A Pipeline Approach to Supervised Error Correction for the QALB-2014 Shared Task

Abstract

This paper describes our submission to the ANLP-2014 shared task on automatic Arabic error correction. We present a pipeline approach integrating an error detection model, a combination of character- and word-level translation models, a reranking model and a punctuation insertion model. We achieve an F1 score of 62.8% on the development set of the QALB corpus, and 58.6% on the official test set.