Selective advantages of grammatical language

(presented at TMBM'99)


The transition from a finite communication system to an infinite grammatical language (like all human languages) is considered to be the last genetically coded major transition in evolution [1]. However, it remains unclear what selective advantages this infinite use of finite set of symbols ("productivity") in language offers, and what evolutionary processes may have been responsible for the growth in complexity of grammar.

Following Hashimoto & Ikegami [2], Steels [3] and others, we propose studying this problem in a framework of simulated language games. We perform such a simulation study to evaluate the effect of different selection pressures on the evolutionary dynamics. The model consists of a small set of agents, that each have a built-in capacity to store a Chomskyan rewriting grammar, to communicate with other agents by deriving and parsing short words (i.e. sequences of 0's and 1's), and to pass the grammar on to offspring. The model resembles the model presented by Hashimoto & Ikegami (1995).

The fitness of agents depends on the scores they get in the game for speaking, recognizing and being recognized. With the scoring scheme proposed by Hashimoto & Ikegami we find three distinct dynamical regimes. In the lexical regime the evolved rules map directly from the start symbol (S) on a sequence of terminal symbols (i.e. 10010). Agents develop large grammars, but the number of parsable words remains limited. In the recursive regime the grammars depend heavily on variables and recursive loops. There are few rules and many parsable words. In the variable regime almost all rules contain variables, but only few of the applied rules are used in a recursive manner. Grammars are quite large, and many words can be parsed.

By manipulating the scores agents receive, we investigated under what conditions these dynamical regimes appear. We considered five scoring schemes that reflect hypotheses on the roles language may play, namely i the total of exchanged information ("communication"), ii the total of received information ("perception"), iii the ability to influence others ("manipulation"), iv the ability to process information ("cognition") and v the ability to intimidate ("intimidation").

Although recursive loops are always only a few mutations away, the development of powerful grammars is not at all trivial. With the "communication" settings no recursive grammars are developed, unless an explicit pressure is put on innovation (i.e. only for words that are novel in one generation scores are counted). With the "manipulation" and "intimidation" settings no sustainable growth in complexity appears. However, both "cognition" and "perception" quickly lead to complicated, recursive grammars.

Although the model starts with some controversial assumptions, the results of the simulation emphasize the need to carefully consider the dynamics of language evolution and functions of language other than just "communication". The success of the "cognition" and "perception" schemes might even be considered preliminary support for theories that view linguistic communication as a side-effect of linguistic reasoning, or theories that view language production as a derivative of language comprehension.

  • [1] John Maynard Smith & Eors Szathmary, The major transitions in evolution, 1995
  • [2] T. Hashimoto & T. Ikegami, The emergence of net-grammar in communicating agents, BioSystems 38:1-14, 1996
  • [3] Luc Steels, Synthesising the origins of language and meaning using co-evolution, self-organisation and level formation, In: Hurford, J., C. Knight and M. Studdert-Kennedy (ed.) Evolution of Human Language, 1997


    home