Edited by:
Reviewed by:
*Correspondence:
This article was submitted to Quantitative Psychology and Measurement, a section of the journal Frontiers in Psychology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
The study of orthographic errors in a transparent language like Spanish is an important topic in relation to writing acquisition. The development of neuroimaging techniques, particularly functional magnetic resonance imaging (fMRI), has enabled the study of such relationships between brain areas. The main objective of the present study was to explore the patterns of effective connectivity by processing pseudohomophone orthographic errors among subjects with high and low spelling skills. Two groups of 12 Mexican subjects each, matched by age, were formed based on their results in a series of
Reading is a cognitive process that requires visually identifying written elements and their respective phonological association to form meaning. In recent years, the study of reading under a neurocognitive perspective has focused on reading disabilities, mainly accuracy and speed (
Some authors like
It has been proposed that the activation patterns of these three regions of the reading system are different between good and deficient readers, such as dyslexics (
However, the vast majority of the previously mentioned studies have been conducted on shallow orthographies. Some authors argue that the consistency of different orthographies is a factor that may directly influence the processing of reading (
Spanish is considered a language with a regular orthography, such as Dutch, Italian, and German, due to its high grapheme–phoneme correspondence for reading; however, for writing, some phonemes may be mapped onto two or three different letters. This is particularly true in Mexican Spanish, given that – in addition to the matches between a phoneme and several graphemes of standard Spanish – other sounds are also equivalent. For example, the phoneme /s/ matches the graphemes “c,” “s,” and “z”; the phoneme /x/ matches “x,” “g,” and “j” the phoneme /j/ matches “y” and “ll”; and the phoneme /b/ matches “b” and “v.” Moreover, Mexican Spanish comprises a great percentage of words originating from the country’s various indigenous languages – now completely integrated into Spanish – many of which involve these types of phonemes and spelling not based on orthographic rules (arbitrary). Because of all this, Mexican Spanish is probably the transparent language where speakers might make the most mistakes when writing pseudohomophones (words with an orthographic error and the same phonology as the correct one), or in the visual recognition of a pseudohomophone as a valid word while reading.
Although these mistakes do not compromise reading comprehension in normal persons in a meaningful way, they do cause the speakers of Mexican Spanish to make numerous pseudohomophone spelling mistakes, something observable in the general population (
The study by
The main objective of the present study was to explore the possible differences in effective connectivity among subjects with high and low spelling orthographic abilities while processing pseudohomophone orthographic errors by using visual word recognition tasks. Another goal of this study is to compare SEM and DCM models given that they entail different approaches to connectivity models and each one reports on different processes to model the brain activities under the Bold signal paradigm.
Twenty-four young adults (age
Prior to the fMRI registrations, five tasks were applied to all the participants intended to assess their handling of homophone spelling in Spanish words (b-v, c-s-z, g-j, ll-y, h-no h) in four different contexts: completing words, dictation (words and text), error detection in a text, and free composition. The tasks used to discriminate the subjects’ performances had yielded an adequate reliability value (α = 0.833) and a very high discrimination capacity in order to distinguish between groups with different orthographic skills (
Out of this large sample, 12 subjects with low spelling skills (LSSs group, 10th percentile) and another 12 with high spelling skills (HSSs group, 90th percentile) were selected to form this study’s sample.
Throughout two experimental tasks, the subjects were exposed to 80 Spanish words, 60 out of which were spelled correctly. Also, 20 words contained a homophone orthographic error [e.g.,
In the first task (blocks A and B, spelling recognition task), the participants were required to indicate whether the word was written correctly or else contained a pseudohomophone orthographic error. In block A, 50% of the words were written correctly and the remaining 50% contained an orthographic error. In block B, 100% of the words were written correctly. In the second task (blocks C and D, visuoperceptual recognition task) the participants were instructed to answer whether the word displayed contained the vowel “i” or not. In block C, 50% of the words were written correctly and 50% contained a pseudohomophone orthographic error. In block D, 100% of the words were written correctly. We should bear in mind, however, that the participants did not have that information, neither in block B nor D.
Both the stimuli and the interval between them were 1 s long. In order to present them, a block design was used: the stimuli were divided into eight blocks of 10 stimuli each and presented pseudo-randomly. The stimuli were presented in an Arial 60 font and were typed in white on a black background with a 300 pixel-per-inch resolution.
Both the words spelled correctly and those with an orthographic error had a high or low frequency according to a frequency dictionary widely used in studies involving words in Spanish (
The total number of stimuli from both categories (words spelled correctly and incorrectly) was divided in half to be presented in both experimental tasks. In each task, four rest blocks were presented with a center fixation dot during which the subjects were not supposed to conduct any activity; the change of color in the fixation dot told the subjects they were about to start watching words and executing answers. Likewise, four activation blocks were presented in each task with ten stimuli each: two of them with words spelled correctly and incorrectly (50–50%), and two blocks only with words spelled correctly.
The participants gave their answer to each stimulus through one out of two buttons, following the requirement of the two experimental tasks:
We used a GE Signa Excite HDxT 1.5 Tesla (GE Medical Systems, Milwaukee, WI, USA) and an 8-channel head coil. For each experimental task, we obtained 32 4-mm thick adjacent axial cuts. We used an echo planar pulse sequence with a Repetition Time of 3 s, echo time of 60 ms, 26-cm FOV, and a 64 × 64 matrix. The voxel size used was 4.06 × 4.06 × 4 mm. From each experimental task, a total of 62 brain volumes were obtained. For reasons of image acquisition time and experimental design, six brain volumes per task were discarded, thus leaving a total of 56 to be analyzed later on (according
The pre-process and the statistical analysis of the images were conducted using the SPM8 computer package (
For the smoothing, a Kernel
There are several statistical conditions involved in the use of SEM to estimate connectivity. Basically, they concern the SEM properties as regards the linear model, and the conditions of range and order that the Path Analysis models must follow. In fact, generating a factorial structure to obtain a score by ROI based on the voxels defined would be, ultimately, a peculiar application of the dimension reduction or, in SEM terms, of a measurement model. Accordingly, it is more than debatable that effective connectivity estimation follows a Path Analysis model strictly.
Regardless of these rather conceptual considerations, several papers have shown the limitations of this technique, which can be summarized in the following aspects. Firstly, SEMs do not allow us to easily analyze the self-impact effects, that is, the β
Consequently, papers like those by
In addition, an effect exists that has received little attention based on the assumption of variance homogeneity between ROIs which can only be solved by standardizing the values, but which is not assumable from the onset (
Such is the case between groups (
The choice of DCM models seems an interesting alternative to the SEM models, given that their statistical properties make them somewhat more malleable. Generally speaking, they are more tolerant with reciprocal effects, with
where t is continuous time,
Therefore, as mentioned above, not specifying the C matrix involves that, given certain conditions – that is,
On the other hand, DCM is strictly linked to ROIs presenting statistically significant activations. In general, the fundamental condition of applying DCM lies on the situation of subject-model specificity when fitting models to specific subjects under defined experimental conditions.
This question has not been overlooked. In fact, several choices of parameter estimation have been generated for it.
It has also been discussed that using DCM entails difficulty given that it only analyzes activated ROIs, and therefore, it overlooks other sources of variation. However, some proposals have tried to generate alternatives to this possible bias effect (
According to the recommendations by
Despite the limited empirical evidence in this field (using orthographic tasks in Spanish-speaking populations) we expected the DCM models to be different for the two groups considered, that is, more complex in the A–B task for the LSS group than for the HSS group. In contrast, for the C–D task, we expected the models fitted in the HSS group to be more complex than those in the LSS group because the former had to complete both tasks (spelling and visuoperceptual recognition).
In any case,
A multivariate analysis of the variance (MANOVA) was conducted by using orthographic competence (High or Low) as a factor between the two groups and the four programmed blocks as an intra-group factor (A, B, C, and D), defining the subjects’ ages as a covariant to extract the components caused by that factor and the following dependent variables: the number of correct answers given in each block, and the simple reaction time in each subject’s answer to the 20 tries in each experimental condition.
Clearly significant was the interaction between Group and Blocks concerning the number of correct answers (
Finally, the main effect linked to the group effect for the reaction time (
Neither the effect of age as a covariant nor its interaction with the block or the group of belonging turned out statistically significant. To prevent the possible “double dipping” effect described by
Descriptive statistical results, mean and standard deviation (SD) for the number of correct answers and the reaction time for each experimental condition.
Group | Number of correct answers |
Average reaction times |
AGE | ||||||
---|---|---|---|---|---|---|---|---|---|
A | B | C | D | A | B | C | D | ||
High spelling skills (HSS) | 17.33 |
15.75 |
17.00 |
18.00 |
825.83 |
780.98 |
688.64 |
649.33 |
22.50 |
Low spelling skills (LSS) | 6.76 |
10.67 |
18.67 |
18.17 |
847.91 |
809.99 |
664.24 |
682.77 |
21.17 |
We conducted an analysis of the SPM algorithm’s linear model (
Definition of Regions of interest (ROI) from activations by group and task.
ROI number | MNI coordinates ranges |
||||
---|---|---|---|---|---|
1 | R/Precentral gyrus (RPCG; |
64/68 | -6/10 | 10/30 | |
2 | L/Inferior temporal gyrus (LITG; |
64/-50 | -32/-50 | 0/-18 | |
3 | R/Middle temporal gyrus (RMTG; |
52/70 | 15/-41 | -2/-22 | |
4 | L/Cerebellum, posterior lobule (LCPL; |
-42/-22 | -46/-36 | -38/-30 | |
5 | L/Middle frontal gyrus (LMFG; |
-50/-22 | -10/10 | 42/58 | |
6 | R/Supramarginal gyrus (RSMG; |
48/62 | -60/-44 | 26/36 | |
7 | L–R/Anterior cingulate (LRAC; |
-4/6 | 30/38 | -10/14 | |
8 | L/Parahippocampal gyrus (LPHG; |
-24/-14 | -18/6 | -22/-14 | |
1 | R/Precentral gyurs (RPCG; |
52/86 | -18/10 | 2/26 | |
2 | L–R/Middle frontal gyrus (LRMFG; |
-16/24 | -26/2 | 46/74 | |
3 | L/Middle frontal gyrus (LMFG; |
-50/-34 | 6/18 | 30/54 | |
4 | L/Precentral gyrus [1] (LPCG1; |
-66/-58 | -18/2 | -6/14 | |
5 | R/Superior frontal gyrus (RSFG; |
4/16 | 50/56 | 22/34 | |
6 | L/Precentral gyrus [2] (LPCG2; |
-46/-26 | -14/-26 | 52/70 |
The extraction of ROIs was data driven. We selected the most significant signal values for each region by defining an area of 5 mm and conducting a Component Principal Analysis to extract the ROI. For each cluster we presented the
Based on the results above and consequently following a data-driven strategy, we extracted the ROIs values for the four first-level analyses conducted through MarsBar by defining, for each significant area, a 5-mm volume around the most statistically significant
Correlation matrix between ROIs for the AB task (spelling recognition) and for the CD task (visuoperceptual recognition) for the two competence groups (High or Low).
Correlation between ROIs for the Spelling recognition Task – A and B blocks –(HSS group – LSS group) | ||||||||
---|---|---|---|---|---|---|---|---|
RPCG | 1 | |||||||
LITG | 0.672∗ |
1 | ||||||
RMTG | 0.722∗ |
0.325∗∗ |
1 | |||||
LCPL | 0.533∗- |
0.355∗∗ |
0.230∗∗ |
1 | ||||
LMFG | -0.737∗- |
0.148∗∗ |
-0.597∗- |
-0.501∗ |
1 | |||
RSMG | 0.666∗ |
0.174∗∗ |
0.566∗ |
-0.012- |
-0.449∗- |
1 | ||
LRAC | -0.545∗ |
-0.122∗∗ |
-0.046- |
-0.812∗ |
0.404∗ |
0.066- |
1 | |
LPHG | -0.533∗- |
-0.116∗∗ |
-308∗∗ |
-0.402∗- |
0.240∗∗ |
-0.266∗∗ |
0.498∗ |
1 |
RPCG | 1 | |||||||
LRMFG | 0.162 |
1 | ||||||
LMFG | -0.215- |
-0.327∗∗ |
1 | |||||
LPCG1 | -0.183- |
-0.251- |
0.500∗ |
1 | ||||
RSFG | 0.294∗∗ |
0.001 |
0.412∗ |
-0.131 |
1 | |||
LPCG2 | -0.509∗- |
0.146 |
0.095 |
0.330∗∗ |
-0.365∗ |
1 | ||
RPCG (Right Precentral Gyrus), LITG (Left Inferior Temporal Gyrus), RMTG (Right Middle Temporal Gyrus), LCPL (Left Cerebellum Posterior Lobule), LMFG (Left Middle Frontal Gyurs), RSMG (Right Supramarginal Gyrus), LRAC (Left–Right Anterior Cingulate), LPHG (Left Parahippocampal Gyrus).
RPCG (Right Precentral Gyrus), LRMFG (Left–Right Middle Frontal Gyrus), LMFG (Left Middle Frontal Gyrus), LPCG1 (Left Precentral Gyrus 1), RSFG (Right Superior Frontal Gyrus), LPCG2 (Left Precentral Gyrus 2).
In order to estimate effective connectivity, each correlation matrix was submitted to the procedure described by
The results of the models with the best fit are summarized in
Fit index for the best models under SEM approach for each task and groups.
Spelling recognition Task – A and B blocks – |
||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Group | Fit index |
Explained variance ( |
||||||||||
χ2 | RMSEA | RPCG | LITG | RMTG | LCPL | LMFG | RSMG | LRAC | LPHG | |||
HSS | 0.786 | 1 | 0.3754 | 0.0–0.09 | 0.543 | 0.668 | 0.436 | |||||
LSS | 6.362 | 7 | 0.4982 | 0.0–0.04 | 0.399 | 0.402 | 0.681 | 0.280 | 0.464 | |||
HSS | 4.140 | 3 | 0.2468 | 0.0–0.07 | 0.474 | 0.429 | 0.521 | |||||
LSS | 0.950 | 1 | 0.3298 | 0.0–0.09 | 0.409 | 0.589 | 0.462 | 0.520 | 0.225 | |||
Out of all the models complying with the above criteria, we selected, for each condition, those offering the best fit and the highest value in the determination coefficient (
RPCG (Right Precentral Gyrus), LITG (Left Inferior Temporal Gyrus), RMTG (Right Middle Temporal Gyrus), LCPL (Left Cerebellum Posterior Lobule), LMFG (Left Middle Frontal Gyurs), RSMG (Right Supramarginal Gyrus), LRAC (Left–Right Anterior Cingulate), LPHG (Left Parahippocampal Gyrus).
RPCG (Right Precentral Gyrus), LRMFG (Left–Right Middle Frontal Gyrus), LMFG (Left Middle Frontal Gyrus), LPCG1 (Left Precentral Gyrus 1), RSFG (Right Superior Frontal Gyrus), LPCG2 (Left Precentral Gyrus 2).
Finally, the path diagrams for each model representing the standardized parameters ML estimation are shown in the
As we can observe in
As for the HSS group, only three ROIs received connections from the other ROIs (right parietal gyrus, LRAC, and RMTG), whereas in the LSS group, five ROIs received connections from the other ROIs involved in the system (right parietal gyrus, LRAC, RMTG, LMFG and RSMG).
For the visuoperceptual recognition task the complexity of the connectivity network is similar in both groups. Eleven paths were defined for both, but the connectivity structure is different: in the LSS group, all the ROIs involved in the connectivity network receive connections from at least one of the other ROIs of the system.
This paper analyzes the DCM models in each of the two groups of orthographic competence (High and Low) in the two tasks presented: A–B (spelling recognition task), and C–D (visuoperceptual recognition task). Therefore, four DCM models were generated based on the average values of each group’s twelve subjects under every condition.
However, the analysis of the results obtained by SEM indicate that, in the case of the subjects with high orthographic competence when solving the A–B task, their levels of significance are of little statistical intensity, except for the ROI defined by the right pre-central gyrus (Zarabozo-Hurtado et al., under revision).
Thus being the case, and in light of the presence of only one statistically relevant ROI, we decided to discard the DCM model for this group and task, so that, ultimately, the remaining three models were studied with several significant ROIs. To do this, we followed the recommended steps by
The estimations of the three remaining DCM models are displayed in
It is also important to mention than the ROIs involved are different in both groups.
We studied the possible differences in detecting homophone orthographic errors in the neurobiological substrate by using two approaches: analyzing the effective connectivity model estimated through SEM and DCM. In order to study this phenomenon, two tasks were used: a spelling recognition task, and a visuoperceptual recognition task. These were applied to two groups of subjects, one with HSS, and another with LSS (Zarabozo-Hurtado et al., under revision).
According to the results while performing the spelling recognition task, the LSS group showed poorer behavioral performance (fewer correct answers and higher reaction times) as compared to the HSS group. Given that the early stages of the reading process involve encoding and orthographical-phonological conversion, these data suggest that, globally, this cognitive process is different in the LSS and HSS groups.
The SEM results of the HSS group suggest that, in the spelling recognition task, two brain areas are involved in the majority of significant effects: the RMTG, and, to a lesser extent, the LPHG. In the LSS group, however, the majority of significant effects are received in the LPHG, and, to a lesser extent, the RSMG. The latter anatomical region has been reported in different studies under transparent orthographies as visual to a phonological encoder (
In this sense, we should mention that the LSS group probably presents activations in the supramarginal gyrus as a compensatory mechanism, since these subjects do not present the temporal-occipital activations shown by the HSS group and which are usually observed in reading tasks in healthy persons. For us this means that the LSS subjects must use this compensatory mechanism to access other later reading processes. This is a consistent result in papers studying reading in transparent languages and, in this sense, our results are similar to those in
We should point out that such compensatory mechanism in readers of transparent orthographies is observed rather generally in persons with reading deficiencies, as shown by the recent meta-analyses by
Additionally, the intensity of these effects both in the most active brain areas during the task and in the rest of ROIs is much less intense in the LSS group than in the HSS group. In the spelling recognition task, the connectivity model is very different between the HSS and the LSS groups. On the one hand, in the former, the ROI receiving the most significant effects is the LPHG, whereas for the LSS group, it is the LMFG. Once again, the most intense effects appear in the HSS group, while they are much more diffuse in the LSS group.
Along with the behavioral data, and given that the early stages of the reading process involve encoding and orthographical-phonological conversion, these data suggest that, globally, we can conclude that the effective connectivity pattern during the spelling recognition task is different between HSS and LSS subjects. The behavioral performance data, which shows that the LSS group performed worse in the task than the HSS group, all of it suggests that the reading process is globally different between the HSS and LSS groups. This had already been proposed in previous neuroimaging studies (
In addition, in a broader sense, our results agree with those of
The DCM results of our study show an effective connectivity pattern quite different from the SEM connectivity pattern. Firstly, in the spelling recognition task in the HSS group, for whom the task is easier due to their probable visual word processing expertise, the DCM model was not estimated due to the fact that only one ROI presented statistically significant activations.
Conversely, in the LSS group, we found significant connectivity patterns between the LMFG, the RMTG, and the LITG toward the RSFG. However, as was the case in the effective connectivity analysis through SEM, the intensity of these effects as well as the intensity of each ROI’s self-activation with itself is small.
These data suggest again that, in both groups, reading is a globally different process in terms of brain activation. The impossibility to estimate DCM in the HSS group is due to the fact that only one ROI reaches statistical activation and in consequence it is impossible to estimate connectivity models.
This situation means that – for this group – there are a smaller number of activated clusters that can explain a statistical model. Strictly speaking, there are no other statistical effects other than the self-correlation effects for this specific ROI. It might be thought that the subjects with high skills found so little difficulty in the spelling task that they did not need special connectivity networks to meet the demand. This would be consistent with some previous results by Zarabozo-Hurtado et al. (under revision) showing similar effects in the estimation of simple effects. The lack of papers on connectivity in this type of task for this specific population prevents us from delving further into the discussion of this aspect.
Conversely, in the LSS group, the DCM pattern is more complex than in the HSS group. In fact, these results are congruent with those found by
Some studies suggest that individuals with reading problems present a series of compensatory mechanisms at brain level when facing the complexity it means for them to execute this cognitive task (
The fact that it is the subjects from the HSS group presenting the more complex DCM model, however, is not as surprising as it might seem. The execution of a relatively simple task, such as finding a letter in a word, which was executed correctly by the low-spelling skills group, might be affected by the automatization of the orthographic processing of words, where the presence of orthographic errors increases the task’s difficulty only for those subjects who have developed a specialization in recognizing orthographic patterns, like the HSS group had.
However, in adults with low orthographic abilities, an orthographic violation is not automatically processed, probably due to weaker orthographic representations in long term memory or to a poorer development of the orthographic lexicon. In other words, the subjects from the HSS group, in the C-D pair of blocks (visuoperceptual recognition task) would be conducting two tasks at once: vowel detection, as requested, and, involuntarily or automatically, orthographic mapping.
Despite the above comments in relation to HSS group, it is important to bear in mind the small sample size used to estimate the statistical effects. There are several DCM models estimated with small sample sizes, but there is not enough evidence about the effects of sample size on the connectivity modelization process (
Our paper presents some limitations that need to be discussed. The main limitation, in our opinion, is the fact that the subjects were selected among students in the senior year of high school and, in light of their reading performance, some subjects from the LSS group might have been dyslexic but, as far as we know, none of them had been diagnosed previously. In other words, some of the participants from this group might have suffered from a relatively mild form of dyslexia that would have been compensated by their own means allowing them to reach the senior year. On the other hand, no measurement or estimation instrument was applied to them for intellect, which might have also influenced their performance in this study’s tasks.
Nonetheless, these limitations need to be clarified. Firstly, the orthographic abilities tests used to form the groups were very thorough, which allowed us to form the HSS and LSS groups with wide knowledge of the subjects’ reading performance at the moment of inclusion in the study, and it also allowed us to have much intra-group homogeneity as regards their current reading skills. Additionally, the fact that all the subjects, both HSS and LSS, were students from the same degree of high school makes it unlikely that there were great differences in the general intellectual functioning of both groups, which makes our results hardly questionable in this sense.
We would also like to note that, to conduct this study, a 1.5T scanner was used with a TR of 3. There is a possibility that, with a 3T scanner, the DCM model could be estimated for the HSS group in the spelling recognition task. However, our data suggest that, in that case, our results would point in the same direction, that is, a very simple DCM model in the HSS group when compared to the LSS group, thus suggesting that the task is easy for these subjects. Still, even if these conditions are not ideal for connectivity studies, recent studies on effective connectivity have used similar equipment to the one used in the present study (
Another limitation of our study is the sample size we selected, which may be considered rather small. However, this should be seen as a relative limitation. The criteria to confirm the groups were strict, and the method of assignment to the groups, following the extreme values technique, allowed us to maximize the possible differences. This made data interpretation rather clear in terms of brain activation despite the relatively small sample size (
Our paper also has some strengths that deserve comment. The most remarkable one is that this is, to our knowledge, the first paper exploring the effective connectivity model, estimated through SEM, and the efficient connectivity model, estimated through DCM, in a visual recognition task of homophone errors in Spanish, while at the same time controlling the subjects’ level of orthographic competence and, consequently, separating those with a high level from those with a low level of competence. As has been commented above, this type of error is characteristic of transparent languages, especially of the variety of Spanish spoken in different parts of Latin America. In this sense, our results are particularly interesting, given that this type of orthographic errors is characteristic and very usual of transparent languages, where reading as a cognitive function has some distinctive features.
More studies should be analyzed to see whether the activation patterns observed in this study are found when facing detection tasks of other types of orthographic errors, or, on the contrary, homophone error detection activates a pattern in good and bad readers somewhat different from other types of errors. In the future, we should obtain more detailed information about brain activities in order to analyze more statistically complex models like Farràs et al. (under revision) suggest.
To conclude, we can say that taken globally, the analyses of the connectivity of the tasks under study through SEM and through DCM present some similarities. The first one is that both the SEM and the DCM models show distinctive connectivity patterns between the HSS and LSS groups. Likewise, both types of analyses suggest patterns with effective connectivity in one case (SEM) and the other (DCM) that are much clearer and with more intense effects in the case of the HSS group as compared to the LSS group. This is much clearer, obviously, in the case of the visuoperceptual recognition task.
Nonetheless, they also present important differences. The DCM models are very different between both groups under study, for the spelling and visuoperceptual recognition tasks. In fact, the DCM models probably reflect somewhat better what happens with the behavioral conduct of the tasks under study. A clear interaction effect was revealed between group and task, so that the HSS group conducted the spelling recognition task more efficiently than the LSS group. Likewise, as we commented above, in the visuoperceptual recognition task, the effective connectivity pattern was more complex in the HSS group than in the LSS group. However, it was due to the fact that the subjects who are good readers probably carry out both tasks at the same time, whereas the LSS subjects would only carry out the visuoperceptual vowel recognition task, their reading being much less automatized than that of the good readers.
Furthermore, we consider it important to remark that it is essential to continue with this research line. It might be interesting to analyze the ROIs that emerge when we request the subjects to conduct an automatic process by differentiating HSS and LSS groups, like the Stroop task. Another important line to continue is the analysis of people who have in fact a real orthographic problem, for example working with dyslexic persons as compared to persons with good competences in orthography.
Finally, these data point in two complementary directions for future research. Firstly, we should approach the estimation of connectivity models when faced with these tasks or similar ones with samples from the same populations but with a larger amount of ROIs, not just the ones generated from data-driven approaches. Instead, more theoretical models should be analyzed and their possibilities of empirical and statistical evidence evaluated for viability. Secondly, we must work in a more structured way on the analysis of limitations and possible improvements of the statistical models we use to estimate connectivity since they involve a special conception of the way connectivity works and it entails a specific way to understand that complex reality.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This study was supported by the