The English dative alternation in vector space: how much does meaning matter?

Abstract Variationist linguists study alternations between different ways of saying the same thing (Labov, William. 1972. Sociolinguistic patterns . Philadelphia: University of Philadelphia Press: 188). Some linguists tend to believe that this sort of optionality is not predicted by theory and dysfunctional. We empirically explore if the two positions can be reconciled by assuming interchangeability between grammatical variants but admitting semantic information about the material in argument slots of competing constructions into variationist modeling. As a test case, we marshal corpus-based methods to study the well-known dative alternation after the verb to give in English. We specifically reanalyze the dataset investigated in Bresnan et al.’s seminal (2007) study. To assess the lexical semantics of the dative constituents, we rely on Semantic Vector Space modeling (Lenci, Alessandro. 2018. Distributional models of word meaning. Annual Review of Linguistics 4(1). 151–171). The meaning distinctions that we thus establish are subsequently submitted to variationist modeling. The distributional-semantic dative alternation model is then benchmarked against a traditional dative alternation model considering mostly higher-level, i.e., fairly abstract formal predictors (e.g., end-weight, definiteness) à la Bresnan (2007). The analysis shows that traditional variationist modeling à la Bresnan (2007) outperforms distributional-semantic modeling.