Practice Talk: “Predicting CEFR proficiency level using grammatical criterial features” by Susie Kim


Join SLS student Susie Kim, who will be giving a practice talk for a corpus linguistics conference.
Friday, September 14, Wells B243 from 3-4pm
Title: Predicting CEFR proficiency level using grammatical criterial features
In a previous study using a Korean EFL corpus, I investigated the use of the grammatical criterial features that have been identified as characteristic of the Common European Framework of Reference (CEFR) levels by Hawkins and Filipović (2012). Research on the use of such level-specific lexico-grammatical features has largely been descriptive. This study contributes to better understanding the relationship between learner language and CEFR proficiency levels by empirically examining how well the use of the set of grammatical features predict learners’ proficiency. The current analysis included 6,042 narrative and argumentative essays written by 3,021 learners. Each learner’s performance had been holistically evaluated on the CEFR scale, ranging from A2 to B2. With the data from the previous study where I examined the occurrences of ten A2-, fifteen B1-, and ten B2-level grammatical criterial features, I tabulated how many different types of the level-specific features each learner used and the average normalized frequencies of the occurrences. In short, the range and the frequencies of these level-specific features adopted by each learner were used to investigate how well these measures predict writing quality.
An ordinal logistic regression analysis was performed with six predictors: range and frequencies of the A2, B1, and B2 features. All six predictors were statistically significant in predicting the proficiency level, and the model showed a good discrimination among the proficiency groups (R2 = .31). The findings indicate that the frequency as well as the diversity in the use of the grammatical features serve as important predictors in determining the proficiency level. The results have implications for CEFR level descriptors and assessment using the CEFR scale.