Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
378627 | Cognitive Systems Research | 2006 | 26 Pages |
Random variables and probabilistic decision making are important elements in most theories of reading eye movements, but they tend to receive little theoretical attention. This paper attempts to address this problem by introducing the Stochastic, Hierarchical Architecture for Reading Eye-movements (SHARE). The SHARE framework formalizes reading eye movements as observable outcomes of a latent stochastic process. By modeling eye movements as time-series random variables, the goal of the model is to uncover statistical regularities in the data, which help to identify conditions and constraints the underlying mechanism must satisfy. In the univariate analysis, it is shown that a 3-component Lognormal mixture model provides a good fit to the marginal distribution function of fixation duration, and a hierarchical model is required for modeling saccade length. As a comprehensive model of reading eye movements, SHARE was implemented as an Input–Output Hidden Markov model. With a few simple hypotheses, SHARE is able to capture reading eye-movement patterns of beginning readers and proficient adults, and to reproduce well-known psycholinguistic effects. The rationale of the model, its relations with other modeling endeavors, and its implications are discussed.