Pragmatics, non-literal language, and language in conversation
Only published papers are included; for preprints, see Papers.
Last Updated: May 2024
Understanding language entails much more than simply decoding the literal meaning of each sentence: our interpretation of each utterance is powerfully shaped by our knowledge of the intent of the speaker, the linguistic and social context of the utterance, and our general world knowledge. This ability to exploit speaker intent and background knowledge to go beyond the literal meaning of the sentence is called “pragmatics”. Pragmatic reasoning is most critically needed during conversational exchanges. We have been investigating pragmatic processing behaviorally and using neural measures for several years now. Below are some things we discovered, but more is on the way.
The language network and the Theory of Mind network are dissociated but show some degree of information sharing
These papers show that the language network robustly dissociates from the Theory of Mind network, although the Paunov et al. (2019, J Neurophys) paper also shows that the two networks show some degree of functional correlation (which may be taken to suggest frequent interactions and information sharing).
Non-linguistic communicative signals are not processed by the language network
This paper shows that non-linguistic communicative information conveyed by co-speech gestures is not processed in the language network.
Discourse structure building draws on the Theory of Mind and Default networks, not on the language network
Inter-clausal connections are critical for many pragmatic inferences. These papers demonstrate that the language network is not sensitive to discourse-level structure. Instead, areas of the Theory of Mind networks and the Default Network appear to support the computations related to discourse-structure building.
Comprehenders are sensitive to co-listeners’ knowledge states
This paper uses event-related potentials to investigate sensitivity to others’ knowledge states and finds that comprehenders are sensitive to whether or not those around them have knowledge that is critical for understanding incoming linguistic input, even if those individuals are not directly involved in the communicative exchange. This phenomenon is known as the “social N400” component.
Non-literal language processing draws on the language and Theory of Mind networks
In this paper, we develop a new method for meta-analyzing activation peaks from past fMRI studies, based on probabilistic functional atlases, and show that the processing of non-literal language (across diverse phenomena) recruits the language network and the Theory of Mind network, but not the domain-general Multiple Demand network. We also find that in the language network, the left-hemisphere regions are more likely to be engaged than the right-hemisphere regions, contra some past claims.
Pragmatic reasoning in large language models
This paper investigates the ability of large language models (LLMs) to perform pragmatic reasoning and finds that some of the larger models achieve quite high performance, which suggests either that distributional linguistic information may suffice for (at least some components of) pragmatic reasoning, or that LLMs implicitly acquire Theory of Mind reasoning capacities.
Processing presupposition violations elicits a general error correction ERP component
This paper investigates the processing of presuppositions using ERPs and finds that presupposition violations elicit a P600 component, which is typically associated with general error correction, but no N400 component, which suggests that comprehenders rapidly integrate presupposed content and that processing these kinds of violations is performed by brain regions outside of the core language network.
