Keynote 4: Judith Degen, Stanford University
The Rational Speech Act framework: an integrative theory of the interaction between literal meaning, world knowledge, and context
In producing and comprehending language, speakers and listeners engage in rich inference processes that involve reasoning about the (often noisy) linguistic signal's literal meaning, world knowledge, and the context of utterance. The development of a fully-fledged theory of meaning, one of the goals of experimental pragmatics, requires precisely describing and formally analyzing these inference processes. But the complexity of establishing which information (whether from world knowledge or from context) is used, and how and when it is integrated in online processing, has presented serious challenges to theory development.
Rational theories of language use starting with Grice have attempted to bring some order to this difficult analytical situation by positing conversational principles that listeners expect cooperative speakers to abide by. More recently, maintaining core insights from the Gricean program while capturing that language production and comprehension are inherently probabilistic and uncertain, the Rational Speech Act (RSA) framework was developed. Its core tenets include a) that language is for communication (i.e., something to be studied by virtue of its use by intentional agents); b) that production phenomena are best captured as resulting from speakers who trade off contextual informativeness and production cost; and c) that comprehension phenomena are best captured as resulting from listeners who integrate beliefs about such speakers with prior beliefs about likely meanings via Bayesian inference.
I argue that RSA is currently our best bet for capturing the systematic effects of the interaction of literal meaning, context, and world knowledge on both production and comprehension. Using case studies including the production of presumed overinformative referring expressions, context effects on scalar implicature, and world knowledge effects on scalar implicature, I will demonstrate the productive application of RSA as both a hypothesis-testing tool for experimental pragmatics and as a pragmatic theory that lends itself to incremental update in the face of evidence while maintaining the above core tenets.
I conclude that, rather than being wastefully overinformative, speakers systematically add modifiers to referring expressions to the extent that doing so reduces their uncertainty about whether the listener will correctly infer their intended meaning. Similarly, rather than being uselessly underinformative, speakers systematically provide less information when listeners can be relied on to make good use of the available contextual information to correctly infer their intended meaning. Together, these findings implicate a linguistic system geared towards communicative efficiency. Finally, I will discuss how RSA, which was developed as a theory of the rational reconstruction of inferences, can be applied to investigate online pragmatic processing.