Determining the alternatives for scalar implicature

Successful communication regularly requires listeners to make pragmatic inferences — enrichments beyond the literal meaning of a speaker’s utterance. For example, when interpreting a sentence such as “Alice ate some of the cookies,” listeners routinely infer that Alice did not eat all of them. A Gricean account of this phenomena assumes the presence of alternatives (like “all of the cookies”) with varying degrees of informativity, but it remains an open question precisely what these alternatives are. To address this question, we collect empirical measurements of speaker and listener judgments about varying sets of alternatives across a range of scales and use these as inputs to a computational model of pragmatic inference. This approach allows us to test hypotheses about how well different sets of alternatives predict pragmatic judgments by people. Our findings suggest that comprehenders likely consider a broader set of alternatives beyond those logically entailed by the initial message.