Leads on the Mechanisms of Communication

James K. Hazy, June 8, 2012

A newly published Institute Interpretive Brief describes recent research that combines information theory with Bayesian logic to provide insights into effective communications among human beings. The two studies described (Frank & Goodman, 2012; Kemp & Regier, 2012) offer distinct yet complementary insights about some possible general principles of language. Their results might finally point to an answer to a philosophical question that goes back to the ancient Greeks: How does naming things with which human beings interact—that is, convention—relate to what is really out there independent of any human interaction with it—that is, objective reality(Matson, 2011)?

Frank and Goodman (2012) implicitly consider this question by describing a “language game.” It is played with two players whose communication is considered in the context of three conditions: The speaker condition, the salience condition, and the listener condition. The first and third are self-explanatory. The second, however, requires some explanation.

The salience condition represents the prior probability that the chosen word refers to a particular thing or relationship. This condition can be thought to reflect the reaction of a third party who is not aware of any prior conventions and therefore has no a priori idea what any particular word or utterance by the speaker might be intended to represent. According to the authors, the salience condition captures “the common knowledge the speaker and listener share, as it affects the communication game” (p. 998). The idea of the Frank/Goodman model is that speakers and listeners communicate to coordinate their efforts within commonly experienced events. Speakers choose words to encode a chosen “partitioning” of mutually observable differences within a shared but ambiguous context. In order “to capture some of the richness of human pragmatic inference in context,” they model communication assuming “a listener can use Bayesian inference to recover a speaker’s intended referent… in context…given the speaker’s uttered word” (p 998).

To explore how systems of words or categories of “partitioning” might be selected, Kemp and Regier (2012) use the categorization (or partitioning) of kinship relationships as the physical and social context that is relevant to communicate among individuals during human interactions. 

Using the basic framework described earlier, the Kemp and Regier (2012) study looks at the effectiveness of communication strategies with respect to kinship relationships. Their results  suggest that evolutionary processes would seem to favor communication strategies that transfers information that partitions kinship relationships in a manner that optimally combines two competing factors: relative simplicity (to minimize cognitive load) and informative content (to enable maximum congruence in the inferences made by both sender and receiver). 

When evaluating their data set of 487 actual approaches to partitioning kinship relationships in world languages, they found that, as predicted by their evolutionary model, actual approaches tended to dominate other possible approaches when plotted, and these fell along what might be called an efficient frontier for communicating, optimally trading-off complexity against information content. Interestingly, this was true in general, but also for specific relationships like paternal-niece. Further, it was more likely to be true for close (more salient) relationships like parent than for more remote (less salient) ones like a sister’s, husband’s daughter.

In the Perspective article in Science that accompanied these studies, Levinson (2012) describes the importance of these two studies adding, “Neither model tells us where the categories come from; they merely place constraints on what is good to think and good to communicate. Here, they could be usefully complemented by another recent line of work in the evolutionary modeling of culture” (p.989). In short, this line of research is promising, but it only represents a beginning.


Frank, M. C. & Goodman, N. D. (2012). Predicting Pragmatic Reasoning in Language Games. Science, 336, 998.

Kemp, C. & Regier, T. (2012). Kinship Categories Across Languages Reflect General Communications Principles. Science 336, 1049-1054.

Levinson, S. C. (2012). Kinship and Human Thought. Science, 336, 988.

Matson. W. (2011) Grand Theories and Everyday Beliefs: Science, Philosophy and their Histories. New York: Oxford University press.