Stress-testing Multilink – or how resting level activation and word frequency interact

Project description

Multilink is one of the leading computational models in bilingualism. It simulates the recognition and production of words of different lengths and frequencies in tasks like monolingual and bilingual lexical decision (i.e., decision about whether stimuli presented on the screen are a word or not, e.g. stone’, ‘rcks’), word naming, and word translation production. A core component of Multilink is the so-called resting level activation (RLA) of a word. In general, words that are encountered or used more frequently have higher RLA levels than words that occur less often. An open question, however, concerns the nature of this relationship. Are RLA and word frequency linearly related (as is often assumed), or is the relationship nonlinear? If the latter is true, what shape is the nonlinearity?

To answer this question, we will conduct a comprehensive exploration of Multilink’s predictions using Approximate Bayesian Computation (ABC). This will not only provide us with estimates of the most likely parameter values for Multilink, but also their uncertainty. Moreover, the results will further elucidate the cognitive processes that underlie e.g. visual word recognition, since different shapes of the relationship between RLA and word frequency necessitate different mechanisms. The project will be undertaken in close collaboration with Prof Kathy Conklin in the School of English.