Pragmatic Back-and-Forth Reasoning

Abstract

We survey a number of game theoretic models that capture speakers' and listeners' pragmatic back-and-forth reasoning about mutual beliefs and linguistic behavior (i.e., utterance choice and interpretation). Two types of models are presented. Firstly, models that rely on rationality of choices and beliefs therein are shown to predict general pragmatic usage and inference patterns. Secondly, we introduce a new probabilistic variant of these reasoning models which parameterizes agents' rationality (and belief therein), thereby enabling fine-grained quantitative predictions of speaker and listener behavior to match data from psycholinguistic experiments.

Coevolution of Lexical Meaning and Pragmatic Use

“…Pragmatic interlocutors instead engage in mutual reasoning to inform their choices. Recent game‐theoretic (e.g., Benz, ; Benz & van Rooij, ; Franke & Jäger, ; Jäger, ) and probabilistic models of rational language use (e.g., Frank & Goodman, ; Franke & Jäger, ; Goodman & Frank, ) capture different types of pragmatic behavior in a reasoning hierarchy. In the following sections, we aim at a general formulation of speaker and listener behavior which is as simple and as practical as possible for our current purposes but still in line with both the game‐theoretic and the Bayesian traditions.…”

Reply to commentaries

“…Although there are plenty of accounts of inferences to likely or stereotypical interpretations (e.g., Parikh 1991Parikh , 2000 and M-implicatures (e.g., van Rooij 2004;Jäger 2014) in the game theoretic tradition, some probabilistic variants of game theoretic reasoning may run into technical problems when they try to add probabilistic choice functions (see Franke and Jäger 2014, for treatments of M-implicature). There are even some empirical problems, but these too can be overcome by extending probabilistic pragmatic models in ways that are, arguably, independently motivated and intrinsically plausible (Bergen, Levy, and Goodman to appear).…”

“…Many game theoretic models that contain higher-order pragmatic reasoning of the kind envisaged by Jäkel and Liu exist (e.g., Franke 2011;Pavan 2013;Rothschild 2013;Jäger 2014), including models that consider probabilistic choice rules at each level of iteration (e.g., Bergen, Levy, and Goodman 2012;Franke and Jäger 2014;Bergen and Goodman 2015; Goodman to appear). The probabilistic speaker and listener rules given in our paper would map onto a level-1 speaker and a level-2 listener in so-called iterated best response models (Franke and Jäger 2014). Generally, a level-(n +1) agent responds (approximately) optimal to the (hypothetical) behavior of a level-n agent.…”

“…Technically, this situation will not occur if the speaker production probabilities are defined by a soft-max function with finite rationality parameter λ based on finite expected utilities. In that case, every message has a positive probability of being used in every state, and the listener will therefore never be surprised by any message (e.g., Franke and Jäger 2014). So, technically the problem is avoided by some versions of probabilistic pragmatics.…”

The Drive towards Application of Pragmatic Perspective in Educational Research: Opportunities and Challenges

“…The philosophical roots of modern-day pragmatism lie in the objections of some features of logical positivism and the celebration of a moderate level of specificity and application of concrete situations (Franke& Jager, 2013;Pierce, 1935). To be useful, a pragmatic approach emphasizes key issues that are important to address for successful implementation or evaluation of specified educational interventions (Hamati-Ataya, 2012).At all times pragmatism has been considered as a process rather than a result, and it is an approach that attempts to validate all participatory variables in a solution specific to educational problems.…”