This is an open access article under the terms of the Creat ive Commo ns Attri bution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Word recognition begins before a reader looks directly at a word, as demonstrated by the parafoveal preview benefit and word skipping. Both low-level form and high-level semantic features can be accessed in parafoveal vision and used to promote reading efficiency. However, words are not recognized in isolation during reading; once a semantic representation is retrieved, it must be integrated with the broader sentence context. One open question about parafoveal processing is whether it is limited to shallow stages of lexico-semantic activation or extends to semantic integration. In the present two-experiment study, we recorded event-related brain potentials in response to a sentence-final word that was presented in foveal or parafoveal vision and was either expected, unexpected, or anomalous in the sentence context. We found that word recognition, indexed by the N400, ensued regardless of perception location whereas identification of the semantic fit of a word in its sentence context, indexed by the late positive component, was only observed for foveally perceived but not parafoveally perceived words. This pattern was not sensitive to task differences that promote different levels of orthographic scrutiny, as manipulated between the two experiments. These findings demonstrate separate roles for parafoveal and foveal processing in reading. Public Significance StatementThere has long been a public interest in the prospect of speed reading, but scientists have repeatedly challenged this possibility, pointing out that there is a tradeoff between reading speed and accuracy. In particular, if a reader skims the text and does not look at all or most of the words (as during "speed reading"), they cannot comprehend the text as well. The present study adds additional support, via neural data, for the idea that looking directly at a word is necessary to perform aspects of the reading process that lead to a higher level of understanding of a word and its context.

Readers extract visual and linguistic information not only from fixated words but also upcoming parafoveal words to introduce new input efficiently into the language processing pipeline. The lexical frequency of upcoming words and similarity with subsequent foveal information both influence the amount of time people spend once they fixate the word foveally. However, it is unclear from eye movements alone the extent to which parafoveal word processing, and the integration of that word with foveally obtained information, continues after saccade plans have been initiated. To investigate the underlying neural processes involved in word recognition after saccade planning, we coregistered electroencephalogram (EEG) and eye movements during a gaze-contingent display change paradigm. We orthogonally manipulated the frequency of the parafoveal and foveal words and measured fixation related potentials (FRPs) upon foveal fixation. Eye movements showed primarily an effect of preview frequency, suggesting that saccade planning is based on the familiarity of the parafoveal input. FRPs, on the other hand, demonstrated a disruption in downstream processing when parafoveal and foveal input differed, but only when the parafoveal word was high frequency. These findings demonstrate that lexical processing continues after the eyes have moved away from a word and that eye movements and FRPs provide distinct but complementary accounts about oculomotor behavior and neural processing that cannot be obtained from either method in isolation. Furthermore, these findings put constraints on models of reading by suggesting that lexical processes that occur before an eye movement program is initiated are qualitatively different from those that occur afterward.

Readers extract information from a word from parafoveal vision prior to looking at it. It has been argued that parafoveal perception allows readers to initiate linguistic processes, but it is unclear which stages of word processing are engaged: the process of extracting letter information to recognize words, or the process of extracting meaning to comprehend them. This study used the event‐related brain potential (ERP) technique to investigate how word recognition (indexed by the N400 effect for unexpected or anomalous compared to expected words) and semantic integration (indexed by the Late‐positive component; LPC effect for anomalous compared to expected words) are or are not elicited when the word is perceived only in parafoveal vision. Participants read a target word following a sentence that made it expected, unexpected, or anomalous, and read the sentences presented three words at a time in the Rapid Serial Visual Presentation (RSVP) with flankers paradigm so that words were perceived in parafoveal and foveal vision. We orthogonally manipulated whether the target word was masked in parafoveal and/or foveal vision to dissociate the processing associated with perception of the target word from either location. We found that the N400 effect was generated from parafoveally perceived words, and was reduced for foveally perceived words if they were previously perceived parafoveally. In contrast, the LPC effect was only elicited if the word was perceived foveally, suggesting that readers must attend to a word directly in foveal vision in order to attempt to integrate its meaning into the sentence context.

No abstract