Quantitative assessment of second language learners' fluency: an automatic approach

This paper describes an experiment aimed at determining whether native and non-native speakers of Dutch significantly differ on a number of quantitative measures related to fluency and whether these measures can be successfully employed to predict fluency scores. Read speech of 20 native and 60 non-native speakers of Dutch was scored for fluency by nine experts and was then analyzed by means of an automatic speech recognizer in order to calculate nine quantitative measures of speech quality that are known to be related to perceived fluency. The results show that the natives’ scores on the fluency ratings and on the quantitative measures significantly differ from those of the non-natives, with the native speakers being considered more fluent. Furthermore, it appears that quantitative variables such as rate of speech, phonation-time ratio, number of pauses, and mean length of runs are able to predict fluency scores with a high degree of accuracy.