Orthographic representation and spelling knowledge

LANGUAGE AND COGNITIVE PROCESSES, 2002, 17 (4), 345–370

Orthographic representation and spelling knowledge V.M. Holmes and C.W. Davis Department of Psychology, University of Melbourne, Parkville, Australia In two experiments, advanced learners read aloud printed words preceded by masked primes. In Experiment 1, for above-averag e spellers, correctly spelt primes facilitated identication more than misspelt primes, not only when the target was correctly spelt, but also when the target was misspelt. Despite being the opposite of standard identity and form priming effects, the latter result shows that the prime that produces maximal facilitation is the one that best matches the individual’s internal orthographic representation. In Experiment 2, when words had been spelt very condently, correctly spelt words were preferentially primed by the correct spelling, while words the person had misspelt were preferentially primed by their own misspelling. The latter result cannot be explained by the view that there is a superior representation developed for reading words that is separate from the one developed for spelling. The quality of the single orthographic representation is indexed by spelling knowledge.

In order to acquire expertise in reading and spelling in an alphabetic language such as English, children must typically learn correspondences between individual phonemes and appropriate graphemes. However, on encountering the same printed words repeatedly, they also create representations in memory for the orthographic specications of these words. It is thought that they begin to do this from a very early point in the literacy acquisition process (Ehri, 1986; Funnell, 1992; Perfetti, 1991; Share, 1995). These memory representations are accessed to allow rapid identication of words during reading and to supply information for spelling. Early representations may often capture only some aspects of a word’s spelling, perhaps containing information about the most crucial letters, such as the consonants. The representations will gradually become

Requests for reprints should be addressed to Dr V.M. Holmes, Department of Psychology, University of Melbourne, Parkville, Victoria 3052, Australia. Email: [email protected]. This research was supported in part by funds obtained from the Australian Research Council. We thank Eve Ng, Naomi Malone, and Tracey Holt for their assistance in conducting the research. c 2002 Psychology Press Ltd http://www.tandf.co.uk/journals/pp/01690965 .html

DOI: 10.1080/01690960 14300026 3

346

HOLMES AND DAVIES

more rened on the basis of the child’s continued experience with words, in addition to explicit instruction about spelling. The mature learner who reads and writes frequently would be anticipated to have developed an extremely large set of precise orthographic representations. Most investigators of normally developing children assume that the representations created for use in reading also function as the representations used in spelling (Bosman & Van Orden, 1997; Ehri, 1986; 1991; Perfetti, 1991, 1997). In support of this notion is the close association typically observed between reading and spelling ability. For example, Juel, Grifth, and Gough (1986) found very high correlations between measures of reading, spelling production, and misspelling recognition in beginning learners. However, it is clear that even mature learners can usually read more words than they can spell with complete accuracy. Investigating individuals whose spelling ability lagged markedly behind their reading ability, whom she termed ‘‘unexpectedly poor’’ spellers, Frith (1980) argued that they could read words successfully based on only partial information in the letter sequence, despite possessing incomplete orthographic representations. To exemplify, someone might be able to read separate, even if their representation were something like sep?rate, while they might spell it incorrectly as seperate, using a more common phoneme– grapheme relation for the ambiguous segment. Thus, learners might be able to read many difcult-to-spell words successfully using partial cues, but they will be unable to spell them exactly unless they can consult fully specied orthographic representations (Perfetti, 1997). An alternative explanation of apparent discrepancies between reading and spelling is in terms of the creation of two separate memory systems— one for reading and one for spelling. The observation of dissociations in spelling and reading performance by brain-damaged individuals has led many researchers to propose that people develop distinct reading (input) and spelling (output) representations for words (Hanley & Kay, 1992; Patterson, 1986; Weekes & Coltheart, 1996). When the dual-representation hypothesis is applied to normal individuals, it is assumed that the information contained in reading representations is often more accurate than that contained in spelling representations (Campbell, 1987). In this way, people would be able to read many words correctly that they could not spell exactly. While there is widespread acceptance of the dualrepresentation view in the neuropsychological domain, there is little evidence for such a dissociation in normal adults. For example, Holmes and Carruthers (1998) found that university students could read words that they could not spell just as rapidly and accurately as words that they could spell. However, when asked to select the correct spelling from among their own misspelling and several other phonologically plausible alternatives, the students could not discriminate between the different versions. By

ORTHOGRAPHIC REPRESENTATION AND SPELLING KNOWLEDGE

347

contrast, they virtually always chose the correct spelling over plausible alternatives when they had spelt the word correctly and condently themselves. If the students had possessed superior information in separate reading representations, then their ability to recognise the correct spellings of words they could not spell should have been better than it was. In the present study, we investigated the nature of the orthographic representations accessed during reading, as well as the relationship between reading and spelling representations, using additional evidence to that based on normal reading and spelling performance. Before explaining the logic of our study, we consider rst why it might be that mature learners sometimes fail to acquire precise spelling representations of words. In addition to acquiring typical phoneme–grapheme correspondences, learners have to master positional constraints on grapheme use, simple orthographic regularities and spellings of common inectional morphemes. However, it is the ‘‘word-specic’’ aspects of spelling that cause the greatest difculty. People have to learn which grapheme occurs in a word when there is a choice between alternative graphemes for a particular phoneme, and no known spelling rule seems to motivate the selection. This problem arises even for short, familiar words, for example, knowing that skip is correct rather than scip, or that heat is correct and not heet. For advanced learners, the difculty of remembering word-specic orthographic information is seen most clearly in multisyllabic words. Such words sometimes allow a choice between different consonants, for example, the rst s in consensus is spelt with a c in similar contexts, such as in the words conceive and census. Even more frequently, these words require a choice between single or double letters for medial consonants, for example, the m in accommodate , and they often contain neutral (schwa) vowels, whose spelling can be reproduced in a large number of ways for different words, for example, the i in hypocrisy. Learners can justify the spelling of many apparently arbitrarily spelt words if they understand the word’s derivational morphological structure and can appreciate its relationship with other words whose spelling more directly represents the phonological form (Fischer, Shankweiler, & Liberman, 1985). For instance, a speller who recognises that hypocrisy is related to hypocrite is less likely to misspell the word as hypocrasy (a common misspelling). However, people vary considerably in their knowledge of such morphological relationships (Fischer et al., 1985; Holmes & Ng, 1993). Moreover, people also differ in the ability to apply their morphological knowledge to spelling. For example, many university students misspell innite as innate, although presumably they would not spell nite as nate (Holmes & Castles, 2001). If they do recognise at some level the morphological relationship between innite and nite, they do not seem to use this knowledge to inform their spelling. At the other end of the

348

HOLMES AND DAVIES

spectrum, the morphological knowledge necessary to explain the spelling of some words may be quite obscure. How many people would remember the a in the middle of separate by recognising the Latin root par from parare (to get ready, arrange), which also appears in prepare? Additionally, the particular spellings of imported foreign words can seem quite idiosyncratic to learners: for example, bourgeois has difcult vowels if one does not know common French phoneme–grapheme conversions, while zucchini has difcult consonants if one does not know common Italian phoneme–grapheme conversions. In short, even advanced learners have trouble learning the word-specic orthographic information necessary to spell many familiar English words. While they might be able to read such words, either from superior reading representations or by partial analysis, their spelling representations for these words would be inadequately specied. As a means of examining the details of a word’s orthographic specication for reading, the present study used a type of priming procedure. Priming occurs when a person processes a stimulus more rapidly if they have encountered the same or a very similar stimulus on an earlier occasion. Rather than employing a situation in which the prime and the target are both visible, recent priming procedures have masked the prime, a technique which reduces the likelihood of the participant becoming alerted to the relationship between the prime and the target (Ferrand & Grainger, 1992; Forster, 1998; Forster & Davis, 1984, 1991). In the masked priming technique of Forster and Davis (1984), participants see a forward masking stimulus for about half a second, followed by a lower-case prime stimulus presented briey (about 60 ms), and nally an upper-case target word to which they have to make a speeded recognition response. Participants can rarely report the prime, most not even realising that the target was preceded by anything other than the masking stimulus. Nevertheless, when the prime is identical to the target, relative to a control prime which is completely different from the target, signicant facilitation in response times to the target is obtained. Additionally, primes which are similar but not identical to the target may also produce facilitation, and this is known as form priming. While form priming is usually of smaller magnitude than identity priming, it can be considered as a special case of identity priming (Forster & Davis, 1991; Forster, Davis, Schoknecht, & Carter, 1987). One possible explanation of masked priming is that, even though the prime stimulus is presented so eetingly, it makes contact with a lexical representation, leaving it in a more accessible state for subsequent retrieval of information. Support for the claim that the facilitation occurs during a process of access to lexically stored information comes from the fact that, when the recognition task is one of lexical decision, priming is not


349

obtained for nonword targets. Furthermore, as the prime and the target are displayed for very different durations, priming is unlikely to be due to an effect of visual summation (Davis & Forster, 1994; Dixon & Di Lollo, 1994). Moreover, priming does not result merely from facilitation at the level of letter identication. Even if the prime and the target have four out of ve letters in common, form priming will not occur unless targets are orthographically distinctive, that is, unless they have few orthographic neighbours (Forster & Davis, 1991). The restriction of form priming to orthographically distinctive target words is regarded as resulting from a different match criterion being set for each word. If a word is not confusable orthographically with many other words, the criterion for matching will be more generous than if the word resembles many other words. In the latter case, the word will be activated only by very close matches, to prevent the unmanageable processing load that would result from the activation of too many perceptual candidates (Forster et al., 1987). The procedure of masked priming provides a method for probing the composition of the orthographic representation consulted to read a word. In the rst experiment, we attempted to determine whether the identity of the prime and target is the critical factor in causing maximal priming to word identication during reading, or whether the correspondence of the prime to the internal reading representation is of greater importance. In the second experiment, we examined which is more effective, a prime whose spelling corresponds to the person’s spelling representation, or one whose spelling corresponds to the putatively distinct reading representation.

EXPERIMENT 1 Priming studies generally assume that participants can both read and spell the target words used in the experiments. When this is the case, it is reasonable to assume that whatever representations they use for reading and spelling, these are fully and correctly specied. If so, then regardless of whether there is a common representatio n or a special one for reading, the typical expectation would be that a correctly spelt word would be maximally primed by the identical version, the correct spelling. A common misspelling should act like a form prime, producing some priming, though less than the identical form. However, consider a situation in which the target word is misspelt in an orthographically and phonologically plausible way. Given that the participant can identify the intended word despite the misspelling, the correct spelling, since it matches better the internal reading representation of the target word, should produce more priming

350

HOLMES AND DAVIES

than the misspelt form, even though it is the misspelt form that is identical to the target. In order to test these predictions, we presented correctly spelt or misspelt targets to be read aloud, preceded by either correctly spelt or misspelt masked primes. We chose fairly common multisyllabic words that we thought the participants would mostly be able to spell. However, each word had at least one segment which could be misspelt plausibly in a different way, for example, imediate for immediate. To check that the participants could distinguish between these spellings, we presented them with a spelling choice task in which they had to select the correct from the misspelt version. Any incorrectly chosen words were removed from each participant’s set of response times. In this way, we restricted the trials to those for which the participant had chosen the correct spelling of the word, and for which a correct spelling representation could thus be assumed. We also used the total number of spelling choice errors as an index of orthographic-processing ability, dividing the sample of participants into an above-averag e and a below-average group. We targeted our predictions principally at the better spellers, as we anticipated that they would be more likely to be able to retrieve all the details of their orthographic representations rapidly in the speeded word identication task.

Method Participants. Participants were recruited from the Psychology 1 subject pool at the University of Melbourne. There were 45 students in total: 35 females and 10 males. They were aged between 17 and 21 years, with a mean age of 18 years 7 months. All were native speakers of English. Materials and design. Target items were 90 multisyllabic words, spelt using common phoneme–grapheme correspondences. Half were presented correctly spelt and half were presented misspelt. They were either 8 or 9 letters in length, and were quite familiar, ranging in frequency from 21 to 83 occurrences per million (KucÏ era & Francis, 1967). For words presented correctly spelt, the mean frequency was 42 and the mean word length was 8.5 letters, whereas for words presented misspelt, the mean frequency was 43 and the mean word length was 8.4 letters. A misspelling was created for each word so that it would typically be pronounced in the same manner as the original. Misspellings were produced in ve different ways. A consonant was spelt with either an added letter, e.g., attatched for attached, or a deleted letter, e.g., exellent for excellent, or a substituted letter, e.g., democrasy for democracy. A consonant was either doubled incorrectly, e.g., memmorial for memorial, or made single incorrectly, e.g., gradualy for gradually. A distinctively pronounced vowel was given an alternative


351

vowel spelling, e.g., simbolic for symbolic, forteen for fourteen. Finally, a neutral vowel was given an alternative vowel spelling, e.g., exable for exible, insurence for insurance. The serial position of the misspelling varied from the second to the eighth letter, but the average position across both the correctly spelt and misspelt presentation conditions was 4.2. Target words were preceded either by a prime that had the same spelling as the target, a prime that had a different spelling from the target or a control prime. For correctly spelt targets, same-spelling primes were the correct spellings and different-spelling primes were misspellings, and for incorrectly spelt target words, same-spelling primes were misspellings and different-spelling primes were the correct spellings. The control prime was a completely different word from the target, although it began with the same letter as the target, so as to avoid a ‘‘Stroop-like’’ onset effect (cf. Forster & Davis, 1991). Each control prime was the same length as its target and was chosen from the same frequency band as the target words. The average frequency for the control primes for the correctly spelt condition was 41 and for the misspelt condition it was 43. Three lists were prepared for presentation to a separate subgroup of participants, so that each of the 90 target words was seen only once by a given participant. The three prime conditions were rotated across the lists in such a way that a participant saw 15 items in each of the six different treatment conditions. Each of the different types of misspelling of either prime or target occurred three times within each condition. Order of presentation of items was randomised within six blocks in which each condition was represented, and the order of blocks and items within blocks was different for each participant. Examples of the six conditions can be seen in Table 1. In order to estimate the participants’ knowledge of the spelling of the target words, a paper-and-pencil spelling choice test was constructed, comprising a list of all 90 target words presented in both their correctly spelt and misspelt form. The two versions appeared adjacent to one other,

TABLE 1 Examples of target words and types of masked prime used in Experiment 1 Prime type Same spelling

Different spelling

Control

Target

Target correctly spelt professor category

proffessor catagory

percentage crossing

PROFESSOR CATEGORY

Target misspelt identicle purchace

identical purchase

impressed painting

IDENTICLE PURCHACE

352

HOLMES AND DAVIES

and the participant had to tick the correctly spelt form. This test was presented after the reading aloud task, and was untimed, the participant merely being instructed to be as accurate as possible. The total number of errors made on this test varied from zero to 10. We eliminated any incorrectly chosen words from the set of response times of each participant. We also used the total number of spelling choice errors to divide the sample at the median into an above-averag e group, with 24 participants, and a below-average group, with 21 participants. The mean percentage of incorrectly chosen words for the above average group was 2.1%, and the mean for the below average group was 7.3%. Procedure and analysis. The experiment was controlled by the DMASTR software program developed by K.I. Forster and J.C. Forster at Monash University and the University of Arizona. Items appeared in the centre of a video monitor. The masking stimulus, which was a sequence of alternating ampersands and percentage signs the same length as the target word, appeared for 484 ms. This was followed by the lowercase prime for 57 ms, and then the target word appeared in uppercase for 484 ms. Response times were taken from when the target word was presented on the screen until the person started to speak. Participants wore a headset with a microphone attached, and when they began speaking, a voice-activated relay was triggered, which sent a signal to the computer to cease timing. Participants were instructed to read each word aloud as quickly and as accurately as possible. They were instructed that some of the words might contain a minor mistyping, but that they should still be able to tell what each word was. They were to ignore the mistypings and just to read the words in a normal manner. A number of practice items preceded the test items. Participants’ spoken output was tape-recorded for subsequent determination of any trials on which they made a reading aloud error. These occurred when they stumbled prior to pronouncing the word, mispronounced it, or failed to respond within four seconds. Trials on which they made a reading aloud error were discarded from the response time analyses. One item in the misspelt target condition had to be eliminated from the experiment when it was discovered subsequently that the correct spelling had not been presented in either the spelling choice task or the word identication task. Any response time which exceeded the mean of all of a participant’s responses by two standard deviations was set at that cut-off value. This procedure inuenced fewer than 1% of responses. Statistical analyses were performed on participant values across items, and on item values across participants, and signicance in both analyses was required for an effect to be considered reliable. The signicance level was set at 0.05 in this and the subsequent experiment.


353

Results and Discussion The mean percentage of spelling choice errors for words that were presented as either correctly spelt or misspelt targets is shown for both speller groups in Table 2. Understandably, below-average spellers chose fewer words correctly than did above-average spellers, F1(1, 43) ˆ 96.71, p 5 .001, and F2(1, 87) ˆ 31.19, p 5 .001. There was a tendency for words that were presented as correctly spelt targets to produce fewer errors than words that were presented as misspelt targets, but the effect was signicant only by participants, F1(1, 43) ˆ 16.41, p 5 .001, and not by items, F2(1, 87) ˆ 1.43, p 4 .05. The interaction of the two factors was not signicant, F1(1, 43) ˆ 2.07, p 4 .05, and F2 5 1. The percentage of trials on which participants produced reading aloud errors on words that had been chosen correctly in the spelling task is also given in Table 2. These are averaged across priming conditions, as analysis showed that neither prime type nor its interactions with speller group and target type approached signicance. Overall, the two speller groups made the same number of reading aloud errors, F1 5 1, and F2(1, 87) ˆ 2.33, p 4 .05. All spellers made more reading aloud errors on the misspelt than the correctly spelt targets, F1(1, 43) ˆ 26.21, p 5 .001, and F2(1, 87) ˆ 7.24, p 5 .01, and this effect did not vary signicantly for the two groups, F1(1, 43) ˆ 2.62, p 4 .05, and F2(1, 87) ˆ 3.21, p 4 .05. This means that the misspelling must sometimes have been different enough from the individual’s reading representation to cause the word to be difcult to identify rapidly. Table 3 gives the mean response times for correctly pronounced target words that were presented correctly spelt or misspelt, as a function of prime type and speller group. The mean amount of facilitation for each of the same-spelling and different-spelling primes relative to the control prime is also indicated. There was a tendency for above-averag e spellers to respond overall more quickly than below-average spellers, but the effect was not signicant by participants, F1(1, 43) ˆ 2.55, p 4 .05, though it was signicant by items, F2(1, 87) ˆ 13.23, p 5 .001. Correctly spelt targets TABLE 2 Mean percentage of errors in spelling choice task and mean percentage of reading aloud errors on correctly chosen words for correctly spelt and misspelt target words as a function of speller group in Experiment 1

Speller group Above average Below average

Spelling choice errors

Reading aloud errors

Spelling of target in reading task Correctly spelt Misspelt

Spelling of target in reading task Correctly spelt Misspelt

1.4 (1.8) 5.8 (3.1)

2.8 (2.1) 8.9 (3.1)

4.4 (5.9) 4.0 (3.6)

8.6 (7.5) 10.7 (8.8)

354

HOLMES AND DAVIES

TABLE 3 Mean response times (in ms) for correctly spelt and misspelt target words as a function of prime type and speller group in Experiment 1

Target type Above-averag e spellers Correctly spelt Misspelt Below-average spellers Correctly spelt Misspelt

Mean response time

Amount of priming

Prime type

Prime type

Same spelling

Different spelling

Control

Same spelling

Different spelling

539 (98) 604 (144)

558 (106) 576 (126)

602 (99) 636 (124)

63 34

44 60

606 (125) 636 (118)

597 (96) 652 (131)

641 (87) 692 (96)

35 56

44 39

Note: Standard deviations are in parentheses.

were read aloud more quickly than misspelt targets, F1(1, 43) ˆ 51.33, p 5 .001, and F2(1, 87) ˆ 13.23, p 5 .001, and this effect was of the same magnitude for both speller groups, both Fs 5 1. This shows that all the participants often noticed the misspelling, and this slowed down their ability to produce the correct pronunciation. Additionally, the main effect of prime type was signicant, F1(2, 86) ˆ 44.90, p 5 .001, and F2(2, 87) ˆ 35.29, p 5 .001, indicating that both same-spelling and different-spelling primes produced facilitation. This overall priming effect was not modulated by target type, and was the same for both above-average and below-average spellers, all Fs 5 1. However, the signicance of the interaction between prime type, target type, and speller group, F1(2, 86) ˆ 6.30, p 5 .01, and F2(2, 174) ˆ 2.96, p ˆ .05, indicates that the pattern of priming effects as a function of target type was different for the two speller groups. For above-average spellers, the prime that had the same spelling as the target produced more facilitation than the prime that had a different spelling from the target when targets were correctly spelt, but the prime that had a different spelling from the target produced more facilitation than the same-spelling prime when targets were misspelt. However, for below-average spellers, this differential priming was not observed: there was little difference between the amount of facilitation from same-spelling and differentspelling primes for correctly spelt targets, and a trend for greater facilitation from the prime that had the same spelling (the misspelt form) than the prime that had a different spelling for misspelt targets. To conrm that this interpretation of the three-way interaction was correct, we performed separate two-way analyses for each speller group. For the


355

above-averag e spellers, target type was signicant, F1(1, 23) ˆ 18.14, p 5 .001, and F2(1, 87) ˆ 12.84, p 5 .001, as was prime type, F1(2, 46) ˆ 44.48, p 5 .001, and F2(2, 174) ˆ 40.41, p 5 .001. Crucially, the interaction between these two factors was also signicant, F1(2, 46) ˆ 8.01, p 5 .001, and F2(2, 174) ˆ 8.07, p 5 .01. For the below-average spellers, both target type, F1(1, 20) ˆ 40.86, p 5 .001, and F2(1, 87) ˆ 11.09, p 5 .001, and prime type, F1(2, 40) ˆ 12.76, p 5 .001, and F2(2, 174) ˆ 8.88, p 5 .001, were signicant. However, the interaction between these two factors was not signicant, F1(2, 40) ˆ 1.20, p 4 .05, and F2 5 1. The results of the experiment accorded with our predictions. For better spellers, the correct spelling produced greater priming than a misspelling not only when it was the same as the correctly spelt target, but also when it was a different spelling from the target. The latter result suggests that when someone has to read aloud a misspelt word, they refer to their own internal orthographic representation in order to retrieve the pronunciation. Weaker spellers did not show these effects. Even for correctly spelt targets, they did not exhibit the standard superiority of identity over form priming: the correct form did not lead to greater facilitation than did the misspelt form. Furthermore, the correct form did not prime more than the misspelt form when the target was misspelt. We assume from the fact that these participants were able to select the correct version over the incorrect spelling in the spelling choice task that their orthographic representations in fact contained correct orthographic information. Yet the prime corresponding to this spelling did not preferentially contact the reading representation. Our supposition that they may not have been able to conjure up the details of their orthographic representations rapidly enough in the word identication task seems to have been borne out.

EXPERIMENT 2 The results of the rst experiment showed that, when people have correct orthographic representations, and can access all their contents rapidly, it is the spelling represented in their memory that governs the amount of facilitation of a briey presented prime, rather than the spelling of the target word. In the second experiment, we asked what would be the consequences for priming on a correctly spelt word if a person’s lexical representation contained incorrect information regarding the spelling. Would this misspelling act as a more effective prime than the correct spelling itself? Is there any evidence that people sometimes base their spellings on incorrect as well as merely indeterminate orthographic representations? Campbell (1987) identied two poorly performing university students who consistently misspelled a large number of even relatively simple-to-spell

356

HOLMES AND DAVIES

words in their spontaneously produced written work. Further, Holmes and Carruthers (1998) found that students who were below-average spellers felt very condent about their spelling about 20% of the time when they misspelled a difcult-to-spell word, suggesting that they too might misspell some words consistently. To verify that the students’ condence judgements reected their spelling consistency, Holmes and Carruthers asked average student spellers to spell the same words on three separate trials, and to rate their condence in each spelling. They found that when people were very condent about the spelling of a word on the rst trial, they hardly ever changed their spelling on later trials. This was true whether their spelling was correct or incorrect. By contrast, when people were less condent about their spelling, they frequently changed misspellings over trials. Although they seldom changed correct spellings that were uncondent, very few responses fell into this category. These ndings are consistent with the idea that when people develop stable spelling representations, most are correct, but some are actually incorrect. Other representations are less stable, and thus more likely to be underspecied or indeterminate. In the present experiment, we identied words that people misspelled with great condence, inferring that the underlying spelling representations contained incorrect letter information. By focusing only on words that the participants could read, we created a dissociation between the accuracy of the spelling representations and the accuracy of the presumed separate reading representations. We preceded correctly spelt targets with primes corresponding either to the correct spelling or to the person’s misspelling. We predicted that, if there is just one representation underlying reading and spelling, and its quality is indexed by spelling, then the persons’ misspelling should be a better prime than the correct spelling itself. However, according to the dual-representation hypothesis, only reading representations are at stake in the priming paradigm. There would thus be no reason to expect the person’s spelling representation to inuence performance: the correctly spelt word should still be a more effective prime than the individual’s misspelling. We also presented words that the participants could spell correctly and condently, for which they could be assumed to have accurate reading and spelling representations. Both the single-representation and dual-representation hypotheses would predict customary identity and form priming effects for correctly spelt words. That is, the correct spelling should be a better prime than a misspelling, although both should prime to some degree. We made the preceding predictions in relation to spellings and misspellings that were very condently produced. But we also included words for which participants had produced less condent spellings, on the assumption that the crucial ‘‘optional’’ orthographic details in their


357

spelling representations for these words would be indeterminate, or if not indeterminate, then difcult to retrieve rapidly enough in the word identication task. It seemed likely that both the correct spelling and a misspelling would facilitate word identication to the same extent, whether the person had initially spelt the word correctly or not.

Method Spelling production pre-test A total of 97 Psychology 1 participants undertook the pre-test, all native speakers of English, aged between 17 and 26 years. They received course credit for participating. The sample was the same as the one used for the pre-test in Experiment 2 of Holmes and Carruthers’ (1998) study. The word list used for the spelling pre-test comprised the 56 difcult-to-spell words selected by Holmes and Carruthers (1998). Frequencies of the words varied from 1 to 79 per million, with an average of 11. Word lengths ranged from 8 to 14 letters, with an average of 9.6. Each word was multisyllabic and orthographically distinctive. The major spelling difculties presented by the words were questions of doubling, such as in harassing, how to represent schwa vowels, such as in plagiarism, and how to represent unusual orthographic patterns for words of foreign origin, such as silhouette. These words were much more difcult to spell than those used in Experiment 1, so that a sufcient number of misspellings would be elicited from the participants. The 56 target words were ordered randomly and tape-recorded in this order. Each word was spoken clearly twice, followed by a pause of 10 s in which the participant was to write down the spelling. Participants were provided with a response booklet in which to record their responses. After writing down the spelling for each word, they were required to rate their condence in their spelling attempt on a ve-point scale ranging from very condent to very uncondent. Because some of the words were very low frequency, we did not wish to include any words with which participants were unfamiliar. For each word, a set of four reasonably common alternative meanings was presented in the response booklet, only one of which was appropriate. For example, for the target word asymmetric, the following words were presented: (a) mysterious, (b) irregular, (c) ambiguous, (d) tasteful. After each condence judgement, participants then had to tick which alternative was closest in meaning to the target word. Participants were given two practice items to become familiar with the procedure. The task took about 40 min to complete. Participants were tested in small groups of from one to 10. On average, the 97 participants knew the meanings of 47 of the 56 words, and subsequent analyses were based only on words for which they chose the correct meaning alternative.

358

HOLMES AND DAVIES

They spelled an average of 47% of known words correctly, with the range being 12% to 83%. For each participant, the spelling attempts for known words were sorted on the basis of the condence ratings into two categories. The rst, called here condent for simplicity, included only those spellings about which the participant was very condent. All less condent judgements were classed together, and called uncondent. Word identification experiment Participants. In selecting participants for the word identication experiment, it was not feasible to include the 22 best spellers, as they did not misspell enough words, while conversely, it was not possible to include the 11 worst spellers because they did not spell enough words correctly. In the remaining average band, there were 62 participants who produced between 31% and 64% correct spellings. Of these, 16 participated in the second phase of Experiment 2 of Holmes and Carruthers’ (1998) study, and so were not considered for inclusion here. From the remainder, we selected 23 participants whose misspellings were mostly phonologically plausible, and who had at least several words falling into each of the spelling accuracy by condence categories. The data of three participants were discarded after they were tested in the word identication experiment, because they made so many reading aloud errors (between 35% and 45%). The nal 20 participants comprised 13 females and 7 males, with a mean age of 18 years and 7 months. Materials, design and procedure. The participants in the nal sample knew the meanings of an average of 48 of the 56 words, ranging from 42 to 55, and the percentage of known words they could spell correctly was 45%. A tailored list was prepared for each participant based on their known words. Three primes were presented for each target word, either the correct spelling, a misspelling, or a control word. When the participant could spell the word, the misspelling presented was the most popular misspelling given by the 62 potential participants from the pre-test. When the participant could not spell the word, the misspelling was their own. The control prime was a different word from the target, beginning with the same letter. Very occasionally, the participant’s misspelling began with a different letter from the target, so in these cases the control word began with the same letter as the misspelt prime. Table 4 shows examples of targets and primes from one participant’s tailored list. In this experiment, because each participant had their own list with their own particular subset of items in the different conditions, it was not possible to test responses to different primes for a given word across different versions of a list. Thus, the participant saw a target word three


359

TABLE 4 Examples of target words and primes from one student’s tailored list in Experiment 2 Prime type Misspelling Spelling accuracy

Correct spelling

Condently produced spellings Could spell bourgeois Could not spell plagiarism

Popular misspelling

Own misspelling

Control

Target

plagerism

belatedly portraying

BOURGEOIS PLAGIARISM

privelidge

constancy publicity

CHARLATAN PRIVILEGE

bourgois

Uncondently produced spellings Could spell charlatan charleton Could not spell privilege

times, appearing with a different prime in each of three separate blocks of items. So that participants would not encounter too many unusually spelt words consecutively, and to increase the distance between repetitions of a target word, 54 regularly spelt ller words were distributed through each list, 18 in a block. Filler words were presented only once. They were preceded by either the identical word, a different control word, or a misspelt form with either one, two, or three letters substituted by another. The word identication experiment was run with exactly the same procedure as in Experiment 1, except that naturally no mention was made of any misspelt targets. Across all participants, a total of seven words was lost owing to mistypings in the stimulus lists. In the determination of reading aloud errors, it was noted that participants did not always make an error on the same word across all three conditions. Only words which were read aloud correctly on all three trials were included in the analyses of the response times. This meant that on average, the number of observations per participant was as follows: 10 of the condently spelt words were correct and 5 were misspelt, while 9 of the uncondently spelt words were correct and 15 were misspelt. To gain an idea as to how inaccurate participants’ spellings were of words they could read correctly, but could not spell, the number of spelling errors in which a letter was either substituted, added, omitted, or misordered was determined. Statistical analyses were performed on participant values across items. In experiments where all participants respond to a xed set of items, additional analyses across items are required to ensure that any effects signicant by participants cannot be attributed to one or two atypical items. Such a problem does not arise in the present design, as each participant saw a different subset of the original items, and these were distributed differently among each of the four categories.

360

HOLMES AND DAVIES

Results and Discussion Table 5 shows the mean percentage of reading aloud errors for the target words. These are averaged across prime type, as there were no effects approaching signicance involving the different primes. However, there was a signicant effect of condence, F(1, 19) ˆ 7.03, p 5 .05, indicating that participants read words accurately more often when they were condent of the spelling compared with when they were uncondent. It is noteworthy that people could read these words aloud just as accurately when they could not spell them as when they could, F(1, 19) ˆ 1.35, p 4 .05, and there was no interaction between condence level and spelling accuracy, F 5 1. Examination of the misspellings of words people had read correctly showed that the average number of incorrect letters was 1.1 for condent misspellings, and 1.7 for uncondent misspellings, a difference which was signicant, t(19) ˆ 10.18, p 5 .001. Thus, people’s condent misspellings were closer to the correct spelling than were their uncondent misspellings. The mean response times for correctly pronounced words participants could and could not spell are given as a function of prime type and condence in spelling in Table 6, with the amount of priming for the correctly spelt and misspelt primes relative to the control prime indicated. Overall, targets that had evoked condent spellings were responded to signicantly faster than targets that had evoked uncondent spellings, F(1, 19) ˆ 21.94, p 5 .001. However, people initiated their responses just as quickly overall when they could not spell a word as when they could, F 5 1. There was a signicant interaction between spelling condence and spelling accuracy, such that the condence effect was larger for correctly spelt than incorrectly spelt words, F(1, 19) ˆ 4.74, p 5 .05. However, this interaction is not very revealing, as it does not take into account any potential effects of the type of prime. There was a signicant main effect of prime type, F(2, 38) ˆ 37.37, p 5 .001, indicating that both correctly spelt and misspelt primes produced TABLE 5 Mean percentage of reading aloud errors for words students could and could not spell as a function of con® dence in spelling in Experiment 2 Condence in spelling Spelling accuracy

Condent

Uncondent

Could spell Could not spell

6.1 (6.1) 7.8 (13.3)

10.3 (8.4) 13.2 (9.7)



361

TABLE 6 Mean response times (in ms) for target words students could and could not spell as a function of con® dence in spelling and prime type in Experiment 2 Prime type Spelling accuracy Condently produced spellings Could spell Could not spell Uncondently produced spellings Could spell Could not spell

Correct spelling

Amount of priming

Misspelling Control

Correct spelling

Misspelling

610 (79) 643 (70)

628 (86) 615 (89)

652 (75) 670 (95)

42 27

24 55

659 (97) 654 (74)

657 (93) 639 (82)

690 (102) 697 (88)

31 43

32 58


facilitation overall. The prediction common to both the single-representation and dual-representation hypotheses was that a prime that had the same spelling as the target would produce more facilitation than a common misspelling when words had been correctly spelt. However, for words that had been incorrectly spelt, while the dual-representation hypothesis assumes that the correct spelling would prime better than the person’s own misspelling, the single-representation hypothesis predicts exactly the opposite effect. The results indicated that different priming effects did occur as a function of whether or not the person could spell the word, as the interaction between prime type and spelling accuracy was signicant, F(2, 38) ˆ 4.59, p 5 .05. The pattern of priming was in line with the singlerepresentation view: the correct spelling primed better than a misspelling for words whose spelling was known, but the person’s preferred misspelling primed better than the correctly spelt word for words which had been misspelt. We had expected that such differential priming would principally be obtained for very condently produced spellings. Inspection of the mean amounts of priming for the uncondently misspelt words suggest that this was the case, since there was no difference between the amount of facilitation for correctly spelt primes and that for misspelt primes for correctly spelt words, though there was a trend for misspellings to prime more than the correct spelling for incorrectly spelt words. Despite this apparently different pattern of priming effects for condently and uncondently produced spellings, the interaction between spelling accuracy, condence in spelling and prime type was not signicant, F(2, 38) ˆ 1.29, p 4 .05. However, separate analyses on condently produced and uncondently produced spellings did provide support for our expectation. For condently produced spellings, the main effect of spelling accuracy was

362

HOLMES AND DAVIES

not signicant, F1(1, 19) ˆ 3.04, p 4 .05, but there was a signicant effect of prime type, F(2, 38) ˆ 16.78, p 5 .001. Crucially, there was a signicant interaction between these two factors, F(1, 19) ˆ 4.53, p 5 .05. For uncondently produced spellings, spelling accuracy was not signicant, F 5 1, but prime type was again signicant, F(2, 38) ˆ 19.21, p 5 .001. However, the interaction between these two factors was not signicant, F(2, 38) ˆ 1.28, p 4 .05. The results of this experiment have conrmed the hypothesis that the orthographic representations of words that people can spell, and about whose spellings they are very condent, are contacted by both a correctly spelt and a plausibly misspelt prime during identication, although the resulting facilitation is greater for the correct spelling. However, if people have incorrect information rmly stored in their orthographic representation, then a prime using this misspelling will provide a better match to the target representation than the identical, correctly spelt word, though it too produces some facilitation. The latter nding is very difcult to explain on the basis of separate reading representations that are superior to spelling representations. Finally, when people have less condence in their spellings, roughly equal amounts of priming are obtained for both identical and misspelt forms, suggesting indeterminate representations rather than clearly specied ones.

GENERAL DISCUSSION The aim of the study was to use the masked priming procedure to provide new information about the nature of orthographic representations. Correct spellings and misspellings appeared as masked primes in a word identication task in which orthographically distinctive target words had to be read aloud. In the rst experiment, we found that for above average spellers, the correctly spelt word acting as prime facilitated identication more than did an incorrect but plausible misspelling acting as prime. This is what would be expected on the basis of previously obtained identity and form priming effects (Forster & Davis, 1991; Forster et al., 1987). Our novel nding was that, when the target word itself contained a plausible misspelling, but was still identiable, the correctly spelt word produced more facilitation than an identical prime. The latter effect represents the reverse of standard effects of identity and form primes in masked priming tasks. Although both types of prime resulted in facilitation of word identication, the prime that provided the better match to the internal representation of the word produced stronger priming, even though it was not identical to the spelling of the target. In the second experiment, we evaluated beforehand the particular spellings and misspellings of individual participants, and used their own


363

misspellings as primes. The condence judgements allowed us to distinguish what we could assume were consistent misspellings from misspellings that were likely to be more variable. We found that when participants had spelt a word very condently and their spelling was correct, the correct spelling facilitated identication more than did a popular misspelling, although this produced considerable priming as well. Again, these effects can be seen as typical identity and form priming effects. However, when the participant had spelt a word very condently but their spelling was incorrect, their own misspelling primed identication more than did the correct spelling, although this too primed identication to a certain extent. These ndings represent the opposite of standard identity and form masked-priming effects. The prime that produced stronger facilitation was the one that corresponded to the orthographic specications stored in memory, even though these specications did not conform to the conventional spelling of the word. Such an effect, which has not previously been reported, is analogous to the reverse priming effect obtained in Experiment 1. The results of the second experiment also provide no evidence for reading representations that are distinct from spelling representations, consistent with the ndings of Holmes and Carruthers (1998). The fact that it was the misspelt form that produced greater facilitation than the correct spelling in our second experiment clearly contradicts the idea that a superior reading representation underlies accurate reading when an individual cannot spell a word. Just because someone can read a word accurately, whether in order to retrieve a pronunciation or to access meaning directly, does not necessarily mean that they have consulted a separate reading representation containing precise orthographic information. Rather, the quality of an orthographic representation can only be determined by nding out how the person spells the word (Perfetti, 1997). At least for the long, orthographically distinctive words used in our experiments, an orthographic representation that was incorrect could still be unambiguously linked with a correct phonological representation and an appropriate meaning. Strictly speaking, our conclusions can only be applied to normal individuals such as the university students that we tested. However, we believe that dissociations observed in the reading and spelling of brain-damaged patients can also fruitfully be considered in this light. A patient who appeared to be accessing different representations for reading and spelling, might actually be accessing a common representation using procedures that have been differentially impaired in some way. In contrast to the preferential priming effects just discussed, both experiments also yielded evidence for equivalent facilitation from correct and misspelt primes. In Experiment 1, for below-average spellers, primes that were correct spellings produced about the same amount of priming as

364

HOLMES AND DAVIES

priming that were misspellings, for both correctly spelt and misspelt targets. Since only words for which the participants chose the correct spellings in the spelling choice task were considered, it would be difcult to argue that their representations of these words were indeterminate. We had hypothesised that the weaker spellers, despite possessing precise representations of the words, might be unable to evoke the ‘‘optional’’ specications of their representations under the constraint of rapid responding. In terms of Ehri’s (1980, 1986) conceptualisation of spelling, the word-specic information necessary to spell ambiguous segments in words might be stored as orthographic ‘‘footnotes’’, rather than as part of the basic specication of the more essential graphemes. If a speller had less well entrenched information in their orthographic footnotes, then they might not have adequate time to recover this information when required to read a word as quickly as possible. Some support for this possibility can be found from the ndings of Holmes and Carruthers (1998). In one task, students were asked to classify individual words as rapidly as possible as correctly spelt or not. The words either had correct spellings or were plausible misspellings. When the students knew how to spell the word, they classied correct spellings as correct over 80% of the time. But rather than classifying popular misspellings as correct less than 20% of the time, they classied them as correct on almost half the trials. By contrast, in a task involving untimed forced-choice selection, the students were able to make almost faultless discriminations between correct and misspelt forms. In other words, it seems that the complete details of an orthographic representation may only be consulted if the task allows adequate time and forces a very deep level of processing. While this account explains why no differential priming might be obtained for people with slower orthographic-processing capability, can it also explain the nding of Experiment 2, that correctly spelt and misspelt primes produced similar amounts of facilitation when people had spelt the words uncondently? We have assumed that low condence in one’s spelling comes about when the spelling representation is imprecise or incomplete, and the indeterminate parts are generated at the time of producing the spelling. However, an alternative view is that all representations are fully specied, and regardless of whether these fully specied representations are correct or incorrect, the disambiguating information might again be difcult to retrieve in speeded tasks. A further idea is that individuals create multiple spelling representations for some words, and when this is the case, primes corresponding to these alternatives produce equivalent amounts of facilitation. A mixed position would be to incorporate the idea of storage of multiple spellings into the single-representation view. For example, in the introduction, we proposed that an indeterminate representation for the word separate might be


365

something like sep?rate. However, the single representation might be more like sep(a)(e)rate, with alternative spellings that the individual considered plausible indicated, perhaps with weightings determining the probability of a particular spelling being chosen. On the basis of our research so far, we cannot distinguish amongst these possible ways of accounting for the lack of preferential priming for the uncondently produced spellings. The priming effects that we have obtained in the present study can be interpreted in the theoretical framework offered by Davis, Forster and colleagues to explain identity and form priming (Castles, Davis, & Letcher, 1999; Forster, 1998; Forster et al., 1987; Forster & Taft, 1994. The word recognition system is regarded as developing to allow the reader to discriminate any presented word from the other words in their lexicon as efciently as possible. Word identication occurs by a two-stage process: a set of candidate lexical entries which match the input closely is activated, followed by an orthographic verication stage in which the exact spelling of the closest matching candidate entry is checked. A prime that is identical to the target will pre-activate the pertinent lexical entry during the initial stage of candidate selection. This allows evaluation of the target word to begin sooner than if its entry had not been activated. Savings are also made as the spelling of the target word is veried, as some of the orthographic information will already have been provided by the analysis of the prime. If the target word is very different from others in the lexicon, a form prime will also make contact with the orthographic representation. Savings will thus occur in the initial candidate selection phase, but not in the orthographic verication phase, where the discrepancy between the prime and the target will be detected. This account applies to masked priming situations where it is assumed that participants have correct and complete orthographic specications of the words that they are to identify. However, it is not difcult to extend the logic to cases where the person’s orthographic representations are not correctly specied. When the representation contains an indeterminate segment, an appropriately misspelt prime and the correct spelling will both pre-activate the same set of candidates and be found to match the internal representation equally closely. Finally, if the person’s orthographic information is precisely specied but incorrect, then a misspelt prime corresponding to this information will provide the best match to the internal representation, and act like an identity prime. By contrast, the correctly spelt form will function like a form prime, and will not yield any savings in the orthographic checking phase. An alternative view of word recognition postulates distributed representations arising within a connectionist system rather than lexical representations. Perhaps distributed representations could offer a natural explanation of some of the present ndings. For instance, learning in such

366

HOLMES AND DAVIES

models typically involves reinforcing multiple mappings for a word. In the face of spelling ambiguity and differing individual experience, this learning process might result in some words having imprecise or even incorrect representations. How would masked priming be modelled in a connectionist system? One possibility is that priming is the product of the modication of connection weights, resulting from an incremental learning procedure being applied each time a word is identied (cf. McClelland & Rumelhart, 1985; Rueckl, 1990). This would mean that any encounter with a word would strengthen the connections between the visual, orthographic, and semantic representations of a word, and so enhance the system’s response the next time that word is encountered. One problem with this approach in relation to the processing of a briey presented prime and a subsequent target is that it fails to explain the short-term nature of the masked priming effect. For example, Forster and Davis (1984) have shown that the masked priming effect lasts for less than a few seconds. It seems implausible that a recognition system would act to change connection weights only to change them back again a few seconds later. Plaut and Gonnerman (2000) have proposed a different connectionist method in which they effectively modelled priming as representing a prepresentation of the target. In order to do this, they rst implemented a time-course of processing by altering the standard feedforward computational process to produce a cascade of unit activation (based on McClelland, 1979). Then they set the network’s orthographic activations to those of the prime word. Following this, the network processed the prime for a set unit of time and the prime’s orthography was replaced by that of the target, without any of the other activations in the network being reset. Processing then continued until the semantic activations satised a stability criterion, settling on the meaning of the target word. The latency of the network’s response to the target was taken as the time from the onset of the target to the time at which the stability criterion was reached. Whether either of these models could reproduce the precise pattern of priming effects obtained in the present experiments is difcult to evaluate, since most connectionist models use iterative non-linear procedures. At present, there would seem to be no reason to prefer one connectionist framework over another. An important conclusion from our study is that a person does not have to be a particularly poor speller to possess some orthographic representations that contain precisely specied but incorrect information. Models of normal reading and spelling development have tended to focus on the lack of completeness (Perfetti, 1997) or lack of stability (Ehri, 1997) of the orthographic representations which underlie inaccurate spelling. Clearly, models will need to incorporate mechanisms to explain the development of incorrect orthographic specications. How is it that a perfectly normal


367

individual would develop faulty orthographic representations in the face of repeated encounters with the correct spelling of a word? Presumably, in an initial attempt to spell a multisyllabic word requiring word-specic orthographic information, the person would guess the composition of any unknown sequence using common phoneme–grapheme correspondences, a procedure which would by denition often result in an incorrect spelling. Given the level of tolerance in the community towards spelling variation, this misspelling would stand a good chance of avoiding correction by an independent source. If the misspelling remained uncorrected, it might well be repeated and gradually become established in the memory specication. When coming across the word during normal reading, the individual would simply fail to check the identity and order of all the component letters, and thus would have no reason to rectify the incorrect representation stored in memory. For people to fail to notice that a correct spelling was different from their own version, the misspelling would presumably be quite close to the correct form, as was the case for the very condent misspellings of our participants. Consistent with this general account, normal individuals have been shown to learn spellings much less effectively from simply reading words than from various types of spelling practice with corrective feedback (Bosman & de Groot, 1992). This effect can be seen as an instance of the more general tendency for recall of an event to be improved when the individual can make reference to enacted or embodied knowledge rather than to a more passive encounter with the event (Cohen, 1989; Glenberg, 1997). Finally, we should point out that we have made the assumption throughout the paper that the advanced learners we used as participants read the target words aloud by consulting lexically stored orthographic information. However, an alternative viewpoint is that priming could have occurred via the lexical phonology rather than the lexical orthography. When individuals were less skilled or less condent, they might not have consulted stored representations of the words’ spelling in the reading task, even though the words were in their written vocabulary. Instead, they might have assembled pronunciations using grapheme–phoneme conversions. Note that the multisyllabic target words we used could not reliably be read aloud correctly if the reader depended solely upon grapheme– phoneme correspondences. For English multisyllabic words, even those that are regularly spelt, there is always a choice as to which syllable should be stressed and which vowels should be reduced or pronounced distinctively. To illustrate, all vowels are pronounced distinctively in a word such as magnetic, and it has its stress on the second syllable. But a word such as symphony has a neutral vowel in the second syllable, and it is the rst syllable which is stressed. Given that there would be no guarantee that the phonological output generated from grapheme–phoneme rules

368

HOLMES AND DAVIES

would be correct, people would have to engage in a further step before being able to produce the correct pronunciation. They would have to match the assembled pronunciation to their set of lexically stored phonological representations, choosing the word that was the closest match to the assembled phonological output. Thus, the priming effects we observed would be regarded essentially as being analogous to cross-modal priming effects: a briey presented orthographic stimulus would be considered as speeding up the time taken to select the appropriate phonological representation. However, the available evidence suggests that cross-modal masked priming of this type only occurs with prime durations long enough for participants to be conscious of some of the letters of the prime (Kouider & Dupoux, in press). In our rst experiment, we asked participants after the priming task whether they ever saw anything ash or icker before the targets, and then whether they ever saw any letters between the mask and the target. Of the 45 participants, 40 said that they saw no ashing nor any letters. Five participants said that they sometimes saw something ash just prior to the target, and only one participant said that she occasionally saw some letters. We therefore feel condent that the priming effects we observed in our experiments came about as a result of automatic consultation of the orthographic lexicon, and not as a result of conscious priming of the phonological lexicon. In sum, the ndings of the present experiments have shed new light on the nature of the memory representations that mature learners develop to support reading and spelling. The results indicate that the masked priming procedure provides a valuable method for investigating these processes, allowing us to supplement and extend the inferences that can be drawn from more standard reading and spelling paradigms. Manuscript received September 2000 Revised manuscript received June 2001

REFERENCES Bosman, A.M.T., & de Groot, A.M.B. (1992). Differential effectiveness of reading and nonreading tasks in learning to spell. In F. Satow & B. Gatherer (Eds.), Literacy without frontiers. Widnes, Cheshire: United Kingdom Reading Association. Bosman, A.M.T., & Van Orden, G.C. (1997). Why is spelling more difcult than reading. In C.A. Perfetti, L. Rieben & M. Fayol (Eds.), Learning to spell: Research, theory, and practice across languages (pp. 173–194). Mahwah, NJ: Erlbaum. Campbell, R. (1987). One or two lexicons for reading and writing words: Can misspellings shed any light? Cognitive Neuropsychology, 4, 487–499. Castles, A., Davis, C., & Letcher, T. (1999). Neighbourhood effects on masked form priming in developing readers. Language and Cognitive Processes, 14, 201–224. Cohen, R.L. (1989). Memory for action events: The power of enactment. Educational Psychology Review, 1, 57–80.


369

Davis, C., & Forster, K.I. (1994). Masked orthographic priming: The effect of prime-target legibility. Quarterly Journal of Experimental Psychology, 43, 673–698. Dixon, P., & Di Lollo, V. (1994). Beyond visible persistence: An alternative account of temporal integration and segregation in visual processing. Cognitive Psychology, 26, 33–63. Ehri, L.C. (1980). The development of orthographic images. In U. Frith (Ed.), Cognitive processes in spelling (pp. 311–338). London: Academic Press. Ehri, L.C. (1986). Sources of difculty in learning to spell and read. Advances in Developmental and Behavioral Pediatrics, 7, 121–195. Ehri, L.C. (1991). Development of the ability to read words. In R. Barr, M.L. Kamil, P. Mosenthal & P.D. Pearson (Eds.), Handbook of reading research, Vol. 2 (pp. 385–419). New York: Longman. Ehri, L.C. (1997). Learning to read and learning to spell are one and the same, almost. In C.A. Perfetti, L. Rieben, & M. Fayol (Eds.), Learning to spell: Research, theory, and practice across languages (pp. 237–269). Mahwah, NJ: Lawrence Erlbaum Associates Inc. Ferrand, L., & Grainger, J. (1992). Phonology and orthography in visual word recognition: Evidence from masked nonword priming. Quarterly Journal of Experimental Psychology, 45, 353–372. Fischer, F.W., Shankweiler, S., & Liberman, I.Y. (1985). Spelling prociency and sensitivity to word structure. Journal of Memory and Language, 24, 423–441. Forster, K.I. (1998). The pros and cons of masked priming. Journal of Psycholinguistic Research, 27, 203-233. Forster, K.I., & Davis, C., (1984). Repetition priming and frequency attenuation in lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 689– 698. Forster, K.I., & Davis, C. (1991). The density constraint on form-priming in the naming task: Interference effects from a masked prime. Journal of Memory and Language, 30, 1–25. Forster, K.I., & Taft, M. (1994). Bodies, antibodies and neighbourhood density effects in masked form priming. Journal of Experimental Psychology: Learning, Memory and Cognition, 20, 844–863. Forster, K.I., Davis, C., Schoknecht, C., & Carter, R. (1987). Masked priming with graphemically related forms: Repetition or partial activation? Quarterly Journal of Experimental Psychology, 39, 211–251. Frith, U. (1980). Unexpected spelling problems. In U. Frith (Ed.), Cognitive processes in spelling (pp. 495–515). London: Academic Press. Frith, U. (1985). Beneath the surface of developmental dyslexia. In K. Patterson, J. Marshall & M. Coltheart (Eds.), Surface dyslexia: Neuropsychological and cognitive studies of phonological reading (pp. 301–330). Hove, UK: Lawrence Erlbaum Associates Ltd. Funnell, E. (1992). On recognising misspelled words. In C.M. Sterling & C. Robson (Eds.), Psychology, spelling and education (pp. 87–99). Clevedon, UK: Multilingual Matters. Glenberg, A.M. (1997). What memory is for. Behavioral and Brain Sciences, 20, 1–55. Hanley, J.R. & Kay, J. (1992). Does letter-by-letter reading involve the spelling system? Neuropsychologia, 30, 237–256. Holmes, V.M. & Carruthers, J. (1998). The relation between reading and spelling in skilled adult readers. Journal of Memory and Language, 39, 264–289. Holmes, V.M., & Castles, A. (2001). Unexpectedly poor spelling in university students. Scientic Studies of Reading. Holmes, V.M. & Ng, E.C. (1993). Word-specic knowledge, word-recognition strategies, and spelling ability. Journal of Memory and Language, 32, 230–257. Juel, C., Grifth, P., & Gough, P. (1986). Acquisition of literacy: A longitudinal study of children in rst and second grade. Journal of Educational Psychology, 78, 243–255.

370

HOLMES AND DAVIES

Kouider, S., & Dupoux, E. (in press). A functional disconnection between spoken and visual word recognition: Evidence from unconscious priming. Cognition. KucÏ era, H., & Francis, W.N. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press. McClelland, J.L. (1979). On the time relations of mental processes: An examination of systems of processes in cascade. Psychological Review, 86, 287–330. McClelland, J.L., & Rumelhart, D.E. (1985). Distributed memory and the representation of general and specic knowledge. Journal of Experimental Psychology: General, 114, 159– 188. Patterson, K. (1986). Lexical but nonsemantic spelling? Cognitive Neuropsychology, 3, 341– 367. Perfetti, C.A. (1991). Representations and awareness in the acquisition of reading competence. In L. Rieben & C.A. Perfetti (Eds.), Learning to read: Basic research and its implications (pp. 33–44). Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Perfetti, C.A. (1997). The psycholinguistics of reading and spelling. In C.A. Perfetti, L. Rieben, & M. Fayol (Eds.), Learning to spell: Research, theory, and practice across languages (pp. 21–38). Mahwah, NJ: Lawrence Erlbaum Associates Inc. Plaut, D.C., & Gonnerman, L.M. (2000). Are non-semantic morphological effects incompatible with a distributed connectionist model of lexical processing? Language and Cognitive Processes, 4/5, 445–486. Rueckl, J.G. (1990). Similarity effects in word and pseudoword repetition priming. Journal of Experimental Psychology: Learning, Memory and Cognition, 16, 374–391. Share, D.L. (1995). Phonological recoding and self-teaching: sine qua non of reading acquisition. Cognition, 55, 115–149. Weekes, B., & Coltheart, M. (1996). Surface dyslexia and surface dysgraphia: Treatment studies and their theoretical implications. Cognitive Neuropsychology, 13, 277–315.