Detecting presupposition failure with EEG

AbstractSentence comprehension in part involves introducing, storing, and retrieving information about individuals. Natural languages provide various means for performing this computational work. One popular idea is that indefinite noun phrases provide instructions for updating the discourse model by adding a new discourse referent, while definite noun phrases presuppose the existence of a discourse referent available in memory, as well as instructions for retrieving it. When no antecedent is available, the definite's presupposition fails to be satisfied, resulting in the so-called 'presupposition failure' and pragmatic infelicity. However, under certain conditions, definite noun phrases can felicitously be used even when no antecedent is available in memory. In such cases, a conversational repair strategy called 'presupposition accommodation' can rescue the discourse by adding the required referent. It is natural to expect greater processing costs for adding a discourse referent with a definite than with an indefinite: although both involve the process of adding a referent, the definite must go through a stage of presupposition failure and a subsequent decision to accommodate. The experimental challenge has been to apply a method sensitive enough to detect expected costs in discourse, even when the participant is unaware of the presupposition failure and repairs it rapidly. The present study addresses this challenge by using EEG to capture temporally fine-grained processing differences between definite and indefinite noun phrases when both introduce new discourse referents in plausible and implausible contexts. Our main finding is that definite noun phrases elicit a LAN compared to indefinite noun phrases both in implausible contexts where there is a sense of oddness and in perfectly coherent contexts. Divergence between signals elicited by the definite versus indefinite conditions continue into later stages, beyond the typical LAN time window. We take our main result to be evidence that the LAN indexes a cognitive stage at which presupposition failure is detected and when an accommodation decision occurs. These findings support the idea that, when encountering a definite, the LAN is tightly linked to working memory processes involving the search for discourse elements that are presupposed to exist in memory. When none are found, definites are subsequently accommodated and bridged to other entities in the discourse.

Return to previous page