Unsupervised mining of lexical variants from noisy text

The amount of data produced in user-generated content continues to grow at a stag-gering rate. However, the text found in these media can deviate wildly from the standard rules of orthography, syntax and even seman-tics and present significant problems to down-stream applications which make use of this noisy data. In this paper we present a novel unsupervised method for extracting domain-specific lexical variants given a large volume of text. We demonstrate the utility of this method by applying it to normalize text mes-sages found in the online social media service, Twitter, into their most likely standard English versions. Our method yields a 20 % reduction in word error rate over an existing state-of-the-art approach.