Computational Modeling of Spoken Language Processing: A hands-on tutorial

Computational models provide a means for concretely specifying theoretical assumptions, and examining their complex interactions via simulation. Ideally, models help explain existing data and provide novel predictions that guide further research. In this tutorial, we will cover principles of conducting simulations, and assessing model success and failure. We will use the domain of spoken language processing for our examples. Models have been particularly useful in the domains of speech perception and spoken word recognition, where theories and the signal are both complex. The TRACE model (McClelland & Elman, 1986) has the greatest breadth and depth of any model in those domains, and despite well-known shortcomings, continues to be used productively (e.g., to model time course data from eye movements). It also has much in common with other models characterized by activation-competition dynamics (e.g., Shortlist [Norris, 1994]). This makes TRACE ideal for introducing principles of computational modeling of spoken language processing. Participants will learn skills for carrying out computational modeling of speech perception with TRACE. Novices will be introduced to principles and practical aspects of using models, and can expect to leave the tutorial prepared to begin modeling their own data. Experienced modelers will add new techniques to their repertoires. Participants will simulate speech experiments and link results to human data using the recently developed jTRACE modeling software (Strauss et al., 2007)