Modeling the Changes of Natural Vowel Systems

Background: In the Iterated Learning Model of language change, faulty language acquisition by children is assumed to be the main driver of change. Many factors have been thought to affect the evolution of a language, and so a speech community which includes a generation of children acquiring language is best modeled as a "complex dynamic system." The system of vowel sounds in a language, in particular, may change fairly rapidly or remain relatively stable, depending upon the system and the parameters that affect learning. Methods: We programmed a computer simulation of a community of speakers which includes generations of adults and children (learners). The "language" of the community was limited to a system of vowel sounds only, characterized by three primary acoustic features. Children were tasked with learning the vowels from the adults, but could only succeed at doing so approximately, with random variation introduced. The "fitness parameters" that influenced learning were the acoustic closeness of vowels and the linguistic importance, or "functional load," ascribed to vowel contrasts. Results: By varying the fitness parameters and number of vowels in the language model, it was possible to obtain realistic effects over many generations, ranging from long-term stability (as in e.g. Spanish) to constant vowel shifting (as in e.g. English). Shifting behavior is positively correlated with both poor learning accuracy and the number of vowels in the language. Conclusions: This study is but a first step toward modeling evolution of natural language sounds in a rigorous fashion. A variety of factors have been speculated to affect this evolution in the past; here we provide evidence that validates some of these proposals.