Classification and Clustering English Writing Errors Based on Native Language

It is important for language learners to determine and reflect on their writing errors in order to overcome weaknesses. Each language learner has their own unique writing error characteristics and therefore has different learning needs. In this paper, we analyze the writing errors of foreign language learners on the language learning SNS website Lang-8 to investigate the characteristics of errors by native language. 142,465 sentences were collected from Lang-8 for analysis. For each native language, the predicted scores of 15 error categories from SVM machine learning models are used as a vector representation of each sentence. These score vectors are then clustered to determine error co-occurrence within the same sentence. The results were then analyzed to determine the error characteristics of different native languages.