Automated Measures of Lexical Sophistication: Predicting Proficiency in an Integrated Academic Writing Task

Background. Advances in automated analyses of written discourse have made available a wide range of indices that can be used to better understand linguistic features present in language users’ discourse and the relationships these metrics hold with human raters’ assessments of writing. Purpose. The present study extends previous research in this area by using the TAALES 2.2 software application to automatically extract 484 single and multi-word metrics of lexical sophistication to examine their relationship with differences in assessed L2 English writing proficiency. Methods. Using a graded corpus of timed, integrated essays from a major academic English language test, correlations and multiple regressions were used to identify specific metrics that best predict L2 English writing proficiency scores. Results. The most parsimonious regression model yielded four-predictor variables, with total word count, orthographic neighborhood frequency, lexical decision time, and word naming response time accounting for 36% of total explained variance. Implications. Results emphasize the importance of writing fluency (by way of total word count) in assessments of this kind. Thus, learners looking to improve writing proficiency may find benefit from writing activities aimed at increasing speed of production. Furthermore, despite a substantial amount of variance explained by the final regression model, findings suggest the need for a wider range of metrics that tap into additional aspects of writing proficiency.