Subtypes Based on Nature of Word Recognition Deficits

Reading Styles

A popular view in "folk psychology" and education is that children can be divided into two distinct subgroups based on whether they learn to read more easily by the visual route or phonological route (Carbo, 1987, 1992; Dunn, 1990). Carbo (1992), for example, divides children into global learners and analytic learners. Global learners or readers are argued to learn to recognize words best through a sight-word approach that makes use of the visual route. Analytic readers, on the other hand, learn to read best by a phonics method that takes advantage of the phonological route. Many in early education also refer to these groups as visual and auditory learners and believe that teachers should identify a child's learning style and teach to that style.

Despite the widespread appeal of reading/learning styles, the evidence is not very compelling that children can be divided into homogeneous subgroups on the basis of their reading strengths (or preferences), or that teaching to these strengths is an effective strategy for improving reading ability (Kavale, Forness, & Bender, 1987; Stahl, 1988; Stahl & Kuhn, 1995; Turner & Dawson, 1978). Relatively few studies have actually addressed this issue. Those studies that have offered some support for reading style subgroups and instruction (Holt & O'Tuel, 1990; Thomasson, 1990) have typically been reported outside the peer-review process. As a result, this work has not had the level of scrutiny and evaluation that is needed in order to effectively guide educational practice. This view is also contrary to most current theories of reading development. Most research suggests the importance of both the visual and the phonological routes in learning to read (Share, 1995; Share & Stanovich, 1995). Children need to have good phonological decoding skills to break the alphabetic code (i.e., self-teach) as well as good orthographic skills to develop accurate and automatic word recognition.

Dysphonetic, Dyseidetic, and Alexic Subgroups

Although evidence supporting the existence of word recognition subgroups in the population as a whole is not strong, there is converging research that indicates that such a classification system may be of value for subgrouping poor readers. There is a long history of poor readers being classified on the basis of individual differences in reading by the phonological versus the visual route (Boder, 1971, 1973; Ingram, 1964). Ingram (1964), for example, grouped poor readers into audio-phonetic dyslexics and visuo-spatial dyslexics. The audio-phonetic dyslexics were argued to have problems in sound discrimination and blending and to be poor in phonological decoding. The visuo-spatial dyslexics, on the other hand, were proposed to have difficulties in visual discrimination and spatial skills and problems reading by the sight-word route.

Elena Boder (1971, 1973) developed a classification system that recognized three subgroups of poor readers based on misreadings and/or misspellings: the dysphonetic, dyseidetic, and alexic. The dysphonetic subgroup has a primary deficit in auditory analytic skills. Children in this subgroup have great difficulty learning and using the phonological route. These children display misreadings and misspellings that are phonetically inaccurate. For example, the dysphonetic reader might pronounce block as book or spell scramble as sleber. Dyseidetic readers, on the other hand, have a deficit in the visual route. Consequently, they have particular problems with exception words (e.g., have, colonel). These words are misspelled or misread as phonetic renditions: for example, reading talc for talk or spelling laugh as laf. Finally, the alexic subgroup have a deficit in both phonetic and visual reading/spelling skills. This subgroup is the most handicapped of the three groups.

The primary evidence for the validity of these subgroups comes from a study of 107 dyslexic children (Boder, 1973). Using an in-depth analysis of reading and spelling abilities, 100 of these children were divided into one of the three subgroups. Boder reported that 67 of the dyslexic children were dysphonetic, 10 were dyseidetic, and 23 were alexic. Boder and a colleague (Boder & Jarrico, 1982) later developed a diagnostic screening test for subtyping dyslexia. Researchers, utilizing this test, have provided some evidence of behavioral and electrophysiological differences between subtypes of dyslexics (Dalby & Gibson, 1981; Flynn & Deering, 1989). Flynn and Deering (1989), for example, found that dyseidetic children demonstrated greater EEG activity in the left temporal-parietal region during reading than did dysphonetic children. They suggested that this was evidence of different processing capabilities between these subgroups. Others, however, have failed to uncover reading-related differences between these subgroups of poor readers (Godfrey, Lasky, Millag, & Knox, 1981; van den Bos, 1982). Godfrey and colleagues (1981), for example, failed to find an advantage in speech perception abilities among dyseidetic dyslexics as compared to dysphonetic dyslexics. Such a difference would be expected if dysphonetic dyslexics had a phonological processing problem.

Deep, Phonological, and Surface Dyslexia

Cognitive neuropsychologists have also considered subgroups similar to those proposed by Boder (Coltheart, Patterson, & Marshall, 1980; Marshall & Newcombe, 1973). This work, however, has used terminology and procedures borrowed from the study of acquired dyslexia. Acquired dyslexia is a reading disability in previously literate individuals following neurological damage. Three syndromes are often identified: deep, phonological, and surface dyslexia. Individuals with deep and phonological dyslexia have considerable difficulty in phonological decoding. They are identified primarily on the basis of their problems pronouncing nonwords such as zun or vope. Such words cannot be recognized by the visual route and must be sounded out using sound-letter correspondence rules. Individuals with deep dyslexia, unlike those with phonological dyslexia, also make semantic errors in reading. For example, when asked to read a word like tulip they might say "crocus" or they might read "sun" for moon. Other symptoms include visual errors (confusing words like wife and life), morphological errors (misreading prefixes or suffixes), and greater facility recognizing content words as opposed to function words (Thomson, 1984). Finally, individuals with surface dyslexia have problems with the visual route. They are identified on the basis of their misreading of exception words. Whereas the terms phonological and surface dyslexia roughly correspond to dysphonetic and dyseidetic readers, the former terms have become more popular in recent years.

Using primarily case studies, cognitive neuropsychologists have subtyped individuals with developmental reading disabilities as phonological or surface dyslexics (Coltheart, Materson, Byng, Prior, & Riddoch, 1983; Holmes, 1978; Marshall, 1984; Temple & Marshall, 1983; see Rayner & Pollatsek, 1989, for review). For example, Temple and Marshall (1983) described a case of developmental phonological dyslexia. This student, a 17 old girl, had considerable difficulty reading nonwords compared to real words. Her responses to nonwords were typically real words that were visually similar to the target words. Marshall (1984) noted that this developmental case was very similar to the case of acquired phonological dyslexia reported by Patterson (1982). Coltheart and colleagues (1983) and Holmes (1978), on the other hand, identified a number of cases of developmental surface dyslexia. Holmes reported on four boys, between 9 and 13 years of age, who had great difficulty reading exception words. They often made phonetic errors, regularizing words like bread as "breed." Coltheart and colleagues (1983) identified a 15-year-old dyslexic girl who had many problems with homophones. For example, she was noted to read "pane" correctly, but to define it as "something that hurts."

Heterogeneity without Clusters

The classification system proposed by cognitive neuropsychologists may lead to the impression that poor readers can be divided into distinct and homogeneous subgroups based on word recognition deficits. Ellis (1985), however, has argued that while there may be heterogeneity among poor readers in terms of word recognition strengths and weaknesses, poor readers do not form distinct subgroups. He proposed that word recognition abilities can be viewed according to two dimensions: one dimension corresponding to reading by the visual route and the other dimension representing reading by the phonological route. He maintained that readers' abilities are distributed continuously along each of these dimensions. Readers may show similar abilities in these dimensions or have abilities in one dimension that are significantly better than those in the other.

Operationally, these abilities can be displayed on a scatterplot in which performance on exception word reading represents one axis and scores on nonword reading constitutes the other. Ellis noted that when plotted like this, the distinct subtype view of cognitive neuuropsychologists assumes that there will be "galaxies" of dyslexics within the scatterplot. That is, phonological dyslexics would be expected to represent a cluster of poor readers who are separated from other readers by their distinct pattern of poor phonological decoding skills and good exception word reading skills. The surface dyslexics, on the other hand, would be predicted to cluster together in this two-dimensional space as a result of their poor exception word reading skills and good phonological decoding skills. Ellis, however, argued that a more valid conceptualization of heterogeneity is one without clusters or galaxies. He suggested that poor readers are more likely to be distributed continuously in this multidimensional space, such that "there will be a complete and unbroken gradation of intermediate dyslexics linking the extreme cases" (Ellis, 1985, p. 192). In proposing this model, Ellis does not deny individual differences, only the homogeneity of subgroups. In other words, he argues that children with RD do not fall into distinct categories in terms of their word recognition skills. While some children can be characterized as surface or phonological dyslexics, these children will differ by degree of impairment and not type of impairment.

Recently, Ellis and his colleagues (Ellis et al., 1996) tested this view of the heterogeneity of word recognition by examining a group of thirteen children with RD. These children, who were 9 to 11 years old, had normal or above normal IQs and a reading age eighteen or more months behind their chronological age. Three control groups, each consisting of thirteen children and matched for reading level to the dyslexic group, were also included. One group consisted of poor readers of the same age as the children with RD, but with lower IQ scores. Another group contained younger children who were reading at a level predicted for their age. The final group was an even younger group of precocious readers, children who were reading well above their age. All children read a list of non words and real words (half of which were exception words). A scatterplot of nonword reading abilities versus sight-word reading abilities showed considerable variability among the dyslexic children. However, there was no evidence of clustering among the dyslexic readers. Instead, the dyslexic children were distributed continuously throughout the scatterplot. Ellis and his colleagues also found similar heterogeneity in the three control groups.

Murphy and Pollatsek (1994) also examined the heterogeneity of word recognition abilities, but in a much larger sample of children with RD. Sixty-five children with RD, 10 to 13 years of age, were administered a variety of measures designed to test children's ability to read by the visual or phonological routes. These included timed and untimed reading of regular, exception, and nonwords; a lexical decision task; and a homophone definition task. Participants' phonological awareness and word retrieval abilities were also assessed.

Despite finding much hetergeneity between poor readers in word recognition abilities, they too failed to uncover distinct clusters of poor readers. Poor readers differed primarily in terms of the severity of deficits, and not in the kind of deficits. Most children with RD were poor at reading by both a visual and phonological route. In addition, a moderate correlation was found between nonword and exception word reading. If discrete subgroups had been present, such a correlation would have been negative, or at least absent. Nevertheless, there were some children with RD who did show a dissociation between phonological decoding and sight-word reading. These children, however, were still part of the same continuum and did not cluster together into discrete subgroups.

Murphy and Pollatsek (1994) further speculated on the reasons for the dissociation in some children with RD. They noted that children fitting the profile of phonological dyslexics performed less well on a phonological awareness task and better on a phonological retrieval task than did children who displayed a surface dyslexia profile. They also speculated that instructional factors may have contributed to individual differences. Several of the surface dyslexics had been enrolled in intensive phonics programs that taught them to read non words and real words, but few exception words. Such instruction could have led to the error pattern of a surface dyslexic.

The results of these studies strongly suggest that poor readers cannot be divided into homogeneous subgroups based on their word recognition abilities. Some poor readers do, however, display a dissociation in their ability to use the phonological as opposed to the visual route. This dissociation may be related to differences in cognitive processing or reading instruction/experience (Murphy & Pollatsek, 1994). The fact that poor readers do display a dissociation despite the absence of distinct and homogeneous clusters suggests that the classification of poor readers on the basis of word recognition abilities might have some clinical/educational validity.

Hard versus Soft Subtypes

For a classification system based on word recognition to be of use, it is necessary to have a reliable procedure to differentiate children with phonological and surface dyslexia. Recall that cognitive neuropsychologists have used the comparison of nonword and exception reading abilities (e.g., Coltheart et al., 1983; Holmes, 1978). Nonword reading relies primarily on phonological decoding, while exception word reading is dependent on sight-word reading abilities. The case reports presented by cognitive neuropsychologists give the impression that phonological dyslexics had poor nonword reading, but normal exception word reading, while surface dyslexics had poor exception word reading, but normal nonword reading. Such "purity" of subtype was seldom the case. Most reports concerned individuals with relative differences in nonword and exception word reading accuracy. For example, the surface dyslexic described by Coltheart and colleagues had problems reading both non words and exception words, but her problem was more pronounced for exception words.

Recently, Castles and Coltheart (1993) investigated different ways of identifying word recognition subtypes. They administered measures of exception word and nonword reading to 53 dyslexics and 56 normal children matched for chronological age. Initially, they divided the poor readers into what can be called "hard" subtypes (Stanovich et al., 1997). According to this approach, dyslexics who performed poorly in exception word reading, as compared to same age peers, but normally in nonword reading were defined as surface dyslexics. Phonological dyslexics were defined as those students who showed poor nonword reading, but normal exception word reading. Castles and Coltheart found that these procedures led to the identification of only 8 phonological dyslexics and 10 surface dyslexics. These numbers were smaller than were expected on the basis of previous reports. Castles and Coltheart noted, however, that many of the poor readers showed a relative difference between nonword and exception word reading. Accordingly, the researchers proposed a statistical procedure that would identify children who showed relative differences, but not necessarily deficits, in one or the other area of reading. These can be called "soft subtypes." This technique involved the use of regression analyses to subgroup children with RD into those with better nonword readding than would be predicted on the basis of exception word reading (i.e., surface dyslexic), or those with better exception word reading than would be predicted on the basis of nonword reading (i.e., phonological dyslexic). In each case, predictions and confidence intervals were based on data from the chronological age-matched control group. Using this approach, Castles and Coltheart identified 16 surface and 29 phonological dyslexics. Thus, most of their poor readers (45 out of 53) showed a relative dissociation between nonword and exception word reading. The researchers argued that although these poor readers, for the most part, did not represent hard cases of surface or phonological dyslexia, the apparent dissociation in word recognition profiles could have important implications for understanding and treating reading disabilities.

Whereas the identification of soft subtypes seems possible, some have questioned how to best evaluate poor readers' relative strengths in nonword and exception word reading. Recall, Castles and Coltheart (1993) used a chronological-age-matched control group to estimate the relationship between nonword and exception word reading. In other words, they used same age peers to determine how many nonwords poor readers should read correctly, given their exception word reading score and vice versa. Stanovich and colleagues (1997) maintained that age-related data may not be appropriate for evaluating the relative strengths of poor readers who are reading at a level well below that of chronological-age-matched children. The relationship between nonword and exception word reading at different reading levels may not be the same. Less skilled readers may, for example, be expected to read fewer exception words for a given nonword reading score than more skilled readers. Stanovich and colleagues (1997) suggested that the more appropriate comparison for poor readers is a younger group of normal children reading at the same overall level as the poor readers.

To evaluate this claim, Stanovich and colleagues (1997) used regression analyses based on both chronological-age-(CA) and reading-level-(RL) matched control groups to divide 68 third-grade children with RD into phonological and surface dyslexic subtypes. When using regression-based predictions from CA-matched children, the researchers found that approximately half of the children with RD (53%) were poor in reading both exception words and nonwords. However, some children with RD showed relative strengths on one or the other set of words. Specifically, 22 percent of the sample performed better on nonwords than exception words (i.e., surface dyslexics), while 25 percent scored better on exception words than nonwords (i.e., phonological dyslexics). When using predictions based on RL-matched controls, again 25 percent of the children with RD could be classified as phonological dyslexics. Using these data, only one child was identified as a surface dyslexic. That is, compared to RL-matched children, surface dyslexia essentially disappeared. Similar findings have also been reported by Manis and his colleagues(Manis, Seidenberg, Doi, McBride-Chang, & Petersen, 1996; Manis, Seidenberg, Stallings, Joanisse, Bailey, Freedman, Curtin, & Keating, 1999).

The findings from these studies provide some insights into the nature of the reading problems of phonological and surface dyslexics. Children identified as surface dyslexics, when compared to CA controls, may best be characterized as showing a developmental lag. These children did not display deviant reading abilities; rather, their nonword and exception word reading was like that of younger normal children. These children appear to be taking longer than same-age peers to learn to read. Stanovich and colleagues suggested that these children may have a mild form of a phonological processing deficit. They further speculated that this deficit when combined with exceptionally inadequate reading experience could result in a surface dyslexic profile. In contrast to surface dyslexia, phonological dyslexia may constitute a true developmental disorder. Phonological dyslexics continued to show a distinctly different pattern of performance when compared to younger normal children. Furthermore, the phonological dyslexics, in contrast to the surface dyslexics, performed less well than the RL-matched children on tests of phonological awareness, working memory, and syntactic processing.

The results of Stanovich and colleagues (1997) are consistent with those of Murphy and Pollatsek (1994), who found that phonological dyslexics had deficits in phonological awareness. Murphy and Pollatsek also suggested that instructional factors contributed to the reading profiles of some of their cases of surface dyslexia. A recent investigation by Vellutino and colleagues (Vellutino, Scanlon, Sipay, Small, Chen, Pratt, & Denckla, 1996) also provides some converging evidence. As will be discussed in more detail in the next chappter, Vellutino and his colleagues found that some children with RD could be "readily remediated" with short-term intervention. These children, who Vellutino and colleagues believed to have instructional or experiential deficits (and may have also had mild phonological processing deficits), may overlap with the surface dyslexics identified by Stanovich and his colleagues. In contrast, Vellutino and colleagues also identified a group of "hard to remediate" poor readers who seem to fit the profile of phonological dyslexics in that these children had deficits in phonological decoding and phonological processing. Finally, Casstles, Datta, Gayan, & Olson (1999) examined the genetic and environmental influences on word recognition subgroups. They reported that phonological dyslexia was more likely to have a genetic basis than was surface dyslexia. The opposite pattern was observed for environmental influences (also see Gustafson, 2001).

In summary, it may seem that we have taken a circuitous and bumpy path in our attempt to understand the often conflicting research on word recognition subgroups. It is important to recognize that although poor readers do not cluster into homogeneous subgroups, they do show dissociations in their ability to use the phonological or visual route. It is these dissociations that may prove to have some utility for understanding and treating word recognition problems. We will elaborate on possible clinical implications later in this chapter. Before doing so, however, we need to consider one further subtyping system related to word recognition abilities.

Rate-Disabled versus Accuracy-Disabled Readers

Research further suggests that poor readers may be subgrouped on the basis of word reading speed and accuracy. For example, Lovett and her colleagues (Lovett, 1984a, 1984b, 1987; Lovett, Ransby, & Barron, 1988; Lovett, Ransby, Hardwick, & Johns, 1989) proposed two subtypes of reading disabilities. One subtype, accuracy-disabled children, was defined as those with significant problems in decoding accuracy, while the other, rate-disabled children, were those with a marked deficit in reading rate despite grade-appropriate decoding ability. To be classified as accuracy-disabled, a child had to score at least one and a half years below grade-level expectations on at least four of five different measures of word recognition. To be classified as rate-disabled, a child had to perform close to, at, or above grade level on four or more measures of word recognition and at least one and half years below grade level on four of five measures of reading speed.

In an attempt to validate the above subgroups, Lovett (1987) administered a battery of oral and written language tests to 32 accuracy-disabled, 32 rate-disabled, and 32 normal children. The children were matched for chronological age, sex, and IQ. The oral language tests measured lexical, morphological, and syntactic knowledge. The written language battery included standardized and experimental measures of single word recognition, decoding in context, reading rate, reading comprehension, sound-letter processing, and related academic skills. The results confirmed the distinctiveness of the three groups. The accuracy-disabled children produced more errors, read more slowly, and showed poorer comprehension than the rate-disabled and normal children. The errors the accuracy-disabled children made in reading nonwords indicated that they still had not acquired basic knowledge of sound-letter correspondence rules. With respect to oral language abilities, the accuracy-disabled children demonstrated deficits in morphological and syntactic knowledge. They also were significantly slower than rate-disabled children in naming serial-letter arrays and analyzing individual speech sounds. Lovett concluded that, "these data suggest that accuracy-disabled children suffer a multidimensional language impairment coupled with specific sound analysis difficulties and a seemingly inability to automatize or consolidate single letter identities and/or names" (Lovett, 1987, p. 257).

The reading abilities of the rate-disabled sample were more selectively impaired. There were no differences between these children and the normal readers in their identification of regular and exception words, suggesting that the groups were equally adept at phonological decoding and sight-word reading. Although these groups were equivalent in accuracy, the rate-disabled subjects exhibited significant impairments in word recognition speed. This was particularly the case in connected text, where these children appear to become functionally overloaded by the demands of processing large units of text. With respect to oral language abilities, the rate-disabled and normal readers were similar with one exception. The rate-disabled children were significantly slower on tasks measuring rapid automatic naming. More recently, Aaron, Joshi, and Williams (1999) also investigated word-reading speed and accuracy in poor readers. They examined 139 children in third, fourth, and sixth grade and identified 16 poor readers who performed more than one standard deviation below the mean in reading comprehension. These children were assessed on various measures of word recognition speed and accuracy and listening comprehension. Most of the poor readers performed poorly in more than one area. However, Aaron and colleagues found that 2 poor readers had significant deficits in word-reading speed but normal decoding and listening comprehension abilities.

The above research thus indicates the existence of another subgroup of poor readers based on word recognition deficits. This subgroup, rate-disabled readers, are accurate in word recognition, but are slow in reading rate. It is unclear from the current research, however, exactly how these children fit into a developmental model of reading. For example, do these children have accuracy problems that later turn into rate problems? It has been our experience that at least some rate-disabled readers do not have a pronounced history of difficulties with accuracy of word recognition. These children appear to develop normally through the primary grades, but then experience significant difficulties in reading more advanced material.

What underlies these children's problems is unclear at present. Some may not have had an adequate amount of reading experience. Because automaticity of word recognition increases with practice, they may lag behind their normal peers in speed of word recognition. Many rate-disabled readers, however, continue to have problems with reading rate despite considerable literacy experience. Reading rate problems may also be related to phonological retrieval deficits. Lovett's results seem to confirm the problems these children have in the rapid retrieval of verbal labels. More recently , Wolf, Bowers, and Biddle (2000) have also reported a link between reading rate and naming speed.

Finally, Lovett's more recent work (Lovett, Benson, & Olds, 1990) is consistent with issues concerning the heterogeneity of clustering. Rather than treat accuracy-disabled and rate-disabled poor readers as distinct subgroups, she and her colleagues have begun to consider the dimensions that underlie these subgroups as continuous variables. For example, they investigated how the continuous variables of reading rate and accuracy are related to intervention outcome.