V  

Writing vs. Spelling, Grammar, And Speech

I. SPELLING: RULES, REGULARITIES, OR RANDOMNESS?

   1.1 The conditions under which the two modalities of speech and writing operate profoundly affect their organization. We saw in IV.3 how punctuating offers a subsidiary signal system whose correspondence to speech patterns is quite complex and variegated. The same state of affairs confronts us in SPELLING: a set of conventions for recording language in a graphic medium. Languages have adopted a range of possibilities. Taylor (1981) distinguishes: (a) logography, based on whole-word units (e.g. Chinese); (b) syllabary, based on syllable units (e.g. Semitic); and (c) alphabet, based on phonemes (in the sense of II.1.5) (e.g. Latin). Most alphabets employ a small number of symbols in a large number of patterns. Most logographies employ a large number of symbols, such as the approximately 3,000 characters regularly used in Chinese and Japanese (Taylor, 1981: 9f, 20). Syllabaries fall in between these two proportions.

   1.2 The view that English spelling is irrational (rife with entrenched inconsistencies and contradictions, III.3.15) has a long history (cf. survey in Venezky, 1976, 1980). In 1789, Noah Webster proposed a “reform” of spelling to fit speech sounds. Some versions from his Compendious Dictionary of the English Language (1806) have become the American standard, such as the endings ‘-or’, and ‘-ic for the older ‘-our’, ‘-ick’¹ [Hyphens show where word-parts are cut off], and while others like ‘ake’, ‘crum’, ‘ile’, and ‘spunge’, are long forgotten. Later reform attempts had even less success . In 1906, Congress forced President Roosevelt to rescind his order that government printing offices adopt 300 reformed spellings. A “Simplified Spelling Board,” funded by Andrew Carnegie and responsible for Roosevelt’s list, lost its subsidy in 1917, and in {215} 1921, the National Education Association withdrew its support from spelling reforms.

   1.3 Reform efforts encountered strong resistance for several reasons. Spelling proficiency is a cherished benefit of middle-class or upper-class schooling. A reformer writing ‘ake’ for ‘ache’ or ‘ile’ for ‘isle’ is easily dismissed as an anarchist or an ignoramus (cf. I.2.11). Printing is an immense, influential industry that shuns the initial effort and expense of a major reform, regardless of long-term benefits. Beyond public attitudes and costs, English poses a still more forbidding obstacle which had defeated Webster himself (cf. Venezky, 1980: 15f): how to find convincing, workable principles to supplant tradition. Society can hardly be asked to replace one inconsistent system with another.

   1.4 Webster advocated the principle of ETYMOLOGY: the historical source and derivation of words. But English sounds have evolved over time much more profoundly than spelling, e.g., the silencing of final ‘-gh’ or the simplification of initial ‘kn-’ and ‘wr-’. English vocabulary was borrowed from a maze of languages, often preserving their spelling conventions, e.g., initial ‘ph-’ or ‘ps-’ from Greek. All these trends undermined the correspondence between spelling and speech sounds. Today, etymology offers no unifying or consistent basis for spelling reform.

   1.5 Another principle advocated by Webster was ANALOGY (in the sense of III. 1.15): spelling a word like other words with similar sounds. But the growing divergence between spelling and sounds also undercuts analogy (V.1.8). A person who hears the word ‘rheum’ for the first time might conceivably spell it to fit many analogies: ‘room’ (‘doom’), ‘rume’ (‘plume’), ‘rom’ (‘whom’), ‘romb’ (‘tomb’), ‘ruem’ (‘rue’), etc., and never hit upon the real version. Even some of the most common words, e.g. ‘one’ and ‘two’, defy analogy. Webster’s reform was doomed from the start.

   1.6 Recently, the rationality of English spelling has been reasserted in terms of rules. In line with the mentalist approach (cf. II.3.6), Chomsky and Halle (1968) argued that the spelling of surface words is determined by a more abstract “underlying” lexical level of form in a “near optimal” way. Gibson and Levin (1975: 172) postulated a set of “rules” that acts as a “higher-order invariant” despite inconsistencies. Henderson and Chard (1980: 89) proposed an “abstract graphemic code” that “is invariant over changes in particular visual form.” Morton (1980: 124) pleaded for “a system of phoneme-grapheme rules” that operate when a processor initiates a spelling action. All these proposals suggest that the failure of English spelling to represent spoken sounds reliably is offset by further functions. If print had to record sounds faithfully, other factors would be blurred. For example, spelling records the grammatical information that ‘-s’ signals a plural and ‘-ed’ a past, though their pronunciation varies according to context (cf. Chomsky & Halle, 1968; Key, 1969; Murrell & Morton, 1974). If spelling had to be a faithful trace of sounds, these other unities would be {216} clouded (cf. V. I.9). Spellings like ‘resign/resignation’ preserve lexical relations no longer obvious from the sounds (cf. C. Chomsky, 1970; VI. 8).

     1.7 Yet these proposals offer no complete set of rules, but only a few that have convincing illustrations. Empirical support for those rules stated by Chomsky and Halle (1968) has failed to emerge (cf. Simons, 1975; Smith & Baker, 1976; Trammel, 1978; Groat, 1979). Nor could the average writer give a list of spelling rules, aside from a few “rules of thumb” (like “‘i’ before ‘c’ except after e ...”) that work for a tiny portion of the language. For the vocabulary of six-to-nine-year-olds, Berdiansky, Cronell, and Koehier (1969) needed 166 rules to predict the pronunciation of words of one or two syllables from their spelling. Hanna, Hanna, Hodges, and Rudorf (1966) aimed at a larger vocabulary of 17,000 words, but even 300 rules were less than 50% accurate, below the level of fourth-graders (Simon & Simon, 1973). Rules for going from sound to spelling are often not reversible (Gould, 1979). Apparently, the search for rules in an inconsistent system is not paying off.

     I.8 Operationally, a rule is an analogy that succeeds over a noticeable set of instances. Whether this process engenders a limited set of reliable rules depends on the amount of asymmetry and redundancy in the system (cf. III.2.27). Since English spelling frequently offers one symbol for several sounds, or several symbols for the same sound (either within one word or within the whole repertory), these two factors run quite high. Therefore, such lexical relations as cited by Carol Chomsky (1970) (V.1.6) are helpful for some cases (‘resign/resignation’), but harmful for others (‘space/spatial’ vs. ‘face/facial’), and fairly infrequent anyway (Simon & Simon, 1973; Frith, 1980). Moreover, the freezing of spelling over centuries (VI.4) has stopped the power of analogy to promote unity. Thus, most “rules” of spelling narrow down the plausible options, but don’t dictate the exact form of the word.

   1.9 If text production is a multi-phase operation (cf. III.2.5-32), spelling should reflect the interaction of levels (cf. J. Baron, 1973; Meyer, Schvaneveldt, & Ruddy 1974; Smith & Baker, 1976; Coltheart, Davelaar, Jonasson, & Besner, 1977; Barron, 1980; Cohen, 1980; Ehri, 1980; Morton, 1980; Wing & Baddeley, 1980). The diverse pressures exerted on the subsystem of spelling foster a disunity of rules. To spell a word, people work not just with individual phonemes, but also with word shapes, letter patterns, phrase formats, derivations, and meanings. Spelling would have different regularities, depending on how the system is operated (P. Smith, 1980: 42; cf. Albrow, 1972; V.1.24, 38). People soon learn that associating sounds with letters is far more efficient than memorizing by rote a sequence of letters for each word. But sound/letter associations may lose out in conflicts with other motives. For instance, spelling could be harder, not easier, if the plural ending had to be written ‘-s’, -z’, ‘-ez’, etc., depending on how each noun is pronounced (VI.6).

   1.10 Until recently, spelling research was sparse, because, like writing {217} and text production in general, spelling is hard to control via experimental procedures (cf. III.1.1, 3ff). The diversity of the system eludes simple theories decidable via external observation. The act of spelling is itself complex, hard to measure in clock time (Sloboda, 1980: 241).¹ [To obtain a “unitary response,” Sloboda had people figure up the number of letters in a word and then press a numbered key.] .The sophisticated instruments available to phonetics have no counterpart for gauging handwriting or print. Still, substantive spelling research is now well under way.

    1.11 One basic requirement for research is a typology of words such as the following. ESTABLISHED WORDS (like ‘waehead’) are found in a consensus of standard dictionaries, while NON- ESTABLISHED WORDS (like ‘poofter’), including NONSENSE WORDS (like ‘cazart!’), are not. The so-called word superiority effect (that established words are processed better than nonsense words (cf. Polf, 1976) has been found for nearly all experimental tasks (cf. survey in V.1.13-29). Among the established words, REGULAR words approximately fit the common correspondences between sounds and letters and yield groups of analogous spellings, e.g., ‘bloom’, ‘broom’, ‘doom’, ‘room’, ‘zoom’, etc. NON-REGULAR words do not, because some letters are either silent e.g., ‘w’ in ‘two’ or ‘gh’ in ‘fight’, or non-typical, e.g., ‘o’ in ‘one’ or ‘s’ in ‘sure’. On another dimension, PRONOUNCEABLE words can be handled with the speech conventions and articulatory habits of the language (like ‘nuclear’), whereas UNPRONOUNCEABLE words (like ‘nucular’) cannot (cf. Rubenstein, Richter, & Kay, 1975). All established English words ought to be pronounceable, though people have trouble with indeterminate letters, more often vowels than consonants (Henderson & Chard, 1980: 110f), or with long words of many syllables (cf. V.1.42, 47). (“Hard words” is sometimes a cover term for irregular, indeterminate or lengthy spellings, particularly in England.) FREQUENT words (like ‘deterrent’) are statistically common in the overall language while INFREQUENT words (like ‘borborygmus’) are not (cf. Kucera & Francis, 1967). FAMILIAR words (like ‘bibliography’ for me) are well-known to a person, whether or not they are frequent in the language, while UNFAMILIAR words (like ‘rozzer’ for me) are the opposite. BORROWED words (like ‘Iraq’, ‘muumuu’, and ‘Hawaii’) are imported from other languages and reflect the latter’s conventions,. NATIVE words (like ‘kick’ or ‘kill’) are accepted as the basic stock of vocabulary.

    1.12 This typology is necessarily fuzzy. The classes overlap and vary from speaker to speaker, according to vocabulary, pronunciation skills, writing experience, and strategies (cf. V.1.26). Children vary in their skills at devising rules, and thus in their opinions about what counts as regular or irregular (cf. Read, 1975). Someone with no knowledge of Latin wouldn’t recognize Latinate borrowings or realize they can’t contain ‘k’ or ‘w’ (Frith, 1980: 505). A rare Greek borrowing like ‘phthisic’ is an established English word many people would consider nonsensical or unpronounceable.¹ [Trammel (1978: 83) tested for rules with established  words like ‘acaleph’ and ‘elenctic’ he assumed his subjects wouldn’t know; and these established words were processed as non- established words.] {218} Therefore, a provisional word typology like that in V.1.11 only outlines general classes that need to be specified for individuals and groups via empirical research.

   1.13 Like many other domains, spelling research is divided between natural writing (done spontaneously in everyday settings) vs. experimental writing (done on cue in controlled situations) (cf. III.1.1). How far the latter can reveal the former is hard to determine (cf. Wing & Baddeley, 1980: 256). The tighter the controls, the more the activity tends to depart from its customary operation (III.I.3). This reservation applies to many of the trends surveyed below.

   1.14 One research direction is to collect spelling errors and to explore the conditions that favor or impede them. Natural writing offers sparse samples: error rates average slightly over 1% of the words (Chedru & Geschwind, 1972; Wing & Baddeley, 1980). Apparently, feedback catches errors, especially because of the slower time rate than speech (Hotopf, 1980: 296f; cf. IV.2.22). An erroneous letter is often corrected before the next is written down—a sign that feedback is open-loop (cf. III.3.2.2). We should therefore compare error rates from experimental writing. On a paraphrase task, my students had to write in their own words the content of a sentence they heard. When we speeded up the presentation rate to make the students work faster, the proportion of spelling errors rose dramatically, about 50%. This finding shows we must distinguish overload from ignorance as sources of spelling errors (cf. I. 3.3; V.1 15).

   1.15 Once errors are collected, they can be sorted into a taxonomy based on writer type or on word position, or on some mixture of these. The writer type taxonomies are focused on dialect (Briggs, 1969; Cronnell, 1973, 1979; Harrison & Trabasso [Eds.], 1976); or on dialect plus IQ and school attitudes (Desberg, Elliott, & Marsh, 1980); or on the writer’s reading skills (Frith, 1980; cf. VI. 30ff). These studies reveal that, despite popular attitudes about literacy (V.2.1), spelling errors do not indicate low IQ (if IQ is a solid notion at all, cf. I.3.2ff) or poor reading skills. Overload and inattention cause errors just as much as ignorance of the real spelling. Though it is sometimes suggested (Hotopf, 1980: 289; Wing & Baddeley, 1980: 268), consistent, repeated misspellings of a word need not be an ignorance-based error—my writers produced them in spite of other evidence that the correct spelling was known. This discrepancy between knowledge and performance might mean that anxiety regarding an error, especially in a shallow phase like sound/letter linearization, makes it happen despite all exertions to suppress it. So much resources are focused on the expected error that the execution of the correct action is blocked.

   1.16 The word position taxonomies make it clear that errors are not equally likely throughout a word or a passage. Errors are more frequent (a) in the middle of a word than at the end; (b) at the end of a word than at the beginning; (c) in longer words, though not in longer sentences; and (d) toward the end of a sentence, though not toward the end of a whole text {219} (Hotopf, 1980: 303f; Wing & Baddeley, 1980: 259, 261, 266f, 9.70). Only at sentence end should processing load be increased by planning out the next stretch of text (cf. IV.2.30). Conversely, attention seems strong at word beginnings; the initial letter is seldom omitted, and if so, its omission is almost always detected (Hotopf, 1980: 303). This finding concurs with speech perception studies where a word-initial phoneme was the most likely to be correctly perceived (R. Cole & Jakimik, 1978); with “tip of the tongue” studies (III.3.12) where the first letter of the reluctant word was immediately recovered in 57% of the cases (R. Brown & MeNeill, 1966); and with reading studies where the letters at the beginning or end of a word were the best cues in early reading for tasks like recall an& comparison (cf. Timko, 1970; G. Mason & Woodcock, 1973; Rayner & Hagelberg, 1975).

    1.17 From a procedural standpoint, error types should reflect the conditions and limitations of processing (Hotopf, 1980). According to Hotopf, pre-activations (III.3.32; IV2.47) (his “anticipations”) span approximately 6.5 syllables, both in speech (with its rate of 6 syllables a second) and in writing (at only .075 syllables a second); but my data show a large number of spelling pre-activations over only two or three syllables—an important issue for memory spans (III.3.35ff). Hotopf found errors due to selective contamination to be rarer in writing than in speech, perhaps because they are normally detected via look-ahead and feedback before they can be expressed. Errors in function words remained unnoticed in 64% of the time, but in content words only 44% of the time—an effect of the core-and-adjunct principle for allotting resources and control (IV 2.6; VI.46).

   1.18 The position of errors is only one way to classify them. We could also inquire whether they violate conventions of sound or of script (H. Nelson, 1980: 480f); whether they reflect the rules and conventions for segmenting words into “phonemes” and “morphemes” (Cromer, 1980: 412); whether sound-based errors are affected by their contexts (Dodd, 1980: 429f); and so on. All three studies concur that such taxonomies are fuzzy and contestable (cf. VI. 12). Still, if the researcher is consistent and discards doubtful cases, a taxonomy can be helpful for both theory and educational practice. In theory, we might speculate that adjacent letters interfere with each other because of a confusion in execution programs (VI.35 42, 45), like the interference of memory traces for adjacent items in list-learning experiments-the “serial position effect” (IV.I.9) (Wing & Baddeley, 1980: 254, 262). Or, we can try to account for errors in sound patterns (Frith, 1978a), among both children and adults (Frith 1979; Bryant & Bradley, 1980), and especially among poor readers (Barron, 1978b; Frith, 1980), by noting how spellers tend to bring the printed form into a more regular correspondence with the sound (Baron, Treiman, Wilf, & Keliman, 1980). In practice, we can warn writers what sorts of errors to anticipate and watch for (VI.49).

   1.19 Experimental writing allows us to vary the conditions of a task. {220} On dictation tests, people must spell the words they hear. We can compare the test with the people’s abilities to read the same words on another occasion (Bryant & Bradley, 1980). We can dictate a list and ask subjects to find one item that differs from the others (Bradley & Bryant, 1978). We can focus on commonly misspelled words (Baron et al., 1980); regular vs. irregular words (Barron, 1980); or non-established words (Secrist, 1976). Test persons only have sounds to guide their spelling of non-established words, while established words can be checked against some entry in mental storage. Whether this storage really resembles a “lexicon” of words is still unclear (cf. I.4.6; III.3.10; V.1.28). Surprisingly, dictation tests show that some people can spell words they can’t read, e.g., children (Bryant & Bradley, 1980) or alexics without aphasia (Beauvois & Derouesne, 1978).

   1.20 We can dictate a word and ask people to decide the number of letters it has. This task elicits a single unitary response and is easier if a mental image of the word is formed (Sloboda, 1980). The image seems to assist the segmentation of words (cf. V. 1.34)—a task that troubles unskilled readers and spellers, as demonstrated on tests where children were taught to tap once for each phoneme (Liberman, Schankweiler, Liberman, Fowler, & Fischer, 1977; Marcel, 1980).

  1.21 On multiple choice tests, spellers must select the correct alternative (Baron et al., 1980), an easier task than dictation (Tenney, 1980: 218). An error is far more likely for phonologically similar alternatives than for dissimilar ones (Sloboda, 1980; cf. III.3.33). A standard experimental method is to elicit same-different judgements, where people must quickly decide if two presentations are spelled the same (Seymour & Porpodas, 1980; cf. II.1.10). This tactic can be done on established words or established-words, or on a mixture of the two.

   1.22 We can have people perform proofreading, either on words or on whole texts. This line of inquiry proves the relevance of reading skills for spelling (Baron et al., 1980: 166; cf. III.1.9; VI.30ff). Fewer errors are detected (a) among words with similar sounds (MacKay, 1972); and (b) toward the end of the sentence, where they are also more likely to occur (VI.16). Proofreading evidently runs on a shallower depth of processing than normal reading for coherent comprehension (cf. Bobrow & Bower, 1969; III.1.13). Extra resources must be expended on the surface text during its brief contact with short-term sensory storage (cf. III.3.7; VI.34); attention is diverted to otherwise automatic tasks (III.2.31). If processing is heavily loaded, e.g., at the end of a sentence (IV.2.30), or in scrambled texts (Cohen, 1980).¹ [According to Cohen (1980: 147), errors in normal texts are easier to find if they sound right but look wrong; in scrambled texts, errors are easier to find if they both sound and look wrong. Evidently, scrambledness weakens visual expectations formed in coherent contexts, so that sounds become a crucial aid.], error detection degrades. Frith (1980: 506ff) suggests that {221) some readers habitually use a full sampling of the text and therefore spell better than those Who read only from a partial sampling.

    1.23 We can ask people whether and how they would reform spellings (Baker, 1980), e.g., so as to yield only one correct spelling for a given word sound (Baron et al., 1980: 180; cf. V.1.2). Baker’s test persons chiefly followed sound-letter correspondences, except to preserve the unity of plural and past affixes (cf. VI.6, 9) and to avoid spelling two distinct words the same way.

    1.24 We can probe learning for transfer effects (II.2.6) from a pretraining to a testing session (Winnick & Daniel, 1970; Murrell & Morton, 1974). For example, Murrell and Morton found that the recognition of ‘sees’ could be primed for the test by having ‘seen’ in the training session, whereas ‘seed’ had no effect: grammatical relatedness was more influential than sound similarity alone. This finding, like those on proofreading (V. 1.22), demonstrated the multi-level nature of spelling within a complex, interactive system (V 1. 9). 1.25 We can present words with case alternation between capital and small letters, e.g. ‘DEvELopInG’ (Frith, 1980: 509). This tactic can measure how far the processing of written words relies on visual uniformity. For example, the word superiority effect (established words processed better than others, VI. 12) was also found with case alternation on a letter-detection task (McClelland, 1976); and on a same-different decision task (Pollatsek, Weil, & Schindler, 1975; G. Taylor, T. Miller, & Juola, 1977). Word recognition was not impaired by alternating cases, but was by zig-zagging print up and down (Tenney, 1980: 223).

    1.26 We can compare different types of spellers, using some classification like those VI.15. Spellers apparently differ in their knowledge of spelling-sound regularities (Boder, 1973); in their reliance on single sounds vs. on whole syllables or words (Baron et al., 1980); in their sampling of cues from print (Frith, 1980; cf. VI.22); and in their use of mental images of word shapes (Sloboda, 1980; cf. Titchener, 1909; VI.20, 33, 48). Training and age also make a difference. Ehri (1980) found that after a year of instruction, children increased greatly in their use of graphic images to symbolize sounds; Marsh, Friedman, Welch, and Desberg (1980) showed a rising use of analogy for spelling from second grade to fifth grade to college. Still other ways to group spellers into types can be devised, though we need a clearer account of what causes the differences. The reliance on sound-to-letter regularities, for example, may result from instructional methods (Marsh et al., 1980: 352f); or from a more general disposition toward or against learning via rules (Baron et al., 1980: 184); or even from the organic development of the two brain hemispheres (Baron et al., 1980: 186; Frith, 1980: 513f; cf. II.2.4). Clinical studies of brain separation suggest that the right hemisphere does not use sound-to-letter correspondences (cf. Gazzaniga & Sperry, 1967; J. Levy, Nebes, & Sperry, 1971). Spelling {222} performance can be observed among people with language dysfunctions such as aphasia (Cromer, 1980; Marcel, 1980); deafness (Dodd, 1980), and dyslexia (Nelson, 1980; Seymour & Porpodas, 1980). However, neurological evidence needs a process model in order to be significant (II.2.3ff).

    1.27 We can block some aspect of processing by imposing distractor tasks (II.2.6; III.1.28), such as shadowing a message (cf. III.1.2.1; III.3.4.10). Silent reading brought lower comprehension if combined with shadowing an acoustic presentation (B. Levy, 1975). Saying a word over and over hindered children on conversion from print to sound, but not in conversion from print to meaning (Barron & Baron, 1977): given a picture of a shirt, the children had trouble recognizing a rhyme (like ‘hurt’), but not a word naming a conceptually associated object (like ‘trousers’). Repeating non-established words interfered with memory for spelling less than copying them down, whether or not test subjects were deaf (Dodd, 1980). Intriguingly, defects due either to dysfunction or to interference tasks can be overcome with proper motivation and practice. The deaf children studied by Dodd derived a sound-based strategy to spell non-established words via lip-reading, so that their hearing defect was bypassed.

    1.28 This broad sketch of spelling research in VI. 14-27 illustrates the special status of an emerging field of research. No consensus obtains as yet about how issues should be categorized and adjudicated. The role of non-established words is a good illustration. Because people can’t consult a mental lexicon for a non-established word, they presumably must apply their general knowledge and strategies for spelling on tasks like dictation, learning, and recognition (VI.19, 24ff). The spelling of non-established words was recalled better if complete printed units were presented rather than letters or syllables spoken individually in a sequence (Farnham-Diggory & Simon, 1975; Ehri, 1980). But the superiority was not merely one of visual over acoustic processing: printed letters presented in a sequence were an even worse aid than spoken letters (Henderson & Chard, 1980: 108f). The superiority was rather one of parallel processing of whole units over serial processing of either sounds or letters one by one. These non-established word studies thus helped narrow down hypotheses about the roles of language modalities (visual/acoustic) vs. processing modes (parallel/serial).

   1.29 A theoretical account of non-established words remains to be found. Non-established words are only rarely encountered in natural communication, e.g., when people (especially young children) meet established words for the first time (cf. Trammel, 1978); or when advertising creates novel brand names to grab attention’ (see survey in Praninskas, 1968).¹ [Conventions of spelling (e.g. ‘Duz’ and ‘No-Doz’ with a rare final ‘-z’), grammar (‘Winston tastes good like a cigarette should’), and coherence (‘I smoke facts’) are emphatically defied, partly for salience, and partly to harness public resentment against purist usage (cf. I. 2.1 I ff).] Moreover, non-established words, like real {223} words, do not constitute a single type: the artificial system inherits the diversities of the natural one (cf. V. I. 12). Pseudowords may resemble established words enough to be easily read and pronounced (cf. Rubenstein et al., 1975), e.g. ‘clasty’ and ‘grilts’² [Except for the Frith example, these non-established words are taken from from Henderson & Chard (1980:99); but the classification here is my own], and can be easily made by slightly altering established words, e.g. ‘rekind’ made from ‘remind’ (Frith, 1980: 500). Regular non-established words, e.g. ‘scrute’, follow the general strategies for mapping between sounds and letters. Children can spell regular non-established words just as well as regular established words, and better than irregular ones (Marsh, Friedman, Desberg, & Saterdahl, 1981). Adults do better on established words of any sort (D. Brown, 1970), probably because of visual familiarity. However, even children were seen to depart from sound-letter regularities to preserve the uniform plural ‘-s’, though not other endings such as for the past tense (Read, 1971; cf. V.1.6, 9, 23). Pronounceable non-established words may be treated either by regular strategies, e.g. ‘belmor’, or by analogies to irregular established words, e.g. ‘kneep’ from ‘knee’ (Baron, 1977; Marsh, Desberg, & Cooper, 1977; Marsh et al., 1980; Frith, 1980). Regularity and pronounceability enable people to agree on how non-established words should be spelled (cf. Frith, 1980: 501); Morton (1980: 126) even speculates that the notion of the “word” is not essential for sound/letter rules. Unmanageable non-established words are those most people can’t read or pronounce, usually because of consonant clusters, e.g. ‘clmpst’, ‘sprnth’, and ‘nmdlbf’. The latter type impedes both spelling as well as matching up two samples as same or different (Barron & Pittinger, 1974; Chambers & Forster, 1975), even if people only have to match initial letters (Barron & Henderson, 1977), or to search for a specified letter (Krueger, 1970).

   1.30 Spelling research could also profit from the large literature on reading, perception, and learning. Children clearly apply to their reading some spelling-related knowledge, such as letter familiarity (Speer & Lamb, 1976; Richek, 1977-78), sound-letter correspondences (Venezky & Johnson, 1973; Mason, 1976; Guthrie & Seifert, 1977), and sound segmentation (Fox & Routh, 1976; Liberman et al., 1977). Experts disagree about how far the reception of script in reading corresponds to the production of script in writing-the operations can’t be simply reversible, i.e., mirror images of each other (cf. III.1.9), because the required actions differ markedly. Another point of comparison could be associative symmetry (reversible access between knowledge you had before and knowledge you just learned as an experimental task), with reading as the learned association and spelling as its reverse. The learned direction seems to be more reliable (Asch & Ebenholtz, 1962) and rapid (Waugh, 1970), just as reading consumes less effort and time than spelling. Still another point of comparison could be memory processes, with reading based on recognition, and spelling on recall {224} (Gibson & Levin, 1975: 335; cf. III.3.5.4). Or, reading could be more heavily automatic, and spelling more heavily attentional (Henderson & Chard, 1980; cf. III.1.21). At least, reading can be done with a smaller portion of the letters than spelling, and on a more abstract and general plane (see also Marsh et al., 1980). Most people read better than they spell; and practice in the one skill does not necessarily improve the other (Frith, 1978b, 1980; Marsh et al., 1981).

    1.31 Like spelling research, reading research must deal with the contribution of visual processing for the graphic aspect of a text; and with that of acoustic processing for the phonological aspect (cf. reviews in Bradshaw, 1975; Perfetti & Lesgold, 1979). Some researchers claim that reading always recodes the graphic version into a phonological one (e.g. Rubenstein, Lewis, & Rubenstein, 1971; Gough 1972). Others say the visual modality is primary (e.g. Bower, 1970; Goodman, 1970; Baron, 1973; Smith, 1973; Green & Shallice, 1976; Barron 1978b); the acoustic one is consulted only if more evidence is needed (Sloboda, 1980: 235). Marcel (1980: 401f) even speculates that “for each learner today, the concept of the phoneme (tacit if not explicit) comes from rather than leads to the particular alphabetic system with which he or she is confronted” (cf. also Templeton, 1981) — a position linguists since Bloomfield (1933) would strongly reject (cf. II.1.5f; V.3.1). This conflict in views matches a curious disparity in experimental findings. On the one hand, speech muscles were observed to become more active during silent reading of unfamiliar or complex materials (Hardyck & Petrinovich, 1970), as if acoustic reconstruction were going on, at least for hard words (cf. V 1.32f). On the other hand, miscuing while reading aloud (cf. IV.2.18, 48) often substitutes words that are visually rather than acoustically similar to the words on the page (Bryant & Bradley, 1980). Distractor tasks indicate at least some cases where people go straight from print to meaning (V.1.27). Apparently, phonological recoding is normal, but not universal, and serves as an aid to difficult tasks.

    1.32 The disparity between visual vs. acoustic should therefore vary to fit reading conditions and reader abilities. Children taught to read without heeding meaning, as Noah Webster advocated, naturally concentrate on converting print to sounds—even today, the most prominent focus of early reading instruction (Kintsch, 1979: 329). Kleiman’s (1975) study with distractor tasks suggested that phonological recoding is more necessary for extended, meaningful texts than for single words: the acoustic modality may support a carry-over from short-term sensory storage into short-term memory and thereby promote comprehension (cf. also Levy, 1978; Cohen, 1980; III.3.7). Unskilled readers, apparently favoring the visual modality (Frith, 1978), might lack an efficient system for the acoustic modality (Ehri, 1980: 321), as indicated by their slowness in accessing it (Barron, 1980); and by their difficulties in reading irregular words or non-established words aloud (cf. Perfetti & Hogaboam, 1975; Frederiksen, 1978), in judging the correctness of a sample (Baron & McKillop, 1975), and in categorizing the sounds common {225} to a list (Bryant & Bradley, 1980). Skilled readers, in contrast, divide up their processing between the acoustic and the visual (Frith, 1980). Presumably, they overcome any visual bias as their skills improve.

   1.33 In spelling, the situation appears to be the reverse (Ehri, 1980; Frith, 1980). Unskilled spellers rely quite heavily on the acoustic modality (Bryant & Bradley, 1980; Marsh et al. 1981), and commit errors on irregular words by trying to match sounds against letters.¹ [ Except for the Frith example, these non-established words are taken from from Henderson & Chard (1980:99); but the classification here is my own.] Skilled spellers use visual mental images more prominently alongside acoustic patterns (cf. VI.20). Thus, there should be a progression from an early imbalance, where reading is heavily visual and spelling acoustic, toward a more mature stage where both activities freely consult both modalities (Barron, 1980: 211; Bryant & Bradley, 1980: 365). Ehri (1980: 321) claims that the decisive shift occurs between the first and second year of reading instruction. Children learn that, in English at least, reading and spelling both work best by operating in two modalities, provided memory is reliable for both sounds and images of vocabulary items.

    1.34 These conclusions are reasonable in terms of linear processing, if we suppose that the immediate execution of the written word runs from one of the buffers maintained in working memory, one most likely based on short-term sensory storage (of one to two seconds’ duration) (cf. III.2.29; III.3.5.4; III.3.33; IV2.12f), and containing a dual-modality representation of the word (Sloboda, 1980: 246; Wing & Baddeley, 1980: 253). Sound is more temporally oriented, and images are more spatially oriented (cf. IV.1.6). The acoustic modality offers a continuous representation, whereas the visual modality offers a discrete one. In exchange, the components of the visual configuration are simultaneously available, while those of the acoustic one are successively available (cf. LaBerge & Samuels, 1974; Doggett & Richards, 1975; Terry, Samuels, & LaBerge, 1976). Therefore, visual knowledge guides segmentation, whereas acoustic knowledge guides sequential ordering. Skilled spellers could tell from the buffer if the correspondence between sound and image is conflicting, or at least indeterminate, so that a compensatory effort is made —particularly for commonly misspelled items (cf. VI. 39, 42ff). Unskilled spellers, relying too heavily just on sounding out the word (VI.20, 33), overlook conflicts and resolve indeterminacies haphazardly.

   1.35 Let’s imagine a writer entering a word in the memory buffer and setting out to spell it. The main thing is to extract enough cues from the buffer before the representation decays too far-rapid decay being a possible reason for the higher frequency of spelling errors at the end of a word (V. 1. 16). A visual scan of the buffer provides spatial, discrete, and segmental instructions; an acoustic scan provides temporal, continuous, and sequentially ordered instructions. Both sets of instructions¹ [Sloboda (1980: 241) opposes “phonemically transparent” words to “phonemically opaque” ones] are passed on to {226} largely automatic execution programs that translate them into motor actions (cf. III.2.31; III.3.33; IV.1.Ilf). Ideally, the proper information is ready when and where it’s needed—otherwise, errors occur without the awareness or intention of the writer, e.g., the linear contamination of adjacent letters (V.1.18, 42ff)—and the word is now successfully spelled. Next, the writer becomes a reader, getting feedback about the goodness of fit between the intended word and the actual result. Here also, errors creep in if visual and acoustic information isn’t brought to bear as it should be. Feedback may be rapid and incomplete, especially if resources are heavily engaged in other phases with concepts, expressions, etc. Then only salient errors would be detected, e.g., in initial position (V.1.16). Sounding out long, uncertain, or unfamiliar words appears to enhance feedback (Ehri, 1980: 316f), perhaps acting as good rehearsal to keep the word in the buffer (cf. Sloboda, 1980: 246).

   1.36 As we see, the question is not whether spelling is either acoustic or visual, but rather how much the two modalities contribute to an interactive process. The current consensus appears to be that the acoustic dominates, and the visual is consulted as a support—just the reverse of reading (Barron, 1980: 212; cf. V.1.31). Simon and Simon (1973) propose that a speller generates acoustically possible alternatives and then uses visual cues to make a selection. Meyer and Gutschera (1975) present a model where visual and acoustic analysis race in parallel, and the first one to finish is used. Barron (1980: 210) speculates that visual memory entries have no procedures for producing spelling, but only for checking after the fact whatever the sound-to-letter strategies have done. Even repeated misspellings could result from a checking failure, not from mere ignorance (cf. V.1.15). Regularity would assist spelling better than familiarity (Marsh et al., 1980: 346), since sound-to-letter correspondences enter earlier and more actively into the operation.

   1.37 However, the visual aspect in spelling is still important, even if only for making a visual judgement about the correctness of a word. At one extreme, Ehri (1980: 315f) claims that memory stores “a fairly exact copy of the printed form” whose “appearance deviates very little across instances.” But this view suggests that variations in print or handwriting should have greater effects than they do (cf. VI.25). At the other extreme, Frank Smith (1971, 1973) claims letters are only patterns of distinctive features (a hoary structuralist notion, II.1.5) needed for perception and recognition. But this view implies an unrealistically large catalog of features in place of the twenty-six letters (Rozin & Gleitman, 1977). Moreover, visual features would not correlate with acoustic ones well enough to assist phonological recoding (VI.31).¹  [Smith (1973) is in fact forced to argue inanely that sound doesn’t mediate the relationship between print and meaning. His extrapolation from “psycholinguisties” impels him to adopt a vastly uneconomical model design (cf. Ill. 1. 31.3) — a huge catalog of features, and, by implication, separate systems of visual vs. acoustic processing.] In between these extremes, we could postulate that visual {227} word memory is integrated with acoustic, and varies in its detail and clarity so as to fit context. Under normal conditions, the image is kept abstract and sketchy enough to adapt to print, handwriting, text mutilation, poor lighting, and so forth. But in doubtful cases, the image could be enhanced by a special reconstructive effort such as imaging and comparing alternatives (cf. Seymour & Jack, 1978). Or, there may be no stored image as such, but a knowledge pattern that can be accessed either acoustically or graphically (Coltheart, 1978) as needed to compute immediate motoric, acoustic, visual, and conceptual representations of the word (Simon, 1976; cf. Ehri, 1980: 323; Morton, 1980: 120).² [It might thus be possible to go straight from meaning to motoric action without passing through all the other levels (cf. D. Simon, 1976; III.3.6).]

   1.38 Whether the multi-level representation is stored or computed, spelling would have the versatility to interact with the other processes in text production as a whole (V.1.6, 9, 24). Though executed on the shallow level of sound/letter linearization, spelling could accept input from the deeper levels of phrase linearization, expression, and conceptual development—a possibility already documented for word perception (III.3.6). Skilled spellers can co-ordinate the shallow levels with the deep ones and allot resources appropriately. Unskilled spellers suffer overload and make errors.

   1.39 The future exploration of these larger issues requires a unified theoretical framework. We might undertake to model spelling in terms of the general linear principles set forth in IV.2. Sound/letter correspondences could fall under the listing principle: say the letters you see, and write the sounds you hear, one by one. Sequences of syllables and words could also be construed as lists to be spelled. The mutual ordering of letters could be a matter of look-back and look-ahead, with the sound sequence providing additional ordering cues (VI.34); errors due to pre-activation and reactivation (V.1.17, 44) would result when these two principles are degraded. The pause principle would be one basis for segmentation into syllables or words especially if the speller deliberately sounds them out. In line with the core-and-adjunct principle, a stressed syllable is the core of the word, and a vowel is the core of a syllable; adjuncts would be the unstressed syllables or the consonants, respectively. The misspelling of short function words (adjuncts) also fits here (VI.17, 46). Heaviness would go to the salient features of a word, such as letters that protrude above or below the line (V. 1.41), or that are marked as trouble spots for being irregular, silent, or indeterminate (V.1.11, 34, 42). Such trouble spots would receive special focus, e.g., by stressing normally unstressed syllables—a common tactic among children (Blumberg, 1976; Ehri, 1980). Disambiguation is required wherever sound/ letter correspondences are indeterminate. However, these principles would have to be worked out in much more detail to account for the whole system of spelling.

    1.40 In educational practice, spelling presents an exaggerated dilemma {228} for literacy programs. Spelling errors can indeed disturb communication disastrously by interchanging similar-looking words:  

(299) Giant roaches set fall practice [coaches] (Highland Park News, July 30, 1981).

(300) AN ITALIAN SINNER will be served at 5:30 p.m. at the Essex Center United Methodist Church. [dinner] (The Vermonter, Oct. 16, 1977)

(301) FORMER PRESIDENT ENTERS DINAH SHORE [entertains] (The Desert Sun, March 28, 1980)  

Poor spellers are popularly judged low in literacy and even in intelligence (I.2.8.1; VI.3). Yet spelling skills are hard to maintain in a culture where reading is not a popular pastime, and the spelling system itself is complicated and flagrantly inconsistent. One of my tests mainly intended to explore mental maps asked Gainesville fourth-graders to list the 50 states of the U.S. Some of the children devised sound-spellings: ‘Colrado’, ‘Indeana’, ‘Mane’, ‘Mishigun’, ‘Missasippy’, ‘Oakalahorna’, ‘Pencilvayna’, etc. Others created versions with a rather vague visual or acoustic resemblance to the actual names: ‘Albemba’, ‘Alksa’, ‘Arozna’, ‘Ieerandna’, ‘Ielllnoy’, ‘Keinteyk’, ‘Morxko’, ‘Soth Carealinen’, ‘Wist Vurginupp’, ‘Wastason’, ‘Calfifurya’/’Callfga’ (California?), ‘Coneiteit’/’Kuneekeen’ (Connecticut?), ‘Calroda’/’Crittordo’ (Colorado?), etc.

    1.41 English teachers are expected to remedy such obstacles as a matter of course. But to do so requires extensive expertise on defects in vision and reading—not the usual professional training in English degree programs. The students feel discouraged and stigmatized by the popular distortion of misspellings as signs of ignorance and low intelligence. In reality, spelling is, as I have tried to show, naturally error-prone. A complex, multilevel system has to be operated with precise timing of memory and actions, yet without much resources (V.1.6, 9, 24, 3 8). Some errors are due to dysfunctions in vision (Marcel, 1982) or in reading (Nelson, 1980; Seymour & Porpodas, 1980). Some are side-products of the mechanics of handwriting (cf. Wing, 1978) or of typing (MacNeilage, 1964; Rumelhart & Norman, 1981).¹ [Two-finger typists like myself get a whole new chance for extravagant typos on a conputer keyboard, the mores so where mistyped command keys can move the cursor accidentally: ‘singihgyst’ (‘insight’) ‘isimiktunmaoies’ (‘simultaneous’), etc.] Some letters are less conspicuous than others; those which do not protrude above or below the line are more likely to be omitted than those which do (Hotopf, 1980: 305). These various factors lead to errors that {229} misrepresent one’s true intelligence or literacy.

   1.42 Most of the samples I have gathered suggest two important error factors (compare samples in Cromer, 1980: 416ff; Frith, 1980. 498ff; Hotopf, 1980. 29Iff; Wing & Baddeley, 1980: 274ff). One is the buffer scan: the writer has trouble getting the needed visual or acoustic instructions from the memory buffer (cf. V.1.35). The other is the execution programs: commands to inscribe a letter one or more times in a given location get confused, disordered, or lost altogether (cf. III.2.31; III.3.32; V.1.35, 45). Indeterminate sounds, quite common in the unstressed syllables of English words (VI.11), seem to impede buffer scan among my students. An unstressed vowel can be displaced by a stressed one from the same word (‘geneus’, ‘obselescence’, ‘seperate’), perhaps as an effect of low heaviness. Or, the vowel ‘a’ is used as a general “schwa” sound /a/ resembling that in the indefinite article (‘differance’, ‘machinery’, ‘definate’). On occasion, an indistinct syllable disappears (‘basicly’ ‘invaluble’, ‘continous’). Silent letters are similarly omitted (‘hussle’, ‘morgage’, ‘goverment’, ‘likly’). Sometimes, though, a special effort is made to watch for them (VI.39). The silent ‘e’ is aided by its functions in lengthening a preceding vowel (‘wine’ vs. ‘win’) or padding certain consonants that hardly appear alone in final position (‘v’ in ‘give’, ‘z’ in ‘freeze’) (P. Smith, 1980: 35ff).

   1.43 Clusters may suffer both in buffer scan and execution programs. If sounds are continuous (V.1.35), the segmentation of adjacent letters might be a special problem (VI.18). Segmentation errors, both between syllables (‘mispell’, ‘wondeful’, ‘dissappear’) and between whole words (‘alot’, ‘aticast’, ‘allover’, ‘a cross’, ‘with out’, ‘never the less’), suggest that the spelling buffer can hold several words, at least if they are closely associated in standing phrases. Errors across words are common mostly among unpractised writers who probably rely on sounds. Smaller-scale segmentation is a more widespread problem. Consonant clusters suffer persistently (Bryant & Bradley, 1980: 364; Frith, 1980: 503; Marcel, 1980.. 378). They get rearranged (‘exist’ for ‘exits’), simplified (‘exeption’) or elaborated (‘excercise’). Vowel clusters are misspelled if their sounds are indeterminate (‘aercraft’, ‘wating’). Resonants, that is, the liquids /1/ and /r/ plus the nasals /m/ and /n/, have the most confusing impact on clusters (Ehri, 1980: 344; Marcel, 1980: 381), probably because their ambivalent status between vowels and consonants hinders both order and segmentation. They are frequently omitted (‘unabiguous’, ‘aswered’, ‘inteference’, ‘wondeful’) or doubled (ammount’, ‘irrationallity’). They seem to blend with nearby vowels, so that spellers are uncertain about what comes first (‘lible’, ‘titel’, ‘perseribe’, ‘preticular’). A nearby vowel may be omitted (‘bachlor’, ‘refrence’, ‘instrments’), sometimes leaving the resonant as the only syllable nucleus (‘untl’, ‘mnknd’, ‘helpfll’); or inserted (‘lightening’, ‘wonderous’, ‘tackeling’).

   1.44 Pre-activations are clearly a slip in the execution programs. Usually, the speller anticipates a letter from a later part of the same word. The {230} mistake is oftentimes caught and corrected at once:¹ [Pointed brackets are for crossouts (IV.2.25).] <fl> fuel’, ‘<st> scientific’, ‘<nowd> nowdays’, ‘<suri> surely’, ‘<cncourg> encourage’. The anticipated letter may be either proximate (‘encourg’) or remote (‘st’), sometimes even from a later word (‘and handful’ for ‘a handful’, ‘slowing counting’ for ‘slowly counting’). Pre-activations are less likely to be detected, e.g. during proofreading, if they preserve the word sound, as in the tenacious misspelling ‘research’. Interestingly, reactivations, where an earlier letter reappears, are less common; probably, the visual trace on the page, or the memory of having produced it, confirms that the action has already been done.

   1.45 However, the reactivations entailed in doubling consonants is a conspicuous error factor. These cases may be due to confusions about the number of iterations an execution action should have. This account could explain why doublings are (a) performed on incorrect letters (‘affraid’, ‘habbit’, ‘slickboddied’); (b) suppressed on correct letters (‘acessible’, ‘dilema’, ‘posible’); (c) displaced from one letter to another (‘necessity’, ‘parralel’, ‘personell’); (d) performed on both correct and incorrect letters (‘assymmetry’, ‘dissappear’); or (e) dropped in one position and retained in another (‘mispell’, ‘acommodate’, ‘embarassed’). There may be an independent doubling execution program that can come loose from the intended letter and be shifted or lost (Rumelhart & Norman, 1981: 14). Similar syllables suffer comparable effects. One may be dropped (a “haplology”) or blended with the other (‘indivual’, ‘scienfic’, ‘bycle’, ‘possibity’, ‘imaging’ for ‘imagining’). Conversely, a syllable may be proliferated (‘censorsored’, ‘creatitivity’, ‘sententences’). Such confusions are rarer than letter-doubling errors, probably because larger changes in word shape are easier to detect and suppress. Many letter doublings hardly affect either shape or sound.

   1.46 Short function words, whose pronunciation in a phrase resembles that of unstressed syllables in a word, are vulnerable to substitutions: ‘is’ for ‘in’, ‘that’ for ‘than’, ‘of’ for ‘or’, and so forth. Although rare in speech (Hotopf, 1980: 297), these substitutions frequently go undetected in writing. Though out of place, the interpolated word is registered as a legitimate item. Sometimes the word comes again in a correct slot within the neighboring context (‘the satellite will the [then] communicate’), so that the error could be in the representation, provided the buffer holds several words (VI.44). Other times, the word is not anywhere near (‘it follows than [that] only a small’), so that the error must be in the execution programs that are doubtless automatic and unsupervised for such frequent words. Either way, these function words get less resources than content words (IV2.6; VI. 17, 39).

   1.47 But these accounts for short words don’t say why errors occur on long words (cf. V.1.16). The latter may be hard to fit into the available {231} buffer space, or to keep there very long (cf. Dodd, 1980: 433). Their disadvantage may relate to their lower frequency (cf. Zipf, 1935), and to their greater difficulty for pronunciation (VI.1.11). They are often built up of word components that naive spellers wouldn’t recognize as such (cf. C. Chomsky, 1970). An error is hard to see because it makes a small difference in the total visual shape of a word being shortened (‘bronchopulmory’, geommucate’, ‘convient’) or lengthened (‘availiable’, ‘surpisered’, ‘feelableness’ for ‘feebleness’). Confusion of similar syllables (V. 1.45) is most likely in long words.

   1.48 We can easily see how the inscription of words misfunctions even when the correct spelling is known. The question is how writers can apply what they know. Radakar (1963) claimed that salient mental images of words make children’s spelling dramatically better (cf. also Sloboda, 1980). Carol Chomsky (1 971) wanted spelling taught before reading, whereas Frith (1980) recommended thorough reading as a prerequisite for spelling. Tenney (1980: 228) advocated writing alternative spellings down and comparing them; or breaking hard words down into parts. Barron (1980: 212) warned that too much emphasis on the “look-say” method (converting print to sound) or the “phonics” method (sounds alone) might cause learners to adopt one-sided strategies and spell badly. Even the traditional spelling bee may hinder good performance because the spoken series of letters obscures the spoken shape of the word (cf. Brooks, 1968; VI.28). Desberg et al. (1980: 80) felt that words should be spelled in contexts rather than in lists to rule out ambiguities.

    1.49 I have been experimenting with an approach where errors are predicted and made the center of focus. I give unskilled spellers large numbers of samples to proofread, focusing on the trouble spots described in V. 1.4248. The goal is to narrow down the set of probable misspellings so that these spots will be given particular attention. The students are also made aware that spelling errors are natural, rather than stigmatizing. Unskilled spellers are advised to use dictionaries continually; to proofread a paper once for spelling only; and to keep a conspicuous list of frequently misspelled items in a visible place in their homes, so that a clear mental image of each word is gradually acquired.

   1.50 Many kinds of misspellings are already obsolete for writers who use a word processor. Since the early 1980s, software vendors have been releasing programs that check the spelling of every word in a text written in a corresponding word processor against a dictionary, the sizes of which are growing immense. And the user can add quantities of additional words, or even create a new dictionary from scratch. Every misspelled word is displayed on the screen and marked in the text for easy searching and correcting. However, if the misspelling forms a known word (e.g. ‘that’ for ‘than’, VI.46), the program has no way to discover it. A really effective program that could check both spelling {232} and context is not currently feasible. Thus, for the near future, advances in school practice still depend on corresponding advances in theoretical research. Better models of how memory runs and how reliable feedback can be gathered will lead to better methods to control spelling, along with other complex, error-prone domains of human performance.

Click here to go to Chapter V, Part 2

 Click here to go to  Main Page