The relationship of dictation errors to learner proficiency.

摘要

The present study used principles of error analysis within a psycholinguistic perspective in order to investigate the relationships of learner errors to measures of proficiency. The research design included two sets of variables to be compared. Such a comparison required the use of canonical correlation analysis, which was an innovation in the field of second language acquisition. An investigation of relationships between dictation errors (E1, phonologically convergent; E2 phonologically divergent; E3 omission) and TOEFL proficiency measures (TL, Listening Comprehension; TG Structure and Written Expression; TR, Vocabulary and Reading Comprehension) represents a first step in the building of a model of second language processing that incorporates findings from speech perception/recognition research and neural network models of learning.; 139 subjects participated in the study. The learners did one dictation per week for five weeks in a language lab equipped with individual access to listening stimuli. The results of a canonical correlation analysis of the proficiency variables indicates a suppressor variable relationship between the TR and the TL and TG sections. This means that performance in the TR section influences the relationship between error types and both TL and TG sections. As for the relationships between proficiency and error, there is a strong relationship between E3 and the TL and a secondary relationship between E1 and the TG section. Both are negative correlations meaning that people with many E3 errors tend to score poorly on the TL subtest, and vice versa, and that people with many E1 errors tend to score low on the TG subtest.; If stronger reading scores suppress scores in listening, the use of reading materials to assist listening activities and other mixed media approaches may be counterproductive when the improvement of listening skill is the goal. A learner who is provided with two channels for extracting meaning from a signal (audio/speech and visual/written text) will rely on whichever decoding skill is the strongest.