Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Individuals with dyslexia use a different visual sampling strategy to read text


Individuals with dyslexia present with reading-related deficits including inaccurate and/or less fluent word recognition and poor decoding abilities. Slow reading speed and worse text comprehension can occur as secondary consequences of these deficits. Reports of visual symptoms such as atypical eye movements during reading gave rise to a search for these deficits’ underlying mechanisms. This study sought to replicate established behavioral deficits in reading and cognitive processing speed while investigating their underlying mechanisms in more detail by developing a comprehensive profile of eye movements specific to reading in adult dyslexia. Using a validated standardized reading assessment, our findings confirm a reading speed deficit among adults with dyslexia. We observed different eye movements in readers with dyslexia across numerous eye movement metrics including the duration of a stop (i.e., fixation), the length of jumps (i.e., saccades), and the number of times a reader’s eyes expressed a jump atypical for reading. We conclude that individuals with dyslexia visually sample written information in a laborious and more effortful manner that is fundamentally different from those without dyslexia. Our findings suggest a mix of aberrant cognitive linguistic and oculomotor processes being present in adults with dyslexia.


Within today’s society, quick and accurate reading skills are essential to participate in societal activities (e.g., education, using social media, communication) and to achieve professional success. Reading is a complex task requiring the eyes, specifically the fovea of the retina, to stop on the written text to successfully encode the sequence of letters that make up each word, and will eventually be assigned meaning (i.e., semantics) during decoding and matching with one’s mental lexicon. The stops in eye movements, termed fixations, are an essential part of reading1,2 and a prerequisite for adequate decoding of a text’s content including successful word selection from the mental lexicon3. Each fixation is complemented by a brief jump in eye position, termed a saccade, which relocates the fovea to the next part of the text to be processed. Saccades are typically 7–9 characters wide1 and take 20–30 ms during reading4. The time available for completing a fixation and its associated linguistic tasks, and planning the next eye movement (a saccade) is approximately 225–250 ms per stop1,4,5. This duration may be even shorter due to the preview benefit of text appearing outside of the fovea6,7. These findings underline the importance of an automatized, rapid interplay of low-level eye movements with higher-level cognitive processing during reading such as lexical, phonological, and semantic processing1,8,9.

One group that is well-known to struggle with developing proficient and fast reading skills are individuals with dyslexia. Dyslexia is a language-based, neurobiological specific learning disorder affecting reading, writing, and spelling10,11, which persists into adulthood10,12,13,14. Specifically, struggles with accurate and/or fluent word recognition and decoding abilities that typically result from deficits in phonological awareness characterize this disorder10,11. Slow reading and a deficit in reading comprehension can be resulting secondary consequences10,11,15. An estimated 5 to 20% of the population are affected11,12,15,16. Dyslexia’s aetiology remains the subject of a heated debate with proponents attributing the main underlying cause to deficits in a variety of systems associated with reading (i.e., phonological awareness, visuo-spatial attention17, magnocellular and cerebellar function18,19,20,21, or a lack of reading experience22,23). While deficits in phonological awareness are considered established24,25,26,27,28, other deficits in low-level sensory processing29,30,31,32,33 and visual attention remain under scrutiny (e.g.,34).

Frequently, researchers who investigate the reading skills and strategies of children and adolescents35,36,37 or adults38,39,40 with dyslexia during sentence reading have focused on reading speed as a measure of performance. These studies report that readers with dyslexia read at a slower rate (i.e., fewer words per minute) compared to readers without dyslexia35,37,38,39,40. The difference in reading rates between affected and non-affected adults with dyslexia can equate to the difference observed in early readers39,40,41. However, reading speed rates neither provide insight into the cognitive mechanisms nor the visual sampling strategy by which readers with dyslexia may decode written text differently.

Recent evidence stresses that at least one subtype of dyslexia is affected by differences in visual processing42, which can be detected within eye movement recordings43,44. For example, Nilsson Benfatto and colleagues43 were able to reliably distinguish between 9–10 year-old children at high- and low-risk of dyslexia using a classification algorithm operating on one-minute eye movement recordings. Specifically, the duration of fixations, and the number of fixations, saccades and regressions were found to be the most predictive eye movements for differentiating between children at high and low risk of persistent reading difficulties. This finding is in line with previous research showing that readers with dyslexia exhibit longer fixation durations45,46,47, an increased number of fixations47, shorter saccades1,46,47,48, and fewer skipped words3,49,50. Conversely, the probability of revisiting a previous part of a text (i.e., expressing a regressive saccade also termed regression) has not proven to be reliably different in dyslexia3,46. It remains unknown, however, if a similar pattern classification is possible for adults with dyslexia across a range of different texts that are presented in different fonts.

Many advances in dyslexia eye movement research have been made in recent decades. Most of our current knowledge about differences in eye movements in dyslexia is provided by researchers investigating either a limited number of eye movement metrics in relation to specific linguistic aspects most often embodied by a target word (e.g.,51,52,53), or is limited by the use of a large variety of often controlled but non-standardized linguistic stimuli in several languages with varying orthographic depth ranging from one character up to one or two sentences54,55,56,57. Hence, a comprehensive profile of the eye movements of adults with dyslexia during naturalistic reading of standardized texts of multiple sentences remains surprisingly unknown. The development of a comprehensive profile would allow to uncover and quantify potential inefficiencies in visual sampling of text that have not come to light using the aforementioned focused, local approach. Therefore, we aim to devise a comprehensive eye movement account of adult dyslexia by investigating how eye movement patterns of individuals with dyslexia differ from those without dyslexia on global (text-based) and local (word-based) reading measures during an ecologically valid silent paragraph reading task in English (Fig. 1a).

Linguistic parameters such as the difficulty of a text58,59, its syntax60,61, word length and word frequency62 can impact eye movements, highlighting the importance of using standardized and validated stimuli. To this end, we employed the International Reading Speed Texts (IReST63). The IReST were developed for standardized reading speed assessment, and consist of ten independent texts, each of about 150 words using novel content. The texts have been equated on a global level for their number of words, syntax, sentence complexity and text difficulty, and were designed to be used in repeated measures within-participant paradigms. Each text was accompanied by one brief multiple-choice question that intended to keep readers attentive and read for comprehension (Fig. 1c,d). The IReST and their accompanying questions were recently validated in a Canadian sample of adult readers attending university64. This validation sample was similar in age and education compared to the present study’s sample, though, it was exclusively comprised of adults without reading disability. Perceptual parameters such as the properties of fonts (e.g., spacing) are another aspect that has been shown to affect reading performance52,65,66,67,68. In an attempt to alleviate the reading struggles of dyslexic readers, designers have developed dyslexia-friendly fonts such as OpenDyslexic69 and Dyslexie70. These fonts omit serifs, increase inter- and intra-word spacing, and have unique letter strokes. Interestingly, these manipulations have not been found to increase reading speed36,71,72. To avoid obtaining an eye movement profile biased by font type, we presented half of the texts in OpenDyslexic (Fig. 1b,c).

Based on previous research (e.g.,13,73), we hypothesize that individuals with dyslexia, compared to an age- and education-matched control group without dyslexia, will take longer to read each text and show slower visual processing speed that is in turn linked to one’s reading duration. Reading texts in the dyslexia-friendly font OpenDyslexic is not expected to result in increased reading speed. In terms of eye movements, we hypothesize that readers with dyslexia will express more eye movements (i.e., fixations, saccades, and regressions), longer fixations and shorter saccades. Scanpaths of readers affected by dyslexia are hypothesized to be longer and to differ in their sequence and duration of eye movement events as a result of increases in reading duration.


In this study, we focused on group-level differences in behavior and eye movements between adults with and without dyslexia. Behavioral analysis included an investigation of the dependent variables reading duration, attention to the text and non-linguistic cognitive processing speed as a function of the two experimental groups (i.e., Dyslexia and Control). Eye movement analyses examined global (i.e., paragraph/trial-based) and local (i.e., word/interest area-based) metrics of eye movement events during reading.

Behavioral results

We constructed a generalized linear mixed-effects model (GLMM) for analyzing reading duration as a function of the predictors: group (Dyslexia and Control) and font (Times New Roman and OpenDyslexic), and their font-by-group interaction on a single-trial level. The predictor font was also included as a random effect, which was allowed to vary by participant. In addition, the predictors group and font and their interaction were included as random effects, which were allowed to vary by text. This model was based on 601 experimental trials (i.e., one value for median reading duration per trial) collected from all 67 participants. Text number five had to be excluded across all participants due to a stimulus presentation issue (67 trials; 10%). Two further trials from the dyslexia group had to be excluded due to recording issues, which resulted in 601 trials being included in all analyses (for details, see “Methods”).

This GLMM demonstrates significant predictive power of the main effect of group on median reading duration with individuals with dyslexia taking longer to read each text (X2 = 13.431, df = 1, p < 0.001; Fig. 2a,b; see Table 1 for detailed model statistics). This speed difference is underlined by a lower words per minute reading rate among readers with dyslexia (MedianDyslexia = 178.09, MedianControl = 248.18; two-sided independent sample t-test: t(65) = 20.51, p < 0.0001; g = 1.67, 95% CIg = [1.486, 1.858]; Fig. 2a). However, this model yields neither a significant improvement in reading duration with font (X2 = 1.41, df = 1, p = 0.235) nor a significant font-by-group interaction (X2 = 0.446, df = 1, p = 0.504).

Figure 1
figure 1

Experimental paradigm and example of stimuli. (a) Pictorial depiction of the sequence of events of one trial. The entire paradigm was participant controlled by pressing the space bar. A drift correction circle (the circle was smaller than depicted) directed the participant’s eyes to the starting location of the text. While fixating at it, participants pressed the space bar to get a text to appear that was then read silently once. Immediately after participants finished reading, they were asked to answer a short multiple-choice attention question relating to the content of the preceding text without time pressure. This process was repeated ten times. (b) Example of one text and its corresponding attention question (panel d) displayed in Times New Roman font. Note that this text was not presented and just constitutes an example comparable to the texts from the commercial reading assessment (IReST63). The original texts are protected. (c) Example of the same text and its multiple-choice question displayed in OpenDyslexic font. (d) Example of the multiple-choice attention question accompanying the texts shown in panels b and c. Attention questions were always presented in the same font than their preceding text.

Table 1 Generalized linear mixed-effects model fixed effect parameter estimates.

Secondly, the multiple-choice questions presented immediately after reading each text served as an incentive for participants to read each text for comprehension—hence, constituting an indicator of attention. Both groups demonstrate attention to the texts clearly above chance level (control: t(31) = 17.67, p < 0.0001; BF10 = 5.1457 × 1014; Min = 66.67%; dyslexia: t(34) = 15.96, p < 0.0001; BF10 = 2.2968 × 1014; Min = 55.56%; all two-sided paired t-tests; Fig. 2c). That is, two participants in the dyslexia group scoring at 55.6% and 57.1%, nine participants across both groups showing performance at 66.7%, and the remaining 56 (out of 67) participants answered more than 75% of the attention questions correctly, with 31.3% of all participants answering all questions correctly. Crucially, our data show no significant evidence for a difference between both groups (two-sided independent samples t-test: t(65) = − 0.34, p = 0.7349; BF10 = 0.2635; Fig. 2c). Altogether, this analysis demonstrates that both groups paid attention to the reading material, as their performance is better than chance across all trials.

Thirdly, we examined if there were any non-linguistic cognitive processing speed differences between our two groups using two subcomponents from the Wechsler Adult Intelligence Scale (Coding and Symbol Search74). This analysis was motivated by previous reports of links between reading speed and slower cognitive processing speed in individuals with dyslexia75. We find that individuals with dyslexia exhibit slower processing speed on the Coding test (two-sided independent samples t-test: t(65) = 5.88, p < 0.0001; g = 1.422, 95% CIg = [0.895, 1.973]; Table 2; Fig. 2d), but not on the Symbol Search test (independent samples t-test: t(65) = 0.399, p = 0.69; g = − 0.1, 95% CIg = [− 0.382, 0.577]; Table 2; Fig. 2d). Our data further show a negative correlation between coding speed and reading duration across all participants (r65 = − 0.51, p < 0.0001, 95% CIr = [− 0.680, − 0.315]; Fig. 2e) suggesting that, in the present study, participants with better coding ability (i.e., faster number related cognitive processing speed) exhibit shorter reading duration. Symbol search speed did not correlate with reading speed across participants (r65 = 0.12, p = 0.33, 95% CIr = [− 0.106, 0.35]). Although these single measures both probe visual processing speed, their separate interpretation warrants caution, since they may not provide a full representation of one’s processing capabilities as outlined in their use76.

Table 2 Descriptive statistics of group characteristics.

These results appear to be the consequence of the control group showing coding speed above the general population average while the dyslexia group shows performance slightly below the general population average (Mpopulation = 10, MControl = 12.84, MDyslexia = 9.40; Table 2; Fig. 2d). One reason for these results might be that the coding task encompasses working memory performance to some degree77. Memorizing digit-symbol pairs only shown at the top of the page more quickly may constitute a strategy for achieving a higher score on this test. Hence, although not explicitly testing working memory performance, these results may be indicative of working memory deficits in adults with dyslexia when compared to a similarly educated non-dyslexia group—in line with previous reports78,79,80,81,82—and their role in achieving age- and education-appropriate reading speed. However, we find no general visual processing speed deficit in dyslexia as both the Coding and Symbol Search test need to be considered in unison76.

In short, our behavioral results show a sustained level of attention to the stimulus material throughout the majority of this study by most participants. Though, readers with dyslexia exhibit generally slower reading speed in line with previous reports. One potential explanation of the observed reading speed deficit might be a difference in the skills probed by the non-linguistic Coding processing speed test but not a general visual processing speed difference.

Eye movement profile

This study aims to devise a comprehensive characterization of the eye movement profile of individuals with dyslexia during natural paragraph reading. This profile comprises eye movement metrics traditionally examined in the field (Fig. 3), and other more recent metrics such as line-initial fixation duration, scanpath similarity, and specific saccades atypical for reading (Fig. 4). The reading-related metrics covered in this study include global (i.e., trial-/text-based) and local (single-word based) metrics. To establish the significance of a group difference in the frequentist sense, we use unbiased effect sizes83 (i.e., the 95% confidence interval of Hedges’ g not including zero; denoted as g in the text; Fig. 5a). A negative effect size indicates a longer duration or larger number exhibited by the dyslexia group and vice versa. High collinearity between some metrics included in our analyses did not allow for the use of a meaningful linear regression approach. In general, where applicable, we report group means alongside group medians to complement this robust measure of central tendency and ensure comparability to previous literature.

Figure 2
figure 2

Behavioral results. (a) Caterpillar plot of median reading speed (words per minute) per participant sorted in ascending order. Blue dots represent participants of the dyslexia group whereas yellow dots represent participants of the control group. (b) Single-trial reading duration in seconds by group. Groups are color-coded as in (a). (c) Attention to the texts as a measure of reading comprehension. Color notation as in (a). (d) Cognitive processing speed from the WAIS-IV scale74 by group. Higher scores illustrate faster processing speed. The left-hand side (i.e., yellow/light color) of each violin plot depicts scores of the control group; the right-hand side (i.e., blue/dark color) depicts scores of the dyslexia group. Red crosses denote group means, black squares group medians. (e) Correlation between reading duration (seconds) and standardized coding processing speed scores across all participants. The shown correlation coefficient (r) and p-value resulted from a robust bend correlation analysis (n = 67). Colors indicate down-weighted data points: red for data in X, green for data in Y and black for data in X and Y dimensions. In each dimension, 20% of the data points were down-weighted. This figure is best viewed in color.

Traditionally, average fixation duration and saccade length have been investigated on a global trial-level as indices of cognitive processing effort. In an effort to replicate previous findings, we computed the same metrics showing that readers with dyslexia stop longer as illustrated by longer mean and median fixation duration (gMean = − 1.07, 95% CIg = [− 1.244, − 0.902]; gMedian = − 0.96, 95% CIg = [− 1.124, − 0.786]; 83% of readers with dyslexia show this effect; Figs. 3a and 5a,b), and scan more area of the text as indicated by a longer total scanpath (g = − 0.54, 95% CIg = [− 0.706, − 0.380]; Figs. 3b and 5a,b). While scanning the text, the dyslexia group exhibits shorter average saccade amplitude (i.e., eye movement “jumps” between fixations; gMean = 1.15, 95% CIg = [0.978, 1.323]; gMedian = 1.10, 95% CIg = [0.930, 1.273]; 100% of readers with dyslexia show this effect; Figs. 3c and 5a,b), and increased variance (i.e., standard deviation) in these saccade amplitudes (g = 0.70, 95% CIg = [0.495, 0.823]). This pattern of results suggests that adults with dyslexia employ a more laborious visual sampling strategy on a global text-level. Longer fixation durations and shorter saccade amplitudes indicate that this group spends more time on information uptake and analysis per stop (i.e., fixation), while also potentially taking up less information per unit of time. A similar pattern of eye movements has previously been associated with individuals who are learning to read2, and those who are considered poor readers56. The observed longer total scanpaths among readers with dyslexia are a logical consequence of this group’s substantial increase in reading duration.

Figure 3
figure 3

Group comparison of selected traditional eye movement metrics. Plots display trial-based eye movement metrics that showed significant differences between groups. Each panel depicts the group comparison collapsed across fonts as a raincloud plot for the respective metric. Kernel density plots depict the frequency of occurrence of a value while scatterplots display the underlying values as in one average value per trial. Boxplots indicate the median, upper and lower quartile, and whiskers the 95% CI. Blue (dark) color represents data of the dyslexia group whereas yellow (bright) color depicts data of the control group. (a) Median fixation duration in milliseconds. (b) Total scan path in degrees of visual angle. (c) Median saccade amplitude in degrees of visual angle. (d) First run dwell time in seconds. That is, the sum of all fixation durations during a first visit to a word if it has not been visited or skipped before. (e) Ratio of words that were skipped during first-pass reading. This excludes all fixations on a word that occurred after a regression to a previous word was completed. (f) Number of regressions. That is, leftward saccades to preceding words that have already been visited or skipped.

Global metrics, however, do not provide more detailed information on specific elements of the visual sampling strategy in relation to single words such as how many of these words get actively fixated and how often. To address these questions, we complemented the reported global metrics with metrics based on the definition of interest areas around single words. We find that individuals with dyslexia spent more time (g = − 1.57, 95% CIg = [− 1.75, − 1.39]; Fig. 3d and First Run Dwell Time in Fig. 5a,b), and skip fewer words (g = 0.40, 95% CIg = [0.235, 0.557]; Fig. 3e and Ratio First Run Words Skipped in Fig. 5a,b) during first pass reading (i.e., the sum of all first fixations on a word in reading direction excluding any revisits or skipped words). In line, readers with dyslexia fixate on more words in a given trial when all fixations are examined (g = − 1.312, 95% CIg = [− 1.488, − 1.135]; Ratio Visited Words in Fig. 5a), and stop more frequently per word on average (g = − 1.27, 95% CIg = [− 1.445, − 1.095]; Number Fixations per Word in Fig. 5a). Further, revisits of earlier parts of a text (i.e., leftward saccades to a preceding word formally called regressions) are a substantial and frequent part of natural reading. We observe that readers with dyslexia express more regressions per text (g = − 0.82, 95% CIg = [− 0.989, − 0.656]; Figs. 3f and 5a,b). However, given the increase in the number of saccades as a result of longer reading durations, this increase in the number of regressions yields no significant difference in the probability of making a regression across an entire text (g = − 0.05, 95% CIg = [− 0.213, 0.107]; Fig. 5a,b).

Furthermore, since the control group shows coding processing speed above the population average, we examined this group’s link between fast coding speed and the traditional eye movement metrics reported above. This analysis shows no correlation between coding speed and any of the reported eye movement metrics (absolute range r30 = 0.016–0.32, all ps > 0.05), which suggests that faster coding speed does not systematically affect the eye movements of readers without dyslexia. We also observe no correlation for the dyslexia group (absolute range r33 = 0.019–0.14, all ps > 0.05).

Taken together, our results on traditional eye movement metrics corroborate previous findings from investigations with readers affected by dyslexia. They demonstrate that these readers examine a given text more slowly and in smaller steps, even without accounting for any revisits of previous words (Fig. 5b). Since efficient reading was found to be characterized by skipping over many words (up to 90%3) during the first rightward scanning of a text in reading direction (termed, first-pass reading), the observed pattern strongly suggests that inefficiencies are introduced by processing less content simultaneously as well as slower information uptake and longer cognitive processing times of text. Crucially, these findings are based on data obtained from natural reading of standardized texts consisting of multiple lines.

Further contemporary metrics of ocular movements during reading

Recently, additional metrics have been proposed to differentiate between oculomotor deficiencies and cognitive, linguistic factors underlying longer fixation times84. Line-initial fixations are one such metric. They constitute the first fixation on one of the first words of a line that is not followed by a leftwards correction within the same line. Uniquely, line-initial fixations do not allow the reader early access to a word’s coarse visual orthographic percept due to absent parafoveal preview. Hence, they have been proposed as an unconfounded indicator of linguistic processing time84,85. By contrasting groups on the duration of line-initial fixations, we find these to be longer in the dyslexia group on a single-fixation (g = − 0.33, 95% CIg = [− 0.383, − 0.276]; MedianDyslexia = 231 ms, MedianControl = 202 ms; Fig. 4a), and single-trial level (g = − 0.81, 95% CIg = [− 0.972, − 0.64]; MedianDyslexia = 256 ms, MedianControl = 216 ms; 89% of readers with dyslexia show this effect; Figs. 4b and 5b). Parker and colleagues84 reported an effect in the same direction when comparing accurate line-initial fixation durations between children and adults without dyslexia, with children showing longer fixation durations. This finding adds to the evidence indicating that readers with dyslexia take longer to process the visual and linguistic information sampled during a fixation. It further supports the notion that the visual sampling strategy of readers with dyslexia resembles the strategy of early readers without dyslexia.

Figure 4
figure 4

Contemporary eye movements typical and atypical for reading. (a) Duration of all identified line-initial fixations by group. That is, the very first “valid” fixation on the first two words of a line not followed by a leftwards corrective saccade. Boxplots indicate the median, upper and lower quartile, and whiskers the 95% CI. Blue (dark) color represents data of the dyslexia group whereas yellow (bright) color depicts data of the control group. (b) Median line-initial fixation duration per trial by group. Color scheme and box plot properties as in (a). (c) Group comparison of the frequency of directional deviations by angle across all trials depicted as polar histograms. Saccade angle and frequency are depicted in bins of 2.75° width. Only saccade angles between ± 35° and ± 145° (displayed as 35/145° and 215/325°) qualified as directional deviations. 0° equals horizontal rightward direction of reading. All other saccades were excluded from this analysis. d) Example of two scanpaths after reading the same text (IReST #10) displayed in Times New Roman font. Scanpaths are depicted over time (y-axis) and word-based interest areas (x-axis). Top scanpath depicts trial by a control participant, whereas the bottom scanpath depicts a trial expressed by a reader with dyslexia. (e) Pairwise, trial-based similarity matrix illustrating differences in spatial and temporal fixation patterns upon normalization by reading duration. Higher scores (darker color) denote higher dissimilarity. Pink square indicates the Scasim dissimilarity score for the two trials shown in (d). (f) Map of clusters of trials in the spatial domain (using the Euclidean distance metric) after multidimensional scaling. Red and black colored circles denote trials allocated to separate clusters. Grey diamonds indicate the centers of these clusters. Trials framed in yellow and blue represent the spatial equivalent of the same trials whose scanpaths are depicted in panel d. Colors mark groups.

To ensure that we selected line-initial fixations accurately, we compared their duration to the overall median fixation duration of a trial. Previous research shows that fixation duration decreases as readers move their eyes towards the end of a line84,86. As expected, the identified line-initial fixations are of longer duration than all fixations considered together (gDyslexia = 0.87, 95% CIg = [0.673, 1.071]; line-initialDyslexia = 256 ms; AllDyslexia = 224 ms; gControl = 0.57, 95% CIg = [0.361, 0.768]; line-initialControl = 216 ms, AllControl = 202 ms). We find this increase in line-initial fixation duration to be larger in the dyslexia group (gGroups = − 0.37, 95% CIg = [− 0.526, − 0.204]; 32 vs 14 ms). The same pattern of eye movement results reported above holds true when analyzing only trials whose attention questions were answered correctly.

Specific divergence from a regular visual sampling strategy

Besides the presented group differences on global and local eye movement metrics, we noticed a clear divergence from a regular left-to-right visual sampling strategy among readers with dyslexia. To quantify these divergences of eye movements that we consider atypical for reading, we examined saccades with angles that would not be expected during the natural reading flow (henceforth, directional deviations).

During left-to-right reading of text most saccades will either be expressed in a horizontal rightward direction in line with the expected reading direction or in mostly horizontal leftward direction as revisits of preceding text or in an occasional return-sweep saccade to the beginning of the next line. Yet, not all observed saccades would satisfy any of these typical categories of eye movements, since their expressed angle deviates substantially from the ones expected as part of the usual reading flow. We examined the number of saccades with an angle that considerably deviates from the expected pattern of saccade angles during typical reading (i.e., angles between 35° and 145° upwards and − 35° and − 145° downwards from the horizontal reading plane). To avoid that falsely programmed return-sweeps were mistaken for directional deviations, we excluded a number of saccades that could be attributed to other eye movements typically involved in reading such as blinks, accurate and inaccurate return-sweeps (for details, see “Methods”). These corrections resulted in 522 saccades being identified as directional deviations.

Readers with dyslexia express directional deviations more than twice as often per trial on average (g = − 0.48, 95% CIg = [0.316, 0.641]; MeanDyslexia = 1.003, MeanControl = 0.441, varDyslexia = 2.08, varControl = 0.61; 69% of readers with dyslexia show more directional deviations per trial than the control group’s average; Figs. 4c and 5b), which signals a more frequent loss of place at unexpected points during the reading process. Remarkably, in the dyslexia group most of the identified directional deviations were directed straight downwards, whereas this pattern was virtually reversed in readers without dyslexia (Fig. 4c). Since even a brief scanning of the area of text just below the current fixation seems rather unintuitive from a cognitive perspective, this finding raises the question whether these directional deviations are the result of occasional issues with oculomotor control previously reported in dyslexia29,31,32,33,87,88,89,90,91,92,93.

The aforementioned differences in eye movements are part of the overall visual sampling strategy of text during reading, termed a scanpath. To investigate whether readers with and without dyslexia differ only on some eye movement metrics or rather use a divergent overall visual sampling strategy, we complemented the previous analyses with a computational similarity analysis of the overall scanpath of each trial. To this end, we quantified the temporal and spatial similarity of the fixations of all scanpaths employing a version of the Scasim analysis94. The aim of this trial-based analysis was to identify clusters of trials with similar scanpath patterns, while achieving independence of the observed group differences in reading time. To identify whether trials of readers with dyslexia were more (dis)similar to those of other readers with dyslexia, we compared the number of trials associated with each group within a given cluster. Similarity scores and clusters were computed separately for each text of the IReST battery and font type, as this coordinate-based analysis is highly sensitive to differences in spacing such as those introduced by text displayed in differently spaced font types (Fig. 1b,c). In this study, trials were equally split between Times New Roman and OpenDyslexic font types. Additionally, all trials were normalized by their reading duration to avoid the introduction of trivial differences between scanpaths of different lengths.

Figure 5
figure 5

Summary of effects and visual sampling strategy group profiles. (a) Effect sizes and their 95% confidence intervals (CIs) of the effect of group on eye movement metrics. Positive effect sizes (i.e., Hedges’ g) illustrate a higher number of events, longer duration or distance or a larger ratio among control participants. A negative effect size illustrates the opposite effect. Red dots denote the effect size and black bars the 95% CI for each effect size. CIs computed using the exact analytical method as implemented in the measures of effect size toolbox83. Effect sizes were considered significant if the 95% CI did not include zero. (b) Radar plot depicting overall group differences in the eye movement and reading profile given selected metrics. Plots depict group averages after all trials of each measure were normalized (i.e., z-scored) for comparability. Counter-clockwise direction follows presentation order as in (a). If two variants of the same metric were present in panel a, only one of them is displayed on the radar plot for simplicity.

Pairwise scanpath similarity scores showed that trials of participants with dyslexia differed from those of participants without, even though participants read identical texts (Fig. 4d–f). Upon normalizing, trial-by-trial similarity scores indicated that dyslexic participants spend a substantial amount of the time (between ~ 34 and 83%) looking at different places on the identical paragraph and/or for different durations compared to their non-dyslexic counterparts (Fig. 4e). These similarity scores were subsequently transferred to the spatial domain (for details, see “Methods”), where we find that the optimal number of group-independent clusters ranges between two and five clusters per text-font pair. Trials of each group were predominantly allocated to separate clusters, and demonstrate a significant difference of association for about 75% of text-font pairs (p < 0.05; for detailed statistics, see Table S3 in the supplementary material). Thus, we find that readers with dyslexia sample identical texts using a different sequence of fixations (i.e., fixating different locations on the text and/or for different durations) than non-dyslexics—even when differences in reading time are accounted for.

To summarize, our findings demonstrate that readers with dyslexia use a generally more laborious and inefficient visual sampling strategy during natural reading. The virtually opposite pattern of directional deviations between groups points towards the existence of occasional deficiencies in oculomotor control that result in dyslexic readers losing their place more often. Replicating previous findings, their laborious strategy is characterized by longer average and line-initial fixation duration, prolonged first run dwell time as well as shorter saccade amplitude and fewer skipped words. Contrarily, the probability of revisiting preceding words was comparable between groups. This pattern of eye movements suggests that prolonged time for cognitive, linguistic processes such as word decoding, lexical access, and/or phonological decoding underlies the behavioral difficulties associated with dyslexia such as substantially slower reading speed; but not an increased need for resolving semantic or syntactic ambiguities through reanalysis of prior text. Altogether, these results indicate that an interplay of linguistic and oculomotor factors underlies the reading struggles in adults with dyslexia.


In this study, we used eye-tracking to devise a comprehensive eye movement profile of the visual sampling strategy of adult readers with dyslexia during naturalistic reading of standardized multi-sentence texts in English (IReST63). Here, combining traditional and contemporary eye movement metrics, we show fundamental differences between readers with and without dyslexia on all but one of the examined metrics. These results, in combination with substantial decreases in reading speed, illustrate a laborious and more effortful reading strategy in adulthood, resembling a pattern observed in beginning2 and poorer readers56.

The idea that eye movements differ between readers with and without dyslexia is not new. Rayner1,48 was among the first to report different eye movements during reading based on anecdotal case studies with only three dyslexics. His investigations were followed by numerous cross-sectional studies using separate samples of readers with dyslexia, and largely varying stimuli in languages with different orthographic depth (for reviews, see2,95). This variety of stimuli, typically consisting of hand-picked single words or short sentences that impose artificial task demands on the reader rather than allowing for an ecologically valid natural reading scenario, constitutes an issue in the field96. The use of standardized and validated multi-sentence texts remains scarce in the literature.

In this work, we were particularly interested in reconciling many previously separate accounts of differential eye movements using the same sample of individuals with dyslexia while also examining specific indicators of oculomotor deficiencies during natural reading. Our results replicate previous findings of differential eye movements in children and adults with dyslexia such as longer fixation durations, fewer skipped words during first-pass reading and repeated fixations on the same word1,3,43,44,45,46,47,48,49,50,97. Oculomotor control has commonly been investigated using a variety of non-linguistic saccade tracking and fixation stability tasks29,33,87,88,90,91,93. Here, we show that specific saccades atypical for reading can be detected during natural reading. A twofold likelihood of expressing such a saccade, termed directional deviation, indicates signs of occasional oculomotor deficiencies in dyslexia—in line with previous reports29,31,32,33,87,88,89,90,91,92,93.

Given that the eye movement profile of children with dyslexia during paragraph reading has previously been exploited for dyslexia screening43,44, it is worth asking whether the inclusion of metrics on separate levels of granularity (i.e., the local single-word and global paragraph level) in adults improve our understanding of eye movements in dyslexia? Our approach differs crucially from these two screening studies on several points. Firstly, both studies were conducted with children around the age of 10 using recordings obtained from reading only one non-validated text with short lines. Secondly, these studies aimed to identify the most parsimonious model that classified recordings accurately as stemming from a child with or without dyslexia. This focus on reducing complexity in the data precluded devising a comprehensive profile, and may have resulted in overlooking smaller but informative differences such as directional deviations. Thirdly, this model-focused approach did not allow for addressing specific hypothesis-driven questions using targeted measures such as line-initial fixations. Hence, these child studies and our adult study complement each other by establishing an eye movement profile of dyslexia at different ages that consists of a diverse set of metrics.

To answer the central question about the source of the reading struggles of individuals with dyslexia, all differences across the entire profile need to be considered. These differences can be explained in the context of established eye movement models that are able to simulate a wide range of reading related patterns including fixation duration, multiple fixations on a word, parafoveal preview benefit, regressions, and spill-over effects—E-Z Reader62,98,99,100, SWIFT101, OB1-Reader6. While these models were devised to explain the reading process of typical readers, they can also provide insight into the processes of beginning100,102 and dyslexic readers3,99. Particularly, the extensive research on the E-Z Reader model, and its prior application to data obtained from readers with dyslexia, makes it an interesting model for our study. It is a serial processing model which posits that processing of the fixated word occurs in four separate stages, beginning as soon as the preceding word (N) is fixated on and attention is allocated. First, information about the upcoming word (N + 1) is extracted from the parafovea during pre-attentive visual processing. Second, a familiarity check is performed upon fixation. The length of this familiarity check is dependent on the word’s frequency and length. Once a word is processed for lexical familiarity, it signals the initiation of an eye movement. Third, completion of full lexical access leads to a shift in attention to the next word (N + 1) in the form of a saccade. Fourth, an integration of the identified word in the syntactic context is carried out simultaneously and may trigger a regressive saccade, if a syntactic or semantic conflict is detected62,98,99.

In the case of individuals with dyslexia who present with longer fixation durations, as seen in the current study, this model posits that these individuals experience slower lexical access, associated with increased lexical processing demands. Where skilled readers require less time to perform the usually fast familiarity check (i.e., finding a match for the letter string making up a word), individuals with dyslexia do not seem to be able to carry out this process equally fast. The dyslexia group also expressed shorter saccades and skipped fewer words during first-pass reading—in line with previous findings1,3,48. These findings, in combination with a higher frequency of fixating on the same word repeatedly, corroborate the notion of needing to process each word or even its sub-components individually, and for longer, when reading for comprehension. Prolonged line-initial fixations of individuals with dyslexia in our study provide more evidence for delays in lexical access (stages two and three). These first fixations on a line do not benefit from any parafoveal preview benefit (stage one) resulting in the sole reliance of information sampled during this fixation for word identification purposes.

While a deficit in lexical processing can explain longer fixation durations for individuals with dyslexia, a deficit in parafoveal processing could likewise explain increased fixation durations, shorter saccade amplitudes and fewer skipped words. The preview benefit takes advantage of orthographic information from parafoveal vision such as word length and word familiarity. It is linked with a reader’s perceptual span, which is defined as the number of distinct characters from which useful information can be acquired in parallel across the fovea and parafovea52,103,104. Should individuals with dyslexia present with such a deficit, removing or having a reduction in this preview benefit could also result in the need for longer processing (i.e., fixation durations) when the next word is being fixated on simply due to reduced pre-processing of its orthographic percept105. Such a smaller perceptual span has previously been associated with reading speed106,107, and reported in dyslexia2,108. This deficit can occur independently of a phonological deficit109, however, previous evidence is inconclusive110,111. In the present study, line-initial fixation duration can serve as an indicator of parafoveal processing when compared to all other fixation durations across a sentence. The former are usually longer due to absent parafoveal preview while the duration of all other fixations decreases towards the end of a line84,85,86. These line-initial fixations were longer for all readers compared to the median fixation duration across a trial. This difference was larger for readers with dyslexia, which suggests that these readers need even more processing time when no preview benefit is available. In other words, readers with dyslexia do not appear to be disadvantaged when the preview benefit for words to the right of a fixation is available. This result demonstrates that a reduction in the perceptual span is unlikely the explanation of the general increases in fixation duration, as seen in this study’s sample with dyslexia.

Another conceivable reason for explaining dyslexics’ reading struggles is that readers with dyslexia encounter increased semantic and syntactic processing difficulties. These difficulties would lead to revisits of previous words, as illustrated by the frequently observed increase in the number of regressive saccades1,45,48,58,89. In the E-Z Reader model, regressions are conceptualized as difficulty with integrating the identified word within the greater sentence context—occurring during a fourth post-lexical integration stage. In line with previous literature, we observed a greater number of regressions per trial in individuals with dyslexia. Vagge and colleagues89, similar to our work, reported this result for a study that also found slower reading speed. However, the higher number of regressions in both studies may simply be a result of longer reading durations, since we did not find a difference in the probability of making a regressive saccade in relation to the overall number of saccades of a trial. This absent regression probability difference replicated an earlier report with dyslexic adults3. With the E-Z Reader model in mind, our findings speak against a deficit in the post-lexical integration stage of reading leading to the reading struggles observed in the present study99.

Although the traditional eye movement metrics reported in this work suggest that prolonged lexical processing time underlies the slower reading speed and visual profile in dyslexia, the role of oculomotor deficiencies is far from being established in the literature19,34,95,112. Their role has been at the center of the on-going dyslexia debate. To investigate potential oculomotor deficiencies during natural reading, as opposed to saccadic tracking and fixation stability tasks, we examined saccades that are unexpected or atypical in the reading flow of a skilled reader. We should consider horizontal rightward, leftward and diagonal return-sweep saccades reading related. One may even include saccades pointing upwards in this list, since they might constitute a sudden double-checking of previously read content. However, readers with dyslexia jumped away from some fixations in directions atypical for reading such as straight downwards more often than their counterparts. We found the directional pattern of these atypical eye movements to be virtually reversed between groups. This indicates that common eye movements such as blinks were not falsely identified as directional deviations in either group, as they should have resulted in a similar pattern of directional deviations (for details, see “Methods”). The nature and frequency (i.e., ~ 1 per trial) in the dyslexia group suggest that these directional deviations may be a residual of oculomotor deficiencies that persist, but only occur occasionally, in adults with dyslexia attending higher education. Thus, differences in oculomotor control appear unlikely to be the origin of their reading struggles—at least in the present sample of individuals with dyslexia.

One conceivable concern of the present study’s design is that the observed eye movement profile could be simply a result of texts that were too difficult for readers with dyslexia, since increases in text difficulty were reported to lead to more eye movements in dyslexic children58, and a pattern similar to the one observed in this study in readers with1 and without dyslexia2. However, given that the IReST are designed at a grade six reading level and all of the participants in the present sample had previously attended or were attending higher education at the time of participation, unsuitable text difficulty is very unlikely to be the reason for the observed differences.

A second concern is that we did not observe a ceiling effect in the responses to our multiple-choice questions—with two participants performing just above chance and 31.3% of all participants answering all questions correctly. Although there are multiple reasons as to why not all participants answered all questions correctly, one is the very specific nature of some questions, which makes those somewhat more challenging for some readers. Importantly, we found no evidence for a difference between the two groups. Such a difference could have raised concerns about it exerting effects on the eye movement contrasts in this study.

Third, we aimed to provide a comprehensive descriptive profile that can be paired with predictions from established models of eye movement control. However, this profile itself may not be used to confirm causality of the deficits underlying the general struggles of individuals with dyslexia. Particularly, given that the present study tested adults and administered only a hypothesis-driven assessment of selected cognitive skills such as visual processing speed, more information could prove useful in future studies to pinpoint the exact origins of the reading struggles and their causal links. It is further conceivable that differences on reading tasks in adults originate from less or reduced reading experience as opposed to neurobiological consequences of dyslexia per se23. Therefore, the presented eye movement profile may serve as a basis for more targeted investigations in the future probing the specific nature of the presumed deficit in lexical access and its direct causal link to the cognitive mechanisms of the reading difficulties. In this respect, the extensive debate on the aetiology of dyslexia is in dire need of more longitudinal investigations starting in kindergarteners to increase the likelihood of being able to establish such causal links reliably through observing the pre-reading state and its change over time.

In summary, the presented eye movement profile of adults with dyslexia demonstrates a laborious and effortful visual sampling strategy when reading multiline paragraphs of text. Specifically, the combination of prolonged fixation duration, shorter saccade amplitude and fewer skipped words suggests deficits in the linguistic processing components of reading such as fast and efficient access to the mental lexicon. Longer line-initial fixation durations were particularly indicative of prolonged lexical analysis. On the contrary, we did not find convincing evidence for a perceptual span deficit or increased difficulties in the semantic or syntactic post-lexical processing stage of reading. An increased number of eye movements atypical for reading shows that the eyes of readers with dyslexia occasionally move to seemingly random places on a page. Hence, occasional oculomotor deficiencies should not be categorically dismissed in dyslexia.



We tested 73 participants: 35 adults with an official diagnosis of dyslexia, and 38 without symptoms of dyslexia. Six individuals from the control group were excluded from all analyses due to large inaccuracies during the calibration procedure (best eye with average error > 0.5° and max error > 1.3°). Hence, the final data analysis was conducted on 67 participants: 35 with dyslexia (female = 23, Meanage = 23.54, SDage = 6.22) and 32 without (female = 32, Meanage = 22.38, SDage = 2.7).

To delineate between control participants who experience dyslexia symptoms but have not been given an official diagnosis, and to get a measure of severity of dyslexia symptoms at the time of participation, all participants completed the Adult Dyslexia Checklist113. This checklist assesses aspects of literacy, language, word finding, and organization skills on a scale of 1–4 (i.e., rarely / occasionally / often / most of the time). As specified by the original authors, a score of 45 or more points indicates mild to severe dyslexia symptoms113. We used a score of ≤ 40 points as a conservative cut-off for our control group, with all formally diagnosed dyslexics allocated to the dyslexia group.

Participants were matched on age and level of education. They were either current or former college or university students at anglophone institutions in Canada. Since participants were recruited in Montréal, a bilingual English-French city, our sample comprises both bilingual and monolingual English speakers (bilingualDyslexia = 18, monolingualDyslexia = 17; bilingualControl = 9, monolingualControl = 23). To avoid introducing a language effect, we compared bilingual to monolingual participants and found no evidence for a difference in reading duration in the dyslexia group (t(33) = 1.829, p = 0.0765; g = 0.6, 95% CIDyslexia = [− 0.064, 1.264]; BF10 = 1.16) nor the control group (t(30) = − 0.5, p = 0.6212; g = − 0.19, 95% CIControl = [− 0.94, 0.56]; BF10 = 0.4). We neither observed a language effect regarding attention to the text in the dyslexia group (t(33) = 1.375, p = 0.1784; g = 0.45, 95% CIDyslexia = [− 0.206, 1.108]; BF10 = 0.67), and insufficient evidence in the control group (t(30) = − 2.241, p = 0.0326; g = − 0.86, 95% CIControl = [− 1.634, − 0.071]; BF10 = 2.16). Based on these findings, groups were collapsed across language for all analyses. Written informed consent was obtained from all participants. Participants could choose between receiving $10 or course credit as compensation. This study adhered to the Canadian Tri-council Policy on ethical conduct for research involving humans114, and obtained approval by the Concordia University Human Ethics Research Committee (certificate: 30003975).


In this study, we used the International Reading Speed Texts as stimuli63. These are 10 texts that have been equated based on their number of words, syntax, sentence complexity and text difficulty (Mtext = 153.6 words, rangetext = 140–160; Msentence = 8.9, rangesentence = 8–11), and are designed for use as repeated measures within participants. Additional to their original validation in a UK sample, the IReST texts have been validated in an adult Canadian sample comparable in age and education to the present sample (validation sample: n = 25; M = 21.72, SD = 4.94, range = 1–41)64. None of the participants in this validation study reported having reading or attention disabilities. No participant took part in both studies conducted at Concordia University (i.e., the Canadian validation and present eye-tracking study). As is common practice in reading studies, each text was accompanied by one short multiple-choice question with three options of which one was correct (for examples, see Fig. 1b–d). This attention question was presented immediately after reading a text. These questions have been used in other reading studies64,115. Five texts and their accompanying multiple-choice questions were displayed in Times New Roman font, with the other five texts and questions being displayed in the specific dyslexia font OpenDyslexic69. Sentence start and end words were identical between the font versions resulting in identical launch and landing interest area sites across font conditions. To equate for differences in physical text size, we varied the nominal font sizes using 20-point size for Times New Roman and 18-point size for OpenDyslexic trials. Text height was comparable. Every text and its respective multiple-choice question were presented in the same font. Due to copyright restrictions by virtue of the text being a commercial reading assessment, we are unable to provide the reader with the 10 IReST paragraphs.

The Wechsler Adult Scale of Intelligence’s Symbol Search and Coding subtests were administered to assess processing speed abilities of all participants (WAIS-IV74). Importantly, both subtests use non-linguistic stimuli. In the Symbol Search task, participants are shown two target symbols and are instructed to identify both of the target symbols within the adjacent search group. This task involves no working memory as the symbols change for each trial (i.e., by horizontal search group of five symbols). Contrarily, the Coding task may involve aspects of working memory77. Participants are shown numbers 1–9 and their unique corresponding symbol at the top of the page. Here, the task is to draw the corresponding symbol associated with each number below a sequence of numbers. The WAIS has an internal consistency score of 0.87–0.98 on processing speed index tasks. The interscorer agreement ranges from 0.98 to 0.99, and intraclass correlation from 0.91 to 0.97116. Correlations between scores on tests that measure similar constructs were in the 0.8 range on criterion-related validity measures117.


The Adult Dyslexia Checklist113 was first completed, after which the experimenter administered the WAIS Symbol Search and Coding tasks74. Subsequently, participants were calibrated to the eye tracker (EyeLink 1000) by looking at a series of 9 equally spaced dots on grey background across the screen (using the SR EyeLink’s inbuilt 9-point calibration procedure with targets in the default locations). A validation procedure using the same 9 points in a randomized order confirmed the accuracy of the eye tracker’s calibration measurements, with participants’ better eye needed an accuracy of < 0.5 degrees on average and no point exceeding 1.3 degrees of maximum error. Upon successful calibration, participants were instructed on screen and verbally to read the 10 texts once thoroughly in silence. They may go back and reread words or entire sentences but once they had reached the end of the text and attempted to start from the beginning they were immediately stopped by the experimenter. Each text was presented separately (five in TNR font, five in OpenDyslexic). We displayed texts in the upper half of the screen with 83.57 characters per line on average and left alignment (Fig. 1a). Once a participant finished reading, they pressed the spacebar, subsequently a multiple-choice question concerning the previously read text appeared. We randomized the presentation order of these 10 texts within-subjects (using the randperm function in MATLAB). An additional text with similar linguistic properties, comparable to the text in Fig. 1b,c and a multiple-choice question was presented as a practice trial in Arial regular font. After completing the reading portion, participants filled out a questionnaire concerning their experience reading the 10 texts as well as demographic information such as age, mother tongue and education (see supplementary information for full questionnaire). In total, the experiment lasted between 30 and 45 min.


Stimuli were presented and data collected using an iMac (2011 27″ i7, 16 GB RAM) with an external monitor (View Sonic G225fb 21″ CRT, 1024 X 768 pixel resolution, 100 Hz refresh rate). A chin rest was used to stabilize head position at a distance of 70 cm from the screen. Eye position was acquired non-invasively using a video-based eye movement monitor (EyeLink 1000 running host software version 4.56, SR Research, Ottawa, Ontario).

Eye tracking analysis

Eye movement data were recorded at a sampling rate of 1000 Hz and stored for offline analysis. DataViewer’s inbuilt algorithms (version 4.1.1, SR Research, 2019, Ottawa, Ontario) were used for the pre-processing of fixations, saccades, and blinks, forming reading-related interest areas and trial-based aggregate measures. An interest period that excluded the first and last 300 ms of each trial was defined in DataViewer to avoid contaminating this analysis with reading unrelated events at the very beginning and end of each trial. The duration of fixations that spanned any of these two cut-off time points was trimmed. Further, the minimum saccade amplitude was set to 0.5°, the fixation merging amplitude to 1°, and the minimum fixation duration to 50 ms. Fixations separated by a blink were not merged. Instead we removed fixations immediately before and after a blink. Fixations beyond display bounds (i.e., the entire screen) were excluded. In general, we analyzed only data of one eye per participant and excluded all samples that were identified by any of the aforementioned criteria from all further analyses. These analysis parameters help to remove outliers caused by random eye movements that are unrelated to reading.

The interest area analysis was word based in that one interest area was associated with each word including five pixels of padding around all sides of a word. A background RGB threshold of less than 350 was chosen to fill gaps between interest areas. Although all fixations were drift corrected by the drift value obtained at the start of each trial, we manually adjusted all fixations of a trial vertically (13.9% of all analyzed trials) if visual inspection showed that fixations exhibited an obvious vertical offset across all lines of a text resulting in them lying on interest area boundaries. Importantly, we neither moved single fixations separately nor adjusted fixations horizontally.

Results of the offline analysis with DataViewer (version 4.1.1, SR Research, 2019, Ottawa, Ontario) were exported for use with custom scripts in MATLAB (version 2020a, The MathWorks Inc., 2020, Natick, Massachusetts). There, we calculated all measures split by experimental conditions (i.e., group and font). We excluded all trials presenting text number five in either font due to a stimulus presentation issue and to avoid any bias (10% of all trials). A further two trials had to be excluded due to recording issues. To quantify and compare the effect of group (i.e., Dyslexia vs Control) in detail, we computed unbiased signed between-group effect sizes (g) and their respective 95% confidence intervals separately for each eye movement metric (using the mes function of the Measures of Effect Size Toolbox83 and its exact analytical method for determining confidence intervals). Following frequentist logic, a significant effect of group was presumed when the 95% confidence interval of an effect size did not include zero. In our design, positive effect sizes represent a higher number or ratio of the respective eye-tracking metric for the control group compared to the dyslexia group and vice versa (Fig. 5a). As well, we estimated the probability density function corresponding to selected eye movement metrics whose between-group comparison yielded a significant effect size employing kernel density estimation in MATLAB (using the raincloud_plot function118). In doing so, we created a probability density heat map for all selected eye movement metrics.

In addition to traditional reading eye movement metrics, we also examined line-initial fixations. These fixations have been proposed to be able to dissociate binocular coordination from linguistic analysis of text/words85. We identified the first fixation on any of the first two words of a line that was not followed by an undersweep corrective saccade to the left of this fixation as accurate line-initial fixation. Group differences within these fixations were then quantified using Hedges’ g.

Furthermore, we compared the two experimental groups on saccades that are unlikely or unexpected during typical reading behavior/flow. Such saccades were the ones showing radical angular shifts with an angle between ± 35° and ± 145°, which we termed directional deviations. These saccades could not be classified as (in)accurate forward, regression, undersweep, return-sweep or blink saccades. In the computation of these directional deviations, we excluded angles of rightward and leftward saccades typically and frequently involved in reading. All saccades with a blink before, during or after the directional deviation saccade in question, as identified by DataViewer’s built-in algorithm (version 4.1.1, SR Research, 2019, Ottawa, Ontario), did not qualify either. Further, saccades representing correctly programmed return-sweeps to the beginning of the next line were excluded85. Return-sweeps were identified by finding potential directional deviations that were launched within 3 words of the final word of a line and landed within the first three words (i.e., interest areas) of the next line. Last but not least, incorrectly programmed return-sweeps (so-called “undersweeps”84) were excluded from this analysis as well, since they are not assumed to be involved in on-going linguistic processing86. These had to be launched from any of the final four words of a line.

As an overall measure of scanpath similarity, we computed the Scasim metric94. This trial-based metric compares the location (x-, y-coordinates) and duration of all fixations that make up a scanpath and can be computed using the scasim function provided by the first author’s GitHub repository119. The resulting score represents a value of dissimilarity. We normalized all scores by their respective reading duration to avoid confounds due to large differences in reading duration. Scasim scores are computed on a pairwise trial-by-trial basis. Since this measure uses the x-,y-coordinates of a fixation’s location, we computed Scasim scores per IReST and font a text was displayed in. This was necessary as differences in font led to words of the same text being displayed in slightly different locations. Trials were analyzed across groups to begin with. However, to quantify scanpath similarity between readers with and without dyslexia, spatial maps of scanpaths were fit on similarity scores using Euclidean distances (dist function) and non-metric multidimensional scaling (isoMDS function from the MASS package120). Subsequently, we used the optimal number of clusters, a result of a calculation of Gaussian mixture models paired with the Bayesian Information Criterion121 (mclustBIC function of the mclust package122), as the number of clusters in a k-means clustering procedure (kmeans function). Since the determined clusters were still comprised of group-independent trials, we employed chi-square and Fisher exact tests of association analyzing whether trials in each cluster belonged to the same or a different experimental group. The entire Scasim analysis was conducted in RStudio123.

Statistical analysis

To investigate potential differences in reading duration as a function of dyslexia (i.e., group factor) and font, we used a generalized linear mixed-effects model (GLMM). Eye movement data was primarily analyzed using unbiased effect sizes (i.e., Hedges’ g; denoted as g in the text) and their exact analytical 95% confidence interval83. For selected eye movement metrics, we also computed the proportion of participants in the dyslexia group whose trial-based average showed performance in line with their group when compared to the average of all participants in the control group (i.e., above or below the control group’s mean).

The GLMM analysis was performed by means of the lme4 package124 and the bobyca optimizer in RStudio123. Reading duration was specified as a continuous dependent variable, and examined, using a gamma model in the family argument of the glmer function, as a function of the two categorical predictors: group (i.e., Dyslexia and Control), and font (i.e., Times New Roman and OpenDyslexic), and their interaction on a single-trial level. The GLMM included the maximal random effects structure justified by the experimental design125. They included all main effects and interactions of our two predictors, group and font, as well as by-subject and by-item random intercepts and random slopes for all relevant main effects. We excluded random correlations for this model. The 95% confidence intervals were calculated for all \(\upbeta\) estimates (using the broom package and Wald method in RStudio123). We accounted for small imbalances in trial numbers of the predictors’ levels by entering all predictors in mean-centred form (deviation coding). All entered predictors were checked for collinearity (using the cor function and model output). Lastly, we used post-hoc likelihood-ratio (X2) model comparisons to quantify the predictive power and exact significance level of all initially significant or trending effects (i.e., p < 0.1) revealed by the GLMM.

To examine potential effects of non-linguistic visual cognitive processing speed on reading duration (i.e., the WAIS subscale scores), particularly in light of a difference in reading duration between experimental groups, we correlated the standardized scores for both processing speed measures with reading duration across participants. This and all other correlations were computed employing robust bend correlations and the default of 20% bending in each direction126.

To investigate each groups’ attention to the reading material separately, we first compared their results on our multiple-choice attention questions against chance (i.e., 50%) using two separate two-sided one-sample t-tests as well as a Bayes Factor analysis. We also contrasted the two experimental groups against each other by means of a two-sided independent samples t-test, under the assumption that both groups would pay equal attention to the reading material. In determining significance, all t-tests were Bonferroni-corrected by the number of t-tests evaluating this dependent variable (n = 3). The Bayes Factor analysis allowed us to quantify the strength of the evidence in support of the null hypothesis of no difference when compared against 0 or between groups. Due to a lack of previous research using Bayes Factors in this area, we used an uninformed prior for the Bayesian analyses with a Cauchy width of 0.7.

Data availability

Data supporting this work is available from the project’s Open Science Framework repository [].

Code availability

Code to reproduce selected figures and analyses is available from the project’s Open Science Framework repository [].


  1. Rayner, K. The role of eye movements in learning to read and reading disability. Remedial Spec. Educ. 6, 53–60 (1985).

    Article  Google Scholar 

  2. Rayner, K. Eye movements in reading and information processing: 20 years of research. Psychol. Bull. 124, 372–422 (1998).

    CAS  PubMed  Article  Google Scholar 

  3. Hawelka, S., Gagl, B. & Wimmer, H. A dual-route perspective on eye movements of dyslexic readers. Cognition 115, 367–379 (2010).

    PubMed  PubMed Central  Article  Google Scholar 

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

    PubMed  Article  Google Scholar 

  5. Sereno, S. C. & Rayner, K. Measuring word recognition in reading: Eye movements and event-related potentials. Trends Cogn. Sci. 7, 489–493 (2003).

    PubMed  Article  Google Scholar 

  6. Snell, J., van Leipsig, S., Grainger, J. & Meeter, M. OB1-reader: A model of word recognition and eye movements in text reading. Psychol. Rev. 125, 969–984 (2018).

    PubMed  Article  Google Scholar 

  7. Snell, J. & Grainger, J. Readers are parallel processors. Trends Cogn. Sci. 23, 537–546 (2019).

    PubMed  Article  Google Scholar 

  8. Dien, J. The neurocognitive basis of reading single words as seen through early latency ERPs: A model of converging pathways. Biol. Psychol. 80, 10–22 (2009).

    PubMed  Article  Google Scholar 

  9. Hauk, O., Davis, M. H., Ford, M., Pulvermüller, F. & Marslen-Wilson, W. D. The time course of visual word recognition as revealed by linear regression analysis of ERP data. Neuroimage 30, 1383–1400 (2006).

    CAS  PubMed  Article  Google Scholar 

  10. Lyon, G. R., Shaywitz, S. E. & Shaywitz, B. A. A definition of dyslexia. Ann Dyslexia 53, 1–14 (2003).

    Article  Google Scholar 

  11. American Psychiatric Association. Diagnostic and statistical manual of mental disorders. (American Psychiatric Association, 2013).

  12. Shaywitz, S. E. et al. Neural systems for compensation and persistence: Young adult outcome of childhood reading disability. Biol. Psychiatry 54, 25–33 (2003).

    PubMed  Article  Google Scholar 

  13. Swanson, H. L. & Hsieh, C.-J.C.-J. Reading disabilities in adults: A selective meta-analysis of the literature. Rev. Educ. Res. 79, 1362–1390 (2009).

    Article  Google Scholar 

  14. Vellutino, F. R., Fletcher, J. M., Snowling, M. J. & Scanlon, D. M. Specific reading disability (dyslexia): What have we learned in the past four decades?. J. Child Psychol. Psychiatry 45, 2–40 (2004).

    PubMed  Article  Google Scholar 

  15. Shaywitz, S. E. Dyslexia. N. Engl. J. Med. 338, 307–312 (1998).

    CAS  PubMed  Article  Google Scholar 

  16. Wagner, R. K. et al. The prevalence of dyslexia: A new approach to its estimation. J. Learn. Disabil. 53, 354–365 (2020).

    PubMed  Article  PubMed Central  Google Scholar 

  17. Lobier, M., Zoubrinetzky, R. & Valdois, S. The visual attention span deficit in dyslexia is visual and not verbal. Cortex 48, 768–773 (2012).

    PubMed  Article  Google Scholar 

  18. Stein, J. F. The magnocellular theory of developmental dyslexia. Dyslexia 7, 12–36 (2001).

    CAS  PubMed  Article  Google Scholar 

  19. Stein, J. F. The current status of the magnocellular theory of developmental dyslexia. Neuropsychologia 7, 12–36 (2018).

    Google Scholar 

  20. Stein, J. F. & Walsh, V. To see but not to read: The magnocellular theory of dyslexia. Trends Neurosci. 20, 147–152 (1997).

    CAS  PubMed  Article  Google Scholar 

  21. Stoodley, C. J. & Stein, J. F. The cerebellum and dyslexia. Cortex 47, 101–116 (2011).

    PubMed  Article  Google Scholar 

  22. Huettig, F., Lachmann, T., Reis, A. & Petersson, K. M. Distinguishing cause from effect–many deficits associated with developmental dyslexia may be a consequence of reduced and suboptimal reading experience. Lang. Cogn. Neurosci. 33, 333–350 (2018).

    Article  Google Scholar 

  23. Goswami, U. Sensory theories of developmental dyslexia: Three challenges for research. Nat. Rev. Neurosci. 16, 43–54 (2015).

    CAS  PubMed  Article  Google Scholar 

  24. Ramus, F. Developmental dyslexia: Specific phonological deficit or general sensorimotor dysfunction?. Curr. Opin. Neurobiol. 13, 212–218 (2003).

    CAS  PubMed  Article  Google Scholar 

  25. Ramus, F. Neuroimaging sheds new light on the phonological deficit in dyslexia. Trends Cogn. Sci. 18, 274–275 (2014).

    PubMed  Article  Google Scholar 

  26. Saksida, A. et al. Phonological skills, visual attention span, and visual stress in developmental dyslexia. Dev. Psychol. 52, 1503–1516 (2016).

    PubMed  Article  Google Scholar 

  27. Snowling, M. J. The development of grapheme-phoneme correspondence in normal and dyslexic readers. J. Exp. Child Psychol. 29, 294–305 (1980).

    CAS  PubMed  Article  Google Scholar 

  28. Snowling, M. J. Phonemic deficits in developmental dyslexia. Psychol. Res. 43, 219–234 (1981).

    CAS  PubMed  Article  Google Scholar 

  29. Raghuram, A., Gowrisankaran, S. & Swanson, E. Frequency of visual deficits in children with developmental dyslexia. JAMA Ophthalmol. 136, 1089–1095 (2018).

    PubMed  PubMed Central  Article  Google Scholar 

  30. Raghuram, A., Hunter, D. G., Gowrisankaran, S. & Waber, D. P. Self-reported visual symptoms in children with developmental dyslexia. Vis. Res. 155, 11–16 (2019).

    PubMed  Article  Google Scholar 

  31. Pavlidis, G. T. Do eye movements hold the key to dyslexia?. Neuropsychologia 19, 57–64 (1981).

    CAS  PubMed  Article  Google Scholar 

  32. Fischer, B. & Hartnegg, K. Stability of gaze control in dyslexia. Strabismus 8, 119–122 (2000).

    CAS  PubMed  Article  Google Scholar 

  33. Biscaldi, M., Fischer, B. & Hartnegg, K. Voluntary saccadic control in dyslexia. Perception 29, 509–521 (2000).

    CAS  PubMed  Article  Google Scholar 

  34. Blythe, H. I., Kirkby, J. A. & Liversedge, S. P. Comments on: “What is developmental dyslexia?” brain Sci 2018, 8, 26. The relationship between eye movements and reading difficulties. Brain Sci. 8, 100 (2018).

    PubMed Central  Article  Google Scholar 

  35. O’Brien, B. A., Mansfield, J. S. & Legge, G. E. The effect of print size on reading speed in dyslexia. J. Res. Read. 28, 332–349 (2005).

    PubMed  PubMed Central  Article  Google Scholar 

  36. Marinus, E. et al. A special font for people with dyslexia: Does it work and if so, why?. Dyslexia 22, 233–244 (2016).

    PubMed  Article  Google Scholar 

  37. O’Brien, B. A., Mansfield, J. S. & Legge, G. E. The effect of contrast on reading speed in dyslexia. Vis. Res. 40, 1921–1935 (2000).

    PubMed  Article  Google Scholar 

  38. Re, A. M., Tressoldi, P. E., Cornoldi, C. & Lucangeli, D. Which tasks best discriminate between dyslexic university students and controls in a transparent language?. Dyslexia 17, 227–241 (2011).

    PubMed  Article  Google Scholar 

  39. Callens, M., Tops, W. & Brysbaert, M. Cognitive profile of students who enter higher education with an indication of dyslexia. PLoS ONE 7, e38081 (2012).

    ADS  CAS  PubMed  PubMed Central  Article  Google Scholar 

  40. Suarez-Coalla, P. & Cuetos, F. Reading difficulties in Spanish adults with dyslexia. Educ. Technol. Res. Dev. 65, 33–51 (2015).

    Google Scholar 

  41. Adler-Grinberg, D. & Stark, L. Eye movements, scanpaths, and dyslexia. Am. J. Optom. Physiol. Opt. 55, 557–570 (1978).

    CAS  PubMed  Article  Google Scholar 

  42. Joo, S. J., White, A. L., Strodtman, D. J. & Yeatman, J. D. Optimizing text for an individual’s visual system: The contribution of visual crowding to reading difficulties. Cortex 103, 291–301 (2018).

    PubMed  Article  Google Scholar 

  43. Nilsson Benfatto, M. et al. Screening for dyslexia using eye tracking during reading. PLoS ONE 11, e0165508 (2016).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  44. Smyrnakis, I. et al. RADAR: A novel fast-screening method for reading difficulties with special focus on dyslexia. PLoS ONE 12, e0182597 (2017).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  45. Razuk, M., Barela, J. A., Peyre, H., Gerard, C. L. & Bucci, M. P. Eye movements and postural control in dyslexic children performing different visual tasks. PLoS ONE 13, e0198001 (2018).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  46. De Luca, M., Di Pace, E., Judica, A., Spinelli, D. & Zoccolotti, P. Eye movement patterns in linguistic and non-linguistic tasks in developmental surface dyslexia. Neuropsychologia 37, 1407–1420 (1999).

    PubMed  Article  Google Scholar 

  47. Hutzler, F. & Wimmer, H. Eye movements of dyslexic children when reading in a regular orthography. Brain Lang. 89, 235–242 (2004).

    PubMed  Article  Google Scholar 

  48. Rayner, K. Eye movements, perceptual span, and reading disability. Ann. Dyslexia 33, 163–173 (1983).

    Article  Google Scholar 

  49. Bucci, M. P., Brémond-Gignac, D. & Kapoula, Z. Poor binocular coordination of saccades in dyslexic children. Graefe’s Arch. Clin. Exp. Ophthalmol. 246, 417–428 (2008).

    Article  Google Scholar 

  50. Jainta, S. & Kapoula, Z. Dyslexic children are confronted with unstable binocular fixation while reading. PLoS ONE 6, e18694 (2011).

    ADS  CAS  PubMed  PubMed Central  Article  Google Scholar 

  51. Hawelka, S., Huber, C. & Wimmer, H. Impaired visual processing of letter and digit strings in adult dyslexic readers. Vis. Res. 46, 718–723 (2006).

    PubMed  Article  Google Scholar 

  52. Rayner, K., Slattery, T. J. & Bélanger, N. N. Eye movements, the perceptual span, and reading speed. Psychon. Bull. Rev. 17, 834–839 (2010).

    PubMed  PubMed Central  Article  Google Scholar 

  53. Slattery, T. J. & Yates, M. Word skipping: Effects of word length, predictability, spelling and reading skill. Q. J. Exp. Psychol. 71, 250–259 (2018).

    Article  Google Scholar 

  54. Drieghe, D., Veldre, A., Fitzsimmons, G., Ashby, J. & Andrews, S. The influence of number of syllables on word skipping during reading revisited. Psychon. Bull. Rev. 26, 616–621 (2019).

    PubMed  Article  Google Scholar 

  55. Degno, F. et al. Parafoveal previews and lexical frequency in natural reading: Evidence from eye movements and fixation-related potentials. J. Exp. Psychol. Gen. 148, 453–474 (2019).

    PubMed  Article  Google Scholar 

  56. Ashby, J., Rayner, K. & Clifton, C. Eye movements of highly skilled and average readers: Differential effects of frequency and predictability. Q. J. Exp. Psychol. Sect. A 58, 1065–1086 (2005).

    Article  Google Scholar 

  57. Engelmann, F., Vasishth, S., Engbert, R. & Kliegl, R. A framework for modeling the interaction of syntactic processing and eye movement control. Top. Cogn. Sci. 5, 452–474 (2013).

    PubMed  Article  Google Scholar 

  58. Trauzettel-Klosinski, S. et al. Eye movements in German-speaking children with and without dyslexia when reading aloud. Acta Ophthalmol. 88, 681–691 (2010).

    PubMed  Article  Google Scholar 

  59. Rayner, K. & Duffy, S. A. Lexical complexity and fixation times in reading: Effects of word frequency, verb complexity, and lexical ambiguity. Mem. Cognit. 14, 191–201 (1986).

    CAS  PubMed  Article  Google Scholar 

  60. von der Malsburg, T. & Vasishth, S. Scanpaths reveal syntactic underspecification and reanalysis strategies. Lang. Cogn. Process. 28, 1545–1578 (2013).

    Article  Google Scholar 

  61. von der Malsburg, T., Kliegl, R. & Vasishth, S. Determinants of scanpath regularity in reading. Cogn. Sci. 39, 1675–1703 (2015).

    PubMed  Article  Google Scholar 

  62. Reichle, E. D., Pollatsek, A., Fisher, D. L. & Rayner, K. Toward a model of eye movement control in reading. Psychol. Rev. 105, 125–157 (1998).

    CAS  PubMed  Article  Google Scholar 

  63. Trauzettel-Klosinski, S. & Dietz, K. Standardized assessment of reading performance: The new international reading speed texts IReST. Investig. Ophthalmol. Vis. Sci. 53, 5452–5461 (2012).

    Article  Google Scholar 

  64. Morrice, E., Hughes, J., Stark, Z., Wittich, W. & Johnson, A. Validation of the international reading speed texts in a Canadian sample. Optom. Vis. Sci. 97, 509–517 (2020).

    PubMed  Article  Google Scholar 

  65. Zorzi, M. et al. Extra-large letter spacing improves reading in dyslexia. Proc. Natl. Acad. Sci. USA 109, 11455–11459 (2012).

    ADS  CAS  PubMed  Article  PubMed Central  Google Scholar 

  66. Dotan, S. & Katzir, T. Mind the gap: Increased inter-letter spacing as a means of improving reading performance. J. Exp. Child Psychol. 174, 13–28 (2018).

    PubMed  Article  Google Scholar 

  67. Hakvoort, B., van den Boer, M., Leenaars, T., Bos, P. & Tijms, J. Improvements in reading accuracy as a result of increased interletter spacing are not specific to children with dyslexia. J. Exp. Child Psychol. 164, 101–116 (2017).

    PubMed  Article  Google Scholar 

  68. Sjoblom, A. M., Eaton, E. & Stagg, S. D. The effects of letter spacing and coloured overlays on reading speed and accuracy in adult dyslexia. Br. J. Educ. Psychol. 86, 630–639 (2016).

    PubMed  Article  Google Scholar 

  69. Gonzalez, A. OpenDyslexic. (2011).

  70. Boer, C. T. Dyslexie Font. (2008).

  71. Kuster, S. M., van Weerdenburg, M., Gompel, M. & Bosman, A. M. T. Dyslexie font does not benefit reading in children with or without dyslexia. Ann. Dyslexia 68, 25–42 (2018).

    PubMed  Article  Google Scholar 

  72. Wery, J. J. & Diliberto, J. A. The effect of a specialized dyslexia font, OpenDyslexic, on reading rate and accuracy. Ann. Dyslexia 67, 114–127 (2017).

    PubMed  Article  Google Scholar 

  73. Spinelli, D., De Luca, M., Judica, A. & Zoccolotti, P. Crowding effects on word identification in developmental dyslexia. Cortex 38, 179–200 (2002).

    PubMed  Article  Google Scholar 

  74. Wechsler, D. Wechsler Abbreviated Scale of Intelligence: WASI-IV. (NCS Pearson Inc., 2008).

  75. Breznitz, Z. & Misra, M. Speed of processing of the visual-orthographic and auditory-phonological systems in adult dyslexics: The contribution of ‘asynchrony’ to word recognition deficits. Brain Lang. 85, 486–502 (2003).

    PubMed  Article  Google Scholar 

  76. Lichtenberger, E. O. & Kaufman, A. S. Essentials of WAIS-IV Assessment (Wiley, New York, 2012).

    Google Scholar 

  77. Joy, S., Kaplan, E. & Fein, D. Speed and memory in the WAIS-III Digit Symbol: Coding subtest across the adult lifespan. Arch. Clin. Neuropsychol. 19, 759–767 (2004).

    PubMed  Article  Google Scholar 

  78. De Jong, P. F. Working memory deficits of reading disabled children. J. Exp. Child Psychol. 70, 75–96 (1998).

    PubMed  Article  Google Scholar 

  79. Savage, R., Lavers, N. & Pillay, V. Working memory and reading difficulties: What we know and what we don’t know about the relationship. Educ. Psychol. Rev. 19, 185–221 (2007).

    Article  Google Scholar 

  80. Beidas, H., Khateb, A. & Breznitz, Z. The cognitive profile of adult dyslexics and its relation to their reading abilities. Read. Writ. 26, 1487–1515 (2013).

    Article  Google Scholar 

  81. Moll, K., Kunze, S., Neuhoff, N., Bruder, J. & Schulte-Körne, G. Specific learning disorder: Prevalence and gender differences. PLoS ONE 9, e103537 (2014).

    ADS  PubMed  PubMed Central  Article  CAS  Google Scholar 

  82. Fostick, L. & Revah, H. Dyslexia as a multi-deficit disorder: Working memory and auditory temporal processing. Acta Psychol. (Amst) 183, 19–28 (2018).

    Article  Google Scholar 

  83. Hentschke, H. & Stüttgen, M. C. Computation of measures of effect size for neuroscience data sets. Eur. J. Neurosci. 34, 1887–1894 (2011).

    PubMed  Article  Google Scholar 

  84. Parker, A. J., Kirkby, J. A. & Slattery, T. J. Predictability effects during reading in the absence of parafoveal preview. J. Cogn. Psychol. 29, 902–911 (2017).

    Article  Google Scholar 

  85. Parker, A. J., Nikolova, M., Slattery, T. J., Liversedge, S. P. & Kirkby, J. A. Binocular coordination and return-sweep saccades among skilled adult readers. J. Vis. 19, 1–19 (2019).

    Article  Google Scholar 

  86. Parker, A. J., Slattery, T. J. & Kirkby, J. A. Return-sweep saccades during reading in adults and children. Vis. Res. 155, 35–43 (2019).

    PubMed  Article  Google Scholar 

  87. Fukushima, J., Tanaka, S., Williams, J. D. & Fukushima, K. Voluntary control of saccadic and smooth-pursuit eye movements in children with learning disorders. Brain Dev. 27, 579–588 (2005).

    PubMed  Article  Google Scholar 

  88. Tiadi, A., Gérard, C. L., Peyre, H., Bui-Quoc, E. & Bucci, M. P. Immaturity of visual fixations in dyslexic children. Front. Hum. Neurosci. 10, 1–7 (2016).

    Article  Google Scholar 

  89. Vagge, A., Cavanna, M., Traverso, C. E. & Lester, M. Evaluation of ocular movements in patients with dyslexia. Ann. Dyslexia 65, 24–32 (2015).

    PubMed  Article  Google Scholar 

  90. Freedman, E. G., Molholm, S., Gray, M. J., Belyusar, D. & Foxe, J. J. Saccade adaptation deficits in developmental dyslexia suggest disruption of cerebellar-dependent learning. J. Neurodev. Disord. 9, 1–8 (2017).

    Article  Google Scholar 

  91. Fischer, B., Biscaldi, M. & Otto, P. Saccadic eye movements of dyslexic adult subjects. Neuropsychologia 31, 887–906 (1993).

    CAS  PubMed  Article  Google Scholar 

  92. Biscaldi, M., Gezeck, S. & Stuhr, V. Poor saccadic control correlates with dyslexia. Neuropsychologia 36, 1189–1202 (1998).

    CAS  PubMed  Article  Google Scholar 

  93. Fukushima, J., Hatta, T. & Fukushima, K. Development of voluntary control of saccadic eye movements I. Age-related changes in normal children. Brain Dev. 22, 173–180 (2000).

    CAS  PubMed  Article  Google Scholar 

  94. von der Malsburg, T. & Vasishth, S. What is the scanpath signature of syntactic reanalysis?. J. Mem. Lang. 65, 109–127 (2011).

    Article  Google Scholar 

  95. Quercia, P., Feiss, L. & Michel, C. Developmental dyslexia and vision. Clin. Ophthalmol. 7, 869–881 (2013).

    PubMed  PubMed Central  Article  Google Scholar 

  96. Schotter, E. R. & Payne, B. R. Eye Movements and comprehension are important to reading. Trends Cogn. Sci. 23, 811–812 (2019).

    PubMed  Article  Google Scholar 

  97. De Luca, M., Borrelli, M., Judica, A., Spinelli, D. & Zoccolotti, P. Reading words and pseudowords: An eye movement study of developmental dyslexia. Brain Lang. 80, 617–626 (2002).

    PubMed  Article  Google Scholar 

  98. Reichle, E. D., Pollatsek, A. & Rayner, K. E-Z Reader: A cognitive-control, serial-attention model of eye-movement behavior during reading. Cogn. Syst. Res. 7, 4–22 (2006).

    Article  Google Scholar 

  99. Reichle, E. D., Warren, T. & McConnell, K. Using E-Z reader to model the effects of higher level language processing on eye movements during reading. Psychon. Bull. Rev. 16, 1–21 (2009).

    PubMed  PubMed Central  Article  Google Scholar 

  100. Reichle, E. D. et al. Using E-Z Reader to examine the concurrent development of eye-movement control and reading skill. Dev. Rev. 33, 110–149 (2013).

    PubMed  PubMed Central  Article  Google Scholar 

  101. Engbert, R., Nuthmann, A., Richter, E. M. & Kliegl, R. Swift: A dynamical model of saccade generation during reading. Psychol. Rev. 112, 777–813 (2005).

    PubMed  Article  Google Scholar 

  102. Mancheva, L. et al. An analysis of reading skill development using E-Z Reader. J. Cogn. Psychol. 27, 657–676 (2015).

    Article  Google Scholar 

  103. McConkie, G. W. & Rayner, K. The span of the effective stimulus during a fixation in reading. Percept. Psychophys. 17, 578–586 (1975).

    Article  Google Scholar 

  104. Rayner, K. & McConkie, G. W. What guides a reader’s eye movements?. Vis. Res. 16, 829–837 (1976).

    CAS  PubMed  Article  Google Scholar 

  105. Kliegl, R., Nuthmann, A. & Engbert, R. Tracking the mind during reading: The influence of past, present, and future words on fixation durations. J. Exp. Psychol. Gen. 135, 12–35 (2006).

    PubMed  Article  Google Scholar 

  106. Häikiö, T., Bertram, R., Hyönä, J. & Niemi, P. Development of the letter identity span in reading: Evidence from the eye movement moving window paradigm. J. Exp. Child Psychol. 102, 167–181 (2009).

    PubMed  Article  Google Scholar 

  107. Ashby, J., Yang, J., Evans, K. H. & Rayner, K. Eye movements and the perceptual span in silent and oral reading. Atten. Percept. Psychophys. 74, 634–640 (2012).

    PubMed  Article  Google Scholar 

  108. Rayner, K., Murphy, L. A., Henderson, J. M. & Pollatsek, A. Selective attentional dyslexia. Cogn. Neuropsychol. 6, 357–378 (1989).

    Article  Google Scholar 

  109. Bosse, M. L., Tainturier, M. J. & Valdois, S. Developmental dyslexia: The visual attention span deficit hypothesis. Cognition 104, 198–230 (2007).

    PubMed  Article  Google Scholar 

  110. Phillips, M. H. & Edelman, J. A. The dependence of visual scanning performance on saccade, fixation, and perceptual metrics. Vis. Res. 48, 926–936 (2008).

    PubMed  Article  Google Scholar 

  111. Silva, S. et al. Too little or too much? Parafoveal preview benefits and parafoveal load costs in dyslexic adults. Ann. Dyslexia 66, 187–201 (2016).

    PubMed  Article  Google Scholar 

  112. Stein, J. F. What is developmental dyslexia?. Brain Sci. 8, 26 (2018).

    PubMed Central  Article  Google Scholar 

  113. Smythe, I. & Everatt, J. Adult Dyslexia Checklist. (2001).

  114. Canadian Institutes of Health Research, Natural Sciences and Engineering Research Council of Canada & Social Sciences and Humanities Research Council of Canada. Tri-council Policy Statement: Ethical Conduct for Research Involving Humans. (2014).

  115. Morrice, E., Johnson, A. P., Marinier, J. A. & Wittich, W. Assessment of the Apple iPad as a low-vision reading aid. Eye 31, 865–871 (2017).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  116. Canivez, G. L. Test review of the Wechsler Adult Intelligence Scale—Fourth Edition. In The Eighteenth Mental Measurements Yearbook (eds Spies, A. et al.) (Buros Center for Testing, Lincoln, 2010).

    Google Scholar 

  117. Schraw, G. Test review of the Wechsler Adult Intelligence Scale – Fourth Edition. in The eighteenth mental measurements yearbook (eds. Spies, A., Carlson, J. & Geisinger, K.) (Buros Center for Testing, 2010).

  118. Allen, M., Poggiali, D., Whitaker, K., Marshall, T. R. & Kievit, R. Raincloud plots: a multi-platform tool for robust data visualization [version 1; peer review: 2 approved]. Wellcome Open Res. 4, 63 (2019).

    PubMed  PubMed Central  Article  Google Scholar 

  119. von der Malsburg, T. Scasim. (2020).

  120. Venables, W. N. & Ripley, B. D. Modern Applied Statistics with S (Springer, Berlin, 2002).

    MATH  Book  Google Scholar 

  121. Schwarz, G. Estimating the dimension of a model. Ann. Stat. 6, 461–464 (1978).

    MathSciNet  MATH  Article  Google Scholar 

  122. Scrucca, L., Fop, M., Murphy, T. B. & Raftery, A. E. mclust 5: clustering, classification and density estimation using Gaussian finite mixture models. R J. 8, 205–233 (2016).

    Article  Google Scholar 

  123. RStudio Team. RStudio: Integrated Development for R (RStudio, Inc., Boston, MA, 2020).

  124. Bates, D., Mächler, M., Bolker, B. M. & Walker, S. C. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67, 1–48 (2015).

    Article  Google Scholar 

  125. Barr, D. J., Levy, R., Scheepers, C. & Tily, H. J. Random effects structure for confirmatory hypothesis testing: Keep it maximal. J. Mem. Lang. 68, 255–278 (2013).

    Article  Google Scholar 

  126. Pernet, C. R., Wilcox, R. & Rousselet, G. A. Robust correlation analyses: False positive and power validation using a new open source matlab toolbox. Front. Psychol. 3, 606 (2013).

    PubMed  PubMed Central  Article  Google Scholar 

Download references


This work was supported by a Concordia University Seed fund and a Concordia University Horizon Postdoctoral Fellowship awarded to L.F.

Author information

Authors and Affiliations



Conceptualization: L.F. and A.P.J. Data Curation: L.F. and Z.S. Formal Analysis: L.F. and Z.S. Funding Acquisition: L.F. and A.P.J. Investigation: L.F., Z.S. and A.P.J. Methodology: L.F. and A.P.J. Project Administration: L.F. and A.P.J. Resources: L.F. and A.P.J. Software: L.F. and A.P.J. Supervision: A.P.J. Validation: L.F. and Z.S. Visualization: L.F. Writing—Original Draft Preparation: L.F. and Z.S. Writing—Review & Editing: L.F., Z.S. and A.P.J.

Corresponding author

Correspondence to Léon Franzen.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher's note

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

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Franzen, L., Stark, Z. & Johnson, A.P. Individuals with dyslexia use a different visual sampling strategy to read text. Sci Rep 11, 6449 (2021).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI:


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing