Encoding transliteration variation through dimensionality reduction: FIRE Shared Task on Transliterated Search

There exist a large amount of user generated Web content in Roman script for the languages which are written in indigenous scripts for various reasons. In the light of this phenomenon, the search engines face a non-trivial problem of matching queries and documents in transliterated space where transliterated content contain extensive spelling variation. This paper describes our proposed method to handle such variation through non-linear dimensionality reduction techniques. The approach achieves MRR as high as 0.84 on the main task of ad hoc retrieval.