UM-Checker: A Hybrid System for English Grammatical Error Correction

This paper describes the NLP 2 CT Grammatical Error Detection and Correction system for the CoNLL 2013 shared task, with a focus on the errors of article or determiner (ArtOrDet), noun number (Nn), preposition (Prep), verb form (Vform) and subject-verb agreement (SVA). A hybrid model is adopted for this special task. The process starts with spellchecking as a preprocessing step to correct any possible erroneous word. We used a Maximum Entropy classifier together with manually rule-based filters to detect the grammatical errors in English. A language model based on the Google N-gram corpus was employed to select the best correction candidate from a confusion matrix. We also explored a graphbased label propagation approach to overcome the sparsity problem in training the model. Finally, a number of deterministic rules were used to increase the precision and recall. The proposed model was evaluated on the test set consisting of 50 essays and with about 500 words in each essay. Our system achieves the 5 th and 3 rd F1 scores on official test set among all 17 participating teams based on goldstandard edits before and after revision, respectively.