Combination of Trigram and Rule-based Model for Singlish to Sinhala Transliteration by Focusing Social Media Text

Language transliteration has become an important area of language processing. Most Sri Lankans use “Singlish”, in which Sinhala (Sinhalese) words are written in English letters in social media. This study evaluates several methods of Singlish to Sinhala transliteration and addresses the problem of transliteration by using a trigram method. A trigram model is combined with the rule-based method which is a system of the UCSC implemented early. The trigram model was trained with Singlish YouTube comments and with their corresponding Sinhala transliteration. The trigram model uses the implementation of the trigram tagger, and the training data set was also formatted as in POS tag dataset. Together the Trigram and Rule base model gives better results for transliteration with the letter-wise accuracy of 77%. This approach is enriched with a natural UX approach to capture the Singlish text content.