Electrical brain potentials reveal temporal dynamics of word prediction during language comprehension

Wing-Yee Chow,Ellen F. Lau,Suiping Wang,C. Phillips

2013-01-01

Abstract:Anticipation of upcoming input has been shown to be critical to the speed and accuracy of many cognitive processes, particularly in domains such as language in which the listener must process an unbroken stream of 3-5 words per second. However, little is known about how the predictions themselves are generated in real time. One important but untested assumption of existing models of language prediction is that all contextual information can immediately feed predictions about upcoming inputs. Here, we demonstrate that this assumption is incorrect. We took advantage of the word order properties of Mandarin Chinese to manipulate the timing with which predictive information became available relative to the critical word in a sentence and used electroencephalography to inspect the effects of prediction on early processing (350-450ms) in three experiments. We demonstrate that, despite its prominence and unambiguity, word order information informs language prediction only at a measurable delay. These findings suggest the need for a shift in emphasis from ‘what’ kinds of information inform predictive processing to ‘when’ different sources of information becomes available to predictive mechanisms, and open a new window into the subcomponents of predictive processes that are crucial for fast and robust language understanding.