Grammatical Error Detection in Automatic Essay Scoring and Feedback
MICHAEL GAMON, MARTIN CHODOROW, CLAUDIA LEACOCK, JOEL TETREAULT · 2013
Leacock and Chodorow (2003) found independent evidence that grammatical errors influence TOEFL scores. They reported a negative correlation between TOEFL essay scores and the grammatical errors that they contained. Leacock and Chodorow found that it is the variety of errors that affects the score, rather than the total error count. They also found that not all errors are equal. The most useful predictors of lower scores were violations of a rule of syntax, such as subject-verb agreement (my friend meet this guy), ill-formed modal verbs (People would said that . . .), and determiner-noun agreement (this things). Somewhat less reliable predictors were commonly confused words (homophones or near homophones such as their, there, and they’re) and confusion of a verb with its nominal form (I will success). Less costly errors involved pronoun use, missing commas, and apostrophes. Some errors appeared to have no effect on the score. For these, Leacock and Chodorow conjectured that the errors could likely be introduced when editing the essay (e.g., a the book or straighten you desk) and that the raters allow for sloppy proofreading.