Generalization of Computer Assisted Prosody Training: Quantitative and Qualitative Findings 论文

摘要

Two experiments investigated the effectiveness of computer-assisted prosody training, its generalization to novel sentences and segmental accuracy, and the relationship between prosodic and lexical information in long-term memory.Experiment 1, using a pretest-posttest design, provided native English-speaking learners of French with 3 weeks of training focused on prosody using a real-time computerized pitch display.Multiple exemplars produced by native speakers (NSs) of French and stored on hard disk provided training feedback.Learners' recorded pre-and posttest productions were presented to NSs for evaluation in two conditions: filtered (unintelligible segmental information) and unfiltered.Ratings using 7-point scales for the prosody and segmental accuracy of unfiltered samples revealed significant improvement in prosody with generalization to segmental production and novel sentences.Comparison of prosody ratings for filtered and unfiltered samples revealed some segmental influence on the pretest ratings of prosody.In Experiment 2, involving a memory recall task using filtered stimuli of reduced intelligibility, learners identified the exact lexical content of an average of 80% of the training sentences based on prosodic cues consistent with exemplar-based learning models.Questionnaire responses indicated a greater awareness of the various aspects of speech and increased confidence in producing another language.Many factors may influence a native speaker's judgment of a second or foreign language (FL) learner's accent including suprasegmental features such as stress, rhythm, and intonation (Munro, 1995).These features began to draw the greatest attention from teachers and materials developers with the advent of the discourse-level focus of the communicative approach to language teaching and remained the principal focus throughout the 1980s (e.g., Morley, 1991;Pennington & Richards, 1986).However, in recent years, the field of pronunciation teaching appears to have adopted a more balanced viewpoint with regard to the importance of both the segmental and suprasegmental aspects of language (e.g., Celce-Murcia, Brinton, & Goodwin, 1996;Derwing, Munro, & Wiebe, 1998).Perhaps, then, an ideal training tool is one that can produce a significant improvement in both levels of the spoken language with evidence of generalization to novel stimuli.The current study investigated such a tool in the form of computer-assisted training with visual feedback.The study was divided into Experiments 1 and 2. Experiment 1 focused on both quantitative and qualitative aspects of the acquisition of French prosody by native speakers (NS) of American English (AE) using computer-assisted training that permits visual display of pitch contours in real time.This experiment was designed to investigate (a) the effects of such training when exemplars from native speakers (NSs) of French were used as feedback rather than models to imitate, (b) the extent to which the training would generalize to improvement in segmental production and to both the prosodic and segmental features of novel sentences, and (c) the potential influence (positive or negative) of segmental quality on the NS ratings of prosody by comparing learners' recorded productions with filtered versions that render the segmental information unintelligible while preserving the prosodic information.In addition to the quantitative findings, observation notes I made during training and the responses to participants' anonymous questionnaires following their training program provided qualitative information