Phonological Restriction Knowledge in Dyslexia: Universal or Language-Specific?

Developmental dyslexia is the most studied and well-documented of the specific learning disabilities in school-age children across languages, which reaches from 5-to-17.5% individuals (e.g., Shaywitz & Shaywitz, 2005; Snowling, 2001). There is now a consensus that developmental dyslexia stems from a genetic neurodevelopmental disorder that does not depend on inadequate intellectual or educational backgrounds (e.g., Lyon, Shaywitz, & Shaywitz, 2003; Sprenger-Charolles, Cole, Lacert, & Serniclaes, 2000; Vellutino, Fletcher, Snowling, & Scanlon, 2004). There is considerable evidence for a phonological deficit as the major correlate of language disabilities in dyslexia, which underpins the cognitive disorder (e.g., Ramus, Rosen, Dakin, Day, Castellote, White, & Frith, 2003; Ziegler & Goswami, 2005). However, an outstanding, long-lasting question that remains unclear, even unanswered, is what underlies the phonological deficit in dyslexia (e.g., Ramus, 2001). Three main directions have been proposed to account for the phonological deficit: 1) limited phonological short-term memory; 2) degraded, under-specified or, conversely, overspecified phonological representations; 3) speech perception disorders. However, the degraded, under-specified phonological representation hypothesis that is basically referred to accounts for the dyslexics’ phonological deficit has been recently challenged: it has been suggested that the dyslexics’ phonological deficit relies on difficulties to store, access, and retrieve the phonological representations (e.g., Ahissar, 2007; Ramus & Szenkovits, 2008; Szenkovits & Ramus, 2005). To date, to reconcile both views, it has been proposed that the phonological deficit results in multi-dimensional difficulties that include difficulties to learn and manipulate the speech units as well as difficulties to store, access, and retrieve the phonological representations (e.g., Snowling, 2001; Ziegler, Castel, Pech-Geogel, George, Alario, & Perry, 2008). Despite this tentative proposal, there is no consensus. Here, I propose to draw an up-to-date portrait of an alternative option that has not been studied so far to disentangle whether another possible source of the phonological deficit in dyslexia may be envisaged: Are dyslexics sensitive to universal phonological knowledge?

In a non-negligible proportion, dyslexics' phonological deficit originates in impairments to process auditory information (i.e., ≈ 50%; Ramus et al., 2003).Typically, to account for the degraded nature of the phonological representations, it has been hypothesized that the dyslexics' perceptual system could not turn to be attuned to the native phonemic categories as shown with impairments in categorical perception (e.g., Adlard & Hazan, 1998;Mody, Studdert-Kennedy, & Brady, 1997;Veuillet, Magnan, Écalle, Thai-Van, & Collet, 2007).The categorical perception refers to the tendency to perceive a sound as a member of a category (e.g., /b/ or /p/).Thus, the variants of the same phoneme within a category are more likely perceived as being similar to each other compared to phonemes from other categories (i.e., /b h / is more likely judged as similar to /b/ than /p/ while /p h / is more likely judged as similar to /p/ than /b/.Scientifically-speaking, the categorical perception can be described as "the degree to which acoustic differences between variants of the same phoneme are less perceptible than differences of the same acoustic magnitude between two different phonemes" (Serniclaes et al., 2004, p. 337).Indeed, dyslexics have been shown to be impaired the processing of relevant acoustic-phonetic characteristics in their native language such as the voicing (e.g., /ba/ -/pa/; Bogliotti, Serniclaes, Messaoud-Galusi, & Sprenger-Charolles, 2008;Hoonhorst, Colin, Markessis, Radeau, Deltenre, & Serniclaes, 2009;Serniclaes, Sprenger-Charolles, Carré, & Démonet, 2001;Serniclaes, van Heghe, Mousty, Carré, & Sprenger-Charolles, 2004).Lower performances in between-categories perception but higher performances in within-categories perception compared to both chronological age-matched and reading level-matched controls have been interpreted as an allophonic mode of speech perception1 .In other words, dyslexics have difficulties to discriminate two phonemes that belong to two different categories as determined by the voicing (i.e., /ba/ vs. /pa/; low between-boundaries performance) whereas they can discriminate two variants of a same phoneme even if one of the variant does not exist in the native language (e.g., /p/ and /p h /; high within-boundaries performance).Hence, dyslexics' phonological representations would be over-specified since dyslexics would maintain acoustic-phonetic contrasts that are irrelevant in their native language and should be deactivated early in life (e.g., Saffran, Werker, & Werner, 2006;Werker & Tees, 1984).To be unable to discriminate relevant acoustic-phonetic duration-based contrasts in their native language (i.e., voicing; e.g., /b/ vs. /p/) would induce degraded, under-specified phonological representations and subsequent difficulties to use grapheme-to-phoneme correspondences (e.g., Bogliotti et al., 2008;Serniclaes et al., 2004).Alternatively, the phonological deficit could stem from difficulties in the time-course aspects of pre-lexical phonetic-phonological processing rather than from impaired phonological-lexical representations (e.g., Blomert, Mitterer, & Paffen, 2004;Nittrouer, 1999).
To determine whether the dyslexic's perceptual system is tuned to process finely-sharpened universal phonological representations (i.e., sound sequences that respect or not the universal phonological well-formedness), I here envisage the universal phonological sonority-related markedness to provide further arguments on the origin of the dyslexics' phonological deficit: universal or language-dependent and degraded/under-specified phonological representations or difficulties to access them?

An unexplored alternative
As hypothesized within the Optimality Theory framework (Prince & Smolensky, 1993;1997;2004), sound sequences that are phonotactically-illegal clusters such as /ʁb/ are more likely rejected compared to phonotactically-legal clusters such as /bʁ/ since all speakers are supposed to have universal phonological knowledge on grammatical restrictions irrespective to their (acoustic-)phonetic properties and phonotactic probabilities.However, whether dyslexics have universal phonological knowledge on grammatical restrictions remain unexplored.

www.intechopen.com
Dyslexia -A Comprehensive and International Approach 50

Phonological markedness and sonority profile
Phonotactic restrictions straightforwardly rule how sound sequences co-occur.It has been shown that sound sequences depend on the sonority of phonemes (e.g., Clements, 1990).Sonority is a scalar acoustic-phonetic property that refers to the sound's "[…] loudness relative to that of other sounds with the same length, stress, and pitch" (Ladefoged, 1975, p. 221).Under this definition, Fig. 1 presents that sonority hierarchically ranks consonants from the high-sonority phonemes (i.e., from liquid to nasal) to low-sonority ones (i.e., from fricative, /f/, /z/, /ʃ/… to occlusive, /b/, /t/, /g/…).Also, the linguistic structures are supposed to conform to a sonority-based organization as proposed by the sonority sequencing principle (e.g., Clements, 1990;Selkirk, 1984): syllables favor a structure with an onset maximally growing in sonority towards the vowel and falling minimally to the coda.Hence, universally-optimal CV syllables that bear high-sonority onsets (e.g., /la/) tend to be avoided in the phonotactics of languages to favor low-sonority ones (e.g., /ta/) whereas, in syllables that do contain a coda, high-sonority codas (e.g., /al/) tend to be preferred to lowsonority ones (e.g., /at/; see Selkirk, 1984).Using a sonority-based distribution of syllables which combines the sonority and the sonority sequencing principle, it is possible to assess the universal phonological knowledge on grammatical restrictions.Fig. 1.Sonority scale adapted from Clements (1990) and Selkirk (1984).

The present study
As I mentioned above, there is plenty of work to refine our understanding of where the phonological deficit comes from.Does the phonological deficit arise from degraded, underspecified phonological representations?If the phonological representations are intact, do dyslexic children have intact universal phonological representations?To provide innovative arguments in speech perception in dyslexia, I designed a preliminary syllable count task to pit the universal phonological knowledge on grammatical restrictions in French dyslexic children.I tested the (mis)perception of marked, grammatically ill-formed unattested onset clusters in French dyslexic compared to chronological age-matched controls and reading level-matched controls.Children were aurally-administered monosyllabic C 1 C 2 VC 3 pseudowords (e.g., /pkal/) and their disyllabic C 1 uC 2 VC 3 counterparts (e.g., /pukal/).All C 1 C 2 clusters within monosyllabic pseudowords were constructed by splicing out the /u/.Onset clusters (C 1 C 2 ) were classified as high-fall, low-fall, plateau, low-rise or high-rise.
Given the markedness constraints (i.e., avoid marked, grammatically ill-formed outputs such as /ʁb/) and the faithfulness constraints (i.e., map the input /ʁb/ to the output /ʁb/), the misperception of C 1 C 2 clusters should increase as markedness increases.Hence, if perceptual confusion depends on universal markedness-related knowledge as determined by sonority profiles, /gmal/ (high-rise SP, the most marked) should be more misperceived as disyllabic than /pkal/ (plateau SP), which in turn, should be more misperceived than /ʁbal/ (high-fall SP, the least marked) in both chronological age-matched and reading levelmatched controls.However, since dyslexics are supposed to have degraded, underspecified phonological representations, phonological sonority-related markedness effects and phonological repair with an illusory epenthetic vowel should not be observed.

Participants
Five French dyslexic children with no comorbid attention deficit hyperactivity disorder (ADHD) were tested in this experiment.Dyslexic children were compared to five chronological age-matched controls and five reading level-matched controls.Control children were recruited from an urban public elementary school.All children were tested after parents returned a consent form.Dyslexic children were diagnosed as dyslexics around two years prior this experiment (M = 29 months; SD = 4 months) by a speech and language therapist.All children were French native speakers with no second language learning, middle class, and right-handed 2 .They reported no hearing disorders.Reading level and IQs were assessed prior to the experiment.Student t tests confirmed that verbal and performance IQs significantly differed between dyslexics and chronological age-matched controls, t(8) = -3.96,p < .005,t(8) = 3.10, p < .02respectively; they also differed on reading level, t(8) = 9.09, p < .0001,but did not differ on chronological age, p > .1.Chronological age significantly differed between dyslexic children and reading level-matched controls, t(8) = 8.71, p < .0001;neither reading level nor verbal and performance IQs significantly differed, p > .1.Chronological age as well as reading level and verbal IQ significantly differed between chronological age-matched and reading level-matched controls, t(8) = 8.92, p < .0001,t(8) = 10.56,p < .0001,t(8) = 2.33, p < .05,respectively.Difference was marginally significant for the performance IQ, t(8) = 2.01, p < .08.Our research was approved by the Regional School Management Office.Profiles are presented in Table 1 3 .Table 1.Chronological and reading level ages, range, verbal and performance IQs for dyslexic children, chronological age-matched, and reading level-matched controls.

Stimuli
Forty stimuli were selected.They were twenty monosyllabic C 1 C 2 VC 3 pseudowords and their disyllabic C 1 uC 2 VC 3 counterparts, which shared their VC 3 rhyme (i.e., /al/) but differed on the structure of their C 1 C 2 clusters (Table 2).Onset clusters were unattested in French.I subdivided them into five sonority profiles (SPs) as follows: high-fall (e.g., /ʁbal/), low-fall (e.g., /fkal/), plateau (e.g., /pkal/), low-rise (e.g., /kfal/), and high-rise (e.g., /zʁal/).Onset cluster markedness progresses from high-fall SPs (the most marked, the grammatically worst ill-formed) to high-rise SPs (the least marked, the grammatically most well-formed).Each SP contained four different C 1 C 2 clusters, repeated eight times within each SP; overall, there were 4 C 1 C 2 x 5 SPs x 8 repetitions x 2 conditions (mono-and disyllabic pseudowords) = 320 stimuli.To exclude some possible phonological biases such as compensation for assimilation or coarticulation, I did not include homorganic consonants (i.e., consonants that share the same place of articulation) and consonants that differ in voicing within C 1 C 2 onset clusters.However, C 1 and C 2 could differ in mode of articulation.Disyllabic C 1 uC 2 VC 3 counterparts were recorded by a female native speaker of French.All sounds were digitally recorded with a Sennheiser e865s microphone through a Tascam US-144MK II external audio interface, sampled at a 44 kHz rate, converted with a 16-bit resolution, and bandpass filtered (0 Hz to 5,000 Hz).C 1 u first syllable in disyllabic pseudowords systematically carried stress.Monosyllabic C 1 C 2 VC 3 pseudowords were obtained by splicing out step-by-step the vowel /u/ with Praat software (Boersma & Weenink, 2011).Visual and auditory inspection of the waveforms minimized the /u/ coarticulation-based traces in the C 1 and C 2 .Mean duration was 197.3 ms (SD = 16.1) for the C 1 C 2 clusters and 79.8 ms (SD = 11.2) for the vowel /u/.
Table 2. Monosyllabic pseudowords used as a function of sonority profiles.

Procedure
This experiment was designed, compiled and run using E-Prime 2.0 Professional software (Schneider, Eschman, & Zuccolotto, 2002) on Sony X-series laptop computers under Windows 7 OS.Children wore Sennheiser HD 25-1 II headphones (16 Hz-22 kHz range, 70 Ω impedance) and were presented pseudowords binaurally at 70 dB SPL.Trials consisted in the presentation of a vertically-centered exclamation mark (i.e., '+') for 500 ms, followed after a 200-ms blank screen by a pseudoword.A 1,000-ms delay separated two consecutive trials.Children were requested to decide as quickly and as accurately as possible whether the pseudoword had one or two syllables (numpad 1 = one syllable, numpad 2 = two syllables).Children were first trained with a practice list of 16 trials with corrective feedback.No feedback was given for the experimental trials.Trials were randomized.The software automatically recorded response times and response accuracy.

Results
I report first the results from two 5 x 2 x 3 mixed-design repeated measures ANOVAs with Statistica software by subject (F1) and by item (F2) on response times and response accuracy (~ 84.1% of the data).ANOVAs were run with Group (dyslexics vs. chronological agematched controls vs. reading level-matched controls) as between-subject factor and Sonority profile (high-fall vs. low-fall vs. plateau vs. low-rise vs. high-rise) and Syllable structure (monosyllabic vs. disyllabic) as within-subject factors.
Neither the Group nor the Syllable structure main effects were significant in response accuracy.The three-way Sonority profile x Syllable structure x Group interaction did not significantly interact in response times, Fs < 1, p > .1 and response accuracy, Fs < 1, p > .1.To ensure that the perceptual confusion response patterns are not due to coarticulationbased artifacts relative to traces of spliced /u/ from the C 1 uC 2 clusters, I examined the nature of the misperception a posteriori.Dyslexic children as well as controls were posttested.Children were asked to report whether or not they heard a vowel, and if so, which one, within monosyllabic pseudowords (n = 160).The task was quite similar, except that for each error, a visual feedback was displayed and children were therefore asked to press on the vowel they thought they heard (i.e., /a/, /i/, /u/, /o/, /e/, /ɛ/, /y/, /ə/, or not a vowel).Response patterns showed that when French dyslexic children misperceived the C 1 C 2 clusters, they reported an epenthetic /ə/ (M = 80.0 ± 4.4) more frequently than other vowels (M = 3.5 ± 4.7), t(4) = 24.69,p <.0001).Response patterns were similar in chronological agematched controls (M = 83.9± 5.5 vs. M = 5.8 ± 2.9, t(4) = 18.37, p < .0001)and in reading levelmatched controls (M = 81.7 ± 6.2 vs. M = 2.6 ± 3.8, t(4) = 27.00,p < .0001).As in the Berent et al.'s studies (2007;2008), I submitted children' response accuracy to the C 1 C 2 clusters to a linear hierarchically-forced stepwise regression analysis4 I first forced in the C 1 C 2 cluster length (in ms); then, I forced in the statistical properties of biphones and triphones respectively (I considered C 1 VC 2 triphones with a vowel /ə/ that was the most reported epenthetic vowel in children), the bigram frequency (Peereman, Lété, & Sprenger-Charolles, 2007), and the phonotactic transitional probabilities (Crouzet, 2000).The analysis revealed that markedness, which was entered last, accounts for significant unique variance in dyslexic children (Adjusted R 2 = .276,p < .0001,β = .62),,chronological age-matched controls (Adjusted R 2 = .258,p < .005,β = .55)and reading level-matched controls (Adjusted R 2 = .394,p < .0001,β = .76).

Discussion
As can be seen throughout this chapter, the dyslexics' phonological deficit has unresolved issues.However, the degraded, under-specified phonological representation hypothesis as a failure in the perception of finely-sharped acoustic-phonetic cues appears to be somehow misleading (e.g., Ramus & Szenkovits, 2008).To solve the intricate problem of the nature of the dyslexics' phonological deficit, I tried to assess whether -and how-the phonological representations are difficult to be accessed, either language-specific or universal, in French dyslexic children compared to chronological age-matched and reading level-matched controls.
The results provide major, innovative responses to a twofold debate: about the nature of the phonological deficit in dyslexics and about the universal phonological knowledge on grammatical restrictions.Crucially, I first observed that the (mis)perception of unattested onset clusters relies on universal sonority-related phonological knowledge on grammatical restrictions.Indeed, response patterns indicate a markedness-modulated misperception of monosyllabic pseudowords as disyllabic ones: as markedness increased from high-rise SP to high-fall SP, perceptual confusion was prone to increase.Also, response patterns were reversed to their disyllabic counterparts: as markedness increased, perceptual confusion decreased.Furthermore, there was no speed-accuracy trade-off: as response accuracy increased, response times decreased.
A posteriori measures confirmed that monosyllabic pseudowords were not perceptuallyconfused due to coarticulation-based artefacts relative to traces of the spliced vowel /u/: monosyllabic pseudowords are more likely phonologically-repaired with an illusory epenthetic vowel /ə/.Since the vowel /ə/ represents a high-frequency vowel in French, a linear hierarchically-forced stepwise regression analysis discarded a straightforward influence of statistical properties and acoustic-phonetic cues on the misperception and the phonological repair by an illusory epenthetic vowel.Neither the C 1 C 2 cluster length, nor the frequency of biphones and triphones explain our results: sonority-related markedness accounts for significant unique variance.Surprisingly, Group effects were absent; French dyslexic children were as sensitive as both chronological age-matched and reading level-matched controls to the phonological sonorityrelated markedness of C 1 C 2 onset clusters and, as well as controls, they phonologically repaired unattested marked C 1 C 2 clusters into attested unmarked ones with an epenthetic /ə/ vowel: this is in accordance with recent results of Maïonchi-Pino, Yokoyama, Takahashi, Écalle, Magnan, & Kawashima (2011) in French adult native speakers (in English, also see Berent et al., 2007;2008).Of interest, dyslexic children did not differ from both control groups on their response accuracy and discrimination sensitivity threshold (d'); however, response times were slower.This suggests that dyslexic children have normal, intact universal phonological constraints and robust phonological representations of their native language; they are able to efficiently recode grammatical ill-formed sequences (i.e., to do that, children insert an epenthetic vowel /ə/ that tends to restore an attested, grammatical www.intechopen.comPhonological Restriction Knowledge in Dyslexia: Universal or Language-Specific? 57 well-formed, phonological sequence) and universal phonological representations to avoid a transgression of grammatical well-formedness of phonological sequence.Thus, the children' misperception of marked onset clusters could be attributed to universal phonologicallyconstrained preferences that follow sonority-related markedness constraints.Since sonorityrelated markedness relies on acoustic-phonetic cues that might require efficient abilities to perceive, store and process brief acoustic-phonetic information (e.g., Hayes & Steriade, 2004; for counter-argument on the phonetic basis of sonority, see Clements, 2006), and since dyslexic children are as sensitive as controls to this phonological marker, our results compete to reconsider the degraded, under-specified phonological representation hypothesis to further explore the phonological access deficit hypothesis (e.g., Ramus & Szenkovits, 2008).

Conclusion
Dyslexic children therefore have intact universal phonological sonority-related sensitivity and efficient language-dependent abilities to underlie both the (mis)perception of phonotactically-illegal clusters and the phonological repair processes, respectively.Further, acoustic-phonetic cues as well as statistical properties do not exhibit straightforward influence, but I do not discard that both contribute to the markedness-related misperception.Further, in our experiment, it remains unresolved whether Peperkamp's position (2007, p. 634-635) is true: "the role of the grammar in phonological perception is not to repair phonologically illegal structure but rather to undo the effect of native phonological processes, and that perceptual repairs take place at a lower, phonetic, processing level".Although I acknowledge that extensive research is important to refine our results, I point out that, as suggested by Ramus & Szenkovits (2008) or Szenkovits et al. (submitted) dyslexics' phonological deficit accommodates with a deficit in storing and accessing the phonological representations.

Fig. 2 .
Fig. 2. Mean response times (in ms) to the Sonority profile x Syllable structure interaction for the dyslexic children (DY), chronological age-matched controls (CA) and reading levelmatched controls (RL).

Fig. 3 .
Fig. 3. Mean response accuracy (in %) to the Sonority profile x Syllable structure interaction for the dyslexic children (DY), chronological age-matched controls (CA) and reading levelmatched controls (RL).