Grouping Spanish-speaking countries by dialect: A corpus dialectometric approach

Abstract

The present study attempts to cluster Spanish-speaking countries into dialect regions by computational means. The frequencies of 592 lexical and grammatical features for 21 countries were obtained the from Corpus del Español-Web Dialects. Principal components analysis and hierarchical clustering analyses used the resulting data to group countries into dialect regions. A number of algorithms were used to rank features in terms of how much they aided in dialect classification, which allowed grouping based on a smaller set of features. Six dialect zones were identified: European (Spain), Southern Cone (Uruguay, Argentina), Southern Central America (Costa Rica, Panama), Caribbean (Puerto Rico, Dominican Republic), Northern Central America (Nicaragua, El Salvador, Guatemala, Honduras), Andean South America (Bolivia, Paraguay, Chile, Peru). However, different subsets of features, and different clustering algorithms produced groupings that varied somewhat. The bulk of the variation dealt with where Cuba, Ecuador, Mexico, Venezuela, Colombia, and the US fit into the dialect regions. The difficulties of the computational approach to dialect classification are discussed. Allowing computer algorithms to determine dialect boundaries appears objective. However, interpreting a principal components analysis entails a degree of subjectivity. Furthermore, the plethora of different classification algorithms allows the researcher to choose the one that produces the desired outcome.