Participants

A group of 107 students with dyslexia (aged between 8.25 and 13.25 years, mean 10.49), 61 students with pre-lingual hearing loss that had received cochlear implants (aged between 6.00 and 13.17 years, mean 9.71), and a comparison group of 785 typically developing students (aged between 6.33 and 12.33 years, mean 8.80) participated in the study. All of them attended primary school in state schools or state-aided private schools in different Spanish cities.

The participants with dyslexia (58 boys and 49 girls), whose first language was Spanish, came from 28 schools. They were all enrolled in regular classes, from 3rd to 6th grade (see Table 1 for details). They had been diagnosed by the Psycho-Pedagogical Guidance Team of their educational district following the criteria established by Spanish educational regulations including: (1) a reading level 1.5 standard deviations below the norms for their chronological age; (2) an intellectual level within the range considered as normal (no one had an IQ below 85); (3) no one was diagnosed as presenting symptoms of ADHD; (iv) intact hearing and a normal or corrected vision.

Table 1.  Demographic information on the participants per Group: Dyslexic, Deaf and Control (age in years). 

The participants with deafness (34 boys and 27 girls) were recruited though the Guidance Team for Hearing-Impaired Students from 27 schools. They attended mainstream classes using spoken Spanish (45 children) or bilingual schools using spoken Spanish and Spanish Sign Language (16 children). All of them were integrated with normally hearing children in regular classes from 1^st to 6^th grade (see Table 1 for details). They had received cochlear implants (CI), 38 children before 30 months (mean = 1.67, range between .67 and 2.5 years) and 23 children after of 30 months (mean = 5.49, range between 2.58 and 11 years).

The comparison group (400 boys and 385 girls) were from 14 schools and attended the grades (1 to 6) corresponding to their chronological age (see Table 1 for details). All of them followed the typical reading acquisition path, and their scores in the reading test (TECLE) were above the mean score -1.5 SD of their age and grade.

Instruments

The tasks used in the present experiment are taken from a previous work by ^{Domínguez and collaborators (Domínguez & Alegría, 2010}; ^{Domínguez et al., 2014}; ^{Domínguez et al., 2016}). In none of the tasks did the participants have to speak. The stimuli were written material, which had to be read silently, and the response consisted of choosing one item from among several options (see https://complydis.usal.es/).

Procedure

The participants were tested in their own classroom, collectively in the group of normally developing children, in small groups (between 1 and 5) in the case of children with dyslexia, or individually in the case of the children with hearing loss. The tasks were explained to the participants with deafness using their preferred method of communication: oral and/or sign language. The first session was devoted to the READ, and the second session to SEM and SNT tasks.

Data analysis

The number of correct responses per participant, corrected for guessing (number correct responses - (number of errors / (number of alternatives - 1)), was calculated in the three tasks (READ, SEM and SNT) and the scores converted into a percentage of correct responses. The method commonly used to compare the performance of two groups in a task consists of considering their mean score in the task. This method, however, requires that the two groups be adequately matched for age (or reading level if this variable is to be controlled). In the present study this method was unsuitable because participants with dyslexia or deafness presented great variability in age as well as in reading level. This led us to adopt a method (described in detail in the Results section) based on regression analysis, which consists of evaluating the performance in each task using delays instead of raw scores. For example, suppose that the READ-score of a ten-year-old participant was reached by the Control Group at the mean age of eight and half years. The participant’s READ-delay would be one and a half years). This method permits individual delays to be calculated for each task by comparing the score of a participant with the whole comparison group represented by the regression equation of the group, rather than considering a few participants who have approximately the same age (or the same reading level) as the participant being evaluated.

Evaluation of the performance in the experimental tasks

The results of the three tasks (READ, SEM, and SNT) are presented following the same scheme. First, the percentage of correct responses (corrected for guessing, as explained above) for each participant and task, was calculated. Two linear regression equations per group (Control, Dyslexic, and Deaf) and task were then calculated, one considering the percentage of correct responses in the task as a function of the age of the participant, and the other (only for SEM and SNT) as a function of their reading level (READ).

Table 2 represents the parameters of the regression equations (intercepts (a) and slopes (b)) per task and group of participants, as a function of age and of reading level. The corresponding determination coefficients (R²) are also included. In order to help with visualizing the results, the data of READ as a function of Age are presented in Figure 1, which shows individual scores and the regression lines per group.

Table 2.  Slopes (b), intercepts (a), and coefficients of determination (R² ) of the linear regression equations, considering the percentage of correct responses as a function of chronological age (in READ, SEM and SNT), and as a function of the reading level (in SEM and SNT), per Group (Control, Dyslexic, and Deaf). 

Figure 1.  Percentage of correct responses in the READ task as a function of Age. The linear regression lines per Group (Control, Deaf and Dyslexic) are included. 

Secondly, individual delays in each task were calculated taking the difference between the age at which a participant reached a particular score in a task (READ, SEM, and SNT) and the age at which the control group as a whole reached the same score.

For example, in the Age Control Analysis, a participant of 120 months of age and with 30 points in the SEM task will have a SEM delay relative to the control participants of the same age (SEM-Da) of -24.47 months. This value was obtained using the regression equation of the Control group considering SEM as a function of age: SEM = .96 x AGE - 61.80). According to this equation, the Control group reached 30 points in the SEM task at 95.53 months, which is 24.47 months before the participant in our example: Age = (30 + 61.80) / .96 = 95.53 months; SEM-Da = 95.53 - 120 = -24.47 months.

The same procedure was used to evaluate the delays in SEM and SNT tasks, relative to reading level (the Reading Control Analysis).

Let us consider that the participant of the above example had a score in the READ test of 25 points. The equation of SEM as a function of READ for the control group (Table 2): SEM = 1.07 x READ - 4.73 shows that the SEM score corresponding to 25 points in READ is 22.02 points: SEM = 1.07 x 25 - 4.73 = 22.02. Consequently, the participant has a delay of + 7.98 points (30 - 22.02) in the SEM task. This value (delay in SEM-points) was translated into delay in months which is easier to grasp. The equation relating SEM to AGE (Table 2) was used to this end. The slope of the equation indicates that the SEM score increases at a rate of .96 points per month. Therefore, the SEM-Dr of + 7.98 points corresponds to a delay of + 8.31 months (7.98 / .96) in SEM.

The mean delays in years per group in READ, SEM, and SNT in the Age Control Analysis (READ-Da, SEM-Da and SNT-Da), as well as in the Reading Control Analysis (SEM-Dr and SNT-Dr), are presented in Figure 2.

Figure 2.  Mean delays, in years, relatively to Age Control (-Da) and Read Control (-Dr) for Dyslexic and Deaf Groups. 

The reading level of Dyslexic and of Deaf Groups

Before coming to the question of reading strategy, it will be useful to examine the reading delay of our Dyslexic and Deaf Groups. Figure 1 shows the reading level (READ) as a function of AGE of the participants and the regression lines per group. Close inspection of the figure shows that both groups, Dyslexic and Deaf, are delayed relative to the Controls of the same age. This is not a surprise in the case of the participants with dyslexia since they were selected due to their delay in reading. More importantly, the deficit in reading tends to increase with age in both groups. The slopes of the regression lines are .46, .59, and .82 in Dyslexic, Deaf, and Age Controls, respectively. Put in other terms, the mean delays in the Deaf Group increases from 8 to 18 months between 3^rd (100 months of age) and 6^th grade (136 months). In the Dyslexic Group the corresponding delays increased from 23 to 39 months in the same period.

Assessment of the Key Word Strategy

The rationale proposed to establish the extent to which participants adopt the KWS consists of comparing the delay in the basic reading test (READ) with the delay in the SEM task. Greater difficulties in the SEM than in the READ test are the signature of using the KWS. Figure 2 shows the mean delays in READ and SEM in the Age Control Analysis. Children with dyslexia show almost identical delays in both tasks (READ-Da = -2.96 and SEM-Da = -2.89; t(106) = -.723; ns). The Deaf Group, on the contrary, shows a deficit in the SEM task almost double that of the deficit in READ (READ-Da = -1.09, and SEM-Da = -1.97; t(60) = 6.238; p < .001, d = .40). This suggests that participants with hearing loss adopted the KWS to some extent, while this tendency in the children with dyslexia, if it existed, did not differ from that of normally evolving readers of the same age.

A more direct way to examine a tendency to use the KWS is to consider the delays of the participants in the SEM task as a function of the reading level (see Figure 2). The results of the Dyslexic and Deaf groups agree with those obtained in the Age Control Analysis. The mean delay of the Dyslexic Group (SEM-Dr = -.05) did not differ from the mean delay of the Control Group, which was zero (t(890) = .497, ns), while it reached almost one year in the Deaf Group (SEM-Dr = -.90), which is significantly greater than the delay of the Control group (t(844) = 7.25, p < .001, d = .96).

Finally, it was interesting to consider the possibility that the tendency to adopt the KWS changes with age and/or reading level. To examine this question the correlation between Age and the gap between READ-Da and SEM-Da was calculated. The r-value in the Dyslexic and Deaf Groups were r = .007; ns, and r = .319; p = .012, respectively. This suggests that the tendency to adopt the KWS increased with age in the Deaf Group, but not in the Dyslexic Group.

Syntactic Ability: dealing with function words

As argued in the Introduction section, the KWS necessarily involves a tendency to disregard the function words. Accordingly, it was expected that the SNT task would be found harder than the READ task. Figure 2 shows the delays in READ and SNT in the Age Control Analysis. These delays are basically identical in the Dyslexic Group (READ-Da = -2.96 years, and SNT-Da = -2.94 years; t(106) = -.177; ns), indicating that the SNT task does not add difficulty to the basic reading test (READ). In contrast, these two delays in the Deaf Group differed significantly (READ-Da = -1.09 years, and SNT-Da = -1.95 years; t(60) = 5.651; p < .001, d = .41), pointing to an extra deficit when dealing with function words.

Another way to examine this question consists of directly comparing the delays in the SNT task in Dyslexic, Deaf and Controls of the same reading level (Figure 2). The absence of any specific difficulty with function words would produce a zero mean delay in the SNT task in all groups. Note that this is necessarily the case in the Control Group. The results are consistent with those obtained when controlling for age. The mean delay of the Dyslexic Group (SNT-D _r = -.14) did not differ from the mean delay of the Control Group (t(890) = 1.39, ns) while it reached almost one year in the Deaf Group (SNT-D _r = -.89) and was significantly greater than the delay of the Control Group (t(844) = 6.47, p < .001, d = .89).

These results agree with those obtained for delays in the SEM task, meaning that the SNT and SEM tasks are more difficult than the basic Read test in both analyses (Age and Reading Control Analysis) for the Deaf but not for the Dyslexic Group. The readers with deafness showed a strong tendency to neglect the function words of the sentences, suggesting that they have poor syntactic ability. This deficit was directly associated with adoption of the KWS.

In order to assess, as directly as possible, the relationships between syntactic ability and the KWS independently of the reading level, two parameters were considered: the SNT-Dr as evaluating the ability to deal with function words, and the SEM-Dr as assessing the tendency to adopt the KWS. It was predicted that these two parameters would be positively correlated_. As expected, the correlation coefficients (Pearson’s r) were r = .524; r = .682; and r = .508 (p < .001, in each case) for Dyslexic, Deaf, and Reading level control Groups, respectively.

The key word reading strategy

The first goal of this study was to examine the reading strategy used by children with dyslexia and with hearing loss to read sentences. Previous studies have shown that adults with deafness who had reached high levels of education, as well as children with deafness with and without cochlear implants, use a reading strategy, the KWS, which consists of identifying the most frequent content words of the sentence, and deriving its overall meaning on this basis (^{Domínguez & Alegría, 2010}; ^{Domínguez et al., 2014}; ^{Domínguez et al., 2016}). Evidence that the KWS was adopted by deaf persons is that the sentence reading task involving foils that were semantically compatible with the meaning of the key words (SEM) was harder than the basic reading test (READ). The new question to which we sought an answer in this study was whether children with dyslexia use the KWS or not, and, whatever the answer to the question - why?

The results are clear-cut. As was expected, compared with controls of the same age, the Deaf Group showed a reading delay in the SEM task twice the delay shown in the basic reading test (-1.97 and -1.09 years respectively). The Dyslexic Group, on the contrary, did not show any additional delay in the SEM task relative to the READ test (-2.96 and -2.89 years, respectively). This suggests that dyslexics, despite having a mean reading delay of about three years, did no rely on the KWS. Based on previous research, it was speculated that dyslexics might compensate for their poor reading ability by exploiting the semantic context of the sentences derived from the meaning of some of its content words (^{Perfetti et al., 1979}; ^{Perfetti & Lesgold, 1977}; ^{Stanovich, 1980}). This was clearly not the case. So, the notion that the KWS is adopted by all poor readers to compensate for a fluency deficit can be rejected. Indeed, both groups were struggling readers but only the deaf children adopted the KWS. The critical difference between dyslexic and deaf children, which might explain why they differ in their reliance on the KWS, is their syntactic ability - an issue that is examined in the following section.

Syntactic processing of sentences: dealing with function words

It was argued in the Introduction that the KWS is probably related with poor syntactic skills, and the results confirm this hypothesis. In the Dyslexic Group the delay in the SNT task was almost identical to the delay in READ (-2.94 and -2.96 years, respectively). In the Deaf Group, however, the delay in the SNT task was about twice that measured in the READ test (-1.95 and -1.09 years, respectively), suggesting that children with deafness do have a specific difficulty in dealing with function words. Nonetheless, the Deaf Group presented an extra deficit in both SEM and SNT tasks relative to the basic reading test, while the Dyslexic Group’s delay in the three tasks, READ, SEM and SNT, was the same. Finally, if it is admitted that poor syntactic ability is a condition for adopting the KWS, then, these two modes of functioning must be correlated. Actually, the empirical index evaluating the (in)ability to deal with function words and the index evaluating the tendency to adopt the KWS were significantly correlated. The former explained 26% and 28% of the latter’s variance in dyslexic and control groups, while this value reached 47% in the Deaf Group. It is tempting to conclude, that the syntactic inability of children with hearing loss is a cause of their tendency to adopt the KWS, and not simply their poor reading skills. Obviously, the present results are merely correlational, and correlations do not allow causal conclusions. Only a longitudinal study demonstrating that successful training in the manipulation of function words reduces the use of KWS in the deaf could be considered as strong evidence of a causal relationship.

The notion of context to explain the use (or not) of KWS remains an intriguing question. Previous experiments on the use of context by poor readers (^{Perfetti et al., 1979}; ^{Perfetti & Lesgold, 1977}; ^{Stanovich, 1980}) typically consisted of showing that the final word of a sentence is not treated but simply guessed. For example, a poor reader presented with a sentence like “The singer took his guitar and intoned a pleasant …. (blues)” is more likely to respond song instead of blues than a good reader. It is worth noting that song is grammatically correct, indicating that poor readers use the semantic context to guess, but they also consider syntactic constrains to elaborate the response. In the present study the foils in the SEM task were all ungrammatical. Dyslexic readers do not adopt the KWS because they do apply the syntactic constraint which eliminate the ungrammatical responses generated by the foils. The deaf readers probably elaborate a context-like model using the key-words and choose the response without taking syntax in consideration.

Controlling for reading level in SEM and SNT reading tasks

The emerging picture when reading level was controlled confirms the conclusions obtained in the Age Control Analysis. The mean delays of the Dyslexic Group were nil when reading level was controlled (-.05 and -.14 years in SEM and SNT tasks, respectively). In contrast, the delays in the Deaf Group decreased from about 2 years in the Age Control Analysis (- 1.97 and - 1.95 in SEM and SNT tasks, respectively) to about one year in the Reading Control Analysis (- .90 and - .89 years in SEM and SNT tasks, respectively). Furthermore, these delays remained constant across reading levels. Nonetheless, when children with dyslexia reached a particular reading level, which happened with a mean delay of three years, their delay in SEM and SNT disappeared. This suggests that the reading mechanisms operating do not differ from those of normally developing readers. The notion that persons with dyslexia, because of their poor word processing skills, adopt the KWS, which favours content word processing, in order to increase fluency and/or reduce memory charge was clearly not supported by the present results. The Deaf Group however, continued to have a significant delay in both SEM and SNT when reading level was controlled, corroborating the view that they have a specific problem with function words. This result confirms previous data obtained with deaf adults with high levels of education (^{Domínguez et al., 2014}). It has been repeatedly shown that persons with deafness have a deficit at a morphosyntactic level (^{Gaustad & Kelly, 2004}; ^{King & Quigley, 1985}; ^{Niederberger, 2007}; ^{Paul, 1998}; see also, in Spanish, ^{Moreno-Pérez et al., 2015}; ^{Rodríguez, García & Torres, 1997}; ^{Stockseth, 2002}). It is likely that readers with deafness use semantically driven strategies to a greater extent than their hearing peers, including dyslexics, because their syntactic ability is poorer.

It remains to be understood why reading activity does not help children with hearing loss to improve their ability in dealing with function words, while it does in children with dyslexia. One obvious reason is that function words would tend to be ignored with the KWS. Consequently, reading activity fails to play its well-established language-teaching role (^{Cunningham & Stanovich, 1991}; ^{Stanovich, 1986}; ^{Stanovich & Cunningham, 1992}). Exposure to speech seems to be indispensable to grasp the role of function words in sentence processing. Content words can be learnt by designation and/or by explicit teaching. Function words, however, are passively acquired through simple exposure to spoken material. Following ^{Marschark, Sarchet, Rothen & Zupan (2010}), it is important to understand what is still missing for cochlear implanted children to develop a basic morphosyntactic ability which would prepare them to exploit reading practice to progress in syntactic development. It may be speculated that syntactic morphemes are generally less salient, shorter and unstressed, than content words, so that they remain less accessible for children with hearing losses.