Predicting judged fluency of consecutive interpreting from acoustic measures: Implications for automatic assessment and pedagogy.

Abstract

The present study presents an experiment to investigate 1) whether judged fluency correlates with judged accuracy of consecutive interpreting performance; and 2) whether and how judged fluency can be predicted on the basis of acoustic measures such as automatically calculated temporal measures of speech quality. In the experiment, ten raters judged six major quality measures related to accuracy and fluency of the consecutive interpreting performance (in this study from Chinese into English) recorded from seven BA trainees and five MA trainees based on the same recorded source speech. Six consecutive interpretations per participants were recorded, two of which were analyzed acoustically by means of the speech analysis tool PRAAT. The pauses detected by PRAAT and the disfluencies transcribed and labeled manually were automatically counted by the text analysis computer program of AWK, based on which 12 acoustic measures of fluency were calculated. The results show that 1) there is a strong positive correlation between judged accuracy and judged fluency in consecutive interpreting; 2) there are strong correlations between judged fluency and objective acoustic measures of fluency in consecutive interpreting; and 3) effective speech rate appears to be the best predictor of judged fluency in consecutive interpreting. The other important determinants of judged interpreting fluency are number of filled pauses, articulation rate, and mean length of pause. The results have practical implications for both developing an automatic assessment tool of fluency facilitating consecutive interpreting testing as well as for consecutive interpreting pedagogy.