The neural basis of language comprehension: Insights from spatiotemporal imaging

为了在嘈杂和模糊的环境中成功地理解语言，人类的大脑必须使用

For language comprehension to succeed in noisy and ambiguous environments, the human brain must use contextual information to actively predict upcoming linguistic inputs. Impairments in top-down prediction are thought to contribute to language and communicative dysfunction in a variety of neurodevelopmental disorders, from the reading disabilities associated with dyslexia, to the profound social and communicative dysfunctions that characterize schizophrenia and autism spectrum disorder. In neurotypical adults, linguistic prediction is known to modulate neural activity within a left-lateralized fronto-temporal network. However, little is known about the computational mechanisms that determine the timing of feedforward and feedback activity across this network. This grant asks whether these neural dynamics can be explained by predictive coding — a unifying theory of perceptual and cognitive function. According to predictive coding, the brain infers the meaning of sensory inputs by minimizing prediction error across multiple levels of the cortical hierarchy. To test this theory, this grant proposes a series of experiments using three complementary neuroimaging techniques –– magneto-encephalography (MEG), electroencephalography (EEG) and functional MRI –– to probe the timecourse and location of neural activity to incoming words during language comprehension. Computational simulations using an implemented predictive coding model of language processing will serve as a powerful complementary research tool, allowing for the testing of explicit, computationally motivated hypotheses. Aim 1 (EEG/MEG) will test the hypothesis that the timecourse and localization of evoked (phase-locked) neural activity within the left temporal cortex can be explained by prediction error at multiple levels of linguistic representation. Aim 2 (MEG/EEG) will use Representational Similarity Analysis to directly capture neural pre- activation of specific words at different levels of linguistic representation in predictive sentence contexts. These methods will also be used to track the timecourse of converging on sharpened neural representations after word onset in both predictive and non-predictive contexts. In both these Aims, computational simulations using the same items will proceed in parallel with these neuroimaging studies, guiding interpretation. Aim 3 (MEG/EEG/fMRI) asks whether the principles of dynamic predictive coding framework can explain how the brain is able to flexibly shift away from prior predictions in order to rapidly infer a new underlying message. Specifically, this Aim asks whether these principles can explain neural activity at the highest level of the fronto- temporal language hierarchy — the left inferior frontal cortex — as well as top-down feedback to lower cortical regions at a later stage of processing. By directly linking the neurobiology of language comprehension to a central theory of human cortical function, this project will identify core neural and computational mechanisms that may be disrupted in multiple language disorders. It therefore lays the foundation for the development of targeted, theoretically motivated neurocognitive strategies for the treatment and prevention of communicative disability.