Existing psycholinguistic models typically describe agreement relations as monolithic phenomena amounting to mechanisms that check mere feature consistency. This eye-tracking study aimed at widening this perspective by investigating the time spent reading subject-verb (number, person) and adverb-verb (tense) violations on an inflected verb during sentence comprehension in Spanish. Results suggest that (i) distinct processing mechanisms underlie the analysis of subject-verb and adverb-verb relations, (ii) the parser is sensitive to the different interpretive properties that characterise the person, number and tense features encoded in the verb (i.e. anchoring to discourse for person and tense interpretation, as opposed to anchoring to cardinality information for number), and (iii) the (local, distal) position of the agreement controller with respect to the verb affects the interpretation of these dependencies. An account is proposed that capitalises on the importance of enriching current sentence processing formalizations using a feature and relation-based approach.

Linguists propose the existence of linguistic trees and define the merge operation to construct complex sentences from simpler elements. Previous neuroimaging studies, primarily utilizing 3T scanners, have identified an extensive fronto-temporal network involved in forming linguistic structures and executing merge operations. Intracranial recordings in these areas reveal a more distributed picture, with adjacent regions undertaking diverse linguistic tasks. We designed a 7T fMRI visual task to investigate the neural coding of syntactic operations. In healthy French-speaking participants, we initially identified the language network using a localizer. Subsequently, we employed short 3-word stimuli, presented briefly (200ms), to explore the response profiles within the language network. These stimuli included control conditions, affirmative statements, and interrogative sentences, all matched for letter and character count. Preliminary results indicate that 200ms is sufficient to differentiate between sentences and non-sentences, and suggest a finely-tuned specialization for syntactic operations within language network subregions.