Effective feedback on L2 pronunciation in ASR-based CALL

Computer Assisted Language Learning (CALL) has now established itself as a prolific area whose advantages are well-known to educators. Yet, many authors lament the lack of a reliable integrated conceptual framework linking technology advances and second language acquisition research within which effective materials can be designed [1],[2]. The CALL world has recently witnessed a flourishing of software applications among which Automatic Speech Recognition (ASR) is gaining growing importance. The reasons for this popularity lie in the opportunities this technology offers for practising oral skills and addressing pronunciation problems, two areas that are hard to improve within traditional class-based settings. ASR-based CALL systems appear to be particularly suited for pronunciation teaching as they allow evaluation of the learner’s speech and provide appropriate, individual feedback in real-time. However, given the lack of guidelines for CALL design, most courseware products often do not provide adequate guidance to the learner [3],[4]. Moreover, owing to the limitations in the state-of-the-art technology, all ASR systems will at times generate errors [4]. The main objective of our research is to study how the frequency and seriousness of feedback errors affect learning. The domain in which we work is the acquisition of pronunciation in Dutch as L2 by adults with different language backgrounds. In our study we will use a Dutch language multimedia course, Nieuwe Buren, to which we will add a speech recognition module previously developed at our department, which is able to recognize and score disfluent non-native speech [5]. Automatic pronunciation evaluation and feedback will focus on segmental and supra-segmental aspects. The system’s functionality will be first tested on a group of experts and subsequently on two experimental groups of different proficiency levels. After three months the pronunciation performance of these groups will be compared to that of two control groups who used the original version of Nieuwe Buren without immediate feedback. This will allow us to establish whether automatic immediate feedback does indeed lead to global improvement in L2 pronunciation. A more detailed analysis of individual results within the experimental groups will then provide insight into the specific effects of different feedback errors generated by the system and their impact on learning. Students and teachers will be finally required to complete a questionnaire meant to establish the feedback system userfriendliness, comprehensibility and adequacy (the control group will, by contrast, report on learning without real-time feedback). The experiment and the tests will be subsequently repeated using an improved version of the system which also takes into account newly emerged needs. In this paper, we first provide an overview of the literature on second language pronunciation teaching and learning in order to derive some general guidelines for effective teaching. Second, we present an appraisal of available CALL systems for L2 pronunciation training with a view to identifying pros and cons. Finally, we describe the choices made for our research on the effects of erroneous automatic feedback on pronunciation.