Evolutionary Explanations of Compositional Communication

Josh Armstrong (University of California, Los Angeles)
Contemporary linguistic theory takes the generative features of language use as a central focus of study. Noam Chomsky has maintained that explaining these generative features requires an appeal to a human language faculty. In this talk, I argue that the evolution of the human language faculty depends upon social facts. In particular, I argue that there are a variety of social preconditions for the biological emergence of the human capacity for language. I argue that this social conception of the evolution of language does a better job explaining the generative features of language use than Chomsky's own favored proposal.

Travis LaCroix (University of California, Irvine)
I propose a formal framework for measuring the amount of information contained in a complex signal. The basis for this model is the relative entropy of the complex signal as a whole compared to the entropy of its constitutive parts when decomposed. On this definition, we have a clear way of determining the degree of compositionality of a complex signal and thus a criterion for whether or not a complex signal is compositional in the first place. Several complicating factors are addressed.

Mélissa Berthet (University of Neuchâtel), Geoffrey Mesbahi, Aude Pajot, Cristiane Cäsar, Christof Neumann, Klaus Zuberbühler
Call sequences have the potential to convey more information than single call utterances. Previous work has suggested that Titi monkeys combine two alarm calls into four different sequences that can provide information about predator type (aerial vs. terrestrial) and location (ground vs. canopy). We reassessed this statement by using field experiments and modeling, and found that Titi monkeys convey reliable information about predator type and location by combining calls into sequences, providing information that is not present in single call utterances. This provides new insights in the evolution of the compositionality.

Shane Steinert-Threlkeld (Universiteit van Amsterdam)
Natural language expressions fall into two categories: content and function words. While function words are essential to compositional semantics, surprisingly little has been said about their emergence. In this paper, I will show that most extant approaches to the emergence of compositional signaling fail to account for the emergence of functional vocabulary. After suggesting some conceptual reasons for this situation, I will present a model and simulation results exhibiting conditions under which such vocabulary can emerge from simple dynamics. This model captures the intuition that function words help aid communication with a limited vocabulary in the presence of contextual variability.

Jeffrey Barrett (University of California, Irvine), Calvin Cochran, Brian Skyrms (University of California, Irvine)
We present a hierarchical model for the evolution of compositional language. It has the basic structure of a two-sender/one receiver Lewis signaling game augmented with executive agents who can learn to influence the behavior of the basic senders and receiver. The model shows how agents with no preassigned representational roles might evolve a compositional language in the context of costly signaling.