Phoneme awareness span : a neglected dimension of phonemic awareness

The importance of phonological and phonemic awareness knowledge in learning to be literate has been demonstrated in many studies. One dimension of the acquisition of this knowledge, the developmental trend from an implicit awareness of rimes to an explicit awareness of phonemes has attracted substantial interest. A second dimension, a trend in the amount of phonemic knowledge that can be manipulated, or 'phonemic awareness span' is examined in the present study. One hundred and sixty children from Prep to Grade 3 completed five phonological tasks; rhyming, onset-rime segmentation, recognition of initial sound, phoneme segmentation and phoneme substitution. Each task involved words of different phonemic length; five each of three, four and five phonemes. As well, the phoneme segmentation and substitution tasks involved words with six phonemes. Over this grade range the amount of phonemic knowledge students can manipulate increases such that students can deal with more complex phonological tasks. Phonemic length influenced performance for each task. The grade levels at which the influence was greatest varied. For the rhyming task, phonemic length influenced Prep and grade one performance, for onset-rime segmentation and initial sound recognition the influence extended to grade two performance, while for phoneme segmentation and substitution, performance at grades one, two and three were affected. Trends in the erroneous responses showed that while younger students were more likely to display phonological errors consistent with substituting sounds in words, older students were more likely to display errors consistent with deleting sounds from words and saying blends as single sounds. These data support the gradual differentiation of the phonological knowledge into a set of phonemic units, with the weightings between the units learnt gradually and leading to an increasingly elaborated system of links. The implications of the findings for subtypes of dyslexia and for the diagnosis of reading disabilities and for the design of optimally effective instruction are discussed. A plethora of investigations over the last three decades has shown the importance of phonological and phonemic awareness knowledge in learning to be literate. Skilled readers are seen as having access to a phonemic awareness knowledge that comprises representations of individual sounds and sound combinations at the sub-word and word levels. This knowledge allows them to detect and manipulate sound patterns that do not necessarily arise independently as separate units in spoken language. Its acquisition suggests the capacity to reflect on, analyse and extrapolate the sound patterns that make spoken words. Development of this knowledge has been in terms of a trend from the capacity to represent the sound properties of spoken words through an implicit awareness of shared rime in words, an explicit awareness of onsets and rimes to an explicit awareness of phonemes (Lenchner, Gerber & Routh, 1990; Vandervelden & Siegel,1995; Yopp, 1992). The latter aspects of the phonemic differentiation begin with the representation of some consonantal phonemes during onset-rime segmentation. Exposure to text and early literacy teaching contribute to this trend (Stanovich, 1986). The developmental sequence has been well documented and used as a basis for both assessment and instruction. A number of tasks have been used to chart the gradual emergence of this knowledge. These examine the ability to detect and produce words that rhyme, to segment one-syllable words into onset and rime, and into separate phonemes, to count the number of sounds in spoken words, to delete sounds from words and to substitute sounds in words. Differential performance on these tasks is assumed to indicate progress through the developmental sequence. Recent studies have recognised the need to distinguish between changes in phonemic knowledge per se and its use in these types of tasks. Wagner and colleagues propose a phonological knowledge base characterised as "phonological processing abilities (that) have coherence and stability and that is characteristic of other cognitive abilities as opposed to representing relatively malleable measures of reading related knowledge" (Wagner, Torgesen, Laughton, Simmons & Rashotte, 1993, page 100). They note the increasing differentiation of this knowledge with age from kindergarten to second grade and that impoverished phonological representations restrict both memory span and phonological analysis performance. Stahl and Murray (1994) show how linguistic complexity as well as task differences account for phonemic awareness task performance. The knowledge base referred to here needs to be distinguished from the 'general cognitive ability' identified by McBride-Chang (1995) as a component of phonemic awareness and operationalised as IQ. A possible model for this knowledge is in terms of a 'connectionist type structure'. Connectionist theories have been used to explain how readers vocalise written words (Plaut & McClelland, 1993; Seidenberg & McClelland, 1989). They propose that readers have a set of orthographic units that they use to process and code letter strings and a set of phonological units that they use to vocalise the data processed. The links between the two sets of units are acquired or 'programmed' during exposure to spoken and written language. A limitation of the connectionist theories is that they have not specified in detail the structure of the phonological knowledge and do not assist in describing how this knowledge develops during childhood (Hulme, Quinlan, Bolt & Snowling, 1995). The present study examines aspects of this phonological knowledge base. It proposes that phonological knowledge can be thought of as consisting of a set of phonological units linked in a weighted network, with the weighting values determined by the types of sound patterns to which the individual had been exposed. Activation of particular units leads to the activation of other units, principally those for which the links have highest weightings. The weightings are learnt gradually, such that young children become aware gradually that some phonemes are more likely than others to co-occur. The networks of phonemes may change from ones comprising few links between units to ones comprising an elaborated system of links. These units are acquired during exposure to spoken and written language. Changes in the phonological network can account for the developmental trends noted earlier. Onset rime segmentation indicates the explicit awareness of sub-word sound units. An awareness of the initial sound in spoken words indicates the awareness of particular consonants. The links between the phonemes in a rime at this time are based on their co-occurrence in particular words. Children may not be aware that each can occur separately from the others. Their knowledge of the "amp" rime informs them that "m" is linked with "p". As their phonological networks differentiate, they learn that any one sound in an onset or rime unit can be linked with one or more sounds other than those in the unit. They learn that particular sounds co-occur, such that some sounds are more likely to precede or follow other sounds, for example, that "m" is more likely to followed by "p" or "b" than by "t". Individual differences in phonemic awareness can be explained in terms of these models. The observations that some children in the course of phonemic awareness development treat certain blends as a single sound when asked to name the first sound in words (Stahl & Murray, 1994) or insert or add sounds to words they are required to segment may be explained in terms of an inflexibility in the phonemic links. The observation that phonological analysis performance is frequently correlated with memory span (Wagner, Torgesen, Laughton, Simmons & Rashotte, 1993) has been attributed to the possibility that both draw upon phonological representation. The present study elaborates this explanation by explaining short term memory span in terms of the phonemic knowledge available for information encoding and links between phonemic units. Span increases with the extent to which links in phonemic knowledge facilitate the ability to chunk phonemic information. The connectionist model of a phonological network leads to predictions about the amount of phonemic knowledge children can manipulate during the acquisition of phonemic awareness. This aspect of the development of phonological knowledge is frequently overlooked. In terms of task performance, this refers to the number of sounds in the words that are to be manipulated. Most studies of the development of phonemic awareness knowledge do not control this aspect (McBrideChang, 1995; Wagner et al., 1993). Some assume implicitly that performance on words of three or four sounds is sufficient to infer the targeted phonological ability. Others use words comprising a range of sound lengths but without control of this variable. The present study proposes that the number of sounds children can manipulate at any time provides a comparative indication of the complexity of their phonological networks at that time. The various phonological awareness tasks require children to perceive the orally presented information, encode it, manipulate it in ways required by the task and display a response. They encode the information using their existing phonological knowledge. Children whose networks are more elaborated can encode phonemically longer words than those who have a less elaborated phonological network and can, therefore, detect the set of separate sounds in longer words. The words they find easier to manipulate have sound sequences that match the links they have formed between phonemes. The more closely the links between phonemes match the sequence of sounds in words, the more able they are to manipulate the sound information relatively automatically. If they lack some of these network links, they need to use other procedures to attempt to retain the sounds in order. These alternative procedures are assumed to be more attention-demanding. If children manipulate sound information in an attention-demanding rather than automatic way, it is likely that they can perform tasks such as phoneme segmentation for words with fewer sounds before they can perform them with longer words. The number of separate phonemes (not in blended form) that children can manipulate at once is referred to as their 'phoneme awareness span'. It is a measure of the longest spoken words the child can segment accurately into separate sounds. The amount of phonemic knowledge that can be manipulated at once influences word reading ability. As noted earlier, connectionist theories propose that a reader's set of phonological units is linked to a matching set of orthographic units (Seidenberg & McClelland, 1989). The model of phonological knowledge proposed in the present study explains the link between phonemic awareness and reading in terms of the extent of differentiation of the reader's phonemic network. Corresponding phonemic and orthographic units must have the same size and can vary in size from words to single letters / phonemes (Van Orden 1987; Van Orden, Pennington, & Stone 1990). Multiple links exist between corresponding orthographic and phonological codes of the same segment size ( Berninger & Abbott, 1994). Each orthographic code (for example, written words, single letters, letter clusters) and each phonological code (for example, spoken words, phonemes, onset and rime, syllables) contribute unique increments of variance in reading real words and pronounceable non words and in spelling dictated words (Berninger, 1994). Earlier studies support indirectly the prediction that the extent of phonemic differentiation influences the number of sounds that can manipulated during phonemic awareness tasks. In a study of the segmentation of nonsense words into separate phonemes by third and fourth grade children, words of three phonemes were easier to segment than five-phoneme words (McBride-Chang, 1995). The interpretation of the findings merit analysis. The grade range of these children would suggest that they were at the final stage of phonemic development. The nonsense stimuli comprised successive phoneme links that ranged from ones that arise frequently in one syllable words to ones that have a low frequency. Combining the data from this range of sound links may introduce a confounding short-term memory influence; manipulation of stimuli that contain sound links that are unlikely to be encountered in words would be expected to make a greater short term memory demand than those that have familiar links. The observation that task performance could be explained by short term memory ability supports the present proposal of a influence of phonemic differentiation on task completion. Because the third and fourth grade children would be unlikely to have phonemic links within their networks for several of the nonsense words used, they would need to remember the links provided in the stimulus. The present study predicted that for children in the course of phonological development, performance on phonological tasks would be influenced by the number of phonemes in a word; words with fewer sounds would be easier to manipulate than longer words. Task difficulty for words of the same phonemic length is assumed to be influenced by the complexity with which the phonological knowledge needs to be manipulated. Its focus is different from that of McBrideChang's (1995) study in that it examines trends for phonemic links that are typical of spoken language. This study also examines the influence of an immature phonological knowledge on patterns in phonemic task performance. An emerging phonological network may lead to characteristic error patterns. Characteristics of the developing phonemic network at any time may include (1) particular phonemic units not being represented; one might expect that students substitute sounds. (2) particular phonemic units being linked in inflexible or restricted ways such that the activation of one sound always leads to the activation of a second sound; this may lead to the insertion or addition of sounds to responses. (3) particular phonemic units not being represented as separate units; one might expect that students treat certain blends as a single sound when asked to name the first sound in words (Stahl & Murray, 1994) or say vowel-consonant or consonant-vowel units as single sound units when they have not fully differentiated vowels. (4) restricted networks of links so that the set of links for some phonemes may not permit the chunking of twoor threephoneme onset units or threephoneme rime units; one might expect students to delete consonants from otherwise correct responses. It is proposed that (1) to (4) indicate a general developmental trend and that errors patterns in students' responses will indicate these.