Query word labeling and Back Transliteration for Indian Languages: Shared task system description

In this paper, we target the problem of word level language identification for Indian languages written in roman script and mixed with English language. In addition to the language identification we also handle transliteration of Indian language words into the native Indic scripts. To address these problems we consider a supervised approach of building a classifier with monolingual samples together with a context-switching probability from Indian Language (IL) to English (Eng). The proposed system is submitted to the FIRE-2013 shared task on Transliterated Search. Our submitted system showed the best performing results.