A Bayesian multinomial model of near-synonymy

Abstract

This paper is a quantitative multifactorial study of near-synonymous constructions let + V, allow + to V and permit + to V based on the British National Corpus. We fit a Bayesian multinomial mixed model with twenty formal, semantic, social, collostructional and other variables as fixed effects and the infinitives that fill in the second verb slot as random effects. The model reveals a remarkable alignment of variables that indicate the formal distance between the predicates, conceptual distance between the events they represent and between the speaker and the main arguments, the social and communicative distance between the interlocutors, as well as the looseness of the relationship between the constructions and second verb slot fillers. These results raise fundamental theoretical questions about the relationships between linguistic form, function and use.