Quantitative methods for corpus-based contrastive linguistics

Abstract

: The present paper makes a methodological contribution to the field of corpus-based contrastive linguistics. Contrary to the large majority of studies in contrastive linguistics that are mainly based on observed (relative) frequencies of (translation) data and are essentially monofactorial in nature, our study leverages more complex contrastive data that do justice to the complexity and multifactorial nature of cross-linguistic phenomena. Specifically, we focus on four challenging notions for the study of cross-linguistic near-synonymy: polysemy, degree of sense distinctiveness, prototypicality and identification of discriminatory variables. Each of these phenomena is tackled by means of a variety of statistical analyses based on two different kinds of input data that offer different kinds of resolutions on the data: (i) annotated concordance data and (ii) Behavorial Profile vectors. In an attempt to add to the toolbox of contrastive linguistics, we pay special attention to visualization techniques for cross-linguistic (dis)similarities such as hierarchical agglomerative cluster analysis, fuzzy clustering, and network analysis. These statistical methods will be illustrated on the basis of a case of cross-linguistic near-synonymy, namely the verb sentir(e) in Romance Languages.