Disfluencies along the garden path: brain electrophysiological evidence of disrupted sentence processing

Bailey and Ferreira (2003) hypothesized and reported behavioral evidence that disfluencies (filled and silent pauses) undesirably affect sentence processing when they appear before disambiguating verbs in Garden Path (GP) sentences. Disfluencies here cause the parser to "linger" on, and apparently accept as correct, an erroneous parse. Critically, the revision process usually associated with GP-disambiguating verbs does not appear to be triggered. In order to verify this effect, we recorded event-related potentials (ERPs) time-locked to disambiguating verbs in spoken GP sentences from 15 adults. A filled pause, silent pause, or no disfluency appeared before the GP-disambiguating verbs. Principal component analysis (PCA) revealed that fluent GP sentences elicited P600, an ERP index that revision of the initial parse was attempted. Crucially, P600 was attenuated for sentences containing a filled or silent pause before the GP-disambiguating verb. However, PCA detected an N400-like activation for these items, suggesting that listeners accepted the original (erroneous) parse and continued integrating at the verb; a conclusion that is tentative and requires further study. A left anterior positivity was also detected at GP-disambiguating verbs flanked by a filled pause. Discussion focuses on what these preliminary findings tell us about how oral comprehension proceeds when the time-course of sentence processing is disrupted.