Integrating Common Ground and Informativeness in Pragmatic Word Learning
- Manuel Bohn, Stanford University, Stanford, California, United States
- Michael Tessler, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States
- Michael Frank, Psychology, Stanford University, Stanford, California, United States
AbstractPragmatic inferences are an integral part of language learning and comprehension. To recover the intended meaning of an utterance, listeners need to balance and integrate different sources of contextual information. In a series of experiments, we studied how listeners integrate general expectations about speakers with expectations specific to their interactional history with a particular speaker. We used a Bayesian pragmatics model to formalize the integration process. In Experiments 1 and 2, we replicated previous findings showing that listeners make inferences based on speaker-general and speaker-specific expectations. We then used the empirical measurements from these experiments to generate model predictions about how the two kinds of expectations should be integrated, which we tested in Experiment 3. Experiment 4 replicated and extended Experiment 3 to a broader set of conditions. In both experiments, listeners based their inferences on both types of expectations. We found that model performance was also consistent with this finding; with better fit for a model which incorporated both general and specific information compared to baselines incorporating only one type. Listeners flexibly integrate different forms of social expectations across a range of contexts, a process which can be described using Bayesian models of pragmatic reasoning.
Return to previous page