Haskins Laboratories Status Report on Speech Research1990, SR-101 /102,28-40

Phonemic Segmentation as Epiphenomenon: Evidencefrom the History of Alphabetic Writing*

Alice Faber

"...we must not be misled in our ontology by the possibilities provided by our metalanguage."

(Anderson 1985: 9)

1. PRELIMINARY REMARKSThere is by now a large and convincing body of

evidence that linguistic units representingacoustic or articulatory steady states need not beincluded as primitives in linguistic represen­tations of phonological structure. Alternatives tosuch segments l being pursued in currentphonological work include both larger units thatin traditional analysis might be treated as morethan one segment and single features that mightbe part of the representation of one or moresegments in a word. Similarly, investigations oflanguage use suggest that many speakers do notdivide words into phonological segments unlessthey have received explicit instruction in suchsegmentation comparable to that involved inteaching an alphabetic writing system. Never­theless, alphabetic writing, writing whose symbolsrepresent individual segments, exists, and islearnable.

Paradoxically, then, alphabetic writing is basedon a phonological unit that is arguably notanatural unit (see also Studdert-Kennedy, 1987:68). How is this paradox to be resolved? To theextent that the paradox has been recognized, thesegmental nature. of alphabetic writing has been

I would like to thank Lloyd Anderson, Cathe Browman,Peter Daniels, Len Katz, Ignatius Mattingly, M. P. O'Connor,and Sue Schmerling for comments, discussion, and suggestionswhich, although not always incorporated herein, arenonetheless greatly appreciated. Extensive discussion withTom Sawallis contributed to the early growth of my ideas onthis topic. Support from NIH grant NS-007237 (to HaskinsLaboratories) during the writing of this paper is gratefullyacknowledged.

28

taken as paramount: Alphabetic writing could nothave evolved if the segmentation on which it isbased were not natural, and, therefore, theexistence of alphabetic writing is itself evidencef?.r the naturalness of phonological segmentation(Ohman, 1979: xviii; Saussure, 1959: 39; Studdert­Kennedy, 1987: 68).

In the absence of a convincing explanation forthe innovation of alphabetic writing that does notrely on the prior existence of segmentation abilityin the human repertoire, explanations of thisnature, however dissatisfying, cannot be rejected.2

My purpose in this paper is to provide just such analternative. After a brief review of some of theevidence regarding segments and segmentationability, I will provide a detailed exposition of howthe Greek alphabet might have developed. On thebasis of this scenario, I will suggest that segmen­tation ability, rather than being a necessaryprecursor to the innovation of alphabetic writing,was a consequence of that innovation.

2. THE STATUS OF PHONEMICSEGMENTATION

As already noted, most twentieth centuryapproaches to phonological analysis rely on theidealization that the speech stream is divided intodiscrete segments, each representing an acousticor articulatory steady state. Nevertheless, as iswell known, there is no. physical basis for thissegmentation.3 As a consequence, there is a longand often under-appreciated tradition of alterna­tive approaches to phenomena not easilyamenable to segmental analysis, most notably ofphenomena not easily localized within a word(e.g., Firth, 1957; Harris, 1944; Palmer, 1970).

Phonemic Segmentation as Epiphenomenon: Evidence from the History of Alphabetic Writing 29

Within "mainstream" theoretical phonology inthe decade following the publication of Chomskyand Halle (1968) (SPE), one of the majorinnovations was the integration of the syllableinto linguistic models. This integration wasprompted by the more elegant and perspicuoustreatment of stress in the newer models (Hayes,1980; Liberman & Prince, 1977; McCarthy, 1979)as well as by the range of apparently segmentalphenomena conditioned by syllable position(Hooper, 1972; Kahn, 1976).4 At least some ofthese alternatives refer also to constituents of thesyllable (e.g., Halle & Vergnaud, 1980: 93; Selkirk,1982; cr. Clements & Keyser, 1983 and Hooper,1972):5 the vocalic nucleus, the pre-nucleusconsonantal onset, and the post-nucleus coda. Thenucleus and the coda are generally groupedtogether as the rhyme. Just how these hierar­chically arranged units should be related topreviously hypothesized units like segments is notexplicitly discussed. Rather, it is assumed thatsyllable constituents dominate segmental nodes,so that a hierarchical syllable display can beunambiguously related to a distinctive featurematrix of the customary sort (Clements & Keyser,1983: 25), a column on such a matrix corre­sponding to a phonemic segment.

Given the demonstrable relevance of syllablecomponents in phonological structure, mainte­nance of segments in syllable-based analyses issurprising. The rich inventory of English syllabletypes gives rise to a tacit feeling that theinventory of phonological primitives would be pro­hibitively large. To my knowledge, only Fujimuraand Lovins (1978) have proposed an analysisbased on demisyllables (approximately, onsets andrhymes) and extra-metrical consonantal affixes.As Fujimura (1980: 122) notes, the memoryburdens imposed by an inventory containing c.1000 elements would not be excessive; Browman's(1980) demi-syllable-based speech synthesisprogram requires c. 850 demi-syllables andaffixes. This increased inventory size (relative totraditional analyses) is compensated for by agreatly decreased need for contextually condi­tioned phonological rules; Browman's model, forexample, requires 20 rules.

One of the major European contributions tophonological theory has been an emphasis on theparadigmatic relationships among the segments ofa language. Thus, for Trubetzkoy (1949), aphoneme is defined by its contrasts with otherphonemes. From here it is a short step to thedistinctive features introduced by Jakobson, Fantand Halle (1969), a formalization of the dimen-

sions along which segments can differ or besimilar. While the Jakobsonian distinctive fea­tures were based almost exclusively on acousticproperties of segments, virtually all currentfeature-based models rely on articulatory features.

In SPE, distinctive feature matrices were simplya two-dimensional array of rows representing fea­tures and columns representing segments. All ofthe specifications for one segment preceded any ofthe specifications for the next. This two-dimen­sional array has evolved in descendants of SPE, sothat the rows are spread out in three dimensions(Clements, 1985; Sagey, 1986; Schein & Steriade,1986). And, given underspecification of featurevalues (Archangeli, 1985), the notion of a columnrepresenting an individual segment is greatly ob­scured; individual feature specifications may, inunderlying representation, precede other featurespecifications, but, without the root node (Sagey,1986: 40-44, 275), to which all specifications for aparticular segment are linked, it would beimpossible to make precedence statements aboutsegments independent of their featuralconstituents.

Statements about the syntagmatic relationshipsamong individual features, independent of theirputative segmental affiliation, are an integral partof the model of articulatory phonology beingdeveloped by Browman and Goldstein (1986, 1987,in press). In this model, the primitives arearticulatory gestures, represented on a gesturalscore, a two-dimensional matrix. In contrast to the[±F] values of more traditional distinctive featuremodels, gestures are either present or absent, and,as already noted, the temporal relationshipsamong gestures on the same score are notconstrained, but must be specified on a languagespecific, or even a style specific, basis.

The approaches just sketched, converging onphonological representations in which segmentsper se play little, if any, role, also allow for moreperspicuous treatment of a class of sounds thathas been chronically insusceptible of principledtreatment in segment-based models-diphthongs,affricates, and prenasalized consonants. These aresounds that, by their very nature, cannot beconceived of as single steady states. As a result oftheir inherently dynamic nature, their existencehas given rise to long and, for the most part,fruitless discussions about whether each is besttreated as "one segment or two," as a complex unitor a simple cluster.6

To summarize, there is little unambiguousstructural support for positing segments aslinguistic units. In particular, there is no need to

30 Faber

appeal to segments in modeling ordinary languagestructure and use. Neither is there evidence that

.incorporation of segments into linguistic modelsleads to more satisfying or parsimonious analyses.In contrast, structural linguistic evidence suggeststhat models based on syllables and syllablecomponents (either onset-rhyme or onset-nucleus­coda) might be more appropriate. However, thereare spheres, notably reading, in which languageusers clearly do display an ability to recognize andmanipulate segments.

The English alphabetic writing system encodes,albeit not always systematically, the segmentalphonemic structure of English. While skilledreaders may not always make direct use of thephonological coding inherent in an orthographicform, they clearly can do so in the case of rare orunknown words. Studies of the development ofreading suggest that at least a minimal awarenessof segmentation is necessary for English speakingchildren to learn to read (Mattingly, 1972: 44).Small children tend to be much more proficient atsyllable counting tasks than at phoneme(segment) counting tasks, while literate older chlI­dren and adults can do both (Liberman &Shankweiler, 1987: 207).

Given a correlation between segmental aware­ness and literacy (in a segment~based alphabet),the question arises whether segmentation ability,arising spontaneously as part of cognitive matu­ration, is an indication of reading readiness, orwhether segmentation ability arises as a conse­quence of specific teaching. Liberman andShankweiler (1987: 210) refer specifically toreading instruction, but widely availablealphabet-related toys and books could easily con­tribute to incipient segmental awareness on thepart of pre-school children. Nevertheless, childrenwho are aware of the alphabet but who cannot yetread may interpret alphabetic symbols differentlythan do proficient readers. Read (1986: 51, 105)refers to two relevant types of studies. In one,conducted in Argentina, preliterate childreninterpreted each letter as representing a syllable.In another pair of studies, conducted in the UnitedStates, some beginning spellers were found towrite consonants only, omitting the vowels.

Further relevant findings concern adult popula­tions. Morais et al. (1986) studied adult illiteratesand students in adult literacy classes in Portugal.Newly literate adults outperformed illiterates ontasks involving phonemic segmentation, but noton comparable tasks involving non-linguisticsegmentation (Le., of notes in a melody).Similarly, Read et a1. (1986) compared two groups

of literate Chinese, those who had never beenexposed to the pinyin Romanization, and thosewho had, even if they were no longer proficient init. The pinyin group outperformed the non-pinyingroup in tasks involving segment addition ordeletion. Likewise, Mann (1986) found thatJapanese elementary school children are lessproficient than their American counterparts atperforming segment-related tasks.7 One basicconclusion can be drawn from these studies:segmental awareness is a result not of cognitivematuration but of exposure to a segment-basedorthography; literacy itselfis not sufficient.

Reading is clearly a metalinguistic ability.There are languages for which there exist neithera standard orthography nor a written literarytradition. Even some languages with long literarytraditions (e.g., Amharic) have large numbers ofilliterate speakers. Furthermore, the wholequestion of 'functional' literacy suggests that thereexist within literate populations substantialdifferences in reading proficiency. Thus readingcompetence cannot be equated with languagecompetence in general. If segmentation abilityresults primarily (if not only) from exposure tosegmental orthography, it too must be treated as ametalinguistic ability, and thus it cannot beascribed to general linguistic competence.8

Language games have been observed in manycultures, and have often been seen as sources ofevidence regarding linguistic structure. Languagegames generally involve systematic insertion ofextraneous linguistic material and/or permutationof structural linguistic units. Some games are bestdefined on syllable components, while others arebest defined on segments; in games defined onsyllable units, initial clusters like sp- would act asa unit, while in games defined on segments, s andp could act independently. Bagemihl (1987), in anextensive survey of documented language games,reports that games based on syllables are far morecommon than games based on segments; the latternaturally presuppose segmental awareness on thepart of their speakers. He further notes (1987: 36)that segment-based games are attested only inlanguages with alphabetic writing systems,suggesting that segmental awareness here too isdependent on exposure to alphabetic writing.

3. ALPHABETIC WRITINGThe evidence just summarized suggests that,

despite their intuitive appeal and longevity,segment-based analyses might not be optimallysuited to represent language asa cognitive sys­tem. A careful examination of the complex

Phonemic Segmentation as Epiphenomenon: Evidence from the History of Alphabetic Writing 31

relationship between orthographic and linguisticunits is thus in order. This examination willproceed in two stages. First, I will motivate atypology of writing systems. Then, on the basis ofthis typology, I will discuss the extent to whichthe development of orthographic systems bears onthe question of phonemic segmentation, and justwhat that bearing is. My goal is, in particular, totrace the development of alphabetic writing, sinceit is the existence of alphabetic writing that isused (implicitly or explicitly) as evidence for theuniversality of segments as a building block oflanguage. My discussion is informed by the viewexpressed by O'Connor (1983: 441) that thestructure of an orthography for a particular lan­guage reflects, albeit not always systematically,native speaker analysis of that language. A scriptcreated for a language ex nihilo will reflect boththe levelof appropriate units and the inventory ofunits at that level. In contrast, a borrowed scriptmay inherit the level of units, but still reflect insome measure the inventory of the borrowinglanguage.

3.1. Typology of writing systemsAn orthography can be defined as a system of

markings which can unambiguously cue a trainednative speaker to produce acceptable utterances inthat language (Faber, 1987). Similar definitionshave been offered by Sampson (1985: 19) ("a givenset of written marks together with a particular setof conventions for their use") and Daniels (1986: 1)("the sets of marks by which utterances are no­tated so as to be precisely reproducible in the ab-

sence of the notator or the notator's instructions").These definitions agree in referring to the arbi­trary yet non-idiosyncratic nature of the signs andto the training involved in their interpretation, aswell as in not restricting the linguistic level of theunits. My definition further excludes notationalsystems that can be interpreted in arbitrarilymany ways by speakers of the same language (or,for that matter, of different languages). Thus, theset of "iconic" international road signs does notrepresent an orthography, despite the relativelyarbitrary nature of the symbols and their suppos­edly unambiguous semantics.

3.1.1. TerminologyDefinition of orthography provides a useful point

of departure for discussion of orthographicsystems. But it is not enough. Before attestedorthographies can be classified in any meaningfultypology, it is necessary to define both thedimensions along which they can differ and theresultant categories. Not to do so invitesconfusion. The debate over whether the 2nd. and1st. millennium BC West Semitic scripts­Canaanite, Ugaritic, Phoenician, Aramaic, OldSouth Arabian, in plain type in Figure 1,9 but nottheir vocalized descendants, in bold type-aresyllabaries (Gelb, 1963; Swiggers, 1983, 1984),alphabets (Daniels, 1986; Naveh, 1982), or neither(as I will argue shortly) hinges not on differingconceptions of the West Semitic scripts but ratheron differing definitions of alphabet and syllabary.Except where noted, the following definitions arebased on Sampson (1985: 32ft).

West Semitic

Canaanite

Old SouthArabian

IEthiopian

Old Hebrew

Syriac later Hebrew

Ugariticcuneiform

Aramaic

Indian scripts(e.g. Devanagari)

Arabic

Figure 1. (Simplified) depiction of relationships among Semitic orthographies. Legend: italic-hypothesized,unattested scripts; plain-scripts representing primarily (or only) consonants; bold-scripts representing consonantsand vowels.

32 Faber

Logographic vs. Phonographic. Orthogra­phies differ in whether the bulk of the symbols inthe system, taken in isolation, are susceptible ofsemantic interpretation. The English symbol <d>has no inherent meaning, while the symbol <+>does. Systems in which symbols like the formerpredominate are phonographic while systems inwhich symbols like the latter predominate arelogographic. Logographic systems code for themost part morphemes, as Sampson notes, andphonographic systems code units of sound.

Complete V8. Defective. An additionaldimension along which orthographies vary is theextent to which they aspire to code all units of therelevant level. Complete orthographies code all (orvirtually all) relevant units, while defective orincomplete orthographies omit some. Orthogra­phies can fail to be complete in two distinct ways.The primary way in which orthographies can bedefective is that certain units may simply not berepresented. This sort of omission is the basis forthe popular misstatement that Hebrew at onetime didn't have any vowels. The earliest Hebrewscript, like the other West Semitic scripts alreadyreferred to, did not contain any symbols for vowelsounds, even though the language had atypologically normal inventory of vowels. Anotherway in which orthographies may be incomplete(one which will not be discussed further here) isthat one or more phonemic contrasts may not berepresented, the same symbol being usedindifferently to represent two or morephonologically distinct units. Thus, the Englishdigraph <th> is used for both voiced and voicelessinterdental fricatives. In addition, it is fairlycommon for orthographies that are otherwisefairly complete not to indicate quantity and tonalcontrasts, even in languages in which suchcontrasts bear a heavy functional load. AsSampson notes (p. 36), no orthography is evertruly complete. Most orthographies notate few orno non-contrastive features; indeed special pur­pose orthographies that do aspire to this level ofcompleteness (e.g., narrow phonetic transcription)are notoriously difficult even for trainedindividuals to read fluently.

Syllabic VS. Segmental. Phonographic scriptsdiffer with regard to the linguistic level that theycode. Thus symbols in syllabic orthographies codesyllables, and symbols in segmental orthographiescode segments)O It is, I believe, plausible to claimthat the syllable is a relevant linguistic unit forusers of a syllabic orthography. But the parallelclaim regarding segmental orthographies is lessplausible. Certainly, users of a complete

segmental orthography must, at some level, beaware of those units represented by its symbols.But, as noted in the previous section, someorthographies are systematically incomplete. Andit does not follow that users of such defectiveorthographies are aware of units or contrasts thatare not reflected in their orthography. In fact, theappropriate implication may well be in the reversedirection; the failure of the early Semiticsegmental orthographies to code vowels mayindicate that users of these orthographies did notsegment syllables like /bal into their constituentparts, despite their awareness of the similarityamong /bal, /bu!, /bil, and even perhaps /tab/.Awareness of this similarity, while prerequisite tosegmentation, does not, as noted above, implyexhaustive segmentation of the speech stream (cf.Mattingly, 1987). Rather, the vowel segments notcoded in these orthographies constitute thebackground for the consonantal segments, whichare coded)!

The question is not, and cannot be, whetherusers of early Semitic consonantal orthographieshad the ability to recognize and manipulate thephonemic segments of their language. It is, rather,which abilities can be imputed to users of thesescripts on the basis of the structure of the scripts.Segmentation ability presupposes an awarenessnot only of the similarity among /bal, /bil, and/tabl, but also of the similarity among /bal, /tal,and naIl. It further implies localization of thesesimilarities in the segments /h/ and /a/,respectively. Evidence that users of anorthography had one of these abilities (in thiscase, the first), does not speak to whether theyhad either (or both) of the other two; they mayhave, or they may not.

Linear vs. Non-linear. Orthographies alsodiffer in regard to linearity. In a linearorthography, the order of the symbols, eitherhorizontally or vertically, iconically mirrors theorder of the sounds that they represent. Thus, in"phonetic languages" like Spanish, the letters insequence reflect the sequence of sounds. Incontrast, English orthography is not as linear. Thecontrast in the medial vowels ofcop Ikapl and copeIkop/ is orthographically conveyed by the presenceof final -e in the latter, so the vowel /0/ isrepresented by the discontinuous sequence ofletters o-e. In more systematically non-linearscripts like Egyptian hieroglyphics or KoreanHan-gul (Sampson, 1985: 131), the placement ofsymbols may be governed by aesthetic rather thanby linguistic principles. Other major non-linearscripts include two segmental scripts that are

Phonemic Segmentation as Epiphenomenon: Evidence from the History of Alphabetic Writing· 33

commonly and erroneously treated as syllabaries:Indian Devanagari (and allied South Asianscripts) and the Ethiopian script. While theorganizational pattern of both is more or lesssyllabic, as Sampson (1985: 66) notes, theindividual graphs clearly encode segmentalinformation. However, as shown in Figure 2,12 theplacement of the graphic modifications for vowelcontent in the Ethiopian script, represented hereby Amharic, is such that a CV syllable may,depending on the vowel, be modified on either theleft side or the right side of the symbol.l3

Even those who recognize the linearity factorassume that the linearity of a phonographic scriptis less significant than the linguistic level encodedin its symbols. That is, the set of syllabic scriptsand the set of segmental scripts can both bedivided into linear and non-linear subsets. I wouldnow like to propose an alternative view. In thatview, linearity is hierarchically superordinate tothe syllabic/segmental dimension. Furthermore,the linguistic level at which a script can be seen aslinear is logically distinct from the linguistic levelof the units that it encodes. Thus, the Ethiopianscript encodes segments but is syllabically linear,while European scripts encode segments and aresegmentally linear.

This property of the Ethiopian scripts isillustrated in Figure 3 with the Amharic word/leba! "thief." Note the non-linear mapping ofgraphs to segments, contrasted with the linear

mapping of graphs to syllables. Of course, anysegmentally linear script will necessarily besyllabically linear, while the reverse is not true.l4

Alphabet. An alphabet can be defined as ascript that is (relatively) complete and, in its idealcase, segmentally linear. By this definition, theearly Semitic scripts, which encode for the mostpart consonants, are not alphabets; since they donot encode vowels, they are systematicallyincomplete. Likewise, the later Semitic systemsthat, like Arabic, do encode some vowels, are not,strictly speaking, alphabets, since they aresyllabically rather than segmentally linear.15 Thediacritic markings that indicate vowel quality inArabic, Hebrew, and Syriac may occur above,below, within or after (to the left of) the symbol forthe preceding consonant sound.l6 I am, in effect,giving special definitional status to just thoseorthographies whose existence has been taken toreflect the innate naturalness of phonemicsegmentation. The orthographies that I define asalphabets turn out to be those that have equallysalient symbols for vowels and consonants. AnEnglish speaker unfamiliar with Russian willhave no a priori intuitions about whether aparticular Russian symbol represents a vowel or aconsonant. One might say, then, that an alphabetis a script which treats vowels and consonantsequally and that, in order to do so, a script mustparadoxically recognize vowels and consonantsboth as separate and equa1. 17

a (g) U 1 a e g (i)/O 0

h U lJ' ~.., l{ U l.f

1 ~ rr. h. ?'\ Cb ~ hom ao lJO., crt;. 011 Olb ~ qo

s n It' r'1. II rb n \1b n n- Il. 11 fb 11 Ilt + oF t :t- ot t tw (J) (J).. 01. q> ~ a:>- ~

y ~ ~ It ~ ~ B ~

~ m m. ITl rn fIb f\l rn

Figure 2. A subset of the Amharic orthography. Amharic, a Semitic language, is the national language of Ethiopia.

34 Faber

le-baI I~Il

I IXl-e-b-a

Figure 3. The segmentally coded, syllabically linearnature of the Ethiopian script, illustrated with theAmharic word LEBA Uthief."

In contrast, Daniels' (1986: 8) definition of analphabet as a system in which "each characterconveys information about just one segment"18implies that the primary subdivision of segmentalscripts should be based not on properties of thesound-to-symbol mapping but rather on the visualautonomy of the symbols participating in it.Naveh's (1982: 11) functional definition of analphabet as an orthography containing a limitednumber of symbols (20-30) that can be arranged ina fixed citation order (A BCD ...), which, likeDaniels', includes the West Semitic scripts, is, Ibelieve, inappropriate, in that it makes explicitreference to metalinguistic, perhaps pedagogical,facts rather than to the structure of the sound-

symbol mapping. It should be noted that mydisagreement with Daniels and Naveh concernsonly whether the term "alphabet" should berestricted to complete, segmentally linearorthographies. I share Daniels' view that theGreek development of a complete, segmentallylinear orthography, by whatever name, is not to becausally linked to the development of Greek, andhence Western, civilization. And, I. mostemphatically would not follow Gelb and, perhaps,Swiggers in treating those segmentally linearSemitic scripts that do not encode vowels assyllabaries. It is not an "either-or" question.The dimensions phonographic/logographic,syllabically/segmentally linear, complete/defective,and syllabically/segmentally encoded define fivecategories of phonographic orthographies, asillustrated in Figure 4.

My focus on which of these categories containorthographies that it is appropriate to label"alphabet" has no bearing on how orthographies inthe other categories should be labeled.19Interestingly, Sampson (1985) offers no definitionof alphabet, but uses the term at various points torefer to the Northwest Semitic scripts. In anycase, the set of alphabets is clearly a subset of theset of segmental orthographies.

logographic

ChineseAncient Egyptian

defective

UgariticPhoenicianAramaicOld South ArabianOld Hebrew

complete(=ALPHABET)Greco-LatinCyrillic

segmen allycodedHebrewSyriacArabicEthiopian

AmharicDevanagari

phonographic

SYllabicall~ar ~ntall linear

syllabicallycodedAkkadianJapanese (kana)

Figure 4. Some representative orthographies arranged according to the types defined in the text.

Phonemic Segmentation as Epiphenomenon: Evidence from the History of Alphabetic Writing 35

3.1.2. Distribution of writing systemsIf the distribution of orthographic types is

charted based on the number of users, alphabeticwriting systems are by no means the mostwidespread. While alphabets are used by about abillion literate individuals,20 world-wide, acomparable number use either logographic writing(Chinese) or syllabically linear systems, used inNortheast Mrica (Ethiopian syllabary), SouthernAsia (Devanagari and allied systems), and Japan.If the Arabic-based scripts (used for Farsi, Urdu,Pashto, Malay, etc., in addition to Arabic) which,as already noted, are syllabically linear whenvocalized even though they encode segments, areincluded, the distribution of syllabically linearscripts is even greater. Thus, (pace Lotz, 1972:119) syllabically linear scripts (which he refers toas syllabaries) are not typologically marginal; boththe Japanese true syllabary and the syllabicallylinear segmental scripts mentioned above havebeen quite stable for 1500 years and more.

The Greco-Latin scripts, including Cyrillic,while geopolitically of great importance, are morecircumscribed demographically. And most of theirexpansion has occurred in the past 500 years. Inthe more distant past, the relative importance ofsyllabically linear scripts was much greater. Thefurther back one goes in the history of writing, thegreater the demographic preponderance of non­linear and syllabically linear scripts: Sumero­Akkadian cuneiform, and Linear B, used forMycenean Greek,21 were syllabically linear andincomplete; Egyptian hieroglyphic writing was, asalready noted, non-linear.

Given the five-thousand year long, incompletelydocumented history of writing it is tempting, and,perhaps, natural to see, as does Gelb, adevelopmental, culturally maturational trendtoward segmentally linear scripts-that is, towardalphabets as defined earlier. However, such a viewwould be inappropriate, not to mentionethnocentric. All known complete, segmentallycoded orthographies (see Figure 4) arose from theearly West Semitic segmentally linear incompletescripts (Sampson, 1985: 77). All of these completedescendants maintained linearity-the Greco­Latin scripts on a segmental level, and thevocalized Semitic scripts and the Ethiopian andDevanagari scripts (if Devanagari is, in fact, to betraced to West Semitic origins) on a syllabic level.Furthermore, the West Semitic segmentally linearirrcomplete script is the only known representativeof its type. In order to attribute the innovation ofeither a segmentally linear script or of analphabet to cultural maturation, we would need

evidence of a recurrent pattern. While suchevidence is not in principle impossible, it is,possibly as a result of an accident of humanhistory, the case that each of these innovations isunique. It is thus in practice impossible todetermine whether either innovation was a sign ofcultural progress or merely the product of thespecial circumstances in which it took place. Tothe extent that explanations for these innovationscan be based on the specific circumstances inwhich they took place, it is unnecessary,unparsimonious, and even imprudent tohypothesize more abstract, putatively causalprinciples.

3.2. The Greek adaptation of theCanaanite system

If it is only use of an alphabet that bespeaksawareness of phonemic segmentation, then it isworth examining the early developme'nt of theonly alphabet ever known to have arisenspontaneously in human history, the Greekalphabet. The Classical Greek orthography of c.400 BC, illustrated in Figure 5, has seventeensymbols for consonants and seven for vowels.Several phonemic contrasts (frY vs. Ig/,22 Ial vs. Ia/,Iii vs. N, lUI vs. 101) are not directly represented,but, aside from these marginal 'defects', compa­rable to those observed in later adaptations of theorthography, the Greek alphabet departs from theideal of a con:tplete, segmentally linearorthography in two respects: (1) the rough andsmooth' breathing marks, post-classicalinnovations which denote /hi and nt onsets to aninitial vowel, are written as diacritics over theletter representing the vowel: a, n. And, (2) thereare symbols for two clusters Iks/ and Ipsl, which,as Sampson (1985: 103) notes, following Allen(1974: 57), are the only two clusters to occur incoda position in ancient Greek.23 The letters B ksiand \lJ psi are used, however, even when theclusters are split by a morpheme boundary. Theorthography of the earliest Greek inscriptions,which date from the eighth century BC, whilearchaic in some respects, is typologicallycomparable to that of later Greek, that is to say,alphabetic.

The accepted wisdom, in, e.g., Jeffery (1961), isthat the Greeks adopted the Phoenician alphabetin the eighth century BC, shortly before the oldestattested inscriptions. However, in light of the vastincrease in recent years in the size of the Greekand Phoenician inscriptional corpora (assummarized in, e.g., McCarter, 1975), as well asincreased sophistication in dating inscriptions in

36 Faber

all relevant languages, the situation is less clear.Isserlin (1983) notes a lack of scholarly consensusregarding the time of the borrowing (eighthcentury BC, as argued by Jeffery, or eleventhcentury BC, as suggested by Naveh [1972]), thegeographical context of the borrowing (Aegean,Levant, or Asia Minor), and the orthographicsource of the Greek alphabet (Phoenician,Aramaic, or (unattested) Canaanite; Kaufman,1987). As all of these potential sources weretypologically comparable in all relevant respects, Iwill, for present purposes, refer only to Canaanite.

Any explanation for the development of theGreek orthography must take into account thatthe development was essentially complete by thetime of the earliest Greek inscriptions; there areno traces available either of Canaanite(Phoenician or Aramaic) representation of vowelscomparable to that attested in Greek24 or of Greek

material in which vowels are not represented.Furthermore, all of the symbols which representGreek vowels (with the trivial exception of the lateQ) are derived from Canaanite letters for soundsthat were, at best, marginal in Greek: n h £ y wi.While it is, indeed, possible that these symbolswere deemed superfluous by the importers of theCanaanite system, and thus used for sounds notrepresented by that system, it is not necessary tobase an explanation for the structure of the Greekalphabet on the unattested existence of anunknown genius.

The names of the Greek letters alpha, beta, etc.,meaningless in Greek, have clear sources in aCanaanite acrophonic tradition, whereby eachsound is associated with an object whose namebegins with that sound. This fixed order of atraditional, invariant list is comparable to modernradio alphabets like able, baker, charlie, etc.

Greek Canaanite Greek CanaaniteA. all A ~ ks/ts

~ f............

b B q ofi 0 0

9 r f\ p TI 7d 6 q s <~) \"e/h E ~ q (9) ~w <F) 'l (1.) r P ~z Z "'2,.. s L: Vv

h(e) H ~ t T tth 8 0 B. u Y ('l)i/y I 1- ph ¢ <0*

k K 1 kh/ks X lI'*

1 1\ L ps/kh qt lI'*

m M 'I C. 0 Q

N 1 #ha (

n a#1a

,a

Figure 5. 'The Greek alphabet with its Canaanite precursor, represented by an Old Aramaic script. Parenthesized Greekletters are found in archaic materials only, and are not preserved in Classical Greek. Parenthesized Canaanite symbolsrepresent deviations from the Semitic citation order. Group A includes Greek letters adopted directly from aCanaanite source and mirroring the source citation order. Where two values for these letters are given, the first onerepresents the Greek value and the second the Semitic value. Group B contains letters based on Semitic prototypesthat were added to the end of the Greek alphabet. 'The alternate values represent variation in early Greek orthographictraditions. 'The starred source symbols are potential South Arabian symbols for Greek letters which have no obviousCanaanite prototype. Group C represents indigenous Greek developments.

Phonemic Segmentation as Epiphenomenon: Evidence from the History of Alphabetic Writing 37

As Sampson (1985: 101) demonstrates, conclu­sively to my mind, transmission of the Canaanitescript, using the acrophonic principle would haveled to the misinterpretation of several Canaaniteconsonant symbols as representing vowelsinstead. The Canaanite words 'lalpa "cow," he "?,"yoda "hand," and fqyna "eye," standing for If:-11,~lhl, 1..1yl, and O/ru, would have beenperceived by speakers of a language in which, asin Greek, these sounds did not occur, as beginningin [a], [e], [i], and [0], respectively.25 Thus,Phoenician [1alpa], with an initial [1] becameGreek [alpa], with no [1]. Identification of [a] and[0] with Greek Ia! and 101 is plausible,26 and theuse of 1.. yod for Iii, not implausible from aphonetic point of view, may be related as well tothe near absence of Iyl in Greek and to earlyCanaanite scribal practices like those observed forUgaritic, in which <y> occasionally represented Iii.The form of Y upsilon appears to be based onthe archaic Canaanite Y wau, whose place in theearly Greek alphabetical order was filled by Fdigamma, the shape of which is in tum based onCanaanite l..yocl (Allen, 1974: 45).27

4. IMPLICATIONSIn this paper, I have demonstrated that it is

necessary neither to impute to users of WestSemitic consonantal scripts segmentation ability,as commonly understood in psycholinguisticsresearch, nor to appeal to such ability to explainthe Greek innovation of alphabetic writing. Infact, segmentation ability as a human skill mayhave been a direct result of (rather than animpetus to) the Greek development of alphabeticwriting. Thus, the existence of alphabetic writingcan not be taken eo ipso as evidence for thecognitive naturalness of the segmentation that itreflects. Given this conclusion, it is necessary topose the question of why virtually all linguistshave fallen prey to what Ladefoged (1984: 92) has,in a comparable context, referred to as a phonemicconspiracy. That is, we as linguists feel that,because we can describe linguistic systems interms of phonemic segments, we must do so. Thatwe are influenced in this practice by the structureof the alphabets that we use in our ordinary livesis obvious. As far as I know, every technicallinguistic tradition that refers to segments arosein an alphabetic milieu or was influenced directlyby such a tradition. There is no segmentalanalytic tradition not supported in this way by anorthographic tradition. In contrast, the indigenousChinese linguistic tradition, described by Halliday(1981), has as phonological primitives syllable

initials and finals, that is, onsets and rhymes.This analytical division is not supported by thelogographic Chinese orthography, a lack whichstrengthens the force of the analysis.

Even though segment-based linguistic modelsare accurate in a wide range of cases, in particularin those in which syllable-constituent modelsreduce to segment models (e.g., CV syllables), theevidence that I have summarized in this papersuggests, however, that they are systematicallylimited in the extent to which they can model awide range of linguistic data. These results lead toa challenge for future research, to develop andrefine a rich set of non-segmental metaphors andnotations, so that models based, like thosesketched in Section 1 above, on differentphonological primitives can be tested against awide range of linguistic data, and revisedappropriately.

REFERENCESAllen, W. S. (1974). Vox Graeca (2nd. ed.). Cambridge: Cambridge

University Press.Anderson, S. R. (1974). The organization of phonology. New York:

Academic Press.Anderson, S. R. (1976). Nasal consonants and the internal

structure of segments. LAnguage, 52,326-344.Anderson, S. R. (1985). Phonology in the twentieth century. Chicago:

University of Chicago Press.Archangeli, D. (1985). Yokuts harmony: Coplanar representation

in nonlinear phonology. Linguistic Inquiry, 16,335-372.Bagemihl, B. (1987). The crossing constraint and 'backwards'

languages. Manuscript, University of British Columbia.Bange, L A. (1961). A study in the use of rowel letters in alphabetic

consonantal writing. Munich: UNI-Verlag.Browman, C. P. (1980). Rules for demi-syllable synthesis using

lingua, a language interpreter. Proceedings of the ICASSP (IEEE).Browman, C. P., & Goldstein, L. M. (1986). Towards an

Articulatory Phonology. Phonology Yearbook, 3, 219-252.Browman, C. P., & Goldstein, L. M. (1987). Tiers in articulatory

phonology, with some implications for casual speech. HaskinsLaboratories Status Reports on Speech Research, SR-92, 1-30.

Browman, C. P., & Goldstein, L. M. (in press). Gestural structuresand phonological patterns. To appear in I. G. Mattingly & M.Studdert-Kennedy (Eds.), Modularity and the motor theoryof speech perception. Hillsdale, NJ: Lawrence ErlbaumAssociates.

Chadwick, J. (1967). The decipherment of linear B. Cambridge:Cambridge University Press.

Chikanza, J. (1986). Borrowing in ShoM. MA thesis, University ofFlorida.

Chomsky, N., & Halle, M. (1968). The sound pattern of English. NewYork: Harper & Row.

Oements, G. (1985). The geometry of phonological features.Phonology Yearbook, 2, 225-252.

Oements, G., & Keyser, S. J. (1983). CV phonology. Cambridge:MIT Press.

Cohen, G. L. (1982). The origin of the letter Omicron. Kadmos, 21,122-124.

Daniels, P. T. (1986). Toward the linguistic study of writing:Aramaic orthographies. Annual Meeting of the LinguisticSociety of America, New York.

38 Faber

Davis, S. (1987). On the arguments for syllable-internal structure.Research on Speech Perception, Progress Report, 13, 265-275.

Faber, A. (1985). Akkadian evidence for proto-Semitic affricates.Journal ofCuneiform Studies, 37,101-107.

Faber, A. (1987). Phonetic and Phonological Interpretation ofOrthographic Forms. Workshop on Language Change andReconstruction Methodology, LSA Summer Institute.

Faber, A. (to appear). Semitic sibilants: A study in comparatilJelexicography. To be published by Eisenbrauns, Winona Lake, IN.

Firth, J. R. (1957/1948). Sounels and Prosodies. Papers inLinguistics, 1934-1951, 121-138. London: Oxford UniversityPress.

Fudge, E. (1987). Branching structure within the syllable. Journal ofLinguistics, 23,359-377.

Fujimura, O. (1980). Modern methods of investigation in speechproduction. In E. Fischer-J0t'gensen &: N. Thorsen (Eds.),Proceedings of the the Ninth Internatio7lll1 Congress of PhoneticScience (Vol. 3. pp. 117-135). Copenhagen: Institute ofPhonetics.

Fujimura, 0., &: Lovins, J. (1978). Syllables as concatenativephonetic units. In A. Bell &: J. B. Hooper (Eds.), Syllables andsegments (pp. 107-120). Amsterdam: North Holland.

Gelb, I. J. (1963). A study ofwriting (2nd. ed.). Chicago: Universityof Chicago Press.

Gordon, C. H. (1980). Forgotten scripts. New York: Basic Books.Halle, M. (1964). On the bases of phonology. In J. A. Fodor &: J. J.

Katz (Eds.), The structure of language (pp. 324-333). EnglewoodOiffs, NJ: Prentice Hall.

Halle, M., &: Vergnaud, J-R (1980). Three dimensional phonology.Journal ofLinguistic Resetlrch, 1,83-105.

Halliday, M. A. I<. (1981). The origin and early development ofChinese phonological theory. In R. E. Asher &: E. J. A.Henderson (Eds.), Towards a history of phonetics (pp. 123-140).Edinburgh: University of Edinburgh Press.

Harris, Z. S. (1939). DeTJelopment of the Canllllnite dilllects. NewHaven: American Oriental Society.

Harris, Z. S. (1944). Simultaneous components in phonology.Language,20,181-205.

Hayes, B. (1980). A metrical theory of stress rules. Bloomington:Indiana UniVersity Linguistics aub.

Hayward, R J., &: Hassan, M. (1981). The Oromo orthography ofShaikh Balai Sakalo. Bulletin of the School of Oriental and AfricanStudies, 44, 550-566.

Herbert, R K. (1986). Language uniTJeTSa1s, 71Ulrkedness theory, andnatural phonetic processes. Berlin: Mouton de Gruyter.

Hooper, J. B. (1972). The syllable in phonological theory. Language,48, 535-540.

van der Hulst, H., &:. Smith, N. (1982). An overview ofautosegmental and metrical phonology. In H. van der Hulst &:N. Smith (Ed.), The structure of phonological representations, (Vol.1. pp. 1-46). Dordrecht: Foris.

Isserlin, B. S. J. (1983). The antiquity of the Greek alphabet.Kadmos, 22,151-163. .

Jakobson, R, Fant, C. G. M., &: Halle, M. (1969/1952). Preliminariesto speech analysis: The distinctiTJe features and their correlates.Cambridge, MA: MIT Press.

Jeffery, L. H. (1961). The local scripts of archaic Greece. Oxford:aarendon Press.

Kahn, D. (1976). Syllable-based generalizations in Englishphonology. Bloomington: Indiana University linguisticsOub.

Kaufman, S. A. (1985). The pitfalls of typology: On the earlyhistory of. the alphabet. Hebrew Union College Annual, 57,1-15.

Kozhevnikov, V. A., &: Chistovich, L. A. (1966). Speech: Articulationand perception,tr. from the Russian. Washington: U.S.Department of Commerce.

Ladefoged, P. A. (1984). Out of chaos comes order: Physical,biological, and structural patterns in phonetics. In M. P. R vanden Broecke &: A. Cohen (Eds.), Proceedings of the TenthInternational Congress ofPhonetic Sciences (pp. 83-95). DordrechtForis. .

Liberman, 1. Y., &: Shankweiler, D. (1987). Phonology and theproblems of learning to read and write. Memory and LearningDisabilities, Adtnmces in Learning and Behll'CJioral Disorders, Suppl.2,203-224.

liberman, M., &: Prince, A. (1977). On stress and linguisticrhythm. Linguistics, 8,249-336.

Lotz, J. (1972). How language is conveyed by script. In J. F.Kavanagh &: I. G. Mattingly (Eels.), Language by ear and by eye(pp. 117-124). Cambridge, MA: MIT Press.

Mann, V. H. (1986). Phonological awareness: The role of readingexperience. Cognition, 24, 65-92.

Mattingly, I. G. (1972). Reading, the linguistic process, andlinguistic awareness. In J. F. Kavanagh &: 1. G. Mattingly (Eels.),Language by ear and by eye (pp. 133-147). Cambridge, MA: MITPress.

Mattingly, I. G. (1987). Morphological structure and segmentalawareness. Haskins Laboratories Status Report on SpeechResearch,SR-92,107-111.

McCarter, P. I<. (1975). The antiquity of the Greek alphabet and theearly Phoenician scripts. Missoula; MT: Scholars Press.

McCarthy, J. J. (1979). On stress and syllabification. LinguisticInquiry, 10, 443-465.

McCarthy, J. J. (1981). A prosodic theory of nonconcatenativemorphology. Linguistic Inquiry, 12,373-418.

Mohanan, K. P. (1986). The theory of lexial1 phonology. DordrechtReidel.

Morag, S. (1972). The oocaliZlltion systems of Arabic, Hebrew, andAramaic: Their phonetic and phonemic principles. The Hague:Mouton. .

Morais, J., Bertelson, P., Cary, L., &: Alegria, J. (1986). Literacytraining and speech segmentation. Cognition, 24, 45-64.

Naveh, J. (1982). Early history of the alphabet. Jerusalem: MagnesPress.

O'Connor, M. P. (1983). Writing systems, native speaker analysis,and the earliest stagl!ll of Northwest Semitic orthography. In C.L. Myers &: M. P. O'Connor (Eels.), The word of the Lord shall goforth (pp. 439-465). Winona Lake, IN: Eisenbrauns.

Ohman, S. (1979). Introduction. In Bjorn Lindblom &: Sven Ohman(Eds.), Frontiers of speech communication resetlrch (pp. xv-xxv).London: Academic Press.

Packard, D. W. (1974). Minoan linear A. Berkeley: University ofCalifornia Press.

Palmer, F. R. (1973) (1970). Prosodic analysis. In E. C. Fudge (Eel),Phonology (pp. 181-189). London: Penguin Books.

Read, C. (1986). Children's creatiTJe spellings. London: Routledge &:KeganPaul.

Read, c., Yun-Fei, Z., Hong-Yin, N., &: Bao-Qing, D. (1986). Theability to manipulate speech sounds depends on knowingalphabetic writing. Cognition, 24, 31-44.

Sagey, E. C. (1986). The representation offeatures and relations in non­linear phonology. Doctoral dissertation, MIT.

Sampson, G. (1985). Writing systems. Stanford: Stanford UniversityPress.

de Saussure, F. (1959). Course in generll1linguistics. C. Bally &: A.Sechehaye (Eds.), (Wade Baskin, Trans.) New York: McGrawHill.

Schein, B. &: Steriade, D. (1986). On geminates. Linguistic Inquiry,17,691-744.

Selkirk, E. O. (1982). The syllable. In H. van der Hulst &: N. Smith(Eds.), The structure of phonological representations (Vol. II, pp.337-383). Dordrecht: Foris.

Phonemic Segmentation as Epiphenomenon: Evidence from the History of Alphabetic Writing 39

Studdert-Kennedy, M. (1987). The phoneme as a perceptuomotorstructure. In A. Allport, D. MacKay, W. Prinz, & E. Schreerer(Eds.), Language perception and production (pp. 67-84).London: Academic Press.

Swiggers, P. (1983). Some remarks on Gelb's theory of writing.General Linguistics, 23, 198-20l.

Swiggers, P. (1984). On the nature of West-Semitic writingsystems. Aula Orientalis, 2, 149-15l.

Trubetzkoy, N. S. (1949) [1939]. Prindpes de phonologie. (JeanCantineau Trans.). Paris: KlincksieCk.

Whitney, W. D. (1889). 5IInskrit grammar. Cambridge, MA:Harvard University Press.

FOOTNOTES"To appear in P. Downing, S. Lima, & M. Noonan (Eds.),

Language and Literacy (Papers from the MilwaukeeSymposium on Language and Unguistics).

II will use the neutral term segment in this paper to refer to anelement representing a steady state, regardless of assumptionsregarding its formal status (e.g., phoneme vs. allophone) in alinguistic model. By segmentation and segmentation ability Imean an exhaustive division of the speech stream into asequence of segments; ability to recognize some segments doesnot imply segmentation ability in this, restricted, sense.

21t has been argued on a utilitarian basis that incorporation ofphonological segments in linguistic analyses is validated bythe light shed by such analyses on complex phonologicalpatterns (Anderson, 1974: 6; Halle, 1964: 325). But sucharguments lose force in the absence of competing non-segmentalanalyses of comparable phenomena.

3The segments identifiable in an acoustic waveform, to theextent that their boundaries are well-defined, are notisomorphic with the segments of phonological analysis or ofalphabetic writing. .

4See the overview in van der Hulst and Smith (1982).5See Fudge (1987) for an up-to-date review of evidencesupporting a hierarchically structured syllable. O. Davis(1987).

6For discussion of prenasalized consonants, see Anderson (1976),Browman and Goldstein (1986), and Herbert (1986). AsHerbert notes, similar considerations apply also to affricatesand diphthongs.

7The older Japanese children could, in fact, perform somesegment manipulation tasks, despite non-exposure to Romaji(Japanese written in the Western alphabet). In her attempt toaccount for this apparent anomaly, Mann does not consider thepossibility that a grid-like presentation of the kana syllablesigns, in which signs for syllables containing the same vowelare listed in the same column, might provide a passiveexposure to segmentation.

8This is not to say that failure after appropriate instruction toacquire segmentation ability might not reflect an underlyingdeficit, as described by Liberman and Shankweiler (1987: 213).

9A note on terminology is in order. Early Near Eastern textualmaterial is available in four different segmental scripts:Phoenician, Aramaic, South Arabian, and Ugaritic..Theserepresent texts in four distinct Semitic languages, but thehistorical relationships among the scripts (for which seeNaveh 1982) do not mirror those among the languages. TheAramaic and Phoenician scripts diverged quite early (c. 1000BC), so it is likely that they had a common ancestor,unfortunately unattested. The South Arabian script, used forinscriptions in now dead languages spoken on the Arabianpeninsula, also developed from this common ancestor. TheUgaritic orthography, in contrast, is a segmental cuneiform.

Its symbols do not resemble those of the other segmentalorthographies, although its citation order was segmentallyisomorphic with that of the Phoenician and Aramaicinventories. This suggests a still earlier orthographiCtradition, which developed in two distinct ways. Strictlyspeaking, it is appropriate to use the term Canaanite for thescript antecedent to the Aramaic and Phoenician scripts, andWest Semitic for the segmental tradition antecedent both tothese scripts and to the Ugaritic cuneiform, but thisdistinction is not maintained in the literature; Naveh (1982)consistently uses the term West Semitic where I would useCanaanite. In what follows, I will use the terms Canaaniteand West Semitic for unattested segmental scripts isomorphic(modulo segment inventories) to the attested Phoenician,Aramaic, and South Arabian scripts.

1000mpson treats the Korean Han-gul as an orthography thatcodes (distinctive) features; my omission of that category herereflects not disagreement with his analysis, but rather theirrelevance of this category for present purposes. It should benoted that analytic recognition of the featural basis of Han­gul does not imply a claim that Korean speakers make use ofthis sub-segmental information in decoding the script.

llSee Browman and Goldstein (1987: 11) for a discussion of thefigure-ground metaphor as a structural representation forqualitative differences between consonant and vowelarticulation within the model of articulatory phonology.

12Non-English fonts in this paper are SuperGreek,SuperHebrew, Sabaean, Ethiopic, and Zakkur (based on a lateninth century Be Aramaic inscription), all copyright byUnguists' Software, and LaserPerfect Phonetique, copyrightby Neoscribe International.

l3The non-linear nature of the Ethiopian script is masked by thesubset of symbols presented by Sampson. In all cases, thesymbols he shows have right-side vowel marking, preservinglinearity. Modification for -" and -a, however, often showsleft-side deviation from the neutral shapes. Note also thatthis modification may involve removal of material, as in"<Ia>, or displacement of the graph, creating the illusion of

left-side deletion, as in III <t'a>. Sampson does not cli<lcu<;s

Devanagari, but conceptually it is identical to the Ethiopianscript; for details, see Whitney (1889: 9-14); d. Mohanan's(1986: 195) description of the Malayalam orthography as "asyllabary that is 'phonemic'."

141 have no information as to whether ambisyllabic consonants,consonants which constitute at one and the same time the codaof one syllable and the onset of the follOWing one, occur inlanguages which are written syllabically.

15But (pace Firth, 1957: 126; Gelb, 1962) they are not syllabarieseither. See below for further discussion.

16See Morag (1972) for the origins and structure of these diacriticvocalization systems.

17While my definition of alphabet might not accord withpopular, non-scientific usage (it is common to speak of theHebrew or the Arabic alphabet), it is, I think, justified.Popular usage, it should be noted, does not provide any termbut alphabet for the writing system of a literate culture,except, perhaps, for "picture writing" and "hieroglyphics" inreference to the Chinese and Egyptian systems.

18Mohanan (1986: 197), in a very different theoretical context,offers a compatible definition of an unmarked writing systemas one in which each orthographic distinction corresponds toexactly one phonemic distinction in the lexical alphabet; thelexical alphabet is the set of symbols in underlyingphonological representation.

19Labeling the West Semitic scripts syllabaries may involve anon-standard notion of syllable. If the orthographic form

40 Faber

<q-b-I-t>, representing IqVbbVltVI, is interpreted as asequence of four syllable signs, then <b> must be taken asrepresenting l.bbV'; and <1> as 1.1./, implying a counter­intuitive syllabification IqV.bbV.l.tVI. This is reminiscent ofthe articulatory syllable of Kozhevnikov and Chistovich(1966: 129ff), which takes as a production unit a vowel and allconsonants intervening between it and the preceding vowel (orthe beginning of the utterance), differing only in treating 11/as a discrete syllable.Hayward and Hassan (1981) describe a curious syllabicallylinear orthography which was developed for Oromo (Galla),a Cushitic language of Ethiopia, in the 1950's by ShaikhBakri Sakalo but which never replaced the Amharic-basedorthography for that language. The Shaikh's system, unlikethe general Ethiopian system, explicitly differentiates singlefrom geminate consonants, writing bllbblz <ba-bba>. Mohanan(1986) notes similar syllabifications by Malayalam speakers,both literate and nonliterate.

20The relative usage of orthographic types was computed on thebasis of population figures and literacy rates in the currentWorld Almanac: for example, given the population of Burma(most recent estimate 38,000,000 people) and a reportedliteracy rate of 66%, 25,080,000 Burmese use a segmentallycoded, syllabically linear orthography. The literacy rates inthe Almanac, especially for developed countries, aresuspiciously high; the United States, for example, is reportedto have 99% literacy.

21 For a description of Linear B, a syllabic script representingMycenean Greek (c. 1500 BO, see Chadwick (1967:75ff). Thetypological properties of the related script, Linear A, cannotbe ascertained, since it has not been determined whatlanguage it represents (Packard, 1974; d. Gordon, 1982: 43ff).

22 See extensive discussion in Allen (1974: 33ff).23 I ignore here Z uta, an affricate Idzl in early Greek (despite

its later OassicaI value of Izdl [Allen 1974: 55]). This letterpreserves the affricate value of its Semitic source ""1- zain, avalue argued for in Faber (1985, MS). Likewise, use of Semitic* samekh for Greek S Iksl (ksi) reflects the originalaffricate value Itsl of the Phoenician symbol; the directlyattested value for cognate sounds in related languages isexclusively lsi, but there is indirect evidence, summarized inFaber (MS), for an earlier affricate value.

24 The earliest attested West Semitic representation of vowels(the Akkadian syllabically coded system of course allowedrepresentation of vowel sounds, although not in isolation) is ineighth century BC Aramaic texts in which some final vowels

and etymologically long vowels are represented ([u]-qualityvowels by Y /w/, £i] quality vowels by -1.. fyI, and [a] qualityvowels by ~ lor '1 ~ /hf). See Bange (1961) for particulars, butwith a different interpretation.

25Comparable reinterpretations are extremely common in lexicalborrowings; borrowed words are regularly restructured inaccord with the host phonological system. So, for example,initial sC- clusters in English loans into Shona, a Bantulanguage of Zimbabwe, are regularly nativized with lei;spoon becomes Cipunu (Chikanza, 1986: 20).

26The differing values for A alpha and 0 o(micron) in Greekmay constitute previously unrecognized evidence for thebacking effect of l'i.l on an adjacent low vowel in Canaanite;comparable effects are commonly observed in LevantineArabic dialects, even though in some Maghrebi dialects laltends toward I £1 in such a context. It is unlikely that thevalue [a] for fayn reflects the Phoenician change of·' to [:>] instressed, originally open syllables, which Harris (1939: 61)places earlier than the Greek adoption of the Phoenicianorthography (on the basis of the letter name iota); ·a in theCanaanite letter name is in a historically closed syllable andthus would not have been subject to this change. Cohen (1982)suggests that the semantic equation of Canaanite fayn andGreek opthalmos, both 'eye', might have led to the Greekvalue for 0 .

27The remaining three symbols in Figure 5 Brepresent sounds notfound (or at least not written) in Phoenician. While Navehdoes not note this, <l> phi, X khi, and in their archaic forms(not those forms given in Figure 5) show a striking resemblanceto early South Arabian characters a> and 'f, conventionallyinterpreted as representing IwI and Ibl (IvI is a conceivablealternate reading of the former). <l>p hi presumablyrepresented I rf'/; Allen (1974: 21) notes that the "classical"value was clearly Iph/, but also (p. 23) the possibility of adialectal pronunciation as IfI as early as the sixth centuryBe. X khi and 'It psi (more similar in their archaic forms)represented Iksl and Ikhl in the Western scripts from whichthe Latin script descended and Ikhl and Ipsl in the easternand classical scripts. At least some of the interchange between<w> and <y> symbols in Canaanite, South Arabian, andGreek reflects the Proto-Canaanite shift of ·w->y wordinitially; the only instances of word initial Iwl in OassicalHebrew and Phoenician are in the morpheme w- "and" and inthe letter name wau, all other instances of initial ·w havingchanged to Iy/. Thus, Hebrew yel elf "child" is cognate toArabic waldo