Cassidy correctly summaries my theory as suggesting that "the child learning to read uses phonetic skills as a bootstrap for training a non-phonetic reading procedure, implemented as some kind of neural network." However, he finds problems with parts of my theory that I will address.
2. Phonetic reading concerns more then sounding-out of the letters of written words by abstract letter-sound rules; it concerns the oral identification of the product of this sounding-out by a reader's "ears" (something indicated in the first sentence of my target paper). In my reply to Coltheart I expanded upon its nature: "I take it to be primarily a route using phonological decoding processes that can identify words by using the phonological information contained in word spelling to access a reader's oral knowledge of how words sound." In hearing speech we recognise words using our oral knowledge as to how they sound; in phonetic reading we create pronunciations (overtly or covertly) which are then recognised using the same oral knowledge as overheard words.
3. There are good reasons not to identify phonetic reading with the first sounding-out process but the second one, involving access to our knowledge as to how words sound.
4. First, there several possible means of recognising words using the sound information contained in their spelling. Cassidy does not specify the nature of his letter-to-sound phonetic reading but his citing the work of Venezky (1970) suggests an approach based largely on grapheme-phoneme correspondences. Other approaches are possible, however. Patterson and Morton (1985) have suggested that in addition to a mechanism using grapheme-phoneme correspondences a separate one uses orthographic-to-phonology correspondences. The latter uses larger units than the former. These consist of "bodies": the terminal subsyllables of words. There are other possibilities. Goswami (1988) has argued that unfamiliar words are pronounced by analogy with known ones (presumably overcoming the letter differences between the known and the unknown words involves deleting some letter sounds and adding new ones, so even this kind of reading involves some phonetic abilities).
5. In addition, context aids in the phonetic decoding of written words (Snowling, 1989; pp 135-137). This should not surprise, us because most overheard words are so poorly articulated (and subject to noise interruptions) that they cannot normally be identified on their own and are only intelligible when heard in context (Pollack & Pickett, 1963). Thus the processes responsible for the oral recognition of words have to be skilled in using context to aid in word identification. The phonetic identification of written words uses (after the generation of a rough pronunciation from a written word's spelling) the same word recognition processes as hearing and so it must also be able to use context. (It should be noted that this use of context differs from that suggested by Stanovitch and colleagues (1984) in that it concerns a process specific to the oral recognition of words rather than some generalised top-down cognitive use of context).
6. Cassidy notes that "if letter-to-sound rules are to be effective for word recognition, they need to have a reasonable degree of accuracy." But this is true only if they are used on their own -- a much "rougher" sound can be used to identify words if it is supplemented with context (see also note 1). Learning readers accordingly need only crude "sounding out" skills to recognise words phonetically. They do not need the level of expertise suggested by Cassidy.
7. Thus Cassidy has identified a nonproblem -- namely, that the recognition of written words using abstract sound-letter rules may be as hard as recognising them nonphonetically -- but nonphonetic methods of recognising words do not need to be trained using such difficult means of recognising words phonetically.
8. This is my main point. However, Cassidy raises another issue concerning "logographic" reading. I do not dispute that children may go through such a stage. Indeed in Skoyles (1987) I suggested this happens. It makes sense: The best way to learn phonetic decoding skills is to have the ability to recognise a few words, for this can train networks to map orthography onto sound. But Cassidy goes further and suggests that logographic reading "could provide the source of the training examples that Skoyles seeks." They can't -- children do not have a large enough logographic vocabulary.
9. First, most words faced by a learning reader are being read for the first time; hence they are visually unfamiliar and must be read nonlogographically. Of 2,747 different words in a grade 1 and 2 reading corpus, 73% were found to occur 5 times or less and 41% only once (Joam & Share, 1983). Another analysis found that 35% of the 240 words used in the first book of a basal reader appeared only once. As Joam and Share (1983) note, "the trend in current reading programs is towards less repetition and a more diverse set of words." Phonetic reading can explain how words are recognised and it can provide a way to bootstrap the child into reading; logographic reading cannot do this, because a child would not have enough familiarity with words to identify them logographically.
10. Second, although children can easily learn a few words by sight, learning a large vocabulary takes a long time. In contrast, although some time might be needed to learn phonetic reading, once acquired it provides readers with a written vocabulary as large as their oral one. For example, Japanese readers who learn both logographic and phonetic scripts take many years to learn logographs (indeed, the rarer logographs might not be fully learnt; Unger 1987: pp. 92-93), whereas they learn phonetic reading in the first school year (Sakamoto and Makita 1973: pp. 446-448).
11. Finally (and this is not a trivial point, even though it is left until last), Cassidy ignores the research showing that training in phonological skills helps in learning to read (Bryant & Bradley, 1985; Wagner, 1988). Moreover, the child's untrained phonological abilities predict later reading success (Wagner, 1988) - something that follows naturally from an account of reading acquisition in which phonetic rather than logographic reading "bootstraps" children into literacy.
12. There is an interesting "natural" experiment which shows that phonetic reading rather than logographic reading is critical to learning to read. Hebrew is written in two forms: a phonologically opaque kind ("unpointed") which does not spell vowels; and a phonologically transparent kind ("pointed") which does so by means of diacritics (hence the name "pointed"). Most adult reading material is unpointed. If Cassidy were right about the role of logographic reading then the pointed variant would have no role in learning to read because it is visually different from adult spelling. In fact, all Israeli children start off reading pointed text, in spite of its being more complex (and different) visually, compared to unpointed text. Attempts to teach children without the indication of vowels by diacritics have not been successful (Shlomo Bentin, personal communication). Moreover, as with English readers, phonological awareness skills predict latter reading success (Bentin & Leshem, in press).
1. Michael Cole and Sylvia Scribner (1981) found an interesting example of context aided phonetic reading among the Vai people of Liberia, Africa. They use a syllabic script which gives only a rough pronunciation of words: readers frequently reread sentences aloud until they "click" on the correct pronunciation of the words they are sounding out.
Bentin, S. & Leshem, H. (in press). On the interaction between phonological awareness and reading acquisition: It's a two way-street. Annals of Dyslexia.
Bryant, P. E. & Bradley, L. (1985). Children's reading problems. Oxford: Basil Blackwell.
Goswami, U. (1988). Orthographic analogies and reading development, Quarterly Journal of Experimental Psychology, 40A, 239-268.
Joam, A. F. & Share, D. L. (1983). Phonological recoding and reading acquisition. Applied Psycholinguisitics, 4, 103-147.
Patterson K. E. & Morton, J. (1985). From orthography to phonology: An attempt at an old interpretation. In K. E. Patterson, J. C. Marshall, & M. Coltheart (Eds.), Surface dyslexia: Neuropsychological and cognitive studies of phonological reading. Hillside. N.J. Erlbaum.
Pollack, I. & Pickett, J. M. (1963). The intelligibility of excerpts from conversation. Language and Speech, 6, 165-171.
Sakamoto, T. & Makita, K. (1973). Japan. In: Comparative reading: Cross- national studies of behavior and processes in reading and writing. Edited by John Downing, pages 440-465, New York: Macmillian.
Scribner, S. & Cole, M. (1981). The psychology of literacy. Cambridge, Mass: Harvard University Press.
Skoyles, J. R. (1988) Training the brain using neural-network models. Nature 333, 401.
Snowling, M. (1987). Dyslexia: A cognitive developmental perspective. Oxford: Basil Blackwell.
Stanovitch, K. E., Nathan, R. G., West, R. F. & Vala-rossi, M. (1985). Children's word recognition in context: Spreading activation, expectancy, and modularity. Child Development, 56, 1418-1429.
Unger, J. M. (1987). The fifth generation fallacy: Why Japan is betting its future on artificial intelligence. New York: Oxford University Press.
Wagner, R. K. (1988). Causal relations between the development of phonological processing abilities and the acquisition of reading skills: A meta-analysis. Merrill-Palmer Quarterly, 34, 261-279.