Disentangling semantic prediction and association in processing filler-gap dependencies: an MEG study in English

Abstract

Understanding language is facilitated by prediction of upcoming words. Sentences with filler-gap dependencies can provide sophisticated cues about an upcoming verb. A sentence beginning with which cat did you … ? Is more likely to end with lift than meow. M/EEG recordings show a diverging response ∼200–400 ms (“N400”) after the onset of unpredictable words vs. predictable words, and similarly for pairs of words with high vs. low semantic association. Previous studies report N400 responses to implausible filler-gap dependencies, however it is unclear whether these findings index verb predictability or semantic association between the reactivated filler and verb. We report on an MEG study examining argument-verb relations in sentences with and without filler-gap dependencies, controlling for lexical association between arguments and verbs. Implausible subject-verb relations showed the characteristic response at 200–500 ms in left frontal cortex, and implausible filler-gap at 600–800 ms in right frontal cortex, suggesting different mechanisms for filler-gap dependencies.