Development of an Automatic Evaluation System of Reading-Aloud and Shadowing Performances by Japanese EFL Learners Using the Latest Speech Technology

虽然影子学习被认为能有效地提高英语学习者的听说能力，但评估影子学习的表现既困难又耗时。为了解决这一问题，利用最新的语音信息技术GOP(Good Of Prontation)开发了一种自动评分系统。计算机在音素的基础上自动将学习者的阴影语音与基于原始文本的声控TIC模型进行比较，并计算指示学习者的阴影语音与原始语音有多近的数字分数。为了验证这种自动评估的有效性，我们进行了一项实验。根据托业考试分数，熟练程度在200到900之间的日本英语学习者被要求在听英语母语人士以每分钟约140个单词的速度阅读以前从未出现过的段落时，将他们的影子记录到计算机中。然后，将自动系统获得的分数与资深英语教师手动评分的分数进行比较。两者之间有很高的相关性(r=0.85)，因此该系统在评估阴影性能时具有显著的有效性，并且节省了大量的时间和精力。由于学习者的自动成绩也与他们的托业成绩高度相关(r=0.82)，影子成绩可以预测他们的总体水平分数。在回归模型的基础上，该系统可以预测和呈现托业考试的成绩水平。因此，这一自动评估系统使英语学习者能够客观地了解自己的跟踪能力，并掌握自己的熟练程度。

Although shadowing is considered to be effective in improving EFL learners’ listening and speaking skills, it is both difficult and time consuming to assess shadowing performance. To solve this problem, an automatic assessing system has been newly developed using the latest speech information technology: GOP (Goodness of Pronunciation). The computer automatically compares the learner’s shadowed speech with an acous tic model based on the original text on a phoneme basis and computes numerical scores which indicate how close the learner’s shadowed speech is to the original. To confirm the validity of this automatic assessment, an experiment was conducted. Japanese EFL learners with proficiency levels ranging from 200s to 900s measured by TOEIC scores were requested to record their shadowing into the computer while listening to passages, which had been never presented before, read by a native speaker of English at a speed of approximately 140 words per minute. Then the scores obtained by the automatic system were compared to those rated manually by veteran EFL instructors. A high correlation (r=0.85) was observed between the two; thus this system is significantly valid and also saves much time and energy in assessing shadowing performance. Since the automatic scores of the learners were also highly correlated with their TOEIC scores (r=0.82), the shadowing performance can predict their overall proficiency scores. Based on the regression model, this system can predict and present TOEIC score levels. Therefore this automatic assessing system enables EFL learners to understand objectively how well they have conducted shadowing and also grasp their own proficiency levels.