Parafoveal N400 effects reveal that word skipping is associated with deeper lexical processing in the presence of context-driven expectations

Abstract

Readers are able to begin processing upcoming words before directly fixating them, and in some cases skip words altogether (i.e., never fixated). However, the exact mechanisms and recognition thresholds underlying skipping decisions are not entirely clear. In the current study, we test whether skipping decisions reflect instances of more extensive lexical processing by recording neural language processing (via electroencephalography; EEG) and eye movements simultaneously, and we split trials based on target word-skipping behavior. To test lexical processing of the words, we manipulated the orthographic and phonological relationship between upcoming preview words and a semantically correct (and in some cases, expected) target word using the gaze-contingent display change paradigm. We also manipulated the constraint of the sentences to investigate the extent to which the identification of sublexical features of words depends on a reader's expectations. We extracted fixation-related brain potentials (FRPs) during the fixation on the preceding word (i.e., in response to parafoveal viewing of the manipulated previews). We found that word skipping is associated with larger neural responses (i.e., N400 amplitudes) to semantically incongruous words that did not share a phonological representation with the correct word, and this effect was only observed in high-constraint sentences. These findings suggest that word skipping can be reflective of more extensive linguistic processing, but in the absence of expectations, word skipping may occur based on less fine-grained linguistic processing and be more reflective of identification of plausible or expected sublexical features rather than higher-level lexical processing (e.g., semantic access).

Headnote

Accepted: 15 October 2024 / Published online: 20 November 2024

© The Psychonomic Society, Inc. 2024

Abstract

Readers are able to begin processing upcoming words before directly fixating them, and in some cases skip words altogether (i.e., never fixated). However, the exact mechanisms and recognition thresholds underlying skipping decisions are not entirely clear. In the current study, we test whether skipping decisions reflect instances of more extensive lexical processing by recording neural language processing (via electroencephalography; EEG) and eye movements simultaneously, and we split trials based on target word-skipping behavior. To test lexical processing of the words, we manipulated the orthographic and phonological relationship between upcoming preview words and a semantically correct (and in some cases, expected) target word using the gaze-contingent display change paradigm. We also manipulated the constraint of the sentences to investigate the extent to which the identification of sublexical features of words depends on a reader's expectations. We extracted fixation-related brain potentials (FRPs) during the fixation on the preceding word (i.e., in response to parafoveal viewing of the manipulated previews). We found that word skipping is associated with larger neural responses (i.e., N400 amplitudes) to semantically incongruous words that did not share a phonological representation with the correct word, and this effect was only observed in high-constraint sentences. These findings suggest that word skipping can be reflective of more extensive linguistic processing, but in the absence of expectations, word skipping may occur based on less fine-grained linguistic processing and be more reflective of identification of plausible or expected sublexical features rather than higher-level lexical processing (e.g., semantic access).

Keywords Reading . Word skipping . Fixation-related potentials

The coordination between the eyes and the brain during natural reading is necessarily a well-oiled machine. In order to continuously add new linguistic information to the queue for cognitive processing, the eyes must move rapidly across a text. The time that readers spend looking at individual words (i.e., fixation durations) is astonishingly brief- around a quarter of a second on average-and some words are skipped altogether (i.e., never directly fixated; Rayner, 1998). The pace of these behavioral responses raises questions about whether they are initiated based on complete linguistic processing. It is assumed that the timing and location of eye movements reflects some aspect of higher-level language comprehension. However, one of the core neural responses associated with semantic processing (Kutas & Hillyard, 1980; see Federmeier, 2022; Kutas & Federmeier, 2011)-the N400 event-related brain potential (ERP)- takes substantially longer to peak (~ 400 ms after fixation or stimulus onset) than the average fixation duration lasts (see Rayner & Clifton, 2009). Based on this timing discrepancy, it appears as though eye movements during reading often precede the completion of semantic access and integration. Some researchers have suggested that eye movement decisions, especially very early decisions such as word skipping, are not closely tied to language processing (McConkie & Yang, 2003a, 2003b; Vitu, 2003). However, there is overwhelming evidence that lexical and contextual variables do influence fixation durations during reading (Rayner, 1998, 2009; Staub, 2015), and that they are processed early enough to affect the majority of fixations (Reingold et al., 2012; Sheridan & Reingold, 2012) and word skipping decisions (see Brysbaert & Vitu, 1998, for meta-analysis; Drieghe et al., 2004, 2005).

Schotter (2018) highlighted two premises that could account for how eye movement decisions precede the completion of word identification: (1) readers can engage in parafoveal preprocessing of upcoming words that lie outside of central vision (i.e., the upcoming text starting from 1° of visual angle away from the point of gaze), and (2) eye-movement behavior in reading represents a "hedged bet" that enough lexical processing will be complete by the time the eyes move to successfully identify the word. By this account, saccade decisions reflect an initial identification threshold that can be reached during parafoveal preview of an upcoming word rather than complete lexical processing (e.g., see computational implementation of this in the E-Z Reader model; Reichle et al., 1998), which occurs later and may be reflected by neural measures such as the N400. There are many proposals about the reasons why readers skip words-ranging from low-level oculomotor accounts (McConkie & Yang, 2003a, 2003b; Vitu, 2003) to high-level lexical accounts (Duan & Bicknell, 2020; Veldre et al., 2020). However, without convergent data from neural measures that connect skipping behavior to linguistic processing, any theory based on eye movement data alone is largely speculative.

To address this question in the present study, we used coregistration of eye tracking and electroencephalography (EEG) to investigate the relationship between behavioral and neural responses in reading and how they are impacted by the (sub)lexical information that can be obtained during parafoveal preview and its relationship to the meaning of the sentence context. In particular, we sought to determine whether word skipping is associated with more extensive processing of the skipped word during parafoveal processing compared with when the reader decides to fixate the word following parafoveal preview. Identifying differences in the brain responses to a parafoveal manipulation, splitting by skipping behavior, could provide novel insight into whether these behaviors are contingent on extensive linguistic processing of upcoming words.

Hedged bets, good enough processing, and the noisy channel framework in reading

Hedged bets (Schotter, 2018) are one mechanism by which the oculomotor system coordinates with the linguistic processing system in the brain in order to sustain the efficiency required for skilled reading. Because saccade plans take ~ 125 ms to execute, they must be initiated before complete word recognition in order to not waste time once the process does complete (see Reingold & Reichle, 2013). However, these somewhat risky behavioral decisions can potentially be made at the expense of word recognition accuracy. Therefore the behavior of reading (i.e., making eye movements quickly) may not necessarily imply deep language comprehension. For example, occasionally readers reach the end of a passage and realize they had been moving their eyes across the text without actually extracting the meaning conveyed by it (i.e., were "zoning out" or "mindlessly reading"; Reichle et al., 2010).

Even when reading for comprehension, it may be that not every word is identified and encoded deeply or even accurately (Ferreira & Patson, 2007). In the context of syntactically misleading (i.e., garden-path) sentences, Christianson et al. (2001) argued that readers do not necessarily extract a perfect representation of the linguistic input, but rather create a representation that is "good enough" to obtain an acceptable interpretation. Similarly, it has been proposed that language processing and semantic access function within a "noisy channel framework" that is well adapted to handling errors in the perceptual input (e.g., Chen et al., 2023). Ryskin et al. (2021) demonstrated that the N400 magnitude is associated with the probability of encountering a word given the context, inferred intended meaning, and the probability that the communicated input has been "corrupted by noise." For example, in the sentence "The storyteller could turn any incident into an amusing anecdote/antidote/hearse," they found a reduction in the N400 amplitude to antidote compared with hearse. Both words were equally implausible in the context, so the authors argued that the orthographic and phonological similarity of antidote to anecdote rendered it "recoverable." They propose that the reduced N400 response to the anomalous word antidote reflects the reader's assumption of a noisy channel and the inference that the intended continuation was in fact the plausible word anecdote.

Schotter's (2018) proposed idea of hedged bets in eye movement planning may reflect decisions based on such inferences given the perceptually "noisy channel" of parafoveal vision. If a parafoveal word is deemed familiar or probable, the reading system may program an eye movement forward based on whether the upcoming word form is a likely continuation of the communicated message. Eye movements may be initiated forward under such a framework even if the eventual recognition of that word proves difficult or results in disruption to overall comprehension. For example, readers are more likely to skip short, high-frequency words, particularly the word the, even when it appears in a sentence in a place where it does not make any sense (Angele & Rayner, 2013). These findings are also not restricted to the very high-frequency article the, but are also observed for implausible high-frequency content words (Abbott et al., 2015; Angele et al., 2014). Factors such as the length and familiarity of a word can influence eye movements even when they are unrelated to the process of accessing its meaning. Therefore, sublexical information may be useful for early stages of word decoding or performing a familiarity check of an upcoming word in order to plan an eye movement (Reichle et al., 1998), but this initial processing may not always translate to successful access or integration of a word's meaning.

A reader may be more likely to make a hedged bet saccade decision, and do so based on only a cursory processing of parafoveal sublexical information, when the sentence context sets them up to have strong expectations for an upcoming word (Schotter, et al., 2015). In highly predictable sentences, substantial orthographic overlap with an expected word can contribute to skipping decisions even when the parafoveal stimulus is semantically inappropriate or is not a real word. For example, skipping rates are higher and reading times are reduced for nonwords or anomalous words when they are orthographically similar to an expected word (Balota et al., 1985; cf. Drieghe et al., 2005). These data suggest that the information used to make these early saccade decisions may not include a semantic analysis of the word, but instead might be based on a comparison of the bottom-up sublexical form to expectations the reader has based on the context.

Contrasting with the account described above, other studies have found evidence that readers can make skipping decisions in a semantically informed way. For example, some eye tracking studies do find that even in low-constraint sentences plausible words are skipped more often than anomalous words (Brothers & Traxler, 2016; Veldre & Andrews, 2017, 2018a, 2018b; Veldre et al., 2020), which implies that high-level semantics can be accessed parafoveally and used to inform saccade decisions. However, not all eye-tracking studies find such effects. For example, a high-powered study using a Bayesian analysis to assess the effect of plausibility on word skipping found strong evidence for the null and concluded that plausibility did not influence skipping probability (Abbott & Staub, 2015). The inconsistencies in this literature may be clarified by incorporating complementary evidence about the ongoing language processing from neural measures, such as ERPs.

The (parafoveal) N400: ERP evidence of sublexical and semantic processing

ERPs may provide insight into potential mechanisms of language processing that eye movements cannot. One of the most well-studied language-related ERP effects is the N400, a less negative amplitude for an expected or contextually appropriate word compared with an unexpected, anomalous, or otherwise inappropriate word in the sentence context that peaks at centroparietal sites around 400 ms after the word is perceived (Kutas & Federmeier, 2011). Although the N400 is broadly accepted to indicate semantic processing, some sublexical manipulations can also affect this component, and these ERP responses can be informative about the nature of processing that might be achieved when a word is fixated. For example, N400 effects are reduced for anomalous orthographic neighbors of expected words compared with violations that are both semantically and orthographically unexpected (Laszlo & Federmeier, 2009), suggesting that semantic expectations can also feed down to expectations about a word's orthographic form. These effects are driven by the presence of expectations because they are not apparent for the same participants when the same stimuli are embedded in low-constraint sentence contexts (Caliskan et al., 2023). Furthermore, in general N400 effects are larger when the sentences have higher constraint (Wlotko & Federmeier, 2012), suggesting that they are driven by the degree to which semantic (and/or sublexical) expectations are met.

The N400 effect is also elicited even when the manipulated target word is perceived in parafoveal vision (Barber et al., 2010, 2013; Milligan, Nestor, et al., 2023b; Payne et al., 2019; Schotter et al., 2023; Stites et al., 2017), suggesting that readers may be able to access semantics parafoveally. If this is the case, it would suggest that semantic access may contribute to the saccade decisions, including word skipping. However, these parafoveal N400 studies primarily use high-constraint sentences (i.e., sentences in which a particular word is rendered highly predictable). Therefore, the parafoveally elicited N400 effects may in fact reflect violations of orthographic expectation rather than true extraction of semantic information from parafoveal vision (Nestor et al., 2024). If so, it would suggest that parafoveal processing serves primarily as a familiarity check and a head-start on processing lower-level features of a word, which informs hedged bets about when and where to move the eyes. Full recognition and integration, on the other hand, may depend more heavily on processing that takes place during (and/or following) foveal fixations.

Orthographic and phonological sublexical processing

So far we have described evidence that a reader's expectations can lead them to plan eye movements based only on a cursory assessment of sublexical features, particularly orthography, but it has also been argued that the "familiarity check" involves phonological processing as well (see Reichle et al., 2006). In eye-tracking studies, readers spend less time fixating semantically inappropriate words when they are homophonous with semantically correct words, compared with words that are only orthographically similar (Jared & Seidenberg, 1991; e.g., The palace had a [thrown/throng] room). These findings also extend to words that are perceived in parafoveal vision (Pollatsek et al., 1992; see Leinenger, 2014; Milligan & Schotter, 2024; Vasilev et al., 2019). Like the role of sentence context in orthographic effects, Milligan and Schotter (2024) found that parafoveal homophony effects are larger when the sentence context is constraining and the reader was likely to have generated predictions about the upcoming word's phonology. Therefore, identification of both orthographic and phonological forms may be crucial for early (presemantic) word identification processes in the parafovea and for making hedged bets about when and where to move the eyes. An open question, however, is whether this benefit from a phonological preview is a benefit to oculomotor programming, providing sufficient information to make hedged-bet saccade decisions, or whether this benefit actually extends to the underlying process of recognizing the word and accessing its meaning. If phonology primarily plays a role in oculomotor programming and informing hedged bets, it may be that phonological properties of a parafoveal preview influence reading behavior (e.g., word skipping) but do not influence the N400 magnitude and downstream semantic processing.

Previous ERP studies have found that, in high-constraint sentences, a (nonword) homophone of an expected word leads to a reduced N400 response compared with a (non)word that is phonologically unrelated to the expected word (Newman & Connolly, 2004). In contrast, in low-constraint sentences, the effect of phonological relatedness appears to depend on the relative familiarity (i.e., lexical frequency) of the homophones and correct words, as demonstrated by Newman et al. (2012). They found no reduction in the N400 effect for homophones compared with orthographic control words when they were lower frequency than the correct word. However, when the homophones were higher frequency than the correct word, the N400 pattern was similar to that reported for high constraint sentences (i.e., was reduced for homophones compared with orthographic controls).

Together, these findings suggest that access to phonological representations that serve a role in semantic access may not be a ubiquitous phenomenon. It may be contingent on the word having a high degree of accessibility, either due to high-constraint preactivating expectations for the word's phonological form, or due to the high frequency of the stimulus allowing for easier accessibility in long-term memory (e.g., due to repetition of retrieval and the strength of the memory trace). Given the somewhat inconsistent findings within both the eye tracking and ERP literatures regarding phonological effects, it is worth further investigating the roles of both orthographic and phonological features in planning eye movements and accessing meaning. It may be the case that orthographic and/or phonological sublexical features are primarily used in the hedged bet familiarity process of planning eye movements, but may also play a prominent role in the more semantically mediated downstream processing that generates the N400.

Coregistration and fixation-related potentials

Recent research utilizing the relatively novel approach of coregistering eye movements and neural activity (i.e., EEG) to create fixation related potentials (FRPs; see Degno et al., 2021; Dimigen et al., 2011) allows researchers to investigate the word recognition processes that continue to unfold after the saccade has been executed, and therefore whether the reader may eventually fully process the word once their eyes have moved on. This may explain why neural word recognition indices like the N400 occur later than the eye-movement plans, even when they are coregistered in an experiment so that all methodological explanations for the timing discrepancy can be ruled out (Kretzschmar et al., 2009; Milligan et al., 2023a, 2023b). Therefore, FRPs may offer further insight into the unfolding of orthographic, phonological, and semantic processing and how such processing occurs before and after the eyes leave a word.

When it comes to behavior-contingent FRP effects, only a single study has yet attempted to tie eye-movement decisions to variations in the underlying neuro-linguistic processing. Metzner et al. (2017) manipulated the semantic and syntactic fit of words in natural sentence reading and found differences in the FRP effects to these manipulations based on regressive eye movements. Their sentence stimuli were in German, for which articles and nouns have a grammatical gender, so they manipulated the semantic plausibility of the target noun or the gender agreement between the article and the noun (e.g., no violation: The_mascdeteriorating farm_masc needs a renovation; semantic violation: The_mascinquisitive farm_mascneeds a renovation; syntactic violation: The_fem deteriorating farm_masc needs a renovation.) The P600 component, which has been functionally tied to syntactic processing difficulty and structural reanalysis (e.g., Osterhout & Holcomb, 1992), was larger in response to the syntactic anomaly condition on trials in which the reader made a regressive eye movement. Therefore, they demonstrated that differences in eye-movement decisions can reflect differences in the underlying brain processes involved in language processing at the individual trial level. In the current study, we employ a similar approach to investigate whether skipping decisions reflect distinctly unique scenarios of deeper parafoveal word processing.

Similar to the relationship between regressive eye movements and FRPs (Metzner et al., 2017), we expected that parafoveal processing may be qualitatively or quantitatively different when the reader decides to skip the parafoveally processed word. Therefore, we included skipping behavior as a predictor in our analysis for the FRPs. Furthermore, investigating the effect of skipping on parafoveal preview effects circumvents this issue of data loss if we were to only analyze trials in which the target word was fixated. To study parafoveal processing of sublexical word characteristics like phonology, we had to use English words with homophones, which tend to be short (i.e., 3-5 characters) and therefore have skipping rates that are relatively high (Milligan & Schotter, 2024).

Current study

The current study coregisters eye movements and EEG to try to get at questions of the role of expectations about sub-lexical form in facilitating parafoveal processing. We analyzed behavior-contingent effects on the FRPs to investigate the nature of the link between eye movements and real-time language processing in the brain during natural sentence reading. We measured FRPs to fixations on the pretarget word while semantically correct, homophone, or orthographic control previews were visible in parafoveal vision. We used the gaze-contingent display change manipulation (Rayner, 1975) to change the text based on the reader's current eye position so that the manipulation (i.e., the homophonous, orthographic control, or identical/ correct preview) was only presented in parafoveal vision and that word changed to the semantically correct target word once the reader's eyes moved to fixate it directly or skip over it.

We also included a sentence constraint manipulation to investigate whether the preactivation of semantic, orthographic, and phonological features from contextually generated expectations influences the contribution of parafoveal preview to downstream semantic processing. We hypothesized that context driven expectations may be relevant when it comes to the mechanisms that influence skipping decisions and that in high-constraint skipping behavior may be more indicative of more extensive semantic parafoveal processing. Meanwhile, in the absence of expectations, skipping of words with subtle form violations may be more reflective of oculomotor constraints and low-level lexical processing informed by the potential plausibility of orthographic and phonological forms, but less reflective of deeper semantic access. To probe the relationship between decisions to execute eye movements and the underlying lexical processing in the brain, we identified trials in which the reader skipped the target word and compared them with trials in which they directly fixated it.

Our primary dependent measure was the parafoveal N400 FRP component, which is time locked to the pretarget word. Prior research has found no parafoveal-on-foveal effects of the preview manipulations (see Brothers et al., 2017). The fact that pretarget fixation durations do not appear to vary based on preview manipulations is advantageous in the current study because it reduces the potential confound of fixation duration or component overlap differences when investigating FRP effects. We expected to find N400 effects (more negative amplitudes) for the homophone and orthographic control conditions compared with the identical preview condition, reflecting the detection of the violation of sublexical and/or semantic expectations. We also expected that there would be reduced N400 for the homophone relative to the orthographic control (particularly in high-constraint sentences) if expectations of semantics feed forward to phonology. Based on the premise that skipping indicates that a substantial amount of lexical processing has occurred based on the parafoveal preview (Gordon et al., 2013; Pollatsek et al., 2006; Reichle et al., 1998), we speculated that the effect of preview condition on the pretarget N400 would be larger when the target was skipped.

Method

Participants

Seventy-three participants were recruited from the psychology department's subject pool at the University of South Florida and compensated with course credit or recruited through flyers and mailing lists and paid $16/hour for their time. All participants were right-handed native English speakers between the ages of 18 and 35 years, with normal or corrected-to-normal vision and no history of reading, learning, or neurological disorders. Participants provided informed consent via an online consent form approved by the University of South Florida. Data from 56 participants are reported in the final analyses of both eye movement and FRPs; four participants were excluded due to synchronization issues with eye-tracking and EEG files, three additional participants did not finish the experimental procedure due to excessive EEG artifacts and disconnected electrodes during recording. An additional six were excluded after data processing because of EEG artifact exclusions for at least 30% of trials, and four were excluded due to eye-tracker/display change errors on at least 20% of trials. All retained participants had at least six observations per condition for both EEG and eye tracking (mean = 18.32, SD = 4.45).

Stimuli and design

The experimental stimuli, taken from Milligan and Schotter (2024) consisted of 112 unique target words presented in a 2 (sentence context: high constraint vs. low constraint) × 3 (preview type: identical, homophone, orthographic control) factorial design; see Examples 1a and 1b for the high and low constraint sentences and their three preview words, respectively:

(1a) The woman likes her eggs to have a runny yolk/yoke/ yore to dip her toast in.

(1b) Patrick always seems to break the yellow yolk/yoke/ yore while trying to flip his fried egg.

The lexical characteristics for the target and preview words are reported in Table 1. The pretarget words were also roughly matched on length and frequency between the two constraint conditions; high-constraint pretarget words were, on average, 6.2 characters long (SD = 2.3) and had a lexical frequency of 3.1 (SD = 2.3), and low-constraint pretarget words were 6.2 characters (SD = 1.9) with a frequency of 3.2 (SD = 1.9).

As described in Milligan and Schotter (2024), sentence stimuli were normed for sentence constraint (i.e., cloze probability for the target word; Taylor, 1953) and plausibility (i.e., a 7-point Likert scale from 1 = very poorly written to 7 = very well written). The proportion of times the target word was produced in the cloze task was, on average, 0.75 (SD = 0.18) in the high-constraint sentences and 0.02 (SD = 0.04) in the low-constraint sentences. The average plausibility ratings on a scale of 1 (very poorly written) to 7 (very well written) in the high-constraint sentences and low-constraint sentences, respectively, were 5.36 (SD = 0.51) and 5.29 (SD = 0.61) for the target word, 3.72 (SD = 0.64) and 3.45 (SD = 0.70) for the homophone, and 3.22 (SD = 0.54) and 3.23 (SD = 0.64) for the orthographic control. The experimental conditions were counterbalanced across items and presented in a random order. Each stimulus item had the same target word, which appeared in both a high- and low-constraint sentence context. Therefore, each participant saw each item, as defined by having the same target word, twice (in both a high- and a low-constraint sentence), but the counterbalancing was such that participants never saw a given item (target word) in the same preview condition. Each member of the homophone pair also served as a target word in separate items. The full experimental stimulus set can be found online (https:// osf. io/ xcq28/). Additionally, there were 58 filler sentences to obscure the design of the study, which were all low constraint and contained a lexical frequency manipulation (taken from Schotter & Leinenger, 2016).

Apparatus and recording

EEG was recorded from 32 Ag/AgCl active electrodes (extended 10/20 system) using an actiCAP/actiCHamp electrode cap and amplifier system (Brain Products) with a 500-Hz online sampling rate. No online frequency filters were used during recording. Horizontal and vertical electro-oculogram (EOG) was recorded from two pairs (bipolar reference) of passive electrodes placed on the outer canthi of each eye and above and below the right eye. The scalp electrode signal was referenced online to the leftmastoid and rereferenced offline to the algebraic mean of the right and leftmastoids. Impedance values were reduced to 10 kΩ or lower at all electrode sites prior to recording.

Eye movements were recorded using an SR Research Ltd. EyeLink 1000 Plus eye-tracking camera in remote desktop mode (sampling rate of 500 Hz). Viewing was binocular, but eye movements were recorded from the right eye. A threepoint calibration was used at the beginning of the experiment and calibration accuracy had to fall within 0.3° of visual angle at each point to be accepted. Recalibration was performed periodically throughout the experiment if accuracy dropped below this level, as determined by an intertrial driftcheck. In the EyeLink recording settings, saccade detection was set to Normal (recommended for cognitive tasks like reading). The sample filter was set to Extra and the link analog filter was set to STD (both the recommended default settings; SR Research, 2017).

Procedure

Participants were seated at a viewing distance of 60 cm from a BENQ XL2540 model LCD monitor with a 240-Hz refresh rate and screen resolution of 1,920 × 1,080 pixels. Participants were instructed to read the sentences normally for comprehension. They were given five practice trials to acclimate them to the task. The sentence text appeared on the screen in black 12-point Courier New (monospaced) font on a light-gray background. Both experimental and filler sentences were followed by yes/no comprehension questions on 25% of trials, with equal numbers of yes and no correct responses, to ensure that participants read the sentences for meaning. Each trial was initiated by the experimenter after a driftcheck of the eye tracker. To trigger the presentation of the sentence, the participant had to make a fixation in a black box on the leftside of the screen at the location of the beginning of the sentence. Participants were instructed to look at a target sticker on the right edge of the monitor (offscreen) once they finished reading for comprehension and the trial was terminated by the participant via a manual button press when they had finished reading.

The gaze-contingent boundary paradigm (Rayner, 1975) was used to present the manipulation of sublexical information only in parafoveal vision; readers therefore only fixated on semantically correct words and were less likely to notice the manipulation and alter their behavior or neural processes for strategic reasons. An invisible boundary was placed immediately after the last letter of the pretarget word, before the space before the target word; prior to fixating the target word, the preview word was visible (one of the preview type conditions). The target word replaced the preview within approximately 5-10 ms of the eye tracker detecting that the boundary had been crossed. After the experiment, participants were asked whether they noticed anything unusual about the sentence display. Three participants reported noticing a specific word change. If they reported noticing nothing unusual, we then asked if they noticed words flickering or changing. When prompted, 24 participants said they noticed some flickering on between 1 and 10 of the trials.

Following the reading experiment, participants also completed a spelling from dictation task for the lower frequency version of each homophone pair presented in the experiment. Each lower frequency homophone word was presented auditorily and used in a sentence and participants were asked to write the correct spelling of the word. This task was conducted to verify that participants were generally aware of the correct orthographic form of the homophone words used as previews in the experiment. Out of the 56 total homophone pairs used in the experiment, on average, participants were able to produce the accurate spelling of the lower frequency homophone member for 51.4 of the words probed in the task (92% accuracy; SD = 4.6, range: 30-56).

Data cleaning and preprocessing

Eye-tracking data were cleaned using the DataViewer program (SR Research Ltd., Version 4.3.1), and fixations in any of the predefined interest areas were merged with a neighboring fixation if their duration was below 80 ms and they were within 0.3° horizontally. Trials were excluded from analyses due to eye-tracking data loss, display change errors, pretarget word skipping, and first-pass refixations on the pretarget word (i.e., were not single fixations). The total number of trials before other exclusions was 12,526 (out of 12,544 total trials presented); the 18 missing trials were from either the participant or experimenter manually terminating a trial accidentally before the sentence was presented. Of these 12,526 trials, 8,646 had single fixations on the pretarget word. Of these 8,646 trials, 230 were excluded due to j-hooks (i.e., oculomotor error in which the saccade overshoots the fixation point and triggers the display change but returns back to the pretarget word prior to fixating), fixations near the right edge of the pretarget region in which a sample that crossed the boundary triggered the display change (despite the fixation being identified as within the pretarget region), unstable calibrations, and any other erroneous early triggering of the display change. These trials were identified based on manual inspection of all eye-tracking trials using the eye movement/ fixation visualization application in Data Viewer, leaving 8,416. EEG epochs were extracted by identifying fixations in the pretarget region of interest that were not preceded within 600 ms by another pretarget fixation and that were followed within 600 ms by a fixation on either the target word (target-fixated trials) or by a fixation on the end of sentence region (target-skipped trials). After merging the eye tracking data with the EEG data, 7,270 total trials were retained that met these criteria. Of these 7,270 trials, 1,115 trials were excluded due to EEG artifacts, leaving 6,155 trials. Finally, trials were excluded if the target word was fixated but followed by a regression out of the target region rather than a forward saccade, resulting in a total of 5,937 trials retained for the pretarget time locked FRP analysis.

EEG data preprocessing was performed using the EEGLAB (v2019.0/v.2021.0; Delorme & Makeig, 2004), ERPLAB (v8.02; Lopez-Calderon & Luck, 2014) and EYE-EEG (v0.85; Dimigen et al., 2011) toolboxes in MATLAB. The EEG was rereferenced offline to an algebraic mean of the right and leftmastoids and band-pass filtered from 0.1 to 50 Hz (- 6 dB), with 0.2-32.8-Hz half-power (- 3 dB) cutoffs, using an IIR Butterworth filter. Ocular artifacts were removed from the EEG using optimized independent components analysis (OPTICAT; Version 2020-01-28), following the procedures and recommendations described in Dimigen (2020). The ICA was trained using band-pass filtered (with a passband edge of 3 Hz) training data that over-weighted spike potentials by a factor of 1. Ocular artifact components were automatically flagged and removed using eye tracker-guided eye artifact component identification (Plöchl et al., 2012), using a variance ratio threshold of 1.1. EEG was epoched into segments from 200 ms before to 1,000 ms after the start of fixations on the pretarget and target words and baseline corrected by subtracting the mean voltage from - 200 to 0 ms for each channel. Epochs containing artifacts were flagged for removal using a moving window peak-to-peak threshold automatic artifact detection algorithm, rejecting epochs with voltage changes of greater than 100 µV within a 200-ms time span, with a 50-ms window step. The epoched data were also inspected manually to confirm that artifact-contaminated epochs were removed.

The resulting dependent variables from the eye tracking data and FRP data were exported from their respective processing software and were merged at the trial level for confirmatory analyses in R.

Transparency and openness

The processed data on which these analyses were performed, the R code for lmer analyses, and the experimental sentence stimuli can be found on OSF (https:// osf. io/ xcq28/). We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study (Simmons et al., 2012) and we follow JARS-Quant (Appelbaum et al., 2018). Some aspects of this study, including the experimental design, target sample size, and hypotheses and planned analyses for target time locked FRP analyses were pre-registered. The preregistration document can be found online (https:// osf. io/ 5fwja/). None of the pretarget time-locked analyses, including splitting FRP trials by target-skipping behavior, were preregistered. We planned a priori to analyze FRPs time locked to the pretarget words as a function of the experimental manipulations (i.e., sentence constraint and preview condition); however, the decision to include skipping behavior as a predictor of interest was made post hoc after identifying that we had relatively high skipping rates of the target word.

Results

Our primary dependent variable was the parafoveal N400 FRP (i.e., the amplitude averaged across a centroparietal ROI [C3, C4, Cz, CP1, CP2] from 300-500 ms postfixation time locked to single fixations on the pretarget word; i.e., the time point at which the preview was visible parafoveally). The ROI was selected a priori to include a representative selection of electrodes surrounding Cz and including centro-parietal sites because the N400 has typically been described as being maximal centro-parietally (e.g., see Kutas & Federmeier, 2011). In order to understand these responses in the context of the larger behavioral responses, we also analyzed pretarget single fixation durations (to assess whether the preview elicited any parafoveal-on-foveal effects) and target word skipping rates (see Table 2 for descriptive statistics). In order to ensure that the eye tracking skipping analysis aligned as closely as possible with FRP analyses, we assessed target skipping rates only for trials with single fixations on the pretarget word and trials in which first-pass fixations on the pretarget and target words were followed by forward saccades (i.e., excluding trials with regressions out of the pretarget or target words).

For all analyses, we used (generalized) linear mixedeffects regression models via the glmer() function (i.e., with a logit link for the binary outcome measure of skipping) and the lmer() function (i.e., for the FRP measures) from the lme4 package (Version 1.1-12; Bates et al., 2015) within the R Environment for Statistical Computing (Version 3.3.1). The fixed effects contained main effects for parafoveal preview condition (homophone, orthographic control, and identical) and sentence constraint (high vs. low), as well as their interactions. Because we were primarily interested in differences in the N400 amplitude for the homophone preview and orthographic control preview compared with the identical (plausible/expected) condition, preview condition was coded as successive differences contrasts, comparing the homophone with the identical preview and the identical to the orthographic control preview. We used these contrasts for both the FRP and eye-tracking measures, even though this differs from the way phonological preview benefit effects have been analyzed in prior eye-tracking studies. Sentence constraint was entered as a treatment contrast, with the highconstraint condition as the baseline so that the main effects of the preview and skipping contrasts represent the main effects at the level of high constraint. For the FRP analyses splitting by skipping behavior, skipping was entered as a treatment contrast with target skipped trials as the baseline. We attempted to include the maximal random effects structure, and simplified the random effects structures due to nonconvergence in a stepwise fashion by removing random effects, starting with interaction slopes, followed by main effect intercepts in the order of least amount of variance accounted for, until the models converged.

Pretarget-word single fixation duration

The analysis of single fixation duration (SFD) for the pretarget word revealed a significant main effect of sentence constraint (b = 6.42, t = 3.22, p < 0.05), with shorter fixation durations in high- compared with low-constraint sentences. Note that this main effect reflects the differences between the sentence contexts themselves, not anything about the preview. A word preceding an expected target word in a high-constraint sentence is likely to be more predictable than an analogous pretarget word in a low-constraint sentence. Therefore, these effects likely reflect more global effects of the constraint manipulation as expectancy builds across the sentence. There were no significant main effects of preview condition, nor were there any significant interactions between preview condition and sentence constraint (i.e., no evidence of parafoveal-on-foveal effects). We also performed a follow-up analysis predicting pretarget SFD by target-word skipping behavior, constraint, preview, and their interactions to assess whether the presence of parafoveal-on-foveal effects depended on target skipping behavior and we still found no main effects or interactions of preview condition on pretarget SFD. This lack of significant preview effects are advantageous for our FRP analyses because they indicate that differences in the timing of saccades away from the pretarget word (and consequently the timing of the initiation of fixations on the target or posttarget words) are unlikely to explain any substantial differences between the preview conditions with respect to fixation timing jitter impacting averaged neural responses.

Target-word skipping rates and fixation durations

In order to maintain consistency of the included data between eye movement and FRP analyses based on skipping, the data included in the skipping rate analyses and figures includes only trials with single fixations on the pretarget word and forward saccades (i.e., excluding trials with regressions out of the pretarget or target regions on the first pass).

For skipping rate, the main effect of sentence constraint was not significant. However, we performed supplementary analyses (see Appendix A) that did not use the filtering criteria described above to align with the analyzed FRP data (i.e., including only single first-pass pretarget fixations and only nonregressive saccades away from the target word) and with these more liberal trial-level inclusionary criteria the effect of constraint on skipping rates was significant. In the high sentence constraint condition (baseline in the model), there was a significant effect for the contrast of the orthographic control compared with the identical preview condition, with higher skipping for the identical preview, but there was no difference in skipping between the identical and homophone conditions. Neither of the interactions between preview condition and sentence constraint were significant (Table 3; see Fig. 1 for a visualization of the data patterns based on preview condition in high and low constraint sentences).

For both fixation duration measures, there were significant effects of constraint, with longer fixation durations on targets in low-constraint than in high-constraint sentences. There were significant contrasts for both the contrast of the orthographic control compared with the identical preview condition, and the identical compared with the homophone condition. However, there were no significant interactions, suggesting that those effects were also present for low-constraint sentences.

N400 FRP effects time locked to pretarget fixations

As expected, the N400 amplitude was significantly more negative in low-constraint compared with high-constraint sentences (collapsing across preview condition; Table 4). In the high-constraint sentences in which the reader skipped the target, there was a significant N400 effect (i.e., more negative N400 amplitude) for the orthographic control preview compared with the identical expected preview. However, there was not a significant difference between the identical and homophone previews and there were no significant interactions between this comparison and constraint or word skipping. There was no main effect of target skipping on the N400 amplitude (collapsing across preview conditions in high constraint sentences). For the orthographic control compared with the identical preview comparison, there were significant two-way interactions between preview and constraint and preview and skipping. The N400 effect of the orthographic violation was larger in high constraint and larger when the reader skipped the target word. There was also a three-way interaction between constraint, preview, and skipping. The pattern of this interaction was such that the effect of preview on the N400 was largest in high constraint sentences when the target word was skipped.

To better characterize the three-way interaction between orthographic versus identical preview, sentence constraint, and target skipping, we conducted follow-up analyses split by target skipping (Table 5). Two separate analyses were conducted for trials in which the target word was skipped and trials in which the target word was fixated, predicting the N400 amplitude by preview condition and sentence constraint. These follow-up analyses used the same contrasts as the primary analysis.

Pretarget FRPs for target skipped trials Mirroring the main analysis, there was a significant effect of sentence constraint collapsing across preview conditions (more negative amplitudes for low compared with high constraint). The orthographic control compared with identical preview effect was also significant for skipping trials, but as reflected in the primary analysis, there was no difference between the homophone and identical preview conditions nor was there an interaction between this comparison and constraint. The interaction between the orthographic preview violation effect and sentence constraint was also significant, demonstrating a larger effect of the preview manipulation on the N400 amplitude (more negative for the orthographic control compared with the plausible/expected preview in high- compared with low-constraint sentences (see Fig. 2 for waveforms and Fig. 3 for topographic scalp maps).

Pretarget FRPs for target-fixated trials In contrast with the target skipped trials, in the target-fixated analysis the only significant effect on the N400 amplitude was sentence constraint (collapsed across preview condition). There was neither a significant orthographic violation preview effect nor a homophone preview effect in high-constraint sentences, and no interaction between either of these effects and sentence constraint. Therefore, in this experiment we found no effect of the parafoveal preview word on the N400 neural response when readers do not skip the target word (presumably because they have not progressed very far into processing; see Fig. 4 for waveforms and Fig. 5 for topographic scalp maps).

Discussion

In the current study, we coregistered eye tracking and EEG to investigate the relationship between eye movement planning and neural responses during silent reading. In particular, we were interested in the relationship between word skipping behavior and the magnitude of brain responses to parafoveally perceived words. We manipulated sentence constraint to investigate the extent to which expectations interact with the sublexical information extracted from the parafovea and influence the underlying neural processes (i.e., the parafoveal N400 FRP response) as a function of eye movement behavior. We found a main effect of sentence constraint on the parafoveal N400 response, with a reduced negativity for predictable compared with unpredictable words (see also Kutas & Federmeier, 2011; Wlotko & Federmeier, 2012). However, the parafoveal N400 effect of an anomaly was only apparent in highconstraint sentences when the target word was ultimately skipped. Furthermore, this effect was eliminated when there was phonological overlap between the anomaly and the expected word (i.e., because they were homophones). Similarly, we found a significant effect of the anomalous previews (for orthographically similar but not homophonic preview words compared with expected words) on skipping rates in high-constraint sentences. However, despite finding no effects of the preview manipulation on the parafoveal N400 when the target word was fixated, fixation durations on the target word were significantly longer for both the homophone and orthographic control preview conditions compared with the identical. We interpret this discrepancy between the patterns observed for the FRPs and the fixation behavior as a consequence of the fact that they are reflecting different aspects of the reading process that are sensitive to different manipulations or properties of the text. We discuss the potential distinctions between what these measures reflect with regard to cognitive processing in more detail below.

Skipping contingency of the parafoveal N400 response to anomalies

The fact that the parafoveal N400 effect for the orthographic control was only present for skipping trials demonstrates that word skipping is tied to deep linguistic processing. Word skipping has been studied extensively in the context of sentence constraint (see Staub, 2015), word length (see Brysbaert & Vitu, 1998), and lexical frequency (see Kliegl et al., 2004); however, up to this point, there has been no way of measuring how word skipping relates to lexical processing and semantic access. Because one of the best predictors of word skipping is word length, even when controlling for lexical frequency (Kliegl et al., 2004; Heilbron et al., 2023), some researchers have argued that skipping is primarily driven by oculomotor constraints (e.g., saccade targeting based on the spatial layout of words in a sentence; e.g., McConkie & Yang, 2003a, 2003b; Vitu, 2003) rather than deeper lexical processing (Brysbaert & Vitu, 1998). However, the fact that we found a larger N400 response to anomalies when a word is skipped (in high-constraint sentences) suggests that word skipping is associated with more extensive lexical processing during parafoveal preview. Therefore, the results of this study demonstrate that skipping decisions do depend, at least in part, on the readers' progress toward parafoveal word identification.

A curiosity of these results is that readers skipped anomalous preview words that they eventually identified as violating their expectations (as evidenced by the downstream N400 response to the orthographic violations). Therefore, although skipping can be taken as an index of greater parafoveal processing, it does not necessarily indicate full identification and integration (or integration failure) that might prompt the reader to terminate the programmed skip. Therefore, word skipping is a somewhat special scenario in which some predictability or semantic activation threshold has been reached by the system that interfaces between language processing and oculomotor programming. In these cases, eye movements must have been initiated based on sufficient lexical processing having taken place for the reading system to "hedge a bet" that the skipped word will be identified (Schotter, 2018), but insufficient identification for fully detecting the semantic violation in time to cancel the skipping decision. Therefore, it may be that when a word is ultimately fixated, parafoveal preview serves to get a head start on assessing whether the upcoming word is familiar or aligns with expectations, but the reader remains reliant on foveal perception for full identification.

Participants skipped the target word quite often even in low-constraint sentences (> 40% of the time, on average), in which the N400 did not differ by preview condition, so apparently some skipping decisions can be initiated even when parafoveal word identification is less thorough. Therefore, different mechanisms may determine skipping decisions in the presence and absence of strong lexical predictions. Skipping, in general, may be primarily governed by a familiarity check that the upcoming word is a known word stored in memory and that the orthographic and phonological forms roughly align with a known word that is a plausible continuation. However, fine-grained identification of the word form and accurate word recognition may only be possible parafoveally when the reader has strong expectations with which to compare the perceived word. The ability to match a bottom-up word form to a predicted lexical form may be necessary for deep enough lexical and semantic parafoveal processing to elicit an N400 response.

When characterizing the cognitive processes that underlie skipping decisions, it is important to note that the oculomotor system is not immune to errors. Saccade targeting does not always result in the reader landing on the intended location, or even the intended word. It has been estimated that up to 20% of all fixations during are mislocated, or not in the intended location (Nuthmann et al., 2005). Therefore, in the current study, it may be the case that some of our trials classified as skips were intended by the reader to be target fixations and vice versa. However, mislocated fixations are often at the boundaries of words and are more likely to result in error correction saccades to the intended location (Nuthmann et al., 2005). This means that mislocated fixations often co-occur with refixations versus when only a single fixation occurs. In the current study we included only single fixation trials, so trials with refixations due to mislocated fixations were not included in the analyses. Additionally, Krügel and Engbert (2010) found evidence from modeling that unintended skips are likely quite rare and that mislocated fixations are more likely in cases when the reader intended to skip but undershot and unintentionally landed on the word instead. Furthermore, as reported in Table 2, the launch positions of saccades away from the pretarget word were remarkably similar across all experimental conditions. Therefore, in the current study it is unlikely that detecting the N400 effect of the orthographic preview violation in the skipping trials is simply due to misclassifying intended target fixations as skips. Also, if anything, these mislocated fixations would add noise and attenuate the N400 effect rather than result in a spurious behavior-contingent N400 effect.

The capabilities and limitations of parafoveal word processing

The parafoveal violations in this experiment were rather subtle. In both display change conditions the preview word had quite a bit of orthographic overlap with the expected word. Laszlo and Federmeier (2009) previously demonstrated that implausible orthographic neighbors of expected words elicit reduced N400 responses compared with nonneighbor violations. Based on these patterns to foveally presented target words, one might expect that the N400 response to subtle form violations would be totally eliminated when viewed in lower acuity parafoveal vision. Given the reduced visual quality and attentional resources during parafoveal processing, the reader might simply not be able to detect the error. However, we demonstrate that even subtle orthographic violations can be identified parafoveally. On the other hand, full recognition of these orthographic violations depends on context and the reader's ability to preactivate (sub)lexical representations and map the bottom-up input onto an expected word form.

In contrast to the pattern observed for anomalous orthographic controls, anomalous homophones never produced a parafoveal N400 effect. This pattern suggests that phonological representations play a substantial role in the processing of words in parafoveal vision and that readers ignore subtle orthographic violations (and potentially large semantic violations) if there is phonological overlap between an expected or plausible word and what they encounter parafoveally. In line with the noisy channel framework (Ryskin et al., 2021), readers may also interpret an anomalous homophone as the correct word, especially if they do not have precise representations of lexical form stored in memory (e.g., if they are weaker spellers and more prone to confusing the meanings of homophone pairs; e.g., Milligan & Schotter, 2024; Veldre & Andrews, 2015). Finally, there is also trial-level variability (presumably based on extraneous factors not manipulated in this experiment) in whether parafoveal processing is extensive enough to fully identify these subtle errors, as demonstrated by the elimination of the N400 effect when the target word is fixated.

The relationship between eye movements and neural responses

The fact that the N400 brain response to semantic violations in the parafovea was only present when a word was skipped and when an expectation was violated is a bit of a conundrum in the context of decades of eye-tracking research on reading that reports longer fixation durations to orthographic and semantic violations even when sentence constraint is not particularly strong (e.g., Balota et al., 1985; Schotter, 2013; Sereno et al., 2018; Veldre & Andrews, 2017). In alignment with prior work, we found longer fixation durations for both the homophone and orthographic control anomalous preview conditions compared with the correct identical preview condition (e.g., Leinenger, 2019; Milligan & Schotter, 2024; Pollatsek et al., 1992; Vasilev et al. 2019), and these effects were not dependent on sentence constraint. In contrast, the N400 response to homophone anomalies did not reflect a semantic disruption under any conditions and the orthographic anomaly effect was contingent on constraint. One possible explanation of these differences is that the effects on fixation durations may be largely due to the display change itself and the recognition that something changed between the preview and target words, even if the preview word itself was not fully identified. In contrast, the parafoveal N400 may reflect more a process of identifying an expectancy or plausibility violation of the parafoveal preview in comparison to the reader's expectations or ongoing interpretation of the sentence. These data support the idea of a hybrid account of parafoveal preview, consisting of both direct processing of the preview and trans-saccadic integration of the preview and target (Schotter, 2018).

If the brain can disregard parafoveal violations that align phonologically with an expected or plausible word to retrieve a meaningful semantic representation, but the eye-movement record shows evidence of disruption due to that violation, it suggests that eye-movement decisions do not always reflect the downstream brain processes involved in comprehension. Eye movements may be more sensitive to subtle form violations and exhibit temporary behavioral disruptions, but if these violations are subtle enough and easily recoverable, the process of comprehension and semantic retrieval appears to proceed without much disruption.

Additionally, in the current study we manipulated the parafoveal preview, but the foveally perceived word, when the reader decided to fixate it, was the expected plausible word. So, again, it may be that eye movements are quite sensitive to subtle violations in the parafovea because efficient oculomotor planning relies on this early head start afforded by parafoveal preview. However, upon foveal inspection, the brain can proceed in the process of comprehending the text with minimal regard for the parafoveally perceived stimulus. In other words, because foveal input is much higher fidelity and receives greater attentional resources than parafoveal input, it may be that the partial lexical representations based on parafoveal input that lead to hedged bets can be largely overridden by foveal input on the subsequent fixation.

Therefore, future theories of the brain-behavior connection between eye movements and underlying neural processing and language comprehension must grapple with the fact that these two measures of language processing appear to be measuring inherently different aspects of the reading process (see Schotter, 2018). We propose that eye movement behavior is governed by linguistic processing, but also by visual and motor constraints, and is determined largely prematurely of full word identification and comprehension. When it comes to eye movements that are influenced by the orthographic and phonological properties of an upcoming word, oculomotor decisions may be quite sensitive to subtle prediction violations of sublexical form, while the underlying brain processes can overcome these subtle parafoveal errors and prioritize making sense of the linguistic representation as a whole.

Conclusions

In the current study we coregistered eye movement and EEG recordings to extract FRPs time locked to fixations on a pretarget word during parafoveal preview of words using the gaze-contingent display change paradigm (Rayner, 1975). We used a novel approach of splitting trials based on the reader's decision to either skip or fixate the previewed target word. We found a parafoveal N400 response to an anomalous word, but only when the reader decided to skip the previewed word, in highly constraining sentences, and when the anomaly did not share the expected word's phonological form. This data pattern suggests that word skipping is associated with deeper lexical processing of upcoming words in parafoveal vision. However, the parafoveal N400 was not elicited in low constraint sentences, regardless of skipping behavior, suggesting that in the absence of expectations, skipping may be determined by separate mechanisms and be less reflective of the depth of word identification. Additionally, fixation durations on the target word were sensitive to the preview manipulation, but the N400 effect was absent on fixated trials, suggesting that eye movement behavior and neural responses index distinct aspects of the reading and language comprehension processes.

Supplementary Information The online version contains supplementary material available at https:// doi. org/ 10. 3758/ s13414- 024- 02984-6.

Author contributions S.M. and E.S. conceptualized and designed the study. S.M., M.J.B., and N.C. programmed the experiment, conducted the study and collected the data. S.M. and M.J.B. processed and compiled the data. S.M. conducted the data analyses and data visualization. S.M. and E.S. contributed to the literature review and writing of the manuscript with input from all authors. All authors reviewed and approved the final version of the manuscript for submission. E.S. provided funding acquisition, supervision, and project administration.

Funding Work on this study was supported by NSF grant 2120507 and a USF Strategic Investment Pool Award awarded to Elizabeth Schotter.

Data availability The processed data on which analyses were performed and the experimental sentence stimuli for this study can be found on OSF (https:// osf. io/ xcq28/).

Code availability The R code used for analyses presented in this paper are available on OSF (https:// osf. io/ xcq28/).

Declarations

Ethics approval The protocol for this study was Approved by the Institutional Review Board at the University of South Florida (Pro. 00042067). All participants provided electronic informed consent prior to participation.

Consent to participate Informed consent was obtained by all participants included in the study.

Consent for publication Consent for publication was obtained from all participants in this study. Participants were informed that their data could be used for publication purposes and agreed to the publication. of their anonymized data. No identifiable personal information will be disclosed in the publication.

Author note Work on this study was supported by NSF grant 2120507 and a USF Strategic Investment Pool Award awarded to Elizabeth Schotter.

Conflicts of interest We declare no conflicts of interest.

Sidebar

* Sara Milligan

[email protected]

Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

References

References

Abbott, M. J., Angele, B., Ahn, D., & Rayner, K. (2015). Skipping syntactically illegal the previews: The role of predictability. Journal of Experimental Psychology: Learning, Memory and Cognition, 41(6), 1703-1714. https:// doi. org/ 10. 1037/ xlm0000142

Abbott, M. J., & Staub, A. (2015). The effect of plausibility on eye movements in reading: Testing EZ Reader's null predictions. Journal of Memory and Language, 85, 76-87.

Angele, B., & Rayner, K. (2013). Processing the in the parafovea: Are articles skipped automatically? Journal of Experimental Psychology: Learning, Memory, and Cognition, 39(2), 649.

Angele, B., Laishley, A. E., Rayner, K., & Liversedge, S. P. (2014). The effect of high- and low-frequency previews and sentential fit on word skipping during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 40(4), 1181-1203. https:// doi. org/ 10. 1037/ a0036 396

Appelbaum, M., Cooper, H., Kline, R. B., Mayo-Wilson, E., Nezu, A. M., & Rao, S. M. (2018). Journal article reporting standards for quantitative research in psychology: The APA Publications and Communications Board task force report. American Psychologist, 73(1), 3-25. https:// doi. org/ 10. 1037/ amp00 00191

Balota, D. A., Pollatsek, A., & Rayner, K. (1985). The interaction of contextual constraints and parafoveal visual information in reading. Cognitive Psychology, 17(3), 364-390. https:// doi. org/ 10.1016/ 0010- 0285(85) 90013-1

Balota, D. A., Yap, M. J., Hutchison, K. A., Cortese, M. J., Kessler, B., Loftis, B., . . . Treiman, R. (2007). The English Lexicon Project. Behavior Research Methods, 39(3), 445-459. https:// doi. org/ 10.3758/ bf031 93014

Barber, H. A., Doñamayor, N., Kutas, M., & Münte, T. (2010). Parafoveal N400 effect during sentence reading. Neuroscience Letters, 479(2), 152-156.

Barber, H. A., van der Meij, M., & Kutas, M. (2013). An electrophysiological analysis of contextual and temporal constraints on parafoveal word processing. Psychophysiology, 50(1), 48-59. https://doi. org/ 10. 1111/j. 1469- 8986. 2012. 01489.x

Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1), 1-48.

Brysbaert, M., & Vitu, F. (1998). Word skipping: Implications for theories of eye movement control in reading. In G. Underwood (Ed.), Eye guidance in reading and scene perception (pp. 125-147). Elsevier Science Ltd.

Brothers, T., Hoversten, L. J., & Traxler, M. J. (2017). Looking back on reading ahead: No evidence for lexical parafoveal-on-foveal effects. Journal of Memory and Language, 96, 9-22. https:// doi.org/ 10. 1016/j. jml. 2017. 04. 001

Brothers, T., & Traxler, M. J. (2016). Anticipating syntax during reading: Evidence from the boundary change paradigm. Journal of Experimental Psychology: Learning, Memory, and Cognition, 42(12), 1894-1906. https:// doi. org/ 10. 1037/ xlm00 00257

Caliskan, N., Milligan, S., & Schotter, E. R. (2023). Readers scrutinize lexical familiarity only in the absence of expectations: Evidence from lexicality effects on event-related potentials. Brain and Language, 238, Article 105232. https:// doi. org/ 10. 1016/j. bandl. 2023.105232

Chen, S., Nathaniel, S., Ryskin, R., & Gibson, E. (2023). The effect of context on noisy-channel sentence comprehension. Cognition, 238, Article 105503.

Christianson, K., Hollingworth, A., Halliwell, J. F., & Ferreira, F. (2001). Thematic roles assigned along the garden path linger. Cognitive Psychology, 42(4), 368-407. https:// doi. org/ 10. 1006/cogp. 2001. 0752

Degno, F., Loberg, O., & Liversedge, S. P. (2021). Co-registration of eye movements and fixation-Related potentials in natural reading: Practical issues of experimental design and data analysis. Collabra: Psychology, 7(1). https:// doi. org/ 10. 1525/ colla bra. 18032

Delorme, A., & Makeig, S. (2004). EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. Journal of Neuroscience Methods, 134(1), 9-21. https:// doi. org/ 10. 1016/j. jneum eth. 2003. 10. 009

Dimigen, O. (2020). Optimizing the ICA-based removal of ocular EEG artifacts from free viewing experiments. NeuroImage, 207, Article 116117.

Dimigen, O., Sommer, W., Hohlfeld, A., Jacobs, A. M., & Kliegl, R. (2011). Coregistration of eye movements and EEG in natural reading: Analyses and review. Journal of Experimental Psychology: General, 140(4), 552-572.

Drieghe, D., Brysbaert, M., Desmet, T., & De Baecke, C. (2004). Word skipping in reading: On the interplay of linguistic and visual factors. European Journal of Cognitive Psychology, 16(1/2), 79-103.

Drieghe, D., Rayner, K., & Pollatsek, A. (2005). Eye movements and word skipping during reading revisited. Journal of Experimental Psychology: Human Perception and Performance, 31(5), 954-969.

Duan, Y., & Bicknell, K. (2020). A rational model of word skipping in reading: Ideal integration of visual and linguistic information. Topics in Cognitive Science, 12(1), 387-401.

Ferreira, F., & Patson, N. D. (2007). The 'good enough' approach to language comprehension. Language and Linguistics Compass, 1(1/2), 71-83. https:// doi. org/ 10. 1111/j. 1749- 818x. 2007. 00007.x

Federmeier, K. D. (2022). Connecting and considering: Electrophysiology provides insights into comprehension. Psychophysiology, 59(1), Article e13940. https:// doi. org/ 10. 1111/ psyp. 13940

Gordon, P. C., Plummer, P., & Choi, W. (2013). See before you jump: Full recognition of parafoveal words precedes skips during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 39(2), 633-641. https:// doi. org/ 10. 1037/ a0028 881

Heilbron, M., van Haren, J., Hagoort, P., & de Lange, F. P. (2023). Lexical processing strongly affects reading times but not skipping during natural reading. Open Mind, 7, 757-783.

Jared, D., & Seidenberg, M. S. (1991). Does word identification proceed from spelling to sound to meaning? Journal of Experimental Psychology: General, 120(4), 358-394. https:// doi. org/ 10. 1037/0096- 3445. 120.4. 358

Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: Finding meaning in the N400 component of the event-related brain potential (ERP). Annual Review of Psychology, 62, 621-647. https:// doi. org/ 10. 1146/ annur ev. psych. 093008. 131123

Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207(4427), 203-205. https:// doi. org/ 10. 1126/ scien ce. 73506 57

Kliegl, R., Grabner, E., Rolfs, M., & Engbert, R. (2004). Length, frequency, and predictability effects of words on eye movements in reading. European Journal of Cognitive Psychology, 16(1/2), 262-284.

Kretzschmar, F., Bornkessel-Schlesewsky, I., & Schlesewsky, M. (2009). Parafoveal versus foveal N400s dissociate spreading activation from contextual fit. NeuroReport, 20(18), 1613-1618. https:// doi. org/ 10. 1097/ wnr. 0b013 e3283 32c4f4

Krügel, A., & Engbert, R. (2010). On the launch-site effect for skipped words during reading. Vision Research, 50(16), 1532-1539.

Laszlo, S., & Federmeier, K. D. (2009). A Beautiful Day in the Neighborhood: An Event-Related Potential Study of Lexical Relationships and Prediction in Context. Journal of Memory and Language, 61(3), 326-338.

Leinenger, M. (2014). Phonological coding during reading. Psychological Bulletin, 140(6), 1534-1555. https:// doi. org/ 10. 1037/ a0037 830

Leinenger, M. (2019). Survival analyses reveal how early phonological processing affects eye movements during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 45(7), 1316-1344.

Lopez-Calderon, J., & Luck, S. J. (2014). ERPLAB: an open-source toolbox for the analysis of event-related potentials. Frontiers in Human Neuroscience, 8, Article 213.

McConkie, G. W., & Yang, S. N. (2003a). Basic assumptions concerning eye-movement control during reading. Behavioral and Brain Sciences, 26(4), 493-494.

McConkie, G. W., & Yang, S. N. (2003b). How cognition affects eye movements during reading. In J. Hyönä, R. Radach, & H. Deubel (Eds.), The mind's eye: Cognitive and applied aspects of eye movement research (pp. 413-427). ScienceDirect.

Metzner, P., Von Der Malsburg, T., Vasishth, S., & Rösler, F. (2017). The importance of reading naturally: Evidence from combined recordings of eye movements and electric brain potentials. Cognitive Science, 41, 1232-1263.

Milligan, S., Antúnez, M., Barber, H. A., & Schotter, E. R. (2023a). Are eye movements and EEG on the same page?: A coregistration study on parafoveal preview and lexical frequency. Journal of Experimental Psychology: General, 152(1), 188-210.

Milligan, S., Nestor, B., Antúnez, M., & Schotter, E. R. (2023b). Out of sight, out of mind: Foveal processing is necessary for semantic integration of words into sentence context. Journal of Experimental Psychology: Human Perception and Performance, 49(5), 687-708. https:// doi. org/ 10. 1037/ xhp00 01121

Milligan, S., & Schotter, E. R. (2024). Do readers here what they sea?: Effects of lexicality, predictability, and individual differences on the phonological preview benefit. Journal of Memory and Language, 135, Article 104480.

Nestor, B., Elaboudi, A. Milligan, S., & Schotter, E.R. (2024). Orthographic and semantic preprocessing in the parafovea: Evidence from ERPs in the flankers task. Manuscript in preparation.

Newman, R. L., & Connolly, J. F. (2004). Determining the role of phonology in silent reading using event-related brain potentials. Cognitive Brain Research, 21(1), 94-105.

Newman, R. L., Jared, D. J., & Haigh, C. A. (2012). Does phonology play a role when skilled readers read high-frequency words? Evidence from ERPs. Language & Cognitive Processes., 27, 1361-1384.

Nuthmann, A., Engbert, R., & Kliegl, R. (2005). Mislocated fixations during reading and the inverted optimal viewing position effect. Vision Research, 45(17), 2201-2217.

Osterhout, L., & Holcomb, P. J. (1992). Event-related brain potentials elicited by syntactic anomaly. Journal of Memory and Language, 31(6), 785-806.

Payne, B. R., Stites, M. C., & Federmeier, K. D. (2019). Event-related brain potentials reveal how multiple aspects of semantic processing unfold across parafoveal and foveal vision during sentence reading. Psychophysiology, 56(10), Article e13432.

Plöchl, M., Ossandón, J. P., & König, P. (2012). Combining EEG and eye tracking: identification, characterization, and correction of eye movement artifacts in electroencephalographic data. Frontiers in human neuroscience, 6, 278.

Pollatsek, A., Lesch, M., Morris, R. K., & Rayner, K. (1992). Phonological codes are used in integrating information across saccades in word identification and reading. Journal of Experimental Psychology: Human Perception and Performance, 18(1), 148-162. https:// doi. org/ 10. 1037/ 0096- 1523. 18.1. 148

Pollatsek, A., Reichle, E. D., & Rayner, K. (2006). Tests of the EZ Reader model: Exploring the interface between cognition and eye-movement control. Cognitive Psychology, 52(1), 1-56.

Rayner, K. (1975). The perceptual span and peripheral cues in reading. Cognitive Psychology, 7(1), 65-81. https:// doi. org/ 10. 1016/0010- 0285(75) 90005-5

Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124(3), 372-422. https:// doi. org/ 10. 1037/ 0033- 2909. 124.3. 372

Rayner, K. (2009). Eye movements and attention in reading, scene perception, and visual search. Quarterly Journal of Experimental Psychology, 62(8), 1457-1506.

Rayner, K., & Clifton, C., Jr. (2009). Language processing in reading and speech perception is fast and incremental: Implications for event-related potential research. Biological Psychology, 80(1), 4-9.

Reichle, E. D., Pollatsek, A., Fisher, D. L., & Rayner, K. (1998). Toward a model of eye movement control in reading. Psychological Review, 105(1), 125-157.

Reichle, E. D., Pollatsek, A., & Rayner, K. (2006). E-Z Reader: A cognitive-control, serial-attention model of eyemovement behavior during reading. Cognitive systems research, 7(1), 4-22.

Reichle, E. D., Reineberg, A. E., & Schooler, J. W. (2010). Eye movements during mindless reading. Psychological Science, 21(9), 1300-1310. https:// doi. org/ 10. 1177/ 09567 97610 378686

Reichle, E. D., & Reingold, E. M. (2013). Neurophysiological constraints on the eye-mind link. Frontiers in Human Neuroscience, 7. https:// doi. org/ 10. 3389/ fnhum. 2013. 00361

Ryskin, R., Stearns, L., Bergen, L., Eddy, M., Fedorenko, E., & Gibson, E. (2021). An ERP index of real-time error correction within a noisy-channel framework of human communication. Neuropsychologia, 158, Article 107855.

Schotter, E. R. (2013). Synonyms provide semantic preview benefit in English. Journal of Memory and Language, 69(4), 619-633.

Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2012). A 21 word solution. Available at SSRN 2160588.

Schotter, E. R. (2018). Reading ahead by hedging our bets on seeing the future: Eye tracking and electrophysiology evidence for parafoveal lexical processing and saccadic control by partial word recognition. In K. D. Federmeier & D. G. Watson (Eds.), The psychology of learning and motivation: Current topics in language (pp. 263-298). Elsevier Academic Press.

Schotter, E. R., & Leinenger, M. (2016). Reversed preview benefit effects: Forced fixations emphasize the importance of parafoveal vision for efficient reading. Journal of Experimental Psychology: Human Perception and Performance, 42(12), 2039-2067. https://doi. org/ 10. 1037/ xhp00 00270

Schotter, E. R., Lee, M., Reiderman, M., & Rayner, K. (2015). The effect of contextual constraint on parafoveal processing in reading. Journal of Memory and Language, 83, 118-139.

Schotter, E. R., Milligan, S., & Estevez, V. M. (2023). Event-related potentials show that parafoveal vision is insufficient for semantic integration. Psychophysiology, 60(7), Article e14246.

Sereno, S. C., Hand, C. J., Shahid, A., Yao, B., & O'Donnell, P. J. (2018). Testing the limits of contextual constraint: Interactions with word frequency and parafoveal preview during fluent reading. Quarterly Journal of Experimental Psychology, 71(1), 302-313.

Sheridan, H., & Reingold, E. M. (2012). The time course of contextual influences during lexical ambiguity resolution: Evidence from distributional analyses of fixation durations. Memory & Cognition, 40(7), 1122-1131.

SR Research. (2017). EyeLink® 1000 Plus user's manual (Version 1.0.12). Author.

Staub, A. (2015). The effect of lexical predictability on eye movements in reading: Critical review and theoretical interpretation. Language and Linguistics Compass, 9, 311-327.

Stites, M. C., Payne, B. R., & Federmeier, K. D. (2017). Getting ahead of yourself: Parafoveal word expectancy modulates the N400 during sentence reading. Cognitive, Affective, & Behavioral Neuroscience, 17(3), 475-490. https:// doi. org/ 10. 3758/ s13415- 016- 0492-6

Taylor, W. L. (1953). "Cloze procedure": A new tool for measuring readability. Journalism Quarterly, 30(4), 415-433. https:// doi. org/10. 1177/ 10776 99053 03000 401

Van Orden, G. C. (1987). A ROWS is a ROSE: Spelling, sound, and reading. Memory & Cognition, 15(3), 181-198.

Vasilev, M. R., Yates, M., & Slattery, T. J. (2019). Do readers integrate phonological codes across saccades? A Bayesian meta-analysis and a survey of the unpublished literature. Journal of Cognition, 2(1), Article 43.

Veldre, A., & Andrews, S. (2015). Parafoveal preview benefit is modulated by the precision of skilled readers' lexical representations. Journal of Experimental Psychology: Human Perception and Performance, 41(1), 219-232. https:// doi. org/ 10. 1037/ xhp00 00017

Veldre, A., & Andrews, S. (2017). Parafoveal preview benefit in sentence reading: Independent effects of plausibility and orthographic relatedness. Psychonomic Bulletin & Review, 24, 519-528.

Veldre, A., & Andrews, S. (2018a). Beyond cloze probability: Parafoveal processing of semantic and syntactic information during reading. Journal of Memory and Language, 100, 1-17. https://doi. org/ 10. 1016/j. jml. 2017. 12. 002

Veldre, A., & Andrews, S. (2018b). Parafoveal preview effects depend on both preview plausibility and target predictability. Quarterly Journal of Experimental Psychology, 71(1), 64-74. https:// doi.org/ 10. 1080/ 17470 218. 2016. 12478 94

Veldre, A., Drieghe, D., & Andrews, S. (2017). Spelling ability selectively predicts the magnitude of disruption in unspaced text reading. Journal of Experimental Psychology: Human Perception and Performance, 43(9), 1612-1628. https:// doi. org/ 10. 1037/ xhp0000425

Veldre, A., Reichle, E. D., Wong, R., & Andrews, S. (2020). The effect of contextual plausibility on word skipping during reading. Cognition, 197, Article 104184.

Vitu, F. (2003). The basic assumptions of E-Z reader are not wellfounded. Behavioral and Brain Sciences, 26(4), 506-507.

White, S. J., Rayner, K., & Liversedge, S. P. (2005). The influence of parafoveal word length and contextual constraint on fixation durations and word skipping in reading. Psychonomic Bulletin & Review, 12, 466-471.

Wlotko, E. W., & Federmeier, K. D. (2012). So that's what you meant! Event-related potentials reveal multiple aspects of context use during construction of message-level meaning. NeuroImage, 62(1), 356-366.