An Efficient Technique for De-Noising Sentences using Monolingual Corpus and Synonym Dictionary

We describe a method of correcting noisy output of a machine translation system. Our idea is to consider di erent phrases of a given sentence, and nd appropriate replacements of some of these from the frequently occurring similar phrases in the monolingual corpus. The frequent phrases in the monolingual corpus are indexed by a search engine. When looking for similar phrases we consider phrases containing words that are spelling variations of or are similar in meaning to the words in the input phrase. We use a framework where we can consider di erent ways of splitting a sentence into short phrases and combining them so as to get the best replacement sentence that tries to preserve the meaning meant to be conveyed by the original sentence.