Corpus-Based Error Analysis of Uzbek EFL Learners’ Academic Writing
Abstract
English academic writing is a frequent source of difficulty for Uzbek learners of English as Foreign Language, particularly regarding learning grammar, vocabulary, and punctuation. For empirical evidence on these difficulties, we approached this issue in a computational way by constructing a small learner corpus consisting of 40 IELTS Academic Writing essays (about 9,000 words) from undergraduate students at one of the Uzbek higher educational institutions. The corpus was converted to digital form and incorporated into Sketch Engine corpus analysis platform, and all texts were manually annotated using a 13-category error-tagging system covering spelling, punctuation, articles, word choice, morphology, and syntax. Our quantitative assessments found most frequent errors and a comparison of error frequencies between first- and last-year undergraduate learners. A gender-based assessment found female learners to have averaged fewer errors, although this was controlled by male learners’ preference for more challenging topics for their writings. In this work, we propose a novel dataset of written learner corpus of Uzbek essays and present experimental results of our computational approach into this dataset analysis.