|
Article Excerpt
The orthographic depth or transparency of a language, which reflects the grapheme-phoneme consistency of words, has an effect on the acquisition of literacy in young children (Seymour, Aro, & Erskine, 2003; Spencer & Hanley, 2003). Liberman, Liberman, Mattingly, and Shankweiler (1980) suggested that all alphabetic orthographies can be classified according to the transparency of their letter-to-phoneme correspondence or orthographic depth, with shallow orthographies demonstrating letters that are isomorphic to phonemes in the spoken word, and deep orthographies having letter-phoneme relations which are substantially equivocal. Based on this concept, Katz and Frost (1992) proposed their orthographic depth hypothesis, stating that literacy acquisition in a shallow language should be based on a single alphabetic process, involving the language phonology, whereas in a deep language a second logographic process, depending on visual-orthographic structure, will be involved. Seymour et al. find that this hypothesis fits well with their reading data, indicating that there appears to be a threshold of orthographic complexity which results in a step change in the way early reading is acquired. Above this threshold are the shallow European orthographies and below are those classed as deep, including Portuguese, French, Danish and English.
Seymour and Duncan (2004) indicate that their scheme is hypothetical or quasi-intuitive, acknowledging that, although the variation in orthographic complexity has not yet been submitted to a comprehensive computational linguistic analysis, there is general agreement that some orthographies are relatively shallow while others are considered to be deep, containing orthographic inconsistencies and complexities, including multi-letter graphemes, irregularities and morphological effects.
Although such quasi-intuitive approaches are useful in comparative studies that only require broad categorizations, there is an inexorable move to more refined definitions of orthographic depth or transparency that take account of the complexity of the construct including, for example, the variation in individual language characteristics according to the direction of the correspondences between graphemes and phonemes. Finnish and Turkish are highly transparent, showing one-to-one mapping in both phoneme-to-orthography (P-O or spelling) and orthography-to-phoneme (O-P or reading) directions. German and Greek are highly O-P transparent, but have greater complexity in the opposite direction, whereas English is opaque in both directions (Goswami, Porpodas, & Wheelwright, 1997), although English consistency mapping may improve when larger units (onset-rime) are considered (Treiman, Mullenix, Bijeljac-Babic, & Richmond-Welty, 1995).
In terms of refined measures of P-O regularity in English, Hanna, Hanna, and Hodges (1966; hereafter Hanna) computed values by identifying the associated grapheme for each phoneme in their 17,310 word corpus. Their results indicated that 73% of phonemes would be spelt correctly if the most regular graphemic form was used, but that only 50% of English words could be spelt correctly by their computer programme that was based on a detailed analysis of the English language. This fine grain analysis provided the basis for computing the regularity or consistency of English words for investigations of spelling, but has been surprisingly little used.
Berndt, Reggia, and Mitchum (1987; hereafter Berndt) presented their scheme that offered a quantitative description, in the opposite O-P direction, of the grapheme-to-phoneme correspondences (GPC) for English reading. The study was undertaken because 'objective information is lacking on the extent of association between specific letters (and letter clusters) and their pronunciation in words' (p. 1), and there was an implicit challenge to the regular/irregular or consistent/inconsistent dichotomous classifications of English words (Glushko, 1979). They also demonstrated that established probabilities for spelling did not reflect those for reading. For example, although the phoneme \or\ is only rarely (.15) spelt (as in auction), the grapheme is almost always (.95) pronounced \or\. As with Hanna, the associated values were not weighted to reflect the frequency of occurrence of individual words in the language (i.e. values were based on type and not token counts).
Carney (1994) combined the approaches of both Hanna and Berndt by taking account of the two directions of correspondence when producing type and token values for O-P (decoding) and P-O (encoding). Unfortunately, as with Wijk (1966) and Venezky (1970), detailed statistical information is not consistently presented, limiting the value of the study as a basis for quantifying orthographic depth or transparency.
Treiman et al. (1995), investigating the special role of rimes in English orthography, agreed with Berndt, pointing out that, while most researchers 'considered consistency to be a dichotomous rather than a continuous variable' (p. 113), it was possible to calculate continuous type and token consistency values for CVC words. They found a large difference in consistency between consonants and vowels, confirming Berndt's result, which had indicated that much of the O-P irregularity of English lies in the vowels. Rimes were shown to be more regular than vowels and to uniquely explain more of the variance for both reaction time and reading errors. However, the study was limited because it considered only a subset of monosyllabic words (CVCs) and computed values only for O-P consistency. The need to provide bidirectional data was demonstrated by Ziegler, Stone, and Jacobs (1996), who evaluated the consistency of French monosyllabic words and demonstrated that, although P-O for the spelling body (rime) was 79% inconsistent, O-P was only 12% inconsistent for rimes. In other words, of the 88% of words that were thought of as being 'consistent' at the O-P rime level of analysis, 77% were actually P-O inconsistent. They contrasted this with English (Ziegler, Stone, & Jacobs, 1997), which was shown to have similar P-O inconsistency (72.3%), but a higher O-P inconsistency (30.7%). The implications of this were elucidated by Stone, Vanhoy, and Van Orden (1997), who demonstrated that both consistency measures matter in visual word perception, concluding that 'perception is a two-way street'. A conflicting phenomenon was observed in foundation literacy (Spencer, 2001) when O-P measures for common words, derived from Berndt, were not significant predictors in a multiple regression model for word reading in young children, whereas P-O values were highly significant.
Kessler and Treiman (2001) revised their earlier study of consistency measures (Kessler & Treiman, 1997) to include both directions, concluding that spelling and reading are not symmetrical and that spelling is always harder because each part of the syllable is less consistent, although spelling is always helped by the sounds in adjacent parts of syllables. They conclude that the most plausible reading and spelling model is one that 'fundamentally operates on the phonemic level, but can take into account the context in which each phoneme is found' (p. 611).
There is now sufficient evidence from inter-language comparisons, using broad classifications, to support the orthographic depth hypothesis (Oney & Goldman, 1984; Frith, Wimmer, & Landerl, 1998; Paulesu et al., 2000; Spencer & Hanley, 2003). There is also increasing evidence from intra-language comparisons that capitalize on the fact that within any deep language, individual words will vary in their relative depth: some words will show one-to-one mapping in a similar manner to words in shallow languages; other words will have components of varying depth. This relative orthographic depth within a language has an effect on both reading and spelling of individual words. Fischer, Shankweiler, and Liberman (1985) and Burt and Butterworth (1996) used a three level categorization of orthographic transparency (high, medium and low) for words in their spelling studies and demonstrated that orthographic transparency had a strong effect on spelling accuracy. Perry, Ziegler, and Coltheart (2002) found a general pattern for spelling: the smaller the contingency value of a phoneme-grapheme relationship, the smaller the likelihood a word containing the grapheme had of being spelt. Treiman et al. (1995) found that their continuous measure of consistency for elements of CVC words accounted for significant percentages of the unique variance for errors in reading.
Although orthographic depth is related to reading and spelling difficulty for words, there is the added complication of grapheme complexity for deep languages and even for some shallow languages. Complexity does not map on to consistency in a straightforward manner (Laxon, Gallagher, & Masterson, 2002), for example, the three-phoneme words 'might' and 'sly' have the same consistency probability for the \ie\ phoneme (0.1), but different levels of grapheme complexity, which is manifested in their word letter-length. In an early study, Gibson, Osser, and Pick (1963) found that three-letter words were easier for young children to read than longer items, but pointed out that increasing letter-length correlates with more complicated orthographic structures within a word. Holding letter-length constant and increasing phoneme-length was found by Rey, Jacobs, Schmidt-Weigand, and Ziegler (1998) to have an intriguing facilitatory effect on word identification times that were shorter for words having a greater number of phonemes. This suggests that grouping letters into graphemes requires additional processing time and, by holding length in letters constant, increasing the number of phonemes amounts to decreasing the average phoneme complexity by reducing the number of letters per phoneme. Rastle and Coltheart (1998) described this as the 'whammy' and 'double whammy' effect of digraphs on naming latencies for non-words, which may provide support for letter by letter serial processing within their dual route cascade computer model of word recognition. This is disputed by Andrews, Woolams, and Bond (2005), who demonstrated an interaction between O-P typicality and the presence of digraphs that was not well simulated by either the dual route or parallel-processing connectionist (Plaut, McClelland, Seidenberg, & Patterson, 1996) models. As Bijeljac-Babic, Millogo, Farioli, and Grainger (2004) suggest, '... the effects of this apparently simple variable are still not completely understood' (p. 411).
Letter-length effects on naming latency were found to be modulated by word frequency by Weekes (1997), with the effect disappearing for high-frequency words. A similar effect has been observed for studies in which neighbourhood size has a facilitatory effect on low-frequency words only (Andrews, 1989; Laxon, Masterson, Pool, & Keating, 1992). Word frequency has long been a major factor controlled for in studies of reading and spelling. Zinna, Liberman, and Shankweiler (1986) found that word reading accuracy was strongly affected by word frequency, which supported the notion that the initial acquisition of word reading skills was accomplished through rote learning, with frequent words usually identified without analysis of their components. Gernsbacher (1984) concluded that over 20 years of research had demonstrated that word recognition was affected by the familiarity of a word, which was operationalized as the frequency with which a word occurs in printed English text. However, in a similar vein to Bijeljac-Babic et al.'s comments on word length, Carroll and White (1973) warned that, 'word frequency may not be the simple variable that it appears to be' (p. 563). Landauer and Streeter (1973) demonstrated that high-frequency words are likely to contain more regularly occurring phonemic and graphemic patterns than low-frequency words, and Gernsbacher suggested that they also differ along semantic and lexicographic dimensions. Balota, Pilotti, and Cortese (2001) find that 'no single variable has been studied more in psycholinguistics and memory research than word frequency' (p. 639), and note that frequency counts are usually based on extensive collections of print samples, which may be quite different from those of estimates of spoken, heard or hand-written frequency. Stuart, Dixon, Masterson, and Gray (2003), recognizing the importance of word frequency in psychological studies, comment on the need to provide recent systematic counts of the contents of children's early reading materials as it cannot be assumed that children's experience with their reading vocabularies does not change over time, nor that it is necessarily similar to that of adults.
Previous studies have demonstrated that the orthographic depth of words, and in particular the level of transparency of individual phonemes, together with word complexity and frequency have significant unique effects on spelling and reading difficulty of words for young children. The present study was undertaken to investigate the relative strength of both O-P and P-O measures of consistency at the fine-grain phoneme-grapheme level, within a narrow frequency band, in predicting spelling difficulty of words for children in the foundation phase of literacy acquisition.
Method
Participants
In the present study, spelling data were collected for all pupils in an urban Hull primary school, which performs at average national levels in English, mathematics and science. The collected data of spelling performance spanned five year groupings (ages 7 to 11 years), for a total of 207 pupils. Data on the average ages and reading quotients for two reading tests for each year group are shown in Table 1. Reading quotients were based on the administration and marking by the school literacy coordinator of Young's reading tests and NFER Group Reading Test II (NFER, 2005). The Group Reading Test (Young, 1999) and Cloze Reading Tests (Young, 1992) are widely used in UK schools for school years 1-3 and years 4-6, respectively. NFER Sentence Completion Forms A or B are recommended for years 2-4, and Forms C or D for years 5 and above (NFER, 2005). A one-way analysis of variance indicated that there were no statistically significant differences in either measure of reading ability across the age range (Young: F(4, 202) = 1.33, p = .26; NFER: F(4, 202) = 1.13, p = .35). The two tests showed statistically significant correlations ranging from 0.68 to 0.75 (p < .001).
Materials and procedure
Previous studies (Spencer 1999, 2000) collected data for words that had been used in national surveys of the UK School Curriculum and Assessment Authority (SCAA) and had a wide variation in frequency. However, this study was concerned with the most frequent 150 words found in British adult print materials (Hofland & Johansson, 1982). These words were selected because they form the basis for adult reading, representing 50% of the total adult token count. The study was designed to track children's spelling over 5 years to the age when...
|
