Tracking the Effects of Eccentricity on the Integration of Orthographic Information From Multiple Words

Introduction

Moving window techniques, in which readers only see letters within a predefined window while reading, have shown that the perceptual span of adult readers extends to 14–15 letters to the right of the fixated word (see Rayner, 2009 for a review). With standard experimental settings (3 letters per visual degree), the span is approximately 5° to the right of the fixated word. Although the question of the quantity of textual information contained within the perceptual span has been adequately addressed through moving window reading tasks, the question of how this information is processed within this span may require the use of alternative techniques.

A recent study by Veldre et al. (2023a) tested the ability of readers to process lexical information at different retinal eccentricities of the fixation point. In this work, lexical decision had to be performed on a central target – which could be a word or a pseudoword – and could be presented at five different locations: in the center, 1° left, 2° left, 1° right and 2° right. Distractor flankers were positioned around the target, which could be a different word, a pseudoword or absent. To allow or prevent eye movement in direction of the target, stimulus display times alternated between 100 and 300 ms. The results showed a decline in accuracy concomitant with an increase in RTs with increasing retinal eccentricities of the target location. RTs were higher and accuracy was lower in the presence of distractor flankers than in their absence, and in the presence of word flankers than in the presence of pseudoword flankers.

The work of Veldre et al. is of particular interest for the present study because it supports the idea that the performance of target word recognition is not distributed uniformly across readers’ visual fields. This important work provided a new way for studying the effects of retinal eccentricity within the perceptual span, isolating the phenomenon to be studied – the eccentricity of the information – while avoiding the extra processing imposed by sentence reading, e.g., of syntax and semantics.

In another note, Dare and Shillcock (2013) introduced a lexical version of the Flankers Task paradigm to study how foveal word processing could be affected by flankers in parafovea. In this paradigm, participants performed lexical decision on a central target word or pseudoword that was surrounded by flankers pairs of letters to the left and right separated by a single space. The results showed that flankers directly impacted processing of the central word processing. Grainger et al. (2014) provided a theoretical framework for their interpretation couched in terms of parallel orthographic processing spanning on spatially distinct orthographic stimuli. Subsequently, the flanker effects were replicated with whole words that could be related or unrelated (rock rock rock vs. path rock path). Results showed that flankers related to the target word led to shorter RTs than the unrelated ones (e.g., Snell et al., 2017; Vandendaele & Grainger, 2023; Cauchi et al., 2020). In this work, we propose to use the flankers relatedness to the central word, to track the integration of orthographic information within the perceptual span.

Method

Material and Design

One-hundred-and-eighty high-frequency words (mean Zipf = 4.57, SD = .62), and 180 low-frequency words (mean Zipf = 3.10, SD = .36) were used in this experiment. All the words were four-letter Dutch words selected from the SUBTLEX database (Keuleers et al., 2010). Names and lemmatized words were discarded from the word selection. Words with the same final letters or letter repetition (e.g., piep, enne) were not selected. For both high and low-frequency words, two sets of target and flanker pairs were constituted in order to manipulate the target-flanker relatedness. In the related set, target and flanker were the same word (e.g., bron bron [source source]). In the unrelated set, target and flanker were different words (e.g., bron staf [source rod]) with as few common letters and semantic relations as possible (The Appendix contains the full list of stimuli.). The set of pseudowords was created from the word set by a single letter substitution. Pseudoword targets, associated with pseudoword flankers, were used as fillers for the purpose of the lexical decision task and were not considered in the analyses.

In order to track the interaction between relatedness and eccentricity, flankers could be located at different locations along a horizontal line covering 4.29° of visual angle on each side. At the first eccentricity, the center of the flanker string was located at 1.65° from the center of the screen (also the target string center), corresponding approximately to a single space separation¹ of .33° away from the central target string, both to the left and right of it. After that, each eccentricity was one space further away from the center, up to the point where the center of the flanker string was located at 4.29° from the center of the screen. In that way there were nine possible eccentricities (see Figure 1).

Figure 1

The experimental design allowed us to test both eccentricity and relatedness as within-subjects factors. Items were counterbalanced across participants and conditions. To do this, we created 18 lists where all 360 items were assigned to one combination of relatedness and eccentricity.

Procedure

An illustration of the procedure is provided in Figure 2. Participants were instructed that three strings of letters would briefly appear on the screen and that they had to quickly categorize the central string as a word or pseudoword, and only that one. After calibration of the eye tracker, the task started with a few trials in a training block which then gave way to the actual task. Each trial started with a centralized fixation dot used as trigger: When the participant’s gaze landed on the fixation dot, the target and flanking stimuli were presented for 170 ms. Participants had a maximum of 4000 ms to make their lexical decision with a right-handed (“p”) button press for word targets and left-handed (“q”) button press for pseudoword targets, using a qwerty keyboard layout. Following the response, a blank screen was presented for 1000 ms before the start of the next trial.

Figure 2

The eccentricity of the flankers varied from 1.65° to 4.29° of visual angle from the central target word. They could be located at 1.65°, 1.98°, 2.31°, 2.64°, 2.97°, 3.30°, 3.63°, 3.96° or 4.29° from the central target word. The 360 trials were presented in random order. A break with a duration determined by the participant was offered halfway in the experiment. The experiment started with a 20 items practice session with positive and negative feed-backs. After the practice session, the participant was notified that the experiment itself was going to start. There was no feedback during the experiment. The experiment lasted approximately 30 minutes.

Results

In order to avoid any artifact caused by low-frequency words that may have been unknown to many participants, all items with an accuracy rate under 70% were discarded. This affected 20 items in the low-frequency words set. The remaining set consisted of 160 items. Pseudoword fillers were not analyzed. Error trials (5.6%) and trials with improbable RTs below 300 ms or above 1500 ms (1.1%) were excluded. RTs were log-transformed to meet normality assumptions. A null model with a random structure including by-participant and by-item random intercepts was fitted to compute standardized residuals from the logarithmic RTs variable. All trials with standardized residuals larger than 2.5 SDs from the grand mean were excluded (2,5%).

The data were analyzed in R (R Core Team, 2022). Linear mixed-effects models (RTs) were fitted with the lmer function from lme4 (v1.1-21; Bates et al., 2014), and generalized mixed-effects models (accuracy rates) with the glmer function. Fixed effects were considered significant if |t| or |z| > 1.96 (Baayen, 2008). Post hoc analyses for significant two-way interactions used cell means coding and single df contrasts, implemented via the glht function from multcomp (v1.4-13; Hothorn et al., 2008) with normal distribution significance testing (see Table 1).

RTs analyses

Relatedness by Frequency by Eccentricity interaction

For the 3-way interaction model, the maximal converging random effect structure included by-participant and by-item random intercepts and slopes among the Relatedness by Eccentricity interaction.

There was a main effect of relatedness (b = .04, SE = .004, t = 7.86) and of word frequency (b = .03, SE = .006, t = 6.77), with faster responses in the presence of related flankers and for high-frequency words. The main effect of the eccentricity did not reach the significance (b = –.001, SE = .001, t = 1.23). The relatedness by eccentricity interaction was significant (b = –.003, SE = .008, t = –3.49). No more interactions were significant in this analysis.

To explore the effect of Relatedness at individual levels of Eccentricity, a post-hoc analysis was conducted (results are shown in Table 1). This analysis revealed a significant relatedness effect for all levels of eccentricity.

Polynomial analysis

The interaction between relatedness and eccentricity appears to be driven primarily by the variation in the unrelated condition, where RTs decrease with increasing eccentricity, while RTs in the related condition remain stable. To investigate this pattern further, we examined the effects of eccentricity separately for each relatedness level.

In the related condition, RTs showed no significant effect of eccentricity (t < 1, p > .10). In contrast, in the unrelated condition, both the linear (b = –.003, SE = .0005, t = –6.51) and quadratic (b = 0.23, SE = 0.07, t = 3.55) effects were significant, confirming a systematic decrease in RTs as eccentricity increased. A model comparison using a likelihood ratio test (anova function, stats package) showed no significant difference (p = 1) between the linear and quadratic models, preventing a clear selection of the best fit. Higher-degree polynomial models (cubic and quartic) were also tested but did not reach significance. The observed and predicted RTs across eccentricities in both related and unrelated conditions are presented in Figure 3.

Figure 3

Discussion

The aim of this work was to track the integration of orthographic information within the perceptual span with a new paradigm. We tested the effects of eccentricity in the perceptual span by presenting related and unrelated flankers to the central word, placed at different locations in the parafovea. We used an eye tracking control and limited the display duration to 170 ms to ensure that participants fixated on the targets.

Our study replicated the typical pattern found in flankers task, where unrelated flankers (e.g., path rock path) lead to longer RTs compared to related flankers (e.g., rock rock rock). These relatedness effects were initially found with bigram flankers (Dare & Schillcock, 2013; Grainger et al., 2014) and later with whole-word flankers (e.g., Snell et al., 2017; Vandendaele & Grainger, 2023; Cauchi et al., 2020). Although the typical pattern was observed in the flankers task without eye movement control, the present study demonstrates that in the context of strict eye movement control – ensuring that gaze does not shift to the flankers – the same patterns emerge as in previous studies. In fact, the magnitude of the relatedness effect at the smallest eccentricity was found to exceed that typically observed in previous studies (e.g., Snell & Grainger, 2018; Cauchi et al., 2022). This could be due to the novel eye-tracking procedure that ensured stimulus onset occurred only when participants were fixating precisely at the center of the display. By constraining participants to strictly focus the center of the target word, this approach likely enhances the sensitivity of our paradigm and may amplify the relatedness effects.

Results showed that RTs in the unrelated flankers condition decreased substantially between 1.65° and 1.98°, with a less pronounced decrease along the rest of the axis, extending to 4.29°. This observation of a non-linear decrease of the unrelated flankers condition is supported by the significance of the quadratic term in the polynomial analysis, although, model comparison could not confirm that a non-linear model provides a better fit than the linear model. Despite this decrease, post-hoc analyses revealed that the relatedness effect remained significant even for the most distant flankers, suggesting that flankers were still processed at 4.29° of visual angle from the central word.

What do the results of this study tell us? They demonstrate that parallel processing extends to large distances from the fovea, reaching the outside boundaries of the parafoveal region. In the context provided by this adapted version of the flankers task, we provide further evidence that – as proposed by Grainger et al. (2014) and recently revisited by Vandendaele & Grainger (2023) – spatial integration of orthographic information occurs across multiple spatially distinct stimuli within a single processing channel for sub-lexical orthographic processing and word identification. Importantly, the present study extends this finding by showing that orthographic information at greater distances from the foveal central word still slow its recognition. This implies that unrelated flankers still exert inhibition, perhaps through lexical competition, at a large distance. This finding would be consistent with the computational model OB1-reader (Snell, van Leipsig et al., 2018; Meeter et al., 2020), which was implemented based on the theoretical framework proposed by Grainger et al. (2014). Indeed, a key feature of this model states that processing is spatially distributed, with acuity decreasing as eccentricity increases. Our results suggest that this principle holds even for very distant flankers, reinforcing the idea that spatial integration is not limited to immediately adjacent orthographic information but can extend further into the parafovea.

Importantly, our data did not support the foveal load hypothesis (Henderson & Ferreira, 1990), which predicts that less attention would be allocated to parafoveal flankers when the foveal target word is a low frequency item. A reduction in parafoveal processing due to foveal load was previously reported by Vignali et al. (2019), who found that flanker processing decreased when the central target was a pseudoword. A similar effect was reported for semantic information: when a foveal word is semantically unexpected or incongruent, parafoveal preview benefits decrease (Payne et al., 2016). This suggests that foveal processing difficulty modulates the extent to which parafoveal words are integrated. However, our study did not reveal any difference in parafoveal processing as a function of lexical frequency: while low-frequency words led to longer RTs than high-frequency words, no interaction was observed between lexical frequency and other experimental factors. This could be explained by the fact that the experimental manipulation employed in these brief presentations primarily engages early orthographic processing mechanisms, i.e., low-level processes which are not sensitive to lexical frequency. Consequently, lexical frequency influenced the foveal target recognition but does not appear to have directly modulate spatial integration in parafovea.

Finally, we highlight that this paradigm, as a laboratory task, does not fully reflect everyday reading. Indeed, in a natural reading context, attention may be diluted by the many words simultaneously visible. The present study builds on the work of Fournet et al. (2023), where we demonstrated that proximal bigram flankers interfere more with target word processing than distant flankers. However, in Fournet et al. we did not quantify these effects in relation to visual eccentricity. The present work is an initial attempt at quantifying the spatial integration of orthographic information within the perceptual span using the change in location of the flanker relatedness manipulation.

Our results are consistent with the findings of McConkie & Rayner (1975) regarding the perceptual span. Specifically, the higher eccentricity tested in this study, 4.29°, corresponds to a spacing of 14–15 characters between the center and the last letter of the flanker. By using this new paradigm, different from the moving window technique, we provide an alternative way of assessing how orthographic information is integrated within the perceptual span.

Conclusion

This study introduced a novel version of the flankers task to examine how orthographic information is integrated across the perceptual span. Our results showed that orthographic information is not processed uniformly across the perceptual span, and we proposed a metric to describe this decline. We suggest that future studies should focus on developing readers to track the developmental progression of orthographic processing within the perceptual span using this paradigm. This study examined the influence of parafoveal information with bilateral rather than unidirectional flankers, given the attentional asymmetry towards the right with Latin alphabets (see Ducrot & Grainger, 2007 for a review), it would be valuable to explore how the central word processing is impacted by the flanker location, whether on the right or left. This could be addressed using unilateral flanker displays, as in Snell and Grainger (2018), or through the presentation of lateralized non-lexical characters (e.g., “#”), as in Cauchi et al. (2024). Furthermore, recent findings from Lázaro et al. (2025) suggest that the contribution of flankers may depend not only on their spatial location but also on the degree of orthographic overlap with the target, emphasizing the need to disentangle spatial and linguistic influences on parafoveal processing. Finally, as underlined by Veldre et al. (2023b), the relationship between eccentricity and orthographic processing is not well defined in current reading models, and we agree that it would be interesting to see whether they could account for these findings.

Data Accessibility Statement

Data, statistical analysis files and materials are available from this OSF link: https://osf.io/2eu7s/?view_only=aa13bc0fb38648b396bcbdec4d188146.

Ethics and consent

Ethical approval for this study was granted by the institutional review board of the Faculty of Behavioral & Movement Sciences of Vrije Universiteit Amsterdam (VCWE #2023-068).

Competing Interests

The authors have no competing interests to declare.

Author contributions

CC: Conceptualization, Methodology, Writing, Data curation.

MM: Conceptualization, Methodology, Writing, Supervision.