Revising parameters for predicting L2 speech fluency and proficiency

Abstract

The aim of the study was to investigate whether<br/>integrating parameters based on pause location<br/>improve the prediction of fluency and proficiency<br/>in L2 Finnish monologic speech. By doing so,<br/>multiple linear regression models were fitted using<br/>two data sets containing L2 Finnish speech and<br/>expert assessments of fluency and oral proficiency.<br/>Separate models were derived for fluency and<br/>proficiency using combined data as well as the<br/>two separate data sets. The comparison of the<br/>models indicate that pause-by-location parameters<br/>can improve the prediction of L2 fluency and<br/>proficiency, but the relevant parameters and their<br/>significance in the regression models depend on the<br/>speech data. Parameters with low incidence work<br/>only in longer speech samples, while parameters<br/>with frequent occurrence can be used even in<br/>shorter samples. The results have implications<br/>for improving automatic assessment of L2 speech<br/>especially in low-resource languages.