Indian Language Transliteration Using Deep Learning

Transliteration of Indian languages has been a challenging problem., given their complex nature., which has conventionally been handled by rule based systems developed by trained linguists. Scalability is a prominent issue with these systems and given the number of language pairs possible for Indian languages., a scalable pipeline is necessary for swift development of these systems. Deep learning systems are pragmatic for building a scalable pipeline., as they are completely data driven. We experimented with LSTMs and Sequence to Sequence models to find an optimal model for the scalable pipeline by comparing the results. The results show Sequence to Sequence models are a better fit for this solution. We also discuss techniques for pre-processing the data and post processing the output for optimal performance.