The psycholinguistic reality of generative theories and construction grammar in language processing

Abstract

Argument realization is a fundamental topic in linguistic theories. A powerful theory should not only describe language grammaticality consistently and cross-linguistically, but also have explanatory power to the field of language acquisition and processing. On the other hand, due to the traditional standards set by the three levels of adequacies, language processing data has not been equally explored in linguistic model evaluation. Therefore, this dissertation surveys different theoretical approaches to argument realization and examines their psycholinguistic reality through language processing evidence. Two major approaches to argument realization have been proposed  the projectionist approach and the constructional approach. The projectionist approach has roots in generative theories, and it assumes argument structure is predictable from the lexical semantic representation of the verb and configures argument realization by projecting verb semantics to grammatical functions through certain mapping algorithm. Construction grammar, by contrast, assumes constructions as the basic language units and tackles argument realization by the fusion between argument structure constructions and verb meaning. Both approaches have received some evidence from language acquisition and processing studies. This dissertation first investigates the psycholinguistic reality of the two approaches by testing the processing correlates of argument structures against the structural analyses in each approach. In order to translate the structural analyses into processing predictions, the dependency locality theory is selected as the compatible processing complexity model for generative analyses. Furthermore, a revised dependency locality theory was proposed in this dissertation with the support of eye-tracking data (Experiment 1) as the compatible complexity model for constructional analyses. Two cross-modal lexical decision tasks (Experiment 2A, 2B) were conducted to examine the processing time of the dative alternations and the locative alternations against the processing predictions based on generative and constructional analyses. The results are not compatible with the predictions from generative theories and construction grammar seems to offer an alternative explanation. Furthermore, this dissertation investigates the psycholinguistic reality of the two approaches by testing the processing correlates of verb semantics against the verb representation in each approach. A lexical decision task (Experiment 3) and a self-paced reading task (Experiment 4) were conducted for this purpose. The results show that construction grammar better predicts the reaction time and the reading time of verbs, suggesting that participant role information is stored in verb representation and accessed during lexical and sentence processing. In conclusion, the processing data of both argument structures and verb representation are compatible with the prediction from the constructional approach, which suggests a better psycholinguistic reality of construction grammar.