Abstract: How much information can be acquired from a single fixation in normal reading? Whether words in a sentence are processed one by one (serially), or two or more words are processed in parallel, is a major dispute between serial attention shift (SAS) and distributed lexical processing models (SWIFT) of eye movement control. Assessing which type of model is more consistent with empirical data has become a hot topic in research examining eye movements during reading. The main purpose of the present study was to determine how processing of characters in the parafovea affects foveal inspection time during Chinese sentence reading.
The boundary paradigm (Rayner, 1975) was used in two experiments designed to examine whether processing of the fixated character was affected by the frequency of the character to the right of fixation. An invisible boundary was positioned between the two characters (N and N-1). Once the reader’s eyes crossed the boundary, the incorrect preview of N was changed to the correct character, ensuring that the preview character was only available for parafoveal processing. In Experiment 1, 30 university students were asked to read 48 sentences. There were two types of preview characters (high- or low-frequency), both of which were congruent with the context. When the eyes crossed the invisible boundary, both previews were changed to the same target character, which was also congruent with the context. In Experiment 2, 34 university students were asked to read 76 sentences. Again, the frequency of the preview character was manipulated (high or low). Both of the two preview characters were incongruent with the sentence context, but the target character was congruent with the context.
In Experiment 1, the results showed that first fixation durations and gaze durations on character n were longer with low-frequency previews than with high-frequent previews. In contrast, first fixation durations and gaze durations on character n-2 were longer in the condition where a high-frequency preview of n was being presented compared to a low-frequency preview. In Experiment 2, there were no significant differences for first fixation durations and gaze durations on character n between high or low frequency preview conditions. However, first fixation durations and gaze durations on character n-2 were longer in the condition where a low-frequency preview of n was being presented compared to a high-frequent preview – the opposite pattern observed in Experiment 1.
The results showed that information about character frequency can be accessed in parafoveal preview. Furthermore, this preview information can affect processing of the currently fixated character to some extent (the results showed a strong preview effect but less stable parafoveal-on-foveal effects). The data from these experiments fit better with the theoretical assumptions of parallel processing models of eye movement control in reading.

Key words: parafoveal, foveal, preview effect, serial processing, parallel processing