Journal
Travaux du Cercle Belge de Linguistique, vol. 17 (2023)
Auteur·e·s
Chiara Paolini, Benedikt Szmrecsanyi, Mariana Montes et Hubert Cuyckens (KU Leuven)
Résumé
The dative alternation in English is one of the most extensively investigated alternations. Many quantitative studies to date have focused on more traditional formal predictors (e.g., complexity, constituent weight) to explain the choice of one variant over the other. In contrast, semantic predictors have been given relatively short shrift in variationist alternation research due to their perceived cost inefficiency. The objectives of this research are on the one hand to determine the importance of fine-grained semantic properties of the theme and recipient nouns for predicting variant choice, and to check on the other hand whether they add to the explanatory power of traditional formal predictors. To accomplish our aims, we make use of automatically generated semantic predictors using distributional models of meaning (Lenci 2018). Analysis shows that while distributional semantics predictors have significant predictive power, traditional predictors are subtly more powerful. Nevertheless, interactions between semantic and formal predictors clearly emerge from statistical analyses, opening the research to further applications and developments.