University of Pennsylvania Working Papers in Linguistics


There is much current debate about processing scalar implicature, but a considerable body of empirical evidence seems to support the idea that it requires additional time and effort on the part of the hearer (e.g. Breheny, Katsos and Williams 2005, Bott and Noveck 2004 and many others). The goal of this study was to contribute to our understanding of the cognitive processes that go on as comprehenders process sentences with and without scalar implicatures. We conducted a visual-world eye-tracking experiment using a picture-verification task, with a novel single-picture display, and asked participants to indicate whether the picture they saw was a good description of the sentence they heard. As a whole, our results suggest that processing scalar implicatures does appear to entail a processing cost. In this paper, however, we take a closer look at a pattern which has also been obtained in several previous experiments (e.g. Noveck 2001, Noveck and Posada 2003), namely, the tendency for participants to split into two distinct kinds of responders in the presence of underinformative descriptions. An example of an underinformative description is “Some giraffes have long necks”, which is not a sufficient description of the reality that all giraffes have long necks.

Existing research suggests that adults respond to underinformative sentences either using a consistent logical interpretation (e.g. “some” always means “some and possibly all”, and thus “Some giraffes have long necks” is judged to be true) or a consistent pragmatic interpretation involving a scalar implicature (e.g. “some” always means “some but not all”, and thus “Some giraffes have long necks” is judged to be false). Although it is widely assumed that participants’ answers reflect their on-line processing (i.e., a logical response means that no implicature was computed, a pragmatic response means that the implicature was computed), our data suggest that participants are aware of scalar implicature regardless of how they respond to underinformative sentences, and in some cases, greater processing can be demonstrated for participants who answer “logically”. We further suggest that the emergence of participant response groups may be due to participants’ sense that they should be consistent within an experimental context, rather than a difference in how underinformative items are interpreted.