Interpreting Predictability Effects in Reading

Adrian Staub

In recent years, the psycholinguistic community has focused intensely on the role of prediction in language processing. It has been suggested that readers and listeners make predictions of both syntactic structure (e.g., Staub & Clifton, 2006) and specific lexical items (e.g., Altmann & Kamide, 1999; Van Berkum et al., 2005). At the same time, influential sentence processing models have proposed a tight connection between the difficulty of processing a linguistic element and the degree to which that element (dis)confirms a prediction (e.g., Hale, 2001; Levy, 2008).

This seminar focuses on a number of core interpretive issues in this literature. These issues include, but are not limited to, the following:

  • Of the various effects of predictability in language comprehension, which should be regarded as evidence of prediction? What are the strengths and weaknesses of data from various experimental methods (e.g., eye movements in reading, the visual world paradigm, ERPs) in demonstrating predictive processing?
  • Is there a useful theoretical distinction between prediction, on the one hand, and the notion of (pre-)activation, on the other?
  • To what extent are predictions in language comprehension discrete, with the comprehender maintaining a prediction for a specific word or structure, and to what extent are they probabilistic, with the comprehender anticipating words or structures in proportion to their subjective probability? Are there two predictive mechanisms?
  • How should we integrate the apparent evidence for predictive processing with the evidence that, except in certain special cases, both listeners and readers show extreme fidelity to bottom-up linguistic input? How should we interpret phenomena in which bottom-up processing appears to be unaffected by a predictive environment?
  • What kind of model is an expectation-based sentence processing model? Is it a cognitive process model, a statistical model, or both? Relatedly, is there evidence that statistically defined predictability (e.g., n-gram probability) actually plays a causal role in processing difficulty?

In discussing these topics, the instructor will discuss his own research on syntactic prediction (e.g., Staub & Clifton, 2006; Staub, Clifton, & Frazier, 2006), lexical predictability effects on eye movements in reading (e.g., Staub, 2011; Staub & Benatar, 2013; Kretzchmar, Schlesewsky, & Staub, submitted) anticipatory eye movements in the visual world paradigm (Staub, Abbott, & Bogartz, 2012), and cognitive modeling of the cloze task (Staub, Grant, Astheimer, & Cohen, submitted), integrating this work with a range of other empirical and modeling literature. Prior familiarity with experimental, computational, and statistical methods in psycholinguistics is not assumed.

More informations