Automatic Detection of Regional Words for Pan-Hispanic Spanish on Twitter.

Languages, such as Spanish, spoken by hundreds of millions of people in large geographic areas are subject to a high degree of regional variation. Regional words are frequently used in informal contexts, but their meaning is shared only by a relatively small group of people. Dealing with these regionalisms is a challenge for most applications in the field of Natural Language Processing. We propose a novel method to identify regional words and provide their meaning based on a large corpus of geolocated 'tweets'. The method combines the notions of specificity (tfidf), space correlation (HSIC) and neural word embedding (word2vec) to produce a list of words ranked by their degree of regionalism along with their meaning represented by a set of words semantically related and examples of use. The method was evaluated against lists of regional words taken from regional dictionaries produced by lexicographers and from collaborative websites where users contribute freely with regional words. We tested the effectiveness of the proposed method and produced a new resource for 21 Spanish-speaking countries composed of 5,000 regional words per country along with similar words and example 'tweets'.