Language as a Complex Adaptive System
John H. Holland
University of Michigan
Tuesday, 11:00 am
Nothing is less real than realism. Details are confusing. It is only by selection, by elimination, by emphases that we get at the real meaning of things. –Georgia O’Keefe
This approach to language acquisition and evolution concentrates on language as a social phenomenon. To this end, it uses agent-based models wherein the agents adapt to an environment consisting of scattered resources and other agents — such models are usually called situated. Interaction between agents already having some linguistic ability, teachers, and agents without linguistic ability, learners, must serve for the propagation of language. In these models, language only evolves if its acquisition by a group of agents enables them to better collect resources — there is no a priori value to language.
Language, for present purposes, is the ability to produce utterance sequences wherein different sequences have different predictable effects on other agents. That is, an agent can produce a wide variety of responses in other agents through combinatoric (grammar-like) use of a limited set of utterances. Agents start with only a few familiar pre-primate capacities
(i) an ability to imitate, (ii) mutual awareness of shared attention when two or more agents are focused on the same salient object, and (iii) an inherent ability to distinguish actions from objects. If these abilities are placed on a quantitative scale, ranging from total lack of the ability to full development of the ability, there may be a sharp inflection point as a vector combination of these abilities increases. This inflection point would offer an explanation of the “sudden” appearance of structured language as we move from closely related primates to humans.
On a larger scale, the models proposed are good candidates for examining several emergent phenomena associated with complex adaptive systems (cas) in general:
(i) Robustness: Despite the established fact that individuals in a language group vary considerably in the grammars and expressions they use, communication proceeds smoothly under a wide variety of conditions.
(ii) Networks of interaction: The language-mediated formation of social groups and the “hub/authority” patterns in the internet are just two examples of generation of networks of increasing complexity and diversity.
(iii) Meanings as equivalence classes over environmental patterns: As languages develop and change, we see an increasing ability to distinguish different repeating patterns in the environment, especially social patterns, ranging from small groups to corporations and nations.
John Holland is a MacArthur Fellow, a Fellow of the World Economic Forum, and co-chairman of the Science Board of the Santa Fe Institute. He is known worldwide as the “father of genetic algorithms” and is the author of HIDDEN ORDER: HOW ADAPTATION BUILDS COMPLEXITY.
Prof. Holland is the guest of the Computation Institute this coming Tuesday, January 29 at 11am, in RI 480 (Research Institute, 5640 South Ellis Avenue), as part of its Deep Disciplinary Dive on language and computation.