Partial and Synchronized Caption Generation to Develop Second Language Listening Skill

Abstract

Captioning is widely used by second language learners as an assistive tool for listening. However, the use of captions often leads to word-by-word decoding and over-reliance on reading skill rather than improving listening skill. With the purpose of encouraging the learners to listen to the audio instead of merely reading the text, the study introduces a novel technique of captioning, partial and synchronized, as an alternative listening tool for language learners. Using TED talks as a medium for training listening skill, the system employs the ASR technology to synchronize the text to the speech. Then, the system uses the learner’s proficiency level to generate partial captions based on three features that impair comprehension: speech rate, word frequency and specificity. To evaluate the system, the performance of Kyoto University students in two CALL classes was assessed by a listening comprehension test on TED talks under three conditions: no caption, full caption and the partial-and-synchronized caption. Results revealed that while reducing the textual density of captions to less than 30%, the proposed method realizes comprehension performance as well as full caption condition. Besides, it performs better than other conditions on new segments of the video without captions.