My research program attempts to answer the following question: What determines when and where you move your eyes as you are reading sentences like this one?  Although this question may seem very narrow or esoteric, to really answer this question, it is first necessary to understand the perceptual, cognitive, and motor processes that play a role in guiding the eyes during reading.  At a minimum, these processes include the visual processes that are necessary to encode the individual words from the printed page, the cognitive processes that are necessary to retrieve the pronunciations and meanings of these words from memory and then use this information to construct the meaning of the text, and the motor skills that are necessary to program and execute the eye movements that move the eyes from one word to the next.  Of course, it is also necessary to understand how these processes become coordinated through years of practice to produce the complex yet systematic patterns of eye movements that are observed with skilled readers.  Thus, by studying eye movements in a well-defined task like reading, I am not just trying to understand eye-movement behavior; I am trying to understand the “eye-mind link,” or how cognition interacts with perception, on one hand, and motor control, on the other.

The focus of my research to date has been directed toward developing a computational model of eye-movement control during reading—E-Z Reader (Pollatsek, Reichle, & Rayner, 2006; Rayner, Ashby, Pollatsek, & Reichle, 2004; Reichle, Pollatsek, Fisher, & Rayner, 1998; Reichle, Rayner, & Pollatsek, 1999, 2003; Reichle, Pollatsek, & Rayner, 2006; Reichle, Warren, & McConnell, 2009; for a review, see Reichle, 2011).  This model is a mathematical description of how perception, cognition, and oculomotor factors determine when and where the eyes move during reading, and can be used to simulate the eye-movement behavior of readers.  The main assumptions of the model are that words are identified one at a time during reading, and that the process of identifying words is the “engine” that drives the eyes forward during reading.  Thus, in contrast to alternative models of eye-movement control during reading (for reviews, see Reichle & Rayner, 2002; Reichle et al., 2003; or the 2006 special issue of Cognitive Systems Research), the E-Z Reader model posits a fairly tight coupling between cognition and eye movements (Reichle, 2006).  These assumptions have made the model important to reading researchers because it is an existence proof that cognition (which tends to be fairly sluggish) can mediate the rapid, moment-to-moment “decisions” about when and where to move the eyes during reading.  By doing this, the model also helps validate the use of eye-tracking technology to study cognitive processes during reading (for reviews of this research, see Rayner & Pollatsek, 1989; and Rayner, 1998, 1999).

I have also been using the E-Z Reader model as a theoretical framework for examining a variety of issues related to reading.  For example, the model has already been used to examine how the meanings of morphemically complex words (e.g., compounds words like “cowboy”) are accessed during reading (Pollatsek, Reichle, & Rayner, 2003).  Similarly, the model has also been used to examine how the syntactic and/or semantic constraints of a words sentence context affect the processing of words that occur frequently versus infrequently in printed text (Rayner et al., 2004), and words that are ambiguous because they have two or more meanings (Reichle, Pollatsek, & Rayner, 2007).  Most recently, I have been working on a version of E-Z Reader that includes a post-lexical (integration) stage of langage processing (Reichle, McConnell, & Warren, 2009; see also my forthcoming book, Reichle, 2012).

Although the E-Z Reader model precisely describes the eye movements of skilled readers, it is agnostic about how this behavior develops.  To address this issue, I have recently started another line of computational modeling to examine how the complex patterns of eye movements indicative of skilled readers might develop as a result of a few simple constraints related to the task (Liu & Reichle, 2010; Reichle & Laurent, 2006).  In this work, an artificial reading “agent” (i.e., an artificial intelligence that is capable of modifying its own “behavior” via a reinforcement machine-learning algorithm) was given the task of learning how to move its “eyes” so as to “read” as quickly as possible.  After extensive practice performing this task, the agent learned to move its eyes in a manner that resembled the eye-movement behavior of skilled human readers.  These simulations have provided new interpretations for some contentious findings in the literature (e.g., why readers sometimes spend more time looking at word n if word n+1 is skipped than if word n+1 is fixated) and may provide an alternative method for exploring questions that are currently the focus of debate among reading researchers (e.g., Is attention allocated to one or more than one word at a time during reading?; see Reichle, Liversedge, Pollatsek, & Rayner, 2009).

More recently, I have been interested in using the model to understand how eye-movement control in reading is different from eye-movement control in other visual-cognitive (but non-reading) tasks (e.g., searching for the letter O in linear arrays of Landolt Cs; Williams & Pollatsek, 2007). This work (Reichle, Pollatsek, & Rayner, 2012; Reichle, Rayner, & Pollatsek, 2012) suggests that the basic perceptual and oculomotor contraints are the same across tasks, but that the signals to initiate saccadic programming versus shift attention may be less tightly coupled in reading (and possibly other highly practiced reading-like tasks).That is, in the context of reading, readers may develop the skill of using word (stimulus) familiarity to initiate saccade programming in a manner that is not possible in non-reading tasks. This use of familiarity affords efficient reading to the extent that indiviual words only have to be fixated long enough to insure that they will be identified, rather than actually being fully identified, per se.

I have also been using other methodologies to examine the relationship between cognition and eye-movement control. For example, a recent experiment on mindless "reading" (Reichle, Reineberg, & Schooler, 2010) instructed participants to respond to periodic computer prompts as to whether or not they had been "zoning out just now?" while they were reading a novel that was being displayed on a computer monitor. The eye movements collected during intervals of mindless reading (i.e., immediately prior to "yes" responses) were different from those collected during intervals of normal reading (i.e., immediately prior to "no" responses; relative to the latter condition, fixation durations during mindless reading tended to be longer in duration and less modulated by variables that are known to affect fixation durtions during normal reading (e.g., word frequency). This suggests that the mind and eyes become decoupled during mindless reading, which in turn reaffirms the conclusion that the two are fairly tightly coupled during normal reading.

Another example involves a recent experiment using event-related (brain) potentials, or ERPs, to examine the time course of lexical processing immediately prior to saccades from one word to another (Reichle, Tokowicz, Liu, & Perfetti, 2011). Consistent with predictions of the E-Z Reader model, this experiment found markers of lexical processing (i.e., differential effects of word frequency measured by ERPs) at approximately 125 ms prior to the saccades, and within 102 ms of stimulus onset. This result suggests that lexical processing occurs very rapidly, leaving enough time (125 ms) to program a saccade from one word to the next during reading (see also Reingold, Reichle, Gaholt, & Sheridan, submitted). Futhermore, analyses to localize the sources of these word-frequency effects suggest that the neural generators are in the left fusiform gyrus, or classic orthgraphic "word form area". The results of this analyses are shown in the figure, below (with active cortical areas indicated in red).