EUROCALL 2014

Full Program »

Automatic Feedback on Spoken Dutch of Low-Educated Learners: An ASR-based CALL study

For controlled research on second language (L2) learning of oral proficiency, we developed a computer assisted language learning (CALL) system that makes use of Automatic Speech Recognition (ASR) technology to provide automatic corrective feedback (CF) on grammar. Practicing speaking is important for developing oral proficiency in the L2. Since opportunities for L2 speaking practice are often limited, CALL systems can be a valuable supplement to traditional classes because they provide learners with individual speaking practice, anxiety free, and at times and lengths that are convenient for them. In addition to offering speaking practice, these systems can provide CF which can contribute to L2 proficiency. Though this claim is the consensus view, it is not uncontroversial, and findings are not consistent across studies. Effectiveness of CF is found to be influenced by many variables, e.g. individual differences, learning context, type of CF and type of error that CF targets. Systematic research into these variables requires a high level of control, which has proven difficult for studies on CF on oral proficiency. Through automatic CF provided with ASR, we can control the learner input and monitor learner behaviour. Knowledge of how CF-variables interact with learner variables could further SLA theory and have implications for (language) pedagogy.
In this paper we discuss the experiments with our CALL system on the influence of level of education and CF on language learning. Low-educated learners do not receive much attention in the literature. As a consequence, there is a relatively large population of language learners of which little is known with respect to efficient pedagogy and learning preferences.
We present our experiment with low-educated learners, and compare the results with those obtained in earlier experiments with an identical setup in which high-educated learners were involved. Since the learners practice individually with the computer, the input for the learners is consistent, and system user-interactions can be compared.
Learners in our system practice speaking Dutch, and receive automatic CF on word-order. In the exercises, learners watch short video clips (approximately 35 s), and then answer questions about the clip by arranging a given selection of 3 -5 segments of a sentence into a grammatically correct sentence. The learners are randomly assigned to either the CF-condition, where they receive automatic CF on their utterance, or to the NOCF-condition, where they are given the chance to re-record or move to the next question. In two pre- and posttests, we measure spoken production and receptive knowledge of the target feature (Dutch V2). We implement questionnaires during and after the experiment to gauge learner appreciation of the practice session.
Besides presenting results on changes in L2 proficiency, as measured in the pre- and posttests, we also discuss learner behavior as recorded by the system, and as observed by the experimenters. We show that we successfully designed an experiment setup that can be used by high and low-educated learners. In addition, we find that practice with our system results in increased speaking proficiency, where the extent of learning is related to individual differences.

Author(s):

Bart Penning de Vries    
Centre for Language and Speech Technology
Radboud University Nijmegen
Netherlands

Bart Penning de Vries has an MA in Linguistics from the University of Utrecht, and is currently a doctoral student at the Radboud University Nijmegen, Centre of Language and Speech Technology. His research interests are speech technology for CALL and second language acquisition.

Catia Cucchiarini    
Centre for Language and Speech Technology
Radboud University Nijmegen
Netherlands

Dr. Catia Cucchiarini is a senior researcher at the Centre for Language and Speech Technology of the Radboud University Nijmegen. Her research activities address speech processing, computer assisted language learning (CALL), and the application of ASR to language learning and testing.

Stephen Bodnar    

Stephen Bodnar is a doctoral student with the Centre for Language and Speech Technology at Radboud University Nijmegen in the Netherlands. His research interests include L2 motivation, speech-interactive CALL systems, and student modelling.

Helmer Strik    
Centre for Language Studies
Radboud Universiteit Nijmegen

Dr. Helmer Strik is associate professor in Speech Science and Technology at the Radboud University Nijmegen. His fields of expertise include phonetics, speech production, automatic speech recognition, spoken dialogue systems, e-learning and e-health.

Roeland van Hout    
Centre for Language Studies
Radboud Universiteit Nijmegen

Roeland van Hout is professor in applied linguistics and variationist linguistics at the Radboud University Nijmegen. He publishes in the fields of sociolinguitics, dialectology and second language acquisition and has a special interest in research methodology and statistics.

 

Powered by OpenConf®
Copyright ©2002-2013 Zakon Group LLC