Labiovelar loss and the rounding of syllabic liquids in Indo-Iranian

This paper analyzes and supports the claim that Vedic Sanskrit preserves traces of the contrast between the Indo-European labiovelars and plain velars—a striking archaism in the Indo-Iranian family, which otherwise collapsed the two velar series. These labiovelar vestiges emerge because of the pervasive labialization of syllabic and consonantal rhotics at all attested stages of the Indo-Iranian family. Two rhotic labialization environments are examined in Indo-Aryan and Iranian: after labial(ized) consonants or before syllables containing u or w. Furthermore, this paper explains the unexpected development of bimoraic Proto-Indo-European *L̥μHμ.C to trimoraic Vedic Ūμμrμ.C by examining the phonetic characteristics of the labializable Indo-Iranian rhotics.


Introduction
Since Burrow (1957), the question has remained unresolved whether traces of the labiovelar series, *Kw, were retained in the outcomes of the Proto-Indo-European (PIE) *KwL̥ H1 as Sanskrit Kūȓ. Because Indo-Iranian shows uniform clayton 10.1163/22125892-0100110021 | Indo-European Linguistics (2022) 1-55 satəm outcomes of inherited velar series (i.e., the PIE labiovelars merged exceptionlessly into the plain velars), only the anaptyxis of rounded vowels in *KwL̥ H sequences enabled evidence of labiovelars to survive during the prehistory of Proto-Indo-Iranian (PIIr.). Under this view, Sanskrit and Proto-Indo-Iranian stand among the handful of PIE branches which retained (partially conditioned) reflexes of all three of the PIE velar series, alongside Albanian (Pedersen 1900;Orel 2000: 66-74 with refs.), Armenian (Stempel 1994), and Anatolian (Melchert 1987(Melchert & 2012. In this paper, I will reexamine the conditioned effects of PIE labiovelars on *-L̥ H-; specifically, I will argue that: (1) a. The distinct reflexes of labiovelars appear only in closed-syllable *KwL̥ H.C-sequences. b. Adjacent rounded segments cause rounding of *L̥ (H) not only in Vedic (Ved.) but also in Iranian. c. The rounding of liquids provides insights not only into the chronology of *L̥ (H) anaptyxis but also into the phonetics of *r̥ in Proto-Indo-Iranian.
This discussion will primarily concern the effects of various rounded segments on *L̥ and anaptyctic *ə.
Though Lubotsky's theories do not directly concern the traces of *Kw, he summarizes the two environments for Ved. ūȓ adduced in the preceding literature; his rules are represented in (2).
With these forms eliminated, the remainder of this section will concern true examples of L̥ H (non-)rounding.

2.1
Vedic ūȓ from *PL̥ H The post-labial rounding environment *PL̥ H > Pūȓ6 in Ved. certainly owes its early recognition by Pāṇini to its exceptionless operation. Some representative examples of rounding are provided in (8). These forms contrast with the nonrounded forms in (9), which illustrate the unconditioned outcome of *L̥ H in Vedic.

2.2
Vedic ūȓ from *L̥ HC0w The case for rounding by a following *w is somewhat more complex than for rounding from a preceding labial; not only is it unclear whether rounding occurs when a consonant intervenes between *L̥ H and *w, but also this environment is susceptible to leveling in a way in which *PL̥ H avoids. For word-internal, non-alternating instances of *L̥ Hw, however, ūȓ appears uniformly, as shown in (10).

9
Indo-European Linguistics (2022) 1-55 | 10.1163/22125892-0100110021 The only potential candidate of *L̥ HCw in the rv is ūrdhvá-'upright' < *(w)r̥ hxdh-wó-, but the reconstruction of this term is notoriously difficult (see discussion in Schrijver 1991: 312-313).13 Hackstein (2018), however, argues cogently that the root should be reconstructed as *werhxdh-with initial *w, which would render ūrdhvá-non-probative because the Ved. u-vocalism may be attributed to the preceding *w. As such, it is difficult to say whether *L̥ HCw acts as a regular rounding environment in Vedic, and so only *L̥ Hw remains certain.

2.3
Vedic ūȓ from *L̥ HC0u In contrast to the easily adduced cases for rounding in *L̥ Hw, the evidence remains scarce whether *L̥ Hu acts similarly; indeed, this sequence faces even more difficulties than the one discussed in the preceding section. In this section and in 2.4 and 2.5, I will support *L̥ Hu as a rounding environment on the basis of indirect evidence from elsewhere in Vedic and from Middle Indo-Aryan. As for direct evidence, *L̥ Hu lacks probative Vedic data for rounding effects. The most obvious candidates, *-u-stem adjectives of the type *R(∅)-ú-given in (12), are all compromised by having a labial or labiovelar before the *L̥ Hu.
Given the uncertainty about inherited *L̥ Hu > uru, it should come as no surprise that no secure evidence exists for *L̥ HC1u. The general dearth of probative forms could argue both for Lubotsky's proposal of *u as a source of regressive rounding and against gurú-as an example of labiovelar rounding, but even that is too confident of an interpretation of these data. Before beginning the labiovelar discussion, the environments set out in (4a) may now be narrowed to (13) with certainty.

2.4
Rounding of r̥ in Middle Indo-Aryan Before I move onto the evidence for rounding caused by labiovelars, the striking continuation of r̥ -rounding in Middle Indo-Aryan deserves discussion, particularly because the earliest stages of Middle Indo-Aryan existed already at the time of the composition of the Vedas and likely represented the household dialects of their composers. As such, the phonetics of Middle Indo-Aryan can shed light on the phonetic milieu of early Sanskrit articulation. The vowel r̥ did not survive as such into Middle Indo-Aryan but was normally replaced by a, i, or u depending on the environment and the dialect.18 In fact, all periods of Middle Indo-Aryan show very similar behavior both to one another and to Sanskrit, namely that r̥ produces u-vocalism after a labial or before a syllable containing u. Some representative examples are given in (14)-(16).

2.5
Vedic ūȓ from *KwL̥ H As demonstrated in the preceding sections, the choice between Ūȓ reflexes in Vedic is by no means arbitrary, a fact which has made the distribution of Ūȓ after velars all the more baffling, in particular near minimal pairs like gurú-< *gwr̥ h2-ú-vs. girí-'mountain' < *gwr̥ hx-í-.20 This problem begs for a new solution, and to that end I propose the formulation in (17): 20 The labiovelar in *gwr̥ hx-í-is reconstructed on the basis of Alb. gur 'stone' < PAlb. *guri (Demiraj 1997: 51 (20) show consistent ıvocalism throughout, as they contain no environment in which rounding is expected.
nom.sg *gwérhx-s, acc.sg *gwérhx-m̥ , nom/voc.pl *gwérhx-es) are removed, giving 4:2 types and 148:82 tokens.31 Since -ir-~-ur-alternations are nowhere attested within a single nominal paradigm, the stem may then have been leveled to gir-on the basis of type and token frequency.32 As mentioned in Section 2.2, the first members of gír-vaṇas-'praise-desiring' and gír-vāhas-'praiseconveyed' are not attested as ×gır°in rv, indicating that the i-vocalism when in the first member of compounds is also not necessarily original. Therefore, regardless of the ablaut reconstructed for this root noun, paradigm-internal leveling can explain the outcome of gir-. Crucially, the root noun's i-vocalism may differ from the u-vocalism in the other forms of √gr̥ -'to praise; welcome' because the root noun stands on the periphery of the verbal system. Finally, √gr̥ -'to swallow' possess only three hapaxes containing outcomes of the zero-grade *gwr̥ h3-, and each requires some attention. The root noun appearing in muhur-gır ('swallowing instantaneously' , rv i.128.3) may have its ī-vocalism via paradigm-internal leveling in the same way as gír-'praise' in the preceding paragraph. The ppp gīrṇám would appear to be a counterexample to (17) as the closed syllable in *gwr̥ h3-nó-should have resulted in ×gūrṇá-. Here I propose that this late-appearing form (rv x.88.2) takes its vocalism analogically from the present stem. Though no presents appear in rv, the forms giráti (av) and gilati (b) develop exactly as expected from *gwr̥ h3-é-ti. If √gr̥ -'to swallow' selected one vocalization for its verbal paradigm as did √gr̥ -'to praise; welcome' and √jr̥ -'to grow old' , then a late Vedic ppp would level to the ivocalism of the present stem. The gir/gil-form of the stem became so standardized that later Indian grammarians report an innovative intensive stem jegilyabuilt to a new root √gil- (Schaefer 1994: 115-116). Also, as Lubotsky (2007: 233) suggests, the *-nó-participles probably spread from the *Ced-to the *CeLHroots later in Indo-Iranian-in this case after the adoption of i-vocalism in the Rigvedic dialect. Indeed, the only attested u-vocalism found in rv form of √gr̥ -'to swallow' appears in the 2sg.intens.pst.act.subj. jalgulas (i.28.1-4). This form does not appear to belong to the dialect of rv, however, having both u- 31 Grassmann (1873: 400-401) 32 One may well object that other text types may have different distributions of cases, but the hymns of rv represent the oldest material preserved and show pervasive influence on each other. If other texts or dialects preferred ū-vocalism for this noun, Vedic poets would be unlikely to import them against the uniform ı-vocalism (that may have overcome an original *á-~∅-ablaut) only to create an -ır-~-ūȓ-alternation otherwise unattested in the nominal system. Furthermore, because this noun is unlikely to appear in the dual with any appreciable frequency, the only pre-consonantal zero-grades in an ablauting paradigm would come from a subset of the oblique plurals (*-bhis, *-bhyas, *-su), an improbable source for paradigm-wide leveling. clayton 10.1163/22125892-0100110021 | Indo-European Linguistics (2022) 1-55 vocalism and l-consonantism found nowhere else within the verbal paradigm. Instead, jalgulas resembles the intensive -jalgulīti found in the ts. Thus, all the forms of √gr̥ -'to swallow' in the rv may be explained by assuming the rv adopted i-vocalism throughout its verbal paradigm and that jalgulas was borrowed from a different dialect which regularized u-vocalism. Note further that jalgulas appears in a hymn of racy and popular character and recurs 4 times as the last word of the verse-final refrain ulūḱhalasutānām ávéd u indra jalgulaḥ 'you, Indra, will keep gulping down the mortar-pressed (soma drops)' (tr. Jamison & Brereton 2014: 126-128). The striking u's and l's of jalgulas may have phonetically mirrored the refrain's first word, ulūḱhala-suta-'mortar-pressed' , especially since the hymn generally concerns the mortar (ulūḱhala-).
In the above section, I have shown that all the outcomes of the velar-initial L̥ H may be explained by applying a simple phonological generalization, (17), with the modification that a given Vedic dialect selected one vocalism for all its verbal forms. Crucially, under this analysis, all * ⁽ ⁾ Ḱ-initial forms show only ı-vocalism, and isolated nominal forms girí-< *gwr̥ hx-í-and gír-< *gwŕ̥ hx-no longer act as exceptions.
Synchronically, the best potential evidence of r̥ -rounding appears in the nasal-infix present ⟨ku-u-n-u-t-i-y⟩ ku(r)nautiy 3sg.prs.act.ind < PIr. *kr̥ náwti of the verb kar-'do, make' < PIE *kwer-. Kent (1942) argues against an older view that r̥ > u / ___n, citing forms like ⟨v-r-n-v-a-t-i-y⟩ vr̥ navataiy 'he believes' 3sg.prs.mid.ind < *(hx)wr̥ -naw-a-tay and ⟨k-r-nu-u-v-k-a⟩ kr̥ nuvakā 'stonecutters' nom.pl < PIr. *kr̥ t-nu-ak-āh from *kart-'to cut' , which do not show the change. He instead prefers to take ku(r)nautiy's u-vocalism by analogy to tunu-'to be strong' < PIr. *tu-ná-H-and the reconstructed OP form *cunautiy 'he hears' 3sg.prs.act.ind < *ćru-naw-ti, which he reconstructs on the basis of YAv. surunaoiti 'he hears' and Modern Persian šonidan 'to hear' . Yet these Iranian comparisons require that the nasal-present *ḱl ̥-né-w-ti have been reformed to *ćru-náw-ti in Proto-Iranian, since Vedic has the expected form śr̥ ṇóti.40 Hoffmann & Forssman (2004: 52) suggest the development of YAv. surunaoiti only occurred in Avestan by analogy with forms like the YAv. ppp sruta-'heard' , but the Modern Persian šonidan leaves open the possibility that *ćru-naw-ti was an earlier innovation. If Iranian did have rounded *r̥ w as I suggest, both the odd shape of PIr. *ćrunáwti and OP ku(r)nautiy could be explained by misperception by speakers of a rounded *r̥ w caused by the *-na-w-~*-n-u-. The effects of the rounding cue in *Cr̥ w-n-u-nasal-infix presents would have further been augmented by analogy with other Iranian Cu-n-u-nasal-infix presents. In this way, the analogical account proposed by Kent and the phonological account of this article may function together to produce the Iranian u-vocalism found in nasal- 40 The claim by Kent (1942: 80) that Sanskrit rebuilt śr̥ ṇóti by dissimilation of u from *śruṇóti is unlikely. clayton 10.1163/22125892-0100110021 | Indo-European Linguistics (2022) 1-55 infix presents. Furthermore, the OP reflex of *kr̥ naw-/kr̥ nu-closely mirrors not only the change from Ved. *kr̥ ṇu-/kr̥ ṇv-> Skt. kuru-/kurv-discussed in section 2.3, but also the change from PIr. *kr̥ naw-/kr̥ nu-> Sogd. kwn-to be discussed in section 3.6. The diachronic evidence of MP rounded ur reflexes of OP r̥ after labials and synchronic evidence of intrusive u in *-naw-presents argues strongly in favor of OP having an allophonic rounded r̥ w in the environments P___ and ___Cu.

Pashto
The situation in Pashto is opaque due to a regular process that lowered stressed *ı̄and *ū̆to ə́in open syllables (Cheung 2011: 199). Nevertheless, Cheung describes how PIr. *r̥ has several different outcomes depending on Pashto stress, syllable structure, and i-and u-umlaut (Cheung 2011: 187-188 & 196-197) The traditional etymology given for the name 'Ormur' comes from a Pashto compound orməṛ-[orməɻ] 'extinguisher of fire' < PIr. *āθr-'fire' + *mr̥ -'to die' , supposedly referring to the Ormuri people's fire-worship (Efimov 2011: 1). The fact that this term is borrowed with u-vocalism could mean either that Ormuri's r-rounding process was productive quite late into the language's phonology or that Pashto had a rounded *-muṛ-which was obscured by the *ū̆> ə lowering rule described in section 3.5. proto-forms should be interpreted as containing u in the syllable following *r̥ . For (27a), he explains the various root vowels by proposing murγə́< *múrγu < *mr̥ γu < *mr̥ gám acc.sg and taking the other forms from the other cases. Such an appeal to the effects of the case endings, however, would not work for (27b). In analyzing the data, Cheung missed a possible generalization: all of the forms in (27) begin with a labial consonant. Under his analysis, the lowering of stressed *ı̄and *ū̆to ə́must have taken place after the after *r̥ produced *ir and *ur. A parsimonious solution to this data would be to say that *r̥ became *ur after labials and *ir elsewhere. Then, after the application of syncope rules *u and i were only retained in closed syllables.44 The explanation using P___ does a much better job with (27), but it unfortunately leaves ___C0u uncertain as a *r̥ -rounding environment. Alternatively following Cheung's u-umlaut solution, *r̥ would be rounded in ___C0u but not necessarily in P___. I prefer P___ as the correct environment, but either solution leaves Pashto with clear evidence for *r̥ -rounding.  Gauthiot (1913: 94-95) already suggested that the *-u-in the nasal suffix was the culprit, saying: 44 The vacillations in (27a) could potentially be due to lowering of vowels before r in unstressed syllables, but Cheung  Il est bien difficile de ne pas voir dans la présence du morphème -nau-: -nu-le point de départ de la naissance du timbre de la voyelle -u-qui s'est substituée à -r̥ -dans la syllabe radicale; en effet le thème du participe passé, dépourvu du morphème -nau-: -nu-, oppose régulièrement *-r-, ou ses représentants normaux, à l'-u-du présent.

Sogdian-Yaghnobi
The fact that both Sogdian-Yaghnobi and Old Persian undergo this exact same change from *kr̥ -nu-> kun-implies that ___Cu was indeed a rounding environment in both languages for the same reasons discussed in section 3.2.46

3.11
The situation in Proto-Iranian The variety of different rounding effects across the Iranian daughter languages indicates that there was no unitary development of rounded *r̥ in Proto-Iranian, but the evidence (summarized in Table 1) shows that *r̥ was certainly phonetically rounded after labial consonants, and probably also before syllables containing *u or *w.
While the above is far from an exhaustive survey, I believe this nontrivial correspondence between Iranian languages (and Sanskrit) to show that Proto-Iranian *r and *r̥ were highly susceptible to the spreading of rounding gestures from adjacent rounded vowels and labial consonants.47 In my investigation of Iranian, nowhere did I find evidence of rounding caused by labiovelars, though very few probative examples of *KwL̥ H.C ap- 47 It is striking that if we were certain that PIE *wr̥ hxdhwó-> PIr. *(w)r̥ dwá-'upright' had lost its initial *w by dissimilation already in Proto-Iranian, several languages would have *r̥ C0w confirmed as a rounding environment. As noted above, this word requires future study.

The diachrony of *L̥ H in Proto-Indo-Iranian
In the previous sections, I have shown the environments for rounding of PIE *L̥ (H) both in Indo-Aryan and Iranian. The formal identity between three of the rounding environments P___ and ___C0{u, w} across both Indo-Aryan and Iranian leaves little doubt that this rounding is an inherited feature. Yet we still need a theoretical discussion of *L̥ -rounding in Indo-Iranian. In the following section, I will describe the development of PIE *L̥ H in Indo-Iranian. Next, I will propose several potential explanations why labiovelars rounded *r̥ in Indo-Aryan.

The development of PIE *L̥ H in Indo-Iranian
In the preceding discussion, most data has been presented from a bottomup perspective, showing that the evidence in the daughter languages supports reconstructing rounding effects on *r̥ in Indo-Iranian. To get a better understanding of the situation in Proto-Indo-Iranian, we must examine the development of PIE *L̥ H into Proto-Indo-Iranian and its descendants. The development of PIE *CL̥ C > PIIr. *Cr̥ C gives little difficulty; between consonants, PIIr. *r̥ does not change in Indo-Aryan and Iranian.49 Take, for example, PIE *kwr̥ -tó-'done; made' > PIIr. / PIA / PIr. *kr̥ -tá-> Ved. kr̥ tá-, Av. kər ə ta-, OP ⟨k-r-t-⟩ kr̥ ta-. PIE *L̥ H, on the other hand, shows more interesting behavior. Before both vowels and consonants, PIIr. *r̥ H produces an anaptyctic vowel before the liquid, as seen in (28). 48 It is also possible that the palatalization of the following *-ya-blocked any rounding effects inherited from the labiovelar, but there is no way to know. 49 When either of the consonants in PIIr. *Cr̥ C is *y or *w, the outcomes become somewhat more complicated (see for instance Lubotsky 1997: 148-149 andByrd 2015: 142-143 The example (28a) shows a very interesting problem, namely that PIE *L̥ HC results in a superheavy syllable *Ūr.C in Indo-Aryan, but a heavy syllable *ar.C in Iranian. This inter-branch discrepancy in syllable weights has never been adequately explained to my mind. Some scholars (e.g., Gamkrelidze & Ivanov 1995: 176-178) have sought to explain the weight discrepancy by continuing the pre-laryngealist practice of reconstructing a 'long resonant' phase of Indo-European and Indo-Iranian (*R̥ H > *R̥). Yet the theory of long resonants does not in itself explain the data so much as hide the problem behind a series of phonemes (*m̥ < *m̥ H, *n̥ < *n̥ H, *r̥ < *r̥ H, *l̥ < *l ̥H) that do not survive in any of the daughter languages. While scholars often reconstruct unattested intermediate phonemes, the long resonants provide no explanation of why long liquids and nasals behave differently (*L̥̄ > *Vr but *N̥ > *ā), why the height of epenthetic vowels before liquids differs (*L̥̄ > PIA *Ūȓ but > PIr. *ar), why *L̥̄ .C becomes superheavy *Ūr.C in Indo-Aryan, but not in Iranian, or why *L̥̄ always results in a prevocalized *Vr in all branches of Indo-Iranian.
To achieve a proper explanation for the development of PIE *R̥ H in Indo-Iranian, we must propose theories which predict the specific outcomes in the daughter languages. Because of the uniform appearance of the vowel before the liquid, Kümmel (2017: 9) and Cantera (2017: 489) propose the rule in (29). Under a moraic approach, the prevocalization of *L̥ H actually produced a moraic increase (bimoraic PIE *L̥ μHμ.C > trimoraic PIIr. *əμrμHμ.C), a step on the way to the trimoraic weight of PIA *Ūμμrμ.C. This moraic increase may just be a side effect of Proto-Indo-Iranian no longer allowing syllabic liquids before laryngeals; the split of *L̥ into two segments, *ə and *r, happened to increase the Indo-European Linguistics (2022) 1-55 | 10.1163/22125892-0100110021 morae by one.50 One might complain that the addition of an extra mora by (29) would be part of a 'Duke of York' sound change from an Iranian perspective: bimoraic PIE *L̥ μHμ.C becomes trimoraic PIIr. *əμrμHμ.C becomes bimoraic PIr. *aμrμ.C again. Iranian, however, has other evidence for the deletion of coda laryngeals from PIIr. *VrH.C-(30a) and *VNH.C-(30b) syllables.
In this way, the theory that PIE *L̥ H.C > PIIr. *ərH.C easily explains the facts of Iranian, but the details of how PIA *Ūr.C occurred require further discussion. As stated above, the PIIr. reconstruction *ərH.C has the advantage of being trimoraic like PIA *Ūr.C, but no account exists for why *ə should raise to *U before *r or how *U became lengthened. Even in the extensive discussion of the distribution of Ved. CUrWV and CŪrWV from PIE *CL̥ HWV by Lubotsky (1997), nowhere does he propose explanations for the height or weight of *CŪŕWV except to say that Ved. Ū́occurs when accented. Regarding only the question of length, he proposes that even inherited PIE *ÚrW lengthened under accent to Ved. ŪŕW, but each of his examples has complications (pp. 143-144). All of this suggests that Lubotsky thinks the Vedic pitch accent also had a durational component which affected only *UrW sequences, but he never explicitly states as much. Since I know of no other Vedic examples of vowel lengthening under pitch accent, such a phonetic motivation seems unlikely, especially because cases of *L̥ H.C where *C ≠ *W show consistent Ūr.C outcomes regardless of accentuation (e.g., PIE *dl ̥h1ghó-> Ved. dīrghá-).52 Therefore, I will discuss what methods can be used to explain the problems of *ə raising and lengthening in Indo-Aryan.

4.2
The 'crossed lines' account of PIE *L̥ H.C > PIA *Ūr.C At first glance, the easiest way to realign the morae of *əμrμHμ.C to *Ūμμrμ.C would be to delete the laryngeal and reassign the mora to the vowel, as shown in (32). When the syllable-final laryngeal deletes, its mora 'leaps over' the intervening liquid to lengthen the preceding vowel.53 Yet this simple explanation faces several theoretical hurdles. First, any explanation which involves the crossing of autosegmental association lines runs afoul of the Well-Formedness Condition (33).

4.3
The 'phonemic geminate' account of PIE *L̥ H.C > PIA *Ūr.C If we cannot simply reassign the laryngeal's mora directly to the preceding vowel as in (32), we might try proposing that *rH temporarily became a geminate *rr before degeminating with compensatory lengthening due to the Sanskrit proscription on geminate r. A schematic version of this account is shown in (34).
Sanskrit has an initially attractive parallel for the creation and immediate loss of geminate *rr with compensatory lengthening, as when /r/ and /s/ in ruki contexts delete with lengthening of the preceding vowel before r (e.g., pátis + rayīṇāḿ (→ *pátir rayīṇāḿ) → pátī rayīṇāḿ; Kobayashi 2004: 99-100). This external sandhi rule (/Us#r/ → [Ū#r]) is not a good parallel for *ərH.C > *Ūr.C precisely because the sandhi rule functions heteromorphemically and heterosyllabically. Since sandhi applied to synchronically derived sequences, no temporary geminate r phase must have existed: the underlying s following a high vowel would normally become r before a voiced consonant (since z and ž were not phonemes in Sanskrit), but because a geminate r would result, the underlying s was instead deleted and its mora transferred to the preceding vowel. At no point in this synchronic process did the illicit *-r r-necessarily exist. Even if this geminate did exist at some phase in the prehistory of Sanskrit, it would have appeared heterosyllabically. Keydana (2014: 277-278) argues in his discussion of Szemerényi's Law that the explanations that analyze **ph2tér-s as going through an intermediate geminate phase **ph2térr fail because there is no evidence that Proto-Indo-European ever allowed tautosyllabic geminates in codas.

4.4
The 'phonetic geminate' account of PIE *L̥ H.C > PIA *Ūr.C If, however, we set aside the idea of a phonemic geminate resulting from the unmotivated assimilation of *UμrμHμ]σ > *Uμrμrμ]σ, Present Day General American English provides a typological parallel for phonetic gemination of coda r and l. When English monosyllables contain a high tense vowel or diphthong followed by a coda liquid, speakers frequently disagree whether the resulting word is mono-or disyllabic, an effect which Lavoie & Cohn (1999)  If the Indo-Iranian *r also involved a dorsal constriction, a similar explanation could be used to explain the lengthening of PIIr. *ərH.C > PIA *Ūr.C. When PIIr. *ə raised to a short high vowel in pre-PIA, the coda *r would prevent its dorsal gesture from overlapping the preceding high vowel and thus become phonetically bimoraic. The laryngeal, now preceded by three morae and followed by another consonant, would then be deleted by stray erasure since extrasyllabicity is not possible word-medially (PIIr. *əμrμHμ.C > **Uμrμμ⟨H⟩.C > *Uμrμμ.C; for discussion of similar laryngeal deletions in Indo-European, see Byrd 2015: 105-123). Yet simply following the r of an already trimoraic syllable cannot have been sufficient to delete a laryngeal, as seen in the -iṣ-aorists where the laryngeal is not deleted (*h1é-tērh2-s-t > *á.tāμμrμHst > *á.tā.riHst » Ved. átārīt 'overcome' 3sg.aor.act.ind). Since the PIIr. laryngeal likely had a dorsal articulation, the preceding bimoraic, dorsal *r could have led to a dissimilatory loss of the laryncharacter [Ψ], but Ball (2017)  geal (*Uμrμμ⟨H⟩.C > *Uμrμμ.C) not found after monomoraic *r (*āμ(μ)rμH.C > *āμ(μ)rμ.Hi.C or *āμ(μ)rμ.iH.C). Crucially for this Indo-Aryan account (just as in English), the extra mora of *Uμrμμ.C is never phonemic but merely a phonetic effect when competing dorsal gestures in high vowels and coda liquids are not allowed to overlap. The morae would then represent, in sequence, the dorsal gesture of the high vowel, the dorsal gesture of the coda *r, and the coronal gesture of the coda *r. After the laryngeal was lost, the *Uμrμμ.C underwent a quantitative metathesis to *Ūμμrμ.C, giving the Indo-Aryan outcome. This quantitative metathesis would result from speakers reassigning the phonetic extra mora of the coda *r to the preceding vowel. English does not have a parallel to the quantitative metathesis *Uμrμμ.C > *Ūμμrμ.C, but this is unsurprising as English high tense vowels are already bimoraic, a fact which precipitated the repair of quadrimoraic ×[piːμμɹμμ] to [pıμɹμμ] or [piːμμ.ɹ̩ μ]. The whole Indo-Aryan process is shown in (36).
Thus, when Indo-Aryan raised the epenthetic *ə, the resulting high vowel conditioned the phonetically bimoraic *r.56 For this analysis to work, much depends on the Indo-Iranian *r having dorsal articulation similar to General American English [ɹ], which is exactly what will be argued in Section 4.7. For the moment though, let us assume that Indo-Iranian *r had the requisite characteristics to allow the lengthening shown in (36). Then the development of PIE *dl ̥h1ghó-'tall' to Ved. dīrghá-and OAv. dar ə ga-may be summarized in (37).

56
Note that under this analysis, the same lengthening of PIIr. *Ur.C > PIA *Ūr.C should occur even without a laryngeal, a theory recently explored and confirmed by Clayton (2022

4.5
The rounding environments of PIIr. *r̥ H We still require an account of why the vowels produced before *r rounded in labial environments. The crucial phonological insight to explaining this phenomenon in Indo-Iranian is the fact that PIIr. *r̥ was phonetically rounded in the environments discussed above: Most relevantly, the vocalization of laryngeals produced a high vowel, ı, in Sanskrit just as *r̥ did, but the vocalized laryngeals never show rounding of the anaptyctic vowel (e.g., Ved. pitár-'father' < PIE *ph2tér-, not ×putár- ;Lubotsky 2018Lubotsky : 1882Lubotsky -1883. Likewise, *N̥ (H) comes out as *ā̆in Indo-Iranian even in the rounding environments reconstructed for *r̥ (e.g., Ved. pūrvajāvarī-'born of old' f < PIE *pr̥ h2-wo-ǵn̥ h1-wer-ih2-, not ×pūrvajūvarī).57 Because anaptyctic vowels before other segments do not show variable roundedness, it stands to reason that some particular property of *r played a role. Therefore, I propose that the lip rounding of a preceding labial or following *w/*u spread to the *r̥ and then to its anaptyctic vowel. These two effects are shown in (39) and (40) There is some evidence that the development of *N̥ HC in Indo-Iranian was not uniform. The outcomes *n̥ HC and *m̥ HC have traditionally been unified in the single outcome *ā (Wackernagel 1896: 14-17;Forssman 1986;Hoffmann & Forssman 2004: 70). Kobayashi (2004: 95-97), however, follows the view that *m̥ HC becomes ām in Vedic, unlike *n̥ HC > ā: -*ḱremh2-> Ved. √ śrami-'be weary': śrāḿyati < *ḱrm̥ h2-yé-ti, śrāntá-< *ḱrm̥ h2-tó- √vam-'vomit':°vaṇtīm < *wm̥ h1-tí-m Kobayashi postulates the following account for the differing outcomes of *n̥ and *m̥ in PIIr.: *N̥ epenthesized a vowel before it, [*ṼN]. For *n̥ , the nasalization cue became reinterpreted as part of the vowel, leading to the loss of the nasal stop and resulting in *Ṽ. *Ṽ was then denasalized to a (cf. Greek *N̥ > α). For *m̥ , however, the lip rounding cue was perceptible enough to require a labial segment to motivate it, and so *m remained. Thus, *n̥ HC > *ṼnHC > *ṼHC > *āC, but *m̥ HC > *ṼmHC > *āmC. Crucially, however, it is the phonetics of the labial *m segment that causes the differing outcome, not the phonetics of the other adjacent segments. Yet, the change *ṼmHC > *āmC, with the laryngeal compensatorily lengthening the vowel at a distance over the intervening m, would run afoul of the same crossed lines and phonemic geminate issues mentioned in Sections 4.2-4.3. Overall, the lautgesetzlich Vedic outcome of *m̥ HC remains problematic. In the data presented for Indo-Aryan and Iranian, it is striking that rounding from a preceding labial functions across the board, while rounding from a following *w/u functions less consistently. A possible explanation is shown in (40), where the spread of lip rounding passes through any intervening consonants, phonetically rounding them in the process. Depending on the particular phase of PIIr. developing into its daughters, the ability of particular intervening consonants to be rounded may vary.58 Contrast this with (39), where, at any phase, the lip rounding need only travel to the immediately adjacent *r̥ without passing through any other segments which could potentially block the effect. As such, the model of roundedness feature spreading predicts the stability of P___ and the instability of ___C0{w, u} as rounding environments. Under this view, the development of PIE *pl ̥h1-nó-into Ved. pūrṇá-'full' proceeds as in (41). (41)

4.6
The development of PIE *KwL̥ H.C-> Ved. Kūr.C-With the analysis of the development of *L̥ H described above, the retention of the effects of labiovelars has yet to be explained. There are three main hypotheses, (42), about how labiovelars could be preserved into Vedic such that PIE *KwL̥ H.C-> Ved. Kūr.C-but PIE *KwL̥ .HV-> Ved. Ki.rV -.
(42) Hypotheses for retention of rounding in *KwL̥ H.C-: a. *r̥ split into two phonemes, *r̥ and *r̥ w, in Proto-Indo-Iranian b. *H split into two phonemes, *H and *Hw, in Proto-Indo-Iranian c. Before *ərH]σ, *Kw did not merge with *K until Proto-Indo-Aryan All of the hypotheses in (42)  the simplest of these is Hypothesis (42a). Yet the notion that Proto-Indo-Iranian would gain a short-lived *r̥ w phoneme which then became phonetic in all the daughter languages is not a parsimonious solution. Furthermore, phonemic labiality contrasts in rhotics are in themselves uncommon. Looking at rhotics using the cross-linguistic database of phonological inventories PHOIBLE (Moran & McCloy 2019), a clear trend emerges. In the languages which contrast coronal rhotics with their rounded variants, /rw/ appeared 17 times, /ɻw/ 10 times, and /ɾw/ 4 times. Each of these languages, however, also has a full series of contrastive labialized coronal stops (/tw/, /dw/, etc.). Since no archaic Indo-Iranian language contrasts rounded and unrounded coronal stops, proposing a labiality contrast in coronal rhotics seems typologically implausible. On the other hand, the dorsal rhotic /ʁw/ occurs in 20 languages, but in 12 of them, it appears as part of series of labialized dorsals.59 The potential for a dorsal place of articulation of PIIr. *r̥ will become important in section 4.7. For the purposes of Hypothesis (42a), however, I see no reason to propose that Indo-Iranian lost its velar-labiovelar contrast while transferring it to a marginal phoneme *r̥ w. In particular, it is unclear why *r̥ would not become *r̥ w in open syllables as well (PIE *gwr̥ hx-éh1 > *gr̥ w.HáH > ×gurā́instead of girā' praise' ins.sg).
Another segment which could become the anchor for a [+labial] feature is the labialized laryngeal *Hw of Hypothesis (42b). Indeed, others have proposed that Proto-Indo-Iranian had the contrast between *H and *Hw before. Khoshsirat & Byrd (2018) and Khoshsirat (2018) argue that the Gilaki causative in -bēand the Vedic causative in -āpaya-could go back to the sequence PIE *-oH-éye-> pre-PIIr. *-oHwéye-> PIIr. *-āHwáya-*/-aːʍaja-/ > Ved. -āpáya-, Gil. -bē-. In support of their proposal, they provide a possible typological parallel for *H > *Hw / o___, in which *-óHe# produces Ved. -au (PIE *dedóh3-e > Ved. dadáu 'gave' 3sg.prf.act.ind; Jasanoff 2003: 61-62). For Khoshsirat & Byrd's argument to work, the *Hw must have become phonemic in Proto-Indo-Iranian when *o lost its [+round] feature when it became *ā. The reconstruction of a phonemic PIIr. *Hw would provide perhaps the most elegant solution to PIE *KwL̥ H.C-> Ved. Kūr.C-. When PIE *KwL̥ H appeared in a closed syllable, *KwL̥ H.C-, the tautosyllabic laryngeal would labialized by the preceding phonetically rounded *rw and then provide a new anchor for the [+round] feature.60 On the other hand, when 59 When taking the data from PHOIBLE, I collapsed entries for different dialects of the same language in an attempt to get a more representative count. 60 For this hypothesis, I see no way of determining whether the creation of phonemic *Hw occurred chronologically before or after *r̥ > ər. If it occurred after, rounding would have to spread through the intervening *ə to *r and then to *H. If it occurred before, then the clayton 10.1163/22125892-0100110021 | Indo-European Linguistics (2022) 1-55 While the previous proposal does get the distribution very nicely, it encounters two drawbacks. First, the labialization of *H would only occur tautosyllabically after *r̥ w, but heterosyllabically after *o (PIE *-o.Hé.ye-> PIIr. *-ā.Hwá.ya-). Secondly, there is much to be worked out in Khoshsirat & Byrd's account of the causatives in *-āHwáya-, which leaves the support for phonemic PIIr. *Hw uncertain. Neither drawback kills Hypothesis (42b), but further investigation of PIIr. *Hw is required.
Finally, Hypothesis (42c) simply allows labiovelars to stick around in one very limited environment. Fortson (2010: 212) has already proposed this idea: Particularly interesting are words like gurú-'heavy' < *gwr̥ h2-u-, where the labialization that induced the u-quality was that of the preceding labiovelar. In other words, at the time of the split of *r̥ H or its immediate descendant into ir and ur, which happened only in Indic (Iranian has a different outcome), the labiovelars were still distinct from the plain velars in at least this environment.
The preceding discussion has shown that gurú-is not a reliable witness of its labiovelar (section 2.3), yet Fortson's suggestion about the conditioned retention of labiovelars is a good one. Since *r̥ must have been phonetically rounded in labial environments already in Proto-Indo-Iranian, the roundedness of the [*r̥ w] could have served to 'reinforce' the labiality of the labiovelars after they disappeared in other environments. Indeed, there is a parallel type of retention found elsewhere in Indo-European. In Luvo-Lycian, the palatal velars collapse together with the plain velars except when followed by a fronting environment (Melchert 2012). The examples are mirrored: Luvo-Lycian marginally preserves palatovelars in a centum language while Sanskrit marginally preserves labiovelars in a satəm language, but conversely the palatovelars collapse with the plain diphthongization would need to occur in front of both *H and *Hw. Since neither of these two situations seems particularly problematic, the matter remains unsettled. velars in the other Anatolian languages just as the labiovelars collapse with the plain velars in Iranian. Philologically, the remnant labiovelars of Indo-Aryan would have escaped detection because they left traces in the vocalism of *L̥ H, while remnant Luvo-Lycian palatovelars appeared as alterations to the consonants themselves.61 If adopting the conditioned labiovelar retention theory, the question remains why they would only be retained in closed *KwL̥ H.C-syllables. A possible solution comes from gestural overlap. Onset consonants are known to have heavy gestural overlap with their nuclear vowels (Browman & Goldstein 1988: 98); in the case of labiovelars, this means the lip rounding gesture of the labiovelar 'spills over' into the following *-ər-sequence. Indo-European has another example where different levels of gestural overlap cause different featural outcomes in vowels. Specifically, Vine (2002: 293-295) proposes that Eichner's Law, the observation that *h2 and *h3 fail to color long *ē, resulted from a failure of the laryngeal coloring gestures to overlap the *ē completely.62 In a similar way, it is possible that the labiovelar's lip rounding gesture failed to overlap an open *Kwər.H-syllable fully. In closed syllables like *KwərH.C-, however, each of the three morae would be commensurately shorter as also found in Hungarian, where the initial vowel of VCC rimes shows the shortest duration of any syllable shape (Cohn 2003: 85-87). In Indo-Iranian, the inherited labiovelar would only be reinforced when its lip rounding gesture fully covered not only the intervening *ə but the following *r.
A second, more promising approach to the conditioned labiovelar retention theory appears in the typology of coda liquids. Gick et al. (2006) conducted an articulatory survey of liquids using ultrasound to map the timings of labial, coronal, and dorsal gestures, and they found that all languages surveyed (Western Canadian English, Quebec French, Serbo-Croatian, Korean, Beijing Mandarin, and Squamish Salish) showed dorsal constriction in coda liquids regardless of the liquids' phonemic velarity or palatality. Furthermore, the dorsal ges-

61
I would like to clarify that I do not think the labiovelars were preserved in all other environments. Parsimony suggests that before palatalizing environments *K and *Kw had already merged in Indo-Iranian. It is conceivable that labiovelars were also preserved in other rounding environments, namely before *-u-and *-w-, but that no difference was shown in the already round vowel. 62 The formulation of Eichner's Law is confusing in terms of the gestural explanation. As an artifact of the history of its discovery, the law is stated negatively: laryngeal coloring fails to affect *ē. From the perspective of the gestural overlap explanation, it would be better to formulate the rule inversely: laryngeals only succeed at coloring in a limited environment, namely when *e is short. Thinking of Eichner's observation in this way may be useful for considering the situation in Vedic. clayton 10.1163/22125892-0100110021 | Indo-European Linguistics (2022) 1-55 tures of the coda liquids frequently occur before the coronal gestures. In onset liquids by contrast, not all languages showed dorsal gestures, and those dorsal gestures that did occur were not significantly before or after the coronal gestures. Assuming that Indo-Iranian behaved similarly and had a dorsal gesture associated with coda *r but not with onset *r, the preservations of the labiovelar rounding gesture could stem from differing syllabifications. Before another consonant, the leading dorsal gesture of coda *r in *Kər⟨H⟩.C-would allow the rounding gesture to spread over the entire nucleus of the syllable, preserving the rounding until Indo-Aryan. Before a vowel, the onset laryngeal could have been deleted early, resulting in the *r resyllabifying into the onset and losing its dorsal gesture. The presence of the dorsal gesture in coda *r would play a crucial role in retaining the rounding gesture, as schematized in Table 3. Following the discussion above, I am inclined to support Hypothesis (42c) since it has a nice typological parallel in Luvo-Lycian. Furthermore the next section will show that there is good reason to believe that Vedic coda and syllabic rhotics had dorsal articulation. While Hypothesis (42b) works very elegantly by providing a new rounded phoneme, *Hw, to take up the standard from the disappearing labiovelars, the existence of *Hw remains highly speculative. Yet two plausible explanations for PIE *KwL̥ H.C-> Ved. Kūr.C-definitely provide sufficient theoretical support for the Vedic data itself.

4.7
The phonological character of PIIr. *r Throughout the preceding discussion, I have been using the traditional symbol *r to represent all the rhotics in the Indo-Iranian languages, but it is implausible to assume that *r represented the apical trill [r] of the IPA throughout the development of Indo-Iranian. I shall remain agnostic about when the PIE *r gained the retroflex /ɻ/ articulation assumed by Ryan (2017) among others. As discussed above, the presence of a dorsal gesture would be sufficient for Indo-Iranian coda *r to serve as a host for a lip rounding gesture in Indo-Iranian and to cause lengthening of a preceding high vowel in Indo-Aryan. Fortunately, the native Sanskrit grammatical tradition provides evidence for the presence of a dorsal gesture. The earliest grammarians describe Vedic r̥ not as retroflex but as jihvā-mūlīya 'produced at the root of the tongue' (e.g., R̥ gveda-Prātiśākhya 1.41; Vājasaneyī-Prātiśākhya 1.65) and as using the jaws and with the tongue tip near the upper back gums (Taittirīya-Prātiśākhya 2.18; Deshpande 1979: 282;Hock 1991: 124-128).63 On the other hand, the Prātiśākhyas describe consonantal r as having a (post)dental place of articulation different from syllabic r̥ 's velar articulation, another comparable feature to English phonology. Walker & Proctor (2019: 6-9) note that, while General American English coda liquids have their dorsal tongue body constriction sequenced before their palatal tongue tip constriction, onset rhotics reach the pharyngeal and palatal constrictions synchronously and onset laterals have the palatal constriction first (which agrees with the findings of Gick et al. 2006). Thus, the differing characterizations of the Sanskrit syllabic and consonantal liquids might represent the grammarians prioritizing whichever articulatory gesture came sequentially first: for consonantal liquids, which would be most commonly described in onsets, the (post)dental target would be reached first, but for syllabic liquids, which would behave more like coda liquids, the dorsal target would be reached first. The 63 The most explicit discussion of the Prātiśākhyas' frequent and unexpected description of r̥ and l ̥ as 'velar' comes from Hock (1992), who explains away this issue as an accident of evolving phonetic theory. Several of the Prātiśākhyas describe r̥ as equivalent to the postdental or alveolar r surrounded by two quarter-moraic a vowels (r̥ = ara = [ ə r ə ] = a 4 + r 2 + a 4 ). Hock proposes that, in the earlier phonetic tradition (represented by the Taittirīya-Prātiśākhya), the a vowel would be considered velar by default since it lacked labial, dental, retroflex, or palatal coloring. He then describes the rise of a new 'glottalic theory' (for instance, in the R̥ gveda-Prātiśākhya), in which a, h, and ḥ were recategorized as glottal. Because of the phonetic descriptions of their surrounding a vowels, r̥ and l ̥ received a velar classification that went unamended when a was reclassified as glottal.
Yet Hock admits the speculative nature of his account, as no treatise preserves the alleged diachronic stage in which a is classified as velar. Furthermore, it seems plausible to me that the surrounding quarter-moraic a vowels described in the treatises could be transitional vocoids perceived on the way to the rhotic articulatory targets. Indeed, the Taittirīya-Prātiśākhya 2.18 describes r̥ as articulated with the jaws close together and the tongue tip near the upper back gums ( jihvāgram … barsveṣūpasaṃharati), which sounds like a rhotic approximant, not an occlusive, meaning the rhotic target could serve as the syllabic nucleus and not as a consonant as Hock seems to prefer. Also, the R̥ gveda-Prātiśākhya 14.38 (& 46) proscribes labial (or palatal) pronunciations of the quartermoraic vowels surrounding r̥ , which shows that r̥ could be labialized (and palatalized) in the relevant contexts (as this paper has demonstrated as an Indo-Iranian-wide feature). The fact that the R̥ gveda-Prātiśākhya explicitly prohibits these coarticulations may indicate that the ara [ ə r ə ] pronunciation of the treatises was overly mathematical and idealized. Indeed, Whitney (1862: i. 37.) doubts the entire notion of ara, saying 'The Tâitt. Pr. does not, any more than the Rik Pr. in the earlier and more genuine part of its text, take any notice of the presence of heterogeneous elements in the ṛ and ḷ vowels' . The phonetic comparison by Allen (1953: 62) of ara [ ə r ə ] to Av. ərə < PIIr. *r̥ may just be a coincidental attempt by Sassanid scribes to represent Av. r̥ in a new alphabet adapted from Pahlavi, which had only consonantal r. ability of the Sanskrit grammarians to provide minute phonetic detail has previously been noted by Catford (2001: 181-183), who also claims that Sanskrit had a dorsal r, though he prefers to reconstruct an English-like [ɹ]. My account does not require [ɹ] for the Indo-Aryan phonological processes under discussion, only the typologically common inclusion of a dorsal gesture in coda liquids.

Conclusions
In the preceding discussion, I have laid out evidence the Indo-Iranian reflexes of PIE *L̥ (H) undergo phonetic coarticulation of lip rounding gestures from nearby labial segments. This behavior shows itself not only in the oldest daughter languages, but extends well into the middle and modern languages. The widespread appearance of this rounding behavior allows us to reconstruct the rounding environments P___ and ___C0{u, w} to Proto-Indo-Iranian and helps to describe the likely phonetics of PIIr. *r/*r̥ . Among these phonological explanations, the environment of Vedic labiovelar traces was further narrowed and clarified, providing insight into the phonological processes of early Indo-Iranian. The finding that Indo-Aryan shows limited reflexes of all three velar series casts further doubt on the use of centum-satəm distinction as a phylogenetically probative category as in Kortlandt (2017).64