What to Count and When to Decide: Open Issues in Frequency-Based Models of Human Parsing

A number of theoretical models compete in explaining initial preferences in human sentence processing, each focusing on different linguistic parameters and applying different sets of principles in order to predict processing effort. Only recently, models have been proposed that base their predictions on frequency information (Mitchell & Cuetos 1991), or probability values derived from relative frequencies of linguistic items (Jurafsky 1996). In this paper, I will present both empirical evidence and theoretical considerations concerning issues of the theoretical specification of these models.