Nonliteral understanding of number words

Abstract

One of the most puzzling and important facts about communication is that people do not always mean what they say; speakers often use imprecise, exaggerated, or otherwise literally false descriptions to communicate experiences and attitudes. Here, we focus on the nonliteral interpretation of number words, in particular hyperbole (interpreting unlikely numbers as exaggerated and conveying affect) and pragmatic halo (interpreting round numbers imprecisely). We provide a computational model of number interpretation as social inference regarding the communicative goal, meaning, and affective subtext of an utterance. We show that our model predicts humans' interpretation of number words with high accuracy. Our model is the first to our knowledge to incorporate principles of communication and empirically measured background knowledge to quantitatively predict hyperbolic and pragmatic halo effects in number interpretation. This modeling framework provides a unified approach to nonliteral language understanding more generally.pragmatics | computational modeling I magine a friend describing a new restaurant where she recently dined. Your friend says, "It took 30 minutes to get a table." You are likely to interpret this to mean she waited ∼30 min. Suppose she says: "It took 32 minutes to get a table." You are more likely to interpret this to mean exactly 32 min. Now, suppose she says: "It took a million years to get a table." You will probably interpret this to mean that the wait was shorter than a million years, but importantly that she thinks it took much too long. One of the most fascinating facts about communication is that people do not always mean what they say-a crucial part of the listener's job is to understand an utterance even when its literal meaning is false. People's ability to interpret nonliteral language poses a critical puzzle for research on language understanding.A rich body of literature in psychology and linguistics has examined how people use and understand nonliteral language (1-4). However, most of the work has been qualitative, with little focus on analyzing aspects of an utterance that predict the quantitative details of people's figurative interpretations. Here, we present a computational model that formalizes and integrates three general principles of language and communication to explain the basis of nonliteral language understanding. First, speakers and listeners communicate with the assumption that their interlocutors are rational and cooperative agents; second, listeners assume that speakers choose utterances to maximize informativeness with respect to their communicative goals; third, speaker and listener use common ground-their shared knowledge of the world-to communicate effectively. The first principle has been formalized by a recent body of work on rational speech act (RSA) models, which views pragmatic language understanding as probabilistic inference over recursive social models and explains a range of phenomena in human pragmatic reasoning (5-8). We go beyond the previous formal wor...

Affective cognition: Exploring lay theories of emotion

“…This effect likely drives the negative correlation of the (literal) Utterance-only model with the judgments of the combinations given by participants. It is worth noting that there has been recent success in applying Bayesian models to study interpretation of non-literal language (Frank & Goodman, 2012;Kao, Wu, Bergen, & Goodman, 2014). Future work could extend our model of affective cognition to include these pragmatic effects.…”

“…This was a formalization of earlier work in belief-desire psychology, describing a lay theory of how an agent rationally chooses actions given his beliefs and desires (Dennet, 1989;Gopnik & Meltzoff, 1997). In a similar fashion, the recently proposed Rational Speech Act (RSA) model (e.g., Frank & Goodman, 2012;Goodman & Stuhlmüller, 2013;Kao et al, 2014) treats language understanding between agents as rational social cognition. Again, this work has its roots in earlier work on communication between rational agents (e.g., Clark, 1996).…”

Pragmatic reasoning through semantic inference

“…As a result, the listener will not gain any information about the weather from the speaker's choice of utterance. This reasoning can be formalized by combining the question under discussion (QUD) extension to RSA proposed in (Kao et al 2014b,a, Lassiter & Goodman 2015 with the lexical uncertainty extension. Using much the same argument, it is possible to show that, if the QUD is common knowledge between the speaker and listener, then the speaker will never communicate information about dimensions which are collapsed (irrelevant) under the QUD.…”

Adjectival vagueness in a Bayesian model of interpretation

“…The interpretive effects of rich speaker goals remain to be explored, as does varying the Question Under Discussion and numerous other possible enrichments and modifications of the model. Likewise, our approach relies crucially on intuitions about reasonable priors; it will be necessary in future work to use empirical measures to validate the choice of priors and check the predictions about their mapping to include inferring the QUD (Kao et al 2014) and choosing the relevant scale with adjectives for which there are several options (Kennedy 1997;Sassoon 2013).…”

“…The choice of QUD is constrained, but not fully determined, by overt questions, the information structure (e.g., prosody) of the utterance, and various other aspects of the history of the discourse. When further specification is required, we would opt for an expansion of the model presented here to include inference of the QUD, as discussed by Kao et al (2014). We will not deal with this additional complication here, though.…”