Evolution of language structure : survival of the fittest in a statistical environment

Abstract

The structure and other characteristics of natural language can be illuminated by an investigation into the real world conditions within which language functions. Language is taken as a medium in which messages are produced, and we investigate how it is best adapted to efficient coding. We examine naturally occurring phenomena and suggest how their exploitation would confer an evolutionary advantage. The main focus of this paper is approached by first examining a related phenomenon: the expected distribution of some linguistic entities. It is shown how this is relevant to a language representation that can facilitate the efficient transfer of information. This leads to an investigation into how naturally occurring events, such as pauses in speech, can be utilised to improve the transmission of information. Using arguments based on comparative entropy measures we show that there seems a certain inevitability to the development of structured language. This is a novel application of well known tools from Information Theory, and demonstrates how a relationship between prosodic information and the organisation of language might be expected to develop.