Description:

This dissertation provides evidence that reading is best explained as rational gathering of visual information to identify words efficiently. Although empirical evidence from human reading research suggests a close link between eye movements and cognitive process, it is not clear how readers decide when and where to move their eyes as a function of their cognitive states, and why they make certain eye movement decisions the way they do. This dissertation presents qualitative evidence that a rational model explains eye movement phenomena that cannot be explained by dominant models, and quantitative evidence that the rational model better predicts eye movement phenomena. In Study 1, we demonstrate that human readers seek visual information about the uncertain part of a word instead of always targeting the word center when they decide where to move their eyes, a finding which cannot be explained by dominant models of eye movements. In Study 2, we demonstrate that the letter position that minimizes reading time varies as a function of the structure of the particular word, which is only predicted by a rational model. In Study 3, we provide quantitative evidence that a rational model predicts humans' decision to skip a word better than a model based on the standard account of word identification. In Study 4, we provide quantitative evidence that predicting readers' comprehension from eye movements through a rational model is more robust than through using the features from a dominant model of eye movements. Altogether, this dissertation provides evidence that the rational model of eye movements in reading, which captures the complicated interaction between visual and linguistic information and optimizes eye movement decisions accordingly, is able to better explain and predict human eye movements than dominant models of eye movements that take a standard account of word identification. Therefore, the eye-mind link between eye movements and online language processing is naturally understood as rational eye movement decision making based on the knowledge from a probabilistic word identification process.