Andrew Wedel

Modeling sublexical contrast maintenance as an emergent effect of lexical category competition.

Abstract. Patterns in complex systems often arise through interacting positive and negative feedback loops, both within and across levels of organization. I present here a model for the development of concise inventories of contrastive sublexical (e.g., phoneme) categories in languages through interacting processes that select for contrast at the higher lexical level.

The transmission of information requires a contrastive code (Shannon1949). Standard linguistic models locate the building blocks of contrast in a set of atomic features which combine to form higher order categories. The system of sublexical contrast cannot be concretely pre-programmed because different languages use widely divergent systems of sublexical categories, including the gestures used for sign. As a consequence, many linguists have theorized that sublexical contrast must be maintained by some innate mechanism that monitors abstract featural contrastiveness itself, intervening to prevent loss of contrast when it would be too detrimental to the function of the overall system.

Within the standard generative model, however, there has always been a conceptual difficulty with this implicit causal connection between contrast at the level of lexical categories and contrast at the lowest sublexical level. This is due to the bottleneck in information flow between levels enforced by the notion that lexical categories consist only of a list of sublexical category labels. As a result of this bottleneck, it has not been clear how success or failure in the communication of meaning feeds back to the sublexical level to influence trajectories of gradual sound change.

However, a growing body of evidence indicates that variation is retained at multiple levels of organization. This cross-categorization of variation between levels opens up what appears at first to be a counter-intuitive possibility: that contrast maintenance arises interior to the chain of levels of organization, rather than at the lowest level. Under this model, any bias in the production or perception of lexical variants toward greater contrast can result in the indirect selection for a system of contrastive sublexical variants as well. This is conceptually similar to natural selection at the level of the individual resulting in global changes in gene variant frequencies in the population. Just as genes that promote successful reproduction of individuals are more highly represented in the next generation, word variants that are more consistently categorized have a greater influence on the trajectory of change in their associated sound categories.

Although contrast is required for language function, the mechanisms that support contrast between lexical categories need not be driven teleologically by the functional need for contrast. As an example, I show how experimentally supported features of categorization indirectly promote contrast maintenance between competing categories. Specifically, a series of simulations show that given (i) a perception-production feedback loop, (ii) cooperative influence of similar memory traces on production and perception, and (iii) slow memory decay, trading of variants at the boundary between competing categories promotes separation between the categories over time.

To illustrate how contrast maintenance at the lexical level can indirectly support contrast maintenance at a sublexical level, I incorporate these mechanisms into a simple model of lexical category evolution where each lexical form is composed of a series of sublexical units. The model incorporates a leniting bias with the result that in the absence of category competition, the lexical and sublexical levels converge to homophony over time. When lexical categories compete for percepts, however, lexical forms tend to remain contrastive, and with them, their component sublexical units. These results raise the possibility that contrastive phoneme inventories arise as an epiphenomenon of processes in usage supporting contrast at the lexical level.