Reranking with Linguistic and Semantic Features for Arabic Optical Character Recognition.

Optical Character Recognition (OCR) systems for Arabic rely on information contained in the scanned images to recognize sequences of characters and on language models to emphasize fluency. In this paper we incorporate linguistically and semantically motivated features to an existing OCR system. To do so we follow ann-best list reranking approach that exploits recent advances in learning to rank techniques. We achieve 10.1% and 11.4% reduction in recognition word error rate (WER) relative to a standard baseline system on typewritten and handwritten Arabic respectively.