Phonological retrieval

The second aspect of phonological processing repeatedly demonstrated to be linked with literacy is the efficiency of accessing and retrieving names of familiar objects or symbols. This efficiency is operationalized with naming tasks that employ stimuli other than written words: letters, digits, colours, or pictures of common objects. Discrete, or

confrontation naming tasks present one stimulus at a time, and the familiarity of stimuli may be varied systematically. This way, both accuracy and latency data may be collected. Serial, or rapid automatized naming (RAN) tasks use small sets (usually around 5 items) of highly overlearned stimuli which are displayed repeatedly in a pseudo-random order and have to be named continuously. With the latter procedure, a high degree of accuracy is assured, and only naming time data are typically collected. Most of the studies investigating the relationship between naming and reading employed the RAN paradigm.

Following the seminal studies of Denckla & Rudel (e.g. 1976), a large and consistent body of evidence has been accummulated which shows that naming abilities are strong concurrent and longitudinal predictor of reading, independent of general intelligence (for review, see Wolf, 1991; Wagner & Torgesen, 1987; Wolf & Bowers, 1999). Moreover, naming and phonological awareness were consistently shown to load on different (albeit usually correlated) factors, each accounting for independent variance of reading. A unique contribution of naming was observed with both normal and

reading impaired participants, yet it is usually greater for disabled readers (Manis, Seidenberg & Doi, 1999). Naming generally explains a smaller amount of reading variance than does phonological awareness, and its contribution may even disappear once stringent controls for reading level are introduced (Torgesen et. al., 1997). The association between reading and naming also diminishes with age, generally faster than the association between reading and phonological awareness (Wagner et. al., 1997). In alphabetic orthographies other than English, however, the relative role of naming seems greater and more enduring (see chapter 3).

Regression studies have shown that phonological awareness and naming not only explain non-overlapping variance of reading, but also that each is best at explaining different aspects of reading. Although both skills predict reading accuracy as well as speed, naming is usually best at predicting (concurrently and prospectively) reading speed and reading comprehension, worse at predicting word identification accuracy and poor at predicting nonword decoding accuracy (Manis, Seidenberg & Doi, 1999; Wolf & Bowers, 1999). The reverse is typically the case with phonological awareness. This suggests some degree of specificity: whereas phonological awareness skills may be most important for the acquisition of phonological recoding, naming tasks may index the ability to identify familiar words quickly. The relation of naming speed to reading comprehension may be indirect: naming contributes to word identification accuracy and speed, which are prerequisites of good reading comprehension (Wolf & Bowers, 1999; Kail & Hall, 1994).

Corroborative evidence for a separate role for naming and phonological awareness in reading acquisition comes from dyslexia studies. Although most dyslexic cases show double deficits (of phonological awareness/recoding as well as naming) smaller, single-deficit subgroups can always be identified (Wolf & Bowers, 1999). Dyslexies with a selective deficit of phonological awareness and/or recoding tend to make more errors in word reading than those with a selective naming speed problem; the latter group, however, takes more time to read (Lovett, 1995, reported by Wolf & Bowers, 1999). Such patterns of difference do not always reach significance, however, (Bowers, 1995 & Wolf, 1997, reported by Wolf & Bowers, 1999). Children with double deficits were consistently shown to perform much worse on all aspects of reading than children with either single deficit. This, however, may reflect the fact that double deficit groups tend to be more severely impaired on both phonological awareness and naming than the respective single-deficit groups. Once single and double groups are matched on

the severity of individual deficits (Compton, De Fries & Olson, 2001), then phonological awareness alone seems to limit decoding accuracy (nonword reading equal in phonological awareness deficit and double deficit groups), and naming alone limits word reading speed and comprehension (they are equal in naming deficit and double deficit groups). When children are grouped by their reading performance, poor readers showing selective difficulties with reading speed (but normal accuracy) exhibit a single deficit of naming, yet those with selective difficulties of reading accuracy (but normal speed) show a double deficit of naming and phonemic awareness (Lovett, 1987). These results imply at least partial dissociation between phonological awareness and rapid naming as predictors of reading.

Some authors (e.g. Manis, Seidenberg & Doi, 1999) presented evidence for naming ability being an especially strong predictor of specifically orthographic skills: judging orthographic legality of nonword spellings (letter string choice), homophone choice, or exception word reading. However, the comparison of severity-matched single and double deficit groups (Compton, DeFries & Olson, 2001) revealed no difference between phonological awareness, naming and double deficit groups on orthographic skills, suggesting that the two types of phonological processing may be equivalent with respect to orthographic ability.

There seem to be two main unresolved theoretical issues regarding the connections between naming and literacy. The first concerns the functional architecture of naming, and the nature of resources it relies on. Some authors (e.g. Wagner & Torgesen, 1987, Wagner et. al., 1993) describe naming tasks as fundamentally phonological. All kinds of naming (including reading) share the same core demand for access and retrieval of phonological information. Common variance of reading and naming reflects differences in efficiency of access and retrieval processes. Any problems with retrieval accuracy and speed are phonological in nature and can be ultimately explained by the poor quality of underlying phonological representations (e.g. Katz, 1986).

Alternative accounts stress that some crucial aspects of naming may not be specifically phonological. This is consistent with the fact that correlations between naming measures and other aspects of phonological processing (e.g. working memory, phonological awareness) are moderate at best (and often non-significant), suggesting a substantial degree of dissociation. Another argument for the importance of non- phonological aspects of naming is inherent in the fact that not all naming tasks are

equally good predictors of reading. Most of the positive findings described so far involved rapid automatized naming (RAN) of alphanumeric stimuli (letters, digits), while weaker or negative results were found in studies using RAN format with pictures and colours and, weaker still, with confrontational naming tasks (for review and discussion see Wolf, 1991; Wolf & Bowers, 1999). This is unsurprising, since serial naming of letters or digits resembles reading activity more than other types of naming (an issue I shall return to in the end of this section). However, there may be some deeper reasons, too. Reading and naming may share not only a phonological processing requirement, but also the demand for automaticity: something that is indexed more directly by naming highly overlearned arbitrary symbols (e.g. letters) than by naming pictures. Several authors, in particular. Wolf and Bowers (Wolf, 1991; Bowers & Wolf, 1993; Wolf & Bowers, 1999) see rapid serial naming as an index of general ability to automatize complex cross-modal perceptual and motor skills. The underlying mechanism of this ability may be the precise timing of all component subprocesses (Wolf, 1991; Bowers & Wolf, 1993). To support this stance. Wolf & Bowers (1999) collated different strains of evidence showing that reading difficulties are related to problems with speed of processing, precise timing and co-ordination that are evident in a variety of tasks, not only naming. Poor readers, when compared to chronological-age matched controls, have difficulties in processing rapidly changing visual information (Lovegrove & Williams, 1993; Lovegrove, 1994) and judging the temporal order of two stimuli presented in rapid succession, both in visual and auditory domains (Kinsbourne, Rufo, Gamzu, Palmer & Berliner, 1991). Similar difficulties in temporal order judgement and perception of very brief tones were also implied in children with speech output and articulation difficulties, which constitute a group at high risk of dyslexia (Tallal et. al., 1996; Merzenich et.al., 1996). Dyslexies also perform significantly worse than controls on motor tasks which are complex and asynchronous, or which have to be performed together with a distracting activity (Fawcett & Nicolson, 1994). Those difficulties cannot, however, be accounted for by a general slowness of processing, since dyslexies are not different from normal readers on simple reaction times (Fawcett & Nicolson, 1994). It seems rather that reading problems are related to difficulties in making fine temporal judgement, or speeded co-ordination of information across modalities. At present, this conclusion remains highly speculative and it remains to be demonstrated whether the apparently disparate difficulties listed by Wolf & Bowers (1999) can indeed be traced to the same underlying problem with temporal and cross-

modal integration. It must be said that the findings of non-verbal deficits in dyslexia reported above are contentious and have been criticised on methodological and theoretical grounds (eg. Wimmer, Mayringer & Raberger, 1999; Studdert-Kennedy & Mody, 1995; Skottun & Parke, 1999). Studies directly exploring the connection between naming ability and nonverbal speeded processing are sparse but promising. Kail and Hall (1994) reported that different nonverbal measures of processing speed: a cross-out (cancellation) task, visual matching, WISC-R coding, are indeed intercorrelated and predict RAN performance, which, in turn, predicts word recognition. Kinsbourne et. al. (1991) demonstrated temporal order judgement deficits for both visual light flashes and auditory clicks in adult dyslexic reading, and showed that performance on these tasks was highly correlated with naming speed. However, Wimmer, Mayringer and Landerl (1999) who used several measures of visual and motor processing speed, alongside with RAN tests found that only naming but not other speed measures could reliably discriminate dyslexies (both surface and phonological) from normal readers. His study did not, however, address the question of whether different measures of processing speed share common variance.

It is important to remember that the two accounts of naming (a phonological code retrieval vs. general timing/integration mechanism) are not necessarily exclusive. Cognitive architecture of naming may incorporate both phonology-specific and cross- modal components; either of them (or both) can constrain reading development (Wolf & Bowers, 1999).

The second controversy regards the hypothesis of naming being specifically linked with the acquisition of orthographic, rather than alphabetic skills. The proponents of this view (e.g. Manis, Seidenberg & Doi, 1999) point out that naming involves

arbitrary connections between visual representations and their names (phonological labels). It is precisely the formation and activation of such arbitrary connections that is required for skilled reading. Alphabetic skills, on the other hand, rely on systematic

letter-sound correspondences, and thus require phonological awareness. Wolf and Bowers (Wolf & Bowers, 1993; Bowers & Wolf, 1999) also developed this view. They emphasised the associative nature of learning to read, which requires formation of stable connections between frequently co-activated units. Those connections are both orthographic (when co-activated letter recognition units merge into familiar orthographic patterns) and phonographic (when different-size phonological elements - single letters, intrasyllabic units, whole words - are co-activated with the corresponding

letter strings). This network of connections cannot form properly unless the corresponding units are activated in sufficiently close temporal proximity. Activation asynchrony will disrupt associate learning and require more practice (greater number of pairings) to establish internal orthographic representations of sufficient quality. The same mechanism may also account for slow naming.

A similar account was also proposed by Wimmer (1999), who pointed out that phonological and orthographic representations of words have distinct anatomical locations. Robust association between phonological and orthographic word-forms can form only if the flow of activation between those two domains is rapid enough.

Theories of this type can successfully accommodate data showing that naming ability is better at predicting specifically orthographic skills than is phonological recoding (e.g. Manis, Seidenberg & Doi, 1999). They are also consistent with studies showing that children who are slow at reading require more practice to familiarise themselves with orthographic patterns. Words and sub-lexical orthographic patterns that are familiar elicit faster responses than matched unfamiliar patterns, yet people slow at naming words require significantly more previous exposures before the same degree of familiarity-related facilitation is observed (for a review see Wolf & Bowers, 1999). If the number of exposures is held constant, then children slow at naming show a smaller increase in the speed of reading isolated words (Bowers & Kennedy, 1993). Slow naming is more detrimental for quick recognition of whole words than of sublexical orthographic units (Levy & Bourassa, 1998, reported in Wolf & Bowers, 1999). Thus, naming difficulties seem related to an impoverished store of sight words.

The hypothesis of a specific link between slow naming and orthographic knowledge has interesting implications for understanding the variability of developmental dyslexia. It is widely agreed that the manifestations of developmental reading difficulties may be classified on the ‘phonological-surface’ continuum. Phonological dyslexia is characterised by particular problems with decoding (evidenced by difficulties in nonword reading) whereas a presenting problem of surface dyslexia is the lack of a good sight vocabulary (evidenced by errors on reading exception words). A standard explanation of surface - phonological difference invokes the dual route models of word recognition developed in the context of adult, skilled reading (see chapter 1). The phonological form of dyslexia is understood to result from underdevelopment of the indirect route responsible for sublexical letter-sound conversion, whereas the surface form stems from underdevelopment of the direct route, linking orthographic,

phonological and semantic representations of whole words (e.g. Ellis, 1993). An alternative explanation is provided within the connectionist framework. The reading profile of phonological dyslexia is best simulated by the network whose capacity to represent phonological structures has been damaged (e.g. through removing connections between phonological units). Reading in surface dyslexia, on the other hand, bears more resemblance to the output of an undamaged, yet ‘inexperienced’ network - one in an early point of its learning curve, before it has mastered the entire training corpus. Surface dyslexia can also be simulated by a network with limited computational resources (i.e. removed hidden units) that is slower at learning all types of stimuli (Harm & Seidenberg, 1999; Stanovich, Siegel & Gottardo, 1997; Manis et. al., 1996). This implies that phonological dyslexia is a developmental deviance, whereas surface dyslexia a developmental delay. Such a formulation is consistent with the human data. Phonological dyslexies usually exhibit a deficit of phonological awareness and nonword decoding (in comparison with reading-age matched controls), whereas surface dyslexies show performance very similar to younger participants of similar reading age, though worse than chronological age controls (Stanovich, Siegel & Gottardo, 1997). Those differential patterns are often explained in terms of the interplay between (innate) deficits of phonological awareness and reading experience (print exposure). Children with large phonological awareness deficits are compromised in all aspects of their literacy development, and develop the phonological form of dyslexia. Children whose phonological awareness deficit is milder, but accompanied by very limited reading experience, will acquire basic decoding skills but have small sight vocabularies (orthographic lexicons) thus demonstrating surface dyslexia. However, since slow naming speed is specifically related to the difficulties in the acquisition of a sight vocabulary, surface dyslexia may also result from a deficit in automatizing grapho- phonological associations. With such a deficit, the process of building the orthographic lexicon from print exposure will be less effective. Even much reading practice may, then, be insufficient to form robust representations of sight words. Unfortunately, most studies of dyslexia subtypes have not included measures of naming speed. However, Murphy and Pollatsek (1994, reported in Wolf & Bowers, 1999) showed that, within a sample of dyslexic children, problems with reading exception words were specifically associated with naming speed, among other factors.

It also seems plausible that the ‘low accuracy’ and ‘low rate’ subtypes of dyslexia, differentiated by Lovett (1987) overlap with phonological and surface

sub types, respectively. Orthographic processing difficulties (problems with acquiring sight vocabulary, or with parallel processing of longer letter strings in general) should manifest themselves in poor reading of exception words - a hallmark of surface dyslexia. The same orthographic difficulties may also require over-reliance on sequential decoding, which should result in slow reading. Orthographies with more regular letter-sound correspondences than English may give few opportunities for misreading irregular words, and there slow reading (accompanying by stubborn tendency to sound words out) may be a primary sign of surface dyslexia (Wimmer, Mayringer & Landerl, 1999). The relationship between the ‘accuracy-rate’ and the ‘phonological-surface’ classifications of dyslexias will be further discussed in chapter 3.

The theory of a specific link between naming and orthographic skills is not without its problems. As noted in the beginning of this section, the dissociation of naming and phonological awareness is only partial, and both skills may be necessary for word reading accuracy (Lovett, 1987). Also, as children’s sensitivity to orthographic patterns and constraints grows with literacy experience, the specific contribution of naming to reading should increase accordingly, yet in fact it decreases with age (e.g. Torgesen & Wagner, 1994). Some studies also demonstrated that slow readers can benefit from repetition practice on naming words just as much, if not more, than the fast readers (Lemoine, Levy & Hutchinson, 1993). Those controversies may perhaps be resolved if a task analysis approach is adopted. This should tease apart distinct components of naming to understand what specific contribution each of them may make into reading. This step was taken by Bowers and Wolf (1999). They hypothesised that those components of RAN tests that are most task-specific are selectively related to lexical and orthographic processing. The non-specific, speed-related components are involved in all aspects of visual and auditory processing and constrain all aspects of reading - for example, they may compromise the development of phonological awareness. A somewhat different analysis of naming tasks was proposed by Manis, Seidenberg & Doi (1999). They emphasise the shared demand for processing arbitrary

sound-symbol mappings as the key reason for the observed associations between naming and orthographic skills. However, they also acknowledge the role of other components that are specific to the continuous (i.e. rapid automatized) naming format. These components (such as sequential eye movements and co-ordination of information from different modalities occurring in rapid sequence) overlap with processes involved in reading connected texts, but are not involved in phonological awareness tasks.

Implicit in Manis et. al.’s (1999) analysis is the possibility that the relationship between rapid naming and literacy is reciprocal - just as the relationship between phonological awareness and literacy. Processes involved in naming may contribute to reading and spelling, but also benefit themselves from reading practice. This is plausible when we consider that even very peripheral and physiologically constrained aspects of reading, such as perceptual span (i.e. the size and shape of effective visual field) are strongly affected by characteristics of the orthography and individual differences in reading skills (Rayner, 1998)^. Also, since reading skills show the strongest relationship with naming of letters, this relationship is probably mediated through the common factor of reading experience. The possible beneficial effect of reading must operate differently for naming than for phonological awareness, however, since explicit training of naming skills does not produce generic gains (i.e. transferable beyond the trained stimuli: de Jong & Vrielink, 2000), unlike phonological awareness training. The possibility of the reciprocal causality between rapid naming and literacy has received little attention so far, and future studies in this area should explore the impact of reading experience on naming more explicitly.