This paper presents a method to apply Natural Language Processing for normalizing numeronyms to make them understandable by humans. We approach the problem through a two-step mechanism. We make use of the state of the art Levenshtein distance of words. We then apply Cosine Similarity for selection of the normalized text and reach greater accuracy in solving the problem. Our approach garners accuracy figures of 71\% and 72\% for Bengali and English language, respectively.