UC San Diego
The interaction of language processing and eye movement control during reading
- Author(s): Abbott, Matthew James Hansen
- Advisor(s): Ferreira, Victor S
- et al.
In this dissertation, I addressed three questions regarding the impact of language processing on eye movements in reading: (1) What information from the sentence context do readers recruit prior to skipping? (2) Are words like "the" unique with regard to skipping because they are so short? (3) Does postlexical integration strictly follow word identification? Throughout, I recruit the E-Z Reader 10 model (Reichle, Warren, & McConnell, 2009) for testable predictions. In Chapter 2, I demonstrate that the influence of the sentence context on skipping is limited to predictability, and not deeper semantic or syntactic processing. Readers skipped over invalid, contextually infelicitous parafoveal previews of "the" more often than valid previews of three-letter words, even when the target words were predictable. In Chapter 3, I show that this effect generalizes to longer function words like "that" and "there", which highlights the role of parafoveal processing relative to oculomotor or contextual constraints (e.g., syntactic or semantic fit) in word skipping. Comparing Experiment 3.2 to Experiment 3.1, it is also evident that readers prefer to skip function words over content words, regardless of their length. The data presented in Chapters 2–3 pose a deeper question about the relationship between language processing and eye movement control that indicates a possible constraint on the architecture: Are word identification and postlexical integration isolated, separable processes? This issue is examined in Chapter 4 by jointly manipulating word frequency and plausibility. I demonstrate probabilistic evidence against an influence of plausibility on word skipping, and in favor of an additive relationship between frequency and plausibility (and against an interactive relationship), by computing Bayes factors. In summary, I paint a picture in which word skipping reflects a hedged bet that identification will ultimately be successful, and all postlexical processing (i.e., of syntax and semantics) follows lexical access.