Hybrid Tamil spell checker with combined character splitting

Abstract

Spell checker is the application, which helps in finding the spelling errors in a given text. Applications like word processors, mails, search engines, speech recognition and social media forums need these kinds of spell checking tools to increase the correctness of the system. Spell checking is completely implemented in languages such as English, French, and Chinese. But as far as Indian regional languages is concerned, very few works have been carried out, that too partially. Tamil is one such Indian regional language, which requires a fully implemented spell checking application as many people started using this language in social media platforms like Facebook and Twitter. Spelling errors fall on different categories in Tamil language, which involves Sandhi errors, Homophone errors (Mayangoli), and misspelt words error. To tackle all these errors, a new ensemble approach is proposed in this paper. The proposed approach consists of Levenshtein's edit distance algorithm, rule-based algorithm, Soundex algorithm along with LSTM (Long Short Term Memory) model. We have used a special feature called combine character splitting of Tamil alphabets for feeding the LSTM model to improve the performance of the system. Proposed system produced an accuracy of 95.67%, which is approved by the Tamil scholar.

Correcting spelling mistakes in Persian texts with rules and deep learning methods

“…They reported an accuracy of 83%, recall of 72%, and F-measure of 76% as the network results for the test data. Some studies 47,48 have examined the combined effect of rules and neural network-based methods, which shows an improvement in the accuracy of the hybrid model. Sampath and Shanmugavel 47 suggested a hybrid approach that combines the algorithm of Levenshtein's edit distance, the algorithm based on rules, the algorithm of Soundex, and the model of LSTM for the spell checker of the Tamil language.…”