Detecting English Grammatical Errors based on Machine Translation

Detecting English Grammatical Errors based on Machine Translation

Many people are learning English as a second or foreign language, and there are estimated 375 million English as a Second Language (ESL) and 750 million English as a Foreign Language (EFL) learners around the world according to Graddol (2006). Evidently, automatic grammar checkers are much needed to help learners improve their writing. However, typical English proofreading tools do not target specifically the most common errors made by second language learners. The grammar checkers available in popular word processors have been developed with a focus on native speaker errors such as subject-verb agreement and pronoun reference. Therefore, these word processors (e.g., Microsoft Word) often offer little or no help with common errors causing problems for English learners. Grammatical Error Detection (GED) for language learners has been an area of active research. GED involves pinpointing some words in a given sentence as grammatically erroneous and possibly offering correction. Common errors in learners'writing include missing, unnecessary, and misuse of articles, prepositions, noun number, and verb form. Recently, the state-of-the-art research on GED has been surveyed by Leacock et al. (2010). In our work, we address serial errors in English learners'writing related to the proposition and verb form, an aspect that has not been dealt with in most GED research. We also consider the issues of broadening the training data for better coverage, and coping with data sparseness when unseen events happen.