Title Linking neural systems to syntax using parsing models and neural decoding
Abstract
“[T]here is absolutely no mapping to date that we understand in even the most vague sense.” So writes David Poeppel in 2012 about the connection between Linguistics and neurobiology. I discuss our attempts to meet this challenge in the domain of syntax. In one path, we compare neural signals from fMRI to the state spaces traversed by different classes of incremental parsers. This approach, which is based on linguistically interpretable models, contrasts sharply with others that use “black box” large language models based on deep neural networks. It provides evidence favoring a particular class of grammatical formalisms and parsing strategies consistent with the key role for prediction in comprehension. In a second, complementary, path, we query the representations that result from such incremental processing. “Neural decoding” is a data analysis technique based on machine learning that identifies brain signals which discriminate between discrete cognitive states. We deploy this approach with electroencephalography (EEG) to decode signals associated with the processing of grammatical features during comprehension. We then use these decoders to interrogate the neural dynamics associated with assigning a phrasal label to a constituent and to pre-activating verbal argument structure.
