auxiliary Words, among which there are 4 fluency indexes and 2 accuracy indexes. And then the paper concludes the regression equation:the predicted value of oral proficiency score y =68.053-30.960* Average Length of Pause+2261* Mean Length of Runs--0.231* Times of Repetition+13.437* Effectiveness of Correction-0.222* Total Errors of Function Words+0.099* Number of Inerrant Auxiliary Words. In the end, relevant analysis results and the scatter diagram both indicate there’s a high correlation between the predicted value y and the average score, implying that this regression equation is able to relatively accurately predict oral proficiency scores.The fifth chapter presents the main conclusions of the paper. First of all, by comparing the 6 text features with conclusions made by English L2 researches and other oral Chinese L2 researches, the paper finds out there are overlapping as well as different indexes, which indicate that most indexes in the conclusion also have good predictive effect in other researches. Then the paper further explores the research paradigm of texts by interviewing three oral Chinese teachers, and as the interview results imply, the paper’s findings are basically consistent with practical teaching situation. In the end, the paper points out that oral language teaching needs to pay attention to some formal problems of language expression, e.g. pause, repetition and correction, attach importance to vocabulary teaching especially function words teaching, and points out the future direction of researches on oral proficiency automatic scoring.