Automated Detection of Usage Errors in non-native English Writing

In an investigation of the use of a novelty detection algorithm for identifying inappropriate word combinations in a raw English corpus, we employ an unsupervised detection algorithm based on the one- class support vector machines (OC-SVMs) and extract sentences containing word sequences whose frequency of appearance is significantly low in native English writing. Combined with n-gram language models and document categorization techniques, the OC-SVM classifier assigns given sentences into two different groups; the sentences containing errors and those without errors. Accuracies are 79.30 % with bigram model, 86.63 % with trigram model, and 34.34 % with four-gram model.