This commentary focuses on two aspects of eye movement behaviour that E-Z Reader 7 currently makes no attempt to explain: the influence of higher order psycholinguistic processes on fixation durations, and orthographic influences on initial and refixation locations on words. From our understanding of the current version of the model, it is not clear how it may be readily modified to account for existing empirical data.

Reichle et al. specify two assumptions as being basic to E-Z Reader: Words are sequentially attended during fixations, and saccades are triggered by a cognitive event. We point out that there is little evidence for the first assumption and counterevidence for the second. Also, the labile/nonlabile stage distinction in saccade preparation seems to be contrary to current evidence. An alternative explanation of saccade onset times in reading assumes that saccades are strategically generated, independent of language processing, but are delayed on a probabilistic basis by processing difficulties.

E-Z Reader fits key parameters from one corpus of eye movement data, but has not really been tested with new data sets. More critically, it is argued that the key mechanism driving eye movements – a serial process involving a proportion of word recognition time – is implausible on the basis of a broad range of experimental findings.

The current model appears comprehensive but is probably not applicable to a writing system like Japanese, which has unspaced text, because the model is mainly based on English. The span size difference (smaller for Japanese than for English) may be a result of high-level working memory-based attentional processing and not of low-level processing. Further, neural correlates of the model are discussed in terms of central executive function.

I discuss evidence for direct linguistic control of refixations and argue that the E-Z Reader model's account of refixations requires elaboration or revision.

Reichle et al. argue that the mechanism that determines where to fixate the eyes is controlled mostly by low-level processes. Therefore, unlike other competing models (e.g., the SWIFT model), the E-Z Reader model cannot account for “global” regressions as a result of linguistic difficulties. We argue that the model needs to be extended to account for regressive saccades.

E-Z Reader achieves an impressive fit of empirical eye movement data by simulating core processes of reading in a computational approach that includes serial word processing, shifts of attention, and temporal overlap in the programming of saccades. However, when common assumptions for the time requirements of these processes are taken into account, severe constraints on the time line within which these elements can be combined become obvious. We argue that it appears difficult to accommodate these processes within a largely sequential modeling framework such as E-Z Reader.

E-Z Reader 7 is a processing model of eye-movement control. One constraint imposed on the model is that high-level cognitive processes do not influence eye movements unless normal reading processes are disturbed. I suggest that this constraint is unnecessary, and that the model provides a sensible architecture for explaining how both low- and high-level processes influence eye movements.

As the number of computational models of eye-movement control in reading increases, so too will their coverage and complexity. This will make their comparison and testing increasingly challenging. We argue here that there is a need to develop a methodology for constructing and evaluating such models, and outline aspects of a possible methodology.

It is argued here that a critical prediction of the E-Z Reader model is that experimental manipulations that disrupt early encoding of visual and orthographic features of the fixated word without affecting subsequent lexical processing should influence the processing difficulty of the fixated word without producing any processing effect on the next word. This prediction is explained and illustrated.

Although the E-Z Reader model accounts well for eye-tracking data, it will be judged by new predictions and consistency with evidence from brain imaging methodologies. The stage architecture proposed for lexical access seems somewhat arbitrary and calculated timings are conservatively slow. There are certain effects in the literature that seem incompatible with the model.

We argue that models of reading should be based on anatomical reality, namely, the fact that both eyes are used in reading; and the observation that the human fovea is precisely vertically split, and projects each half of a fixated word to the contralateral hemisphere.

We tested whether the E-Z Reader model can be generalised to the French language. The simulation showed that the model can account for the frequency effect. The predictability effect is moreover accurate for word skipping, but not for fixation times. We think that this model is psychologically plausible for certain aspects of reading and we have used it to evaluate the performance of dyslexic readers.

The E-Z Reader 7 model is powerful but incomplete. When programming the saccade to the next word, we take into account the familiarity of the letter sequences at the beginning of that word. This landing position effect is well established, but is neglected in the model. A possible locus for the effect is suggested within the E-Z Reader framework.

This commentary focuses on the two basic assumptions of the E-Z Reader model, discussing the possibility that adjacent words in reading may be processed in parallel rather than serially, and presenting evidence against a cognitive control of eye guidance in reading.

The issues the commentators have raised and which we address, include: the debate over how attention is allocated during reading; our distinction between early and late stages of lexical processing; our assumptions about saccadic programming; the determinants of skipping and refixations; and the role that higher-level linguistic processing may play in influencing eye movements during reading. In addition, we provide a discussion of model development and principles for evaluating and comparing models. Although we acknowledge that E-Z Reader is incomplete, we maintain that it provides a good framework for systematically trying to understand how the cognitive, perceptual, and motor systems influence the eyes during reading.

In this commentary on Stanovich & West (S&W) we call attention to two points: (1) Freud's original dual process theory, which antedates others by some seventy-five years, deserves inclusion in any consideration of dual process theories. His concepts of primary and secondary processes (Systems 1 and 2, respectively) anticipate significant aspects of current dual process theories and provide an explanation for many of their characteristics. (2) System 1 is neither rational nor irrational, but instead a-rational. Nevertheless, both the a-rational System 1 and the rational System 2 can each have different roles in enhancing evolutionary fitness. Lastly, System 1 operations are incorrectly deemed “rational” whenever they increase evolutionary fitness.

Stanovich & West (S&W), following all relevant others, define the rationality debate in terms of human performance on certain well-known problems. Unfortunately, these problems are very easy. For that reason, if System 2 cognition is identified with the capacity to solve them, such cognition will not enable humans to meet the cognitive demands of our technological society. Other profound issues arise as well.

Individual differences are indeed an important aid to our understanding of human cognition, but the importance of the rationality debate is open to question. An understanding of the process involved, and how and why differences occur, is fundamental to our understanding of human reasoning and decision making.

We did not, as Brakel & Shevrin imply, intend to classify either System 1 or System 2 as rational or irrational. Instrumental rationality is assessed at the organismic level, not at the subpersonal level. Thus, neither System 1 nor System 2 are themselves inherently rational or irrational. Also, that genetic fitness and instrumental rationality are not to be equated was a major theme in our target article. We disagree with Bringsjord & Yang's point that the tasks used in the heuristics and biases literature are easy. Bringsjord & Yang too readily conflate the ability to utilize a principle of rational choice with the disposition to do so. Thus, they undervalue tasks in the cognitive science literature that compellingly reveal difficulties with the latter. We agree with Newton & Roberts that models at the algorithmic level of analysis are crucial, but we disagree with their implication that attention to issues of rationality at the intentional level of analysis impedes work at the algorithmic level of analysis.