Fast and Robust Arabic Error Correction System

Abstract

In this paper we describe the implementation of an Arabic error correction system developed for the EMNLP2014 shared task on automatic error correction for Arabic text. We proposed a novel algorithm, where we find some correction rules and calculate their probability based on the training data, they we rank the correction rules, then we apply them on the text to maximize the overall F-score for the provided data. The system achieves and F-score of 0.6573 on the test data.