Computer-assisted detection of 90% of EFL student errors

Abstract

ABSTRACTSoftware can facilitate English as a Foreign Language (EFL) students’ self-correction of their free-form writing by detecting errors; this article examines the proportion of errors which software can detect. A corpus of 13,644 words of written English was created, comprising 90 compositions written by Spanish-speaking students at levels A2-B2 (inclusive) of the Common European Framework. A total of 1,310 language errors were detected by the researcher. It was found that approximately 21% of these errors were spelling errors. A further 58% were characterised as either two-word phrases (45%), three-word phrases (9%), or four- and five-word phrases (4%) which are either absent from or rare in a large corpus of English which is known to be correct. The nature of software which can detect such words and phrases and bring them to students’ attention with a view to self-correction is briefly described. Of the remaining 21% of errors not detected by such software, most were found to be either errors of tense (7%), misuse of false friends (4%) or problems with determiners (3%). Again, software which can help students detect and correct such errors is outlined. The limitations and pedagogical significance of the research are then briefly discussed.