Defining Morphological Isoglosses: The Broken Plural and Semitic Sub-classification (1) Robert R. Ratcliffe Tokyo Gaikokugo Daigaku, [Transcription Table ƒˆ= glottal stop `= voiced pharyngeal fricative (Arabic `ain) H= voiceless pharyngeal fricative (Arabic Ha) G= voiced velar fricative (Arabic Gain) q= voiceless uvular stop K= ejective velar stop S= voiceless pharyngealized sibilant (Arabic Sad) Z= voiced pharyngealized sibilant (Arabic Za) T= voiceless pharyngealized dental stop (Arabic Ta) D= voiced pharyngealized dental stop (Arabic Dad) $= voiceless palatal sibilant (English 'sh') @= voiceless dental fricative (English 'th' in 'thin') &= voiced dental fricative (English 'th' in 'then') vowels: i u e * o E O a note that *= short central vowel 'schwa'] 1.Morphological Isoglosses and Semitic Sub-classification The question of Semitic sub-classification, particularly the question of the place of Arabic within Semitic, has become controversial since Hetzron (1974, 1975, 1976) argued, on the basis of isoglosses in the verbal morphology, that Arabic should be classed with Northwest Semitic in a Central Semitic sub-branch.(2) Earlier researchers had generally accepted Brockelmann's (1908) classification of the Semitic languages into three subgroups ƒ~ East, Northwest and Southwest, the last including Arabic with Ethiopian Semitic and the languages of South Arabia. A minority opinion (represented by Cantineau (1932), Leslau (1959) W.W.Muller (1964), and Diakonoff 1965, among others) favored a division into four branches, in which Arabic (SW Semitic) is a separate both from NW Semitic and the "Southeast" Semitic languages of South Arabia and Ethiopia. Hetzron's classification of Arabic with NW Semitic has been adopted by a number of Semiticists (Goldenberg 1977, Faber 1980, Voigt 1987, Huehnergard 1991, Rodgers 1991), but has been challenged by others (notably Werner Diem 1980 and Zaborski 1991) on the grounds that it gives priority to one set of morphological isoglosses while ignoring others ƒ~ƒ~ notably the noun plural morphologyƒ~- which earlier researchers had deemed significant. Hetzron (1976:102) dismisses the "broken" plural system as a shared retention: "The traditional unit of South Semitic is untenableƒ~ƒ~ even though Arabic does share some features with South Arabian and Ethiopic (the articulation f for northern p, the existence of 'broken plurals'-- definitely a common retention.)" Rabin (1955) hinting at a similar classification twenty years earlier had dismissed it as an innovation: "Some of the results of Rabin as well as a recent study by al-Yasin suggest connections of Arabic with North-West Semitic. Arabic has generally been considered to form a separate branch of the Semitic languages-- called South Semitic or SW Semitic. Among the features distinguishing this branch are... the broken plurals. ...The broken plurals are almost certain to be a late development which has little value as a genetic criterion." >From a methodological point of view Hetzron is correct, and Rabin is mistaken. If the system is innovative, it does indeed have great value as genetic criterion. On the other hand, Rabin's statement does appear to be an accurate reflection of scholarly opinion up to that time. Whereas Hetzron's assertion to the contrary would seem to require more substantial support. Diem (1980:76) sums up the problem when he says: "Die situation ist nicht ganz so einfach. Die Existenz einfacher Formen innerer Plurale im Nordwestsemitischen beweist zunachst nur, dass, was schon lang bekannt ist, die Erscheinung des inneren Plurals dem Ursemitischen nicht Fremd war, aber nicht, dass das Ursemitisch ein derartiges reiches System wie das Sudostsemitisches und vor allem das Arabische kannte ...(3) After evaluating the isoglosses which argue in favor of both classifications, Diem concludes (p.82): "Die Zentrale Frage ist die Interpretation der inneren Plurale des Arabischen und Sudostsemitischen. Von ihr Beurteilung hangt die Entscheidung ab."(4) In previous research (Ratcliffe 1992) we have attempted a thorough comparison and reconstruction of the internal noun plural in Semitic and Afroasiatic. In what follows we intend to take up the research project proposed by Diem, to look more thoroughly at the internal plural in Semitic languages, and to bring this evidence to bear on the classification debate. 2. Methodological Issues There are two principal reasons why the noun plural morphology has disappeared from recent classification debates. The first is that the broken plural system has always been a poorly understood area of Semitic morphology, both synchronically and diachronically. The second is that there is no clear consensus in historical linguistics generally about the appropriate method for comparative work in morphology. To address the first issue first, the broken plural problem is one which shows up a fundamental inadequacy of the traditional pedagogical 'rootƒ† and 'patternƒ† grammar as a tool of synchronic or diachronic analysis. The fact that the traditional grammars offer very little analysis of relationships between forms or patterns has led to a mistaken assumption that such relationships can not be categorized or do not exist. Thus, with regard to the issue at hand, a traditional assumption has been that the association of singulars to broken plurals is largely random or unpredictable and that there is no formal relationship between singulars and plurals. The first assumption is mistaken, as statistical studies such as those of Murtonen (1964) and Levy (1971) have shown, while the second assumption largely depends on the theoretical or analytical framework adopted. Recent theoretical work under the rubric of autosegmental or prosodic morphology has greatly enriched the formal apparatus available for analysis of morphological systems like those found in Arabic and has consequently opened the way for a clearer understanding of the formal basis of the relationship between singulars and plurals (cf. McCarthy 1983, McCarthy & Prince 1990, Hammond 1988, Ratcliffe 1990, 1992, 1996). A widespread view in historical linguistics is that historical reconstruction in morphology must be based on comparison of phonologically cognate forms or formatives. I believe that this view is actually quite mistaken. Comparison in morphology can and should be approached in a way entirely analogous to phonological comparison. That is, we compare elements in the systems of different languages on the basis of distribution. The observation that both English and German have a voiced dental/alveolar stop 'd' does not reveal anything about the history of the languages or of their relationship to each other. On the other hand the fact that English 'd' is distributed in a way very similar to formally different German 't' (first C of words meaning day, daughter, etc.), gives us evidence that the languages are related and that a sound change has taken place in one language or both. To take a morphological example, if we compare the highly allomorphic noun case systems of Greek and Latin, we could do so on the basis of formal similarities. This would lead us to speculate about whether the accusative plural meaning of the suffix -os (Latin), for example, or the genitive or nominative singular meaning (Greek) was primary and why and how the ending spread from one meaning to another. This would clearly be a bizarre and fruitless procedure. The proper approach would be to compare those case endings which share the same distribution defined both in terms of function and phonologically-defined stem (or declination) class. We would then see that Greek -os corresponds to Latin -us (in the role of nominative singular for one class of nouns) and to Latin -is (in the role of genitive singular for another class). This discrepancy is evidence that a formal change of a definable type (phonological merger in Greek, conditional split in Latin, or analogical reshaping) has in fact taken place. To put it another way comparing phonologically similar or cognate forms can only provide evidence that a shift in distribution or function (=meaning) MAY have taken place. Comparison on the basis of form, can not, however, provide evidence of analogy or of sound changes which are not already known. By contrast, comparison on the basis of distribution and function can not provide evidence of shift in distribution or function, but is the only way to discover evidence of formal changes due to sound change or analogy. We have to control one of the variables (form or function/ distribution) in order to discover evidence of change in the other. The approach to the diachronic study of the broken plural system in Semitic languages has almost always been based on a comparison of formal rather than distributional similarities. Where reconstruction was attempted such an approach necessarily led to an explanation in terms of semantic shift. Hence the theory that the ƒformsƒ‚ of the broken plurals were originally ƒcollectivesƒ‚ or ƒabstractsƒ‚ was widely accepted (e.g. by Moscati, Spitaler, Ullendorf, and von Soden 1964) in spite of the facts that there is no Semitic language in which one finds internal and external plurals contrasting in a productive way (as true plurals vs. collectives for example), and that the vast majority of the plural ƒformsƒ‚ do not appear in any function other than the plural in any Semitic language.(5) In the domain of classification, the form-based approach leads to irrelevant comparisons of phonologically similar but non-corresponding forms and to obscuration of the evidence for innovations which may be essential for sub-classification. In order to identify isoglosses in the noun plural systems of the Semitic languages and in order to uncover evidence of innovatory changes relevant for classification, it is necessary to consider the relative distribution of the various plural forms in regard to their respective singular types. 3. The evidence In the case at hand we should distinguish two questions. 1) The question of the existence of an internal plural beside an external one: At what historical depth can we reconstruct the existence of an internal plural, and why should both types exist and 2) The question of the rich allomorphy within the internal plural of Arabic (and possibly other Semitic or Afroasiatic languages?): At what depth can we reconstruct a formally complex internal plural system comparable to that of Classical Arabic? 3.1 Internal vs. external plurals On the basis of the relative distribution of internal and external plurals in Semitic languages it should be possible to decide what the original distribution of internal and external plurals was (that is the original basis for the distinction) in Proto-Semitic and what type of shift or analogic spreading has affected the distributional range of the two types in the attested languages. Chart (1) summarizes the facts of distribution of internal and external plurals through singular stem classes (see below) in the Semitic languages. (1) Distribution of internal and external plurals in Semitic(6) sg. form Ar. OSA Ge. MSA Ti. He. Arm. Ak. CVCC i i i,e i i i+e i+e e CVCCat i i e i i i+e i+e e CVCCVC i i i i i e e e CVVCVC i,e i,e i,e i,e i,e e e e CVCVVC i,e i,e i,e i,e i,e e e e On the basis of these distributional isoglosses alone, the Semitic languages fall clearly into three groups. In the Eastern group the external plural is general. In Hebrew and Aramaic (NW Semitic) underived triconsonantal nouns have plural forms with a plural suffix which also show evidence of an intercalated vowel between the second and third consonants. Derived nouns and nouns with more than three consonants in these languages have external plurals only. The third group (SW Semitic) is characterized by an obligatory internal plural for both quadriconsonantal nouns and triconsonantals with no long vowel and by the existence of optional internal plurals for triconsonantals with a long vowel. The relative distribution of internal and external plurals for long-vowel singulars in these languages (as most clearly represented in Arabic and Tigre) is that productively derived nouns (active participles, verbal nouns and adjectives, diminutives in Arabic) obligatorily take an external plural, and that basic nouns which have these patterns (including lexicalized derived nouns-- nouns which have acquired a meaning not predictable from their source) obligatorily take an internal plural. Thus the relative distribution of internal and external plurals in Proto-Semitic seems quite clear. In all languages in which internal plural morphology is productive, it is more or less obligatory for base, underived triradical (and biradical) nouns (CVCC, CVCVC, CVC(V)Cat) with no long vowel, and optional or non-existent elsewhere. The external plural is obligatory for derived nouns. The most plausible conclusion is that the internal plural has spread by analogy to quadriliteral and lexicalized derived nouns in some languages (i.e. SW Semitic), while in others the external plural has spread by analogy to the base noun, either replacing the internal plural entirely (E. Semitic), or being pleonastically superimposed upon internal plural forms (NW Semitic). 3.2 Allomorphy in the form of the internal plural 3.2.1 The plural system of Arabic With regard to the problem of allomorphy, again the question is what is the relative distribution of the internal plural allomorphs through their respective singulars. A basis for the discussion of this problem has been laid by the statistical analyses of Murtonen (1964) (based on Lane's dictionary of Classical Arabic ) and Levy (1971) (based on Wehr's 1960 dictionary of Modern Written Arabic).(7) What these studies show is that the principal factor determining the form of the plural in Arabic is the form of the singular. The system IS highly allomorphic: For a given singular pattern, two different plural forms may be equally frequent, and there may be no way to predict which of the two a particular singular will take. For some singulars as many as three further statistically minor patterns are also possible. And there are a handful of nouns whose plural is completely anomalous. The range of allomorphy, though, is in general from two to five. The figure of 27 to 30 'formsƒ† of the broken plural (given in traditional grammars like those of Wright) is a meaningless number because no singular has the possibility of pluralizing according to all of them. Starting from these statistical studies, which still take semantic features into account, we have shown (Ratcliffe 1992) that internal plural allomorphy (like case suffix allomorphy in Latin) is a function of stem class, where stem class is defined in terms of three criteria-- syllable structure, (stem-internal) vowel quality, and presence or absence of a suffix -at. This gives us six groups of singular/ plural pairs as represented in Table (2a).(8) Forms in parentheses here and in the following charts indicate minor patterns (fewer than 11% but more than 5% of all of the plurals of a given singular, taking into account in the case of Arabic both Levyƒ†s and Murtonen's data). Underlining here and in the following charts indicates a geminate consonant. 2a) The Arabic Broken Plural System 1) Ci/uCC >> ƒˆaCCaaC, CuCuuC CaCC >> CuCuuC, ƒˆaCCaaC, CiCaaC, (ƒˆaCCuC) CaCaC >> ƒˆaCCaaC 2) Ci/uCCat >> Ci/uCaC CaCCat >> CaCaCaat, CiCaaC, Ci/uCaC CVCVCat >> CaCaCaat, CiCaaC 3) CVCCV(V)C >> CaCaaCi(i)C 4) CVVCVC(at) >> CawaaCiC CVCVVC(at) >> CaCaaƒˆiC 5) CaaCiC >> CuCCaaC, CaCaCat, CuCCaC 6) CVCaaC >> ƒˆaCCiCat, CuCuC CaCuuC >> CuCuC, ƒˆaCCiCat CaCiiC (n.) >> CuCaCaaƒˆ, ƒˆaCCiCaaƒˆ; CuCuC CaCiiC (adj.) >> CiCaaC Table 2b gives examples of each type. This chart includes one example of an unmarked collective noun (labelled 'C') and a corresponding singulative derived by suffixation of -at. Although not logically part of the broken plural system, the singulative/collective pattern is formally similar to the type 2 feminine plural. As the two classes have merged or partially merged in some Semitic languages, comparative treatment requires that we consider them together. 2b) The Arabic Broken Plural System, examples 1) ƒˆu&nun >> ƒˆaƒˆ&aanun "ear" qirdun >> quruudun "monkey" `aynun >> `uyuunun, ƒˆa`yun "eye" waznun >> ƒˆawzaanun "measure" kalbun >> kilaabun "dog" farasun >> ƒˆafraasun "horse" 2) Gurfatun >> Gurafun "room" jabhatun >> jabahaatun "face" farxatun >> firaaxun "hen" C) naHlatun << naHlun "a bee" << "swarm of bees" 3) `aqrabun >> `aqaaribu "scorpion" 4) Saa`iqatun >> Sawaa`iqu "thunderbolt" jaziiratun >> jazaaƒˆiru "island" 5) Taalibun >> Tullaabun, Talabatun "student" $aamixun >> $ummaxun "lofty, proud" 6) `amuudun >> `umudun, ƒˆa`midatun "column" baxiilun >> buxalaaƒˆu "miser" waliiyun >> ƒˆawliyaaƒˆu "governor" kabiirun >> kibaarun "big" 3.2.2 The plural systems of the other southern languages The questions for comparative study are the following: (1) Do other Semitic languages show allomorphic variation in the form of the internal plural? (2) If so, is this allomorphic variation a function of stem classes defined in the same way as those of Arabic? (3) And if so, how much formal similarity is there among the distributionally corresponding forms of the internal plural? The answer to the first question is that only the ancient and modern languages of South Arabia and North Ethiopia show a range of allomorphic variation in the form of the internal plural comparable to that of Arabic. In order to answer the second question (and if necessary the third), we have to have the same sort of statistical distributional analysis as has been done for Arabic. For Old or Epigraphic South Arabian (OSA), specifically Sabaic and for two Modern South Arabian (MSA) languages I have recently done such an analysis based on counting singular-plural pairs in the available dictionaries (Biella 1982 and Beeston, Ghul, MŸller, Ryckmans 1982 for Sabaic, Johnstone 1977 for Harsusi and Johnstone 1981 for Jibbali). For Ge`ez I rely on Leslauƒ†s 1991 dictionary but only counted the forms of the base CVCC and CVCVC nouns, since otherwise plurals are predictable according to the available reference grammars (i.e. Dillmann 1903 and Lambdin 1978). For Tigre I rely on Raz's (1983) grammar, supplemented by count of base noun plurals based on the information given in Palmer (1962). There is some latitude for interpretation in MSA because of complicated vowel correspondences (cf. Johnstone 1975 for a full discussion), and in OSA because of the absence of vowel-notation. The answer to the second question ("Is allomorphy a function of stem class?") turns out to be yes as indicated in tables 3-7, and following discussion. The answer to the third question (regarding formal similarity of corresponding classes) is discussed in section 4. (3a) The Ge`ez system 1) C*CC >> ƒˆaCCaaC, C*CaC, sfp. CaCC >> ƒˆaCCaaC, ƒˆaCC*Ct, sfp., (ƒˆaCC*C,ƒˆaCCuuC) CaCaC >> ƒˆaCCaaC, sfp. ƒˆaCCuuC 2) CVCVCt >> sfp. [=CVCVCtaat, CVCVCaat] 3) CVCCVC >> CaCaaC*Ct, CaCaaC*C CVCCVCt >> CaCaaC*C 4) CVCVVC >> CaCaaw*Ct, CaCaaw*C, CaCaay*C 5) C*CaaCii >> CaCaCt 6) CaCiiC, CaCaaC >> CaCaCt CVCaaC(n) >> sfp. (3b) Ge`ez examples: 1) ƒˆ*zn >> ƒˆ*zan "ear" `ayn >> ƒˆa`y*nt "eye" wagr >> ƒˆawg*r "hill" faras >> ƒˆafraas "horse" hagar >> ƒˆahguur "town" 2) 0 3) `aqraab >> `aqaar*bt "scorpion" 4) w*Hiiz >> waHaay*zt "river" xaTiiƒˆt >> xaTaaw*ƒˆ, xaTaay*ƒˆ "sin" 5) saraaqi >> saraqt "thief" 6) `abiiy,`abaay >> `abayt "great" Notes: The abreviation 'sfp.' indicates sound feminine plural. The Ge`ez short central vowel '*' corresponds regularly with the Arabic short high vowels 'i' and 'u'. The Ge`ez long vowels and short 'a' correspond with the similar vowels in Arabic. Thus although the Ge. ƒˆaCC*Ct could be interpreted as cognate with Arabic ƒˆaCCiCat, distributional analysis shows that is not correct, and that Ge. ƒˆaCC*Ct must reflect historical ƒˆaCCuC-t. Note too that the the Ge.C*CaC, which is a group (1) masculine plural does not correspond with the phonologically cognate Ar. group (2) feminine Ci/uCaC plural. (4a) The OSA (Sabaic) system 1) CCC >> ƒˆCCC, CCC, CCCt 2) CCCt >> CCC 3) CCCC >> CCCCt, CCCC CCCCt >> CCCC 4) CCC >> CCCwC(t), CCCyC(t) CCC >> CwCC 5) CCC >> CCCt, CCC 6) CCC >> ƒˆCCCw, CCCw CCC >> ƒˆCCCt (4b) OSA examples: 1) ƒˆ&n >> ƒˆ&n "ear" hgr >> ƒˆhgr "town" ƒˆb >> ƒˆƒˆbw, ƒˆbh "father" 2) Zlt >> Zll "roofed tomb" Hrt >> Hyr "encampment" 3) mHfd >> mHfdt "tower" tb$rt >> tb$r "announcement" 4) xTƒˆ >> xTyƒˆ "sin" xrf >> xrwf, xryf(t),(ƒˆxrft) "year" ƒˆmr >> ƒˆwmr "sign" 5) rkb >> rkbt, rkb "rider" 6) xrf >> ƒˆxrft (xrwf, xryf(t)) "year" rbb >> ƒˆrbbw "person under protection" Hs'r >> Hs'rw "steersman" (Hofner) "poorer class" (Beeston) Notes: In the absence of vowel notation, the plural patterns of OSA must be interpreted in light of Ge`ez and Arabic. In the context of a discussion of sub-classification, this approach is admittedly somewhat circular. Basically we assume that it is a priori plausible that OSA, Ge`ez and Arabic all had a similar plural system and that the burden of proof falls on those who wish to develop a contrary hypothesis. Thus we assume that where an OSA plural form written only with consonant signs can be interpreted as reflecting a plural pattern found in the same distribution in Ge`ez or Arabic, it should be so interpreted. We assume that the hypothesis is challenged only by the existence of OSA forms which can not possibly be interpreted as cognate with corresponding patterns in Arabic or Ge`ez. (In fact there are no examples of such forms.) Group (1): OSA ƒˆCCC most plausibly reflects Ge. and Ar. ƒˆaCCaaC, but could also reflect Ge. and Ar. ƒˆaCCuC or Ge. (and Tigre) ƒˆaCCuuC, compare Ge. hagr >> ƒˆahguur "town" with the second example in (4b). That OSA also had a group 1 plural without alif-prefix like Ar. CvCaaC or Ge. CvCaC is most apparent in plurals of biradicals which show a default third consonant (here 'h') like ƒˆb >> ƒˆabh, cf. Ge. ƒˆab >> ƒˆabaw. Group (2): That OSA , like Arabic, Tigre and MSA, but unlike Ge`ez, had an internal plural for feminine base nouns is apparent from alternations involving nouns with final geminate or medial-glide radicals like those in the examples above. Thus Zlt >> Zll is plausibly reconstructed as *Zvllat >> *Zvlal, and Hrt >> Hyr as *Hairat or *Heerat >> *Hvyar Group (4): Comparison of the OSA singulars which show these plurals with cognates in other languages combined with semantic analysis of the words involved shows that most of the singulars must be of the CvCvvC type (for the examples above cf. Ar. xariifun "autumn," Ge. xariif "year," Ar. xaTiiƒˆat >> xaTaayaa Ge. xaTiiƒˆt >> xaTaayeƒˆ, xaTaaweƒˆ "sin" ). Thus the most cautious assumption is that plurals CCyC(t) and CCwC(t) reflect CaCaaGvC(vt), which is the default plural for singulars of this type in Arabic, Ge`ez, and Tigre. An alternative interpretation of these forms as reflecting *CvCiiC, or *CvCuuC plurals is suggested by Hofner (1943) on the basis of similarity of form without regard for distributional correspondences. Ignoring the philological problem of whether OSA graphs 'y' and 'w' can be interpreted as reflecting long vowels, the difficulty for this interpretation is that CvCiiC does not appear as a productive plural in any Semitic language, and CuCuuC appears only in Arabic, but not as a plural for CvCvvC nouns. Hofner's analysis would thus force us to assume that the OSA system was closer to Arabic than to Ge`ez, but basically quite different from either. Group (5): For CCC nouns which are semantically active participles, plurals with suffix 't', presumably cognate with Ar. CaCaCat or Ge. CaCaCt, are attested. Apparent plurals of participles with no overt mark-- CCC, indicate that an alternative internal plural may have existed in this class, possibly cognate with Arabic CuCCa(a)C. Group (6): A handful of plurals with final 'w' are the only plausible cognates throughout southern Semitic to the Arabic plurals CuCaCaaƒˆu, ƒˆaCCiCaaƒˆu.(9) Again based on semantics and comparison of cognates, the OSA singulars which take these plurals can be analyzed as having the pattern CaCiiC. (The cognates to xrf have been noted. For rbb, cf. Ar. rabiibun >> ƒˆaribbaaƒˆu, "step-child.") (5a) The Tigre System 1) C*C*C >> ƒˆaCCaaC, ƒˆaCCuC, (C*CaC) CaC*C >> ƒˆaCaCC*C(at), ƒˆaCC*C(t), ƒˆaCCaaC,ƒˆaCCuC CaCaC >> ƒˆaCCuC, ƒˆaCaCC*C(at), ƒˆaCCaaC; 2) C*CCat >> C*CaC CaCCat >> C*CaC, C*CaaC CaCaCat >> C*CaaC 3) CVCCVC >> CaCaaC*C, CaCaCC*C CVCCVVC >> CaCaaCiC, CaCaaCuC 4) CVVCVC >> CawaCC*C, CawaaC*C CVCVVC >> CaCaay*C 5) CaaC*C >> sfp. CaaCCay >> CaaCCat 6) C*CaaC >> ƒˆaC*CCat, (C*C*C) (5b) Tigre, examples: 1) m*d*r >> ƒˆamdaar,ƒˆamdaaraat "land" war*K >> ƒˆawarr*K "silver" Kab*r >> ƒˆaKb*r(t) "tomb" daKal >> ƒˆadkul "mast" 2) H*frat >> H*far "hole" kar$at >> k*ra$ "belly" Karbat >> K*raab "water skin" C) n*hbat "bee" n*h*b "bees" 3) janjar >> janaaj*r "chain" d*rf*n >> daraff*n "ram" 4) mezan >> mawazz*n "scales" w*Hiz >> w*Haay*z "river" 5) kaadmaay >> kaadmat "worker" 6) s*gaad >> ƒˆas*gdat "neck" Notes: Tigre '*' corresponds to Arabic short 'u' and 'i'. Tigre 'i' and 'u' correspond to Arabic long 'ii' and 'uu'. Tigre has ƒˆaC*CCat corresponding to Ar. group (6) ƒˆaCCiCat and ƒˆaCC*C(t) (where the 't' appears only in juncture) corresponding toe Ge. group (1) ƒˆaCC*Ct. An interesting development is the compensatory gemination (geminates indicated by underlining) which optionally substitutes for original vowel length CVV >> CVC in some plurals of groups 1, 3, and 4. (6a) The Harsusi system 1) *CVCC >> CeCeeC, CeCooC, CeCewweC, eCCooC 2) CVCCe(e)t >> CeCeCten, CeCaCten CVCCoot >> CeCeeC, CeCaC CVCeCe(e)t >> CeCeeC 3) CVCCV(V)C >> CeCooCeC, CeCooCeCet CVCCiiC >> CVCCooC 4) CVVCVC >> CewVVCeC, CewVCCeC CVCVVC >> CeCaayeC, CeCayyeC, CeCawweC 5) CeeCeC >> CeCeCet (m), sfp. 6) CeCiiC >> CeCooC CeCiiC,CeCeCet >> CeCeeC CeCeC, CeCeCet >> CeCeeC (6b) Harsusi examples: 1) bark >> bereek "knee" kawb,kob >> keloob "wolf" Ham >> eHmoo "lower belly" Harf >> Herewwef "gold coin" Hoobel >> Hebewwel "hobbling rope" 2) $ebdeet >> $ebadten "liver" kezebet >> kezeeb "coconut" C) nexeleet >> nexel "date palm" 3) menxel >> menooxel "sieve" mesHeToot >> mesaHHeT "slaughtered beast" kebkiib >> kebkoob "star" 4) Gooreb >> Gewooreb "base of the neck" Gaaber >> Gewabber "pregnant she camel" meGiig >> meGaayeg "camel with young" &eraa >> &erawwet "forearm" 5) keehen >> kehenet "clever" 6) refiif >> refoof "horse's mane" KeSiir, KeSeret >> KeSeer "short" xefef, xefefet >> xefeef "light" Notes: In Harsusi original long 'aa' usually appears as 'oo' or 'ee'. Original short stressed 'a' may also appear as 'oo,' as in Hoobel, cf. Ar. Habl. The sequence 'ewwe' may reflect original long 'uu' (cf. Johnstone 1975). (7a) The Jibbali system 1) CVCC >> vCCO/E/uC, CVCƒ\/E/uC CaCC >> ECCeCt 2) CVCCet >> CVCOCt* CVCCVt >> CVCVC 3) CVCCiC >> CVCCuC, (CVCEbC*C) CVCCeC >> CVCCOC, (CVCEbC*C, CoCoCuC) CVCCVC >> CVCEbC*C CVCCVC(V)t >> CoCoCuC 4) 0 5) CoCVC >> CVCECt 6) *CVCaaC >> ECCeCVt, CVCOC, CVCC*t* *CVCiiC >> CVCECt CVCiC >> CVCiCEh (7b) Jibbali, examples: 1) bErk >> EErOk "knee" Harf >> OHrOf "gold coin" kOb >> kOlOb "wolf" Ham@ >> aHm*@ "lower belly" bek*r >> Ebkert "animal with one young" 2) z'efret >> z'ofOrt* "tress" ka`lEt >> ka`el "swollen testicle" C) t*mrEt >> tu~r "date palm" 3) maHfer >> moHofur "basket" k*bkeb >> k*bkOb "star" munxul >> minEbx*l "sieve" 5) gohul >> ghElt "ignorant" 6) bOkOr >> Ebkir*t "pile" s'irOx >> s'irx*t* "root" miger >> mOgOr "skin for milk" xfif, xfift >> xfEft "light" s'hid >> s'hidEh "witness" Notes: distinctive vowel length is lost in Jibbali. Long 'aa' appears to be generally reflected as O, o, or E. The phoneme 'b' is deleted intervocalically. Jibbali, like Tigre, maintains a distinction between group 1 plurals like Ebkert, corresponding with the Ge. ƒˆaCC*Ct type and group 6 plurals like Ebkir*t, corresponding with the Ar. ƒˆaCCiCat type. 3.2.3 NW Semitic In NW and E Semitic an external plural is general, with different forms for the masculine and feminine. In Hebrew there are two widespread exceptions to this general rule, both of which plausibly reflect Proto- Semitic features (Brockelmann 1908/13:430ff., Wallace 1989, Corre1991): 1) The plurals of underlying CVCC (termed 'segolate') nouns regularly show an inserted 'a' between 2C and 3C in the plural form. 2) There is a class of feminine singulars with masculine plurals which appear to go back to older patterns of opposition between singulatives marked with an -at suffix and unmarked collectives. The second pattern is found also in Aramaic, and there is sufficient internal evidence that an older form of Aramaic had the rule of 'a' insertion in segolate plurals as well (cf. Rosenthal 1983:27). (8a) Productive Exceptions to Suffix Pluralization in Hebrew surface reconstructed 1 ) CeCeC >> CeCaƒCi:m *CvCC >> *CvCaC+iim 2 ) CvCCa: >> CeCa:Co:@ *CvCCat >> *CvCaC+aat C) CvC(v)Ca: >> CvC(v)Ci:m *CvC(v)Cat << *CvC(v)C(+iim) (8b) Hebrew, examples: 1 ) melex >> mela:xi:m *malk >> *malakiim "king" sefer >> sefa:ri:m *sipr >> *sipariim "book" qo&e$ >> qo&a$i:m *qud$ >> *quda$iim "sanctuary" 2 ) malka: >> mela:xo:@ *malkat >> *malakaat "queen" $ifHa: >> $efa:Ho:@ *$ipHat >> *$ipaHaat "maid" `orla: >> `ora:lo:@ *`urlat >> *`uralaat "foreskin" C) be:Sa: >> be:Si:m (Ar. baiDat << baiD) "egg" nema:la: >> nema:li:m (Ar. namlat << naml) "ant" s'e`o:ra: >> s'e`o:ri:m (Ar. $a`iirat << $a`iir) "barley" timmo:ra: >> timmo:ri:m, (Ar. tamrat << tamr) "date palm" timmo:ro:@ ( pl. tamaraat, tumuur) 3.2.4 Other Afroasiatic One of the strongest reasons for assuming that some form of internal plural system is reconstructable for Proto-Semitic is that internal plurals are found in non-Semitic Afroasiatic languages. Since Greenberg (1955) first identified these possible parallels, most scholars (e.g. Petracek 1960, Diakonoff 1965) have taken a 'minimalist' approach-- trying to explain away the similarities. In Ratcliffe (1992:372-628) I took a maximalist approach-- trying to explain any plural forms which were remotely similar to or which could plausibly be taken as deriving (historically) from a form remotely similar to a Semitic internal plural as actually deriving from Proto- Afroasiatic. I was unable to find anything that looked like the (3), (4), (5), (6) or 'alif- prefix forms of southern Semitic. In the case of pan-Afroasiatic comparisons the argument from silence is admittedly weak. All sorts of things may be lost or unattested. Yet it is arguably significant that for every aspect of the plural system which NW Semitic and the southern languages (Arabic, Old and Modern South Arabian, and North Ethiopian Semitic) share-- internal 'a' in CVCC nouns, expansion of biradicals to 3C structure in the plural,(10) singulative/ collective (CVCCat/ CVCC) contrasting with feminine plurals (CVCCat >> CVCaC), masculine and feminine forms of the external plural-- parallels can be found in one or more Chadic, Berber, or Cushitic language. Yet for those features which the southern languages share against NW Semitic, only the tendency to form 'u' plurals from 'a'-singulars finds plausible parallels in non-Semitic Afroasiatic languages. Thus even by the most liberal interpretation of the Afroasiatic data (assuming that all of the internal plurals found in some Afroasiatic language which resemble Semitic internal plurals are retentions from Proto-Afroasiatic) the system we would reconstruct for proto-Afroasiatic is basically identical-- except for the 'u' plural-- to the system we would reconstruct for proto-W Semitic on a conservative interpretation (assuming only shared features are original): that is a system in which the internal plural was obligatory for bi-and triradical underived nouns and the external plural was obligatory everywhere else. 4. Evaluating the Isoglosses Chart 9, below, summarizes the data presented above in the form of an isogloss table: (9) Formal isoglosses in the plural system other Ar.OSA Ge. Ti. Ha. Jb. He. Ak. AFAS 1) CVC(V)C >> CVCV(V)C + + + + + + + - + (i/u) > CiCaC - + + ? ? + - (+) CiCaaC + - - + (+) - - (+) (a) > CuCuuC + - - + (+) (-) - + ƒˆaCCV(V)C + + + + + + - - - (i/u) > ƒˆaCCaaC + + + + + - - - (a) > ƒˆaCCuuC - + + (+) (+) - - - ƒˆaCCuC + + + - - - - - ƒˆaCCuCt - + (+) - + - - - ƒˆaCaaCeC - (-) + - - - - - 1-2) CVC >> CVCV(V)C + + + + + + + + + 2) CVCCVt >> CVCV(V)C + + - + + + + - + (i/u)> Ci/uCaC + - + (+) (+) - - (+) (a) > CiCaaC + - + (+) (+) - - (+) (a) > CaCaCaat + - + + + + - + C) CVCCat < CVCC + ? (+) + (+) (+) (+) + + 3) CVCCVC >> CVCaaCVC + + + + + + - - - CVCaCCVC - - - + + - - - - CVCabCVC - - - - - + - - - CVCCiiC >> CVCCaaC - - - - + + - - - 4) CVCVVC >> CVCaaGVC + + + + + - - - - 5) CaaCiC (a.p.) >> CaCaCat + (+) + + + + - - - CVCCaaC + ? - (-) - - - - - 6) CaCiiC >> CuCaCaaƒˆ + + - - - (+) - (-) - ƒˆaCCiCaaƒˆ + + - - - (+) - - - CiCaaC + - - + + - - - CaCaC(V)t - + - - + - - - CVCaaC >> ƒˆaCCiCat + + - + - + - - - CuCuC + ? - (+) - ? - - - AR.OSA Ge. Ti. Ha. Jb. He. Ak. AFAS 4.1 Isoglosses defining SW Semitic against NW Semitic As the chart shows the isoglosses linking Arabic with the southern languages and separating it from other Semitic and Afroasiatic languages are quite stark. Where both NW Semitic and the southern languages have internal plurals, the Arabic forms correspond closely with those of the southern languages. A rule of vowel insertion in plural formation: CVCC >> CVCa(a)C is obligatory for group (1) and (2) nouns in West Semitic generally. Beyond this basic pattern NW Semitic has a characteristic pleonastic plural suffix (Huehnergard 1991, Corr1991). A characteristic feature of SW Semitic, by contrast, is that plurals of the base noun in these languages more often than not have a prefix or prothetic ƒˆa- attached to the plural base. According to Beeston 1962, the plural with alif prefix accounts for more than half of all the plurals in Sabaic. The pattern ƒˆaCCaaC is also the most common plural for CVCC nouns in Arabic, Ge`ez, and Tigre, as indicated on chart 10 by the numbers in parentheses under each type, which is the total number of examples of each type in the sample. In addition SW Semitic languages allow variation both in the quality (a vs. u) and length of the inserted vowel. As chart (10) shows, in Arabic, Ge`ez, and Tigre, the internal 'u' plurals are predominately associated with singulars which have an 'a' vowel. This distribution suggests that the ancestor of these langauges had a rule of vowel dissimilation or apophony. (10) Relative distribution of a and u group (1) plurals (Arabic (based on Levy 1971) CiCaaC ƒˆaCCaaC CuCuuC ƒˆaCCuC (123) (654) (401) (27) CuCC 5, 9 17, 73 6, 16 6, 2 CiCC 5, 7 23, 67 13, 23 12, 3 CaCC 29, 12 26, 27 73, 49 82, 6 CvCvC 5, 6 33, 85 6, 9 -, - Geƒ^ez (based on Leslau 1991, my count) CeCaC ƒˆaCCaaC ƒˆaCCeCt ƒˆaCCeC ƒˆaCCuuC sfp (30) (116) (46) (10) (8) CeCC 100, 28 35, 38 17, 7 10, 1 -, - ,13 CaCC -, - 35, 30 83, 28 90, 6 62, 4 ,18 CaCaC -, - 30, 68 -, - -, - 38, 6 ,20 Tigre (based on Palmer 1962) C*CaC ƒˆaCCaaC(aat) ƒˆaCCuC ƒˆaCC*C ƒˆaCaCC*C(vt) (5) (89) (42) (15) (32) C*C*C 100, 6 80, 76 36, 16 7, 1 3, 1 CaC*C -, - 10, 20 17, 16 80, 27 53, 38 CaCaC -, - 10, 20 48, 44 13, 4 44, 31 In cases where Arabic and the southern languages have internal plurals, while other Semitic does not, the forms are also quite close. A key isogloss defining SW Semitic is the obligatory CaCaaCiC plural of quadriliteral CVCCVC base nouns. Another significant isogloss is that all of the languages (except Jibbali) have the option of forming a plural of triconsonantals with a long vowel according to the quadriconsonantal pattern (CaCiiC >> CaCaaƒˆiC or CaCaayeC). In addition to these three productive and regular plural patterns (CVCC >> ƒˆaCCaaC, CVCCVC >> CVCaaCiC, and CVCVVC >> CaCaaGiC), there are a number of statistically minor or irregular patterns which are shared by Arabic and one or more of the languages further to the south. In group (1) Arabic and a number of other southern languages have a plural in 'u' (or a central vowel '*' reflecting short 'u'). The productive 'u' plurals are, however, different in Arabic and the Ethiopian languages. Arabic has ƒˆaCCuC and CuCuuC; the Ethiopian Semitic languages have ƒˆaCC*C (=*ƒˆaCCuC), ƒˆaCC*Ct and ƒˆaCCuuC. Only the ƒˆaCCuC pattern, which is non-productive and statistically minor wherever it occurs, is shared. In the forms of the feminine base noun (group 2), Arabic, Tigre and MSA all show a pattern whereby feminine nouns fall into two phonologically defined classes, which take different plural types. In Arabic singulars with high vowel generally form a plural by simple insertion of short 'a' (Ci/uCCat >> Ci/uCaC), while singulars with an 'a' vowel either have a pleonastic suffix in the plural (CaCCat >> CaCaCaat) or show dissimilation of the stem vowel combined with insertion of long or short 'a' (CaCCat >> CiCaC, CiCaaC). In Tigre, the high-vowel singulars (C*CCat) consistently show (historically) short 'a' intercalation only in the plural (>> C*CaC), while singulars with stem vowel 'a' (CaCCat) frequently undergo stem vowel dissimilation in the plural and, as in Arabic, have either long or short 'a' inserted (>> C*CaC, C*CaaC). In the MSA languages, too, tri-consonantal feminine nouns fall into phonologically defined classes, of which one takes the simple internal plural only, and the other has a 'mixed' type with vowel insertion plus a feminine plural suffix. For the (lexicalized) active participle, Arabic, MSA, Ge`ez, Tigre and seemingly OSA share a frequent irregular plural CaCaCat (Ge`ez CaCaCt, Tigre CaaCCat), although the Ge`ez (CeCaaCii) and Tigre (CaaCCay) singular forms are somewhat idiosyncratic and although in Arabic this alternates with another pattern CuCCaaC, which is not clearly reflected in other Semitic languages. For singulars CVCaaC a frequent irregular pattern ƒˆaCCiCat or ƒˆaCiCCat is attested in Arabic, Tigre, Jibbali, and apparently OSA. For the Arabic pattern CaCiiC >> CuCaCaaƒˆ, ƒˆaCCiCaaƒˆ, the only SW Semitic parallel is found in OSA.(9) (The parallel Jibbali pattern CVCiC >> CVCiCEh is probably borrowed from Arabic, since it only appears in religious or legal terms borrowed from Arabic). 4.1.1 Hebrew revisited Several efforts have been made (most recently that of Wallace 1989, but going back at least to Salter-Brooks 1883) to identify possible other remnant forms of the broken plural in Hebrew, beyond the productive and well-attested patterns discussed in section 3.4. Brockelmann (1908/13) gives the following list of possible remnant forms in Hebrew. (11) Hebrew Residual Forms (?) (Brockelmann 1908/13:427-8) (Group 1 'u' plural ?) CVCuXC(at) `*ve& >> `evudda *`abuddat "slave" paqi& >> pequdda *pvquddat "officer" gibbor >> gebu:ra *gvbuurat "hero" zaxar >> zexu:r *zvkuur "male" (Group 5 plural ?) CaaCiC >> CaCC roxev >> r*xev *rakb "rider" rome$ >> r*m*$ *ram$ "creeping" $oreS >> $*r*S *$arS "" Wallace (1989) adds the forms pesel >> pesi:li:m (*pasiil) "idol." The key point about these forms is that they do not show a close formal match with any of the productive broken plurals found in Arabic or other southern languages. The patterns CaCC (<< CaaCiC) and CaCiiC (<< CaCC) do appear in Arabic but only very marginally. (Murtonen's sample includes 10 examples of CaCC << from CaaCiC as opposed to 71 examples of CaCaCat << CaaCiC and 8 of CaCiiC << CaCC, as opposed to 399 cases of CuCuuC << CaCC and 312 cases of ƒˆaCCaaC << CaCC.) Thus, although these forms may represent residual proto-Semitic forms of the internal plural in both Hebrew and Arabic, Arabic has largely replaced these forms by innovative forms of the internal plural. And it is these innovatory forms which find close cognates in the other southern languages, but seemingly not in NW Semitic. (11) The plurals with 'u' vowel are particularly interesting. The existence of a 'dissimilatory' plural with 'u' (from CaCC singulars) is the only feature shared by the southern languages but not shared (productively) by NW Semitic which is well-attested in non-Semitic Afroasiatic. Thus the existence of these apparent remnant forms with 'u' (although from singulars of various types) conforms with expectations derived from pan-Afroasiatic comparisons. Here again most of the forms are not very close to those found in Arabic. All of them can perhaps be taken as reflecting an older CvCuuC pattern subject (irregularly) to two essentially analogical changes: addition of a pleonastic -at suffix (a widely attested development throughout southern Semitic, but particularly in Ethiopian Semitic) and gemination of a consonant to compensate for a long vowel. The last development never appears in Arabic but is common in Tigre and Tigrinya, and may have occurred also in MSA. One example of a possible CaCaaCiC plural in Hebrew is given by Wallace (1989) and by Corre (1991)-- the place name `aro:`e:r. Place names need not reflect the language in which they are recorded, of course; and when they reflect the ONLY evidence for a purported phenomenon in a language, they should not be given much weight. Corre's derivation of the Hebrew plural ya:mi:m (sg. yo:m) from an Arabic type plural *ƒˆayyaam through an unmotivated loss of the initial syllable is completely idiosyncratic and ad hoc. Obviously plurals like ya:mi:m << yo:m ƒdaysƒ‚, `a:ri:m << `i:r ƒcitiesƒ‚ and ba:tti:m << be:@ ƒhousesƒ‚ are regular segolate plurals with infixation of short 'a' and deletion of the medial glide in accordance with the phonotactic rule productive in most (all?) Semitic languages which deletes a glide between short vowels (except where the first vowel is high and the second is low). Thus *yawm >> **yawam => *yaam shows the same process as that found in Ar.impf. yaquumu >> pf. *qawama => qaama, ƒstand.ƒ‚ 4.2 Isoglosses linking Southern Arabian and Ethiopian Semitic against Arabic There are no isoglosses in the plural system which would link Ethiopian Semitic and Southern Arabian in such a way as to require a classification of these languages as belonging to a branch of Semitic or West Semitic distinct from that which contains Arabic. At most the isoglosses argue for a division within SW Semitic. That is, all the isoglosses represent modifications of patterns which these languages share with Arabic, but which are not found in NW or E Semitic. Table (9) shows two isoglosses shared by one or more south Arabian language and one or more Ethiopian Semitic language but not found in Arabic. The most significant of these, in terms of number of words affected, is the form of the 'u' plural. The pattern ƒˆaCC*Ct (<*CaCC) links Ethiopian Semitic and Jibbali (as opposed to Arabic which has fu`uul). Yet such a form as plural of CVCC nouns is not clearly attested in Harsusi or OSA. Harsusi, moreover, has a productive plural CeCewweC, which seems to reflect Arabic CuCuuC. Tigre, Tigrinya, and Ge`ez also have a plural ƒˆaCCuuC, which is completely unknown in Arabic. Some of the VCCVVC types of Harsusi and the VCCVC types of Jibbali could reflect this pattern as could some of the ƒˆCCC forms in OSA. The second isogloss is the treatment of adjectives of the pattern CaCiiC. Ge`ez and Jibbali have the plural CaCaCt, although the Ge`ez and Jibbali feminine forms are not cognate. Arabic has a plural CiCaaC, as well as CuCaCaaƒˆ and ƒˆaCCiCaaƒˆ, and occasionally the type 4 plural CaCaaƒˆiC. Harsusi has both a feminine (*CaCiiCat) and plural (CiCaaC) of the Arabic type for adjectives of this class. And in Tigre, although the feminine form of these adjectives is formed, as in Ge`ez, by internal vowel alternation, the plural is formed on the regular type 4 pattern (CaCaay*C), as indicated in Chart 12, below. (12) Feminine and plural forms of CaCiiC adjectives in Arabic, Harsusi,Ge`ez, Jibbali, and Tigre m.CaCiiC f.CaCiiCat pl.CiCaaC f.CaCaaC pl.CaCaCt f.CaCCaaC pl.CaCaaGiC Ar.xafiif xafiifat >> xifaaf Ha.xfef xefefet >> xfeef Ge.`abiiy `abaay >> `abayt Jb.xfif xfift >> xfEft Tg.`abii `abbaay >> `abaayi Essentially both of these isoglosses can be reduced to a single innovation-- i.e. a tendency to add a pleonastic 't' to broken plural forms. This tendency is not unknown in Arabic, it may simply have gone further in other languages. A third isogloss links Harsusi with the modern Ethiopian language Tigre (and also Tigrinya). This is the existence of quadriliteral plurals where the heaviness of the second syllable is created by a geminated third consonant rather than a long vowel (CaCaCCiC, CaGaCCiC, CaCaGGiC). This isogloss is disturbing because such a form is apparently not found in Ge`ez. Even if this must be regarded as an innovation shared by Ethiopian Semitic and southern Arabian, it too merely represents a modification of patterns (CaCaaCiC, CaGaaCiC, CaCaaGiC) shared by all of SW Semitic, but not found in NW Semitic 5. Explaining the isoglosses- innovations or retentions? Comparison of the noun plural systems of the Semitic languages reveals a rich series of isoglosses linking Arabic with the geographically contiguous languages of South Arabia and Northern Ethiopia. Some of these isoglosses reflect retentions shared by NW Semitic and by a number of non-Semitic Afroasiatic languages. However, by the accepted criteria for evaluating both internal and comparative evidence, the ƒˆa- prefix group 1 plurals and all of the 3,4,5,and 6 plurals can only plausibly be explained as the result of innovations shared by SW Semitic languages. The argument from internal evidence could be made on the basis of Arabic, Ge`ez, Tigre, possibly even OSA. Where we find two or more allomorphs showing the same distribution, the assumption is that, other things being equal, the more productive form is newer, while an older allomorph which is non-productive or lexically restricted is likely to be older. (This is Kurylowicz's 4th law of analogy). On this basis we can identify at least two strata. The younger stratum is represented by the statistically productive and formally relatively regular or transparent forms (1.ƒˆaCCaaC, 3.CaCaaCiC, 4.CaGaaCiC, CaCaaGiC). The older stratum is represented by less productive and less regular forms (1.ƒˆaCCuC, 5.CaCaCat, 6.ƒˆaCCiCat, ƒˆaCCiCaaƒˆ). The forms defined as innovative by this criterion show close formal correspondences in all the southern languages. The hypothetically older forms also show partial correspondences in the older (by attestation) southern languages, but these too are not attested outside SW Semitic. As we have said, Diem (1980:70) notes rightly that it is not the existence of internal plurals but of a rich and complicated system of internal plurals which distinguishes SW Semitic from NW and E Semitic. He also says, mistakenly as it turns out, that what characterizes SW Semitic is the existence of a "Typ des inneren Plurals, der dadurch gekennzeichnet ist, dass die Pluralform in keinen formalen Verhaeltnis zur Singularform steht" (Diem 1980:70).(12) If such were in fact the case, if the languages merely shared a handful of idiosyncratic plural patterns conventionally or lexically associated with their singulars, one might reasonably suspect that these could be shared retentions of great antiquity. But such is not the case. The patterns which are shared by SW Semitic languages are not merely the irregular or non-productive patterns, but rather the most productive and regular patterns in each language. The historical record (represented by the difference between classical Arabic and the modern dialects, or less exactly by the difference between Ge`ez and Tigrinya, cf. Corriente 1971:20, 47ff.) shows these forms spreading at the expense of others. Borrowed words which conform with the patterns of the corresponding singulars also generally pluralize according to these patterns. Thus the range of the productive forms has tended to expand and this expansion must have begun in the common SW Semitic period. The comparative argument follows from the facts of relative distribution. Assuming that Arabic and the southern languages derive from a common proto-SW Semitic ancestor the 3,4,5,6, forms and the alif-prefix group 1 forms can be seen as innovations due to analogical extension and sound changes which took place in this proto-language, while the mixed or double plurals of Hebrew and Aramaic can be seen as the result of innovations characteristic of proto-NW Semitic. The discrepancies between North, South and East Semitic are then easily explained as analogical extension working in different directions in different subgroups after their separation from each other. In SW Semitic the method of formation of the internal plural would have been extended to CVCCVC nouns and to lexicalized nouns with the patterns CVCVVC and CVVCVC. In East Semitic the method of formation of the external plural would have been extended to CVCC(at) nouns. In NW Semitic the internal plural would have been maintained in its original domain, but a pleonastic plural suffix would have been superimposed upon the inherited plural forms, as also upon the inherited collectives. The proposed development is indicated in chart (13), where boldface forms represent hypothetical innovations: (13) P-Sem 1) Ci/uCCv >> CvCa Cv internal CaCCv >> CvCuCv plural 2) Ci/uCCatv >> CvCaCv domain CaCCatv >> CvCaCaatv C) CvCcatv << CvCCv 3) CvCCvCv >> CvCCvCvv external 5) CvvCvCv >> CvvCvCvv plural 6) CvCvvCv >> CvCvvCvv domain // P-NW Sem P-SW Sem 1) CvCC >> CvCaCvv Ci/uCCv >> CvCa(a)Cv ƒˆaCCaaCv [CaCC >> CvCuCv] CaCCv >> CvCu(u)Cv ƒˆaCCu(u)Cv 2) Ci/uCCatv>> CvCaCaat Ci/uCCatv >> CvCaCv CaCCatv >> CvCaCaat CaCCatv >> CvCaCaatv C) CvCCatv << CvCCvv CvCCatv << CvCCv 3) CvCCvCv >> CvCCvCvv CvCCvCv >> CvCaaCvCv 5) CvvCvCv >> CvvCvCvv CvvCvCv >> CvvCvCvv CaCaCatv CvGaaCvCv 6) CvCvvCv >> CvCvvCvv CvCiiCv >> CvCiiCvv CvCaaCv CvCaaGvCv CvCaaCv >> CvCaaCvv ƒˆaCCiCatv, CuCaCaaGv CvCaaGvCv If we assume that Arabic and NW Semitic derived from a common Central Semitic ancestor, then Proto-Semitic or Proto-West Semitic would have to have had a plural system like that reconstructed for Proto-SW Semitic in Chart 13. We would have to assume the following : 1) The extension of the method of internal plural formation to CvCCvC, CvvCvC, and CvCvvC nouns took place in Proto W Semitic or Proto Semitic, yielding the innovative group 3,4,5,and 6 plurals, 2) that the variation in the forms of the group 1 plurals involving vowel length and the presence or absence of alif-prefix also developed in proto-Semitic, 3) that these innovative plurals (along with the archaic 'u' plurals) were lost in proto-NW Semitic returning the system back to the original situation, and 4) that THEN a second wave of analogic extension took place in NW Semitic, resulting in the mixed or double plurals. In other words the NW Semitic facts can not be interpreted as a simplification of a hypothetically archaic system preserved in SW Semitic. If the SW system is proto-Semitic, then the NW Semitic system would represent a mysteriously incomplete simplification, combined with an unmotivated complication. As noted above, the superimposition of a pleonastic 't' in Ge`ez and some other southern languages has left the inherited SW Semitic plural system still recognizable and intact. If such a system had been inherited by NW Semitic, superimposition of the sound plural suffixes should have had the same result. 6. Conclusion Hetzron (1976:98-99) argues that morphological isoglosses should be given priority over phonological ones, and that shared innovations should be given priority over shared retentions. These principles seem fundamentally sound. But in the case at hand they do not provide a decisive criterion for classification, because the shared morphological innovations are contradictory. This is not in fact all that surprising. The classification paradoxes in Semitic are typical of the paradoxes one finds in any group of closely related languages spread out over a relatively contiguous territory. Anttila observes of Baltic Finnic (Estonian, Finnish, Karelian, etc.): "The complicated relations within Baltic Finnic cannot be rendered through a tree" (Anttila 1989:301n). Crowley (1992:240-249) makes the same point about a set of Aboriginal languages of Australia, and about the Melanesian languages spoken on the islands of Vanuatu. The same points can be made for Germanic (Hock 1991:444-450, Anttila 1989:303, Crowley 1992:247) and even for Indo-European generally (Hock 1991:450-455, Lehmann 1992:119-123, Anttila 1989:303-306). Bynon (1977:239-256) gives cases of intimate grammatical borrowing in languages of Western Europe, the Balkans, East Africa, Central India, and Southeast Asia. And finally Zaborski 1991 has most recently raised this issue with regard to the conflicting isogloss patterns in Semitic. The solution in such cases seems to be to adopt a relationship model which combines the family tree with the wave model along the lines suggested by Southworth (1964). Classification is, after all, not a goal in itself but a way of drawing inferences from language data about prehistoric movements of, or contacts among, populations. It can not be denied in light of the ]work of Hetzron, Rabin, Faber and others, that Arabic shares important morphological features with NW Semitic. Arabic also shares important morphological features with SW Semitic. The simplest plausible explanation of these facts is one of the following: 1) The linguistic ancestors of Arabic speakers belonged to a Proto-SW Semitic speech community. They separated from this community and came to live in close proximity with speakers of a NW Semitic language. 2) The linguistic ancestors of Arabic speakers belonged to a Proto-NW Semitic speech community. They separated from this community and came to live in close proximity with speakers of a SW Semitic language . The actual facts, of course are likely to have been much more complex, perhaps involving several migrations and contacts or perhaps involving different dialects of Arabic influenced to different degrees by neighboring languages. In order to decide which of these scenarios is more plausible we must ask "Of the shared features, which are likely to be due to contact or convergence and which to common inheritance?" The principle "Central Semitic" isoglosses represent simplifications, specifically reductions of redundancy. If we assume that Proto-West Semitic inherited the Proto-Semitic perfect suffixes -tu (1), -ka (2m.) -ki (2f.), the tendency to eliminate redundant contrasts would favor levelling of the paradigm to -tu,-ta,-ti or to -ku, -ka, -ki. Since there are only two possibilities, the chance of two languages making the same change independently is quite high. Alternately (Proto-) Arabic may have replaced original -k suffixes with -t suffixes due to contact with NW Semitic. A reverse development (-t >> -k) has happened indepently in the modern Arabic dialect spoken in rural north Yemen, under the influence of contact with an unknown South Arabian language (Testen 1992). The loss of gemination on the second consonant of the imperfect represents an equally 'natural' development. Once the inherited stative with suffix conjugation had become established as the usual indicator of the perfective tense/aspect in West Semitic, the gemination of the second radical (yvCvCCvC) which originally served to distinguish the imperfective from the old prefix perfective (yvCCvC) became redundant, since the tense/aspect contrast was now adequately marked by the difference between suffix and prefix conjugations. Loss of gemination has also happened independently in southern Ethiopian Semitic (Greenberg 1952, cf. Leslau 1953). Thus an explanation of these isoglosses in terms of borrowing or convergence is quite plausible. The broken plurals can be plausibly interpreted as the result of an analogic development but the resulting system is nonetheless too idiosyncratic and full of redundancy to be the result of independent developments in the individual languages. Borrowing is possible (and has occurred in MSA for example), but the detailed complicated isogloss network is not consistent with a borrowing explanation. Isoglosses include idiosyncratic marginal forms (reflecting the first wave of analogy?) and the most productive forms in the languages in question. Therefore I would conclude that Arabic is a SW Semitic language strongly influenced by contact with NW Semitic. I don't presume that the preceding analysis represents the last word on this complicated issue. But I do hope that it has laid a foundation for reintroducing the rich evidence from noun plural morphology into the classification debate. References Antilla Raimo. 1989. Historical and Comparative Linguistics. Amsterdam/Philadelphia: John Benjamins. Beeston, A. F. L. 1962. A Descriptive Grammar of Epigraphic South Arabian. London:Luzac & Co. Beeston, A. F. L.; M.A. Ghul; W.W. MŸller; and J. Ryckmans. 1982. Sabaic Dictionary (English-French-Arabic). Publication of the University of Sanaa, Yemen Arab Republic. Louvain-la-Neuve: Editions Peeters and Beirut: Librarie de Liban. Biella, Joan. 1982. A Dictionary of Old South Arabic, Sabaen Dialect. Harvard Semitic Studies 25. Chico: Scholars Press. Brockelmann Carl. 1908/13. Grundriss der vergleichenden Grammatik den semitischen Sprachen. Berlin: Reuther &Reichard. Bynon, Theodora. 1977. Historical Linguistics. Cambridge: Cambridge University Press. Cantineau, J. 1932. Accadien et sudarabique. Bulletin de la SocietLinguistique de Paris 33:175-204. Corr Alan D. 1991. Hebrew-- Some Modest Proposals. In Semitic Studies in Honor of Wolf Leslau on the occasion of his eighty-fifth birthday, ed. Alan S. Kaye. Pp. 245-51. Wiesbaden: Otto Harrosowitz Corriente, Frederico. 1971. Problematica de la pluralidad en semitico. Madrid: Instituto B.A. Montano. Crowley, Terry. 1992. An Introduction to Historical Linguistics. Auckland, New Zealand: Oxford University Press. Diakonoff, Igor, M. 1965. Semito-Hamitic Languages. An Essay in Classification. Moscow: Nauka Publishing House. Diem, Werner. 1980. Die genealogische Stellung des Arabischen in den semitischen Sprachen: ein ungelšstes Problem der Semitistik. In. Studien aus Arabistik und Semitistik A. Spitaler zum 70.Geburtstag, ed. W.Diem and S. Wild. Pp. 65-85. Wiesbaden: Otto Harrassowitz. Dillmann, A. 1903. Grammatik der ƒZthiopischen Sprache, 2nd ed. Leipzig: Wiegel. Faber, Alice. 1980. Genetic Subgroupings of the Semitic Languages. PhD. dissertation: University of Texas at Austin Garbini, Giovanni. 1984. Le lingue semitiche- studia di storia linguistica. Napoli. Goldenberg, Gideon. 1977. The Semitic Languages of Ethiopia and their Classification. BSOAS 40:461-507. Greenberg, J. H. 1955. Internal a-plurals in Afroasiatic. In Afrikanistische Studien Diedrich Westermann zum 80.Geburtstag gewidmet, ed. J. Lukas. Pp. 198-204. Berlin: Akademie Verlag. ---------- 1952. The Afro-Asiatic (Hamito-Semitic) Present. JAOS 72: 1-9. Hammond, Michael. 1988. Templatic Transfer in Arabic Broken Plurals. Natural Language and Linguistic Theory 6:247-270. Hetzron, Robert. 1976. Two Principles of Genetic Classification. Lingua 38:89-106. ---------- 1975. Genetic Classification and Ethiopian Semitic. In Hamito-Semitica ed. James Bynon and Theodora Bynon. Pp. 103-127. The Hague: Mouton. ---------- 1974. La division des langues smitiques. In Actes du premier congrs international de linguistique smitique et chamito-smitique, ed. A.Caquot and D. Cohen. Pp. 182-194.The Hague/Paris: Mouton, . Hock, Hans Heinrich. 1991. Principles of Historical Linguistics. Berlin/ New York: Mouton de Gruyter. Hšfner, Maria. 1943. AltsŸdarabische Grammatik. Leipzig: Otto Harrasowitz Huehnergard, John. 1987. Three Notes on Akkadian Morphology. In Working With No Data: Semitic Studies Presented to Thomas O. Lambdin. ed. David M. Golomb. Pp. 181-193. Winona Lake, Ind.: Eisenbrauns. ---------- 1991. Remarks on the Classification of the Northwest Semitic Languages. In The Balaam Text from Deir `Alla Re-evaluated: Proceedings of the International Symposium Held at Leiden 21-24 August 1989, ed. J. Hoftijzer and Gvan der Kooij. Pp.284-293. Leiden: E.J. Brill. Johnstone, T.M. 1981. Jibbaƒli Lexicon. London: Oxford University Press. ---------- 1977. Harsuƒsi Lexicon and English Harsuƒsi Word-List. London: Oxford University Press. ---------- 1975. The Modern South Arabian Languages. Afroasiatic Linguistics 1/5:1-29 (=93-121). Lambdin, Thomas O. 1978. Introduction to Classical Ethiopic (Ge`ez). Harvard Semitic Studies 24. Missoula, Mont.:Scholars Press. Lane, Edward W. 1863-93. An Arabic-English Lexicon. London: Williams and Norgate. Lehmann Winfred. 1992. Historical Linguistics. London/ New York: Routledge. Leslau, Wolf. 1991. Comparative Dictionary of Ge`ez. Wiesbaden: Otto Harrasowitz. ---------- 1959. The Position of Ethiopic in Semtic: Akkadian and Ethiopic. Akten des 24. int. Orientalisten-Kongresses MŸnchen. Pp. 251-253. Wiesbaden: Otto Harrasowitz. ---------- 1953. The Imperfect in South-East Semitic. JAOS 73:164-166. Levy, Mary M. 1971. The Plural of the Noun in Modern Standard Arabic. Doctoral dissertation: University of Michigan. McCarthy, John J. and Alan Prince. 1990a. Foot and Word in Prosodic Morphology: The Arabic Broken Plural. Natural Language and Linguistic Theory 8:209-283. McCarthy, John J. 1983. A Prosodic Account of Arabic Broken Plurals. In., Current Trends in African Linguistics, ed I.Dihoff. Pp. 289-320. Dordrecht: Foris. Moscati, Sabatino, Antoine Spitaler, Edward Ullendorf, and Wolfram von Soden. 1964. An Introduction to the Comparative Grammar of the Semitic Languages. Wiesbaden: Otto Harrassowitz. MŸller, Walter W. 1964. †ber Beziehungen zwischen den neusŸdarabischen und abessinischen Sprachen. Journal of Semitic Studies 9:50-55. Murtonen, A. 1964. Broken Plurals, the Origin and Development of the System. Leiden: E. J. Brill. Nšldeke, Theodor. 1904. BeitrŠge zur semitischen Sprachwissenschaft. Strasbourg: Karl J. Trubner. Palmer, Frank R. 1962. Morphology of the Tigre Noun. Cambridge: Cambridge University Press. Rabin, Chaim. 1963. The Origin of the Subdivisions of Semitic. In Hebrew and Semitic Studies Presented to G. R. Driver, ed. Winton Thomas and W. D. McHardy. Pp.. Oxford: Oxford University Press. ---------- 1955. The Beginnings of Classical Arabic. Studia Islamica 4:19-37. Ratcliffe, Robert R. 1996. Drift and Noun Plural Reduplication in Afroasiatic. BSOAS 59/2: . ---------- 1992. The Broken Plural Problem in Arabic, Semitic and Afroasiatic: A Solution Based on the Diachronic Application of Prosodic Analysis. Doctoral dissertation: Yale University. ---------- 1990. Arabic Broken Plurals: Arguments for a Two-Fold Classification of Morphology. In Perspectives on Arabic Linguistics II, ed. Mushira Eid and John McCarthy. Pp. 94-119. Amsterdam/ Philadelphia: John Benjamins. Raz, Shlomo. 1983 Tigre Grammar and Texts. Malibu: Undena Publications. Rodgers, Jonathan. 1991. The Subgrouping of the South Semitic Languages. In. Semitic Studies in Honor of Wolf Leslau on the occasion of his eighty-fifth birthday, ed. Alan S. Kaye. Pp. 1323-1335. Wiesbaden: Otto Harrosowitz. Rosenthal, Franz. 1983. A Grammar of Biblical Aramaic. Wiesbaden: Otto Harrosowitz. Salter-Brooks. 1883. Vestiges of the Broken Plural in Hebrew. Dublin. Southworth, Franklin C. 1964. Family-Tree Diagrams. Language 40:557-65. Testen, David. 1992. Extra Arabic Affilitations of k-Yemeni. In. Perspectives on Arabic Linguistics IV, ed. Ellen Broselow, Mushira Eid and John McCarthy. Pp.77-90. Amstersdam/ Philadelphia. John Benjamins Voigt, Rainer M. 1987. The Classification of Central Semitic. Journal of Semitic Studies 32:1-21. Wallace, Constance V. 1989. Broken and Double Plural Formations in the Hebrew Bible. Doctoral dissertation: New York University. Wehr , Hans. 1960. ed. and trans.by J. Milton Cowan. Dictionary of Modern Written Arabic, 3rd. edition. Ithaca, N.Y.: Spoken Language Services. Wright, William, ed. and trans. 1896. A Grammar of the Arabic Language, 3rd. edition. London: Cambridge University Press. Zaborski, Andrzej. 1991. The Position of Arabic within the Semitic Dialect Continuum. In Proceedings of the Colloquium on Arabic Grammar, ed. Kinga DŽvŽnyi and Tam‡s Iv‡nyi. Pp. 365-375. Budapest: The Arabist, Budapest Studies in Arabic. NOTEs 1) A version of this paper was presented at the North American Conference on Afroasiatic Linguistics in Philadelphia in March, 1996 and a much abridged Japanese version at the African Linguistic Perspectives conference in Tokyo, June 1996. Thanks to Robert Hetzron, Robert Hobermann, Gene Gragg, Carleton Hodge, Lionel Bender, Alan Kaye, David Testen, Akio Nakano, and others present at both conferences for their comments. Thanks also to John Huehnergard for originally suggesting this line of research and for help with bibliography. 2) The key isoglosses for Hetzron's classification are (1) the first, second and third person singular suffixes of the perfect-- tV in Hebrew, Aramaic and Arabic, -kV in South Arabian and Ethiopian Semitic, and (2) the presence of gemination on the second consonant in the imperfect stem in Akkadian and MSA and Ethiopian Semitic, vs. its absence in Arabic and NW Semitic. 3) The situation is not so simple. The existence of simple forms of the inner plurals in NW Semitic only proves what has long been known-- that the pheomenon of inner plurals was not foreign to Proto-Semitic-- but not that Proto-Semitic knew such a rich system as Southeast Semitic and especially Arabic. 4) The central question is the interpretation of the inner plurals of Arabic and Southeast Semitic-- the decision depends upon a critical evaluation of them. 5) For a more detailed summary and critique of the compartivist/ diachronic literature treating the problem of recronstruction of the broken plural, see Ratcliffe 1992:175-241. 6) The designation "i+e" indicates that the plural form regularly shows both internal and external morphemes of plurality simultaneously. The designation "i,e" indicates that there are both internal plural forms and external plural forms which occur regularly with this singular class. Listed singular forms represented the assumed underlying forms. 7) A summary of Levy's and Murtonen's data-- total percentages of plurals for each singular and singulars for each plural-- in chart form can be found in Ratcliffe 1992:97-98. 8) Actually there is a seventh group which includes a small, closed set of adjectives like ƒˆaHmaru >> Humrun "red." We leave these out of consideration here, for the sake of simplifying the discussion. We also leave out of consideration those plurals which are regularly associated with singulars which are traditionally treated as having a 'weak'-- i.e. glide-- radical: for example CuCaat (<< CaaCii = CaaCiGu) as in (ƒˆal)-qaaDii >> (ƒˆal)-quDaatu "judge," CiiCaan (<< CaaC) as in baabun >> biibaanun "door," CiCaCat (<< CiiC) as in diikun >> diyakatun "chicken." 9) Although there is a possible E Semitic parallel, see note 11 below. 10) The data and and an explanation for this 'shared anomaly' of Afroasiatic is presented in Ratcliffe (1996). 11) The same line of argument holds for the adjective plurals with suffix -a:ƒˆ which Huehnergard (1987) identifies in Akkadian and connects with Arabic CuCaCaaƒˆu (<< CaCiiC). Given the Akkadian evidence, one can not rule out the possibility that the final -aaƒˆu sequence on the Arabic form reflects an inherited Proto-Semitic adjective plural suffix. However, the existence of this plural type is in any case not an important isogloss for the SW Semitic hypothesis, first because the form is found only in the Arabian peninsula languages Arabic and OSA, second because the restricted distribution of the form (to a closed class of nouns in Arabic, to no more than four examples in OSA) would, even without the Akkadian evidence, lead us to suspect that this is a retention rather than an innovation. 12) A type of inner plurals, which is characterized by the fact that the plural stands in no formal relationship to the singular. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - @@An expanded and revised version of this paper appears in the Journal of Near Eastern Studies (University of Chicago Press, Chicago, Il., U.S.A.) vol. 57, no.2 (April,1998), pp. 81-123 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -