42 Wednesday, March 14: Poster Abstracts Effects of frequency, predictability, & length in a rational model of eye movements in reading

Eye movement control in reading is one of the best and most naturalistic examples of the adaptive deployment of linguistic knowledge for real-time comprehension. Here we present a model of eye movement control in reading as rational action, which seeks to understand reading behavior as resulting from the efficient way to achieve reader goals. Specifically, we describe empirical benchmarks for the first rational model of eye movements in reading that makes predictions for the full range of the eye movement record. Unlike E-Z Reader (Reichle et al., 2003), SWIFT (Engbert et al., 2005), and Glenmore (Reilly & Radach, 2006), the model directly optimizes its behavior to best achieve the reader’s goals, here characterized as accurate and rapid identification of the contents of the text. Unlike Mr. Chips, the only previous rational model in this domain (Legge et al., 1997), the model makes predictions about not only fixation locations but also fixation durations. In the model, readers use Bayesian inference to combine two sources of information – (1) probabilistic language knowledge (the prior), and (2) noisy perceptual input about the text (the likelihood) – to form and repeatedly update a posterior distribution over possible text contents. At each discrete time step, readers can continue fixating the eyes ’ current position to obtain more perceptual input, move the eyes to a position of the model’s choice, or stop reading and end the trial. The model includes realistic physical constraints including motor error, saccade planning time, and an asymmetric visual acuity curve. Readers obtain noisy letter-identity informatio