Agreement variation in different grammar architectures: Challenges and perspectives

This paper compares two grammar architectures in the context of intralingual variation. Using the case of agreement variation in Russian, we highlight the advantages and limitations of approaches that imply a binary division of linguistic data (grammatical vs. ungrammatical) as well as those that consider gradient judgments. In the first case, modelling of variation is limited: existing models either presume a single definite derivation, thereby excluding any variation, or fail to restrict the distribution of variants. The second approach entails ranking grammatical constraints: in particular, agreement variation is subject to factors of varying strength that influence the selection of the variants. However, this approach is used inconsistently and lacks sufficient empirical justification. To integrate variation into the model of grammar, we advocate shifting the methodological framework from descriptive to parametric. We take the parametric nature to be the main requirement for the model of intralingual variation: in the case of agreement, it is implied that all parameters influencing the choice of the agreement variant are taken into account. The assessment of the contribution of individual parameters or their interaction is possible by means of multifactorial quantitative approaches. In particular, the experimental
method in combination with meta-analysis techniques allows us to generalize both the effects of multiple factors within a single construction and the effect of a given factor across various constructions. Moreover, quantitative methods offer novel insights into how grammatical constraints are distributed across individual grammars of speakers. Thus, the multifactorial approach yields profoundly innovative generalizations about variable linguistic phenomena.