Data Mining Methods to Compare Englishes

Abstract

The paper presents the results of the corpus-based research of noun cryptotypes in 20 varieties of English (Englishes). The data for this research collected from Mark Davies’ corpora GloWbE and NOW enabled us to focus on variation in the covert classification of nouns in modern Englishes. A noun cryptotype introduced by Whorf is approached as ‘a covert type of classification of nouns, marked by lexical selection in a syntactical classifier rather than a morphological tag’. The purpose of the study has been to compare and contrast the covert classification of basic 23 emotions in 20 Englishes (64,702 tokens). 20 Englishes have been clustered with the help of Data Mining methods (such as k-means clustering and a self-organizing Kohonen map). There are six clusters that appeared to be corresponding to geographic areas: American cluster (American and Canadian Englishes); Australian cluster (Australian and New Zealand Englishes); European cluster (British and Irish Englishes); Asian cluster (Indian, Pakistani, Singapore, Hong Kong, Malaysian, Bangladeshi, Sri Lankan, and Philippine Englishes); African cluster (Kenyan, South African, Nigerian, Ghanaian, and Tanzanian Englishes); Caribbean cluster (Jamaican English). The correlation coefficients among Englishes in the Asian and African clusters (the Outer Circle in the World Englishes Paradigm of Braj B. Kachru) range from 0.74 to 0.8 due to little contact among the varieties inside these clusters. The correlation coefficients between Englishes in the American, Australian and European clusters (the Inner Circle, Kachru) range from 0.92 to 0.933, which indicates a high consistency of these varieties owing to the long lasting, enduring linguistic contacts.