Reading aloud is difficult for L2 learners; they must decode and interpret written information and almost simultaneously produce sounds with accurate pronunciation, rhythm and intonation suitable to the text. However, repeatedly reading aloud the same passage enables learners to enhance their automatized processing and produce more fluent performances.
Coefficient of Variance (CV), which indicates how learners' language processing is automatized, is often used in recent research focusing on reaction time (RT), automaticity, fluency, etc. CV, calculated by variance of RT divided by mean RT, decreases as learners' proficiency levels become higher, because their language processing changes from controlled to more automatized and variance decreases.
A new CALL system has been developed in the present study which can assess reading aloud both in terms of accuracy and fluency. The former is evaluated by using one of the latest speech information technologies named GOP (Goodness of Pronunciation). The latter is assessed by utterance rate: the number of syllables spoken per second. This is almost the first trial to develop a CALL system evaluating the degree of automaticity of reading aloud as well as accuracy, fluency, and content comprehension during repetition.
Japanese EFL learners, classified into high, middle and low groups, were requested to read aloud a passage five times that they had never read before and answer multiple-choice questions on the content of the passage. The time needed to read aloud each individual sentence was regarded as RT. CV was calculated by the variance of RT divided by the mean of RT.
The statistical analysis revealed that RTs and CVs across the three groups significantly decreased as repetition time increased and that RTs and CVs were observed to be smaller as the proficiency levels went up. These results are consistent with prior research. Thus the validity of the developed system was confirmed.
