A Scoring Method for Second Language Writing Based on Word Alignment

Automatic evaluation reduces the burden on second language (L2) teachers when checking for errors made by L2 learners. One evaluation approach is to automatically classify L2 learner sentences into either natural or unnatural sentences. However, because the distinction between natural and unnatural sentences is complex, such a binary classification cannot accurately reflect the naturalness of L2 learner sentences. To solve this problem, we developed a method that assigns scores (continuous values) to L2 learner sentences by examining the degree to which L2 learners use erroneous expressions arising from unnatural literal translations of phrases from a L2 learner's first language. Experimental results suggest that our scoring method is effective for the automatic evaluation of L2 learner sentences.