Proto-Aztecan Vowels: Part I Author(s): Lyle Campbell and Ronald W. Langacker Source: International Journal of American Linguistics, Vol. 44, No. 2 (Apr., 1978), pp. 85-102 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/1264522 Accessed: 24/02/2010 19:51

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=ucpress.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to International Journal of American Linguistics.

http://www.jstor.org InternationalJournal of AmericanLinguistics

VOLUME44 April 1978 NUMBER2

PROTO-AZTECAN VOWELS: PART I1

LYLE CAMPBELL ANI) RONALDW. LANGACKER

STATE UNIVERSITY OF NEW YORK, UNIVERSITY OF CALIFORNIA, ALBANY SAN DIEGO

0. Introduction quality (either **i or **e) has been the 1. Reconstruction subject of continuing controversy. 2. Residual patterns and problems The one exception to the relative uni- 3. Implications formity of the Aztecan vowel systems is Pochutec, once spoken in Pochutla, Oaxaca and now 0. At first the vowel systems of the extinct. Pochutec was on the Aztecan languages appear to offer little of verge of extinction when Boas did his interest to the comparative Uto-Aztecanist. salvage work in 1912, represented in his "El Dialecto Members of the Aztecan subfamily are paper Mexicano de Pochutla, the first very similar and seem to point to a recon- Oaxaca," article in the first issue of IJAL. This is the struction of *i, *e, *a, and *o (with only substantial source of information on distinctive length) for the vocalic system of Pochutec. Pochutec is Proto-Aztecan; this system predominates strikingly divergent from the rest of Aztec, even in the daughters and in no case is a compared to Pipil of Central America, which is often considered daughter system drastically different. The the most diver- The distinctness only notable aspect of this system with gent. of Pochutec is most respect to Uto-Aztecan as a whole is that probably notable in the vowel system, where three traits in Aztec *i apparently reflects both **i and particular stand out: **u of Proto-Uto-Aztecan (henceforth (1) distinctive vowel length has apparently been Pochutec PUA). The vowels *a and *o of Proto- lost;2 (2) o often corre- to i or e in the rest of Aztecan (henceforth PA) continue the sponds Aztec; and stress falls on the final PUA vowels of the same quality, and PA (3) vowel in in contrast to the has been assumed to agree with the other Pochutec, penultimate stress found in Classical southern Uto-Aztecan subfamilies (with and other modern dialects. the exception of Pimic) in having *e as the Our reflex of the fifth PUA vowel, whose purposes in this article are to recon- struct the vowel system of PA, taking the 1 Each of the authors claims responsibility for Pochutec evidence into account, and to the insights captured in this article and blames consider the implications of this recon- the other for errors or infelicities that any may struction for Aztecan and Uto-Aztecan remain. We wish to acknowledge helpful com- ments from William Bright, Una Canger, Karen subgrouping and reconstruction generally. Dakin, Kenneth Hale, and Thelma Sullivan. This The implications of Pochutec are substan- research was supported in part by NSF grant tial. For example, it is probable that #20-692-A. 2 "Creo que no hay vocales largasen Pochutla" [IJAL, vol. 44, no. 2, April 1978, pp. 85-102] "El Dialecto Mexicano de ? 1978 by The University of Chicago. (F. Boas, Pochutla, 0020-7071178/4402-0001 $01.51 Oaxaca," IJAL 1 [1917]: 9-44, esp. 10). 85 86 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44

Proto-Aztecan (PA)

Pochutec (Po) General Aztec (GA)

Classical Nahuatl Tetelcingo Zacapoaxtla Pipil (CN) (T) (Z) (Pi) FIG. 1

Pochutec is coordinate with all the other a task for which neither our energies nor members of the Aztecan subfamily, for available data are sufficient.4 which we adopt the term General Aztec 4 Our sources of data are: for Pochutec, Boas (GA) (for lack of a more imaginative label). (n. 2 above); for Pipil dialects of Cuisnahuatand The vowel system *i, *e, *a, *o, while Santo Domingo de Guzman, both in El Salvador, L. Vocabulario valid for GA, cannot be justified for PA as Campbell, Pipil (San Salvador: of for a for which we must reconstruct Ministry Education, forthcoming); whole, Tetelcingo, F. Brewerand J. G. Brewer, Vocabu- vowels not directly attested in any of the lario Mexicano de Tetelcingo, Morelos ( Aztecan languages. The PA vowel system City: Instituto Lingiistico de Verano, 1962) and that we hypothesize bears importantly on R. S. Pittman, A Grammar of Tetelcingo (Morelos) the of reconstruction and sub- Nahuatl, supplement to Language 30 (1954), problem and "The Phonemes of in the Uto-Aztecan 'Tetelcingo (Morelos) grouping family. Nahuatl," in A William Cameron Townsend The languages and dialects that we will (1961), pp. 643-51; for Zacapoaxtla, D. F. consider are shown in figure 1. We choose Robinson, Aztec Studies II: Sierra Nahuat Word these both for availability of materials and Structure(Norman, Okla.: Summer Institute of for their inherent interest, and we Linguistics, 1970) and H. Key and M. Key, linguistic Vocabulario de la Sierra de confine our attention to them. With this Mejicano Zacapoaxtla, (Mexico City: Summer Institute of overall classification (we ignore possible Linguistics, 1953); for , F. A. subgroupings within GA) we find our- de Molina, Vocabulario en Lengua Castellana y selves in considerable agreement with Mexicana (1571; reprint ed., Mexico City: Whorf and Lastra de but differ Editorial Porrua, 1970), H. Carochi, Compendio Suarez, del Arte de la Mexicana ed. I. de from Hasler.3 We believe our classification Lengua (1645), Paredes (Mexico: Imprenta de la Biblioteca and analysis of the PA vowel system to be Mexicana, 1759), R. Simdon, Dictionnairede la correct; however, we attempt no complete LangueNahuatl ou Mexicaine (Paris: Imprimerie resolution of all the loose ends and minor Nationale, 1885), M. Swadesh and M. Sancho, since that would involve us in the Los Mil Elementos del Mexicano Cldsico (Mexico details, Universidad Nacional Aut6noma de extreme of Aztec City: complexities dialectology, Mexico, 1966), and J. R. Andrews, Introduction 3 See B. L. Whorf, "The Milpa Alta Dialect of to Classical Nahuatl (Austin: University of Texas Aztec," in LinguisticStructures of NativeAmerica, Press, 1975). ed. H. Hoijer et al. (New York: Viking Fund, We purposely avoid the vexed question of 1946), pp. 367-97; Y. Lastrade Suarez,"Apuntes where the boundary between dialect differences sobre Dialectologia Nahuatl," Anales de Antro- and language differenceslies. We are certain that pologia 11 (1974): 383-98; and J. A. Hasler, Pochutec was a separatelanguage, quite different "Tetradialectologia Nahua," in A William from the others. It also seems that Pipil is a CameronTownsend en el VigesimoquintoAniver- separate language, though rather closely related sario del Instituto Lingiiistico de Verano,ed. B. to the rest of GA. We venture no guesses as to Elson and J. Comas (Mexico City, 1961), pp. the status of the other members of GA, though 455-64. in any event they are not very differentfrom one NO. 2 PROTO-AZTECANVOWELS 87

1. Previous attempts at stating the Pochutec but "do not attempt a detailed sound correspondences between Pochutec treatment." They note that PUA **u and GA have been brief and unsystematic. "remains as a " in Pochutec, Voegelin, Voegelin, and Hale5 (henceforth namely o, rather than being fronted to i as VVH) recognize the divergent character of in GA; this is an important point, par- ticularly with respect to the word "re- mains," as we will see. They also note the another. As for Pipil, though it is quite similar basic between GA oo and and to other there correspondence lexically phonologically GA, Po as well as the of **a to Po e are grammaticaldifferences of some significance. u, change Glottochronology (for whatever it may be in certain environments. However, they worth) suggests a time depth of fifteen minimum omit the crucial fact that GA e often centuries (or ca. A.D. 500) for PA, with Pochutec corresponds to Po o rather than Po e, and splitting off first. GA has a time depth of eleven for the most do not consider minimum centuries A.D. when part they (ca. 800), Pipil vowel since were "not as split off (see T. Kaufman, Idiomas de Meso- length, they yet america [Guatemala, 1974]). This time correlates able to reconstruct vocalic length for UA well with the ethnohistorical reconstruction of generally."6 Pipil migrations (see W. Jim6nez-Moreno, Being more concerned with relationships "Mesoamerica before the Toltecs," in Ancient within Aztecan than with those between Oaxaca, ed. J. Paddock [Stanford, Calif.: Pochutec and Boas7 provided a Stanford University Press, 1966], pp. 4-82 for PUA, details). The Pipil were scattered in pockets generally more accurate account of the throughout Central America. Central to the correspondences, in each case treating the ethnohistorical reconstruction is the Nicarao Pochutec vowel as innovative, derived of statement recorded (Pipil Nicaragua) by from the GA counterpart: short a becomes Torquemada that the to Central migration short e becomes and i America took place seven to eight "edades" or e; o; long becomes "vidas de viejo" before. This is generally inter- long o. Boas must be credited with the preted as the period called "huehuetiliztli" of discovery of the o-e correspondence be- 104 years (two fifty-two-year cycles), giving tween Po and GA, but he was wrong in roughly A.D. 800 (note the glottochronological the that the date). As the reconstructions have it, the making gratuitous assumption Pipils Pochutec left Cholula (in Puebla, Mexico) to escape the vowels were necessarily innova- tyranny of the historical Olmecs; they came to be tive. In particular, his GA ii > Po oo in Veracruzin the cultural province of Cerro de change contradicts VVH's statement that las Mesas and Los Tuxtlas briefly,but were again PUA **u "remained" a back rounded to establish themselves in displaced, going vowel, instead that it became a Xolotan (the Soconusco coast of implying Chiapas). back rounded vowel However, to escape historical Olmec oppression after having pre- once again, they left Soconusco, and passed viously been fronted.8 through Guatemala and El Salvador, leaving Five years later, in his IJAL paper, Boas settlements as far as Nicaragua, Costa Rica, and greatly refined these correspondences and Panama.This ties in well with dialectinformation, data. since shares several similarities with Gulf provided supporting Unfortunately, Pipil he still Coast Nahua, e.g., Mecayapan (see Lastra de treated the Pochutec vowels as Sudrez [n. 3 above] for details). This being the case, it would be surprising if 6 Ibid., pp. 61-62. Pipil were not a separate language. Perhaps the 7 F. Boas, "Phonetics of the Mexican Lan- surprising thing is the degree of similarity still guage," Proceedings of the International Congress existing after so many years separation. of Americanists 18 (1912): 107-8. 5 C. F. Voegelin, F. M. Voegelin, and K. L. 8 We emphasizethat this is an extrapolationof Hale, Typological and Comparative Grammar of Boas's claims, since he was concerned only with Uto-Aztecan: I (Phonology), IJAL Memoir 17 Aztecan and not with the UA source of Aztecan (1962). vowels. 88 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44 derived from those attested in GA, specifi- There are eleven basic correspondences, cally the modern Aztec spoken in the though it will be possible to reconstruct Valley of Mexico, and his correspondences fewer than eleven distinct vowels for PA. differ significantly from those reported in They are divided into five sets on the basis his earlier paper. He still claimed correctly of likely PA quality distinctions. It seems that short a becomes e in Pochutec, while evident that distinctive vowel length must long aa remains as (short) a. In regard to be reconstructed for PA, although the the o-e correspondence, he revised his length distinction has apparently been lost statement that the change affects only in Pochutec and has been realigned in short e and claimed instead that it affects Tetelcingo as a distinction of vowel only stressed e, regardless of length. More- quality. We now consider in turn the five over, while he originally said that long ii sets of vowels in table 1. shifts to oo in Pochutec, in the later work he reversed himself, saying that almost all 1.1. The obvious reconstructions for the the instances of i that become o are short. vowels of set 1 are *a and *aa for the Boas also noted a relation between GA u respective correspondences e-a-a-a-a and and Pochutec o, without refining his state- a-aa-o-aa-aa. For the most part these ment in terms of length or other condition- correspondences are straightforward, with ing factors. some exceptions to be dealt with below (2). In view of these conflicting analyses, it A set like 42 deer is particularlyinstructive, is apparent that we must approach the in which the first vowel reconstructs as *a matter anew, taking into account the wider and the second as *aa: Po mesat, CN range of data presently available. In masaatl, T masatl, Z masaat, Pi masaa-. table 1, we give the basic correspondences Other sets illustrating the regular e-a-a-a-a correspondence for PA *a are 2, 8, 12, 17, TABLE 1 22, 28, 29, 33, 37, 43, 46, 49, 52, 53, 56, 58, 70, 72, 80, 83, 95, 98, 100, 105, 108, 112, Po CN T Z Pi 116, 119, 124, 128, 132, 133, 138, 140, 142, 1 ...... e a a a a 147, 149, 159, 163, 168, 170, 175, 181, 182, a aa o aa aa 183, 185, 194, 198. Other sets illustrating 2...... o o o o u the regular a-aa-o-aa-aa correspondence U 00 U o UU for PA *aa are 4, 10, 25, 31, 48, 62, 64, 68, 3 ...... i I i i 69, 96, 106, 110, 124, 130, 133, 157, 160, i ii i ii ii 179, 186, 189, 197. 4...... o i I i i o ii i ii ii 1.2. Since we know length is distinctive, 5 ..... e e e e e the most likely reconstructions for the o e e e e vowels of set 2 are *o for the o-o-o-o-u cor- e ee ie ee ee respondence and *oo for the u-oo-u-oo-uu correspondence. Long *oo was raised to for the five daughters under consideration. u(u) in some of the daughters; in Tetel- These correspondences are supported by cingo, this was part of the general process the cognate sets assembled in section 5 whereby length distinctions were realigned (to appear in a later issue of IJAL); as quality distinctions. Since there was no various refinements and revisions to these phonemic distinction between *o(o) and correspondences will be considered below. *u(u), there being only one back rounded NO.2 PROTO-AZTECANVOWELS 89 vowel (occurring either long or short), or not we can find reasonable conditioning variation in quality between o(o) and u(u) environments that determine when *i(i) is is to be expected.9 The Pochutec situation reflected in Pochutec as i and when it is is fairly complex and will be considered reflected as o. Second, since the attestation carefully in 2 below. A cognate set illus- of the o-ii-i-ii-ii correspondence is sparse trating both correspondences is 39 coyote, at best, can a length distinction be main- with *o for the first vowel and *oo for the tained? Third, if one or more vowels second: Po koyud, CN koyootl, T koyutl, distinct from *i(i) must be reconstructed Z koyoot white man, Pi kuyuut. Other for this set, what is their vocalic quality ? regular sets with *o are 1, 9, 18, 26, 74, 77, The answer to the first question is that 104, 116, 123, 126, 134, 142, 145, 151, 171, no plausible conditioning factor(s) can be 175, 178, 181, 198. Other regular sets with isolated, hence PA vowels distinct from *oo are 7, 23, 24, 40, 41, 63, 127, 128, 152, *i(i) must be reconstructed to represent the 163. fourth set with Pochutec o. An examina- tion of possible conditioning environments 1.3. For the i-i-I-i-i correspondences (accent, word boundaries, and adjacent and i-ii-i-ii-ii, the best reconstructions are segments) reveals that none of them per- clearly *i and *ii respectively. The first mits any general statement predicting is illustrated twice in 167 correspondence when Pochutec will have i and when it will in the root and once in thorn, once *wi?- have o. the absolutive Po wisti, suffix, -t(l)i:10 In the case of accent, we must consider CN wiCtli,T blCtlI, Z wiCti,Pi wiCti.Other both the penultimate stress of Classical sets for *i are 3, 6, 20, 41, 47, 52, regular Nahuatl,11 presumably reconstructible for 61, 73, 79, 81, 83, 94, 95, 106, 130, 156, PA (see below), and the word-final stress 187. The root 161, 165, 172, 173, 174, 184, of Pochutec.12 Word-final stress as a con- vowel of 36 cornfield shows the second ditioning factor is ruled out, since both o correspondence: Po mil, CN miilli, T mill, (from the relevant correspondences) and i Z miil, Pi miil. Other regular sets for *ii occur freely in Pochutec final syllables are 2, 17, 27, 32, 55, 98, 113, 147, 155. (e.g., 1 noco all, everything, 63 sucot 1.4. The fourth set, with the correspon- flower, 68 nayom four, 87 kiskom how dences o-i-I-i-i and o-ii-i-ii-ii, is problematic much, how many, 3 asi arrive, find, 84 eti in several ways. First, is it really distinct heavy, 184 nki want, 187 ig2ti (ig'ti) weave). from the previous set or can both be derived Nor is the Pochutec o/i distinction corre- from *i(i)? The answer hinges on whether lated with PA penultimate accent, since we find both o and i for vowels that were 9 For some discussionof this matter, see penultimate in PA (e.g., 37 osket cotton, H. Seiler and G. Zimmermann, "Studies in the Phonology and Morphologyof ClassicalNahuatl: 11See W. Bright, "'Accent' in Classical I," IJAL28 (1962): 243-50; and Carl Wolgemuth, Aztec," IJAL 26 (1960): 66-68. "Isthmus (Mecayapan) Nahuat Laryn- 12 "El acento cae casi siempre en la iltima geals," in Aztec Studies I, ed. D. F. Robinson silaba" (Boas, "El Dialecto Mexicano," p. 13). (Norman, Okla.: SummerInstitute of Linguistics, We have not investigated the accent shift of 1969), pp. 1-14. Pochutec in any detail, but almost certainly it 10This form (167) is the only one in which Po involved (1) loss of final (unstressed) vowels in has -ti; elsewhere, the Po absolutive is -t or 0. certain instances, leaving stress on the final The origin of Aztecan tl is still uncertain; below, syllable, and (2) regularizationof stress to make we present evidence suggesting that tl was a it consistently final. Vowel loss will be discussed feature of PA. in 2 below. 90 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44

140 -ote see, 2 ?iket ant, 155 til soot) and a-a (short *a should become e in Pochutec), for vowels that were unstressed in PA and os-sii (via metathesis, but a different (e.g., 61 micomfish, 63 sucot flower, 3 asi metathesis than that discussed below). arrive,find, 83 keki hear). These difficulties suggest that the Pochutec Word boundaries are also ruled out, form is not cognate. However, if it were to since we find the relevant o in both word- prove to be a valid cognate, it is still initial and word-final position (e.g., 37 possible that the o could come from epen- osket cotton, 138 ostet salt, 1 noco all, thesis after deletion of the first stem vowel everything, 75 toso grind), while the same (see below). The second set that might be is true of i (e.g., 55 istface, 187 ig?ti weave, suggested is 172 throw,stir, move: Po uluni, 83 keki hear, 130 tali put, place). As for CN oliinia, T ulinIa, Z ooliiniya, Pi uuliini. adjacent segments as possible conditioning The problem here is the Pochutec u rather factors, our data is sufficient to establish than the required o. One could hypothesize the contrast beyond reasonable doubt, that it had been o, but due to harmoniza- since we have examples of Pochutec o and i tion with the regular u of the first syllable contrasting both before and after a number it became u. In view of the lack of other of consonants. To take just one example, examples supporting the o-ii-i-ii-ii corre- k can precede both o and i, and it can spondence, it is certainly doubtful that this follow both: 87 kiskom how much, how irregular case supports it. Even in Pipil many, 104 okost male, 83 keki hear, 93 there is variation in some forms, for koski itch, 45 mok die, 92 -tok in, inside, example, the past participle uuliin-tuk, 113 sikt navel, 147 kWike-sing. uuluun-tuk, where there has never been The second question is whether a dis- any question of any Pochutec-like o tinction can be made between the similar correspondence. correspondences o-i-I-i-i and o-ii-i-ii-ii in Since the evidence is doubtful at best for view of the slight attestation of the latter. the o-ii-i-ii-ii correspondence, we pursue Since all other vowels definitely show a the matter no further. It may well be that length contrast, we might expect the dis- there was a length contrast which was lost tinction here too. The o-i-I-i-i correspon- before PA times. If such a length contrast dence is amply attested, for example, 45 die persisted into PA, our data are insufficient shows this correspondence for the first to demonstrate it. We leave this as an : Po mok, CN miki, T mIkI, Z miki, question.13 Pi miki. Other sets with this cognate 13 Another set for the are possible cognate correspondence 1, 30, 41, 57, 59, 61, 63, o-ii-i-ii-ii correspondenceis ear of corn CN sintli 67, 68, 75, 87, 92, 94, 104, 106, 137, 140, (lengthuncertain), Z siinti, and Pi sinti. The length 144, 151, 173, 193. In the case of o-ii-i-ii-ii, variation is unexplained. Both VVH and W. R. on the other hand, only two cognate sets Miller (Uto-Aztecan Cognate Sets [Berkeleyand even the and Los Angeles: University of California Press, suggest correspondence, reconstruct**sunu for and is neither is without 1967]) PUA, **u(u) problems. the source of GA i. There are problems, however. One cognate set that might be suggested The same root may be reflected with a short to support the o-ii-i-ii-ii correspondence is vowel in the compound seed T sInoatlI,Z sin-aac, 71 gourd vessel (jicara): Po oskast, CN Pi sinaac womb(cf. set 141). Furthermore,other dried corn sikalli, T sikallI Z siikal, Pi siikal. cognates are related: 50 ear of corn, cup, Po CN T Z senti. These it seems that the Po form is not a son/-sn-, sentli, sentlI, However, may all be early loans from Otomanguean real cognate. If it were, we would have to languages (Proto-Otomanguean *sen, from C. assume irregular correspondences of s-l, Rensch, "Comparative Otomanguean Phonol- NO. 2 PROTO-AZTECANVOWELS 91

The third question is, what vowel should highly dubious, especially in view of a more be reconstructed for o-i-I-i-i? We cannot plausible alternative.15 Furthermore, the reconstruct long or short *i, since these evidence indicates that the back rounded underlie other correspondences, and we can vowel may have varied allophonically rule out *o(o) for the same reason. The PA between *o(o) and *u(u) already in PA, vowel representing this correspondence which suggests that PUA **u(u) had set had to have been capable of evolving to already been unrounded to *i by that time. both i (in GA) and to o (in Po). Moreover, To summarize, we reconstruct PA *i for since this PA vowel reflects PUA **u(u) the o-i-I-i-i correspondence. It reflects (and became i in GA), we can plausibly PUA **u(u), though the length distinction conclude that it was high. We exclude *ii probably disappeared before PA times. because the development *ii > o is quite PA *i was fronted to i in GA; it became o unlikely (not to mention ii being completely in Pochutec, probably through the route foreign to UA). This leaves *u and *i as *i > *o > o. possible choices. 1.5. This brings us to the last set of The choice between *u and *i is basically correspondences, e-e-e-e-e, o-e-e-e-e, and a of The question chronology. development e-ee-ie-ee-ee. The best reconstruction for of PUA **u to GA i probably did not come the last of these is *ee, with diphthongiza- about in one step; **u > *i > i is much tion in Tetelcingo and shortening in more likely.14 The important question is Pochutec. A good illustration of this whether the intermediate step *i arose in correspondence is 109 moon: Po mest, CN PA or after the between subsequently, split T Z Pi Po and GA. The choice between *u and *i meeCtli, mieCtlI, meeCti, meeCti. Other sets are 13, 26, 54, 88, 100, for PA thus on whether the regular depends 101, 107, 111, 118, 135, 150. Pochutec o in derives directly question For e-e-e-e-e and o-e-e-e-e, we face from *u or comes instead from the inter- again the of whether the mediate *i. question correspon- stage dences are real, and whether are The evidence indicates that Pochutec o they really distinct. That is, must we reconstruct two did indeed descend from *i and not directly to underlie these from *i the best reconstruc- contrasting segments **u, making Boas's claim that GA e tion for PA. Several factors to be made correspondences? becomes o in Pochutec when accented16 evident below suggest this, so for now we needs to be investigated. While we do not merely note the following simple but, we accept his proposed development *e > o argument. Positing both hope, persuasive for Pochutec, it may be that stress is a > o and > u for Pochutec is *u(u) *o(o) relevant factor. In fact, it is the case that only stressed vowels participate in the ogy" [Ph.D. diss., University of Pennsylvania, o-e-e-e-e correspondence, with only mar- 1966], p. 629). The forms cited by VVH and ginal exceptions (to be discussed presently). Miller are all from southern languages, so we suspect that **sunu is not a valid PUA recon- 15Most cases of "flip-flop" rules prove to be struction (though it may be valid for Proto- either erroneous or not to have come about in a Southern-Uto-Aztecan). single stage (i.e., X > Y, Z > X, Y > Z, rather 14 The partial fronting of *i may be another than directly X > Z, Z > X). Such an inter- intermediatestep; in Hopi, for example, i (from mediate stage is what we propose in our alterna- PUA **i) has a markedly front articulation, tive (**u(u) > *i > o). occasionallyeven vaguely suggestingI to speakers 16 "e del mexicano, con acento, se vuelve o en of English. Pochutla"(Boas, "El Dialecto Mexicano,"p. 12). 92 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44

In most cases, the vowel in question was CN tle?ko, T tlehko, Z tehko, Pi tehku. stressed both in PA and in Pochutec, for Other sets are 58, 89, 118.19 example, 5 ashes: Po nost, CN nestli, T We have seen, then, that stressed vowels -nestlI, Z -nes, Pi nesti. Other sets for result in the correspondence o-e-e-e-e which this is true are 12, 34, 50, 76, 103, unless they are word-initial or followed 115; 132, 162, 194. directly by saltillo, where e-e-e-e-e occurs. There are instances of stressed e in both But what about unstressed vowels? In GA and Pochutec, but they are easily some cases the Pochutec reflex seems to be explained. The vowel is reflected as e in e, in others it is zero. There are not many Pochutec only when it is either word-initial examples with the e reflex in Pochutec, and or followed directly by "saltillo" 7 in PA. most have minor irregularities that leave There are two sets involving word-initial room for doubt. If this e could be explained vowels: 16 blood Po est, CN estli, T yestll, away in unstressed position, we would be Z esti, Pi esti; and probably 84 heavy Po eti, left with the generalization that this vowel CN etiik, yetiik, T yetik, Z etiik, Pi etik becomes o when accented and zero other- (see below). (Contrast these with sets 135 wise in Pochutec. If the Pochutec cases roasting ear and 169 three, with initial *ee.) with e reflexes cannot be explained away, Although the Tetelcingo forms begin with then we are left with e-e-e-e-e as an addi- ye (which might be interpreted as ie, tional correspondence in unstressed posi- suggesting a long vowel), several factors tion. argue that the vowel was in fact short. Consider the cases; the reflex is e in 88 First, it is recorded as short in Zacapoaxtla huipil keskemt, 123 owl tekolot, 149 skin, and other dialects. Second, external com- leather kWetest, and 180 two omem. parison with Uto-Aztecan in general Pochutec keskemt huipil seems to have indicates a short vowel. Third, Tetelcingo been a crucial case for Boas's e-e corre- lacks word-initial e except for a few spondence in unstressed position. It corre- borrowings (from Spanish); y in these sponds to kec-keemit(l) in central Mexico, forms comes from either a Tetelcingo which is not a huipil, but a triangular tendency to lengthen initial vowels, or from neck-piece, worn ornamentally over the the general tendency in GA for y to be huipil (native blouse). Its distribution may inserted before word-initial e(e) with have been limited largely to the area around varying degrees of consistency.18 The the state of Puebla.20 Its etymology, kec- correspondence e-e-e-e-e before saltillo is neck and -keemi-tl cloth, reflects its use in illustrated by 131 raise, ascend: Po teku, central Mexico, but Pochutec huipil has no inherent relationship to "neck," just as 17 Aztec saltillo is h in most dialects and ? in Po kos- neck may have little to do with the h in others; we have arbitrarilywritten it as our kes- of keskemt. The quechquemitl If we assume that vowel-initial portion reconstructions. was a of in Oaxaca and words in Pochutec really began with saltillo (a symbol authority plausible assumption in view of general Uto- its use was highly restricted. The Spanish Aztecan tendencies), the two environments in suppressed it to maintain control, and thus which stressed vowels take part in the e-e-e-e-e 19 correspondence could be collapsed, since both Another example is Po -ye mother; cf. involve adjacent saltillo. However, it is not clear Mecayapan -ye? and PUA **yPi (254). whether such a generalizationwould be genuine 20 D. B. Cordry and D. M. Cordry, Costumes or spurious. andTextiles of theAztec Indians of the Cuetzalan 18 See Lastra de Suarez (n. 3 above) for some Regionof Mexico,Southwest Museum Papers 14 discussion. (1940). NO.2 PROTO-AZTECANVOWELS 93 other Oaxacan languages have no attested seems to have added the plural suffix word for it.21Thus it is likely that Pochutec (-mV), which PUA **woo- lacked; how- keskemt is a post-Conquest borrowing ever, Pochutec seems to have added a from GA.22 As for 123 tekolot owl, it not second plural -m (omem) and even a third only has the e which we are questioning, (?) in some cases (omemet kig'laskit two but also o in the final syllable, where we women). The ending for other numbers is expect u by the regular correspondences -om (eyom three, nayomfour) in Pochutec, (cf. CN tekolootl, T tekolutl, Z tekoloot, which makes the -m or -em of omem Pi tekuluut). These vowel problems could suspicious. Thus it is possible that two be due to a number of things. Perhaps the omem is borrowed into Pochutec from GA, Pochutec form is borrowed from Spanish or that Boas's informants were somehow tecolote or from GA; perhaps Boas mis- influenced by the GA form (Boas may have recorded it; perhaps vowel harmonization suggested it). Though these speculations accounts for the o in the last syllable; etc. may be poorly founded, it is nevertheless In any case, Po tekolot is not free of undeniable that the Pochutec form omem problems. Skin, leather, 149 kWetest(cue- two has complications that make it text), is also not without complications. dubious. In short, the evidence is not For example, Tetelcingo has an otherwise sufficient to demonstrate either the exis- unexplained I, kWItlastlI.Since this seems tence or absence of the e-e-e-e-e corre- not to have cognates outside of Aztecan, spondence in unstressed position; it is, the vowel cannot be checked by compara- however, controversial enough to cast tive evidence. Besides the e-e-e-e-e potential suspicion on the correspondence. correspondence, one can imagine many There is, on the other hand, a clear other possible explanations, for example, 0-e-e-e-e correspondence, as illustrated in the possibility that the Pochutec form is 44 descend: Po -d&mu(d'mu), CN temo(a), borrowed from GA, the possibility that T temo, Z temowa, Pi temu. The 0-when- Po e is the result of vowel harmonization unstressed-and-o-when-stressed pattern is or vowel epenthesis (kwtest > kWetest?). also shown in Pochutec morphophonemic Finally, 180 two omem is not entirely variation, for example: 194 woven mat felicitous either. If it should prove to be an mo-pot your mat, b?te-t (b'te-t) mat (plus accurate cognate, it would be the only case absolutive suffix); kea raise (ctza), k6sk where we would have to posit an e-e-e-e-e raised (cozc); etc. Other examples of the correspondence before the shift of stress 0-e-e-e-e correspondence are 9, 50, 112, to final syllables in Pochutec. The Pochutec 132, 184, 194, etc. o is also irregular, not the expected u (cf. There are a couple of exceptional cases, CN oome, T ume, Z oome, Pi uume). PA where o-e-e-e-e occurs even unstressed, for example, 93 itch and 193 work. One can 21 B. Jordan La de Dahlgren, Mixteca: Su imagine a number of potential explana- Cultura e Historia Prehispdnicas (Mexico: tions for Pochutec o in these cases. One Imprenta Universitaria, 1954). 22 Po keskemt has undergone some uniquely particularly appealing possibility is that Po changes (c > s/ C, i > 0); these changes, Pochutec was prevented from losing its however, appear to be quite late and not very vowel (if the 0-e-e-e-e correspondence in profound. Many contemporary GA dialects also unstressed position is maintained) when to change (?, c) fricatives (s, s) before the loss would have resulted in initial consonants. The Po loss of i may be late, since one would expect nasal assimilation (not m-t) if consonant clusters of more than two the change were deeply embedded in the language. members (i.e., a prohibition against the 94 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44 formation of #CCC-); this would prevent but also from PUA **a in certain circum- Po tokdo work and koski itch from losing stances (to be discussed further), suggesting the first vowel. a mid central vowel in the reconstruction. In summary, the main correspondence Furthermore, reconstructing *o and posit- appears to be o-e-e-e-e in stressed position; ing the change *o > o for Pochutec has e-e-e-e-e, when stressed, is confined to the additional support of relating this word-initial position or the position innovation to another Po innovation, the directly before saltillo. The main corre- change of PA *i to Po o (through the spondence in unstressed position may be intermediate stage *a). We reconstructed *i 0-e-e-e-e, though we cannot entirely rule for the correspondence o-i-I-i-i together out the possibility of e-e-e-e-e also. In any with the sound change *i > *3 > o, with event, these correspondences are not in the hypothetical intermediate stage *' to contrast, and a single PA vowel is to be make the change phonetically plausible. reconstructed to represent them. (Vowel The rounding of *a to o in Pochutec is loss will be dealt with again below.) (apparently) the same change, and that the It remains to choose a protosegment to two can be collapsed in Pochutec is addi- represent these correspondences. Since we tional strong support for our reconstruc- already have *ee to represent e-ee-ie-ee-ee, tions. we might naturally consider *e. But this 1.6. In table 2 we summarize the corre- will not work. While it would make our spondences and reconstructions we have PA system more symmetrical with con- established. Table 2 does not accommodate and short it does not trasting long vowels, various secondary developments to be account satisfactorily for Pochutec o. A considered in the next section, nor vowel of *e to o would be change hardly plausible loss to be taken up below. It should be as a single-step development, nor is there noted that the vowels *i and *a have been any evidence to suggest a multiple-step reconstructed for what we believe to be A much more development.23 plausible compelling reasons, even though they are and alternative is that PA had likely *a, not preserved unchanged in any of the which evolved to Pochutec o through daughters; additional supporting argu- circumstances des- rounding (in the just ments are presented later. cribed above) and became e in GA (and in 2. Examination of the sets Pochutec word-initially or before saltillo) cognate assembled be in section by fronting, in both cases eliminating the (to presented 5) marked segment *a in favor of an un- TABLE 2 marked segment. This is supported by the fact that this PA vowel normally develops PA Po CN T Z Pi from PUA **i (which we support below), *a e a a a a 23 *e > *o > o is hardly likely, since the *aa a aa o aa aa change must be restrictedto stressed vowels; the *o o o o o u centralization of a peripheral vowel like *e is *00oo u oo u 00 uu more commonly associated with lack of stress. *i i i I i i *e > *ew > o is perhaps not totally out of the *ii i i ii ii question, but there is no evidence whatever to corroborate this kind of diphthongizationof *e, *i o i I i i is our and indeed, such a development (to '* o/e/0 e e e e knowledge) attested nowhere else in Uto- *ee e ee ie ee ee Aztecan. NO. 2 PROTO-AZTECANVOWELS 95 reveals many instances in which the generally are shortened before consonant expected reflex does not appear in one or clusters, including initial vowels, which more daughters. In this section, we try to otherwise are generally all long. This provide an account of the discrepancies. accounts for the variant length in 87 and 129. Furthermore, the vowel of a final 2.1. In some instances, the only dis- unstressed syllable is also generally short, is the occurrence of a short vowel crepancy and length in this position can only be in Classical Nahuatl where a long vowel is determined in derivationally related forms expected. We suspect that most of these in which the syllable is no longer in final reflect the fact that was not simply length position (e.g., masat deer, masaa-?in little indicated in the orthography of generally deer). This accounts for the short vowel of Classical Nahuatl, that the vowels were in 15, 30, 84, 85, 135, 139, 152, 161. Remain- fact long, but that length is not attested in ing cases of problematic length in Pipil are the documents.24 sets with this Cognate 28, 46, 56, 66, 82, 121, 146, 154, 169. Part of for Classical Nahuatl type discrepancy of the problem may be Pipil's very mori- are: for *aa, 8, 43, 74, 141, 176; for *oo, bund condition; the few remaining good 18, 19, 33, 35, 56, 129, 146, 172; for *ii, 23, speakers in Cuisnahuat have difficulty 43, 152, 156, 182. recalling whether a form has a long or Length in Pipil poses problems in a short vowel, especially in infrequent words. number of forms. For the most part, the Apart from these, there is a lesser Pipil length discrepancies reflect special number of sets in which Tetelcingo, developments in this language and do not Zacapoaxtla, or Classical Nahuatl (when affect the PA reconstructions. Vowels length information is available) shows a in relation to the 24 The basic source of information on vowel length discrepancy length is Carochi (n. 4 above). However, while testimony of the other daughters. Examples Carochi harps on the desirability of careful are 4, 7, 10, 20, 25, 28, 38, 47, 62, 66, 73, pronunciation,he does not always indicate length 80, 86, 91, 139, 154, 157, 164, 170, 179. (he marks the saltillo more consistently). More- Several of these (28, 38, 73, 157) involve rather than a over, since his work is a grammar medial w and reflect the of dis- dictionary, length information is not provided difficulty for any word which he does not happen to use as tinguishing such sequences as oo, ow, oow. an example. Swadesh and Sancho (n. 4 above) The worst case we have of medial w also give length indications for cases where they influencing the length of adjacent vowels not their source of have data, but they do specify (or its perception and transcription) is we cite vowels for CN on their information; long 153 snake: Po CN T kowatl, authority in 13, 17, 27, 40, 48, 62, 64, 68, 69, 84, kuet, kooaatl, 85, 88, 98, 101, 106, 110, 111, 113, 127, 133, 135, Z koowaat, Pi ku(u)wat; CN aa, T o, and 147, 154, 155, 163, 170. We cite length on the Z aa are unexpected given our reconstruc- authority of Andrews (n. 4 above), who also does tion *koowa-, which the other daughters in not indicate his source of information, 2, 4, 7, amply justify. This number of length 10, 23, 30, 32, 41, 46, 54, 66, 86, 148, 152, 157, is not more than one would 160, 161, 164, 172, 179, 197. We have used discrepancies indirect evidence for length (length cited for expect from dialect variation and difficulties parallel or related forms) in 79, 139, 166, 190, of consistent recording. 192. In all these cases, other daughters being Other Tetelcingo length exceptions are considered corroborate length information. In only due to distributional restric- Nahuatl apparent, addition, elicitation with North Puebla tions. We out above that short e corroborates the length for 68, 79, 98, 110, 133, pointed 139, 190; Nahuatl for 64, 68, 98, 135, does not occur initially in Tetelcingo, so 139, 192. the initial ye in 16 yestlI blood and 84 yetik 96 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44 heavy does not contradict the reconstruc- several forms, we attribute the exceptional tion with a short vowel. Similarly, Tetel- vowel to harmonization with an adjacent cingo has no i/I contrast in word-initial vowel: 35, 54, 105, 143, 178 (?), 185. In position, which accounts for the quality of 60 tet fire ai merged to e; in 161 iyek the first vowel in 37, 47, 120, 138, 168. stinking ya became ye by assimilation; and Other apparent irregularitiesfor the GA we assume the first two vowels of 66 daughters are relatively few, and most have pusonelfoam were interchanged. The first natural explanations. Vowels have been vowel in 87 and 90 is left unexplained.25 lost due to morphological developments in Now we turn to exceptions involving 19, 30, 66, 164, 196. Discrepancies that we back rounded vowels. The changes we have attribute to the harmonization of adjacent established for back rounded vowels in vowels are found in 34 (?), 59, 75 (?), 103; Pochutec are *o > o and *oo > u. Both Pipil 103 atimet louse may show vowel of these are strongly attested, yet there are interchange (cf. CN atemitl). Various many forms that seem to violate them. natural assimilations and mergers have Most, however, prove to be predictable. In occurred, such as -yi > -y, aw > ow, four cases, where the vowel was adjacent ya > ye, and ay > ey > i (14, 30, 60, 78, to m in PA, *oo becomes Po o rather than 112, 117, 169, 174, 188). The only sets in u: 19, 65, 170, 180(?), for example, 170 which one of the GA daughters has a throw Po mote, CN mootla. It may be, vowel that can be explained only by appeal therefore, that *oo > u was inhibited by to sporadic change or analogy are 10, 35, an adjacent m, though there are exceptions 149. (e.g., 127 kumt pot, jug). There are many more examples where 2.2 Discrepancies in Pochutec, the other *o becomes u, rather than remaining o as branch of Aztecan, can be explained with we predict. The great majority of these fall comparable success. The fate of *o was under the special rule: *ow > u. Most considered above and will be discussed commonly, the *ow which shifts to Po u is again later, and so will not be dealt with part of the common verb ending -owa; the further here. The discussion of other -a of these cases and others is fundamen- vowels is less than definitive; a careful and tally the transitivizing suffix that occurs extensive examination of Pochutec pho- especially in present tense forms (i.e., the nology would be most welcome, though thematic vowel of present, imperfective, the language's state of decay (very near and conditional items) and is lost in Po extinction), uncertainty of the informants' due to either phonetic or morphological knowledge, and Boas's orthography may developments (compare Po and GA forms make some uncertainties inevitable. We in 13, 20, 38, 41, 44, 93, 94, 99, 101, 102, present what we believe the Pochutec 106, 122, 125, 130, 136, 143, 165, 172, 195). phonetic values to be, in symbols consis- An example is 122 open Po tepu, CN tent with those we use for the other tlapoa; notice the Po past tense form Aztecan representatives; where Boas's tepok, where the transitional w after o and orthography is ambiguous or uncertain, we before the truncated -a is absent and list the form in his orthography, as he Pochutec consequently has o. Other it, in parentheses. 25 presented These two residue forms are similar enough We will to with the mis- try dispense to suggest some phonetic or analogical explana- cellaneous cases first, before turning to the tion: 87 how much, many kiskom; 90 iguana more general classes of exceptions. In kWisom. NO. 2 PROTO-AZTECANVOWELS 97 examples of *ow > u are 11, 22(?), 28(?), 144, 196; there are no exceptions in our 44, 73, 99, 101, 102, 125, 136, 143, 195, data. The presence of saltillo is therefore 196(?).26 (Possible exceptions to this rule a conditioning factor in two sound changes are 38, 116, 145.) (*%> o/e, *a > e) and possibly in the We have only four other cases of irregular *o > u. *o > Po u: 23, 131, 159, 190. In 131 and A second regularity is the correspon- 159, *o was contiguous to saltillo, which dence between GA -aawak and Po -ek in we have seen to be relevant to vowel shifts adjectival forms (79, 166, 190, 192), for in Pochutec: 131 raise, ascend Po teku, example, 79 hard, strong Po cikek, CN CN tle?ko; 159 squashPo eyut, CN ayo&tli. cikaawak. The discrepancy is apparently In 23 and 190, the environments for the explained by the fact that -aawak adjectival exceptional u are so similar that a phonetic forms in GA often also have a "short" or analogical explanation might be sought: form in -ak, especially in compounds and 23 buzzard ?upilut; 190 white, clean, clear when followed by another morpheme cupek (cf. 214). The k of 131 may be rele- within the word. Pochutec seems to reflect vant (see n. 25 above), and there is evidence the short version. (For example, compare for positing saltillo in 23, though not Pi cipaknah very white, Pi cipak-tuk enough at present to establish such a cleared, CN cipak-kaltik very clean, T reconstruction definitively; note Zaca- cIpak-tIk clean, which occur alongside the poaxtla ?ohpiiloot (North Puebla Nahuatl cipaawak forms.) also has saltillo in this form). Regardless Several cases remain of *a > a, not the of the proper analysis of these forms, they predicted e. The situation seems basically are not sufficiently numerous to cast doubt irregular, as one might expect in the case of on the general correspondences we have a sound change in progress, which this shown. may have been. Though we have not This leaves only the sets in which *a attempted to explore the grammatical remains a in Pochutec, instead of becoming correlates of this sound change, it is e as expected. We consider the predictable possible that many of the remaining regularities first. One is that *a remains a exceptions have to do with the final -a of before saltillo, for example, 3 arrive, find verbs, which in many cases is not part of Po asi, CN a?si. Other examples are 6, 51, the verb stem, but is the present tense, transitivizing suffix.27 26 It is not clear how Boas's always orthog- Remaining exceptional instances of a is to be perhaps the change raphy interpreted; the are 64, was *ow > w (Boas's orthographic u), and (not expected e) 10, 11, 20, 21, perhaps w > u by glide vocalization. In a case 69, 103, 120, 158, 164(?), 177, 192, 196. such as 57 fall Po o?o, GA we?i (PA *wo?i),it is While no explanatory rule seems possible, not clear whether Po lost *a (via changes to be we can speculate about possible reasons which became presented below), yielding *w?o for some of these cases. For example, via vocalization > and vowel o?o glide (w u) 158 Po has two harmonization; or whether Po simply mono- malage-, -meleg- spindle phthongized *wo to o. This is suggested by Po variants, the latter regular, the former enolak I came (from e- (perfective), *nV- (I), wala- (come), -k (past)), where monophthongiza- 27 See Boas, "El Dialecto Mexicano," pp. 21- tion (wa > o) seems to have taken place. The 22. Also note 4, 24, 25, 40, 81, 145, 160, 196. vocalization of w is suggested by the morpheme Po kokoa sick (145) is probably a loan from of intimate possession (which most possessed CN kokoa. If this were directly descended from forms in Pochutec have), which is -w in GA, but PA *koko(wa), we would expect Po koku by accented -u in Po. the changes previously discussed. 98 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44 irregular, but supplied by different infor- 3. We have reconstructed for PA the mants. We might speculate that those who vowel system shown in table 2, with the knew the language well gave the regular short vowels *a, *o, *i, *i, *o and the long -meleg-, while those who did not know it vowels *aa, *oo, *ii, *ee (doubtfully *ii). so well gave malage under the influence of In this section, we discuss the implications Spanish malacate, no doubt the form Boas of this reconstruction for both Aztecan used to elicit these responses. Of course it is and the Uto-Aztecan family as a whole. possible that there was simply free varia- 3.1. For the Aztecan subfamily, several tion, as well. For 64, 69, and 164(?) we major conclusions emerge. First, we must might guess that the irregular a is due to reconstruct for PA penultimate accent (as harmonization with the regular, stressed a in Classical Nahuatl) rather than the final of the As for 10, it is following syllable. accent of Pochutec, since that the a is from *i after k penultimate possible (see stress interacts with several sound and thus not a true changes below), exception. with vowel Second, Form 20 which 21 is have (especially loss). (to related) may saltillo must also be reconstructed for PA had sporadic lengthening (cf. Z taatiya). despite its absence in Pochutec, for it plays Finally, perhaps the w (or labial com- a role in several Pochutec phonological ponent of kW)inhibited the change in 177, developments. Third, there is good evi- 196. In the remaining cases, the unchanged dence that Pochutec is parallel to all the a is adjacent to a voiceless obstruent, other Aztec dialects and languages (what which may have had some effect, though we have called General Aztec), as shown in certain similar environments apparently figure 1. facilitated the change, for example, k t This third conclusion may require some (2, 8, 37, 72, 105, 108, 117, 133, 168). elaboration. While parallel innovations Finally, though vowel loss is discussed may complicate matters slightly, the sound in detail later, we should point out some implied by table 2 provide per- Pochutec kinds of loss. One changes specifically suasive evidence for the coordinate status kind is the loss of final *i or *i in nouns of Pochutec and GA. Pochutec has under- before an absolutive suffix, for example, gone major sound changes shared by no 103 louse Po atomt, CN atemitl. Other other daughter, notably *i > *a > o (in- instances are 19, 88, 97, 127, 176, etc. This cluding PA *o > o) and *a > e. Similarly, loss predates the shift from apparently all of GA has undergone two major sound penultimate to final stress, since otherwise changes not shared by Pochutec, namely, the vowel would be stressed and pre- *i > i and *o > e.28 These are basic and sumably from deletion. Final i protected distinctive innovations counting heavily in drops from intransitive verbs in 18, 45, 48, determining subgroupings. They contrast, 64, 156, though these verbs also have in this regard, with those innovations in GA which lack the final allomorphs common to Pochutec and GA because of vowel under certain conditions (see below). parallel developments. One such innovation is the 2.3. In summary, we maintain that parallel of to in Tetelcingo, most apparent exceptions to our corre- raising *o(o) u(u) Pipil, Since there was no *u in PA can be accounted for in a and Pochutec. spondences with the of *o to u in reasonable manner. While a few uncertain to contrast *o, raising cases remain, the basic validity of the 28 Pochutec did undergo *o > e, but only as a analysis has been established. minor change word-initially or before saltillo. NO. 2 PROTO-AZTECANVOWELS 99 various daughters can hardly be grounds The basic changes that derive Pochutec for grouping them into a subfamily; in from PA are *a > e, *oo > u, *i > *o fact, allophonic variation between o and u (and on to o/0), *o > o/e/0 (e if followed may have been a feature of PA. Our view by saltillo or word-initial, o if stressed, of the independent shift of o to u finds 0 unstressed), and neutralization of vowel strength in the fact that the details of the length. The changes *a > e and *oo > u realignment differ in the daughters in must precede the loss of length distinctions. which it occurred. In Pipil it affected both The *i first became o, merging with *o, and long and short *o; it affected only *oo in then becoming o, etc. Tetelcingo as part of the general shift from We summarize the evolution of PA a length distinction to a quality distinction; vowels in figures 2 and 3. Figure 2 shows and in Pochutec it affected *o and *oo, but ii ii in differentways under differentconditions, i i as we have seen. ee 00o ee 00 Another parallel innovation, the loss of -- vowel length, differs again in details, 0 o e 0 insuring independent development. The aa aa Tetelcingo development clearly stands a a apart in that the length distinctions were FIG. 2 uniformly transformed into quality distinc- ii tions (*aa > o, *oo > u, *ee > ie, *ii > i, u *i > I). In Pochutec, on the other hand, i i i length distinctions were for the most part ee o0 simply neutralized, but the loss of length had to follow certain other uniquely -I 0 -f 0 e o Pochutec changes which were sensitive to a length (in particular, *a > e). Assuming that Pochutec is coordinate \1a with GA, we can now trace the evolution a from PA of the daughters under considera- FIG. 3 tion. Two sound changes derive the GA the from the vowel system from that of PA: *i > *i and changes leading PA system the to the GA *o > *e. No ordering relation can be (on left) system (on the The evolution from PA to Pochutec established for these two changes, and it right). is sketched in 3. Note that may be significant that they both involve similarly figure in both the evolution results in a fronting. They yield for GA the vowel instances, more unmarked system *i, *e, *a, *o, plus distinctive vowel relatively system, lacking nonlow central vowels. length. This system is retained unmodified in Classical Nahuatl and Zacapoaxtla. In 3.2. We turn now to the implications of Pipil an additional change occurred, the our reconstruction for the entire UA raising of *o(u) to u(u). The Tetelcingo family. As noted earlier, there is general changes all play a role in realigning the agreement that a five-vowel system, with length contrast to quality distinctions: distinctive length, can be reconstructed for *aa > o, *oo > u, *ee > ie, *i > I, *ii > i. PUA. It is also generally agreed that four The change of *i to I must precede *ii > i of the five vowels are **i, **a, **o, and (or be simultaneous with it). **u. For the fifth vowel Miller follows 100 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44

* Sapir in reconstructing *e, while Langacker Langacker implicitly assumed that im- agrees with VVH in reconstructing **i.29 plausible developments such as **i > *e > i Our analysis bears directly on this con- or **e > *i > e were to be avoided if troversy. possible, so the necessity of reconstructing Although there is some evidence that a *i for Proto-Takic was taken as casting basic division can be made between doubt on the likelihood of **e for PUA. Northern Uto-Aztecan and Southern Uto- However, it is not a necessary assumption Aztecan,30 for purposes of the present that the vowel shifts of a language family discussion we can assume that the UA show a certain coherence in their overall family consists of eight subfamilies: direction, so the fact that Cahuilla e Numic, Tiibatulabal, Hopi, Takic, Pimic, derives from Proto-Takic *i does not Taracahitic, Corachol, and Aztecan. Four preclude the possibility of the latter in of these subfamilies show i as the reflex of turn deriving from PUA **e. Nor do the fifth PUA vowel (Numic, Tiibatulabal, numbers provide compelling evidence one way or the other. With *i for Proto-Takic, there are five subfamilies that **i (Sr) support for the fifth PUA vowel Tibatula- i i u (Numic, bal, Hopi, Takic, Pimic) and three that (Ca)e *- a - o(L) suggest **e (Taracahitic, Corachol, Az- (Cu) tecan). This ratio slightly favors **i over but the is but a **e, advantage anything decisive, since it is conceivable that a sound FIG. 4 change **e > *i spread among the five northernmost subfamilies and left the three show e as the Hopi, Pimic); generally original **e unaffected only in the south. reflex of this vowel (Taracahitic, Corachol, These arguments have assumed without and one shows a Aztecan); (Takic) variety question that *e is the proper reconstruc- of reflexes (Serrano i, Luisefio o, Cupeno o, tion for the PA descendant of the fifth Cahuilla Langacker showed that *i e). PUA vowel. Though *e is indeed accurate must be reconstructed for Proto-Takic; *i for GA, the picture for PA changes was retained in Serrano but lowered to (Sr), drastically when Pochutec is taken into *o in Proto-Cupan; *o in turn Proto-Cupan account. The whole question of **i versus was retained as in o Cupefio (Cu), fronted **e for PUA therefore needs to be re- to e in Cahuilla (Ca), and rounded to o in examined in light of the more accurate Luisefio (L). A composite picture of the reconstruction of the PA vowel system that development of Proto-Takic *i is sketched we have proposed. We argue that the case in figure 4. for **i as the fifth PUA vowel is greatly strengthened by this additional evidence. 29 Miller 13 See (n. above) and E. Sapir, The PA reflexes of the fifth PUA vowel "Southern Paiute and Nahuatl: A Study in Uto-Aztecan," Journal de la Societe des Ameri- are not *ee and *e as generally assumed canistes de Paris 10 (1913): 379-415 and 11 previously (this is valid only for GA), but (1919): 443-88. Also see R. W. Langacker, "The rather *ee and *--a crucial difference. Vowels of Proto Uto-Aztecan," IJAL 36 (1970): The is that the fifth PUA vowel and VVH 5 implication 169-80, (n. above). was toward in the southern 30 J. Heath, "Uto-Aztecan Morphophonem- evolving *e(e) ics," IJAL 43 (1977): 27-36, and Kaufman (n. 4 branches of the family, but had not com- above). pleted this shift by PA times. We have no NO. 2 PROTO-AZTECANVOWELS 101 reason to doubt that *ee is the best PA **e > (*o) > *i for the five northernmost reconstruction for the long vowel, but the subfamilies to account for *i as the reflex of short vowel was still central in PA times. the fifth PUA vowel. The conflict with the The evolution of PUA **i in Aztecan is Takic evolution involves not just one stage, therefore seen to parallel its evolution in then, but two linked stages; we find it Takic; **i first lowers to *3 and is then highly implausible that Takic would agree either fronted to e or rounded to o (com- with Aztecan and the other subfamilies in pare figs. 3 and 4). Since the direction of regard to half of this integrated vowel shift this evolution in Takic seems firmly (*i > *o > o)31 but disagree with it in established, the plausibility of the similar regard to the other half (*i > *o > e development we posit for Aztecan is versus **e > *3 (> *i)). Moreover, if it is greatly enhanced. conceded that *3 is the probable inter- A much less reasonable picture emerges mediate stage in the change of **e to *i in when **e (rather than **i) is considered as the northern subfamilies, the following the PUA candidate. One would have to unlikely evolution would have to be posited claim either that PUA **e was backed to for Cahuilla: PUA **e > *a > Proto- *a in PA and then generally reverted to e, Takic *i > Proto-Cupan *a > Cahuilla e. or that PUA **e was retained as PA *e, With **i, on the other hand, only **i > which was backed to *o in Pochutec and *a > e is required. then rounded to o. The former alternative Our analysis, then, leads to a picture of requires **e > *3 > e, which appears con- Aztecan and UA vocalic evolution that is trived and fails to reflect the natural much more coherent and much less con- Aztecan evolution, namely, that the overall trived than any readily available alterna- tendency to reduce markedness eliminates tive. *i (both long and short) must be *i and *o in favor of less marked e or o. The reconstructedas the reflex of the fifth PUA latter alternative requires that stressed *e vowel in the five northern subfamilies; *e be centralized to *o (> o) in Pochutec, not can be reconstructed only for Taracahitic what one would expect on the grounds of and Corachol; while for Aztecan, *ee is the phonetic naturalness. best reconstruction for the long vowel and Both the **e alternativeshave the further *o for the short vowel. PUA **i was disadvantage of accounting for a distribu- evolving toward *e (through *a) in the tion of vowels similar to Takic in a manner southern branches of the family. (The simi- diametrically opposed to that required for lar Takic development was independent.) Takic. Compare figures 3 and 4 once again. The evolution ran its course in Proto- We take it as established that *i > *a > e/o Taracahitic and Proto-Corachol, but in is the proper direction of evolution for Proto-Aztecan the reflexes were still *ee Takic. Exactly the same vowels figure in the and *o. Six of the eight subfamilies now Aztecan scheme, and in our analysis a provide evidence that e was not the primi- precisely analogous derivation is proposed tive quality of the fifth PUA vowel, includ- (with minor differencesof detail). However, ing the southernmost as well as the five if **e ratherthan **i is posited for PUA, we northern subfamilies.The case for **i rather would have *e > *a rather than *o > e in than **e is therefore very strong indeed. Aztecan, just the opposite of the Takic 31 Since PA *i reflects PUA the The conflict is even more **u, change development. *i > *3 > o is needed for Aztecan, specifically startling when we broaden the scope of the Pochutec, regardless of whether **i or **e is analysis. With **e, we would have to posit reconstructedfor the fifth PUA vowel. 102 INTERNATIONALJOURNAL OF AMERICANLINGUISTICS VOL.44

i i +- u i i u e <- (a) t- i t i +- u - (o) 0 e U <-- 0 a a

FIG. 5

But the case for **i is even stronger; structed, for we find that **i evolved further support is provided by the vocalic toward *e in the southern subfamilies, changes in Corachol (Cora and Huichol). eliminating a marked segment for an Besides **i (> *o) > *e, Corachol has unmarked one. This change triggered the **u > i and **o > u. If **e rather than other vowel shifts in Corachol in the classic **i were reconstructed as the fifth PUA "drag-chain" fashion shown in figure 5. vowel, no change involving **e would be Our analysis thus establishes an intrinsic required, since *e is the Proto-Corachol relation between the controversial i/e reflex. Hence no relation would be evident vowel in UA and the innovation of *i between the fifth UA vowel (**i or **e) and (from **u) in Corachol.32This is additional the Corachol sound change **u > *i. In strong support for **i as the fifth PUA particular, **u > *i would be totally inde- vowel. pendent of the **e > *i shift needed for 32 The Corachol chain shift transforms a the northern subfamilies, and one would vowel system with more back vowels (and fewer have to argue that two different groups of front vowels) into a system with more front subfamilies disrupted the symmetrical and vowels (and fewer back vowels). We might that the of the oral unmarked vowel system **i, **e, **a, **o, suspect asymmetry cavity, with more at the front and less space at the **u to introduce the marked vowel space highly back, would favor changes toward less crowding *i by totally differentmeans. No such coin- in the vowel space of back vowels, increasing the cidence need be posited if **i is recon- number of front vowels.