Context-Sensitive Arabic Spell Checker Using Context Words and N-Gram Language Models

This paper addresses real-word spell checking using context words and n-gram language models. A corpus that consists of different Arabic topics is collected. A collection of confusion sets is normally used in addressing real-word errors. Twenty eight confusion sets are chosen in our experiments. These sets were collected from the most common confused words made by non-native Arabic speakers and from OCR misrecognized words. The probabilities of the context words of the confusion sets are estimated using a window-based technique. N-gram language models are used to detect real-word errors and to choose the best correction for the errors once found. An automatic context-sensitive spell checking prototype that detects and corrects real-word errors in Arabic text is implemented. The experimental results showed promising correction accuracy.