Auditing the Auditor’s Economy

Much of linguistic pragmatics is built upon the idea of contrasting pressures, where the speaker must choose an utterance to minimize her cost of articulation while at the same time maximizing informativity (or minimizing the cost to the listener.) Out of these pressures arise many of the inferences that can be derived by a listener about what a speaker means by what she says. Each of these competing pressures is necessary, as communication would be infeasible without the balance. To ignore the cost of articulation would result in the speaker being endlessly verbose, as saying more makes it more likely that the listener will correctly interpret your utterance. To ignore the cost of interpretation would result in the speaker saying nothing, but the listener therefore being left without any information at all. Zipf (1949) provided the initial insight into this phenomenon, and used it to derive an entire theory of human behavior. He characterizes the trade-off based on the number of words mapped to each meaning in a language or the number of words a speaker would have to know in order to produce language. From this, he derives his famous power law that captures a stable relationship between the length of a word (which may be related to the cost of producing it) and the frequency of occurrence of the word in the language. This property reflects what later became known as communicative efficiency, as it predicts that a language evolves to minimize effort by making the more frequently produced words shorter. Later, Grice (1975) captured this in his famous maxims of pragmatics from which he derives a theory of how the interpreted meanings of words can go beyond the literal and interact with context. Horn (1984) returned to the Zipfian duality, and reframed these Gricean maxims into a pair of competing forces–the Q and R principles–which capture Horn’s claim that the optimal utterance is both necessary and sufficient. The Q-principle states that you should say as much as is necessary and the R-principle states that you should say no more than is sufficient. The R-principle has an obvious motivation for the speaker, namely that she minimize her own production effort, whereas the Q-principle is instantiated less directly, since it is based on improving the listener’s chance of correctly interpreting the speaker’s intended meaning. If we view this interaction game-theoretically, we see that the speaker has no direct reason to care about the listener’s utility, as it is the listener’s and not hers.