通过预测学习者理解度比较英语不同变体的清晰度--一种语音实验方法

Comparing the Intelligibility of Different Varieties of English Through Predicting Learners Comprehension:A Phonetic Experimental Approach

本实验研究着重于使用自动语音系统测量英语发音的清晰度,并在该可行性实验研究中进行系统的测试。实验挑选18名来自不同国家的受试者,代表6种类型的英语(中国、越南、埃及、印度、南非和菲律宾),这些受试者均能代表一系列语音的清晰程度。实验中使用了两种不同的方式——转录和无意义——测试和衡量英语发音的清晰度。其中基于超分段测量的口语能力评分的计算机自动模式被用来预测清晰度得分。人类评估得分和计算机预测得分之间的皮尔森相关性为0.743,无意义得分为0.819。转录得分的可靠内部评分Cronbach’s alpha为0.943,无意义可解性得分为0.945。根据可理解性测度的类型,计算机利用不同的超分段测量来预测得分并且使用了11个无意义理解度的指标和8个衡量转录分数的指标,两种结构方式显示出只有两类特性是共性的。这些计算机实验模式的分析和结果可为二语研究者在今后衡量发音可理解性的研究中提供不同的视角。

In this paper, we focus on the need for measuring the intelligibility of English pronunciation using an automated speech system, and the system proceeded in this feasibility study was tested with 18 speakers coming from six countries representing six types of English(China, Vietnam, Egypt, India, South Africa and the Philippines). Those test candidates were selected carefully to stand for a series of intelligibility, two different approaches, transcription and nonsense, were utilized to test and measure their intelligibility. An automated computer pattern developed for speaking proficiency scoring based on suprasegmental measures was set to predict intelligibility scores. The Pearson’s correlation was 0.743 for transcription construct and 0.819 for the nonsense construct between the human assessed and computer predicted scores. The reliable inter-rater Cronbach’s alpha for the transcription scores was 0.943 and 0.945 for the nonsense intelligibility scores. Basing on the type of intelligibility measure, the computer utilized different suprasegmental measures to predict the score. The computer used 11 measures for the nonsense intelligibility score and eight for the transcription score. Only two features were common to both constructs. These analyses and results of this computer experimental pattern can provide researchers of L2 different perspectives of measuring intelligibility in research afterwards.