Linking brain systems with syntax requires characterizing the dynamics by which phrase-structure is built as a sentence unfolds word-by-word. A majority of prior work tackles this challenges with “complexity metrics” that quantify processing load incrementally as a function of, e.g., phrasal complexity, dependency length, predictability etc. Here we take a complementary approach by querying the representations that result from incremental processing. “Neural decoding” is a data analysis technique based on machine learning that identifies neural 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 language comprehension. We use these decoders to interrogate the similarity in neural responses to grammatical information across languages, the role of phrasal-labels in incremental processing, and the time-course of verb prediction.
