Second language writing classification system based on word-alignment distribution

The present paper introduces an automatic classification system for assisting second language(L2) writing evaluation. This system, which classifies sentences written by L2 learners as eithernative speaker-like or learner-like sentences, is constructed by machine learning algorithmsusing word-alignment distributions as classification features for detecting word-bywordtranslated expressions. The experimental results demonstrated that our classificationsystem provided adequate classification results with respect to (i) high classification accuracy,(ii) classification results reflecting the properties of L2 learner sentences, and (iii) identificationof learner-like expressions. These results suggest that our classification system can beused to evaluate L2 learner sentences.