Out-of-Vocabulary Pashto Spell Checker using Morphological Operations

A spell checking model detects spelling errors in the input text, generates possible corrections for each error, and organizes them based on their relevance. Such tools are essential for writing, editing, and publishing in a language. In literature, different variants of edit distance and language models are used for dealing with spelling errors. The existing approaches have improved for efficient use of computational and memory resources without compromising accuracy. Exploiting the efforts made for other low-resource languages, we have proposed the first spell checking model for Pashto language. A common issue with the existing spell checker models of low-resourced languages is having many false positives due to limited vocabulary. Our proposed approach makes use of morphological operations to generate out-of-vocabulary words and is coupled with a filtering mechanism to retain the words with more repeating syllable patterns only. The n-gram probabilities are combined through linear interpolation to detect and correct spelling errors. The test samples with spelling errors are generated by randomizing characters for non-word errors and words in higher-order n-gram for real word errors. The proposed approach achieves detection accuracy of 85.5% and correction accuracy of 75.0%, respectively.