Linguistic choices vs. probabilities – how much and what can linguistic theory explain?

A question of general theoretical interest in linguistics is what is the relationship between naturally produced language, evident in e.g. corpora, and the posited underlying language system that governs such usage. This concerns on the one hand the use and choice among lexical and structural alternatives in language, and on the other the underlying explanatory factors, following some theory representing language as a cohesive system. A subsequent subservient methodological challenge is how this can be modeled using appropriate statistical methods. The associated question of general theoretical import is to what extent we can describe the observed usage and the variation it contains in terms of the selected analytical features that conventional linguistic theory incorporates and works upon. The practical purpose of this paper is to present a case study elucidating how multivariate statistical models can be interpreted to shed light on these questions, focusing on a set of near-synonyms as the particular type of linguistic alternation. With multivariate modeling, I mean two distinct things. Firstly, I imply the use of multiple linguistic variables from a range of analytical levels and categories, instead of only one or two, in order to
study and explain some linguistic phenomenon. Secondly, I mean with this term the use of multivariate statistical methods such as polytomous logistic regression. In the following introduction, I will first present research demonstrating that one and the same linguistic phenomenon can be associated with, and appear to be explainable in terms of a wide range of different variables from various levels of linguistic analysis.
Next, I will note research indicating that satisfactory explanations of such linguistic phenomena requires mult ivariate (mult icausal) models, i.e. the incorporat ion of all of these variables at the same time in the analysis. This leads us to the final and central question of how much of the phenomena we can in the end account for with the fullest set of explanatory variables available to us in current linguistic analysis.