Classification of English language learner writing errors using a parallel corpus with SVM

Abstract

In order to overcome mistakes, learners need feedback to prompt reflection on their errors. This is a particularly important issue in education systems as the system effectiveness in finding errors or mistakes could have an impact on learning. Finding errors is essential to providing appropriate guidance in order for learners to overcome their flaws. Traditionally the task of finding errors in writing takes time and effort. The authors of this paper have a long-term research goal of creating tools for learners, especially autonomous learners, to enable them to be more aware of their errors and provide a way to reflect on the errors. As a part of this research, we propose the use of a classifier to automatically analyse and determine the errors in foreign language writing. For the experiment in this paper, we collected random sentences from the Lang-8 website that had been written by foreign language learners. Using predefined error categories, we manually classified the sentences to use as machine learning training data. This was then used to train a classifier by applying SVM machine learning to the training data. As the manual classification of training data takes time, it is intended that the classifier would be used to accelerate the process used for generating further training data.