In this paper, we deal with transliterations from non-standard forms of bengali words written in english to their standard froms. Familiarity of users with standard english keyboard makes it easy for them to represent bengali words with english letters (i.e Roman Script). Starting from a list of commonly used bengali word compiled from a corpus, we obtain a pronouncing dictonary using a grapheme-to-phoneme converter. We propose a novel method based on heuristic search thecniques over the pronouncing dictionary for transliteration of a word written in bengali using roman script to its standard bengali from application of this thecnique include design of phonetic keyboards from for bengali, automatic correction of casually written bengali in roman english, query correction for bengali search over the web, and serching loosely transliterated named entity. The techniques used are generic and can be readily adapted to other languages.