We Predicted That Typical Readers Would Show Significantly Higher Reading Accuracy

Loading metrics

Open Access

Peer-reviewed

Inquiry Article

Holistic processing of faces and words predicts reading accuracy and speed in dyslexic readers

Nuala Brady,
Kate Darmody,
Fiona Due north. Newell,
Sarah M. Cooney

Published: December 15, 2021
https://doi.org/10.1371/journal.pone.0259986

Figures

Abstruse

We compared the performance of dyslexic and typical readers on two perceptual tasks, the Vanderbilt Holistic Face Processing Task and the Holistic Word Processing Chore. Both yield a metric of holistic processing that captures the extent to which participants automatically attend to data that is spatially nearby merely irrelevant to the job at mitt. Our results show, for the commencement time, that holistic processing of faces is comparable in dyslexic and typical readers only that dyslexic readers bear witness greater holistic processing of words. Remarkably, we evidence that these metrics predict the operation of dyslexic readers on a standardized reading chore, with more holistic processing in both tasks associated with higher accurateness and speed. In contrast, a more holistic style on the words task predicts less authentic reading of both words and pseudowords for typical readers. We talk over how these findings may guide our conceptualization of the visual deficit in dyslexia.

Citation: Brady Northward, Darmody K, Newell FN, Cooney SM (2021) Holistic processing of faces and words predicts reading accuracy and speed in dyslexic readers. PLoS ONE xvi(12): e0259986. https://doi.org/10.1371/journal.pone.0259986

Editor: Randi Starrfelt, Kobenhavns Universitet, DENMARK

Received: April xvi, 2021; Accepted: Oct 30, 2021; Published: Dec 15, 2021

Copyright: © 2022 Brady et al. This is an open access article distributed under the terms of the Artistic Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The data are available at Figshare: 10.6084/m9.figshare.16943806.

Funding: The publication of this study was supported by University College Dublin (UCD) in the form of funds to NB. UCD makes funds available to academic staff contingent on their research output. The funders had no office in study design, information drove and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests be.

Introduction

Developmental dyslexia is characterised past difficulties in learning to read that are unexpected in light of a child'south cerebral abilities and educational opportunities [1]. These difficulties can persist into adulthood, and reading may be slower and new vocabulary challenging to higher students with dyslexia [2, 3]. While differences in visuo-temporal processing are considered integral to dyslexia past many [4, 5], research on dyslexia has focused predominantly on phonological processing, with reported impairments in phonological coding, rapid naming and exact short-term memory [half-dozen].

The view that visual problems in dyslexia are secondary to a 'core phonological deficit' [7] endures, in part, because it resonates with the dual-route theory of reading [8]. By this account, learning to read involves the acquisition of distinct phonological and orthographic skills. Phonological coding, crucial in early reading, establishes a mapping between messages and their associated sounds. Orthographic coding refers to the representation of the visual course of words–including groupings of letters that betoken spelling regularities–and, in fourth dimension, enables word recognition without the need to access phonological information at the pre-lexical level [ix]. This lexical route to give-and-take sounds is assumed to underlie fluency. It has been proposed that poor phonological processing in dyslexia may hinder the development of spelling-audio mappings, thus preventing children from learning precise orthographic information almost words, and ultimately from attaining fluency [10].

This accent on the primacy of phonological deficits in dyslexia has been challenged in recent years. For instance, while the business relationship has much appeal for languages with opaque orthographies such as English—where spelling-audio correspondences are particularly obtuse [eleven] - dyslexia is also common in languages with more than transparent orthographies. Using a cluster analysis of WISC-Iv data from over 300 Italian children with dyslexia [12] report ii distinct groups, both with harm in visual processing, but with but one group having additional impairment in phonological processing. The argument for a more straight role of visual damage in dyslexia, specifically in visual attending, is as well fabricated by Valdois and colleagues [13]. [14] show deficits in visual attention span in large samples of English and French dyslexic children, independent of damage in phonological processing. Such findings fence for a reconceptualization of dyslexia as a multifaceted disorder, one in which anomalous visual processing may occur independently of or in conjunction with poor phonological processing. However, the nature of the visual deficit in dyslexia is yet poorly understood. Here we investigate whether 'holistic processing', defined as obligatory attention to all parts of a stimulus is dissimilar in dyslexia.

Virtually pertinent to the enquiry we nowadays in this paper are a number of recent studies reporting subtle deficits in visual noesis in dyslexia which propose that dissonant visual processing is less specific to words than previously considered. These studies were inspired in function by reports of hypoactivation in left fusiform gyrus in both adults and children with dyslexia [15–17]. This region of the brain includes the visual word grade area (VWFA) which responds preferentially, just not exclusively, to printed words [18] and which is next to regions that respond preferentially to faces. Therefore, deficits in both word and face recognition, reflecting a general impairment in ventral stream processing, might occur in dyslexia [19, twenty].

Sigurdardottir et al. [20] investigated whether dyslexic and typical readers differ in their confront and object recognition abilities. Nineteen self-reported dyslexic and controls were tested on the Cambridge Face up Retentivity Examination (CFMT), the Vanderbilt Holistic Face Processing Exam (VHFPT), and the Vanderbilt Expertise Exam (VET). Dyslexics showed poorer retention for faces in the CFMT, beingness less accurate whether the task was performed with upright or inverted faces. As face recognition was comparably compromised across groups when the faces were inverted—a manipulation thought to induce a switch from holistic to function-based processing—this suggests that the poorer performance of dyslexics does non reflect a specific impairment to holistic processing of faces. Similarly, in the VHFPT dyslexics were less accurate overall than controls. Finally, as dyslexic readers were less accurate than controls on the VET simply not on a control colour recognition task, this also suggests that hypoactivation in left fusiform gyrus may result in subtle impairments in within-category object discrimination.

Employing a number of challenging perceptual tasks, [19] found that dyslexics were slower than typical readers in matching faces beyond unlike viewpoints, but that the groups were similarly hindered when matching between upright target faces and inverted exam faces, suggesting that holistic processing of faces is non specifically dumb. Similarly, dyslexics were less accurate in discriminating pairs of morphed images of faces but non in discriminating pairs of morphed images of cars. Finally, [21] asked participants to match images of 3D modelled faces and novel object. Briefly, in Exp 1 accuracy in the face up matching task predicted reading problems in a sample of academy students, although not distinguishing within groups of competent readers or within groups of poor readers. In Exp 2, operation in the novel objects matching job did not predict whether participants were dyslexic or typical readers, but operation in the face matching task did. The authors conclude that visual problems in developmental dyslexia are specific to loftier level tasks involving words and faces with which people have extensive experience or expertise.

In this newspaper we ask whether 'holistic processing', a grade of visual processing which is considered a hallmark of perceptual expertise by some [22, 23] is dissonant in dyslexia. Specifically, and for the offset time, we compare holistic processing of words and of faces in participants with dyslexia and age-matched controls and nosotros show that holistic processing of both faces and words predicts reading performance in the dyslexic merely not in the typical reader group.

Holistic or configural processing has been proposed to underlie both face [24] and discussion [25] recognition. In the example of faces, it is generally agreed that an authentic representation of second-society facial configuration—the precise geometric arrangement of features in the face—underlies expertise in recognition [26]. Although ofttimes used synonymously with 'configural processing', the term 'holistic processing' is frequently reserved to describe the automatic processing of facial features as a perceptual whole or gestalt which makes individuation of features difficult and it is in this sense that we utilize the term in this paper. This automatic processing of facial features as a perceptual whole is illustrated by the composite confront illusion whereby a single confront, fabricated by aligning images of the superlative and bottom half faces of dissimilar individuals, is perceived equally a unmarried facial identity [27, 28]. Even when directed to ignore one half of the composite, participants typically fail to selectively attend and some course of perceptual integration occurs. As expected, the blended face up result is considerably reduced when the two half faces are misaligned. The composite paradigm has recently been extended to the report of discussion recognition by [25] who show that expert readers are unable to ignore i role of a discussion when asked to nourish to the other part of that word in a matching task. This suggests that holistic processing is not specific to confront perception, but instead may occur as a result of repeated exposure or visual expertise with objects.

The current study explores whether 'holistic processing'–every bit measured for both faces and words using comparable tests of performance—is anomalous in adults with dyslexia. For faces, we use the VHFP Test [29], a modern variant of the face composite test that dispenses with the alignment condition and focuses exclusively on the principal effect of congruency of the aligned faces. For words we use Wong's Holistic Word Processing Chore [25] which is based directly on the original confront composite test and involves matching words under conditions which vary in congruency and alignment equally described below. As in [25] Study 1, we define the congruency outcome as the departure in operation betwixt congruent and incongruent trials in the aligned condition, which matches the metric of [29]. These two measures of holistic performance are then used as predictors of participants' scores on a standardized reading examination.

Methods

Participants

Of 62 participants who took part in the study, data from 3 were excluded; one's data were missing a very loftier proportion of trials (over 30%) and two had very high error rates coupled with very fast RT's or alternate yes/no responses suggesting that the participants did not engage seriously with the job. Analyses were conducted on the final sample of 59 adults, 30 students with a formal diagnosis of dyslexia (17 female person) and 29 students (19 female) who served every bit controls. Power analysis, using PANGEA [30] indicated that a sample size of 30 per group (Dyslexic/Typical Readers) would provide 98% ability to observe a medium effect size (d = 0.45) for a two-mode Group*Congruency ANOVA pattern. Participants were recruited from both University College Dublin and Trinity College Dublin, the students with dyslexia existence registered with inability support services at their university which requires a formal diagnosis of dyslexia to be provided by a clinical or educational psychologist. Of the 30 students with dyslexia, one completed the words task only and 1 completed the faces job only due to fourth dimension constraints.

The dyslexic and typical readers participants had a mean age of 25.0 years (SD = 8.1) and 25.86 years (SD = 11.0) and a t-test revealed no significant difference in age betwixt the groups, t(57) = 0.35, p = .73. All participants cocky-reported normal or corrected to normal vision. While all participants reported 'normal' or 'corrected to normal' visual acuity for the purpose of the written report, there were more reports of corrected vision and of other issues with vision among dyslexic participants every bit documented in Table 1. Using a binary classification of 'normal vision' and 'other', Pearson'south Chi-squared examination showed Ten ² = 6.13, df = one, p = 0.01.

The study was approved by the UCD and TCD Enquiry Ethics Committee; in accordance with the Proclamation of Helsinki all participants gave written, informed consent and were advised of their correct to withdraw from the study at any time without prejudice.

Materials and procedure

Reading tests.

All participants completed 2 subscales of the Wechsler Private Achievement Test (3rd Edition), the Give-and-take Reading and the Pseudoword Decoding tests. In the Discussion Reading exam, participants were asked to read aloud 74 words from a test sheet and the number of words read at xxx seconds was noted as a measure of reading speed. Words read fluently were awarded 1 point and words pronounced incorrectly were awarded 0 points. The examination was discontinued if the participant read iv consecutive words incorrectly and participants were given a further opportunity to read any incorrectly pronounced words at the finish of the session. The same procedure was followed for the Pseudoword Decoding test using a examination sheet of 52 pseudo-words. All participants completed the reading tests start, after which the order of the faces and words tests was randomised beyond participants.

Vanderbilt Holistic Face Processing Job (VHFPT).

We used the VHFPT 2.0 version of the Vanderbilt Holistic Face Processing Test described and tested in [29]. As reported by the authors, the VHFPT 2.0 shows superior psychometric properties relative to prior holistic face processing measures, with higher internal consistency (0.56) than the composite task and with exam–retest reliability of 0.49 (R = 0.94) subsequently a 6 month delay. It produces large average effect size for holistic processing (η2p = 0.75) and is normally distributed in an developed population [29]. The stimuli, with society counterbalanced so that half the participants completed the words tasks first and the other half completed the faces job first, were presented on a 22-inch colour monitor (1280 x 1024 resolution) using a Dell PC running Presentation® software. Viewing distance was ~50cm.

The test utilizes grayscale images of composite faces, made by combining images from two individuals' faces from a fix of 360 unfamiliar Caucasian faces. The iii-alternative forcefulness choice (3AFT) task involves looking at a target region of a study face, while ignoring the rest of the face up, and locating the matching identity in the same target region of one of three test faces, where ane is the correct exam face, and the two others are foils. There were nine target segment conditions: lesser two thirds (BTT); peak two thirds (TTT); bottom third (BT); meridian third (TT); bottom half (BH); superlative one-half (Th); eyes; oral cavity; nose. At that place were 20 trials (x congruent, x incongruent) per target segment and 60 trials (30 congruent, thirty incongruent) per face size (small, medium and large) as described past for a total of 180 trials. The target segment of the study face up and the target segment of the (right) test face were taken from two different images of the same person on both congruent and incongruent trials. On congruent trials the distractor segment of the right test face was also matched in identity to the distractor segment of the written report face. Still, on incongruent trials, the distractor segment of the (correct) exam face was non matched in identity to the distractor region of the study face. Specifically, the target region in both the report face and the (correct) test faces are from Person A. On the coinciding trial, the non-target region of both the study and (correct) test face up are from Person B. Withal, in the incongruent trial, while the target regions are matched in identity (Person A), the non-target regions of the study face and the (right) test face are from two different identities (Persons A and C). See S1 Fig for graphical details.

On each trial, a report face which was a composite image of two different face images was presented for 2000ms with the target region of the face delineated by a cherry box. Participants were instructed to only focus on the target region and to ignore the rest of the face. A blank screen followed for 1000ms. Three examination faces were then displayed, positioned horizontally, left, centre and right, until the participant fabricated a response to point which one had the matching target region. Each of the iii test faces were marked on the target region with a ruby box. Only one of the examination faces contained the right target segment identity (correct face) the two other test composites were wrong foils. Participants were required to signal which of the test faces contained the target segment of the study confront by pressing one of iii response keys on the keyboard, left image, centre image, correct paradigm, using keys F, M, & H. The experiment was preceded by three practise trials using composites created from Muppet faces that were presented in colour.

Discussion recognition task.

The stimuli were identical to those used past [25] and were given freely by the commencement author for use in this report. The stimuli consisted of four-alphabetic character words created from x sets (xl words in total). Each prepare was fabricated up of four words from which the left and right halves could be alternated, east.chiliad., as shown in S2 Fig left halves 'br' and 'sl' tin be combined with right halves 'im' and 'ow' to create four distinct words, 'brim', 'brow', 'slow' and 'slim'. Iv test conditions were created. On congruent trials the study and test stimuli were entirely the same or different. On incongruent trials one-half of the study and test stimuli were the same and half were dissimilar. Each word was presented as both test and report stimulus equally in each of the four conditions. Half of the trials were presented in aligned atmospheric condition and half were presented misaligned in which the non-cued one-half of the word was moved approximately 1.7° vertically.

On each trial a fixation was presented for 500ms, followed by a written report word for 400ms. This was replaced by a mask for 500ms, afterwards which a cue appeared to the left or to the right of the mask for a further 300ms to indicate the target half of the study word. The test word, also cued on the same side, followed for 1500ms later which the screen went blank until the participant responded. Participants were required to indicate if the cued half of the test stimulus was the aforementioned or different as the same half of the study stimulus past pressing either "same" or "different" keys on a Cedrus RB-844 response box. Post-obit [25] the study contained a total of 640 trials, with sixteen blocks of twoscore trials. Presentation of alignment weather condition was counterbalanced, so that one-half the participants completed the aligned condition first and the other half completed the misaligned chore starting time. All other conditions were randomized across participants. Participants completed 20 practice trials in accelerate of the experiment.

Results

Data were analysed in R [31]. Welsh's t-examination is used past default for between-grouping comparisons and corrected degrees of freedom reported [32]. Outcome sizes (Cohen's d) are interpreted as originally suggested with d = 0.2, 0.5, 0.8 as small, medium and large consequence sizes. For ANOVA, Greenhouse-Geisser corrections are used when Mauchly's Test for Sphericity was significant and effect sizes are given by partial eta squared (η ⁱⁱ _p). We follow a conservative arroyo to removal of RT 'outliers' using exploratory data analyses (box- and-whisker and qq-normal plots) to note RTs which are plainly as well fast (anticipatory errors) or much likewise deadening (so there is a noticeable break in the upper extremes of the data) suggesting that the participant was not attending properly on the trial. Other methods, such as removing whatsoever RT above a fixed number of SDs above the mean for each person can be very problematic (despite their regular utilize as heuristics), as RT data are by and large asymmetric with long right-tailed skew [33–35] and so that distinguishing outliers from 18-carat high RTs is problematic.

Reading tests

Fig 1 plots pseudo-give-and-take accurateness confronting word accuracy and pseudo-discussion reading speed against give-and-take reading speed for both dyslexic and control participants.

Fig 1. Pseudo-word accuracy and pseudo-word speed is plotted confronting word accurateness and discussion speed for both dyslexic and control participants.

The shaded areas bear witness the standard error bounds. Accuracy is reported as the number (out of 74) read accurately, Speed is the number read accurately within 30secs so that lower speed corresponds to poorer operation.

https://doi.org/ten.1371/journal.pone.0259986.g001

Accuracy scores for pseudo-words and words were highly correlated for both dyslexic, r = 0.74, df = 28, p < 0.0001, 95% CI [0.51, 0.87] and command, r = 0.67, df = 27, p < 0.0001, 95% CI [0.40, 0.83], groups, and for the combined groups, r = 0.82, df = 57, p < 0.0001, 95% CI [0.71, 0.89]. Similarly, speed scores for pseudo-words and words were highly correlated, for both dyslexic, r = 0.77, df = 28, p < 0.0001, 95% CI [0.57, 0.88], and control, r = 0.78, df = 27, p < 0.0001, 95% CI [0.58, 0.89], groups and for combined groups, r = 0.seventy, df = 57, p < 0.0001, 95% CI [0.54, 0.81]. While accuracy clearly discriminates the dyslexic and control groups for both words and pseudowords as shown in the Fig 1, the groups perform comparably with respect to discussion speed but not with respect to pseudoword speed where dyslexic students are slower (see also Table 2). Paired sample t-tests showed significant differences between groups in both discussion accuracy, t(34.87) = 5.89, p < 0.0001, d = 1.51, and pseudo-word accurateness, t(44.57) = 5.71, p < 0.0001, d = 1.47. The difference between groups in word reading speed was non significant, t(54.06) = -0.36, p = 0.71, d = -0.09, whereas for pseudo-word reading speed the control grouping were faster, t(54.96) = 3.05, p = 0.004, d = 0.79. Summary statistics are provided in Table two.

Faces examination

Accuracy & response time.

The overall error rate was comparable for control (36.seven%) and dyslexic (37.8%) groups and in keeping with the high rates reported by [29] who explain that the job is purposively challenging. Exploratory data analyses highlighted a minor number of RTs less than 400ms or greater than 20000ms (less than 0.05% of all trials) that were removed equally outliers. Fig two plots accurateness past group and congruency which suggests an issue of congruency only. In contrast, although this was not an explicit reaction timed task, the plot of RT on right trials suggests an upshot of congruency and group.

Fig two.

RT (left) and accurateness (right) are plotted by congruency and past grouping in the faces task. Error bars testify 95%CI about the mean. Accuracy is expressed as a proportion (0.0 to 1.0) and chance performance in this task is 0.33.

https://doi.org/10.1371/periodical.pone.0259986.g002

For accurateness, mixed-furnishings ANOVA with a within-subjects factor of Congruency and a between-subjects gene of Group yielded master effects of Congruency, F _{(i, 56)} = 231.38, p < .0001, = .81, where accuracy is college for congruent [Thousand = 71.4%, SD = 8.4%] than incongruent [M = 54.0%, SD = 6.6%] trials. Neither the primary event of group, F _{(1, 56)} = 0.45, p = .fifty, = .008, nor the Congruency*Grouping interaction, F _{(one, 56)} = 0.33, p = .57, = .006, were pregnant. For RT, the main effects of Group, F _{(1, 56)} = 3.83, p = .055, = .06, and of Congruency, F _{(1, 56)} = 104.35, p < .0001, = .65, are of notation while the Congruency*Group interaction, F _{(one, 56)} = 0.67, p = .41, = .01, was non meaning. With respect to group, controls were faster, 1000 = 2456ms, 95% CI [2218, 2698], than dyslexics, Grand = 2892ms, 95% CI [2669, 3116]. At the request of a reviewer, this analysis was repeated using median RT and showed a pregnant principal effects of Group, F _{(i, 56)} = 4.61, p = .036, = .08, and of Congruency, F _{(ane, 56)} = 66.72, p < .0001, = .54, while the Congruency*Group interaction, F _{(1, 56)} = one.34, p = .25, = .02, was not significant.

Words test

Response time & sensitivity.

Exploratory data analyses highlighted a minor number of RTs less than 200ms or greater than 8000ms (less than 0.04% of all trials) that were removed as outliers. Overall, errors were made on 5.34% of trials, 2.83% for dyslexic and 2.51% for typical readers. Fig 3 plots RT on right trials past group and by conditions. Dyslexic participants are slower than controls overall, and both groups are slower in the incongruent than in the congruent condition.

Fig 3. RT for right trials is plotted in separate panels for command and dyslexic participants, with condition of alignment indicated on the ten-axes and status of congruency indicated by line type.

Error confined 95%CI almost the hateful.

https://doi.org/10.1371/journal.pone.0259986.g003

Mixed effects ANOVA showed a significant primary effect of Group, F _{(ane, 56)} = 10.91, p = .002, = .16, with dyslexics, M = 832.8ms [SD = 495.5ms], slower than controls, Chiliad = 650.4ms [SD = 278.7ms]. In that location was a significant main effect of Congruency, F _{(one, 56)} = 55.22, p < .0001, = .50, with slower performance on incongruent, M = 771.9ms [SD = 378.9ms], than on coinciding trials, Yard = 711.5ms [SD = 378.9ms]. And, as expected from [25] there was a significant Congruency * Alignment interaction, F _(i,56) = xx.78, p < .0001, = .27, the issue of congruency beingness greater in aligned than misaligned trials. Additionally, the Grouping * Congruency interaction, F _(1,56) = 9.76, p = .003, = .xv, was pregnant and is explored further below in the Congruency Effect section. The three way Group*Alignment*Congruency interaction was not significant at conventional levels, F _(1,56) = 1.69, p = .197, = .03. Using median RTs, as in a higher place for the faces job, the analysis showed a significant main effect of Group, F _{(1, 56)} = 11.53, p = .001, = .17, and of Congruency, F _{(one, 56)} = 38.25, p < .0001, = .41. There was a meaning Congruency * Alignment interaction, F _(1,56) = 4.12, p = .047, = .07, and a meaning Grouping * Congruency interaction, F _(one,56) = 9.01, p = .004, = .xiv. The 3 fashion Group*Alignment*Congruency reached significance at conventional levels, F _(1,56) = 7.xix, p = .02, = .11. For dyslexics, the Alignment*Congruency result was pregnant, F _(1,28) = 21.17, p < .0001, = .43, with a more marked difference in RT betwixt congruent and incongruent trials in the aligned that in the misaligned condition. And for typical readers, the Alignment*Congruency effect was significant, F _(one,28) = 8.35, p = .007, = .23, with a more marked divergence in RT between congruent and incongruent trials in the aligned that in the misaligned condition.

D-prime number (d'), a measure of sensitivity that is independent of response bias [36] was calculated. Mixed furnishings ANOVA revealed no significant main consequence of Group, F _{(1, 56)} = 0.34, p = .561, = .006, a significant main upshot of Congruence, F _{(1, 56)} = 74.26, p < .0001, = .57, with higher sensitivity on coinciding than incongruent trials. The Congruence * Grouping interaction was significant, F _(1,56) = 4.87, p = .03, = .08, equally was the Congruence * Alignment interaction, F _(1,56) = thirty.05, p < .0001, = .35. Regarding the quondam, planned comparisons prove an effect of Congruency for both dyslexic, F _(ane,28) = 54.63, p < .0001, = .66, and controls, F _(1,28) = 22.15, p < .0001, = .44, that is more marked for dyslexic participants. Regarding the latter, planned comparisons show an effect of Congruency in aligned (p < .0001) and misaligned (p = .0003) conditions, that is more marked in the aligned condition.

Congruency consequence

Faces & words.

The congruency effect in the faces chore is defined as the deviation in accuracy on incongruent and congruent trials, and serves as a metric of 'holistic processing', operationalized in terms of obligatory attending to all parts of the face [29]. Participants who tin can attend solely to the highlighted region of the face should perform with comparable accuracy on congruent and incongruent trials and will have a low congruency effect. In contrast, those with a more than holistic way of processing will exist more easily distracted by information from the irrelevant or 'to exist ignored' face region and and then be less accurate on incongruent than on congruent trials leading to a college congruency effect. By similar logic, the congruency effect in the words task is defined as the difference in RT between incongruent and congruent trials on aligned trials and serves as an index of how much the irrelevant data interferes with observers' judgments [25]. Fig 4 plots the congruency event in the faces (left) and in the words (right) job for both dyslexic and typical readers, with. In both tasks, both groups show bear witness of holistic processing, of comparable magnitude in the faces tasks only with dyslexic participants showing a stronger effect than controls in the words task. One participant from the dyslexic grouping was removed as their congruency effect on the words task was over four standard deviations from the group mean in the positive direction, i.eastward., they showed an extremely loftier congruency effect. They are non represented in Fig iv nor in the analyses below.

Fig 4. Congruency effect on the faces (left) and words on aligned trials (correct) task for both participant groups.

The congruency issue for faces is defined with respect to accuracy and the congruency upshot for words with respect to RT, meet text for details. The violin plots include individual subject points and show 95% CI nigh the mean.

https://doi.org/10.1371/journal.pone.0259986.g004

In the faces task, ANOVA showed no outcome of Grouping, F _{(i, 56)} = .34, p = .56, = .006, the congruency effect being comparable for dyslexic [M = .18, SD = .09] and control [G = .17, SD = .09] participants. Nevertheless, in the give-and-take task, ANOVA in that location was a significant event of Grouping, F _{(1, 55)} = 11.90, p = .001, = .eighteen, the congruency effect being greater for dyslexic [M = 107.0, SD = 77.3] than controls [M = 51.four, SD = 37.5] participants. Although not pertinent to the analysis below, where the focus is on the congruency consequence on aligned trials, nosotros note that a further mixed effects ANOVA shows a significant effect of both Group, F _(1,55) = 10.38, p = .002, = .17, and of Alignment, F _(1,55) = 19.52, p < .0001, = .26. The Group* Alignment interaction was non significant (p = .28). This shows that both dyslexic and control participants were susceptible to interference from the unattended part of the stimuli, and more so when the two halves of words were properly aligned.

Congruency effect equally predictor of reading scores

Fig 5(A) and 5(B) plots the congruency event past each of the four WIAT reading score metrics in the faces and words tasks respectively, and statistics are reported in Table three. Considering first the faces chore, for dyslexic participants greater holistic processing in the faces task is associated with ameliorate reading scores in both word and pseudoword accurateness and in word and pseudoword speed. This is not the case for the typical readers, where holistic processing in the faces job shows no obvious clan with any of the reading metrics. Turning to the words task, greater holistic processing is associated with better reading scores in all four metrics for the dyslexic groups, whereas for the typical reader group greater holistic processing in the words task is associated with poorer performance in discussion and pseudoword accuracy but unrelated to speed. Respectful of encouragement to move away from the null-hypothesis significance testing framework [37] nosotros plot estimated regression coefficients (slopes) with their 95% confident intervals using dot-and-whisker plots [38] in Fig 6. A clear blueprint is evident whereby a higher congruency outcome for dyslexic readers is predictive of ameliorate reading scores in all four metrics (give-and-take accurateness, pseudoword accuracy, word speed and pseudoword speed) and this is the example for both the faces and the words task. In contrast, for the typical readers, the congruency result in the faces chore is not predictive of reading scores while a higher congruency event in the words chore is predictive of lower word and pseudoword reading. Therefore, automated and obligatory attention to all parts of a stimulus, as measured in the faces and words tasks, conspicuously relates to reading strategy every bit discussed below.

Fig 5.

Besprinkle plots of the congruency upshot in the faces job (A) and the words job (B) past each of the four WIAT reading score metrics with dissever plots for dyslexic (magenta) and typical (bluish) readers. Accuracy is reported every bit the number (out of 74) read accurately, Speed is the number read accurately within 30secs then that lower speed corresponds to poorer performance.

https://doi.org/x.1371/journal.pone.0259986.g005

Fig half dozen.

Dot-and-whisker plots showing the gradient coefficients (with 95% CIs) from linear models regressing reading score metrics on the congruency effect for faces (left) and for words (right). Symbols key: Give-and-take Accuracy (square), Pseudoword Accuracy (circumvolve), Discussion Speed (triangle), Pseudoword Speed (asterisk).

https://doi.org/10.1371/periodical.pone.0259986.g006

Further exploration using multiple regression on the dyslexia data only (performance on the faces task non being predictive of any reading metric for the typical readers) is summarized in Table 4. In the case of Give-and-take and Pseudoword Speed, the ii different Congruency Effects (in the words task and in the faces job) acts as independently predictors. However for the Word and Pseudoword Accuracy score, the improver of the Congruency Effect in Words Task adds only marginal extra prediction (p<0.x).

Word

We compared the performance of college students with dyslexia and historic period matched typical readers on two perceptual tasks, the Vanderbilt Holistic Face up Processing Job (VHFPT) and the Holistic Word Processing Task (HWPT), that each yield a measure of holistic processing known as the 'congruency effect'. This metric captures the extent to which participants automatically nourish to information that is spatially nearby but irrelevant to the chore at hand. In both the VHFPT and the HWPT, the extraneous or irrelevant data may benefit performance when information technology is congruent with the data that participants are asked to attend to, or may disadvantage functioning when it is incongruent. The congruency upshot is calculated as a difference score for performance on 'coinciding' and 'incongruent' trials and serves as an index of holistic processing. Our results show, for the first time, that holistic processing of faces is comparable in dyslexic and typical readers but that dyslexic readers evidence greater holistic processing of words, at least for the specific tasks at hand. Furthermore, nosotros evidence that these measures of holistic processing predict performance on a standardized reading task, the WIAT-3, with a more holistic fashion in both the faces and words task associated with improve reading scores—specifically, more accurate and faster reading of both words and pseudowords—for dyslexic readers. In contrast, a more holistic style on the words task predicts less accurate reading of both words and pseudowords for typical readers.

Below we discuss how these findings compare to recent research on anomalous visual processing in developmental dyslexia and to a rapidly evolving literature on the office of visual attention in dyslexia. Finally, nosotros consider how our finding of enhanced holistic processing in dyslexic readers–where holistic processing is defined in the strict sense of automatic attention to the whole stimuli–may guide our conceptualization of the visual deficit in dyslexia.

Starting with the faces chore, many aspects of our findings (specifically, with the command participants) replicate straight those of [29] while also revealing interesting similarities between the dyslexic and command groups. Firstly, error rates are comparably loftier (~35%) to previous reports and are equal across dyslexic and typical readers. 2d, accuracy is considerably higher on congruent than on incongruent trials equally expected from [29] and this was the case for both dyslexic and typical readers. Although this was non an explicitly timed chore, the response time data show that dyslexic participants are slower than typical readers to correctly friction match the target regions beyond the written report confront and the exam faces. Nevertheless this grouping divergence was not modulated past congruency, with both dyslexic and typical readers showing comparable reward on congruent trials.

This finding is consistent with contempo reports of a general impairment in ventral stream processing in dyslexia that may atomic number 82 to subtle differences in face up processing but not to specific impairments in holistic processing [19, 20]. [twenty] establish that typical readers showed an advantage over dyslexia readers on the Cambridge Face up Retention Test, just that both groups were comparably impaired past stimulus inversion suggesting that at that place is no specific harm in holistic processing in dyslexia. These authors too used the VHFPT and report that typical readers performed with higher accuracy (61.iii%) than dyslexic readers (59.vii%)–a result reported as 'marginally significant'—simply as this advantage was not specific to the coinciding condition, they argue that this cannot be attributed to poorer holistic processing in dyslexic readers [20]. While noting that dyslexic and typical readers performed with comparable accuracy in our study, differently than in [20], our findings that the congruency outcome is comparable betwixt the groups strengthens previous conclusions that holistic processing of faces is not impaired in dyslexia. Inquiry by [xix] has been similarly motivated past the question of whether dissonant visual processing in dyslexia is specific to words or extends to other classes of visual objects. They report slower response times by dyslexic readers compared to typical readers in matching faces merely not in matching cars, suggesting that visual impairments in dyslexia extend beyond words. Withal, as inverting the stimuli led to comparably slower performance in both groups there is no proposition of a specific impairment in holistic processing.

Turning to the words job nosotros notation that aspects of our findings (specifically with the control group) map directly onto those of [25]. Participants are slower in the incongruent than in the congruent condition and this 'congruency result' is greater for aligned than misaligned trials equally reported by in their Written report 1 [25]. With regard to group differences we find that, while dyslexic participants are slower than controls overall in that location is no departure in sensitivity between the groups. This is consequent with findings from our previous research [39] which reports that dyslexic participants are slower to respond than typical readers only testify comparable sensitivity in a novel non-reading job that encompasses aspects of the 'discussion superiority' and 'word inversion' paradigms.

In the electric current report nosotros notice that dyslexic participants prove a stronger 'congruency effect' than controls on the words task. Specifically, while both dyslexic and control participants were susceptible to interference from the unattended part of the stimuli, and more so when the two halves of words were properly aligned, dyslexics were more than susceptible to this interference than controls. While it is hard to directly compare with the findings of [25] is notable that in their Study 2—which compared the performance of native English speakers with those for whom English is their second linguistic communication–the native English speakers showed a more marked 'congruency effect'. This suggests that readers with more experience use more than holistic processing than those with less experience. While all participants in the current report were college students—and reading is an integral office of college life–information technology would be difficult to argue that dyslexic students are the more expert readers. Interestingly, a recent paper past [40] shows that adults with dyslexia recognize Chinese characters with stronger holistic processing than controls.

Similarly, it is besides difficult to directly compare the findings of the current study to those of [39] who utilized a very dissimilar task to compare the use of holistic processing between dyslexic and typical readers. As in the face perception tasks utilized by [20] and [nineteen] stimulus inversion was used by [39] as a way to explore holistic processing of words. Specifically, participants were asked whether pairs of words (which were identical or varied by i alphabetic character, and which were intact or jumbled) were the aforementioned or different and word pairs were presented in both upright and inverted orientation. [39] show a more marked inversion outcome for command than for dyslexic participants. Specifically, for short 4-letter words, response times to discriminate inverted stimuli was comparable across the two groups whereas for upright stimuli dyslexics were markedly slower than the typical readers suggesting that they benefit less from holistic cues. Although both groups showed clear evidence of holistic processing in that written report, typical readers showed more marked holistic processing than dyslexic readers. In contrast, in the electric current written report the dyslexic participants bear witness a definite congruency event that is an accepted marker of holistic processing and a more marked effect than their peers in the typical reader group. It may exist that typical readers have more flexibility in how they perform word processing tasks and tin can switch more easily between holistic and analytic processing as required.

A key finding of this inquiry is that the congruency result, as measured in both the faces task and in the words task, is predictive of dyslexic participants' reading scores with more holistic processing in both tasks associated with higher accuracy in reading words and pseudowords and in faster reading of words and pseudowords. This is evident in Fig half-dozen where we plot estimated regression coefficients (slopes) with their 95% confident intervals. Across both tasks, the obligatory attention to extraneous information captured by the congruency effect is predictive of better—faster, more than accurate—reading in dyslexic readers. In contrast, holistic processing in the faces tasks is not predictive of reading performance in the control group, and holistic processing in the words task is merely predictive of reading accurateness and that association runs counter to the pattern seen for the dyslexic readers. For the typical readers more holistic processing in the words task is associated with less accurate word and pseudoword reading.

Interestingly, a recent report [41] that a higher congruency effect for words is associated with more efficient performance of typical readers in a lexical decision task. In that study, participants were asked to indicate as apace equally possible whether a presented alphabetic character string was a real word or not, the stimuli consisting of iv, five and 6-letter words (and associated pseudowords) of both depression and high frequency. The negative correlation betwixt the magnitude of the discussion-frequency effect (where a smaller give-and-take frequency effect is associated with more than efficient word processing) and the word congruency issue ways that the more than efficient readers were less able to selectively ignore extraneous data in the word composite task. This finding appears to run counter to the present finding that a college congruency effect on the word task is associated with less accurate reading of words and pseudowords. Plain, the two tasks–reading words and pseudowords aloud in the current study and making a speeded lexical decision task in- are quite unlike, the reading aloud task requiring sub-lexical phonological processes. Information technology would be interesting to see whether the electric current findings on the relationship between the congruency effect for faces/words and performance on the reading tasks reported for dyslexic readers holds also for lexical decision tasks.

In response to a reviewer's comments we note that the tasks we used to measure holistic processing of words and faces differ in a number of means. For instance, the VHFPT utilises a 3AFC task in which participants are asked to attend to one specific target region of a study face and to subsequently locate the matching identity in the aforementioned target region from 3 examination faces. In dissimilarity, the HWPT or give-and-take composite job uses a aforementioned-dissimilar epitome in which a report word is presented and participants are subsequently cued to which side of the examination discussion they should attend to in deciding whether it is the same or different than the respective half of the study give-and-take. However, common to both tasks is the requirement that participants selectively attend to 1 office of a complex stimulus while ignoring a spatially next office of the stimulus, and the congruency effect is a measure of their power or inability to ignore this extraneous information. We stress that we are non claiming that 'holistic processing' (equally a perceptual mode) is the same for faces and for words, but but that the congruency effects serve as a measure of the involuntary tendency to integrate data across the stimulus in these different tasks.

Before considering the implications of these findings, we also draw attention to the fact that, while all participants in the current study reported normal or corrected-to-normal vision, it is notable that those with dyslexia (22/xxx) evidence a greater incidence of refractive errors and other issues with vision than those without dyslexia (11/29). Myopia or short-sightedness is marked in the dyslexic (17/30) compared to the typical reader (five/29) sample. Reduced visual acuity has been previously reported every bit beingness significantly associated with dyslexia [42] simply others study no association between refractive error and dyslexia [43]. It is likewise possible that the typical readers in this study had unusually low rates of myopia, as national statistics show prevalence rates of ~nineteen% in children aged 12–thirteen years which would be expected to be higher in college aged young adults [44].

Past way of general conclusions, our results bring together others in showing subtle impairments in high level visual processing, including in memory for faces, perceptual matching of faces, within-category discrimination of other objects [20, 21] and in recognition and matching of words and faces [19]. Collectively, these findings suggest that visual deficits underlying dyslexia are more than 'domain general' than 'domain specific' in that they touch on the recognition of objects other than words [xix]. Interestingly, similar findings have been reported in cases of alexia, an acquired damage in reading following brain injury and historically also referred to 'letter-past-alphabetic character reading', 'word incomprehension', 'word course dyslexia' and 'acquired dyslexia' [45]. For instance, [46] describe iv patients with pure alexia, arising for unilateral harm to left inferior occipitotemporal lobe, who bear witness poorer performance on a confront matching job than controls. Similar to the results now emerging in research on developmental dyslexia, these impairments in face processing in alexic patients are described as 'mild'. Interestingly, brain imaging inquiry points to a common dysfunction in left occipitotemporal cortex (the visual word form area) in both acquired and developmental dyslexia [47, 48].

A second conclusion is that the visual deficit in dyslexia has a strong attentional component, and nosotros base this ascertainment on our findings that holistic processing of both words and faces strongly predicts word and pseudo give-and-take accuracy and speed in dyslexic readers. In contrast, holistic processing of faces is unrelated to reading scores in typical readers, and where holistic processing of words is related to reading accuracy, the predictions run counter to those for the dyslexic group. While 'holistic processing' is an elusive concept in both definition and measurement [49], the tasks nosotros use in this report operationalize holistic processing in terms of selective attention. Variously described every bit measuring obligatory attending to all parts of an object or, analogously, as a failure of selective attention to parts of an object [25, fifty] this grade of perceptual processing is traditionally associated with expertise, see [51, 52] for debate.

In a comprehensive review of accounts of dyslexia, [thirteen] notation the heterogeneity of the dyslexic population and present evidence that anomalous attentional processing may be the core deficit in a subset of dyslexic children. Since then, the independence of deficits in phonological processing and in visual attending disorders as contributing factors to dyslexia has been demonstrated in both French and English speaking samples [14]. These studies, notable for their use of larger sample sizes that are necessary to explore heterogeneity in the disorder, join others emphasising the role of visual factors in dyslexia. For example, using cluster analysis with a sample of 316 Italian children [12] bear witness distinct groupings, both of whom bear witness impairment in visual tasks just only one of whom shows phonological impairment. The authors conclude that visual impairment is central to dyslexia which cannot exist explained with reference to a main phonological damage. And specific to adult readers, [53] study that college students with dyslexia show poorer performance than their peers in tasks involving visual discrimination of novel grid-like patterns and in visuospatial working memory tasks which are known to crave attentional command.

How might these findings inform the interpretation of the results from our current study? The use of visuo-spatial tasks with reports of dissonant attentional factors provide a common theme to these diverse studies. In the current inquiry, both the VHFP and the HWPT may exist conceptualized as tasks of selective attention with dyslexic participants showing a comparable tendency toward holistic processing in the faces task and a greater tendency toward holistic processing in the words task. Furthermore, while the congruency outcome on the faces task and on the words task are both predictive of reading scores for the dyslexic grouping with a more holistic fashion associated with improved reading, the association betwixt holistic processing and reading operation for the control grouping is only seen in the case of the words tasks where a more holistic style is associated with poorer accuracy in word and pseudoword reading. Information technology is important to consider these very unlike patterns of association for the dyslexic and typical reader groups in lite of their very different operation on the WIAT reading tests (Fig 1, Table 1); these groups differ essentially in their reading performance with typical readers attaining significantly higher levels of accuracy on average.

Reading involves the analysis of visual word forms at different spatial scales, including noting letter combinations at both fibroid and fine scales that indicate spelling regularities [ix] and this combined use of global and analytic processing is central to attention-focused models of reading [thirteen]. Particularly in languages with opaque orthographies such as English language, we suggest that efficiency or fluency in reading may be associated with the ability to switch strategy as needed, rather than with an exclusively holistic strategy. This consideration is likely relevant to agreement the broader question of what underlies reports of 'mild' impairments in non-reading tasks in dyslexia that hint to differences in ventral stream processing underlying 'perception expertise', e.g., [19, twenty]. Equally noted by [51] might more usefully be conceptualised in terms of attentional as well as purely perceptual factors.

Conclusions

Nosotros replicate recent findings that dyslexic readers show mild impairment in visual, non-reading tasks including in a confront perception and a word perception task that both yield a metric of holistic processing. Further we show that this metric, the 'congruency outcome', predicts reading performance in dyslexic readers with a more holistic style associated with better accurateness and speed scores. In contrast, a more than holistic manner on the words task is associated with poorer give-and-take and pseudoword accuracy scores in typical readers. This suggest that selective attention plays a different part in the reading strategies of dyslexic and typical readers.

Supporting information

S1 Fig. The upper panel depicts each of the 9 possible target regions in the red segments.

The lower panel contains an example of a congruent (upper row) and an incongruent trial (lower row) using the top half target region which is highlighted in carmine. On the congruent trial, the target segment in the study image and in the (correct response) test prototype are two unlike images of the aforementioned person (Person A) and the not-target region (the bottom half of the confront) are also images of the aforementioned person (Person B). On the incongruent trial, the target segment in the study prototype and in the (correct response) test epitome are two unlike images of the same person (Person A) only the non-target regions are images of different people. Target segments are outlined in colour for illustration purposes only and were not used in the actual experiment.

https://doi.org/10.1371/journal.pone.0259986.s001

(TIF)

S2 Fig.

(A) Examples of four-alphabetic character study and examination words in congruent and incongruent conditions where the first two letters of the report and test are the same or different. (B) The temporal sequence of the stimuli presented. Based on Wong et al. (2011).

https://doi.org/10.1371/journal.pone.0259986.s002