Capturing L2 Oral Proficiency with Composite CAF Measures: A Focus on Fluency
Despite an emphasis on oral communication in language courses, the resource-intensive nature of speaking tests hinders regular oral assessments. A possible solution is the development of a (semi-) automated scoring system, as the consistency of computers can complement human raters’ comprehensive judgments and increase efficiency in scoring. In search of objective and quantifiable variables, a number of studies have reported that some utterance-fluency variables (e.g., speech rate) are strongly correlated with overall L2 oral proficiency. While these studies focused on finding a single fluency variable as a predictor, given the complex nature of L2 oral proficiency, it is also important to examine a composite variable to predict learner proficiency. Consequently, this study investigated the relationship between complexity, accuracy, and fluency (CAF) variables and L2 oral proficiency. Utilizing audio samples from the Oral Proficiency Interview (OPI), a wellestablished speaking test by the American Council on Teaching Foreign Languages, this study analyzes spontaneous speech samples collected from 170 L2 Japanese learners with varied proficiency levels. The first part of the study investigated the relationship between CAF variables and learners’ oral proficiency. The results revealed that speech speed and complexity variables demonstrated strong correlations to the OPI levels, and moderately strong correlations were found for the variables in the following categories: speech quantity, pause, pause location (silent pause ratio within AS-unit), dysfluency (repeat ratio), and accuracy. The second part investigated an optimal composite measure that could best predict the OPI levels. A series of multiple regression analyses revealed that a combination of five measures (effective articulation rate, silent pause ratio, repeat ratio, syntactic complexity, and error-free AS-unit ratio) can predict 72.3% of the variance in the OPI levels. This regression model includes variables that correspond to three categories of fluency (speed, breakdown, and repair) and variables that represent CAF.