Here I present a translation of pp. 18–23 of the Фонетика volume of the Сравительно-историческая грамматика тюркских языков ed. E. R. Tenišev (Moscow: Nauka, 1984). This is a work which I appreciate more and more as time goes by, and I hope to bring further portions of it into English in future.
Close ẹ in the Common Turkic vowel system
This problem arises in the reconstruction of Proto-Turkic vocalism, and its solution depends on solving the question of how many cardinal vowels there were in Proto-Turkic: 8 or 9. Theoretically the following hypotheses are possible:
the Proto-Turkic system had ä (wide) and ẹ (narrow);
there was a qualitative and quantitative opposition of these vowels, i.e. ä versus ẹː (wide short versus narrow long);
there was ä (short) and äː (long): this variation of the reconstruction is actually very similar to the second if one takes into account that phonetically a long vowel is usually more close than the corresponding short one;
there were äː and ẹ (wide long and narrow short);
there were ä, äː, ẹ and ẹː;
at an early stage Proto-Turkic had ä̂, äː, ä and ệ, ẹː, ẹ, but in Common Turkic (with the exception of Chuvash) ä̂ and ệ fell together into ä̂, while ä and ẹ fell together into ä (cf. variant 5).
All of these variants have been discussed in specialist literature.
First of all, one must observe that in the modern Turkic languages there are not two (open and close) but several phonetic variants of phonemes which can be presented in transcription as æ, ä, ɛ, e. Furthermore, in each specific system or type of system they have their own particular origin and status.
Thus in languages of the Kipchak type (Kazakh, Karakalpak and Nogay), where a comparatively regular raising of mid vowels occurred, the variant e was established, which could have originated in ɛ or ẹ. In Tatar and Bashkir, this e (< ä, ẹ) shifted to i, but in the affixal subsystem it is represented by ä (a front variant of a). Tatar and Bashkir also developed a secondary ä from a in the environment of dorsal j, z, ẟ, ž, š, ǯ, č, ž and both ä fell together:
|ä/ä in suffixes||} ä|
|ä (vowel harmony variant of a)|
In Turkish (taking its dialects into account) there resulted ɛ and e and even ä, e, ẹ, though in the literary language ɛ and e did not develop into independent phonemes.
In Turkmen a new opposition between ɛ and äː arose, whereas earlier eː, apparently through a stage ei, gave iː. The stage e was preserved in the Khorezm dialects of Uzbek: eːr ‘early’ (~ Turkmen iːr), and in Turkmen dialects (eːr ‘early’, eːl ‘country’, beːl ‘small of the back’, geːč ‘late’).
The opposition of long and short e (ɛ vs. eː) can be found in Azeri, but now it is not quantitative (ä vs. eː) but qualitative (äl ‘hand’ vs. el ‘country’; Azeri er ‘early’, bel ‘small of the back’, geǯ ‘late’ ~ Turkmen iːr ‘country’, biːl ‘small of the back’, giːč ‘late’).
Apart from this, in Azeri (and Turkmen) a shift ä > e took place in the environment of j, attested already in ancient languages, and also in rare instances of assimilation before a following i (ä > e): Azeri jet‑ ‘arrive, reach’, jer ‘earth’, cf. Old Turkic jetirü ‘until’ and jer ‘earth’ in the Brahmi texts, Azeri ešik ‘door’, Turkmen iːšik ‘door’, but Azeri dämir ‘iron’ (Turkish demir), gämi ‘boat’, where there is no influence from i.
In Yakut the quantitative and qualitative opposition between ɛ and eː changed into an opposition between ɛ and i͜e, i.e. between a relatively short vowel and a diphthong: än ‘you’ versus i͜en ‘width’. Furthermore, there is also a dialectal variation i ~ e (is‑ ~ es‑ ‘wade’, ilt‑ ~ elt‑ ‘lead’, iliː ~ eliː ‘hand’) and ä, e > i under the influence of j (> ǯ > č > s): sir‑ ‘reject’, sit‑ ‘reach, attain’, and also i of a following syllable: tirit‑ ‘sweat’ (< tär ‘sweat’), tiriː ‘leather, hide, skin’ (< *täriɣ), diriŋ ‘deep’ (< *däriŋ), timir ‘iron’ (< *tämir).
In Chuvash ɛ is of recent origin. It is a substitution for Tatar ä in loanwords and the front variant of the wide vowel in suffixes.
In Chuvash a and i correspond to the Common Turkic phoneme e (ɛ and ẹ). Thus since a can be found instead of the mid variant of the vowel, i.e. ɛ > ä > a, and i is usually found instead of high ẹ (ẹː) or a diphthong, one could imagine that Chuvash reflects more accurately the ancient qualitative opposition between ä (äː) and e (ëː ?). For every case of a in Chuvash, at an earlier stage of Common Turkic there must have been ä (or äː), and wherever Chuvash has i earlier Common Turkic had e (eː).
In the remaining Turkic languages one must consider the qualitative opposition between ä and e to be lost and explain the various reflexes of these vowels as traces of a quantitative opposition, i.e.:
|Azeri ä < ä, e; e in many cases < äː, eː|
|Turkmen e < ä, e; iː < äː, eː|
|Yakut ä < ä, e; i͜e < äː, eː, etc.|
Analysing Chuvash examples, we find that Chuvash a reflects ä from Common Turkic eː and ä, and also from ä in some loanwords.
Turkmen giːč ‘late’ (gẹːč), Turkish geč, Azeri g’eǯ, Yakut ki͜ehä ~ Chuvash kas’; Turkmen ber‑ ‘give’, Turkish dial. beːr‑, Yakut bi͜er‑ ~ Chuvash par‑; Turkmen iŋ ‘width’, Yakut i͜en, Azeri en ~ Chuvash an; Turkmen iːn‑ ‘go down’, Azeri en‑ ~ Chuvash an‑;
Turkmen ek‑ ‘sow’, Azeri äk‑ ~ Chuvash ak‑ (cf. Hungarian eke ‘plow’ < Bulgarian); Turkmen θeθ ‘voice’, Azeri säs ~ Chuvash sas̬ə;
Turkish eš‑ ‘trot’ ~ Chuvash aš‑ (i.e. the shift e‑ > ä‑ > a‑ took place even in relatively late loans).
The reflex i in single-syllable roots in Chuvash is found instead of Common Turkic e, but also e < ä, including early and late loans:
Turkmen eẟ‑, Azeri äz‑, Tatar iz‑ ~ Chuvash ir‑ ‘crush’; Azeri jet‑ ‘arrive’, Turkmen jet‑, Tuvan čeʰt‑, Tatar ǯ́it‑, Bashkir jët‑ ~ Chuvash s’it‑; Azeri g’äl‑ ‘come’, Turkmen gel‑, Tuvan, Yakut kel‑, Tatar kil‑ ~ Chuvash kil‑;
Azeri sez‑ ‘feel’, Tatar siz‑, Bashkir hiẟ‑ ~ sis (< Tat.); Turkmen em ‘medicine’, Tat., Bashkir, Khakas im ~ Chuvash im (< Tat.); Kyrgyz, Altay, Tuvan er ‘use’, Bashkir ir (Tat. irlə̈) ~ Chuvash ir ‘use, gain’ (< Tat.).
One finds instances where Common Turkic e < ä (next to j) gives ə̈ in Chuvash (as in Bashkir): Turkmen, Turkish, Azeri jer ‘earth’, Kyrgyz ǯer, Tuvan čer, Tat. ǯ́ir, Yakut sir, Bashkir jə̈r ~ s’ə̈r ‘earth’; Turkish jen‑, Turkmen jeŋ‑, Kyrgyz ǯeŋ‑, Tat. ǯ́iŋ‑, Khakas čiŋ‑, Bashkir jə̈ŋ‑ ‘defeat’ ~ s’ə̈n‑.
In a number of Chuvash words the vowel i corresponds to Common Turkic ẹ (~ Turkmen iː, Yakut i͜e): Turkmen bäːš ‘5’, Yakut bi͜es, Turkish, Azeri beš, Tat., Bashkir biš ~ Chuvash pilə̈k; Turkmen, Yakut biːl ‘small of the back’, Azeri, Kyrgyz bel, Tat. bil, Khakas pil ~ Chuvash pilə̈k; Turkmen ir ‘early’, Azeri er ~ Chuvash ir.
According to Doerfer, the last example illustrates the assumption that Chuvash i goes back to Common Turkic e, as in Mari we find the word er ‘morning, early’ which was borrowed from Chuvash. Mari e represents an earlier stage of development (in Hill Mari e > i).
Turkic borrowings in Mari like el ‘country’ (~ ? Chuvash jal), en ‘most’, ertäš ‘go past’, pelčän ‘sow thistle (genus Sonchus)’, teŋə̈z ‘sea’, terə̈s ‘manure, fertilizer’, terke ‘plate’, keremet ‘evil spirit’, seŋäš ‘defeat’, s’erə̈p ‘heavy’ show that the raising of e > i involved words not from Ancient Chuvash but rather representing a general Turkic stock in the Middle Volga that goes back to a single source.
In iranianized Uzbek dialects we find e (narrow) and æ (a very wide variant of the vowel e).
In Uyghur, which has the so-called i-umlaut, we find wide ä and e (e and ë) secondary in origin, originating from ä and aː under the influence of a following i.
Close and open variants of e (ẹ and ɛ) are apparently found in the language of the Yenesei runic inscriptions, as e is depicted by a special grapheme. A distinction was made between these two variants also in the texts in the Brahmi script. Worth noting are the Brahmi-Azeri parallels ket‑ ‘leave, go away’ ~ g’et‑ (Turkmen gider ‘he goes out’), keŋ ‘wide’ ~ g’en (Turkmen giːŋ), ber‑ ‘give’ ~ ver‑ (Yakut bi͜er‑), beš ‘5’ ~ beš (Yakut bi͜es), el ‘tribe’ ~ el (Turkmen iːl), which confirm that in a portion of words Azeri e reflects the quantity of the Proto-Turkic vowel.
E (as a variant of ä) before and after j is found in Turkish dialects, Azeri, the Brahmi texts, Yakut, etc. Cf. e.g. Brahmi jel ‘wind’ ~ Azeri jel; Brahmi jer ‘earth’ ~ Azeri jer ~ Yakut sir; Azeri jet‑ ‘arrive’ ~ Yakut sit‑; Azeri jerik ‘cravings of a pregnant woman’ ~ Yakut sir‑ ‘reject’, etc.
In Yakut this (short) ẹ narrowed to i (sir, sit‑).
Because combinatory and positional variation of the type ä ~ ɛ ~ e and e ~ i, and thus ä ~ ɛ ~ ẹ ~ i is typical of many modern-day Turkic languages and dialects, one can assume it also for earlier stages of their development. Nonetheless one cannot neglect the rich attestations of dialect mixing, reflected in many (if not all) Turkic vowel systems, cf. e.g. the systems of Chuvash, Khakas and West Siberian Tatar dialects. Both of these factors have led (including in the literary standards) to irregular correspondences: Turkmen lit. bäːš ‘5’, dial. beš, Azeri beš (< beːš) ~ Chuvash pilə̈k; Turkmen äːr ‘man’, Azeri är, Tat., Bashkir, Khakas ir ~ Chuvash ar (on the basis of the Chuvash and Azeri forms one can reconstruct *är); Turkmen mäːẟ ‘gland’, Turkish, Kyrgyz, Kumyk bez ~ Chuvash par (Azeri väz) (on the basis of the Chuvash form one can reconstruct *bär); Turkmen gäːt‑ ‘break off, away’, Turkish get‑ (gedik), Kyrgyz ket‑, Tat. kit‑ ~ Chuvash kat‑ (on the basis of the Turkish and Chuvash forms one can reconstruct *kaːt‑). Thus Turkmen äː corresponds to Common Turkic ä and ẹ̈ː.
Also noteworthy are correspondences between Chuvash and Common Turkic: Chuvash alək ‘gate, door’ (< *äːlik), cf. Turkish, Azeri ešik (where ä > e under the influence of a following i?), Tat., Bashkir išə̈k, Khakas dial. izə̈k, Khakas ə̈zə̈k; but Turkmen iːšik (< *eːšik); Chuvash at‑ ‘do’ (< *ät‑) (cf. Turkmen eder ‘he does’), where d < t after an initial long vowel), but Azeri et‑ points to a protoform *eːt‑; Chuvash ilt‑ ‘hear’ (< *elit‑), Turkmen, Azeri ešit‑, Turkish išit‑ (e > i under the influence of i), Tat., Bashkir išə̈t‑, Yakut ihit‑; Chuvash i, Azeri e point to Common Turkic *e.
Thus an ancient qualitative opposition of ä and e is reflected in the Chuvash system, where we have a < ä and i < e. Only traces remain of a quantitative opposition ẹː > i, cf. pilə̈k ‘5’. Significantly more frequently long eː and short ä are reflected as a: kas’ ‘evening’ (< *käːč < *kẹːč); ak‑ ‘sow’ (< äk‑).
The whole Common Turkic map is tainted with subsequent dialect mixing and positional-combinatorial variation of the vowels ä, e, i.
For showing the ancient quantitative opposition of e sounds, the Turkmen and Yakut data are the most reliable: in Turkmen iː < ẹː, as a rule, corresponds to the Yakut diphthong i͜e, for example: Turkmen giːč ‘late’ ~ Yakut ki͜ehä ‘evening’, giːŋ ‘wide’ ~ ki͜eŋ (but käŋä‑ ‘widen’), iːn ‘width’ ~ i͜en (but ? äŋäj‑ ‘spread out’), iːt ‘lead’ ~ si͜et‑ ‘take by the hand, by a leash or rope’ (but sätiː ~ si͜etiː ‘leading a blind person’).
The regular nature of these correspondences is undermined by such examples as Turkmen biːl ‘small of the back’ ~ Yakut biːl (and not i͜e), bäːš (and not biːš) ‘5’ ~ bi͜es (but bähis ‘fifth’).
The alternation of i͜e ~ iː possibly arose within Yakut, cf. also Yakut iːt ‘load a rifle’ ~ ? Turkmen et‑ ‘do’ (< *eːt‑, indicated also by the d in eder ‘he does’); Yakut tiːl ‘calf or colt nursing from an unrelated female’ ~ Kyrgyz tel.
As far as the correspondence bäːš ~ Yakut bi͜es is concerned, it is well known that numerals are often characterized by phonetic peculiarities due to their function in speech, such as emphatic gemination of consonants. It is also well known that in Turkmen dialects one also encounters the phonetic variants beːš, beš. Chuvash pilə̈k and Volga Bulgarian *bielim may also attest to the length and close character of ẹ in beːl ~ beš.
Thus the materials we have examined allow us to speak with a high level of probability of the existence in Proto-Turkic of short ä and long ẹː and of the combinatorial variation of ä (ɛ, e) in different phonetic environments at a late stage of the protolanguage and at various points in the history of the modern Turkic languages all the way to the present day.
 See e.g. Scherbak 1970, 28–33, which contains a detailed analysis of almost every proposed hypothesis, and Doerfer 1971, 240–247.
 If the shift ɛ > e > i had occurred in suffixes, then the variations of some affixes, e.g. ‑di and ‑də̈, could merge, as i in open final syllables tends to be lowered. Indeed, in the Kasimov dialect ä > i even in affixes: bir8in ‘he gave’, bə̈zdi ‘on us’.
 Azeri ešik possibly goes back to *eːšik, and not *äšik, though the latter reconstruction is suggested by Chuvash alək ‘door’.
 As pointed out by D’jakovskij (1971, 98–99), the second part of the i͜e diphthong is more close than short ɛ.
 The shift ä > a occurred after the shift of Common Turkic a > ao > o. Note that in the period of Permian-Bulgarian contacts (8th–9th centuries) Chuvash still retained ä in opposition with e: ban, bam ‘cheek’, but s’i̮l ‘storm’.
 Cf. Turkmen äːr ‘man’, Azeri är ~ Chuvash ar, where the long vowel in the Turkmen word points to äː in the protolanguage, see e.g. the reconstructions of Poppe (eːr) and Doerfer (ä̂r or är).