Notes on Franke 2013: Game Theoretic Pragmatics
The general logic of a game theoretic explanation of a pragmatic phenomenon is this: (i) the conversational context is modelled as a game between speaker and hearer; (ii) an adequate solution concept then selects the to-be-explained behavior in the game model.
Franke thinks of game theory as the study of strategy games—games where player outcomes depend on the actions of other agents, and thus player decisions hinge on predictions about these future actions. (“Mutual futures modeling.”) He gives a simple example of a coordination game using rubber boots and blue suede shoes, then writes:
Telling stories, giving commands and asking questions, also listening and figuring out what Joe meant when he said ‘‘kind of’’ — all of these linguistic activities have something essential in common with the rubber-or-suede game. Here and there, several agents have to make a decision how best to act given their preferences and their expectations about the others’ behavior. As speakers we usually have a preference for getting our point across, or for persuading our hearers of our point of view. A speaker’s action choices are naturally constituted by the set of possible utterances given by our grammatical competence, or some relevant subset thereof that interlocutors commonly attend to. Hearers may seek entertainment or information.
Grice’s maxim of concision (or “quantity”) falls naturally out of the pragmatic frame—it is less an imperative or norm, and more an expectation that emerges naturally from a desire for efficient, efficacious communication. Given this efficiency, we as listeners can assume that unnecessary information provided is “doing something”—is there for a reason—an idea that sometimes gets called Horn’s division of pragmatic labor:
When a marked (relatively complex or prolix) expression is used, when an alternative (simpler) expression is available, this is typically interpreted as conveying a marked message which the unmarked, simpler alternative would or could not have conveyed.
Franke argues that Schelling was critical in inspiring early game-theoretic linguistics:
David Lewis was the first who applied game theoretic ideas to philosophical questions about language in his seminal work on convention (Lewis 1969). Inspired by the work of Schelling (1960) [Strategy of Conflict], Lewis tried to give a non-circular naturalistic grounding of the notion of conventional meaning in terms of repeated plays of signaling games, thus answering challenges to conventionalism put forward by, among others, Russell (1921) and Quine (1936).
Franke distinguishes between an evolutionary model of game-theoretic pragmatics, and a more rationalistic model. He then further distinguishes between evolutionary microdynamics and macrodynamics. Microdynamics are learned or trained, a “dynamic change in terms of how each agent adapts her behavior over time.” We might alternatively call this training process “cybernetic.” Macrodynamics, meanwhile, involves preferential replication of successful strategies over successive generations—standard natural selection stuff.
Rational pragmatic language theory was pioneered by Prashant Parkih in the 1990s and 2000s. Parkih advances, among other arguments, that (1) for a view of communication in which an “equilibrium between the speaker’s intended meaning and the hearer’s interpretation” is critical; (2) the central task of the listener, in interpreting speech, is disambiguation; (3) the speaker and hearer together construct a “game of partial information.” In the rationalistic frame, behavioral predictions follow from an understanding of speaker beliefs, dispositions, rationality, etc.