Reading disability is associated with phonological problems which might originate in auditory processing disorders. The aim of the present study was 2-fold: first, the perceptual skills of average-reading children and children with dyslexia were compared in a categorical perception task assessing the processing of a phonemic contrast based on voice onset time (VOT). The medial olivocochlear (MOC) system, an inhibitory pathway functioning under central control, was also explored. Secondly, we investigated whether audiovisual training focusing on voicing contrast could modify VOT sensitivity and, in parallel, induce MOC system plasticity. The results showed an altered voicing sensitivity in some children with dyslexia, and that the most severely impaired children presented the most severe reading difficulties. These deficits in VOT perception were sometimes accompanied by MOC function abnormalities, in particular a reduction in or even absence of the asymmetry in favour of the right ear found in average-reading children. Audiovisual training significantly improved reading and shifted the categorical perception curve of certain children with dyslexia towards the average-reading children's pattern of voicing sensitivity. Likewise, in certain children MOC functioning showed increased asymmetry in favour of the right ear following audiovisual training. The training-related improvements in reading score were greatest in children presenting the greatest changes in MOC lateralization. Taken together, these results confirm the notion that some auditory system processing mechanisms are impaired in children with dyslexia and that audiovisual training can diminish these deficits.
Failure to acquire adequate reading skills (reading being slower or less accurate than expected for age) is one of the most common neurobehavioural problems affecting children. Although there is some debate about the precise definition of the term ‘dyslexia’ (Frith, 1999), abundant research has demonstrated that one of its primary features is defective development of the phonetic skills necessary to identify and properly use the constituent sounds of written words. The concept of auditory processing disorder argues that a sensory temporal processing deficit affects the sensory input needed for the proper phonological coding critically required for reading (Mody, 2003). Such a deficit could prevent the learning of precise relations between word sounds and letter sounds, leading to difficulties in associating the printed letter (grapheme) with the appropriate speech sound (phoneme). This could arise in part from left hemisphere dysfunctions, more particularly located in the left perisylvian region (see Demonet et al., 2005 for a review). These deficits are aggravated in the presence of background noise, suggesting that a noisy listening condition, such as often prevails in the classroom, is more particularly deleterious for such children (Bellis, 1996; Chermak and Musiek, 1997; Bradlow et al., 2003; Ziegler et al., 2005). Electrophysiological studies confirm such vulnerability to noise. Cunningham et al. (2001) observed abnormal cortical but also reduced brainstem potentials in response to speech stimuli presented in noise in children with reading-based learning disability. King et al. (2002) reported that the noise selectively degrades cortical responses in learning-impaired children who present abnormalities in neural brainstem synchrony, and that training could increase neural resistance to noise for these children with abnormal brainstem processing. Of particular interest is that, after an auditory training program, cortical (Hayes et al., 2003; Warrier et al., 2004) and brainstem responses (Russo et al., 2005) became more resistant to the deleterious effect of background noise. Assuming that evoked potentials, which reflect the response pattern of neurons responsible for encoding the acoustic complexities of speech, express neural synchrony, and given that noise is known to alter the timing of morphological features of the waveform, it can be supposed that noise excessively desynchronizes auditory firing in learning disabled children.
The neural mechanisms underlying speech intelligibility-in-noise have not yet been well identified, but the medial olivocochlear (MOC) system is a possible candidate. Previous animal studies have shown that efferent bundle activation can improve hearing-in-noise by exerting an antimasking effect (Kawase and Liberman, 1993; Kawase et al., 1993). In humans, weak MOC functioning is correlated with poor tone detection in background noise (Micheyl and Collet, 1996; Micheyl et al., 1997a, b) and reduced speech intelligibility-in-noise in both adults (Giraud et al., 1997) and children (Kumar and Vanaja, 2004). The MOC inhibitory fibres originate from a part of the superior olivary complex and project onto the outer hair cells, forming the last step of a descending auditory pathway which originates in the cortex (Huffman and Henson, 1990). In humans, there are several lines of functional evidence for cortical influence on the cochlear micromechanical properties reflected by evoked otoacoustic emissions, which are sounds thought to be generated by the outer hair cells (Kemp, 2002). This corticofugal modulation may be mediated by the MOC fibres. Such an effect has been demonstrated in cortically resected patients (Khalfa et al., 2001) and by electrical cortical stimulation in epileptic patients (Perrot et al., 2006). Moreover, the MOC fibres which originate from the ipsilateral superior olivary complex (uncrossed MOC system) present a pattern of functional asymmetry influenced by handedness (Khalfa et al., 1998)—a pattern found to be absent in schizophrenic patients (Veuillet et al., 2001). It could thus be argued that abnormal cortical functioning results in impaired cortical feedback to the brainstem and cochlea and that dysfunction in the MOC system, which is partly under auditory cortex control, thus reflects cortical alterations. Finally, the MOC system has been shown to function more strongly in professional musicians (Micheyl et al., 1997a, b; Perrot et al., 1999; Brashears et al., 2003), suggesting the possibility that sound conditioning could strengthen these auditory descending pathways.
We therefore wondered whether the exacerbated deficits under background noise found in children with learning problems might stem from impaired MOC functioning. In a previous experiment we had observed clearly and significantly impaired uncrossed MOC system functionality in learning-impaired children selected for their academic difficulties, in a context of multiple phonemic confusion between voiced/voiceless French phonemes (Veuillet et al., 1999). Fine sensitivity to voicing requires optimal phonological encoding of voice onset time (VOT) duration. In a previous study, Giraud et al. (2005) found that dyslexic adults with auditory discrimination deficits for voiced–unvoiced contrasts presented specific time-coding impairment of the successive components of the acoustic signal. VOT is a temporal component of speech and, as with the perception of speech in noise, the neurophysiological processes involve stimulus-locked synchronous firing. Probably the two processes share the same mechanism, as shown by Wibble et al. (2005) who observed that children with degraded brainstem timing presented greater degradation of cortical responses in noise. Banai et al. (2005) suggested that this could be the result of abnormal cortical feedback.
Motivated by these findings, the present study sought to clarify the nature of the auditory deficit in dyslexia by investigating MOC functioning, coupled with a quantitative assessment of voicing sensitivity using a categorical perception test. MOC functioning was easily and non-invasively explored by an objective procedure. The first experiment was conducted to test the hypothesis of an efferent feedback deficit and to study a categorical perception task assessing the/ba/-/pa/VOT contrast. Coupling these measurements enabled links to be sought between MOC functioning and both voicing sensitivity and reading skills. In the second experiment, both of these parameters were measured before and after an audiovisual training program focused on voicing contrast. It was hypothesized that, compared to average-reading children, children with dyslexia would present abnormalities in MOC functioning which could be associated with an altered sensitivity to voicing and that audiovisual training of voicing could improve voicing sensitivity and MOC functioning. The impact of such training on the performance of children with dyslexia in a word-reading test having been published elsewhere (Magnan et al., 2004), the present study reports only the links between the impact of training-induced changes on auditory parameters and reading skills.
Materials and methods
A group of 46 school-aged children served as subjects. Twenty-three were formally classified as dyslexic on neuropsychological and speech-therapy assessment. These children with dyslexia (mean age, 10yr 11m; range, 8yr 4m to 13yr 11m; 11M, 12F) had a consistent history of persistent specific literacy difficulties, with reading levels at least 18 months behind chronological age, but with a performance Intelligent Quotient above 80 on the Wechsler Intelligence Scale for Children – Revised (3rd Edition) (WISCIII-R). None of them presented an attention-deficit hyperactivity disorder. Twenty-three average-reading children (mean age, 10yr 10m; range, 8yr to 14yr; 10M, 13F), all academically average or above average, having never had to repeat a grade, and without any signs of learning disability, were included as controls: they had normal-for-age reading skills on the French ‘L’Alouette’ reading test, which evaluates reading level in terms of both word and non-word decoding and reading speed (Lefavrais, 1965); none had previously undergone speech therapy. For practical reasons (availability of test material and time limitations) non-verbal intelligence was assessed by the Raven's Progressive Matrices (PM38, Edition 98) (Raven et al., 1984) and not by WISCIII-R as for the children with dyslexia.
The children were all monolingual native French speakers; none suffered from neurological, psychiatric or emotional disorder or were educationally disadvantaged. All had audiometric pure-tone thresholds not exceeding 20 dBHL at octave frequencies in the 250–8000 Hz range and normal middle-ear function, and none had had significant middle-ear infection during infancy. Speech intelligibility in quiet was in all cases 100% between 80 and 50 dBHL for both right and left ears. Handedness was assessed on the short 10-item version of the Edinburgh Handedness Inventory (Oldfield, 1971): all scored between +0.71 and +1, indicating a right-hand manual preference; mean laterality quotient did not differ significantly between the two groups. Testing involved the clinical assessment procedures routinely carried out on children with learning problems consulting in our hospital department. The average-reading children and their parents gave their written informed consent to participate in the research.
Table 1 gives descriptive data of both group and test statistics.
Characteristics of the 23 participants of both groups in Experiment 1
Children with dyslexia
P (df = 44)
Age in months
130 ± 4 (96–168)
131 ± 4 (100–167)
Performance Intelligence Quotient
Age reading in months
132 ± 4 (97–159)
94 ± 4 (72–140)
Reading difficulties (delay in months between reading age and chronological age)
−2.26 ± 1 (−15 to 9)
36 ± 4 (18–76)
Mean range ± SE (range). See text for the tests used. The reading retardation corresponds to the difference between the child's age and reading age as measured with the ‘L’Alouette’ test. Concerning the children with dyslexia: 10 present mixte dyslexia and the other (n = 13) phonological dyslexia.
Behavioural and physiological measurements
Categorical perception test
Stimuli and instrumentation
In this task the children were asked to label token stimuli from a phonemic continuum based on the bilabial stop consonant [b:] plus the vowel [a:] in which only the VOT differed. A categorical perception task was used because it is the most basic linguistic task focusing on phonemes, which are the smallest speech units that can change the meaning of a word (Scarborough and Brady, 2002). VOT, essential to stop-consonant discrimination, refers to the time between onset of ‘voice’ (laryngeal vibration) and its release from the mouth closure. French speakers, produce vocal cord vibration during ‘voiced’ stops, explaining why the VOT for voiced stops (normally perceived as/b/) has a long negative value while for voiceless stops (normally perceived as/p/) it has a short positive value (whereas in English, initial voiced plosive such an/b/are only partially voiced). Stimuli were taken from a synthetically created natural-speech voiced-voiceless continuum designed to range from/ba/to/pa/by step-wise deletion of most of the initial low-frequency segment of the voiced consonant–vowel (CV) sequence/ba/pronounced by a female native French speaker (see more technical details in Laguitton et al., 2000). So that the test would not be excessively long, 16 VOTs (ms) were selected: −110, −96, −73, −55, −26, −20, −14, −9, −3, +3, +11, +17, +23, +29, and +35. These 16 token stimuli were generated via a compact disc player connected to an AC40 audiometer, and were binaurally presented at 60 dB HL at random. The continuum was presented 5 times consecutively.
Procedure. The child was seated at a table in a quiet room, wearing a pair of Telephonics TDH39P headphones. A single-interval, 2-alternative forced-choice identification procedure without feedback was adopted. The two possible responses (letters B and P) were printed on a white sheet stuck onto the table, and the stimulus was identified by pointing and giving an oral response. The experimenter recorded the responses manually on a data sheet. For the rare cases when the two responses (pointing and oral) were discordant, the pointing response was recorded in order to better target the phoneme–grapheme association and avoid false responses due to problems of articulation. Prior to beginning the test, children performed a short training trial with feedback to familiarize them with the phonemic contrast, and then the whole continuum from the longest to the shortest VOT was administered without mixing the CVs, in order to sensitize the child to the voicing fade-out. During the test, the VOT stimuli were presented randomly to avoid order effect.
Data processing. The percentages of B and P answers per VOT were calculated and identification functions plotted for each child by graphing the percentage/BA/and/PA/responses as a function of VOT (−110 through +35 ms). These identification curves were fitted with the Matlab® software using a sigmoid function (hyperbolic tangent). Two parameters characterizing each identification function were extracted from this mathematical model: (1) the phoneme boundary, calculated as the 50% point of the fitted labelling curve (maximum confusion point) and (2) the identification function gradient (slope).
MOC system functioning
The Otodynamics® ILO92 apparatus (software V3.94L) was used for recording the cochlear responses (Kemp, 2002). Following the MOC system exploration protocol described in detail elsewhere (Collet et al., 1992; Veuillet et al., 2001), evoked otoacoustic emissions were recorded monaurally at five stimulus intensities ranging from 57 to 69 dB SPL in 3 dB steps, in random presentation order, with and without contralateral acoustic stimulation consisting of 30 dBSL continuous broadband noise (speech-like noise) produced by an audiometer.
The test was carried out in a soundproof room with the children sitting quietly watching silent video cartoons of their own choosing or playing on a Gameboy with no sound. Both ears were tested in random order.
The groups did not differ significantly in emission amplitude in either right or left ear. The amplitude reduction with contralateral acoustic stimulation was quantified in terms of stimulus-equivalent attenuation, defined as the mean decrease in the ipsilateral stimulus (dB) which causes the same reduction in emission amplitude as does a 30 dBSL contralateral broadband noise (for more details, see Veuillet et al., 1999): the more negative the equivalent attenuation, the more functional the MOC system. An asymmetry index quantifying the functional lateralisation of the MOC system was calculated as the difference between right and left ear: the more negative the index, the more MOC system functioning predominated in the right ear, and the more positive, the more in the left.
Statistical analysis was performed with SigmaStat® software (SPSS). All data were expressed as mean (± SE). The data were evaluated by parametric t-tests, or by mixed-design repeated-measures analysis of variance (ANOVA) with group effect as between-subjects factor and stimulus condition (VOT duration, Ear) as within-subject factors. The significant ANOVAs were followed up by post hoc Bonferroni-adjusted t-tests. Linear regression analyses were conducted on the relationship between MOC functioning and VOT sensitivity. Although there was no significant difference between groups for non-verbal performance or chronological age (Table 1), partial correlations were even co-computed to control for these factors. The significance threshold for all tests was set at P < 0.05.
Comparisons between average-reading children and children with dyslexia
Averaged category labelling functions for the VOT series (percentage identified as/ba/or/pa/, plotted as a function of VOT) are presented for each group in Fig. 1a. Both average-reading and dyslexic children identified stimuli along the VOT acoustic–phonetic continuum in a categorical-like fashion with well-established plateaus at the extremes and steep gradients around the phoneme boundary. Significant differences between groups were predominantly observed for the VOT around the phoneme boundary, where the function showed a rightward shift in the children with dyslexia as compared to the average-reading group. The graph in Fig. 1 shows that the functions differed significantly between groups with respect to phoneme boundary [F(1,44) = 12.81; P < 0.001] but not to gradient [F(1,44) = 1.48; P = 0.23). In average-reading children, phoneme perception tipped over from ‘ba’ to ‘pa’ earlier than in children with dyslexia. A two-way ANOVA for repeated measures on one factor (duration) using the percentage of ‘ba’ categorizations for each subject was significant for the VOT duration effect [(F(15,660) = 478.95; P < 0.001], for group [F(1,660) = 13.79; P < 0.011)], and for the interaction [F(15,660) = 4.29; P < 0.001]. Post hoc pairwise multiple comparison procedures (Bonferroni t-test) indicated that, whichever the group, the percentage of ‘ba’ responses obtained with the −9 ms VOT stimulus differed significantly from that with all other stimuli, but also revealed significant differences in percentage ‘ba’ responses between average-reading and dyslexic children on both sides of the phoneme boundary [−14, −9 and −3 ms VOT (P < 0.001) and −3, +3, +11 ms VOT (P < 0.01)] and also for the 35 ms VOT (P < 0.05). For all these cues, average-reading children gave less ‘ba’ responses. Thus, the fall-off in ‘ba’ responses became significant as of the −14 ms VOT in this group, but did not appear before the −9 ms VOT in the group of children with dyslexia.
Comparison between groups of average-reading children l and children with dyslexia for categorical perception and MOC functioning. (a) Mean percentage stimuli identified as ‘ba’ (solid lines) or ‘pa’ (dotted lines) as a function of VOT for the control (open squares) and dyslexic (filled squares) groups. Enclosed Fig. 1a: Mean ± SE and individual phoneme boundary values for the dyslexic (filled squares) and the control (open squares) children. (b) Comparison between right ear (RE) and left ear (LE) evoked otoacoustic emission suppression (expressed as Equivalent Attenuation) for children with dyslexia (filled symbols) and average-reading children (open symbols). (c) Means and individual values of the interaural difference in Equivalent Attenuation (EA): Asymmetry Index: EARE-EALE) in children with dyslexia (filled squares) and average-reading children (open squares). Bars represent the standard error of the mean. (*P<0.05; **P<0.01, ***P<0.001).
MOC system functioning
The mean equivalent attenuation values for groups and ears are presented in Fig. 1b. A contralateral suppression effect was present in both groups for both ears. To analyse the results, a two-way ANOVA for repeated measures on one factor (Ear) revealed no significant effect of the factor Ear, a tendency for a significant effect of Group [F(1,44) = 3.10; P = 0.08], but a significant interaction between Ear and Group [F(1,44) = 27.11; P < 0.001]. Post hoc comparisons (Bonferroni t-test) indicated that, in average-reading children, the MOC system was much more functional in the right than the left ear (P < 0.001), but predominated in the left ear in children with dyslexia (P < 0.01). Moreover, a significant difference was observed between the two groups in the right ear (P < 0.001), suggesting a deficit of MOC functioning in the right but not the left ear in children with dyslexia. These differences in MOC function lateralization between the groups are confirmed in Fig. 1c, plotting mean and individual asymmetry index values: a t-test revealed a significant difference between the groups (t44df = 5.21, P < 0.001). In addition, one sample t-test showed that the mean laterality score for average-reading children (−1.4) was significantly less than 0 (P < 0.001), indicating right ear predominance, whereas the mean score of children with dyslexia (+1.07) was significantly greater than 0 (P = 0.004), indicating a left ear advantage in MOC functioning. Analysis of individual data showed a right ear advantage in MOC functioning for only four children with dyslexia, in contrast to 21 of the 23 average-reading children.
Relations between MOC functioning, categorical perception and reading ability
Linear regressions between MOC functioning (for equivalent attenuation in right and left ear and asymmetry index) and phoneme boundaries were computed for each group separately. Significant correlations emerged exclusively for the group of children with dyslexia, and concerned the three indices of MOC functioning (Fig. 2): the stronger the contralateral suppression effect in the right and left ear, and the more strongly positive the asymmetry index and more strongly negative the phoneme boundary even after controlling for non-verbal performance [respectively: r = 0.43, r = 0.48 and r = −0.44 (P < 0.05)] and age (respectively r = 0.42, r = 0.48; r = −0.44 (P < 0.05)]. After the exclusion of two outlying children presenting the most prolonged boundaries, less MOC functioning and also the most severe reading difficulties (reading levels at 67 and 76 months behind chronological age), these significant correlations persisted and were even strengthened for equivalent attenuation in left ear (r2 = 0.21; P = 0.03) and asymmetry index (r2 = 0.25; P = 0.02).
Scatter-grams showing the relation between phoneme boundary and left MOC functioning (a) and Asymmetry Index (b) in children with dyslexia (filled symbols) and average-reading children (open symbols). The lines represent the line of best regression fit (solid for children with dyslexia and dashed for average-reading children). The two circled points (outliers) were not included in the regression. *P<0.05.
For the group of children with dyslexia, a single significant link was found between reading difficulties and phoneme boundary (r = 0.42; P = 0.04): the more severe the retardation, and more strongly positive the phoneme boundary even after controlling for non-verbal performance and chronological age (respectively r = 0.42 and 0.45; P < 0.05).
These results show that some children with dyslexia have altered voicing sensitivity which is sometimes associated with abnormal MOC functioning with more particularly abnormalities in MOC system lateralization. Moreover, the children with the most severe reading problems (i.e. greatest reading difficulties) were those who were the most deficient in categorical perception.
Categorical perception deficits have been widely reported in children with dyslexia, who are generally described as being ‘less categorical’ (for more recent studies, see Adlard and Hazan, 1998; Serniclaes et al., 2001, 2004) or ‘less accurate’ (Manis et al., 1997; de Gelder and Vroomen, 1998; Joanisse et al., 2000; Breier et al., 2001; Chiappe et al., 2001) than average-reading children in phonetic contrast identification tasks. In the present study, all children presented identification functions with a clearly regular S-shape. The mean slopes of the boundary were both equally steep, suggesting that no difference existed in the consistency of phoneme categorization between the two groups. The mean phoneme boundaries, however, differed significantly, average-reading children labelling VOT signals as/pa/which were still being identified as/ba/by the children with dyslexia. These children were also less accurate than the average-reading children in identifying clear instances of/p/at the end of the continuum. The two groups may thus have been using different criteria for VOT identification, children with dyslexia showing abnormal sensitivity to voicing, which they continued to perceive even when it was so short as to be no longer perceptible for the average-reading children, as if their neural encoding of voicing was altered leading to erroneous VOT perception. It could be due to deficits in inhibitory processes or excessive noising in the auditory pathways. This altered sensitivity could also be related to the higher degree of allophonic perception found both in dyslexic children (Serniclaes et al., 2004) and in those with gap-detection deficits (Van Ingelghem et al., 2001; Hautus et al., 2003). Electrophysiological studies also revealed that learning-impaired children present abnormalities in response to speech syllables that originate specifically in the brainstem (Cunningham et al., 2001; King et al., 2002; Wibble et al., 2004) and that dyslexic children process verbal stimuli in a different manner than normal-reading children at the cortical level (Kraus et al., 1995; Breier et al., 2003) with a link between these brainstem and cortical responses (Wibble et al., 2005).
Altered MOC functioning in children with dyslexia is in agreement with our previous study (Veuillet et al., 1999) and also with an experiment conducted on learning-disabled children with auditory processing disorder (Muchnik et al., 2004). Such a deficit was not found in children with specific language impairment (Clarke et al., 2006) but was present in children with selective mutism, where abnormal MOC functioning in the right ear was also found (Bar-Haim et al., 2004). Functional studies, both in animals subjected to electrical stimulation of the inferior colliculus (Zhang and Dolan, 2006) and in humans undergoing cortical stimulation (Perrot et al., 2006), now support the assumption that the MOC system is under central control. Several previous studies argue for abnormal patterns of cerebral activation in dyslexia more particularly at the level of the auditory cortex: for example, diminished left temporoparietal cortex activation (for a recent review, see Demonet et al., 2005), reduced left temporoparietal cortex activation paired with increased temporoparietal activity specifically during reading (Simos et al., 2000a, b), and altered M100 asymmetry pattern (Edgar et al., 2006). The MOC system is driven by cortical areas and thus the present study provides new arguments for the notion of abnormal cerebral lateralization in dyslexia. The robust and significant link observed between voicing sensitivity and left-ear MOC advantage in the group of children with dyslexia suggests that some such children may, perhaps through compensatory activation, develop right-hemisphere dominance for speech consonants. The question which then arises is whether such abnormal lateralization of MOC functioning is linked to deficits in mechanisms which are the cause and/or consequence of reading impairment. Longitudinal and maturational studies will be necessary to examine this. Probably the descending system linking brain and cochlea exerts ‘top-down influences’ (Davis and Johnsrude, 2007), important for the preservation of the neural encoding of auditory information, especially when temporal processing is required as is the case with VOT perception. It has been suggested that there may be an interaction between ascending and descending corticofugal auditory pathways (Suga et al., 2002), and perhaps the significant links found in our group of children with dyslexia between the phoneme-boundary VOT value and MOC functioning may be taken as indicating a role of the auditory efferent system in inhibiting extraneous auditory cues. However, a statistical relationship between two variables does not necessarily entail an underlying causal link between them, and one must be very cautious in interpreting correlation data.
Materials and methods
Eighteen children with dyslexia participated in this study, three of whom had been included in the first Experiment. The selection criteria were the same as for previous experiment except for the performance Intelligent Quotient, which was never under 70. The children were divided into two groups of equal size. One group (the ‘Trained group’: mean age, 10yr 6m; range, 9yr 2m to 12yr 8m; 7M, 2F) completed 5 weeks’ training, with pre- and post-training assessment. The second group (the ‘Non-Trained group’: mean age, 10yr 4m; range, 9y to 11y 10m; 4M, 5F) also received two assessments with a 5-week interval, identical to those of the trained group. These two assessments are referred to as ‘pre-test’ and ‘re-test’, respectively. Each child and their parents were fully informed on all study procedures and gave their written consent.
Table 2 summarizes the results on inclusion tests and statistical tests. None of the parameters differed between the groups.
Characteristics of the children with dyslexia in the Experiment 2
P (df = 16)
Age in months
126 ± 5 (110–152)
124 ± 4 (108–142)
Performance Intelligence Quotient
Age reading in months
86 ± 2 (77–92)
85 ± 3 (73–106)
Reading difficulties (delay in months between reading age and chronological age)
40 ± 4 (19–65)
40 ± 5 (18–63)
(N = 9 in each group). Same legend as for Table 1. Concerning the children of the trained-group: five present phonological dyslexia, three with mixed dyslexia and one with surface dyslexia. In the non-trained group: six with mixed dyslexia and three with phonological dyslexia.
The training procedure consisted of an exercise embedded in an audiovisual computer game (Play-On) designed by Danon-Boileau and Barbier (2000), played individually by each child of the training group 4 times a week for 5 weeks. The goal was for each child to complete 20 training sessions, each lasting 30 min. The training did not disrupt their regular classroom lessons, and all the children, whichever the group, continued to receive their usual special education and written language therapy during the study period. The 10-h training module was based on an alternative choice with feedback, and consisted in listening to consonant-vowel (CV) syllables and choosing which of two printed letters, differing only in voicing, represented the initial consonant. This training, practiced as a game-like exercise, focused on the voicing opposition between two items in six pairs of phonemes:/b/-/p/;/t/-/d/;/k/-/g/;/f/-/v/;/s/-/z/and/ch/-/j/. After auditory presentation of the syllable (for example,/ba/), a basketball fell from the top of the screen and the child pressed one of two keys to drop the ball into the basket corresponding either to ba or to pa. This audiovisual game, discriminating the phonetic feature of voicing, provided training with ortho-phonological units and was expected 1/to help dyslexic children to specify phonemic representations, and 2/to improve matching between visuo-orthographic pattern (letter name) and phonological unit (letter sound), so that children would not confuse phonetically close phonemes, which could facilitate their written-word processing.
Auditory and reading assessments
Two forms of auditory assessment were given to the children: a categorical perception test and an exploration of MOC functioning. The tests were performed as described earlier (Experiment 1), ahead of training (pre-test) and post-training (re-test) for the trained group, or simply after a 5-week interval for the non-trained group (re-test), to assess training-related changes. To evaluate the impact of training on reading ability, a word recognition test (Ecalle, 2003) was also administered but to only seven of the nine children, two being absent at the time of evaluation. It consisted in finding the target word in a list of five items consisting of the orthographically correct word (e.g. bateau, meaning ‘boat’), and four pseudowords: namely, a homophone (bato), a visually similar item (baleau), an item sharing the same initial letters (batte) and an item containing an illegal letter-sequence (btaeu). A composite score (called ‘recoding score’) of the first two responses (i.e. target word and homophone) was calculated (max.: 36).
Statistical analysis was performed as in Experiment 1 to compare MOC functioning, VOT sensitivity and reading scores between the two sessions (pre- and re-test) for each group (trained and non-trained).
Pre- and re-test identification performance for the trained (left panel) and non-trained (right panel) groups is shown in Fig. 3a. No pre- to re-test change in identification was found for the non-trained dyslexic group; audiovisual training, on the other hand, resulted in a clear and substantial shift of the identification function, more particularly in the region around the phoneme boundary. Repeated measures ANOVA, comparing the pre-test performance of the two groups, with group (trained or non-trained) as between-subjects variable and VOT duration as within-subject variable, found no significant group difference but a main effect of VOT duration [F(15,240) = 113.48; P < 0.001] with no significant interaction between group and VOT, indicating that both groups divided the continuum into/ba/and/pa/categories in a similar manner before training. Separate two-way repeated measures ANOVAs were conducted for each group (trained and non-trained), with test session (pre- and re-test) as the independent variable. For the trained group, this analysis revealed a main effect of test session [F(1,120) = 13.22; P = 0.007], a main effect of VOT duration [F(15,120) = 144.65; P < 0.001], and also a significant interaction between test session and VOT [F(15,120) = 1.98; P = 0.02]. Post hoc Bonferroni t-test revealed significant differences between pre- and re-test values for three VOT stimuli located within the phoneme boundary (−9, −3 and +3 ms VOT) but also for one located at the extremity of the continuum (+35 ms). There were no significant test-session differences for any of the VOTs in the non-trained group, with a main effect of VOT [F(15,120) = 80.39; P < 0.001] but no significant interaction between test session and VOT—indicating that, for the non-trained children, mean identification functions showed no difference in waveform morphology between the two test sessions over an interval of 5 weeks. A repeated measures ANOVA comparing re-test VOT identification between groups showed a significant group difference [F(1,240) = 6.68; P = 0.020], a main effect of VOT [F(15,240) = 132.84; P < 0.001] and a significant Group × VOT interaction [F(15,240) = 1.79; P = 0.04].
Effect of training on categorical perception. (a) Mean/b/-/p/identification functions of trained (circles) and non-trained (squares) children with dyslexia. (b) Mean pre-test and re-test phoneme boundary values for trained (circles) and non-trained (squares) children with dyslexia. Comparisons between pre-test (filled symbols) and re-test (open symbols). Bars indicate standard error of the mean. *P < 0.05, **P < 0.01, ***P < 0.001. (c) Individual subject changes in phoneme boundary between the pre- (before) and re-test (after) sessions for the trained (circles) and non-trained (squares) children. Post-test changes are plotted relative to normalized pre-test session values. The diagonal hatched line denotes equal performance, so that points below the lines show individual improvement between the earlier and the later test. *P < 0.05.
This training effect also impacted mean phoneme boundaries (Fig. 3b). (It is to be noted that, for one child in the non-trained group, the phoneme boundary was not measurable at any moment.) A repeated measures ANOVA with test session (pre- and re-test) as within-subject factor and Group (trained or non-trained) as between-subjects factor produced no main effect of Group but a significant test session effect [F(1,15) = 11.15; P = 0.004], and a significant interaction [F(1,15) = 7.09; P = 0.018]. Post hoc Bonferroni t-test revealed that the phoneme boundaries, which were comparable between groups at the pre-test session, were significantly modified from pre- to re-test only in the trained group (P < 0.001): as shown in Fig. 3c, the correlation between pre- and re-test phoneme boundaries was significant only in the non-trained group (P < 0.05). Individual subject analysis found that six of the nine trained-group subjects exhibited a substantial decrease in phoneme boundary (more than 5 ms) compared to only one child in the non-trained group. All the other children maintained constant phoneme boundaries over time; this included three trained children, who would appear to have been ‘resistant’ to training, but two of these ‘resistant’ children had no impaired categorical perception in the first place.
MOC system functioning
Contralateral suppression effects were present in both the trained and non-trained groups for both ears on both pre- and re-tests. Two-way ANOVAs with repeated measures were separately conducted on the right and left ear to separate the effects of test-session (within subject) and group (between subjects). No significant main effects for these factors and no significant interaction were found. To further investigate a possible lateralized training-induced effect on MOC functioning, asymmetry indexes were compared between groups and across test sessions (Fig. 4a). They were positive for both the trained and non-trained group in the pre-test session but reached −0.24 ± 0.56 on re-test for the trained children, indicating a change in the asymmetry of MOC functioning. A two-way repeated measures ANOVA on one factor of repetition revealed no significant effect of group, a tendency for a significant effect of test-session [F(1,16) = 3.53; P = 0.08] and a significant interaction [F(1,16) = 5.66; P = 0.03]. Post hoc Bonferroni t-test revealed a significant decrease in asymmetry index between pre- and re-test sessions only in the trained group (P = 0.008). Close analysis of the individual asymmetry index data revealed that five of the nine trained children showed a decrease in this acoustic parameter after training, with a particularly great change for one of the children, whereas no change was ever observed in the non-trained subjects. The individual changes in equivalent attenuation (right and left ear) from pre- to re-test for both groups are plotted in Fig. 4b. In the right ear, they were significantly correlated between the two sessions whichever the group, whereas the left-ear correlation was not significant in the trained group—suggesting that training had the effect of attenuating the suppression effect in this ear.
Effect of training on Equivalent Attenuation results. (a) Mean Asymmetry Index for trained (circles) and non-trained (squares) subjects during the pre- and re-test sessions. Bars indicate standard errors of the mean. (b) Individual results. Comparisons of pre-test (before) and re-test (after) values for right ear and left ear. The diagonal hatched line denotes equal values, so that points above or below the line show individual changes between the earlier and the later test.
Relationships between change in MOC functioning and shift in phoneme boundary
To discover whether these training-induced modifications in phoneme boundary and MOC functioning were related, linear regressions were calculated between the training-induced shifts in phoneme boundary and MOC functioning; however, none reached significance. Closer analysis of individual results revealed that the asymmetry indexes of five of the six trained children who presented a clear shift in phoneme boundary towards average-reader values and who can be qualified of ‘dyslexic responder’ shifted towards an RE advantage for MOC functioning. This decrease involved improved MOC functioning in the right ear after training in three children and/or decreased MOC functioning in the left ear in five children. Those who associated both profiles tended to be those showing the largest shift in phoneme boundary. For the sixth ‘responder child’, MOC functioning increased in both ears but most especially in the left. After training, this child no longer differed from the children with dyslexia in the previous experiment, who were characterized by normal phoneme boundaries and strong MOC inhibition in the left ear. For the three ‘resistant’ children, no common profile for change in MOC functioning emerged.
Reading skills and relationships with auditory processing changes
The audiovisual training had a beneficial effect on recoding scores, which significantly increased between the pre- and re-test sessions (respectively, 28.57 ± 2.29 and 32.71 ± 1.29; t6dl = −2.82, P = 0.03) with improvements being evident for the five trained children present at the reading evaluation. A significant link existed between changes in training-induced scores and in asymmetry index. The more asymmetric the MOC system became in favour of the right ear (asymmetry index decrease), the more the word-recognition score improved (Fig. 5). There was no significant correlation between the training-induced changes in phoneme boundary and in word recognition score. Analysis of individual data revealed that the reading improvement observed in five children was accompanied by a shift to the left in the phoneme boundary in only three trained children, the other two showing no change. One of the two children with no improvement in word-recognition score showed a decrease in phoneme boundary, and the other showed no change.
Scatter-grams showing the relation between training-induced changes in MOC functioning (Asymmetry Index) and recoding score. The line represents the line of best regression fit. *P<0.05.
After training, the patterns of voicing sensitivity and MOC system lateralization shifted towards average-reading values and the improvement in reading scores was as large as the change in MOC functioning in favour of the right ear.
In the audiovisual training experiment, the identification curves of the non-trained children did not vary between test sessions, indicating that neither task repetition or on-going speech therapy were exerting significant effects. In the trained group, the significant shifts in identification function produced new phoneme boundaries which were closer to those found in the group of average-reading children as tested in the first experiment. Thus, training was effective in enhancing French listeners’ ability to perceive the/b–p/contrast, in agreement with previous behavioural (for recent studies, see Bradlow et al., 1997; Bedoin, 2003) and neurophysiological (Kraus et al., 1995; Tremblay et al., 2002) studies. This training-induced phoneme boundary shift suggests that audiovisual training enables a better perceptual representation of voicing to be constructed in the auditory pathways. However, our experiment also confirmed that this rehabilitation strategy based on improving the correspondence between speech contrasts and graphemes does not prove efficacious in all children (Bishop et al., 2005; Tijms and Hoeks, 2005) and indeed sometimes has no effect (Pokorni et al., 2004); the next step will be to understand why. MOC functioning did not differ significantly between sessions, attesting to the good reproducibility of these measures. The significant decreases in asymmetry index, which tended to be greater in the trained group, suggest an effect of training on the lateralization of MOC system functioning, which tends to become more negative, as in the average-reading children of the first experiment. This MOC lateralization change was shown to be significantly correlated with improvement in reading skills. Our sample size was relatively small, but this training-induced change appeared in seven of the nine trained children. Using the same remediation technique as ours, Santos et al. (2007) observed a beneficial effect on pitch discrimination deficit, supported by an increase in P300 amplitude. Other studies, conducted on learning-disabled children with different types of training, have reported functional brain changes along the auditory pathways from brainstem to cortex. At sub-cortical level, training was found to improve neural synchrony (Hayes et al., 2003; Russo et al., 2005). At cortical level, the reading improvement induced by training was associated with increased left-hemisphere involvement, corresponding to a ‘normalisation’ of underactivated left-hemisphere regions (Simos et al., 2002; Aylward et al., 2003; Shaywitz et al., 2004). But as in the present study, these changes were not systematically observed and when observed were not always very large. The greater left auditory cortex plasticity reported in the above training studies may be in agreement with the lateralized training effect observed in the present study, especially as voicing is known to be preferentially processed in the left hemisphere and the MOC functioning advantage in favour of the right ear is thought to reflect the hemispheric dominance of the left auditory cortex in language processing.
First, this study confirms the existence of deficits in auditory processes for some children with dyslexia, showing links between altered voicing sensitivity and reading difficulties. Secondly, audiovisual training was found to modify not only VOT categorization but also MOC lateralization, this effect on descending auditory pathways being correlated with the training-induced effect on reading. Thus, these results argue for auditory deficits in dyslexia with a positive effect of training.
However, it has been commonly underlined that an auditory deficit is neither a necessary nor a sufficient condition for disturbed reading acquisition (Bailey and Snowling, 2002). This is well confirmed by the present results, where a large overlap in phoneme boundaries was observed between groups. Our children with dyslexia all had considerable experience of prior speech-therapy which may well have had some effect on their voicing categorization even if it had failed to fully remedy their dyslexia. It must also be underlined that not all children with dyslexia have a categorical perception deficit (Manis et al., 1997). Our study shows that certain ‘average-reading’ children performed no better than certain dyslexics on this task, and yet were average readers—suggesting that other factors must be involved which are able to compensate any such deficit in phonemic perception. However, the children with dyslexia presenting the greatest alteration in VOT sensitivity were those who presented the greatest reading difficulties. Manis et al. (1997) observed that the children with lower phonemic awareness had shallower phoneme identification than children with higher phonemic awareness, and our reading-impaired children may have presented this pattern. At the individual level, however, some average-reading children are comparable to children with dyslexia in their categorical perception, even if dyslexics are often ‘over-trained’, which indicates that other abilities, doubtless involved in phoneme sensitivity, come into play, confirming the multidimensional nature of reading disorders.
During the MOC exploration, a greater differentiation between the two groups emerged. This reinforces the importance of the use of this test as a biological marker for abnormal lateralized central processing in children with reading acquisition deficits in a context of normal hearing. MOC system investigation could be an easy complementary and generally objective tool to pinpoint children and monitor the effect of training. In the present experiment the audiovisual training focused on the improving correspondences between speech contrast and graphemes. The significant changes observed in VOT categorization and MOC system functional lateralization argue that both the representation of speech sound and MOC functioning are plastic, and that these changes may reflect processes of normalization in the neural activity which underlies speech representation. Recently, Hayes et al. (2003) reported that after training the cortical response became more resistant to the deleterious effect of noise. Such improved resistance to noise may be related to the changes in MOC functioning observed in the present study. This rehabilitation strategy, however, did not prove efficacious in all children with dyslexia and the co-variations between the training-induced changes in auditory parameters and reading scores differed from one child to another. This heterogeneous effect of training is imputable to the particular profile of each child with dyslexia before training: two of them presented no deficit in VOT sensitivity (perhaps because over-trained), another presented a maximal recoding score before training, and two children showed an MOC system lateralized in favour of the right ear. Thus, in spite of a selection which tried to be as rigorous as possible, an unwanted heterogeneity subsisted and was increased by the diverse type of dyslexia and the diverse forms of rehabilitation that these children were following, and also probably by maturational processes (Bishop and McArthur, 2005)—even if the effect of age was controlled for in the present study.
In conclusion, our results support the notion that children with dyslexia present different degrees of MOC functioning deficit, which may in some cases be associated with an altered perception of voicing that is proportional to their reading acquisition difficulties. Some of these auditory abnormalities, however, are reversible by training. MOC system investigation could be an easy and objective tool to characterise children with dyslexia and monitor the neural changes induced by training, but further research is needed to better understand the involvement of the corticofugal system in auditory perception. Lastly, because of the great heterogeneity of the learning-impaired population, probably only an individual approach will enable better specification of a given child's disability, opening up the possibility of more efficient rehabilitation tailored to the precise disorder, the ultimate challenge being an improvement in literacy skills among them the reading speed and fluency. Ahead of any attempt at therapy, the pervasiveness of these auditory deficits will need to be determined by assessing their form and extent.
We thank Sophie Jéry and Isabelle Comte-Gervais (respectively speech therapist and neuropsychologist) for help collecting the psycho-educational data; Dr Isabelle Soares-Boucaud (child psychiatric) for allowing us to test the children with dyslexia and for giving us some helpful advice during the setting up of the training experimentation. We are very grateful to students and more particularly Norbert Maionchi-Pino for help collecting the experimental data. Lastly, this study would not have been able to be made without the children who always showed enthusiasm and motivation and we thank them infinitely.