In reading, the eyes progress on the line of text with forward saccades of variable sizes. Two out of three words are fixated with a forward saccade, some words being skipped during a first eye pass. The position where the eyes initially land in a word is generally located between the beginning and the middle of the word, but in some instances it can be located at the beginning or end of the word.

During the last 30 years, a large amount of research has been devoted to understanding what determines the variability in both word skipping rate and initial landing sites in words. Results as a rule show that visuomotor factors associated with saccadic programming (such as the length of the next word and the distance from the word where the eyes are launched from) contribute largely to this variability. In contrast, linguistic factors such as the processing difficulty of the words located in parafoveal vision seem to participate to a smaller extent in the variability. As noted by Brysbaert and Vitu (1998), most studies have failed to find an influence of ease of processing a parafoveal word on the likelihood of skipping it, and when effects are reported, they are either small in amplitude or can be attributed to some other variables.

Concerning initial landing sites in words, Beauvillain, Doré and Baudoin (1996) reported a difference in initial landing sites of less than a letter between words that differ in the frequency of their initial letter sequence. On the other hand, Everatt and Underwood (1992) showed an initial landing difference as a function of where the uniquely identifying information in the word is located. But this effect could never be repeated, and it has been attributed to either artifacts of the eye tracking system or confounding variables such as the luminance of the letters. Finally, Dubois and Sprenger-Charolles (1988) reported an effect of semantic context on initial landing sites, when paired comparisons were made between sentences that were identical except for the target word (that was or was not semantically related to a preceding prime word). Since target words differed in both the visual attributes and frequency of occurrence in the language, conclusions can hardly be drawn as to the role of semantic factors.

At the same time, there is evidence in the literature that processing of parafoveal words occurs at least up to a lexical level. Both naming time and gaze duration (or total time the eyes spend on the word) are shorter when the word was visible in parafoveal vision before being fixated than when it was masked. In addition, the resulting parafoveal preview benefit is larger for high- than low-frequency words and for words that are predictable from the prior semantic and/or syntactic context. Whether there is a semantic preprocessing of the parafoveal words is still an open question.

Thus, although some information is extracted from words located in parafoveal vision, it does not seem to have a clear effect on where to move the eyes next. According to O’Regan (1990), the paradox results from both the slowness of parafoveal word identification processes and the use of an autonomous scanning strategy that is independent of these processes and relies only on low-level visual processes. The eyes could be guided through the line of text by a strategy attempting to drive the eyes from the center of one word to the center of the next word. Variability in both word skipping rate and initial landing sites then result mostly from low-level visuomotor factors that affect saccadic accuracy; the rare instances in which ongoing linguistic processes influence the eye landing could correspond to cases where the prior fixation duration is longer than usual.

This oculomotor-control view is challenged by a more cognitive view, that suggests that parafoveal word processing and the programming of the next saccade overlap in time, and that ongoing linguistic processes can modify the planned saccade. According to this view, all inter-word saccades initially aim for the center of the next word, but they deviate from the target when processing of the word is terminated before the saccade is computed. Thus, the likelihood that a saccade is influenced by ongoing processes then depends on the ease of processing associated with the parafoveal word.

The purpose of the present study was to distinguish between oculomotor- and cognitive-control views of eye movements in reading, and to determine the extent to which initial landing sites in words are sensitive to ongoing perceptual and linguistic processes. In particular, the present experiments tested whether the position where the eyes initially land in a word depends on the word’s predictability from a prior semantic context and whether the likelihood of observing such an effect is a function of the ease of processing associated with the parafoveal word. If context effects are obtained in conditions that facilitate parafoveal word processing, then this argues against the hypothesis that inter-word saccades are determined by an autonomous scanning strategy that is independent of ongoing linguistic processes. At the same time, this would provide further evidence for the case of semantic parafoveal word processing, and it would bring some insights on the question of where in the time course of word identification semantic processes intervene. This question has indeed been extensively debated in the last 20 years, and it is still not clear whether semantic context effects emerge before or after lexical access has been completed. As a saccade is programmed within 250 ms of an average fixation duration in reading, linguistic effects on saccades then suggest that semantic context effects can be shown at an earlier stage than in lexical decision and naming tasks, which take about 500 ms and which, in addition, involve post-access strategies.

In the present experiments, eye movement data were recorded in two experiments that were part of a larger study. Participants were presented isolated sentences with one or two semantically (un)related prime(s) preceding a target word. In both studies, conditions were selected to maximize the occurrence of early context effects (i.e., on initial landing sites). First, we used for related prime(s) and target words, pairs of words that were strongly associated. Second, we used both high- (Experiments 1 and 2) and low-frequency target words (Experiment 1 only). It has been shown in prior studies that high-frequency words are more likely to be preprocessed in parafoveal vision, and therefore they may enhance the occurrence or the amplitude of context effects. Third, we determined a posteriori the launch site (i.e., the last position of the eyes before the target word), since context effects may emerge only for close launch sites that favor parafoveal preprocessing. In addition, to ensure that context effects could not be the result of the target words having different visual attributes, pairs of, respectively, related and unrelated sentences contained the same target word, but differed by the preceding prime being used.

Participants. Eighteen students (11 from the University of Leuven, Belgium, and seven from Paris, France) who were between 18 and 30 years old, were paid to participate in the experiment at the University of Leuven, Belgium. They were all native French speakers (first language) and had normal or corrected-to-normal vision (only with glasses).

Material. Two sets of 72 sentences were constructed. Each sentence contained a prime and a target word. In half the sentences, both words were semantically related, and in the other half, both words were unrelated. Both related and unrelated sentences were matched except for the prime word which was either semantically related to the target or not. The related and unrelated primes were matched in length up to a two-letter difference.