In order to find evidence of word processing strategies employed by the present students, the misspellings produced have been analysed formally. The approach was to identify the formal difference between the word as written and its correct spelling. Therefore, the small number of misspellings which I have been unable to interpret with confidence are excluded from the analysis.

Some misspellings lend themselves easily to formal description. For example, the misspelling brovided results from writing the letter �b� instead of the letter �p�. On the other hand, how could one describe the transition from �says� to seaes? There is no doubt that this is a misspelling, and also that its locus is vocalic rather than consonantal. Formal analysis can go no further. In the case of gavernemte, it can be said that �a� has been written instead of �o�, and also that �e� has been added to the end of the word. However, although it is possible to isolate -memt- as a misspelling, does it involve the omission of the first or second �n� of �government�? Certainly there is one consonant missing from the word, and this can be included in the analysis. One can further analyse the misspelling by isolating -em- and stating not the reductio ad absurdum that �e� has been written for �m� and �m� for �e�, but rather interpret that the string �m e� has been transposed.

This approach to formal analysis has two important features. First, it may be possible to isolate in a single misspelling one or more formally accountable errors, and the plausibility of isolation will in many cases depend on personal interpretation. Second, different errors can be described with differing degrees of delicacy. Therefore, it is not possible to give a complete and specific analysis of frequency of error types. On the other hand, it is possible to identify those error types which occur most frequently, and to study in more depth some error types which are of particular present interest.

The consonant-vowel dichotomy seems obvious to an English speaker, and reader, although that would presumably not be the case with speakers of languages, like Arabic, whose orthography is partly �disemvowelled�. However, since misspellings in English are being considered, this dichotomy seems appropriate. Of course, in English the distinction between vowel and consonant is both phonetically and orthographically arbitrary. In English orthography, the letter �y� is often held to be either vocalic or consonantal, depending on its context: vocalic in �say�, but consonantal in �yes�. In the present study I have followed this no doubt unsatisfactory division to some extent. Certainly, our students are evidently taught that �there are five vowels� in English (their pronunciation suggests that they are also taught that there are only five vowel phonemes as well, some of which they avail themselves of). I have drawn a distinction between what I shall call pure consonantal letters, such as �p� in �pig�, and consonantal letters such as �w� or �r�, which I shall for present purposes and at great risk of censure call pure consonantal in �with� and �red�, where they represent approximants, but vocalic-consonantal in �saw� and �door�.

Using this crude and slightly arbitrary distinction between vocalic and consonantal letters, the data reveals that the majority of spelling errors are attributable to vocalic letter errors. Misspellings attributable to pure consonantal errors only (that is, not accompanied by other error types in the same word) account for 20.0% of types and 17.9% of tokens. If vocalic-consonantal errors are added, misspellings attributable to consonantal errors only are 22.9% of types and 20.2% of tokens. When those misspellings are added in which a vocalic error co-occurs, misspellings wholly or partially attributable to consonantal errors account for 33.1% of types and 27.5% of tokens. These figures for consonantal errors indicate the high frequency of vocalic errors.

Misspellings in which vocalic errors occur represent 77.1% of types and 79.8% of tokens. There are essentially four types of vocalic error: omission, addition, transposition and substitution, and I shall consider each in turn.

Vocalic omission accounts wholly or partially for 21% of misspelling types and 22.9% of tokens, and is therefore as a group comparable in importance to consonantal errors. Vocalic omission errors can be categorized into two types: intraconsonantal and final. Each group accounts for 10.5% of types, but the former constitutes 9.5% and the latter 13.4% of tokens.

lntraconsonantal vocalic omission could plausibly be caused by phonetic processing. In this case, the misspelling would be considered a graphic representation of a reduced or omitted vowel sound. Obvious examples from the data are custmers and travling. On the other hand, in tempreture it is also plausible to consider that the second �e� was in fact written as the letter originally seen between �p� and �r�, but the couple �e r� was transposed; then the suffix �ture� was added. In other cases, such as trvellers, the vocalic omission is more probably visually inspired, especially in view of the fact that the word is text derived. In one third of the examples of vocalic omission, the omitted vowel letter represents either a fully stressed vowel sound or one which can only be slightly reduced.

Final vocalic omission is the omission of the letter �e� either at the end of a word or before a suffix such as �s� or �ment�. Examples of the latter could be considered intravocalic, but I have included them here in order to compare final �e� omission with, principally, final �es�, �ed�, and �ly�. �Final� should therefore be taken as morpheme-final. In fact, 67% of the examples are word-final �e� omission, 12% �es� omission and 9% �ed� omission.

As with intraconsonantal vocalic omission, final vocalic omission could plausibly be phonetically inspired. In some cases, the final �e� plays no obvious phonetic role; examples include accurat and befor. In other cases, the phonetic role is clear, for example in translat and prepar. However, for vowel-phoneme poor students, the unaided �a� in these examples may be considered to suffice; also, the final �t� of translat would seem to be a logical completion of a phonetic transcription. In addition, many of the examples are words familiar to the present students which they have heard and no doubt spoken often, and there are relatively few clearly textderived words. Therefore, phonological processing is entirely plausible.

On the other hand, there is some evidence that final vocalic omission may be visually inspired. If one considers the letter which precedes the vocalic omission, the two most common letters are �r� and �t�, which account for 27.5% of the examples. In the corollary of final vocalic omission, final vocalic addition (see below), the two most common letters which precede addition are also �t� and �r�, and they account for 33.2% of the examples. Furthermore, the five most common preceding letters in final vocalic addition, �t�, �r�, �n� �k� and �l�, comprise 63.6% of examples, and these same letters precede 52.5% of examples of final vocalic omission. It is therefore possible that a student has a visual memory of pairs of word-endings such as �-t� and �-te�, �-r� and �-re� and so on, and that he selects one without any phonological input.

Vocalic addition, or epenthesis, accounts for 11.5% of types and 11.3% of tokens in the data. Therefore, a student is nearly twice as likely to omit as to add a vowel letter. As with omission, vocalic addition can be classified into two groups: intraconsonantal and final. However, unlike omission, where the two groups were of similar size, the former group constitutes only 2.5% of misspelling tokens whereas the latter accounts for 8.7%.

A quarter of the examples of intraconsonantal vocalic addition reflect actual or potential native speaker pronunciation, all but two of them being syllabic �l�, as in articals. Other examples may reflect actual or potential Arabic speaker pronunciation, as in insturucti, where the constring is graphically, and may be phonetically, broken. A much larger number of examples of this feature are produced by transposition, and I shall therefore consider this question later.

It is interesting to note which vowel letter has been chosen to break the consonant string. If a native speaker were asked to transcribe an epenthetic schwa using only the English alphabet, in a novel for example, the vowel letter of choice would surely be �e�: he would probably write �insteruction� and �techenical�. In 38% of the student examples, the epenthetic vowel is the same as the contiguous transconsonantal vowel. This can be seen in insturuction and techinical, where the vowel letter after the consonants �r� and �n� respectively is mirrored immediately before the consonant. Such palindromic infixes, which also occur as a result of other error types, may be evidence of right-left orientation embedded in left-right orientation, and would therefore suggest an element of visual processing.

There seem to be three possible significant processing methods involved in intraconsonantal vocalic addition. First, the addition may be a reflection of a potential or actual pronunciation. Second, the error may be a visual mirroring of a letter in close proximity. Finally, and most likely in most cases, initial phonological processing is completed visually.

The majority of examples of vocalic addition are in the final position. Whereas the corollary, final vocalic omission, was seen to be potentially phonologically inspired, this seems less plausible with final vocalic addition. In most examples, and more so than in final vocalic omission, the final �e� plays no phonetic role. What is the �e� in affecte other than an elegant decoration? In some cases, there is a phonetic element. For example, the �e� in plane (for �plan�) establishes the vowel phoneme, in this case incorrectly. The student is clearly retrieving the word �plane�, which figures inordinately in the Preparatory Year writing component, from his visual memory. Therefore, final vocalic addition seems to be preponderantly visually inspired.

It should be remembered that vocalic omission of final �e� is most common after the letters �t�, �r�, �n� �k� and �l�, and that these letters account for 52.5% of examples of final vocalic addition. This fact, and the significantly larger number of tokens of final than of intraconsonantal vocalic addition, lends support to the hypothesis that final vocalic omission is a result of visual processing, during which a choice is made from pairs of word-endings such as �-t� and �-te�, �-r� and �-re� in the visual memory.

Vocalic transposition is the incorrect positioning of a vowel letter in a word. This feature occurs in 8.8% of misspelling types and 9.5% of tokens. Vocalic transpositions can be categorized into three types according to their formal cause. First, a vowel and a contiguous consonant may be transposed, as in kinves and froks (knives and forks). Of course, this could be considered a consonantal rather than a vocalic transposition. However, since I have reserved the term consonantal transposition for those instances involving consonants only (examples of which are designed and Enlgish, and which account for 1% of tokens), I describe misspellings involving the transposition of a contiguous vowel and consonant as vocalic. This first type comprises 67% of all vocalic transpositions by type, and 54% by token.

In only four cases does this type of transposition produce a plausible native speaker phonetic representation, and these all involve syllabic �l�, as in peopel and sampels. However, in 38.8% of types, and 43.4% of tokens, the transposition serves to break up consonant clusters, as in toursim. As mentioned before, the same happened in some cases as a result of intraconsonantal vocalic addition. This function of an epenthetic vowel, known whimsically in generative phonology by the heirs and successors to Chomsky as �cluster busters�, may reflect Arabic phonology.

According to Al-Ani (1970), an Arabic word may have two contiguous consonant phonemes medially or finally, but never initially. Three contiguous consonant phonemes are impossible, and if such a string is formed across word boundaries, an epenthetic vowel breaks the cluster. Taking all examples of epenthetic vowels which formally break consonant strings, 41% of types and 46% of tokens break a three-consonant or four-consonant cluster, which are impossible in Arabic; the others break a two-consonant cluster. This is strong evidence of phonological processing.

Two features may temper judgement of this type of epenthesis as proof of phonological processing. Whereas initial consonant clusters are illegal in Arabic, only 7% of tokens involve the breaking of initial clusters, and in 13% of the tokens, Arabic-illegal initial clusters either remain intact, as in transilated, or are formed by the epenthesis itself, as in claculations. Also, of the medial and final two-consonant strings which are broken, only five types break clusters which are considered non-existent in Arabic according to Al-Ani. Moreover, the breaking of any cluster usually involves the formation of another cluster, as in siklls or exucses. Intuitively, I suspect that some visual processing is involved in these two, and other examples. Nevertheless, it seems most probable that a vowel breaking a consonant cluster to produce a misspelling is a result of phonological processing.

On the other hand, the second and third types of vocalic transposition are apparently the result of visual processing. The second type of vocalic transposition is the transposition of two contiguous vowel letters, and it accounts for 25% of types and 40% tokens in the vocalic transposition group. The much greater incidence of tokens is caused by the single type thier. In fact, well over 80% of tokens of this type involve transposition of the strings �e i� or �i e�, and this constitutes 3% of all misspelling tokens. Other examples include jiont, mian and fuor. The first two contain strings (�i o�, �i a�) which are common in English; the string �u o� is less common. It seems clear that misspellings of this type are the result of visual processing. The final type, also, can plausibly be considered visually inspired. It consists of a small number of misspellings involving transvocalic transposition, such as bigen, complixety and costums (�customs�), where two vowel letters are interchanged across a consonant or consonant cluster.

Vocalic substitution is the replacement of a single intraconsonantal vowel letter by another. It accounts for 23.1% of all misspelling types and 19.5% of tokens. The following table shows the distribution of vocalic substitution according to the vowel letter written and the correct vowel letter; the numbers are the number of types and, boldface in parenthesis, the number of tokens.

Of the 30 substitutions possible, 24 actually occur. By far the most common substitutions are e-->i (�e� written for correct �i�), i-->e, and a-->e. These three comprise 41.6% of vocalic substitution tokens. When the next three most common substitutions, e-->a, a-->o, o-->a, are included, 67.3% of tokens are accounted for. I shall therefore consider these substitutions in the light of possible phonological and visual processing methods.

A vowel letter may be said to have a canonic phonetic role. For example, in the word �better�, the �e� represents the phoneme /e/ and in �middle� the �i� represents /i/. On the other hand, the less the vowel phoneme is stressed, the less canonic it will be. This is the case for �e� in �specific� (perhaps schwa or /i/) and for the second �i� in �ministry� (also perhaps schwa or tending towards /i/). For speakers, this may offer a measure of the extent to which a vocalic substitution misspelling is phonologically inspired. For example, a 15-year old British schoolgirl of my acquaintance wrote enfisis (instead of �emphasis�). The vowel letters chosen for the unstressed second and third vowel phonemes are a credible representation of those phonemes by their canonically associated letters. This is clearly a phonologically inspired misspelling, and it is significant that the assimilation is also represented orthographically as �n�. On the other hand, had she written imphasis, the misspelling would clearly be a result of visual processing, since it can be assumed that she is aware of the canonical association of �e� and /e/.

However, for an Arabic learner of English it cannot be assumed that his English vowel-phonemic system is as rich or well defined as for a native speaker. The misspelling well (for stressed �will and testament�) by a native speaker would be visually inspired. However, the same misspelling by an Arabic speaker could plausibly be the result of an attempted graphic representation of an Arabic-influenced vowel phoneme. In other words, a student may have chosen from three or four vowel letters the one which he believes best represents his pronunciation of the vowel phoneme. Therefore, in the present study it is not possible to determine the extent to which vocalic substitution is the result of phonological processing.

On the other hand, the potential influence of visual processing can be examined to some extent. Given that a student may have difficulty choosing the correct letter for a stressed vowel phoneme, and has a much wider choice for reduced stress phonemes, he may choose a vowel letter that is in his short-term visual memory, a letter from the word he is looking at in the text or in his long-term visual memory. This could lead to a palindromic infix, where the same vowel letter is mirrored about a consonant or consonant cluster, as in cannat (�cannot�).

Disregarding monosyllabic words, where visual input would be across word boundaries and for which data has not been prepared, the percentages of tokens containing palindromic infixes within the six most common vocalic substitutions are as follows:

This table suggests, apart from the striking fact that the string �e (consonant) e� occurs in 27.6% of all tokens exhibiting vocalic substitution, that not all examples of vocalic substitution should necessarily be considered a result of phonological processing. The misspellings cannat and componies are plausibly phonologically inspired, the former to a greater degree than the latter. However, an element of visual processing also seems entirely probable.

As detailed above, consonantal errors are far less frequent than vocalic errors in the misspelling data. Like vocalic errors, consonantal errors can be categorized into four groups: substitution, addition, omission, and transposition. The numbers of tokens in each group as percentages of all consonantal errors are roughly 45%, 33%, 16% and 6% respectively. Such a categorization is fraught with methodological difficulties. However, it helps to identify specific commonly occurring features. I shall therefore not describe or define the methods used to produce these categories, but rather consider only those three features which are frequent enough to be of interest: substitution involving a limited number of letters, common examples of omission, and omission or addition in identical double consonant strings.

Consonantal substitution occurs in 17.0% of all misspelling types and in 14.3% of tokens. The most frequent consonantal substitutions are p-->b (that is �p� written for �b�), of which there are 32 tokens, and s-->c, with 23 tokens. Conventional wisdom has it that the writing of �b� for �p� is a common Arabic speaker error, and Ibrahim (1978), explaining the absence of /p/ in Arabic, gives four examples, each of which involves b-->p substitution. For the record, the present data includes only twelve tokens of this substitution, or 0.58% of all tokens, and b-->p and p-->b together account for 2.1% of tokens. Given the absent /p/ in Arabic, one could expect phonological processing to produce misspellings such as cabablity rather than apility, both examples from the data. In fact, the latter is nearly three times more common than the former. Also, there is a potentially visually inspired palindromic infix in cabablity. Furthermore, it is difficult to imagine how the misspelling doupt could be a result of phonological processing. Another absent phoneme in Arabic is /v/. The data has 9 tokens involving f -->v and 3 tokens with v -->f substitution. I conclude that errors connected with absent Arabic consonantal phonemes are not as frequent as is commonly thought, and that they are not necessarily phonologically inspired.

The second most common substitution is s-->c. This and its opposite c-->s account for 41 tokens. In all but two of the tokens, the letter written represents /s/, and this is clearly a result of phonological processing. The students have chosen from the pair �s c� to represent the phoneme. In the other two examples, racing (�raising�) and plasec (�places�), the substituted letter corresponds to /z/. While the first example is still plausibly phonologically inspired, the second seems to be rather a misorientation, the string �ces� being reversed, and therefore caused by visual processing. However, these substitutions are no doubt overwhelmingly phonological.

The second statistically important feature is consonantal omission, which occurs in 8.3% of all misspelling types and 6.8% of tokens. The most frequently occurring examples consist of omission of consonant letters which are part of the orthographic representation of sibilants or fricatives. In these cases, the correct orthographic representation consists of a consonant cluster, and the students have omitted one of the consonants. For example, in Englis, the final phoneme is represented by the single letter �s�, the �h� being omitted. Similarly, the first phoneme in �the� is minimally represented in reater (�rather�), and likewise the first phoneme in �chop� in sckiched (�sketched�). This last example shows how, in one case by addition and in the other by deletion, two phonemes, /k/ and the first phoneme in �chop�, have been represented by consonant strings which in other common lexical contexts, such as �back� and �much�, would be correct. In the first type of consonantal omission, it appears that a choice has been made from visual memory between various consonant letter graphemes potentially representing a given phoneme, but only part of that grapheme has actually been written.

The third important feature, found in 7.1% of all misspelling types and 5.5% of tokens, occurs across the consonantal addition and omission groups. In the addition group, there are misspellings such as writting, where a single consonant is duplicated, and from omission, examples like writen, in which one of a pair of identical consonants has been omitted. In the two examples given, the single or double consonant has a phonetic role, and this applies to at most 10% of the tokens. However, since these misspellings produce an incorrect phoneme, they cannot be considered phonologically inspired. It is scarcely plausible that the use of a single �t� in writen is an attempt at orthographic representation of the vowel phoneme in �write�.

In over 90% of the tokens, the single or double consonant plays no phonological role. The doubling of a consonant inflected forms to preserve the vowel phoneme, as in �sit-->sitting�, or the retention of a single consonant for the same purpose, as in �site-->siting�, appear to be rules which the present students apply arbitrarily without reference to their phonemic significance. In other words, they have a visual memory of pairs of single and double consonants, from which they make a phonologically uninformed choice.

To sum up, this analysis has highlighted the most common types of miand discussed textent to which visual or phonological processing may be involved in the production of both vocalic and consonantal misspellings. Less frequent types of misspelling have not been accounted for, and the relative importance of visual and phonological processing cannot be quantified precisely. However, it is clear that visual processing, and right-left orientation interference, plays a significant role in the production of misspellings by the present students.