On the interaction between syntax, semantics and working memory during sentence processing: an eye-tracking study

Abstract

During language processing, we constantly make predictions about the upcoming words (their semantics, syntactic category). Similarly, during agreement processing different features of information of the antecedent have to be predicted/retrieved. But what is the timing at which these prediction processes interact? Previous research has shown evidence supporting either a syntax-first model (Friederici,2002,2011) which considers each source of information separable from each other at early stages, and integrated at later ones or an interactive model (Hagoort, 2003) that considers all types of information processed simultaneously from early points of processing. In the present study, an eye tracking experiment was conducted with 24 Spanish native speakers in order to explore the course of interaction between lexico-semantic predictability effects with gender agreement between full DP object arguments and object-clitics. In addition, the role of Working Memory Capacity (WMC) during reading comprehension was also tested. The Spanish sentences used factorially combined grammaticality (grammatical vs ungrammatical object-clitic gender agreement) and verb predictability (high-cloze vs. low-cloze verbs) manipulations. The results revealed that in First Fixation Duration, considered an early timing measure, no interaction in the computation of verb-cloze probability and gender agreement was found. By contrast, late measures such as Total Fixation Duration showed interaction effects at the critical verb + clitic region. In addition, WMC interacted with predictability at early and late measures, increasing First Fixation Duration and Total Fixation Durations rates for participants with larger WMC. These findings contribute to the understanding of the processing of syntax-semantics interplay in correlation with WM and suggest that semantic and syntactic information interact only at late stages of the linguistic processing, favouring the syntax first model (Friederici, 2002; Wicha et al., 2014) and offering counter evidence for the interactive theoretical proposal (Hagoort, 2003). Likewise, they also prove that WM interactions constrain reading comprehension and predictability from early phases of processing.