Routes to read by
One of the most extensively studied issues in research on reading aloud is the use, in various tasks, of grapheme phoneme conersion GPC, also referred to as graphophonic transposition GPT or phonological recoding. Under discussion is first, whether there is a pathway that takes readers through a rule-based system that converts written strings into forms for pronunciation; second, whether this conversion happens prior to lexical access; and third, the extent to which the use of such conversion is under strategic control.
One extreme position concerning the grapheme–phoneme route is that it is obligatory, and involves what is known as an articulatory loop between the visual buffer used in visual processing and the lexicon. As a consequence, the pronunciation of words we read has to become available, even if only inwardly, before we can recognise them from our lexicon. Inward rehearsal of the spoken forms of words is known as subocalisation. Less extreme positions do not insist on this subvocalisation, but still maintain that there is conversion between spelling and sound during reading, even when this is silent reading.
A key question is therefore whether lexical access in visual processing is mediated or at least influenced by sound forms. I.e. do we first recognise the graphemes, then the phonemes corresponding to those graphemes, and then from there follow the same route as is used in spoken word recognition, as shown in Figure 9.4?
Orthographic written forms are of course historically and developmentally secondary to spoken forms. Children normally learn spoken words first and only subsequently learn written words and thus make the connections between written forms and meanings. It is therefore tempting to think that for visual word recognition to be successful a grapheme phoneme conversion system just needs to be plugged onto a pre-existing spoken word recognition system. However, a link between the written and spoken forms of a word does not have to involve recoding of subword units such as conversion of letters to phonemes. Instead, the whole written word may give access to a spoken form of the complete word, which then leads to lexical access, as in Figure 9.5.
On top of this, even if the process of learning to read suggests some kind of mediation of sound forms as in Figure 9.4, the mature skill of reading may not be based on mediation, but may rather proceed from whole word forms as in Figure 9.5.
Additionally, of course, the irregularity of grapheme–phoneme correspondences in many languages as well as differences in writing systems as outlined in Chapter 7 casts doubt on the reliability of the grapheme- phoneme route for those languages.
Evidence for grapheme–phoneme conversion
We can pronounce nonwords. For instance, you can probably find a pro nunciation for <florp> even though you may not have encountered this form before, and it is likely that native speakers will agree on the pronunciation of this form. This suggests that there may be some element of subword translation that takes us from spelling to pronunciation. Note that this need not involve recoding from individual letters to individual sounds, but might instead involve larger subword elements such as the onsets of words (e.g. the initial consonant cluster <fl> and the rhymes or word bodies (e.g. <orp> ).
The regularity effect noted above also supports some sort of spelling to-sound recoding. The processing advantage for words with a regular spelling–sound correspondence falls naturally out of a system in which the regularities are represented in terms of rules used during the processing task. In early research using a naming task (Baron & Strawson, 1976), it was found that participants can initiate their reading aloud of words with regular grapheme–phoneme correspondences more quickly than that of words with irregular correspondences (compare cowl-fowl-howl and bowl; boss-loss-mass and gross).
Note though that the task involves both the recognition of the written form and the production of a spoken form of the word. The regularity effect could therefore originate in the production process rather than in recognition. It turns out that if participants’ productions of the word are delayed (i.e. they are not allowed to say the words as soon as they see them) then the advantage for regular words is removed (Gough & Cosky, 1977). This suggests that the regularity effect is indeed in the initial recognition of words, not in their production. However, other research removed the speaking component altogether, using a visual lexical decision task with regular and irregular forms. Such tasks showed no consistent effect of regularity on real word decisions, but a clear effect on nonword decisions (Coltheart, 1978).
A particularly robust effect turned out to be the pseudo homophone effect (Rubinstein Lewis & Rubinstein, 1971). Pseudo homophones are non-words that would sound like real words if they were pronounced, such as <blud>. Such nonwords take longer to reject in a lexical decision task than other nonwords, indicating that there is some obligatory spelling–sound recoding that makes the corresponding real word available (in this case blood), and this then interferes with the lexical decision response. The effect is clearly influenced by the spelling–sound correspondences that speakers have been exposed to, since the pseudo homophone effect is not found for word bodies that do not exist in actual words, even though their pronunciation might be predictable. For example, the effect is found for <beaf> (as a pseudo homophone for <beef>) but not for <befe> (Vanhoy & Van Orden, 2001).
Such a collection of findings has been taken to suggest that there is both a direct route from the written input to the lexicon, without any spelling–sound conversion, and a mediated route which uses spelling sound correspondences. This second route is used in nonword recognition and in certain tasks that require reading aloud. The pseudo homophone effect then arises when the mediated route produces a form that o l be a real word, and which has to be re-checked against the written form, leading to longer processing times.
Dual-route models
Such a dual-route model (Coltheart, Rastle, Perry, Langdon & Ziegler, 2001), with a whole-word route for practised forms, and a grapheme phoneme route for novel forms, is shown in Figure 9.6. This type of model has received widespread discussion in the context of how children should be taught to read, in particular in the debate between an alphabetic or phonic method, which involves identifying spelling–sound correspondences, and a whole-word or look-and-say’ method, in which children learn to associate the sound of a word with its overall visual pattern. Our focus will be on the role of such a model in adult reading, and in particular on the part it has played in the discussion of various types of acquired dyslexia see later in this chapter.
Evidence that both routes are used comes from studies that looked at the interaction of regularity and frequency effects in naming tasks. The regularity effect reported in naming tasks is robust only for infrequent words Seidenberg, Waters, Barnes Tanenhaus, 1984. If the words are of high frequency then irregular words (<have>) are read aloud just as rapidly as regular words (<make>). The account given for this is that although both routes shown in Figure 9.6 are used, the direct route from the written word to the stored word is faster, particularly for high frequency words with high activation levels or low recognition thresholds. There is no reason to suppose that the access speed to stored words via this whole-word route is different for regular and irregular words, and this is confirmed by the absence of a regularity effect for the high-frequency words. For low-frequency words, however, it is argued that the spelling–sound conversion route might produce a candidate for pronunciation at about the same time as the whole-word route, which is slow for these low-frequency words. The two candidate pronunciations are then checked against one another. If they agree, as in the case of low-frequency regular words, then the pronunciation of the word takes place. For irregular words, though, the two outputs disagree. For example, the low-frequency irregular word <deaf> would be predicted to have the pronunciation /dif/ rhyming with lea following the spelling–sound rules, but will have the pronunciation /def/ according to the pronunciation information for this word that is contained within its lexical entry. This disagreement needs to be resolved before the word can be pronounced, and this leads to a delay in naming low-frequency irregular words compared to low-frequency regular forms.
Reading for meaning
During normal reading we clearly do not just recognise the form of a word, but we also need to access its meaning. One question is whether access to meaning, i.e. reading for meaning, is phonologically mediated, even in silent reading. One way in which this has been tested is through a category monitoring task (Van Orden, 1987). In this task, participants see individual words and have to indicate whether or not each word belongs to a particular semantic category. For instance, if the category they were classifying for was types of food’, then participants should answer yes’ to <meat>. The results show misclassification of words that were homo phones of within-category words, so that <meet> was for instance often wrongly classified as a type of food. This effect was even found for pseudo homophones, with <soop> also being classified as a type of food.
Such results indicate that phonological mediation does take place. However, the category monitoring task may be sufficiently different from normal reading that it might allow unusual effects to arise. For instance, the task involves a late post-lexical decision about whether the word is a member of the target category. This means that a decision has to be made after the word is accessed from the lexicon, which typically results in a slower response than in many other tasks. During the time taken for this task, phonological forms generated by a spelling–sound conversion process might also become available for lexical search, so that at least some words are accessed based on the routes typically used for spoken word recognition. That is, phonological mediation during visual word recognition may be a rather indirect and marked process. It is nevertheless noteworthy that word activation based on spelling–sound correspondences takes place even when it is counter-productive.
Other evidence for the influence of meaning on word recognition comes from a range of studies showing that words with richer semantic representations are recognised more easily or more rapidly than those with weaker semantic representations. So there are processing advantages, for instance, for words with greater imageability (Balota, Cortese, Sergent-Marshall, Spieler & Yap, 2004), or those with a greater number of related meanings (Rodd, Gaskell Marslen-Wilson, 2002).
Visual access to words
During normal silent reading, it is almost certainly the case that practised readers access familiar visually presented words via a direct route, without spelling-to-sound coding. If it is in fact the case that most real words are recognised visually without such coding, then a further question concerns precisely how they are accessed visually, and whether this involves letter units. That is, do readers get from the written word to the stored word using the right-hand grapheme identification route in Figure 9.7, or via a more direct whole-word recognition route And if words are accessed via grapheme identification, what is the nature of these letter-sized representations, i.e. how abstract are the letter shapes, allowing for variation in font and handwriting (Rapp, Folk & Tainturier, 2001)?
Whole-word recognition relies on visual evidence concerning the word-form as a complete unit, such as its overall word shape. Evidence that word shape is important in word recognition includes the finding that words are recognised more quickly if they are presented in lower case than if they appear in upper case. It is argued that lower case letters, with ascenders (as in the letters <l, t, d>, et). and descenders (as in <p, q, y>, etc). make the word shape more distinctive than PPER CASE, where all letters are the same height. Similarly, typing mistakes are less likely to be detected if they preserve the overall word shape. So for the intended word es, an oversight during proof reading is more likely for <tesf> than for <tesc>. Note though that both these sources of evidence could also be evidence for easier discrimination and faster recognition of individual letters, and hence that words are recognised letter-by letter.
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
