Identifying the deficits responsible for developmental dyslexia has been a central objective of reading research over the past decades. This research has pointed to a causal role of deficits in the phonological skills required for reading acquisition (for a review, see Vellutino, Fletcher, Snowling, & Scanlon, Reference Vellutino, Fletcher, Snowling and Scanlon2004). Dyslexic children perform poorly in comparison to reading level (RL) controls in tasks evaluating phonemic and phonological awareness,Footnote 1 phonological short-term memory, rapid access to phonological forms in the mental lexicon, and phoneme–grapheme conversion (for a review, see Snowling, Reference Snowling2001). Alternative explanations to the phonological deficit hypothesis have been proposed, implying the causal role of auditory (Tallal, Reference Tallal1980; Temple et al., Reference Temple, Poldrack, Protopapas, Nagarajan, Salz and Tallal2000), visual/magnocellular (Stein, Reference Stein2001) or cerebellar deficits (Nicolson, Fawcett, & Dean, Reference Nicolson, Fawcett and Dean2001). There is nonetheless considerable agreement that phonological deficits constitute a primary cause of developmental dyslexia, and these other deficits are considered by most as being associated (Sprenger-Charolles, Colé, & Serniclaes, Reference Sprenger-Charolles, Colé and Serniclaes2006).
Deficits in phonemic awareness also characterize dyslexic adults (Ben-Dror, Pollatsek, & Scarpati, Reference Ben-Dror, Pollatsek and Scarpati1991; Bruck, Reference Bruck1990, Reference Bruck1993; Chiappe, Stringer, Siegel, & Stanovich, Reference Chiappe, Stringer, Siegel and Stanovich2002; Pennington, van Orden, Smith, Green, & Haith, Reference Pennington, van Orden, Smith, Green and Haith1990). Phonemic awareness in dyslexic readers does not appear to evolve across their lives. For example, Bruck (Reference Bruck1992) demonstrated that dyslexic readers do not develop phonemic awareness in accordance with either their chronological age (CA) or their RL. Dyslexic readers apparently reach a level of phonemic awareness beyond which they can no longer improve (Miller-Shaul, Reference Miller-Shaul2005). In sum, phonological deficits are persistent and constitute the core characteristic of developmental dyslexia. However, the expression of the phonological awareness deficit may vary across orthographies (Vellutino et al., Reference Vellutino, Fletcher, Snowling and Scanlon2004; Ziegler & Goswami, Reference Ziegler and Goswami2005). While English dyslexic adults are outperformed by both CA and RL controls in phonological awareness tasks (Bruck, Reference Bruck1992, Reference Bruck1993; Chiappe et al., Reference Chiappe, Stringer, Siegel and Stanovich2002; Pennington et al., Reference Pennington, van Orden, Smith, Green and Haith1990), French adult dyslexics are outperformed only by CA controls but not by RL controls (Martin et al., Reference Martin, Colé, Leuwers, Casalis, Zorman and Sprenger-Charolles2010).
MORPHOLOGICAL AWARENESS IN DYSLEXIC CHILDREN AND ADULTS
Whereas the role of phonemic awareness in reading development and in dyslexia has been clearly demonstrated, the role of another metalinguistic ability, that is, morphological awareness, is less well documented and understood. Morphological awareness is defined as the capacity to reflect on and explicitly manipulate orally the morphological structure of words (Carlisle, Reference Carlisle and Feldman1995). Because morphological awareness plays a role in reading acquisition (see, e.g., Carlisle, Reference Carlisle and Feldman1995; Roman, Kirby, Parrila, Wade-Woolley, & Deacon, Reference Roman, Kirby, Parrila, Wade-Woolley and Deacon2009; Wolter, Wood, & D'Zatko, Reference Wolter, Wood and D'Zatko2009), it is important to gain knowledge about its development in the dyslexic population. The goal of the present research is to evaluate morphological and phonological awareness in dyslexic university students with the objective of determining the level of development of these reading-related skills in this population. To situate the metalinguistic abilities of the dyslexic university students properly, we compare their performances with those of appropriate CA and RL controls. As proposed by Bryant and Bradley (Reference Bryant and Bradley1985), CA controls allow the performances of the dyslexic readers to be situated with respect to the RL normally expected on the basis of CA, and RL controls on the basis of RL. These comparisons have often been used in the study of dyslexic adults (e.g., Ben-Dror et al., Reference Ben-Dror, Pollatsek and Scarpati1991; Bruck, Reference Bruck1990; Chiappe et al., Reference Chiappe, Stringer, Siegel and Stanovich2002; Martin et al., Reference Martin, Colé, Leuwers, Casalis, Zorman and Sprenger-Charolles2010; Pennington et al., Reference Pennington, van Orden, Smith, Green and Haith1990). Moreover, our study, like others in the literature, focused on derivational morphology, because this type of morphology plays a central role in word formation and visual word recognition (see, e.g., Rastle, Davis, Marslen-Wilson & Tyler, Reference Rastle, Davis, Marslen-Wilson and Tyler2000). Because our study was conducted on a population reading a language with an alphabetic system (French), our literature review will only include studies concerned with derivational morphology in alphabetic systems (see Muter, Hulme, Snowling, & Stevenson, Reference Muter, Hulme, Snowling and Stevenson2004, for a review on inflectional morphology).
Morphological awareness has been shown to be important in typical reading development and literacy acquisition (Carlisle, Reference Carlisle and Feldman1995; Casalis & Louis-Alexandre, Reference Casalis and Louis-Alexandre2000; Mahony, Singson, & Mann, Reference Mahony, Singson and Mann2000; Roman et al., Reference Roman, Kirby, Parrila, Wade-Woolley and Deacon2009; Wolter et al., Reference Wolter, Wood and D'Zatko2009). Morphological awareness contributes to decoding skills, word recognition skills, and reading comprehension independently of phonological skills (Carlisle, Reference Carlisle and Feldman1995; Casalis & Louis-Alexandre, Reference Casalis and Louis-Alexandre2000; Deacon & Kirby, Reference Deacon and Kirby2004; Mahony et al., Reference Mahony, Singson and Mann2000; Nagy, Berninger, & Abbott, Reference Nagy, Berninger and Abbott2006; Roman et al., Reference Roman, Kirby, Parrila, Wade-Woolley and Deacon2009; Singson, Mahony, & Mann, Reference Singson, Mahony and Mann2000) and vocabulary (Casalis & Louis-Alexandre, Reference Casalis and Louis-Alexandre2000; Nagy et al., Reference Nagy, Berninger and Abbott2006; Singson et al., Reference Singson, Mahony and Mann2000). Some English regression studies have found that the contribution of morphological awareness to predicting word reading ability increases across the school years (Carlisle, Reference Carlisle and Feldman1995; Singson et al., Reference Singson, Mahony and Mann2000), while that of phonemic awareness decreases (Singson et al., Reference Singson, Mahony and Mann2000). However, more recently Roman et al. (Reference Roman, Kirby, Parrila, Wade-Woolley and Deacon2009) did not replicate this dissociation between phonological and morphological awareness with English-speaking children in Grades 4, 6, and 8. Rather, they observed a constant and specific contribution of both abilities across these grades.
Although studies of the development of morphological awareness in dyslexics are relatively rare, there are some findings that have compared the morphological awareness of dyslexic children with that of CA and RL controls. Dyslexic children have been shown to perform less well than CA controls on morphological awareness tasks involving derivational morphology, both in French (Casalis, Colé, & Sopo, Reference Casalis, Colé and Sopo2004; Casalis, Mathiot, Bécavin, & Colé, Reference Casalis, Mathiot, Bécavin and Colé2003) and in English (Champion, Reference Champion1997; Tsesmeli & Seymour, Reference Tsesmeli and Seymour2006). However, these studies also reported that dyslexic children performed equally well or even better than the RL controls. Taken together, these results indicate that, contrary to phonological awareness, dyslexic children develop morphological awareness to at least the level expected from their RL.
Data on the morphological awareness of dyslexic adults are even scarcer despite their obvious utility for understanding its development across the life span of dyslexic readers. Only a few studies of morphological awareness in dyslexic university students or adults in alphabetic languages have been conducted, and these have always been in English. Tractenberg (Reference Tractenberg2002) found that reading-disabled university students performed worse than typical university students in morphological awareness tasks. Leong (Reference Leong1999) found that learning-disabled university students performed similarly to RL controls in terms of processing speed, but they were more accurate. Finally Rubin, Patterson, and Kantor (Reference Rubin, Patterson and Kantor1991) found that English-speaking adults with literacy problems performed similarly to 7-year-old normal readers in morphological tasks. However, there are some shortcomings in these studies. In Tractenberg's (Reference Tractenberg2002) study, the dyslexic university students and the typically reading university students were not matched by CA. In Leong's (Reference Leong1999) and Rubin et al.'s (Reference Rubin, Patterson and Kantor1991) studies, the experimental groups, labelled “learning-disabled” and “poor readers,” may have included participants who suffered from other deficits than developmental dyslexia. In addition, the tasks used by Leong and Tractenberg required participants to read, which could have affected negatively the performance of dyslexic readers. In these tasks, the reading difficulties of the dyslexic readers in Leong (Reference Leong1999) probably slowed down their stimulus identification and decreased the validity of the comparison of the level of morphological awareness with that of the control groups.
In sum, dyslexic children do not develop phonemic awareness to a level that might be expected on the basis of their RL. This deficit, which is assumed to be causally linked to their reading deficits, persists into adulthood. In contrast, dyslexic children develop morphological awareness to a level that is lower than that of their CA but that appears to correspond at least to their RL. These difficulties are assumed not to be responsible for their reading deficits. In dyslexic adults, the level of morphological awareness was seldom assessed, and the existing studies had methodological flaws that did not allow clear conclusions to be drawn. Thus, the question of whether morphological awareness difficulties persist into adulthood is still unanswered, and it should be addressed using a carefully designed methodology.
RELATION BETWEEN PHONOLOGICAL AND MORPHOLOGICAL AWARENESS
The relation between phonological and morphological awareness has been examined by several researchers, but no clear picture has yet emerged. In typically developing readers, performances in morphological and phonological awareness tasks are often correlated (Carlisle, Reference Carlisle and Feldman1995; Casalis & Louis-Alexandre, Reference Casalis and Louis-Alexandre2000; Colé, Royer, Leuwers, & Casalis, Reference Colé, Royer, Leuwers and Casalis2004; Roman et al., Reference Roman, Kirby, Parrila, Wade-Woolley and Deacon2009). For example, Carlisle and Nomanbhoy (Reference Carlisle and Nomanbhoy1993) showed that the performance of first graders in a phonological awareness task was correlated with that in morphological awareness tasks, suggesting that the developmental trajectory of the two abilities is related (or that they rely on some common processes). However, as we have stated above, morphological and phonological awareness generally contribute independently to reading ability. Thus, Casalis and Colé (Reference Casalis and Colé2009) showed in a training study that while some processes are shared by both metalinguistic abilities, each appears to have its own specificity and may have partially independent developmental trajectories. It has been shown that the overall developmental trajectory of morphological awareness has a much longer span than that of phonological awareness (Berninger, Abbott, Nagy, & Carlisle, Reference Berninger, Abbott, Nagy and Carlisle2010).
Differences between phonological and morphological awareness were evidenced with dyslexic readers too. Dyslexic children who differ in their phonological skills do not differ in their morphological awareness skills, as long as phonological processing required in the morphological task is not too great (Casalis et al., Reference Casalis, Colé and Sopo2004). Elbro and Arnbak (Reference Elbro and Arnbak1996) also showed that after specific training on morphological awareness, dyslexic children improved more than controls on morphological awareness tasks but not on phonemic awareness tasks.
PRESENT STUDY
In this study of dyslexic university students, we expect the levels of phonological awareness and morphological awareness to differ. We expect morphological awareness to develop further as compared with phonological awareness, whose development is curtailed (Bruck, Reference Bruck1992; Miller-Shaul, Reference Miller-Shaul2005) in the dyslexic population. An explanation for these different developmental trajectories for the two kinds of metalinguistic abilities in our population may well lie in the semantic dimension of morphology. Morphemes convey meaning, and dyslexic readers do not suffer from semantic deficits (Vellutino, Scanlon, & Spearing, Reference Vellutino, Scanlon and Spearing1995). Dyslexic readers rely heavily on semantics when available, as has been demonstrated in word identification studies with dyslexic university students (Bruck, Reference Bruck1990). Thus, dyslexic readers may depend on the semantic dimension of morphology to develop their morphological awareness, which is obviously not possible for the development of phonological awareness. This hypothesis is further supported by a longitudinal study showing that derivational morphological awareness continues to develop after the fourth grade in typically developing readers (Berninger et al., Reference Berninger, Abbott, Nagy and Carlisle2010). Consequently, one can expect that dyslexic adults will show higher levels of morphological awareness compared with phonological awareness.
To test this hypothesis, we assessed the levels of phonological and morphological awareness in dyslexic adults with a high level of academic achievement. The dyslexic adults involved in our study were university students or had attended a university in the past. University students generally have considerable exposure to print and have read throughout their lives, which is likely to enhance the development of their metalinguistic skills. Moreover, several authors have suggested that dyslexic university students may have developed compensatory mechanisms to be able to pursue university studies despite their persistent reading difficulties. In particular, Elbro and Arnbak (Reference Elbro and Arnbak1996) proposed that dyslexic readers may rely on their morphological skills to compensate for their reading difficulties. Dyslexic university students thus constitute a privileged population for identifying to which extent phonological and especially morphological awareness can develop in readers with dyslexia.
We compared the performance of these dyslexic university students with that of both CA and RL controls. The use of these two control groups allowed us to situate precisely the performance of the dyslexic students with respect to normal reading development and CA. Our study was conducted in French, a language for which the prevalence of dyslexia is not significantly different from that of English-speaking countries (i.e., 5% to 10%; INSERM, 2007).
Comparisons of the phonological awareness of dyslexic university students with that of both CA and RL controls have already been made in French (Martin et al., Reference Martin, Colé, Leuwers, Casalis, Zorman and Sprenger-Charolles2010). Because our phonological awareness tasks and group characteristics are largely comparable to those in Martin et al. (Reference Martin, Colé, Leuwers, Casalis, Zorman and Sprenger-Charolles2010), we expect to replicate their results. That is, we expect dyslexic university students to exhibit poorer phonological awareness skills in comparison with their CA controls but to perform at the level of their RL controls.
Comparisons of the morphological awareness of dyslexic university students with that of carefully defined CA and RL control groups will be made for the first time in this study. Our goal is to determine to what extent dyslexic university students have developed morphological awareness. Following Elbro and Arnbak (Reference Elbro and Arnbak1996), we expect dyslexic adults to show a more developed morphological awareness performance than predicted, based on their reading skills. Thus, we expect dyslexic university students to consistently outperform their RL controls in the morphological awareness tasks. On the comparison with CA controls, we can put forward two alternative hypotheses. First, we can assume that the semantic dimension of morphological knowledge leads dyslexic university students to develop their morphological awareness to a level equal to that of the CA controls. Second, the persistent phonological difficulties of dyslexics could limit their development of morphological awareness, so that dyslexic university students would be outperformed by their CA controls.
METHOD
Participants
Three groups of 44 participants took part in the present study: dyslexic university students, CA controls, and RL controls. All participants were native monolingual speakers of French and had a nonverbal IQ within the normal range (i.e., at least equal to or above the 10th percentile; Raven's matrices: Raven & Court, Reference Raven and Court1995; Raven's Coloured Progressive Matrices; Raven, Court, & Raven, Reference Raven, Court and Raven1998). The participants’ nonverbal IQ ranged from the 10th to above the 95th percentile.
Dyslexic university students were recruited at the University of Geneva by means of posters, in the Geneva and Lausanne areas (Switzerland) by means of an advertisement in a free local newspaper, at the University of Savoie (Chambéry, France) by a public e-mail sent to all students, and at the Health Center of the University of Grenoble (Grenoble, France) by means of a letter sent to their dyslexic students. To be included in the study, dyslexic participants had to be current or past university students. They also had to (a) be native, monolingual speakers of French; (b) lack any known neurological/psychiatric disorders and report normal or corrected to normal hearing or vision; and (c) have a nonverbal IQ within the normal range (Raven's matrices; Raven & Court, Reference Raven and Court1995). The majority of our participants had been diagnosed as dyslexic by a speech therapistFootnote 2 in their childhood. Those with no diagnosis by a speech therapistFootnote 3 at the time of our study (6 out of the 44 participants) had to score 2 SD below the control mean on a reading task. We relied on this criterion (see Szenkovits & Ramus, Reference Szenkovits and Ramus2005, with self-identified French dyslexic university students for a similar criterion) because no standardized reading test exists in French for adults older than 16 years. The dyslexic university students with no diagnosis by a speech therapist all reported major difficulties in learning to read in their childhood. Moreover, their reading age on the Alouette Test (see below) never exceeded 12 years and 2 months. Finally, on aloud reading of isolated pseudowords, the reading latencies of all these participants were 2 SD below the control mean.
Reading scores were obtained using the Alouette Test (Lefavrais, Reference Lefavrais1967),Footnote 4 a standardized French reading test employed in the study of developmental dyslexia in French (e.g., Bogliotti, Serniclaes, Messaoud-Galusi, & Sprenger-Charolles, Reference Bogliotti, Serniclaes, Messaoud-Galusi and Sprenger-Charolles2008; Casalis et al., Reference Casalis, Colé and Sopo2004; Martin et al., Reference Martin, Colé, Leuwers, Casalis, Zorman and Sprenger-Charolles2010; Sprenger-Charolles, Colé, Kipffer-Piquard, Pinton, & Billard, Reference Sprenger-Charolles, Colé, Kipffer-Piquard, Pinton and Billard2009; Ziegler et al., Reference Ziegler, Castel, Pech-Georgel, George, Alario and Perry2008; Ziegler, Pech-Georgel, George, & Lorenzi, Reference Ziegler, Pech-Georgel, George and Lorenzi2009). This test requires participants to read aloud a meaningless text within a 3-min time limit. The text consists of real words in meaningless but grammatically correct sentences. Each participant's performance is converted into a reading age according to a standard procedure taking into account both the time taken to read and the number of errors. The Alouette test is standardized for the reading performance of children aged from 5 to 14, so the highest possible reading age is 14 (which was reached by the CA controls). So as to avoid ceiling effects, we calculated a composite score (called the “reading score”) that took both accuracy and speed into account. To do so, for each participant, we calculated a z score for speed and a z score for accuracy, with CA controls as the reference group, and we then averaged these z scores. If the participant did not finish reading the text within the 3-min time limit, his/her raw performance in terms of speed and accuracy was adjusted to the whole text on the basis of the performance for the number of words read. Finally, the z scores signs were adjusted so that negative z scores correspond to a bad performance (higher number of errors or longer reading time).
The dyslexic university students were matched with two control groups, one of the same CA and the other of the same RL. In both control groups, the participants had normal literacy skills and no previous history of any learning disability. In addition, dyslexic participants were matched pairwise with both control groups by gender and country of education. All three groups were composed of 44 participants each (25 females, 19 males), of which 13 were educated in Switzerland and 31 in France. The CA controls were recruited from the university student bodies of the University of Geneva, the University of Savoie, and the University of Provence. The RL controls were recruited from elementary schools and middle schools in Geneva, Annemasse (France), and in the Academy of Aix-en-Provence (France).
The dyslexic participants were between the ages of 18 and 47 (M = 25.56, SD = 6.91). They all reported histories of reading difficulties, and their mean reading score was –3.38 (SD = 2.12), which corresponds to a mean reading age of 11.3 years. Twenty-seven had received a diagnosis of dyslexia during childhood and 11 after the age of 14.
The age of the CA controls ranged from 18 to 48 years (M = 25.56, SD = 6.64), and their mean reading score was 0.00 (SD =0.70). The mean reading score of the dyslexic participants differed significantly from that of CA controls, t (86) = 10.03, p < .001, whereas the CA of the two groups did not, t (86) < 1, ns.
The age of the RL controls ranged from 7 to 15 years (M = 11.40 years, SD = 1.75). Their mean reading score was –3.48 (SD = 2.49), which did not differ significantly from the dyslexic participants’ score, t (86) < 1, ns, whereas their CA did, t (86) = 13.18, p < .001.
Experimental tasks
We used three existing tasks to test phonological awareness, and we developed three new oral tasks to evaluate morphological awareness.
Tests of phonological awareness
Three tests were used to assess phonological awareness: the trisyllabic (TRISYL) test, the consonant–vowel–consonant (CVC) test, and the consonant–consonant–vowel (CCV) test. These were taken from EVALEC, a computerized battery of tests of reading and reading-related skills for French elementary school children (Sprenger-Charolles, Colé, Béchennec, & Kipffer-Piquard, Reference Sprenger-Charolles, Colé, Béchennec and Kipffer-Piquard2005). All involved the deletion of the first part (syllable or phoneme) of a pseudoword and measured both accuracy and speed.
TRISYL TEST
In this test participants had to delete the first syllable of a trisyllabic pseudoword. This test includes 10 items with a simple syllabic structure (consonant–vowel; e.g., povidu). Items were prerecorded in order to ensure identical stimuli and to avoid lip-reading. They were presented to the participant through headphones. The test was computer driven, and each trial was carried out as follows: the experimenter clicked on the mouse to trigger the item presentation. As soon as the participant gave his/her answer, the experimenter clicked on the mouse again to stop the chronometer and measure response times. Before the experiment began, participants received one example and one training trial with feedback.
CVC TEST
In this second test participants had to delete the first phoneme of a pseudoword composed of three phonemes. This test included 12 items with a CVC structure (e.g., puf). The procedure was identical to the TRISYL test.
CCV TEST
This test was similar in all respects to the CVC test, except that the 12 pseudowords had a CCV structure (e.g., klo).
Tests of morphological awareness
We developed three computer-driven oral tasks to evaluate the morphological awareness of our participants: the suffixation decision task, the prefixed word detection task, and the suffixed word detection task. These tasks tested explicit morphological awarenessFootnote 5 and measured the participants'performance in terms of accuracy and also in terms of speed for the suffixation decision task. They were designed to meet several criteria. First, we limited the phonological difficulty of these tasks, which is known to affect the dyslexics’ performance in morphological awareness tasks (Casalis et al., Reference Casalis, Colé and Sopo2004; Fowler & Liberman, Reference Fowler, Liberman and Feldman1995; Shankweiler et al., Reference Shankweiler, Crain, Katz, Fowler, Liberman and Brady1995). To do so, we limited to three the number of words the participants had to remember to perform the task. Second, we ensured that the participants’ response could not be based exclusively on the form properties of the words. To do so, all the monomorphemic items we used were pseudoaffixed. Pseudoaffixed words contain affixlike sequences that are not actually suffixes, as in the case of pseudosuffixed words (e.g., mistral), or not prefixes, as in the case of pseudoprefixed words (e.g., endive).
The three morphological awareness tasks were always presented in the same order: first, the suffixation decision task; second, the prefixed word detection task; and third, the suffixed word detection task. Before testing the participants, the experimenter informed them of the definition of suffixes and prefixes and trained them on this distinction. No item was used in both the training and the experimental tasks.
MORPHOLOGICAL TRAINING
At the beginning, the experimenter gave a definition of an affix. He/she then gave an example of a suffixed and of a pseudosuffixed word, and explained each time why the word was suffixed or not. The same was done with a prefixed and a pseudoprefixed word. At the end of this training the experimenter asked the participant to decide whether a suffixed and a pseudosuffixed word were suffixed, and whether a prefixed and a pseudoprefixed word were prefixed. The experimenter gave feedback to the participants for each response, explaining if necessary why their answer was wrong. At the end of this training phase, the participants could start with the morphological awareness tasks.
SUFFIXATION DECISION TASK
In this task, participants made a speeded manual forced choice of words heard in isolation. In each trial, they had to decide whether the word they heard was suffixed or not. The material consisted of 24 bisyllabic and 24 trisyllabic words, half being suffixed (e.g., frontal) and half being pseudosuffixed (e.g., mistral) words. All the words were of low written (M = 2.08, SD = 2.16) and spoken frequency (M = 0.94, SD = 1.26) according to Lexique, a lexical database for adults in French (New, Pallier, Ferrand, & Matos, Reference New, Pallier, Ferrand and Matos2001). Suffixed and pseudosuffixed words were matched pairwise for word endings, number of syllables, and grammatical class, and did not differ in their oral frequency, t (46) < 1, ns, and written frequency, t (46) = 1.13, p > .20. Suffixed words had a root ranging from low to high frequency (range of occurrences per million: oral frequency = 2.46–570.67, written frequency = 0.88–105.61), with a greater majority of high-frequency roots (66% oral frequency and 92% written frequency). The frequency of the root was always higher than that of the derived form, in terms of either written or oral frequency. Because the pseudosuffixed words did not contain a root, it was not possible to match the suffixed and pseudosuffixed words based on this criteria.
Words were prerecorded and played through headphones. The task was computer driven, and the procedure was as follows: a cross appeared in the middle of the screen for 250 ms and then the participant heard the item. Participants were instructed to answer as accurately and as quickly as possible. Their response was given by pressing a green button if the item was suffixed and a red one if the item was not suffixed, the “suffix” reponse being given with the preferred hand. There was a 1100-ms intertrial interval separating the participant's response and the beginning of the next trial. The items were presented randomly, and both accuracy and reaction times were measured. The experiment was preceded by a training phase using eight items whose characteristics were similar to those of the target items.
PREFIXED WORD DETECTION TASK
In this task, participants heard triplets of words and had to indicate which of the three was prefixed. Twelve word triplets sharing the initial phonemes (e.g., endive, envol, and enclume) were selected, with 1 prefixed and 2 pseudoprefixed words. All the words were of low frequency according to both their spoken frequency (M = 2.08, SD = 2.25) and their written frequency (M = 3.18, SD = 2.76; New et al., Reference New, Pallier, Ferrand and Matos2001). Words within triplets were strictly matched by word onset, number of syllables, and grammatical class. Words were also matched by their spoken frequency, t (22) < 1, ns, for all comparisons; and written frequency, t (22) = 1.39, p > .15, for the comparison of the prefixed word and one pseudoprefixed word list, t (22) < 1, ns, for the remaining comparisons. Prefixed words had a root ranging from low to high frequency (range of occurences per million: oral frequency = 4.75–766.2; written frequency = 6.69–923.45), with a greater majority of high-frequency roots (83% in terms of oral frequency and 92% in terms of written frequency). The frequency of the root was always higher than that of the derived form, in terms of either written or oral frequency. Because the pseudoprefixed words did not contain a root, it was not possible to match the prefixed and pseudoprefixed words based on this criteria.
Words were prerecorded and played over headphones. The task was computer driven, and the procedure was as follows: participants first heard the three words in succession with a 750-ms pause between them. Immediately after the third word, a cross appeared in the middle of the screen. Participants were instructed to answer as quickly and as accurately as possible once the cross appeared on the screen. Participants responded by means of the response box, by pressing the left button to select the first word, the middle button to select the second word, and the right button to select the third word. The target prefixed words occupied each of the three positions in equal proportion. The order of words within the triplets was fixed across participants, but the triplets were presented randomly. There was a 1000-ms intertrial interval before the next trial began. The experiment was preceded by a training phase composed of three triplets of words whose characteristics closely matched those of the target item. Accuracy was measured.
SUFFIXED WORD DETECTION TASK
In this task, participants had to do the same as in the previous task, but this time with respect to the suffix (e.g., lanière, tanière and glacière). The 12 word triplets were made up of one suffixed word and two pseudosuffixed words. All the words were of low frequency according to both their spoken frequency (M = 2.16, SD = 3.21) and their written frequency (M = 3.24, SD = 2.65). Words within triplets were strictly matched by word-onset, number of syllables, and grammatical class. Words were also matched by their oral frequency t (22) < 1, ns, for all comparisons; and written frequency, t (22) = 1.39, p > .20, for the comparison of the suffixed word and one pseudosuffixed word list, t (22) = 1.15, p > .20, for the comparison of the two pseudosuffixed word lists, t (22) < 1, ns, for the remaining comparison. Suffixed words had a root ranging from low to high frequency (range of occurences per million: oral frequency = 1.27–605.75, written frequency = 15.61–923.45), with a greater majority of high-frequency roots (92% in terms of oral frequency and 100% in terms of written frequency). The frequency of the root was always higher than that of the derived form, in terms of either written or oral frequency. Because the pseudosuffixed words did not contain a root, it was not possible to match the suffixed and pseudosuffixed words based on this criteria. The procedure was the same as in the prefixed word detection task.
Statistical analyses
All results were analysed with the R software (R Development Core Team, 2007). We used mixed-effects regression models to run our analyses because they have many advantages for analyzing data with high inter- and intrasubject variability, such as that typically obtained with dyslexic adults or children (see Baayen, Davidson, & Bates, Reference Baayen, Davidson and Bates2008; Goldstein, Reference Goldstein1987, Reference Goldstein1995; Quené & van den Bregh, Reference Quené and van den Bergh2008; Rasbash & Goldstein, Reference Rasbash and Goldstein1994, for details on mixed-effects regression models applied to psycholinguistic data). As with simple linear regression, mixed-effects regression models allow us to test whether a standard fixed predictor accounts for the data. In addition, these models have the advantage of accounting simultaneously for the random variation induced by specific words and by specific participants. In this manner any data averaging across participants or items is avoided. Thus, in all our analyses, we entered both participants and items as random variables, which always significantly improved the final models according to likelihood ratio tests (all ps < .001).
To fullfill the aim of our study (i.e., situate the performance of the dyslexic university students compared to the CA and RL controls), we entered group as a predictor and compared the performance of the dyslexic university students to that of the CA and the RL controls. Because these two comparisons are nonorthogonal, we lowered the significance level to 0.025 following Bonferroni correction. The effect of group (and predictors in general) was tested by comparing the model with and without this predictor by means of likelihood ratio tests.
On the suffixation decision task, in addition to group, several relevant variables were included in the analysis to account for the data: the number of syllables (two vs. three), the type of item (suffixed vs. pseudosuffixed), the response time for responding correctly in the TRISYL task, the physical duration of the item, and the interaction between the type of item and the group. We tested the effect of these variables to ensure that any group effect would not be confounded with such effects. The number of syllables, the type of item, and the interaction between the type of item and the group were entered as categorical predictors for both accuracy and reaction times analyses. The response time for responding correctly in the TRISYL task was also entered as a covariate in both cases (centered), because the phonological difficulties of the dyslexic participants in these morphological awareness tasks may have prevented them from performing this task quickly and accurately.Footnote 6 Finally, the physical duration of the item was entered as a covariate for the analysis of reaction times (centered). Only factors that were significant were retained in the final models.
Accuracy analyses were realized by running generalized linear mixed-effects models underlied by a binomial distribution (statistical package lme4) on the data sets, with accuracy as a dependant variable. To analyze speed, we considered the response times for correct answers only. We removed short response times, which were likely to correspond to anticipated responses (8 datapoints below 750 ms in the suffixation decision task). Then we ensured that the reaction/response times followed a normal distribution. To do so, we performed a visual inspection of the reaction times. This inspection showed that the distribution was right skewed in all tasks. This skewness was removed by taking out the longest response times when necessary (TRISYL task: 7 points above 5000 ms; CCV task: 4 points above 7000 ms; suffixation decision task: 2 points above 15000 ms) and by performing a reciprocal transformation of the remaining data points (following the Box–Cox test; Box & Cox, Reference Box and Cox1964). In this manner, only a few data points were removed from the raw data, and participants were never eliminated from the sample. Mixed-effects models analyses (statistical package lme4) were run on these data sets with centered, reciprocally transformed reaction/response times as a dependent variable. Following Baayen et al. (Reference Baayen, Davidson and Bates2008), residuals having an absolute value higher than 2.5 were considered outliers and duly removed (comprising at most 2.56% of the data).
To sum up, we relied on mixed-effects regression models to run our analyses. For each model, both participants and items were entered as random variables, and group was entered as a categorical predictor. In this manner, we tested the effect of the group while accounting simultaneously for the random variation induced by each item and by each participant. Precision analyses relied upon generalized linear mixed-effects models with accuracy as a dependent variable. Speed analyses used mixed-effects models analyses with centered, reciprocally transformed reaction/response times as a dependent variable.
RESULTS
Phonological awareness
On the phonological awareness tasks, data were missing for one RL control. Data from this participant and the matching dyslexic and CA controls were removed from the data sets. The results (mean scores and standard deviations) of the three reading groups in the phonological awareness tasks are presented in Table 1.
Mean (standard deviation) error percentages and response times related to phonological awareness on the TRISYL, CVC, and CCV tests
Note: TRISYL, trisyllabic; CVC, consonant–vowel–consonant; CCV, consonant–consonant–vowel; CA, chronological age; RL; reading level.
*p< .025. **p < .01. ***p< .001.
Reliability
Cronbach alphas were calculated to measure reliability. When Cronbach alphas did not reach high values, we dropped problematic items to improve reliability. All reliability scores exceed 0.60. On the TRISYL task, one item was dropped to improve reliability, which reached 0.88 and 0.62 for response time and accuracy, respectively. On the CVC task, the Cronbach α was 0.94 for response times (analyses were not conducted on accuracy because participants reached performances at ceiling). On the CCV task, Cronbach αs were 0.86 and 0.72 for response times and accuracy, respectively.
TRISYL test
Group was a significant predictor of accuracy, χ2 (2, 5) = 9.83, p < .01, and the probability of answering correctly was higher in CA controls than in dyslexic university students (β = 1.97, z = 2.51, p = .012), but it did not differ between dyslexic university students and RL controls (β = 0.37, z < 1, ns). In response times, group was a significant predictor, χ2 (2, 6) = 65.75, p < .0001, with the dyslexics being slower than the CA controls (β = –0.08, t = –7.16, p < .0001), but not differing from the RL controls (β = 0.02, t = 1.34, p > .15).
CVC test
Because all groups performed at ceiling, only analyses of response times for correct answers were carried out. Group was a significant predictor, χ2 (2, 6) = 40.81, p < .0001, with the dyslexics being slower than the CA controls (β = –0.10, t = –5.96, p < .0001), but not differing from the RL controls (β = –0.00, t < 1, ns).
CCV test
On the CCV task, reaction times were missing for one RL control. Data from this participant and the matching dyslexic university student and CA controls were removed from data sets. Accuracy was predicted significantly by group, χ2 (2, 5) = 11.33, p < .001, with the dyslexics being less accurate than the CA controls (β = 1.08, z = 2.73, p < .01), but not differing from the RL controls (β = –0.22, z < 1, ns). For response times, group was a significant predictor, χ2 (2, 6) = 76.88, p < .0001, with the dyslexics being slower than the CA controls (β = –0.14, t = –8.96, p < .001) but not differing from the RL controls (β = –0.00, t < 1, ns).
In sum, in phonological awareness tasks, the dyslexic university students were less accurate than their CA controls in both the TRISYL and the CCV tasks, and all groups performed at ceiling in the CVC task. For response times, dyslexic university students were slower than their CA controls in all tasks. The performance of the dyslexic university students never differed from that of their RL controls for the three tasks, neither for accuracy nor speed. In sum, dyslexic university students were outperformed by their CA controls but performed like their RL controls, both in terms of speed and accuracy.
Morphological awareness
Reliability
Cronbach alphas were again calculated to measure reliability; when the alphas did not reach high values, we dropped problematic items to improve reliability. All reliability scores exceed 0.60. On the suffixation decision task, Cronbach αs were 0.91 and 0.69 for response times and accuracy, respectively. On the prefixed word detection task, two items were dropped to improve reliability, which reached 0.69. On the suffixed word detection task, one item was dropped to improve reliability, which reached 0.60.
Suffixation decision task
Concerning accuracy, the type of item was a significant predictor, χ2 (1, 6) = 13.21, p < .001, with participants responding more accurately to pseudosuffixed words as compared to suffixed words. Accuracy was also significantly predicted by group, χ2 (2, 6) = 30.48, p < .0001. Dyslexics did not differ from CA controls (β = 0.08, z < 1, ns), but they were more accurate than RL controls (β = –0.59, z = –4.70, p < .001).
Concerning reaction times, the duration of the item was a marginal predictor, χ2 (1, 7) = 3.60, p = .06. Group was a significant predictor of reaction times, χ2 (2, 7) = 42.48, p < .0001, with dyslexic participants being slower than CA controls (β = –0.11, t = –4.55, p < .001) but faster than RL controls (β = 0.06, t = 2.36, p < .025).
Detection of prefixed word
Accuracy was significantly predicted by group, χ2 (2, 5) = 21.56, p < .0001. The dyslexic university students were marginally less accurate than the CA controls (β = 0.78, z = 1.96, p = .05) and were more accurate than the RL controls (β = –1.00, z = –2.87, p < .01).
Detection of suffixed word
Accuracy was significantly predicted by group, χ2 (2, 5) = 28.90, p < .0001. The dyslexic university students did not differ from the CA controls (β = 0.34, z = 1.25, ns) and were more accurate than the RL controls (β = –1.07, z = –4.18, p < .001).
In sum, in the morphological awareness tasks (Table 2), dyslexic university students did not differ in accuracy from their CA controls in either the suffixation decision task or the detection of suffixed word task. They were marginally less accurate than their CA controls in the detection of prefixed word task and slower than their CA controls in the suffixation decision task. The dyslexic students always outperformed their RL controls, both in terms of speed and accuracy.
Mean (standard deviation) error percentages and response times related to morphological awareness on the suffixation decision task, detection of prefixed words, and detection of suffixed words
Note: CA, chronological age; RL; reading level.
†p < .05 (marginal). **p < .01. ***p< .001.
DISCUSSION
In this study, we evaluated the levels of phonological and morphological awareness in French-speaking dyslexic university students. We employed two control groups, CA controls and RL controls, to accurately assess the performance of the dyslexic university students in each of these abilities. In the phonological awareness tasks, dyslexic students were outperformed by their CA controls and performed at the level of their RL controls. In contrast, in the morphological awareness tasks, dyslexic students clearly outperformed their RL controls and almost reached the proficiency level of their CA controls. These results have several implications that are discussed below.
Concerning phonological awareness, as expected, we have replicated Martin et al.'s (Reference Martin, Colé, Leuwers, Casalis, Zorman and Sprenger-Charolles2010) results using a similar population and similar phonological awareness tasks. In phonological awareness tasks, French-speaking dyslexic university students appeared to perform better than their English-speaking counterparts, who were outperformed by their RL controls in such tasks (Bruck, Reference Bruck1992, Reference Bruck1993; Chiappe et al., Reference Chiappe, Stringer, Siegel and Stanovich2002; Pennington et al., Reference Pennington, van Orden, Smith, Green and Haith1990). To explain these differences, Martin et al. (Reference Martin, Colé, Leuwers, Casalis, Zorman and Sprenger-Charolles2010) proposed that the relative orthographic transparency of the two languages played a role. From the point of view of reading, French orthography is more transparent than English orthography (Peereman & Content, Reference Peereman and Content1998; Peereman, Lété, & Sprenger-Charolles, Reference Peereman, Lété and Sprenger-Charolles2007). Consistent correspondences between graphemes and phonemes may facilitate the development of phonemic awareness, particularly for dyslexics. A good example of the positive effect of orthographic consistency can be found in German, which is considered to have a transparent orthography. Studies by Wimmer (Reference Wimmer1993, Reference Wimmer1996) with German dyslexic children across several grades showed that only in first grade did the children have trouble with their phonemic awareness, which was no longer observed by the end of fourth grade. In a similar fashion, the relative transparency of French orthography might have helped the dyslexic participants to develop phonemic awareness, although this development remained limited.
Concerning morphological awareness, our results indicate that dyslexic university students can develop far better morphological awareness than that which might be expected from their RLs. This finding for dyslexic university students mirrors what has been previously observed in French dyslexic children (Casalis et al., Reference Casalis, Colé and Sopo2004). It adds to the evidence suggesting that the development of morphological awareness is unlikely to be deficient in French dyslexic university students. Several hypotheses can be proposed to explain how morphological awareness could have developed so well in our population.
As we proposed in the introductory section, because the semantic abilities of people with dyslexia are not deficient (Vellutino et al., Reference Vellutino, Scanlon and Spearing1995), this population could probably rely on the semantic dimension of morphology to develop morphological awareness. Some studies also reported a relationship between morphological awareness and vocabulary knowledge in normally developing readers (Carlisle & Fleming, Reference Carlisle and Fleming2003; Colé et al., Reference Colé, Royer, Leuwers and Casalis2004; Kirby, Derochers, Roth, & Lai, Reference Kirby, Derochers, Roth and Lai2008). However, the exact nature of this relationship is still under discussion. A vocabulary measure was not included in our study, but it would be useful in future studies to test if the level of morphological awareness is related to that of vocabulary knowledge.
That morphology is encoded in the orthography could also be important for the development of morphological awareness. In French, as described by Jaffré (Reference Jaffré2006), morphemes are systematically encoded in print, even though they are frequently not fully pronounced. For example, morphograms (letters or groups of letters at the end of words) that often encode morphological relations establish a visible link between different words sharing a morphological relationship, but they are not always pronounced.Footnote 7 Morphograms are likely to help French readers to grasp the morphological relations between words and to develop awareness of the morphological structure of the language. Moreover, although morphograms introduce inconsistencies at the graphophonological level, they represent consistencies at a graphosemantic level, which could be particularly relevant for dyslexic readers. This hypothesis about the role of orthography in the development of morphological awareness is not limited to French because alphabetic languages all encode morphology in print to some extent. Across such languages, there are nonetheless differences in consistency between speech and print in morphological encoding. The involvement of orthography in the development of morphological awareness across languages may depend precisely upon these differences in consistency. More generally, the involvement of orthography in the development of morphological awareness is supported by the fact that dyslexic readers appear to be sensitive to the morphological information contained in written language. The few studies conducted on alphabetic orthographies with dyslexic children and teenagers show that they process the morphological structure of the written language in word identification and that this processing does not seem to be deficient (Burani, Marcolini, De Luca, & Zoccolotti, Reference Burani, Marcolini, Luca and Zoccolotti2008; Carlisle et al., Reference Carlisle, Stone and Katz2001; Elbro & Arnbak, Reference Elbro and Arnbak1996; but see Deacon, Parrila, & Kirby, Reference Deacon, Parrila and Kirby2006, for contrary results with dyslexic university students; see also Deacon, Parrila, & Kirby, Reference Deacon, Parrila, Kirby, Reid, Fawcett, Manis and Siegel2008, for a literature review of morphological processing by dyslexic readers).
That our target population consisted of university students must also be taken into account when interpreting the high degree of morphological awareness we observed. First, dyslexic university students are not representative of dyslexic adults. Data indicate that university students with learning disabilities (including dyslexia and other learning disabilities) represent less than 2% of the student body (Horn & Berktold, Reference Horn and Berktold1999), which contrasts with the prevalence of dyslexia in the general population (5% to 10%; INSERM, 2007). Second, several authors propose that dyslexic university students develop compensatory strategies to be able to pursue university studies despite their persistent reading difficulties. Elbro & Arnbak (Reference Elbro and Arnbak1996) suggested that morphological knowledge may support one such compensatory reading mechanism. The level of morphological awareness of our target population, which is better than expected from their RL, is consistent with the idea that this skill does play a compensatory role in reading. Unfortunately, our study was not designed to test for the existence of a compensatory mechanism, which would have required other experimental conditions. Moreover, this level of morphological awareness is not specific to dyslexic university students because it has also been observed in dyslexic children (Casalis et al., Reference Casalis, Colé and Sopo2004). Whether these children also possess compensatory reading strategies remains an open question. In sum, the low representativity of our population and the possibility that morphological awareness has a specific compensatory function in this population make it difficult to generalize our results to the entire population of dyslexic adults. However, our results raise the very interesting possibility that morphological awareness may play a specific compensatory role in our population.
Third, the comparison of phonological and morphological awareness in our target population, using RL controls as a baseline, showed differences between these two skills, with an advantage in favor of morphological awareness. The level of morphological awareness of the dyslexic university students clearly surpassed that which was expected from their RL. In contrast, their level of phonological awareness corresponded to that which was expected from their RL. One could argue that, in morphological awareness tasks, dyslexic university students were faster than their RL controls because they were much older and thus benefited from an advantage in their motor and cognitive development (rapidity of processing and response). However, if it were the case, this should also have been true for phonological awareness tasks. In phonological awareness tasks, the dyslexics were never faster than their RL controls. Thus, the dyslexic university students were genuinely more proficient in completing morphological awareness tasks than their RL controls, and this effect could not be attributed to differences in their general level of cognitive and motor development. The higher level of morphological awareness, as opposed to phonological awareness, observed in our dyslexic university students adds to the body of evidence that suggests that these two metalinguistic abilities can, to a certain extent, develop independently.
In conclusion, the present study has allowed us to show that French-speaking dyslexic university students exhibit higher levels of morphological awareness than phonological awareness by using CA and RL controls as a baseline. Future studies should clarify the role of semantics, vocabulary, and orthography in the development of morphological awareness in dyslexic population. Moreover, it would be essential to understand if the level of morphological awareness demonstrated by our university student population is also found in dyslexic adults in general or if it is somewhat exceptional. It would also be crucial to establish if it constitutes a specific compensatory mechanism. It is important to identify the abilities and procedures that can improve reading in dyslexic readers, especially because they could provide a solid basis for intervention. It is likely that morphological instruction could be beneficial to dyslexic readers because it has already proved to be beneficial to learners, and in particular to poor readers (see Bowers, Kirby, & Deacon, Reference Bowers, Kirby and Deacon2010, for a review of the literature, and Goodwin & Ahn, Reference Goodwin and Ahn2010, for a meta-analysis).
