Improving sentence reading performance in Chinese children with developmental dyslexia by training based on visual attention span

Deficits in the visual attention span (VAS) are thought to hamper reading performance in dyslexic individuals. However, the causal relationship between VAS deficits and reading disability remains unclear. The present study attempts to address this issue by using a VAS-based intervention to explore the possible influence of VAS on reading processes in Chinese children with dyslexia. Given the influence of the heterogeneity of dyslexia on intervention effects, VAS-impaired dyslexic and VAS-intact dyslexic individuals were separately trained. Therefore, there were five groups of participants in this study, including 10 trained dyslexic individuals with VAS deficits and 10 untrained dyslexic individuals with VAS dysfunction as the baseline reference, 10 trained and 10 untrained dyslexic individuals with an intact VAS, and fourteen age-matched normal readers for reference of normal level. All participants completed reading measures and a visual 1-back task, reflecting VAS capacity with non-verbal stimuli and non-verbal responses, before and after VAS-based training. VAS-based training tasks included a length estimation task regarding the bottom-up attention, visual search and digit cancelling tasks targeting top-down attentional modulation, and visual tracking tasks to train eye-movement control. The results showed that visual training only helped improve VAS skills in VAS-impaired dyslexic individuals receiving training. Meanwhile, their silent sentence reading accuracy improved after training, and there was a significant relationship between training improvements in VAS function and reading performance. The current findings suggest that VAS-based training has a far-transfer effect on linguistic level (i.e., fluent reading). These findings suggest the possibility that VAS-related training may help children with dyslexia improve their reading skills.


Supplementary materials for results in the training procedure Analyses about the learning effects during the VAS-related intervention
The training tasks in the present study involved two types of attentional capacities underlying VAS, that is, bottom-up stimulus-driven attention abilities and top-down controlled attention. In details, a VAS estimation task was applied to train the visual short-term memory storage regarding the bottom-up attention, and visual search and digit cancelling tasks were used to target the top-down attentional modulation and control. In addition, given that eye-movements during VAS tasks might exert an influence on the final performance, the present intervention also included visual tracking tasks to train eye-movement control.
Repeated ANOVAs on training sessions (10 sessions) were separately submitted to the performance in tasks of visual search, digit cancelling, visual attention span, line/maze puzzle tasks within the two groups of training dyslexics. The results of onesample Kolmogorov-Smirnov tests exhibited no significant effects (ps>.1), suggesting normal distribution for all the dependent variables. The results of Levene's test for the homogeneity of variance showed that the variances were equal for all the measures (ps>.1). Post hoc analyses were conducted in which the relevant p-values were corrected by Bonferroni adjustment.

Visual attention span estimation
For the training dyslexics with VAS deficit, the training effect was significant [F(9, 63)=2.41, p=.02, η 2 =.19], and post hoc analyses showed that the visual attention span in the 5 th session was larger than that in the other sessions except for the 10 th session (ps<.05), and VAS in the 10 th session was larger than that in the other sessions except for the 5 th and 6 th sessions (ps<0.05), and there were no any other significant differences (ps>0.05).

Visual search
The training effect of the dyslexic children with impaired VAS was significant [F(9, 63)=2.40, p=.02, η 2 =.23]. Post hoc analyses showed that the accuracy in Session 5 to 10 was higher than that in Session 1 to 3 (ps<.05); meanwhile accuracy in Session 9 was higher than that in Session 4 (p=.03).
Moreover, it has been suggested that VAS skills were related to visual search loading more visual attention rather than to the visual search in a pop-out condition (Lallier et al., 2013). Given that the training task of visual search in the present study adopted three types of target ovals, which had the same width as the circle and included three heights (i.e. 1/4, 1/2, and 3/4 of the circle's diameter), and identifying the target oval with 3/4 height of the circle's diameter from the background circles required great involvement of visual attention due to the high similarity between this type of target oval and background circle. Therefore, we further explored the learning effect in this condition within the two trained groups. The VAS-impaired dyslexics showed significant differences across sessions [F(9, 63)=2.98, p=.009, η 2 =.43], in which the accuracy in Session 5 to 10 was higher than that in Session 1 to 4 (ps<.05) and mean accuracy in Session 1 was lower than that in Session 2 to 4 (ps<.05). As to the VASintact dyslexics, there was not a significant effect of session [F(9, 63)=.91, p=.52, η 2 =.12].

Potential improvement during the VAS-related intervention and its relation to the training effects of VAS skills for the VAS-impaired DDs
The above results revealed that VAS-impaired dyslexic children rather than VASnormal dyslexic children exhibited the significant learning effect in training tasks of VAS estimation, visual search, and digital cancelling, and there was not significant change in visual tracking performance which was not included into the following analysis. We used slopes of the learning curves in the training tasks as the estimation of their learning effect. Functions of linear, exponent, logarithmic, quadratic and cubic were tried to fit the datasets of all the 10 training sessions for these training tasks except the length estimation task. Especially, performance in the length estimation task reached the highest level in about Session 5/6, seemed to reduce from Sessions 6 to 9, and finally (i.e., in Session 10) returned to a level which did not differ from that of the highest level in Session 5/6. This trend of learning curve might be associated with the fatigue effect during the training procedure. Accordingly, we used the datasets from Session 1 to Session 5 in the length estimation task to conduction the curve fitting. It should be noted that the choice of only using a part of training sessions might lead to inflated positive findings. Future studies with setting a longer period for the training procedure are required to further ensure the learning effect in VAS-related tasks (especially for the length estimation). Table S1, the logarithmic function (y = aln(x)+b) was finally selected because it could fit the learning curve of each training task better than the other functions. The ''y'' included the performance of each session in the three visual training tasks of visual search, digital canceling, and length estimation which all exhibited significant improving trends during the intervention; the ''x'' represents the session number; the ''a'' indicates the slope of the learning curve that we would utilize in the following analysis, and b represents the constant. .96 Note. S1-S10 means Session 1 to Session 10 for the training procedure. Visual search-3/4, performance in searching the oval with a 3/4 height of the circle during the visual search task. +, p<.1; *, p<.05; **, p<.01; ***, p<.001.

As shown in
Given the significant improvements in mean accuracy in the visual 1-back task, as well as the d-prime values when the target was presented in the 1 st , 3 rd , and 5 th positions of a string, we further examined the relationship between the improvement of VAS skills and the learning benefit of the three visual training tasks through Spearman correlation analysis (Table S2). Results showed that there were close relationship between improvements in performance of the visual 1-back task and learning-related changes in the three training tasks of VAS estimation, digital cancelling, and visual search. Especially, the increased d' values in the 3 rd position was correlated with learning effects of digital cancelling and length estimation tasks. .34 Note. Dif_VASacc, the subtraction of mean accuracy in the visual 1-back task between the postand pre-tests; Dif_VASd'1, the subtraction of d-prime values between the post-and pre-tests when the target was presented in the 1 st position of a string in the visual 1-back task; Dif_VASd'3, the subtraction of d-prime values in the 3 rd position between the post-and pre-tests; Dif_VASd'5, the subtraction of d-prime values in the 5 th position between the post-and pre-tests. *, p<.05; **, p<.01; ***, p<.001.