A Neurolinguistic Approach to Noncompositionality and Argument Structure

Abstract

Understanding the neural bases of language comprehension is to understand the implementation of language processing in the brain and how it affects language performance. Within a neurolinguistic study, we can examine the connection between linguistic competence and language performance at the cerebral level and whether the distinctions that we draw in linguistic theory map on to particular brain systems. Recently there has been an increase in psycholinguistic and neurolinguistic research using naturalistic stimuli following Willem’s (2015) encouragement to investigate the neural bases of language comprehension with greater ecological validity. Along with naturalistic stimuli, applying tools from computational linguistics to neuroimaging data can help us gain further insight into naturalistic, online language processing as computational modeling makes it easier to study the brain responses to contextually situated linguistic stimuli. (Brennan 2016). Utilizing this approach, in this dissertation I focus on two topics: noncompositional expressions (MWEs) and verbal argument structure. Across seven studies, I show how we can utilize various models and metrics from computational linguistics to operationalize cognitive hypotheses and help us better understand the neurocognitive bases of language processing. This dissertation is based on a large-scale fMRI dataset based on 51 participants listening to Saint-Exupéry's The Little Prince (1943), comprising 15,388 words and lasting over an hour and a half. While previous work has examined individual types of noncompositional expressions (such as idioms, compounds, binomials), this work combines this heterogeneous family of word clusters in a single analysis. Association measures are metrics from corpus and computational linguistics to identify collocations. This research contributes a gradient approach to these noncompositional expressions by repurposing association measures and demonstrates how they can be adapted as cognitively plausible metrics for language processing, among other findings. This dissertation also investigates the neural correlates of argument structure and corroborates previous controlled, task-based experimental work on the syntactic and semantic constraints between a verb and its argument. Another finding is that the Precuneus, not traditionally considered a core part of the perisylvian language network, is involved in both processing noncompositional expressions and diathesis alternations for a given verb. Overall, based on this interdisciplinary approach, this dissertation presents empirical evidence through neuroimaging data, linking linguistic theory with language processing.