A computational analysis of the constraints on parallel word identification

The debate about how attention is allocated during reading has been framed in as: Either attention is allocated in a strictly serial manner, to support the identification of one word at a time, or it is allocated as a gradient, to support the concurrent processing of multiple words. The first part of this article reviews reading models to examine the feasibility of both positions. Although word-identification and sentence-processing models assume that words are identified serially to incrementally build larger units of representation, discourse-processing model allow several propositions to be co-active in working memory. The remainder of this article then describes an instance-based model of word identification, Über-Reader, and simulations comparing the identification of single words and word pairs. These simulations indicate that, although word pairs can be identified, accurate identification is restricted to short high-frequency words due to the computational demands of both memory retrieval and limited visual acuity.