HRskins LAbortItories Sbltus Report on Sp«eh Rnerrrc:h1992, SR-109/110, 129-140

Linguistic Awareness and Orthographic Form*

Ignatius G. Mattinglyt

INTRODUCTION: THE TAXONOMY OFWRITING SYSTEMS

To impose some pattern on the vast array ofwriting systems, present and past,l severalinvestigators have proposed typologies of writing(Gelb, 1963; Hill, 1967; Sampson, 1985;DeFrancis, 1989; see DeFrancis for a review).While typology for its own sake may seem adubious goal, these proposals bring to noticecertain interesting questions.

Consider first the problem posed by logograms.It is generally recognized that the signs found inwriting fall into two broad categories: logographicand phonographic. Logograms stand for words, ormore precisely, morphemes. Thus, in Sumerianwriting, there is a logogram that stands for themorpheme ti, 'arrow.' Phonographic signs stand forsomething phonological: syllables or phonemicsegments. Thus, in Old Persian, there is a sign fort?e syllable cia, and in Greek alphabetic writing, aSIgn for the vowel a. This distinction suggests thatwriting systems might be classified according towhether they are logographic or phonographic.But the attempt to impose such a classification isembarrassed by the fact that while the manysystems in the West Semitic tradition are indeedes~e?tially phonographic and have no logograms,wntmg systems of all other traditions use bothlogograms and phonograms. There have been nopurely logographic systems: phonographic signsare found in all traditions.

In these circumstances, Gelb sets up a hybridcategory "word-syllabic," in which he includesSumerian, Egyptian (whose phonographic signs hetakes to be syllabic2), and Chinese. Other

Preparation of this paper was supported in part by NICHDgrant HD-ol994 to Haskins Laboratories. Alice Faber, LeonardKatz, Alvin Liberman, and Yi Xu gave helpful comment andcriticism on an earlier draft.

129

orthographic taxonomists allow a writing systemto belong to two different categories. Thus for Hill,Egyptian is both "phonemic" and "morphemic" andfor Sampson, Japanese is both "phonographic" and"logographic." DeFrancis, recognizing thatlogograms are neither necessary nor sufficient foran orthography, more sensibly treats logographyas an optional accompaniment to variousphonographic categories. But the question ofinterest is why logograms should play only thissecondary role, why there have been no purelogographies.

A second problem arises in sorting out thephonographic categories. Here one might recog­nize, with DeFrancis, systems like Sumerian orLinear B, in which the phonographic signs standsfor syllables; systems like Egyptian or Phoenicianin which they stand for consonants; and system~like Greek or English, in which they stand forboth consonants and vowels (plene systems).

The distinction between consonantal and plenesystems, however, proves to be less than rigid. InEgyptian, the letters for j. w, and 1 are used towrite i, u and .. respectively, in foreign names(Gelb, 1963). Phoenician, indeed, is a strictlyconsonantal, but the other "consonantal" systemsderiving from it all have some convention fortranscribing vowels when necessary. For example,in Aramaic, the letters yodh, waw, and he (oraleph) were used to write final i, U, and a,respectively, and to render vowels in foreignnames (Cross & Freedman, 1952). In MasoreticHebrew, Arabic, and various Indic systems,vowels are regularly indicated by diacritic markson consonant letters. And, of course, the firstclearly plene system, the Greek alphabet, is adevelopment from the Phoenician consonantalsystem. The taxonomist thus has to decide whereto draw the line between essentially consonantalsystems, hybrid systems, and undoubted plenesystems. Perhaps the wisest course is the onefollowed by Sampson: simply to classify all thesesystems as "segmental."

130 Mtlttingly

Syllabic systems, in contrast, are clearly aseparate category and present no problem to thetaxonomist. There is no writing system that mustbe regarded as a hybrid between a syllabic and asegmental system. Syllabic systems show notendency to analyze syllables into segments. Whatis found, rather, is that when analysis becomesnecessary, complex syllables are analyzed intosimpler syllables. Thus, neither theMesopotamian nor the Mayan syllabaries hadsigns for all possible C1V1C2 syllables in theirrespective languages. Instead, such syllables werewritten in Mesopotamian as if they were C1V1 +VlC2 (Driver, 1976) and in Mayan as if they wereC 1V1 + ~V1 (Kelley, 1976». Similarly, GreekCIC2Vl ...syllables were written in Linear B asCIV1+ ~Vl +...(Ventris & Chadwick, 1973). Nor,despite suggestions to the contrary by Gelb andDeFrancis, has a syllabic system ever developedinto a segmental system, or conversely.S It cannotbe excluded that the Egyptians may, as DeFrancissays (following Ray, 1986), have gotten the idea ofwriting from the Sumerians. But there is certainlyno reason to believe that they borrowed the idea ofsyllabic writing from the Sumerians and thenadapted it to consonantal writing, in the way thatthe Greeks may be said to have borrowed the ideaof consonantal writing from the Phoenicians andadapted it to plene writing. The variousorthographic traditions are remarkably self­consistent in this matter. The Mesopotamian,Chinese, Cretan and Mayan traditions began andremained syllabic; the Egyptian and West Semitictraditions began and remained segmental.

If the main purpose here were to arrive at ataxonomy of writing systems, the conclusionwould have to be that there are two Primarycategories: syllabic and segmental. Either of thesemayor may not be accompanied by logograms.Transcription of vowels in segmental systems is amatter of degree, with Phoenician at one end ofthe scale and Greek at the other. The interestingquestion, however, particularly given the degree ofoverlap or hybridization that is found betweenlogographic and phonographic categories, andbetween consonantal and plene categories, is whythe syllabic and segmental categories haveremained so distinct.

In an attempt to answer the questions justposed, it is necessary to consider why anorthography can make reading and writingpossible, what constraints there are on the form oforthographies, how orthographies could have beeninvented, and what happens when orthographiesare transmitted from one culture to another.

WHY READING AND WRITING AREPOSSIBLE4

When a listener has just heard an utterance in alanguage he knows, he has available for a brieftime not only his understanding of the semanticand pragmatic content of the utterance (thespeaker's message), but also a mental repre­sentation of its linguistic structure. The basis forthis claim is that a linguist, by analyzing theintuitions of informants about utterances in theirnative language (such as that two utterances areor are not the same word, or that a certain word isthe subject of a sentence), can formulate acoherent grammar, consistent with grammarsthat would be formulated by other linguistsworking with other informants on the samelanguage. This holds true even if, as is typicallythe case for a language with no writing system,the informants are quite unaware of the linguisticunits into which utterances in their language canbe analyzed. Because the informants' intuitionsare apparently valid, they must be based onlinguistic representations of some kind.

While linguists are not in total agreement aboutthe nature of the linguistic representation of anutterance, it seems reasonably clear that such arepresentation must include the syntacticstructure, the selection of lexical items and theircomponent morphemes, the phonological struc­ture, and the phonetic structure. The linguist'ssyntactic diagrams and phonological and phonetictranscriptions are formal reconstructions ofdifferent levels of the representation. These levelsare not independent of one another. Syntaxconstrains lexical choice, lexical choice determinesmorphology and phonology, syntax and phonologydetermine phonetic structure. The representationthus has extensive inherent redundancy.

The linguistic representation is strictlystructural rather than procedural. The listenerhas no access to the many intermediate steps hemust presumably go through in the course ofparsing the utterance, so that these steps arenot represented. Acoustic details such as formanttrajectories are not part of the linguisticrepresentation, simply because the listenerdoes not perceive them as such, but only thephonetic events they reflect. Other aspects of theutterance, such as individual voice quality,speaking rate, and loudness, which the listenercan hear, must be presumed to be excludedbecause they are not linguistic at all and neverserve to mark a linguistic difference between twoutterances.

Linguistic AwamJeSs and Orllwgraphic Form 131

Access must be distinguished from awareness.All normal laneuage users, it has been claimed,have access to the contents of linguistic represen­tations. This means that they have a potentialability to introspect and report on significant de­tails of the representation, and to regard it as astructure of phrases, words, and segments, notthat they can actually do so. The representation isa complicated affair, and a person who is not"linguistically aware" can no more be expected tonotice its characteristic units and structure thanan electronically naive person can be expected toappreciate the units and structure of a circuit dia­gram (Mattingly, 1972). Linguistic awarenessmust in large part be acquired. The principalstimulus for linguistic awareness in modem cul­tures is literacy (Morais, Cary, Alegria, &Bertelson, 1979). Unlike illiterate adults or prelit­erate children, those who have learned to read canreadily report on and manipulate at least thoseunits of the linguistic representations of spokenutterances to which units of the orthography cor­respond (Read, Zhang, Nie, & Ding, 1986).However, there must certainly be other sources oflinguistic awareness: Long before writing wasknown, poets composed verse in meters requiringstrict attention to subtle phonological details.

It is not agreed how linguistic representationsare created. On one view, they are a byproduct ofthe cognitive processes by which utterances areanalyzed. Linguistic information, recovered stepby step from the auditory image of the inputsignal, is temporarily represented in memoryuntil, at a later stage, the speaker's message canbe computed (Baddeley, 1986). The difficulty withthis view is that, as has been noted, the languageuser seems to have no access to the supposedlycognitive analytic steps that must precede theformation of the representation or to thesubsequent steps by which the message is derivedfrom this representation. An alternative view isthat the representation, as well as the messageitself, is not a byproduct but a true output of aspecialized, low-level processor (the "languagemodule") whose internal operations, beinginaccessible to cognition, have no cognitivebyproducts (Fodor, 1983). This view implies thatthe linguistic representation must have somebiological function other than communication, forwhich the message alone would suffice. What thisfunction might be is unclear (but see Mattingly,1991, for some speculations).

So far, the cognitive linguistic representationhas been considered just as the product of theperception of utterances. But such representations

are produced in the course of other modes oflinguistic processing as well. Thus, a linguisticrepresentation is formed in the production of anutterance, so that the speaker knows what it is hehas just said. And when one rehearses anutterance in order to keep it in mind verbatim,what presumably happens is that the linguisticprocessor uses a decaying linguistic repre­sentation to construct a fresh version of therepresentation, and incidentally, of the message.This seeming defiance of entropy is possible forlinguistic representations (as it may not be formental representations in general) because oftheir high inherent redundancy.

Consideration of rehearsal also shows that thelinguistic representation can be an input to aswell as an output from the linguistic processor.Even more significantly, for the present purposes,a representation not originally produced by pri­mary processes of perception or production can besuch an input. An introspective, linguisticallyaware person can readily compose a "synthetic"linguistic representation according to some arbi­trary criterion: the first five words he can think ofthat begin with /bI, for example. This is obviouslya very partial representation: just a sequence ofphonological forms drawn from the lexicon, with­out explicit phonetics or syntax. But if this se­quence is rehearsed, the phonetic level, togetherwith whatever syntactic structure or traces ofmeaning may be accidentally implicit in the se­quence, will be computed, just as if the sequencewere what remained of a natural representationresulting from an earlier act of production, per­ception, or rehearsal. All that is required for asynthetic representation to serve as input forcomputing a natural one is that it contain enoughinformation so that the rest of the structure of theutterance is more or less determined.

These various considerations suggest how it isthat one linguistically aware language user cancommunicate with another, not by means ofspeech, but by means of synthetic representations,provided a way of transcribing suchrepresentations, that is, an orthography, isavailable. The writer speaks some utterance (atleast to himselfl, creating a linguistic repre­sentation. The orthography enables him totranscribe this representation in some very partialfashion. From this transcription, the readerconstructs a partial, synthetic linguistic repre­sentation. Such a representation is enough toenable the reader's linguistic processor to computea complete, natural representation, as well as thewriter's intended message.

132 Mattingly

Ifwe compare what happens between writer andreader with what happens between speaker andhearer, it can be seen that the difference is muchmore than merely a matter of sensory modality. Inspeech perception, there is a natural and uniqueset of "signs"-the acoustic events that the humanvocal tract can produce-and they are already in aform suitable for immediate linguistic processing(Liberman, this volume). Only the output of thisprocessing is a linguistic representation. The inputspeech signal is in no sense a partial linguisticrepresentation, but rather a complete representa­tion of a very different kind. Moreover, the specifi­cation of the complex relation between the phonet­ically significant events in the signal and the unitsof the linguistic representation is acquired pre­cognitively (Liberman & Mattingly, 1991); it doesnot have to be learned. Indeed, as has been re­marked, the hearer has no access to the acousticevents, and may have little or no awareness of theunits of the linguistic representation. In reading,on the other hand, there is no one, natural set ofinput symbols. Linguistic processing must there­fore be preceded by a stage having no counterpartin speech perception: a cognitive translation fromthe orthographic signs to the units of the syntheticlinguistic representation. The beginning readermust therefore deliberately master the mappingbetween the signs and the units, and for this hemust have an awareness of the appropriate as­pects of the linguistic representation.

CONSTRAINTS ON ORTHOGRAPHICFORM

What psychological factors constrain the form ofan orthography? Gelb (1963) makes a usefuldistinction between "outer form"-the shape of thevisible symbols and their arrangement in a text­and "inner form"-the nature of thecorrespondence of the symbols to linguistic units.Beyond the trivial requirement that the symbolsbe visually discriminable, there appear to be noparticular psychological constraints on outer form.The shapes of the signs in the writing systems ofthe world and the way they are arranged areextremely various, and such limitations as existare to be accounted for not by cognitive orlinguistic factors but by practical ones, such as thenature of the writing materials available andwhat patterns are easily written by hand, or byesthetic ones, such as the beauty of particularstroke patterns. This variety is possible because,as has just been seen, a cognitive translation isrequired for reading and writing in any event.

This price having been paid, outer form can varyalmost without limit.

Inner form, on the other hand, is highlyconstrained. In the first place, the orthographymust correspond to the linguistic representation,because there is no other cognitive path tolinguistic processes. This is the reason thatproposals to treat spectrographic displays ofspeech as, in effect, an orthography the deaf couldlearn to read (Potter, Kopp, & Kopp, 1966) are notlikely to succeed. On the one hand, the reader ofspectrograms cannot process the visually­presented spectral information as a listener canprocess the same information in the auditorially­presented and biologically-privileged speechsignal. On the other hand, the spectrogram readerhas no natural cognitive access to raw spectralevents, and, a fortiori, no awareness of them.Therefore, even ifhe could somehow synthesize acognitive spectral representation from the visibleone, there is no reason to believe it could be aninput to linguistic processes. All he can do is toapply his cognitive knowledge of acousticphonetics to the task of inferring the linguisticrepresentation from the spectrogram. Because therelation between spectral patterns and even themost concrete level of this representation, thephonetic level, is extremely complex, and a greatdeal of extraneous information is present,"reading" spectrograms is a slow and unreliableprocess. Analogous observations, obviously, couldbe made with respect to other records of physicalactivity in which linguistic information is implicit,such as the speech waveform or traces ofarticulatory movements. What has to betranscribed, then, is some level or levels of thelinguistic representation itself.

However, certain levels of the linguisticrepresentation are seldom or never transcribed intraditional orthographies. For example, syntacticstructure is never transcribed. The few features oforthography that might be considered syntactic,such as punctuation and sentence-initialcapitalization, are more reasonably regarded astranscriptions ofprosodic elements. Why is syntaxthus avoided? It is not just that tree diagrams arecumbersome to draw and nested brackets difficultto keep track of, but that the syntactic structurealone would be insufficient to specify a particularsentence: Each possible phrase marker is sharedby an indefinitely large number of sentences. Itwould therefore be necessary that a syntacticorthography also transcribe in some way theparticular lexical choices. But if this is to be done,the phrase-marker itself becomes redundant,

Linguistic Awareness and Orthographic Form 133

because (barring some well-known types ofstructural ambiguity, such as those discussed byChomsky, 1957) the words, and the order in whichthey occur, are themselves sufficient to specifysyntactic structure.

Again, someone who supposed that speech andwriting converged at the lowest conceivable level,given the difference of modality, might expect thatthe most efficient form of writing would be a nar­row phonetic transcription (see Edfeldt, 1960).This transcription would correspond to the outputof the phonological component of the grammar,presumably the level of the linguistic representa­tion closest to the speech signal itself. Owing tocontextual variation, higher-level units such asphonemes, syllables, morphemes, or words are notconsistently transcribed or explicitly demarcatedin such a transcription. But, in contrast to thesyntactic orthography just considered, more thanenough linguistic information to specify the lin­guistic representation would nevertheless be im­plicit. Why is such an orthography not found? Apartial answer is that because, as has been sug­gested, writing and speech are not, in fact, so sim­ply related, there is no particular advantage to alow-level, phonetically veridical representation.Moreover, it seems more difficult to attain aware­ness of phonetic details insofar as they are pre­dictable. Once the language-learner is able to rep­resent words phonemically, the phonetic levelseems to sink below awareness. But as will beseen, there is a still more fundamental reason whya narrow phonetic transcription would be imprac­tical.

It is important to distinguish between thelinguistic unit used for the actual processing of anutterance by writer and reader, and the linguisticunits to which the various graphemic unitscorrespond. Elementary graphemic unitscorrespond to phonemes (English letters ordigraphs), syllables (Japanese kana5), or mor­phemes (simple Chinese characters). These areusually organized into complex units that havebeen called "frames" (Wang, 1981). A spelled wordin English, a complex Chinese character, agrouping of Egyptian hieroglyphics are examples.Frames are usually demarcated by spaces inmodem writing, but other demarcative symbolshave been used. Sometimes the frame is implicit:The structure of the frame itself may be sufficientto demarcate it from adjacent frames, as inJapanese, where a kaIiji logogram or logograms isregularly followed by kana syllable signsspecifying affixes. Some orthographies, such asthose early alphabetic orthographies in which

there is no demarcative information of any kind,have no frames larger than their elementarysigns. Frames often correspond to linguisticwords, but not always: In Chinese and Sumerian,they correspond to morphemes.

By "unit of transcription" is meant the linguisticunit that the writer actually transcribes and thereader cognitively translates to form the syntheticlinguistic representation. One might expect thatthe units of transcription for a particularorthography would be those to which its framescorresponded. Thus, in English, the frames are

- consistent spellings of words, and the experiencedreader's intuition is surely that he reads word byword and not letter by letter, as he would if thetranscription unit were the segment. Thisintuition is borne out by demonstrations of "wordsuperiority." In these experiments, it is found, forexample, that subjects can recognize a letterfaster and more accurately when it is part of areal written word than when it appears alone or ina nonword (Reicher, 1969). This result suggeststhat in the case of a real word, subjects can usethe orthographic information to recognize theword very rapidly, and then report the letters itcontains. If the segment were the transcriptionunit, the letters corresponding to the segmentsshould be recognized and reported faster than thewords.

However, it is possible that the unit oftranscription does not really depend on the frameused in a particular orthography, but is in factalways the word. One reason for believing this isthat the word has to be the most efficient unit oftranscription, because words are the largestlexical structures. Anything smaller would requireprocessing more units per utterance; anythinglarger could not be readily coded orthographically.

Chinese writing allows a test of this possibility.A Chinese word consists of one or moremonosyllabic morphemes. In the writing,characters are the frames and correspond to thesemorphemes. Words as such are not demarcated.There is some evidence, however, that the unit oftranscription is nonetheless the word. In a recentexperiment (Mattingly &: Xu, in preparation),Chinese speakers were shown sequences of twocharacters on a CRT. In half the sequences, one ofthe characters was actually a pseudocharacter,consisting of two graphic components that inactual writing occur separately as components ofother characters, but not together in the samecharacter. Of the sequences in which bothcharacters were real, half were real bimorphemicwords and half were pseudowords. The subject's

134 Mattingly

task was to reb;,.;ond "Yes," if both characters in asequence were genuine and "No: if either was apseudocharacter. Subjects performed this taskfaster for words than for pseudowords, and it waspossible to show that this was not simply an effectof the higher transitional probabilities of the wordsequences, but rather a valid "word superiority"effect. This result, like that of an earlierexperiment by C. M. Cheng (1981, summarized inHoosain, 1991) suggests that despite morphemicframing and the absence of word boundaries, theword is the transcription unit for Chinese readers.Other writing systems in which words are notframed remain to be investigated.

But if word-size frames are not essential forreading word by word, why is a narrow phonetictranscription an unlikely orthography? The reasonmust be that the shapes of words in such a tran­scription are context-sensitive and thus difficult torecognize. (Notice what happens to JbcDcl', hand,in [bcutuwlz], hand tools, [bcQGrouejdl, handgrenade, [b&m.plkt],hand picked, etc.). The readeris therefore forced to process the transcriptionsymbol by symbol, a slow and arduous procedure.In Chinese, on the other hand, though word­boundaries are absent, the form ofan orthographicword is constant, or at least not subject tocontextual variation. It is suggested that this is aminimal constraint that all writing systems mustmeet, so that words can serve as units of tran­scription.

Although words are the transcription units,writing always employs graphemic unitscorresponding to linguistic units smaller than theword. It might seem possible, in principle, to havea pure logographic system, consisting simply ofone monolithic symbol for each word. But thedifficulty with such a system is that while thelexicon of a language is, in principle, finite, it is inpractice, indefinite: New words are continuallybeing coined or borrowed. In some cases-a nonceword or an unusual foreign name, for example-itwould make little sense to provide a speciallogogram. A writer could thus find himself with nomeans of writing a particular word because nologogram for it existed. Or, of course, he could bestuck simply because he did not know the correct10gog"1.m. An actual writing system lLJUreS thatthe iter will never be in this situation bypr, \1g a system of spelling units. Theav ity of the spelling system guarantees thatth Jgraphy will be "productive," that is, thattht er who has mastered the spelling ruleswill a,Nays have some way (though it may not be

the "correct" or standard way) to write every wordin the language (Mattingly, 1985).

The only linguistic units that have served as thebasis for spelling units are syllables andphonemes. It might be thought that morphemescould be the basis of a spelling system and some(e.g., Sampson, 1985) have argued that Chinesehas such a system, because the characterscorrespond to morphemes. This is true, but, as hasalready been noted, these morphemic units areframes: Relatively few of the characters in theinventory are simple logograms. Over 90% arephonetic compounds, each consisting of twographic components that (in general) occur also asseparate logographic characters. One of these, the"phonetic" stands, in principle, for a particularphonological syllable, and the set of phoneticsthus constitutes a syllabary. The other, the"semantic," is one of214 determiners that serve tomitigate the extensive homophony of Chinese: Thenumber of monosyllabic morphemes far exceedsthe number of phonologically distinct syllables.The situation is complicated, however, becausethere is usually more than one phoneticcorresponding to a particular phonological syllable(there are about 4000 in all for about 1300phonologically distinct syllables), and because,through various accidents of linguistic history, aphonetic often has different phonological values indifferent characters. But these circumstancesshould not obscure the highly systematic,syllabographic nature of the spelling, any morethan the existence of several spelling pattems forone sound, and numerous inconsistencies in letter­to-sound correspondence, should obscure thesystematic, alphabetic nature of English spelling(DeFrancis, 1989).

Words can indeed be analyzed into morphemesas well as segments and syllables, but theinventory of morphemes in a language, like theinventory of words itself, is indefinitely large andsubject to continual change. While logograms thatare morphemic signs can have a valuablesupplementary function in orthography, theycould not constitute a productive spelling system,and there is no orthography in which they playthis role.

Syllables and segments, on the other hand, haveseveral properties that make them suitable as abasis for spelling units. First, a word can alwaysbe analyzed as a sequence of phonologicalelements of either type. Second, the inventory ofsyllables may be small (and indeed was small inall the languages for which syllabic spelling

Linguistic Atwreness tmd Orthographic Form 135

developed independently) and the inventory ofsegments is always small. Third, the membershipof these inventories changes only very slowly. Noother linguistic units have these convenientproperties, save perhaps phonological distinctivefeatures (Because a diacritic is used to indicatevoicing, it could be maintained that features havea marginal role in Japanese spelling).

In sum, every orthography needs to have aspelling system and a spelling system isnecessarily phonographic. It is not accidental thatall orthographies spell either syllabically orsegmentally: there is probably no other way tospell.

THE INVENTION OF WRITING6Writing was invented, probably several times,

by illiterates. From what has been said already, itfollows that what had to be discovered was one orthe other of the two possible spelling principles,the syllabic or the segmental, and that this musthave required awareness of these units of thelinguistic representation. How could the inventorshave arrived at such awareness?

Some linguistic units seem to be more obviousthan others. Awareness of words can perhaps beassumed for most speakers, even if they are pre­literate or illiterate. It probably requires only avery modest degree of awareness to appreciatethat an utterance is analyzable as a sequence ofsyntactically functional phonological strings, ifonly because sequences consisting ofjust one suchstring are quite frequent: Words may occur inisolation. Certainly preliterate children have nodifficulty in understanding a task in which theyare to complete a sentence with some word, and alinguist's naive informant readily supplies thenames of objects. Awareness of syllables as count­able units may also be fairly widespread. The syl­lable is the basis for verse in many cultures; pre­literate children can count the number of syllablesin a word. This kind of syllabic awareness, how­ever, is probably not the same thing as beingaware (if such is indeed the case) that the sylla­bles of one's language constitute a small inventoryof readily demarcatable units.

These limited degrees of linguistic awarenessare probably readily available to speakers of alllanguages. But more subtle forms of awarenessmay well have arisen only because they werefacilitated by specific properties of certainlanguages, including, in particular, those forwhich writing was originally invented.

Consider, first, Chinese. In the Ancient Chineselanguage, words were in general monomorphemic,

there being neither cOD)pounding nor affIXation.Morphemes were monosyllabic and a particularmorpheme was invariant in phonological form.Because of restrictions on syllable structure, theinventory of syllables was small. Homophony wastherefore very extensive, one syllable correspond­ing to many morphemes (Chao, 1968).7 The num­ber of different characters in the Chinese writingsystem sharing a particular phonetic componentgives some notion of the degree of homophony inAncient Chinese, and this number often exceedstwenty. Chinese thus contrasts sharply withEnglish and other Indo-European languages, inwhich morphemes vary in phonological form, maybe polysyllabic, and may not even consist of an in­tegral number of syllables; syllable structure iscomplex; the number of possible syllables is rela­tively large; and homophony is therefore amarginal phenomenon.

Since words coincided with morphemes inChinese, awareness of morphemes required noanalysis, and the use of logograms, i.e.,morphemic signs, was an obvious move. Theextensive homophony made "phoneticborrowing"-using the sign for one morpheme towrite another morpheme with the same syllabicform8-a strategy that was both obvious andproductive; when a writer needed to write amorpheme, a sign with the required sound wasvery likely to be available. It thus became obviousthat the number of different sounds was in factsmall, yet every morpheme corresponded to one ofthem. Awareness of demarcatable syllable unitsthus developed. Of course, the same extensivehomophony that fostered the discovery of theseunits also meant that their signs had to bedisambiguated by the use of logograms asdeterminers, as in the large class of characterscalled "phonetic compounds," described earlier.

Chinese morphophonological structure thusencouraged the discovery of the syllable; on theother hand, it did not encourage the discovery ofthe phonemic segment. There was nothing aboutthis structure that would have served to isolatephonemes from syllables or morphemes.

Sumerian was an agglutinative language. Aword consisted of one or two monosyllabic CVCmorphemes and various inflectional and deriva­tional affixes. Its phonology had certain propertiesthat imply a preference for a CVCVC...VC syllabi­fication. There were no intrasyllabic consonantclusters; a cluster simplification process deletedthe first of two successive consonants across syl­lable boundaries, resulting in such alternations astil, do 'life'; and final vowels were deleted (Driver,

136 M4ttingly

1976; Kramer, 1963). In other relevant respects,however, Sumerian resembled Chinese and, likeChinese, favored awareness of morphemes and ofsyllables as demarcatable units. Aside from theeffects of the syllable-forming processes just men­tioned, a root maintained an invariant phonologi­cal form. A root could be repeated to indicate plu­rality. Because the morphemes were monosyllabic,and because of the restricted syllable structure,the number of possible distinct syllables wassmall. These circumstances, resulted, again, in ex­tensive homophony.

For a speaker of Sumerian to become aware ofmorphemes was perhaps not quite as easy as for aspeaker of Chinese. He would have had to noticethat words with similar meanings often hadcommon components, for the most partcorresponding to syllables. This stage ofawareness having been achieved, morphemicwriting is possible. From this point on, the story isquite similar to that for Chinese, homophonyleading to phonetic borrowing, and then to syllablewriting supplemented with determiners.

There is, however, one striking differencebetween the Sumerian and the Chinese writingsystems. While Chinese makes no internalanalysis of syllables, Sumerian does. A sign for aCIVIC2 morpheme could be borrowed to write aCIVICS morpheme, e.g., the RIM sign was used towrite riD. A VC syllable sign could be used as apartial phonetic indicator after a logogram, e.g.,GUL + UL. For many of the CIVIC2 syllables, ashas been mentioned, there was no special sign;instead, such a syllable was written with the signfor the CIVI followed by the sign for VIC2.Thusthe syllable ral is written RA AL (examples fromGelb, 1963). A possible explanation of thesevarious practices is that in spoken Sumerian,consistent with its preference for CVCVC...VCstructure, some form of vowel coalescence tookplace when two similar vowels came together, sothat CIVI + VIC2 sequences became phoneticallyCIVIC2, and thus homophonous with originalCIVIC2 syllables. Such homophony could havesuggested analyzing and so writing the latter asCIVI + VIC2. Again CV signs as well as VC signswere used to indicate the endings of CIVIC2morphemes. For example, because of multiplesemantic borrowing, the logogram DU could standnot only for du, 'leg,' but also for gin, 'go,' gub,'stand,' and tum, 'bring'. Which of the latter threewas intended was indicated by writing DU NA forgin, DU BA for gub, and DU MA for tum (Driver,1976). This practice perhaps arose because thephonological final vowel deletion made CIVIC2

and CIV IC 2V 2 sequences homophonous,suggesting that what followed CIV I could bewritten in either case as if it were CIV2. Thus theSumerians may have viewed CIVIC2 morphemeseither as CIVI + VIC2 or as q,VI + C2V2, either ofwhich was entirely consistent with their syllabicphonological awareness.

With Egyptian, in contrast to Chinese andSumerian, the morphology and phonology of thelanguage of the language favored segmentalawareness. In Afro-Asiatic languages, the rootsare biconsonantal and triconsonantal patternsinto which different vowels or zero (that is novowel at a11) are inserted to generate a largenumber of inflected forms. Because the vowels ofEgyptian are unknown, it is easier to illustratethis point with an example from another Afro­Asiatic language, e.g., Hebrew. From the Hebrewroot k+b are derived kItab, 'he wrote'; yikklteb, 'hewill be inscribed'; kItob 'to write'; kltub, 'written';mJttab, 'letter; and many other forms. Because ofphonological restrictions, the number of differentconsonantal patterns in Egyptian was relativelysmall, and there were consequently numeroushomophonous roots, e.g., n-f-r, 'good'; n-f-r, 'lute'(Jensen, 1970).

It is not difficult to imagine an Egyptiannoticing that many sets of semantically similarwords in his language had a common consonantalground and a varying vocalic figure, though atfirst he may not have individuated theconsonants. Accordingly, signs for root morphemeswere devised. The homophony of Egyptian thendid for phonetic segments what homophony inChinese and Sumerian did for syllables. Amorphemic sign was frequently borrowed to writea homophonous morpheme, e.g., NFR, the sign forn-f-r, 'lute', used to write n-f-r, 'good,' or WR,'swallow,' used to write w-r, 'big.' The signs werenow generalized to stand for consonantalsequences that were not morphemes, e.g., WR <WR was used to write the first part of w-r-d,'weary.' And because in some cases roots wereactually uniconsonantal, and in other cases thesecond consonant had become silent, some signscame to stand for single consonants, andconstituted a consonantal alphabet. Thus the d inw-r-d could written with the sign D < DT, the finalconsonant in d-t, 'hand,' being actually thefeminine suffix, not part of the root. Finally,logograms were employed as determiners toclarify ambiguous transcriptions: the spelling MNN H for the word m-n-b being followed by thedeterminer for 'plants' when this word had thesense 'papyrus plant,' the determiner for 'men'

Linguistic Awamu!Ss and Orthographic Form 137

when it had the sense 'youth,' and the determinerfor 'minerals' when it had the sense 'wax'(examples from Jensen, 1970). In this fashion, theEgyptians arrived at a consonantal spellingsystem.

If the Egyptians had thus achieved segmentalawareness, why did they not transcribe the vowelsas well as the consonants? It is not likely that theywere unable to hear the different vowels. Theexplanation is rather that because the vowelsordinarily conveyed only inflectional information,the writing was sufficiently unambiguous withoutsuch indications, just as English writing issufficiently unambiguous without stress marking.But as has already been noted, there was aconvention for writing vowels when necessary.Such writing is found very early in the history ofEgyptian writing (Gelb, 1963).

The Egyptians could hardly have arrived at asyllabic system instead. Because zero alternatedwith vowels in the generation of words, there wasno obvious correspondence between morphemesand syllables or syllable sequences. And becauseof such alternations, a syllabic orthography wouldhave resulted in a number of dissimilar spellingsfor the same morpheme.

These examples suggest that the phonologicalawareness required for the invention of writingdevelops when morphemes have a highly re­stricted phonological structure-monosyllabic, inthe case of Sumerian and Chinese; consonantal inthe case of Egyptian-that results in pervasivehomophony. Speakers of such languages are natu­rally guided to the invention of writing by thesespecial conditions. (A corollary is that it is notnecessary to propose a derivation of Egyptianfrom Sumerian to account for parallels in the de­velopment of the two systems.) On the other hand,Indo-European languages and many others lackany such restrictions, and would not have favoredphonological awareness in this way. Indeed, onehas to wonder whether, for such languages, writ­ing could have been invented at all.

In the early discussion of the psychology ofreading, the precise role of phonologicalawareness in learning to read appeared equivocal.Is phonological awareness a prerequisite forreading? Or, on the other hand, does theexperience of reading engender phonologicalawareness (Liberman, Shankweiler, Liberman,Fowler, & Fischer, 1977)? It was later seen,however, that both statements must be true: Thebeginning reader must, indeed, have some degreeof awareness, but this awareness is increased anddiversified in appropriate directions as a result of

his encounter with the orthography (Morais,Alegria & Content, 1987). In the same way, theinvention of writing must have been anincremental process, beginning with an initialawareness of morphemic structure. Theexperience of working out ways to transcribemorphemes for which there were no logograms ledto awareness of the syllabic or phonemic structureof. these morphemes, and then to awareness ofsuch structure generally.

To say that the process was incremental is notto say that it was not quite rapid. It is noteworthythat in all three of the writing traditions justconsidered, evidence of spelling is found veryearly: in Sumerian writing from the Uruk IVstratum (Gelb, 1963); in Chinese writing of theShang dynasty (DeFrancis, 1989); in Egyptianwriting of the First Dynasty (Gelb, 1963). Thesefacts are consistent with the proposal that forgeneral-purpose writing, a purely logographicsystem is impractical. As has been argued, anorthography is not productive without a spellingsystem: The invention of the one requires theinvention of the other.

To the extent that this account of the inventionof writing is plausible, it supports the dichotomybetween syllabic and segmental spelling proposedearlier, for what had to be invented was one or theother of the two spelling principles that providethe basis for the classification. It should also benoted that the segmental principle did not developin Egypt by elaborating on the syllabic principle,but rather by generalizing from the segmentaltranscription of morphemes: The syllable playedno role. And, conversely, when Sumeriansanalyzed complex syllables, they did not resolvethem into their constituent phonemes, but ratherinto simpler syllables. The discovery of onemethod almost seems to have guaranteed that theother would not be discovered. In effect, speakersof these languages come to regard them as asessentially syllabic or as essentially segmental,and their writing systems reflect one of these twophonological theories.

TRANSMISSION OF WRITINGSYSTEMS

It has already been noted that orthographictraditions are either consistently syllabic orconsistently segmental. Some explanation for thisconsistency is required. It seems natural enough,perhaps, that a segmental tradition should notbecome syllabic, for this would appear to be abackward step. But that no syllabic tradition

138 Mattingly

should have become segmental is puzzling, themore so because there have been at least twooccasions when such a development mightreasonably have been expected. The first waswhen speakers of Akkadian, an Mro-Asiaticlanguage with consonantal root structure similarto that of Egyptian and Hebrew, borrowedSumerian syllabic writing. A proper awareness ofthe morphophonology of their language wouldhave suggested that they convert the Sumeriansystem into a consonantal system. But instead,the Akkadians preserved the syllabic character ofthe borrowed writing, even though to write thesame triconsonantal pattern in different waysdepending on the particular inflectional vowelsobscured the roots of native words. Similarly, theMycenaean Greeks borrowed Minoan syllablewriting, and instead of making an alphabet out ofit, as would have been sensible, given theextensive consonant clustering in Greek, theycontinued to write with signs that stood for CVsyllables, either ignoring the "extra" consonants orpretending that they were syllables. This resultedin such bizarre transcriptions such as ARE KU TURU WO for alet1ruGq 'cock' (Ventris & Chadwick,1973). What can have happened to linguisticawareness in these cases?

The explanation begins with the observationthat the mismatches between language and writ­ing observed for Akkadian and Mycenean Greekare not unparalleled; they are simply fairly ex­treme cases. While an originally invented writingsystem clearly reflects the morphophonologicalstructure of the language it was invented to write,this situation is obviously exceptional. In general,the system used at a particular time to write aparticular language has been inherited from anearlier stage in the history of that language, orhas been adapted from a system (itself perhaps anadaptation) used for some other language, or,most commonly, both. The consequence, in manycases, is that the writing often seems very poorlysuited to the spoken language. If Akkadian andMycenaean Greek illustrate the risks of borrow­ing, the English writing system is a good illustra­tion of the effects of orthographic inheritance. Thephonology of English has changed considerablysince the fifteenth century, most notably in conse­quence of the Great Vowel Shift, but the writingsystem has remained very much as it was then(Pyles, 1971). As a consequence, the system has anumber of features that must seem very peculiarto the foreigner learning English: For example,the same letter is used to write phonetically dis­similar vowels, a tense vowel is denoted by an E

after the following consonant, and a lax vowel isdenoted by the doubling of this consonant. A simi­lar account could be given for Chinese writing,which corresponds more closely to ClassicalChinese than to any modem dialect.

It cannot be doubted, given what has beenleamed in recent years about the relation betweenorthographic structure and learning to read inmodem languages, that such complications placea heavy burden on the learner (Liberman,Liberman, Mattingly, and Shankweiler (1980).What is surprising, given the close connectionbetween literacy and awareness of linguisticrepresentations, a connection clearly essential inthe invention of writing, is that readers andwriters have so often happily accepted (once theyhave learned it) an orthography that seems poorlymatched to their language. It might have beenexpected that Akkadian cuneiform would havebeen rejected as soon as it was proposed, and thatEnglish orthography would by now have beenabandoned as obsolete. But, instead, it is reportedthat the Akkadians believed their writing systemto be of divine origin (Driver, 1976), and Chomskyand Halle (1968) say that "conventional [English]orthography is...a near optimal system for thelexical representation of English words" (p. 49).

In the case of inherited orthographies, the ex­planation may be that the orthography itself maydetermine not only which aspects of linguistic rep­resentations are singled out for awareness, butperhaps, indirectly, the character of these repre­sentations themselves. This could come about ifthe orthographically based, synthetic input repre­sentations were taken seriously by the languageprocessor as evidence about the structure of thelanguage, and thus led to adjustments in the be­ginning reader's morphophonology. It will be re­called that according to the sketch of the readingand writing process given earlier, the processordoes not distinguish synthetic representationsfrom natural ones. Consistent with this possibilityis the fact that orthographic conventions some­times mimic phonology: The conventions formarking English tense and lax vowels invite thereader to assume that underlying lax vowelsbecome tense in open syllables and underlyingtense vowels become lax before underlyinggeminate consonants. Such pseudophor . ')gicalrules, as well as derivational morpl gicairelations as those between heal, h".th ortelegraph, telegraphy, though at first havingmerely orthographic status, may acquire linguisticreality for the experienced reader.9 For such areader, the orthography corresponds to linguistic

Unguistic AWQmteSs and Orthographic Form 139

representations because the representationsthemselves have been appropriately modified, andEnglish orthography now indeed seems "nearoptimal.-

In the case of borrowed orthographies, a similarexplanation may apply. The phonologicalawareness of a borrowing group, such as theAkkadians or the Greeks, was not guided bypeculiarities of their own spoken language, as wasthe awareness of the original inventors of writing,but by the writing system they were borrowing.This is hardly surprising: The borrowers were notsophisticated consumers, comparing competingtechnologies to decide which was better for theirparticular needs. They did not realize that therewas a choice that could be made between the twodifferent spelling principles and the theories ofphonology implicit in each. They simply embracedunquestioningly the spelling principle-syllabic inthe cases considered above-used by the cultureunder whose influence they had come, just asbeginning readers accept the principle of thewriting system they inherit. This principle havingbeen accepted, the morphophonologies of theborrowers adjusted so that their linguisticrepresentations became, in fact, a good match totheir syllabic orthographies.

If this account is correct, it has to apply to thetransmission of segmental systems, as well. Asegmental system has obvious advantages over asyllabary for languages with complex syllablestructure. But the spread of the alphabet isperhaps to be explained by an appeal to the forcesof tradition rather than to those of reason.

An orthographic tradition can perpetuate itselfbecause it offers a particular brand ofmorphophonological awareness ready-made. Theprocesses of introspection needed to invent writingin the first place are not demanded. The kind ofawareness offered may be poorly matched to aparticular language, but this does not impede theprocess. Whether the writing system is borrowedor inherited, the morphophonology of the newreader adjusts to meet the presuppositions of thesystem.

CONCLUSIONSIt has for some time been widely agreed that the

notion of linguistic awareness is essential for anunderstanding of the reading process, the acquisi­tion of reading and reading disability. This notionis likewise essential for an understanding of theinvention and dissemination of orthographies.There are really only two possible ways to write,the syllabic method and the segmental method,because only by using one of these two methods is

the writer assured of being able to write any wordin his language. But for an illiterate to discover ei­ther of these methods, and thus be in a position toinvent writing, requires awareness of the appro­priate unit of linguistic representations.Awareness of syllables, or, on the other hand, ofsegments, is fostered by special morpho­phonological properties found in those languagesfor which writing systems were invented, thoughby no means in all languages. But once it hasbecome established, the writing system itselfshapes the linguistic awareness, and even thephonology, both of those who inherit the systemand of those who borrow it to transcribe someother language. Thus, in the history of writing,syllabic and segmental traditions are clearlydistinguished.

REFERENCESBaddeley, A. D. (1986). Working memory. Oxford: Clarendon Press.Chao, Y.-R. (1968). LAngUAge lind symbolic systems. London:

Cambridge University Press.Cheng, C. M. (1981). Perception of Chinese characters. Ada

PsychologiCI' TIliWllnia, 23, 137-153.Chomsky, N. (1957). Syntactic structures. The Hague: MoutonChomsky, N., &r Halle, M. (1968). The sound pattern ofEnglish. New

York: Harper &r Row.Cross, F. M., &r Freedman. D. N. (1952). FArly Hebrew orthography:

A study of the epigraphic eTJidence. New Haven: AmericanOriental Society.

DeFrancis, J. (1950). NationAlism and langUAge reform in Chil1ll.Princeton. NJ: Princeton University Press.

DeFrancis, J. (1989). Visible speech: The diTJeTSe oneness of writingsystems. Honolulu: University of Hawaii Press.

Driver, G. R. (1976). Semitic writing: From pictogTllph to alphabet.London: Oxford University Press.

Edfeldt, A. W. (1960). Silmt speech and silmt rmding. Chicago:UniVersity of Chicago Press.

Edgerton, W. F. (1952). On the theory of writing. 10u,.,l111 of NmrEAstern Studies, 11, 287-290.

Fodor, J. A. (1983). The modularity of mind. Cambridge, MA: MITPress.

Gelb, I. J. (1952). A studyofwriting: Thepundations ofgrtmmllltology.Chicago: University of Chicago Press.

Gelb, I. J. (1963). A study of writing (rev. ed.). Chicago: Universityof Chicago Press.

Hill, A. A. (1967). The typology of writing systems. In W. M.Austin (Ed), Pllpers in linguistics in hotwrofLeon Dostert (pp.92­99). The Hague: Mouton.

Hoosain, R. (1991). Psycholinguistic impliCtltions fur linguisticrrlatitrity: A CIISe study of Chinese. Hillsdale, NJ: LawrenceErlbaum.

Hyman, L. (1975). Phonology theory lind I1l1Illysis. New York: Holt­Rinehart-Winston.

Jensen, H. (1970). Sign, symbol lind script: An IIccount ofman's effurtsto write (3rd ed.). Tr. G. Unwin. London: G. Allen &r Unwin.

Kelley, D. H. (1976). Deciphering the Mllyan script. Austin. TX:UniVersity of Texas Press.

Kramer, S. N. (1963). The SumerlIIns: Their history, culturr, lindchaTllder. Chicago: University of Chicago Press.

Uberman, A. (1992). The relation of speech to reading andwriting.

140 Mattingly

Liberman, A. M., &: Mattingly, I. C. (1991). Modularity and theeffects of experience. In R. R. Hoffman &: D. S. Palermo (Eds.),Cognition lind the symbolic processes: IIpplied lind ecologiulperspedit1eS (pp. 33-38). Hillsdale, NJ: Lawrence ErIbaum.

Liberman, 1. Y., Liberman, A. M., Mattingly, I. C., &: Shan}cweiler,D. (1980). Orthography and the beginning reader. In J. F.Kavanagh &: R. Venezky (Eds.), Orthography, reading, anddysle:riR (pp. 137-153). Baltimore: University Park Press.

Liberman, 1. Y., Shankweiler, D., Liberman, A. M., Fowler, C., &:Fischer, W. F. (1977). Phonetic segmentation and recoding inthe beginning reader. In A. S. Reber &: D. A. Scarborough(Eds.), Towllrds Il psychology of retlding (pp. 207-225). Hillsdale,NJ: Lawrence Erlbaum.

Mattingly, I. C. (1972). Reading, the linguistic process, andlinguistic awareness. In J. F. Kavanagh &: I. C. Mattingly,LAnguage Uy eIIr Ilnd Uy eye: The reilltionships between speech Ilndretlding (pp. 133-147). Cambridge, MA: MIT Press.

Mattingly, I. C. (1985). Did orthographies evolve? RASE RemedilllRnd Specild EduClltion, 6(6),18-23.

Mattingly, I. C. (1987). Morphological structure and segmentalawareness. CPC uhiers de psychologie cognitive, 7, 488-493.

Mattingly, I. C. (1991). Reading and the biological function oflinguistic representations. In I. C. Mattingly &: M. Studdert­Kennedy (Eds.), Moduillrity and the Motor Theory of SpeechPerception (pp. 339-3(6). Hillsdale, NJ: Lawrence Erlbaum.

Mattingly, I. C., &: Xu, Yi ( in preparation). Word superiority inChinese.

Morais, J., Alegria, J., &: Content, A. (1987). The relationshipsbetween segmental analysis and alphabetic literacy: Aninteractive view. CPC CIlhiers de Psychologie Cognitive, 7,415­438.

Morais, J., Cary, 1., Alegria, J., &: Bertelson, P. (1979). Doesawareness of speech as a sequence of phones arisespontaneously? Cognition, 7, 323-331.

Potter, R. K., Kopp, C. A., &: Kopp, H. C. (1966). Visible speech (2nded.). New York: Dover.

Pyles, T. (1971). The origins and development of the English umguage(2nd ed.). New York: Harcourt Brace Jovanovich.

Ray, J. D. (1986). The emergence of writing in Egypt. WorldArchtu:ology, 17(3), 308-316.

Read, C., Zhang, Y.-F., Nie, H.-Y., &: Ding, B.-Q. (1986). The abilityto manipulate speech sounds depends on knowing alphabeticreading. Cognition, 24, 31-44.

Reicher, C. M. (1%9). Perceptual recognition as function of themeaningfulness of the stimulus material. journlll ofExperimentlllPsychology, 81, 275-280.

Sampson, C. (1985). Writing systems A linguistic introduction.Stanford, CA: Stanford University Press.

Ventris, M., &: Chadwick, J. (1973). Documents in Mycenttelln Creek(2nd ed.) Cambridge: Cambridge University Press.

Wang, W. S.-Y. (1981). Language structure and optimalorthography. In O. J. L. Tzeng &: H. Singer (Eds.), Perception ofprint: Retzding resellrch in experimental psychology (pp. 223-236).Hillsdale, NJ: Lawrence ErIbaum.

FOOTNOTES-In L. Katz &: R. Frost (Eds.), Orthography, phonology, morphology,

and ?naming (pp. 1-16). Amsterdam: Elsevier Science Publishers(1992).

t Also University of Connecticut, Storrs.1It will be assumed here, following Celb (1%3), Jensen (1970),

DeFrancis (1989) and others, that there are six majororthographic traditions: (1) Mesopotamian cuneiform, beginningwith Sumerian (c. 3100 B.C.) and including Akkadian, cuneiformHittite, Urartian, Hurrian, Elamite, Old Persian; (2) Cretan,including Minoan Linear A, Mycenaean Creek Linear B,

Cypriote, and Hittite hieroglyphics, all probably derived from acommon source (c. 2000 B.C.); (3) Chinese, beginning withChinese itself (c. 1300 B.C.) and including Korean nonalphabeticwriting and JapaMR; (4) Mayan (c. 300 A.D.); (5) Egyptian(c. 3000 B.C); (6) West Semitic, beginning with Phoenician (c.1600 B.C.) and including Ras Shamrah cuneiform, Old Hebrew,South Arabic, Aramaic, and Creek alphabetic writing. FromAramaic derive Hebrew, Arabic, and many others; from Creekderive Etruscan, Latin, and many others. Cermanic runes andKorean alphabetic writing probably belong in this tradition also,though the derivations are not clear. All but the most dogmaticmonogeneticists would agree that the Mesopotamian, Cretan,Chinese, and Minoan traditions are probably independentdevelopments. But some scholars (e.g., Driver, 1976; Ray, 1986)would derive Egyptian writing from Mesopotamian, and some(e.g., Driver, 1976), with somewhat greater plausibility, wouldderive West Semitic from Egyptian.

2Egyptologists and molIt other students of writing believe thatEgyptian phonographic signs stand for consonants, the vowelsnot being regularly transcribed. But according to Celb, theystand instead for generalized syllables, e.g., the Egyptian signusually interpreted as consonantal w actually stands for wa, wi,we, wu, or wo, according to context. It is obviously difficult todistinguish these two accounts empirically. The only supportCelb offers for his position is that "the development from alogographic to a consonantal writing, as generally accepted bythe Egyptologists, is unknown and unthinkable in the history ofwriting" (Celb 1%3, p. 78). But this argument is clearly circular(Edgerton, 1952; Mattingly,1985).

3Celb (1952, 1963) proposed some cases in which syllabic systemsare supposed to have developed into segmental systems; but seeEdgerton (1952). Ethiopic writing, derived from the West Semiticconsonantal tradition, might be viewed as a syllabic systemderived from a segmental system, because the signs docorrespond to syllables. But, with a few exceptions, each signactually consists of a consonant letter plus a vowel mark, exceptthat a is left unmarked. As in the case of Indic systems, onecould argue about whether this is a consonantal or a plenesystem, but it is certainly not a syllabic system (Sampson, 1985).

4The proposals in this section are developed in more detail inMattingly (1991).

SJapanese kana conesp<Jnd., strictly speaking, to moras, which arenot equivalent to English syllables. But they do belong to ageneral class of phonological units that can be called "syllables"(see, e. g., Hyman, 1975).

6An earlier formulation of some of the proposals in this sectioncan be found in Mattingly (1987).

7DeFrancis (1950), protesting against the "monosyllabic myth,"has suggested that there actually were many polysyllabic wordsin Ancient Chinese, just as in Modem Chinese, but that only oneof the syllables in a word was transcribed in the writing. Thus,morphemes that appear from the writing to be monosyllabichomophones may actually have been polysyllabic morphemeswith common homophonous syllables. Y.-R. Chao's (1%8)response was that "so far as Classical Chinese and its writingsystem is concerned, the monosyllabic myth is one of the truestmyths in Chinese mythology" (p. 1(3). For the present purpose,however, it does not matter whether the myth is true or false.DeFrancis's partial homophony will serve as well as the totalhomophony more usually attributed to Ancient Chinese.

8Ot, on DeFrancis' (1950) view, another morpheme having asyllable in common.

9These changes in the morphophonologies of individual readershave, by hypothesis, no basis in the spoken language and aretransmitted only from writer to reader, and not from mother tochild. Thus, though psychologically real, they are not part of thegrammar of the language as usually conceived of.