As children become more proficient at reading, their knowledge about letter-to-sound correspondences becomes more developed. Even children taught by the whole-word method will become aware of the similarities in pronunciation that exist between words that are spelled similarly. They therefore develop, and use, two routes to accessing the meanings of written words-one of them via letter-to-sound correspondences, and the other directly. What changes as the child develops is how much each is relied on. Proficient adult readers rely more on the direct route, although individuals (and especially children) may differ quite markedly in how much reliance they place on each of those two routes. But is this the only difference between adult and child readers? It sounds somewhat counter-intuitive to suggest that what adult readers do is most similar to what beginning readers do (relying mainly on a sight vocabulary), and is most different from what intermediary readers do (placing increased reliance on letter-to-sound correspondences). And indeed, this is the subject of considerable controversy, as exemplified by this question: is the `direct' route used by adults the same route that children employ when they recognize a word by sight alone? Some researchers believe it is, and others believe it is not.
The idea that adults, and children too, rely on two different routes to accessing the meanings of written words is not without its theoretical difficulties-how, for instance, do you decide which route to use? And why go to all that effort to establish letter-to-sound correspondences if, as an adult, you will apparently throw much of it away in favour of a more whole-word approach? Is it inevitable that things work this way, or is this just some psychological oddity? In fact, it is inevitable, and this inevitability explains why what children do in China, when they learn to read, is not much different from what children do anywhere else.
On learning that ink blots have meaning
The child's task, when learning to read, is to associate written words with their meanings. This, after all, is the purpose of the written wordto convey meaning. The meaning is provided, so to speak, by the teacher; either by saying the word, or by pointing to a picture, or by having attached a label with that word onto the thing that the word stands for, although to begin with it is nothing more than just a complicated ink blot. This does riot present a problem, because this ink blot (or however much of it the child attends to) can still become associated with the meaning. The way this works is simple: things which tend to occur together gradually become associated with one another. Encountering one thing (e.g. the ink blot) will, after learning, evoke a pattern of neural activity that would be similar to that evoked by the other (e.g. the meaning).
This learning mechanism is driven by things occurring together, and more specifically, the occurrence of one thing predicting the occurrence of the other (see Chapters 9 and 13). But reading involves more than just ink blots and meanings. If each word is said out loud as part of the teaching method, the sounds of each word will also be present. The word-as-ink-blot can therefore become associated both with its meaning directly, and with its corresponding sounds, which are, in turn, associated with that meaning.
But the associations do not stop there, because certain sounds will tend to occur only when certain parts of the various ink blots are present. These parts (the individual letters) occur in different combinations in different words, so the word-as-ink-blot is soon replaced by the word-as-combinations-of-letters. And because these letters will tend to predict those different sounds, they will become associated with those sounds. These associations may form either with help, in the case of children explicitly taught the letter-to-sound correspondences, or without, in the case of children who have not yet been taught those correspondences but who none the less develop an awareness of them.
One complication, in English at least, is that letters which have tended to occur with only certain sounds may, on certain occasions, occur with different sounds-E, t, and G, for example, do not normally occur with the sound that corresponds to the sequence EIG in REIGN. In this case, the combination of letters (or, again, whichever letters the child attends to) will become associated with the actual sounds that are heard, because it is this combination that predicts the appropriate sounds. Things become more complex still when one considers that combinations of letters may become associated not just with the sounds that the child hears, but with the sounds that the child produces when pronouncing each word. Consequently, combinations of letters may become associated with the gestures of pronunciation-with syllables, for instance (see Chapter 10 for more about syllabic gestures).
So there are a lot of different associations, all happening, potentially, at the same time. It sounds terribly complex, with lots of things being associated with lots of other things. But with these complexities comes a very definite advantage-it does not matter whether the script is alphabetic, syllabic, or logographic; the manner in which we are able to learn associations will apply equally well to each one. This is why Japanese, Chinese, Arabic, and English children (in fact, all children) probably start off doing pretty much the same thing when learning to read. The result of their learning will look quite different, though, because the different written forms of these languages mean that different features of the ink blots predict different aspects of the spoken language (predicting either phonemes, syllables, whole words, or a combination of these, as is required in, for instance, Chinese).
This is all very well, but what about the fact that proficient readers of English appear to abandon the alphabetic principle-the letter-to-sound correspondences-when accessing the meanings of written words? On the assumption that individual letter-to-sound pairings are very much more common than individual word-to-meaning pairings, we should expect the letter-to-sound associations to be very much stronger, and very much more influential, than the word-to-meaning associations, because a basic property of associative learning is that the more common the pairings, the stronger the associations. Yet it is those word-tomeaning associations that are apparently favoured in adulthood, not the letter-to-sound ones. How can this be explained?
This is a controversial issue. In all likelihood, a number of factors contribute to the pattern. One of these may be that the direct associations between print and meaning (that is, bypassing sound) are simply more consistent than those from letters to sounds, because of the irregularities of languages like English. Another reason may be that as children get older, they stop reading aloud as part of their reading classes, and start to read for meaning-for themselves. This means that their task, as learners, is not to predict from the print on the page the correct pronunciation for that word, but to predict its correct meaning. The associations from letters to sounds allow such prediction only indirectly. And if we assume (as before) that associations form, and are strengthened, in order to allow such predictions to be made, it follows that the direct associations from print to meaning will be strengthened with each successive pairing of a word and its meaning. So for any given word, the child may progress from a stage when only the letter-to-sound and sound-to-meaning associations are strong enough to enable the word's meaning to be retrieved, through a stage when these associations work together with the direct print-to-meaning associations to retrieve that word's meaning, to a stage when the word's meaning is effectively retrieved on the basis of the direct print-to-meaning associations alone.
An important aspect of this gradual shift is that although the print-tomeaning associations predominate, both sets of associations will be operating, so to speak; it is just that in certain cases (e.g. commonly encountered words) the direct word-to-meaning associations will be stronger, and in certain other cases (e.g. infrequently encountered words) the word-to-sound-to-meaning associations will be stronger. Even in the case of commonly encountered words, because both kinds of association are operating, aspects of the sound structure of the spoken version of each word (aspects of its pronunciation, perhaps) will be retrieved, but not as a prerequisite to retrieving the word's meaning. What this means, in answer to an earlier question, is th
at the direct route used by adults to retrieve the meaning of a word is not quite the same as the route used by children when they recognize words by sight alone. The adult route is a rich, but delicately balanced, combination of associations.
Even though, as adults, we do not generally retrieve meanings on the basis of simple letter-to-sound correspondences, these correspondences are not lost. If they were, we would not be able to read new words that we had never encountered before, or words which we encountered so rarely that the word-to-meaning associations had not formed strongly enough to allow the meaning to be activated on the basis of the visual information alone. This may also contribute, in part, to another paradox we encountered earlier-that the phonics approach to the teaching of reading, not the whole-word approach, apparently leads to better reading later on. Children taught this way are able to read new words they have not seen before, which in turn means that when practising their new-found skill, they are likely to access the appropriate meanings more successfully than children who, having been taught by the wholeword method, cannot cope so well when left to their own devices. As always, practice makes perfect, but only when the practice is successful.
The next step is to attempt to verify all this.
Getting it wrong, yet again
One of the most productive research tools for discovering how children read is an analysis of the errors they produce. If beginning readers recognized words according to letter-to-sound correspondences, we would expect errors to happen when the child applied a letter-tosound correspondence that just happened, because of the irregularities of English, to be inappropriate (e.g. `fatty' for FATE, `fairy' for FARE, or perhaps `ray-gun' for REIGN). We would also expect such children to be able to read new words that they had not seen before, but only when they had regular spellings. But if young readers recognize words by their visual shape, independently of the sounds associated with the individual letters, then we would expect the errors to be related by shape and not sound (e.g. perhaps `dad', for BED). In addition, these readers should do badly with new words, whether regular or irregular. And if children focus more on certain letters in the word (e.g. the first letter) than certain others, then we would expect the errors to share either their sound, or their shape with those letters, but not necessarily with the others. Seeing how these patterns of errors change over time provides an important indication of what kinds of reading strategies the child adopts at different stages in his or her reading development.
Working out what adults do is harder, because they rarely make errors. There are circumstances, though, when even adults make large numbers of reading errors. Admittedly, these circumstances are confined in the most part to psycholinguistic laboratories. In the mid-1980s, Guy van Orden at the University of California in San Diego took advantage of the fact that when adults are asked to say, as fast as possible, whether a word like FEAT is a part of the body, they will often say `yes', even though this response would only be appropriate for the likesounding word FEET. Similarly for HARE and HAIR. A clever feature of this work was that van Orden ensured that such errors are not due to just the physical similarity of the words. If they were, adults should also mistakenly judge FELT, or PAIR, to be a part of the body. But they do not. (The actual examples used by van Orden differed from the ones shown here, although the principle is exactly the same.)
The fact that FEAT-FEET errors were made seems to contradict the view that the sound of a word does not influence the access of that word's meaning. It turns out, in fact, that these errors only occur for words which are infrequent in the language-they tend not to happen for very frequent words in pairs like SUN-SON.
The overall picture with proficient adult readers, then, is that the meanings of the most commonly encountered words are accessed directly, and not via sounds. Less frequently encountered words, and novel words that have never been encountered before, are accessed using letter-to-sound correspondences.
It is beginning to look as if there is not that much more to say about reading. For the printed word to become associated with meaning requires processes that are essentially the same, irrespective of which script, or which language, we read in. As usual, what we do as adults is very much a by-product of the way in which we learn, as children.
An interesting example of one such by-product is the strong feeling we often have that our reading is accompanied by some kind of inner speech (often referred to as subvocalization). Various studies have shown that there is some muscular involvement in this inner speech: we do not just imagine it. They also show that subvocalization happens to a far greater extent in earlier reading than in later reading-earlier reading tends to place greater emphasis on overt vocalization. So reading can be a quite active process, involving more than just the route (whichever route that is) from print to meaning. But the story is not over yet. There is still one more very active ingredient to reading which, although essential, is all too easily overlooked-the role of the eyes.
Moving eyes
Eye movements are a little like breathing-we can take conscious control of them, but we can also let go of that control and allow automatic processes that are unavailable to our consciousness to take over. Much of what happens when we move our eyes during reading is under such automatic control. It is for this reason that we are generally unaware, for instance, that the time we spend looking at any one word can depend on factors such as: its length (shorter times for shorter words); its frequency of occurrence in the language (shorter times for more frequent words, even once length is controlled for); its grammatical function (longer times, for instance, on verbs than on nouns); its predictability from the context (shorter times for more predictable words); and the complexity of what is being read (see Chapter 7 for an example of how this last effect can be used to study language processing).
The information that we receive from our eyes is complex. And whilst it is easy enough to say that the information on a page is associated with sounds, or with meaning, life is not quite that simple. For a start, the same letters can have different shapes (g, G, or g, for example), and different fonts (such as bold or italic) add to the range of different shapes that each letter can have. And yet, once we have learned the lower-to-uppercase correspondence, we have little difficulty interpreting all the different varieties of print, even varieties that we may never have seen before. So when talking about associations between letters and sounds or meanings, we do not mean associations with specific visual images of letters (in which case even s, s, and s would each be different). Instead, what must happen is that, with experience, these different versions of the letter `s' come to evoke the same pattern of neural activity. And it is this pattern, which in effect represents the concept of the letter `s', that is associated with sound or meaning. Exactly how these different images give rise to the same concept is unclear-in all likelihood it is a process that is very similar to the one described in Chapter 9. The idea there was that the patterns of neural activity that are evoked by something come to reflect the contexts in which that something can occur. If things occur in similar contexts, then they will evoke similar patterns of neural activity. The context may include the physical attributes of the things in question, but also other aspects of the context in which these things occur. So no matter what the `s' looks like, each version has something in common, physically, contextually, and predictively, and it is this commonality that is picked up on.
But different letter shapes is the least of the problems that the reader has to contend with. Far more serious is the fact that the print we read is constantly moving. Well, not quite. But whereas the page tends to stay still, our eyes do not. And because our eyes move across the page, the image of that page on our retinas also moves. On the face of it, the way in which the eyes move will necessarily influence the nature of the image that reaches the brain, and will therefore have important consequences for the way in which information can be `lifted', so to speak, from the page.
Perhaps the most basic fact about eye movements
in reading is that the eyes do not move smoothly across the text-the movement is instead quite jerky. Typically, for an alphabetic language like English, the eyes will move in jumps (called saccades) which are about eight characters long and last around 25-30 milliseconds. Most jumps are rightwards in a left-to-right language like English, but between 10% and 15% are leftwards. The size of each jump depends on the script being read-in Chinese it is about 2 characters; in Japanese it is about 3.5 characters; in Hebrew it is about 5.5 characters. What this pattern shows is that the more information you can extract from the characters, the shorter the jumps. Hebrew is informationally more dense than English, because the written words often lack vowels, and the grammatical function words often exist as inflections attached to the content words. So more information gets packed into the same space (as defined by the number of characters). Japanese is denser still, as it uses a combination of logographic and syllabic symbols. So you can pack even more into that same space. And Chinese is the densest, with each character (each logogram) corresponding to an entire word. Interestingly, the size of the script makes little difference to how far the eyes move.
A quite puzzling property of eye movements is that whilst the eye is moving, it is effectively `switched off-information seems to be taken up from the printed page (or whatever we happen to be looking at) only when the eyes are still. This is probably just as well, as little sense could be made of the rapidly moving and exceedingly blurred image that exists during the saccade. Dave Irwin, a vision scientist working in Illinois, has pointed out that we make around 100 000 saccadic movements in the typical day (whilst awake), and if each takes between 25 and 30 milliseconds, that adds up to a total of 50 minutes of `down time' every day.
The Ascent of Babel: An Exploration of Language, Mind, and Understanding Page 21