Different sources of predictions during natural reading: an EEG and Eye-Tracking co-registration study

Constraint satisfaction in large language models

“…It will be necessary to create a mechanistic model of the implicit decision processes (e.g., Hoover et al, 2022) that can translate the processing done in LLMs to human cognitive processes. It is worth noting in this regard that these human measures all operate on different timescales and have not always been related clearly to each other, despite all of them coexisting for decades in psycholinguistic and neuroscience research (though see Bianchi et al, 2020 andHimmelstoss et al, 2020 for interesting work relating different comprehension measures). It is possible that continued efforts to relate human data and LLM models, for example studies of the relationship between N400s in evoked potential research and neural attention mechanisms, could eventually lead to a better understanding of the relationships between the various human measures of comprehension.…”