Online processing of scalar implicatures in Chinese as revealed by event-related potentials

Abstract

During sentence processing, whether pragmatic information is integrated immediately and automatically or at a delay is a subject of debate in experimental pragmatics. One test case is that of scalar implicatures, which occur in statements like "some of the students are hardworking", which have both a logical meaning ("at least one is hardworking") and a pragmatic meaning ("not all of them are hardworking"). Default processing accounts hold that the pragmatic meaning of some comes online immediately and effortlessly, whereas context-based processing accounts propose that this meaning is not generated until after the logical meaning. Previous event-related potential (ERP) studies on scalar implicatures typically investigated critical words downstream of the quantifier and were thus not able to address the possibility of immediate construction of scalar interpretations at the moment the quantifier is encountered. Furthermore, effects of lexico-semantic processing and real-world context make it difficult to interpret effects observed in these studies. The present study adopts a picture-sentence design to make the violation immediately detectable when the quantifier is read and to control the context in which the sentence is understood. Participants saw pictures in which several characters are either performing the same activity or different activities, followed by sentences using "some" or "all", yielding a 2x2 design including both pragmatic violations ("some" sentences after "all" pictures) with matched controls, and purely incorrect assertions ("all" sentences after "some" pictures) with matched controls. Crucially, the pragmatic violation cannot be recognized as a violation until after the pragmatic meaning of some is computed. Pragmatic violations and purely logic violations elicited an early N400 effect and a right-lateralized negativity in the 600-900 ms time window, whereas purely logic violations elicited qualitatively different effects in at least the late time window. These results demonstrate that the pragmatic meaning of some, which relies on the generation of a scalar implicature, is available to the processor immediately; furthermore, they show that errors based on pragmatic expectations and errors based purely on logic elicit qualitatively different electrophysiological responses. I conclude that these findings are consistent with a default processing account, although they do not rule out a context-driven account.