The Effect of Speaker-Specific Information on Pragmatic Inferences

2 Introduction It is commonplace to observe that utterances can convey more information than they explicitly encode. Indeed, the field of pragmatics has grown from the insight that speakers exploit communicative conventions in order to say more with less. Less attention has been paid to the burden this places on perceivers. Because speaker meaning is underspecified, perceivers must infer implicit content in order to interpret, and successfully situate, utterances within the discourse. These inferences 1 appeal to many types of knowledge including entailment relations, world knowledge, and the discourse context. The last of these presents a particular challenge to investigators because contexts are dynamic and utterance meaning can be context sensitive. For instance, the utterance in (2) implies something like (2a) if it is a response to (1a) and something like (2b) if it is a response to (1b). 2 (1) a. What time is it? b. How good is the party? (2) Some guests are already leaving a. It must be late. b. The party is not much fun. The ease and prevalence of such inferences facilitate efficient communication, but at the same time create difficulty for models of language understanding. A central puzzle is how the extrasentential context is combined with intrasentential information to ultimately yield an interpretation of a sentence. The present paper addresses this issue by exploring a particular dependency between the referential environment and linguistic form. Suppose a speaker wishes to refer to one member of a set of entities belonging to the same nominal category in the current discourse. They must use a modified expression in order to refer successfully. For instance, if one cup is the intended referent in a context containing two cups, the speaker must use a restrictive modifier such as " the cup on the left " or " the red cup. " Significantly, this dependence appears to be bidirectional. Upon encountering a restrictively modified noun phrase (NP), such as the tall cup, two sets are invoked in the immediate discourse: (1) a target set corresponding to the literal denotation of the 1 Here and throughout we use the term "inference" to refer to information that is communicated to the perceiver via the utterance of an expression, but which is not part of its asserted content. This would include both accommodated presuppositions and implicatures. We adopt this term because it is neutral with respect to whether the inferred content arises …