Anatolian evidence suggests that the Indo-European laryngeals * h2 and * h3 were uvular stops

In this article it will be argued that the Indo-European laryngeals * h2 and * h3 , which recently have been identified as uvular fricatives, were in fact uvular stops in Proto-Indo-Anatolian.AlsointheProto-AnatolianandProto-Luwicstagesthesesoundsprob- ably were stops, not fricatives.


Background
It is well-known that the Indo-European laryngeals *h2 and *h3 have in some environments survived in Hittite and Luwian as consonants that are spelled with the graphemes ḫ (in the cuneiform script) and h (in the hieroglyphic script).1 Although in handbooks it was usually stated that the exact phonetic interpretation of these graphemes is unclear,2 in recent years a consensus seems to have formed that they represent uvular fricatives (Kümmel 2007: kloekhorst Indo-European Linguistics 6 (2018) 69-94 331; Simon 2014;Weiss 2016), an interpretation that is based on the way in which Hittite and Luwian lexemes containing these sounds (especially personal names) are rendered in other languages of the Ancient Near East; I refer to the works mentioned for the details.3 From a comparative perspective, this uvular interpretation of the outcomes of *h2 and *h3 in Anatolian is interesting, since it is quite generally assumed that at the Proto-Indo-European stage these sounds were rather pharyngeal fricatives.4 Weiss therefore cogently states that "[i]f the Anatolian reflexes [of *h2 and *h3] were uvulars, then odds are that the second and third PIE laryngeals were themselves uvulars, because directionality favors the development from uvulars to pharyngeal (Aramaic, Hebrew (eventually), Dathina) over the opposite development […]. We can have our cake and eat it too if we suppose that Nuclear Proto-Indo-European underwent a uvular-to-pharyngeal shift" (Weiss 2016: 337).
In other words, Weiss assumes that in the original mother language the laryngeals *h2 and *h3 were uvular fricatives, a situation that was retained as such in Anatolian, whereas all non-Anatolian languages underwent a common innovation, namely a shift from uvular fricatives to pharyngeal fricatives. In this way, Weiss' scenario forms an argument in favor of the Indo-Anatolian5 hypothesis, which states that Anatolian was the first branch to have split off from the mother language. In the case of *h2 and *h3, we can schematize this relationship as follows (note that the language stages that Weiss called "PIE" and "Nuclear Proto-Indo-European", respectively, are here rather called "Proto-Indo-Anatolian" and "Classic Proto-Indo-European"): 3 According to Weiss (2016: 335, 337), the fact that Luwian ḫ is in Neo-Assyrian sometimes rendered as q (e.g. Neo-Ass. Q(a)ue/i = Luw. Ḫii̯ au̯ a and Neo-Ass. Qalparunda = Luw. Ḫal-paruntii̯ a) indicates that "in some dialect" of Luwian the uvular fricative [χ] was hardened to a uvular stop [q], which was then written in Neo-Assyrian with q. However, Simon (2014: 887) has cogently argued that the geographic and chronological distribution of these names with q for Luw. ḫ indicates that they were probably transmitted to Neo-Assyrian via Aramaic: in Aramaic, no [χ] existed, so that the Luwian ḫ = [χ] was substituted by its plosive variant [q]. This attractive analysis makes it unnecessary to follow Weiss' assumption that in some (undefined) dialect(s) of Luwian, a hardening of *[χ] to [q] has taken place, which is a typologically difficult development anyway (see also section 3.1, below). 4 E.g. Beekes 1995: 148;Tichy 2000: 31 (with hesitation) ;Fortson 2004: 58;Weiss 2009: 50. 5 Although often called the 'Indo-Hittite hypothesis' , using the term originally coined by Sturtevant (1933: 30), a better term to describe the theory that the Anatolian branch split off from the mother language is the 'Indo-Anatolian hypothesis' . From now on I will therefore use this term, and use the name 'Proto-Indo-Anatolian' for the mother language from which both Proto-Anatolian and Classic Proto-Indo-European descend.
Indo-European Linguistics 6 (2018) 69-94 Proto-Indo-Anatolian *h2 / 3 = uvular fricatives innovation retention Classic Proto-Indo-European Proto-Anatolian pharyngeal fricatives uvular fricatives In this article, I will go into more details regarding the phonetic value of the outcomes of *h2 and *h3 in the Anatolian languages, and will argue that for Proto-Anatolian (and Proto-Indo-Anatolian) we should not reconstruct them as uvular fricatives, but rather as uvular stops.

The outcome of *hand *h3 in Lycian and Carian
Although in Cuneiform Luwian and Hieroglyphic Luwian the outcomes of *h2 and *h3 are rendered with an ḫ (in cuneiform) and an h (in hieroglyphic), which, following the arguments of Kümmel (2007: 331), Simon (2014), and Weiss (2016), represent uvular fricatives, the value of their corresponding sounds in the two other relatively well known Luwic languages, Lycian and Carian, are different.

2.1
Lycian In Lycian, we find four consonants that are generally regarded to reflect a consonantal outcome of *h2 and/or *h3, namely χ, k, g, and q.

2.2
Carian Our knowledge of the phonology of Carian is much less advanced than that of Lycian, but nevertheless, also for Carian some facts are known regarding the outcome of PIE *h2 and/or *h3.
Car. k: It seems now rather generally accepted that Car. k is the outcome of *h2 and *h3 in at least word-initial position (Car. kδou-'king(?)'~CLuw. ḫantau̯ at(i)-'id.' < *h2ent-; but cf. also the Carian gloss κόον 'sheep' (Adiego 2007: 455)~CLuw. ḫāu̯ a/i-'id.' < *h3eu-i-).12 Synchronically, there can be no doubt that Car. k represents a voiceless velar stop.13 Car. q: Also this sound reflects *h2: it is the outcome of the sequence *-h2u̯ -in There is debate on the exact phonetic interpretation of Car. q: Adiego (2007: 244) assumes that it had the value of a uvular stop, [q] (thus also Simon 2011: 539f. and Brosch 2016: 9-10), whereas I myself have rather proposed a labiovelar stop [kw] (Kloekhorst 2008a: 138).14 Whatever be the exact value of Car. q, it is clear that Car. k was a voiceless velar stop [k], which seems to be the regular outcome of PIA word-initial *h2 / 3. It thus matches Lyc. χ, which is a voiceless velar stop [k], as well.
Indo-European Linguistics 6 (2018) 69-94 will only focus on the unlenited outcome of *h2; the value of its lenited counterpart will be discussed in section 8; and cf. section 9 for a more detailed discussion of *h3 In theory, there are three options.

Option 1: PLuwic *[χ]
The first option is that the Proto-Luwic value of *h2 was the same as in the Luwian languages, namely *[χ]. This implies that we have to assume a development of PLuwic *[χ] > Lyc./Car. [k], i.e. a "hardening" of a fricative to a stop and a subsequent fronting of the uvular to the velar place of articulation (or first a fronting of uvular *[χ] to velar *[x] with subsequent hardening to [k]). This development is not self-evident, however, since the hardening of a velar / uvular fricative to a stop is a cross-linguistically rare phenomenon. In his book on the typology of consonantal change, Kümmel (2007: 147-148) remarks that an unconditioned hardening of fricatives is almost exclusively attested for the dental-interdental place of articulation. In the case of velar and uvular fricatives, Kümmel (2007: 148) states that a hardening of a fricative to a plosive articulation is usually conditioned ("teilweise auf den Anlaut oder andere Positionen beschränkt"), and that "eine Entwicklung zur Aspirata häufiger [ist]". And indeed, of the examples of hardening of velar / uvular fricatives that Kümmel cites (2007: 148), in most cases they develop into an aspirated velar stop [kh]. In the case of Lycian and Carian, however, the velar stops were clearly unaspirated: if they would have been aspirated, they should have been rendered in Greek with the aspirated velar stop χ [kh], which they never are. Therefore, the languages in which a velar / uvular fricative develops into an aspirated [kh] cannot be used as an argument in favor of postulating a PLuwic *[χ] that would have undergone a hardening into Lyc./Car. [k]. The only examples of a hardening of a velar / uvular fricative to an unaspirated stop cited by Kümmel (2007: 148) are either conditioned15 (and thus cannot be used as a 15 The development of "x > k /_R" in Upper Sorbian and Central Lower Sorbian, and the development of "x > k /_#" in Polish dialects (Lesser Polish).
Indo-European Linguistics 6 (2018) 69-94 parallel for PLuwic *[χ] to [k], which seems to have been unconditioned), or are dubious.16 All in all, assuming an unconditioned hardening of a PLuwic uvular fricative [χ] to a velar stop in Lycian and Carian does not seem to be supported by cross-linguistic evidence.
Of course, one could argue that Lycian and Carian have undergone a heavy substratum influence that eliminated fricatives in general, and thus did cause a hardening of *[χ] to [k]. However, this would be contradicted by the fact that synchronically in Lycian we do find fricatives, like labial b = [β] (or [ɸ]), dental d = [ð] (or [θ]), but also velar g = [ɣ] (or [x]), which is the outcome of the lenis variant of PIA *h2 (perhaps Carian had fricatives too, but their status and origins are much less clear). So there definitely never was a general hardening of fricatives in the prehistory of Lycian (and probably neither in Carian), which means that postulating a development of *[χ] > Lyc./Car. [k] on the basis of heavy substratum influence is not satisfactory either.

Option 3: PLuwic *[q]
Since the PLuwic value of *h2 can hardly have been identical to its value in the Luwian languages, [χ], and can certainly not have been identical to its value in Lycian and Carian, [k], it is best to assume that it was a sound different from both outcomes. I therefore want to propose that it in fact was a voiceless uvular stop, *[q]. This means that for Lycian and Carian we would have to 16 The development "?*x > k /_" as cited for Proto-Baltic is regarded by Kümmel himself as dubious (cf. the question mark). The development of "x > k /_" as cited for Montenegrin (with reference to Popović 1960: 436), is in fact not unconditioned, but rather an auslaut phenomenon (cf.

Additional arguments
Thus far, the argumentation has been typological. However, in order to make a definite case in favor of interpreting *h2 as a uvular stop at the Proto-Luwic, Proto-Anatolian, and Proto-Indo-Anatolian levels, we need other types of arguments as well. At this moment, I can think of three such arguments in favor of this interpretation.
If we would assign to *h2 the value of a voiceless uvular fricative for the Proto-Anatolian or Proto-Indo-Anatolian stage, it would be difficult to understand why this would yield a long, fortis consonant in Hittite and CLuwian, whereas the other fricative, *s, which was voiceless as well, yielded a short, lenis consonant. As far as I am aware, this problem has never been properly addressed, let alone solved. I now want to argue that the difference in outcome between *h2 and *s can be explained by assuming that *h2 originally was a voiceless stop, *q. In this way, we could compare its outcome as a fortis consonant in Hittite to the outcome of the other voiceless stops of PIA, which in Proto-Anatolian and in Hittite and CLuwian yielded long, fortis consonants as well, e.g. PIA *t > PAnat. *[tː] > Hitt./CLuw. [tː], spelled -tt-, etc.
In other words, we can now set up the following developments, in which the outcome of *h2 = *q as a geminate spelled -ḫḫ-in Hittite and CLuwian can be explained.
One could in principle explain the development of *sh2V-to Hitt. išḫV-by stating that in this environment fricatives were treated as stops. This may be contradicted, however, by the development of *ssénti > Hitt. šašanzi 'they sleep' (only if this development may be viewed as regular, which admittedly is quite uncertain).

5.3
Argument 3: Aḫḫiiau̯ ā vs. Ἀχαιοί It is nowadays generally accepted that the Hittite toponym Aḫḫiiau̯ ā refers to Mycenaean Greece, and that this name is related to the Greek term Ἀχαιοί 'Achaeans' .20 However, the equation between Hitt. Aḫḫii̯ au̯ ā and Gr. The equation between Aḫḫii̯ au̯ ā and Ἀχαιοί becomes easier if Hitt. -ḫḫderives from an earlier *[qː]. We could then assume that Aḫḫii̯ au̯ ā goes back to an earlier form *[aqːiaw-], which is much closer to the Greek stem *[akhaiw-].

6
A counter-argument Next to these additional arguments in favor of assuming that *h2 was a uvular stop at the Proto-Indo-Anatolian, Proto-Anatolian and Proto-Luwic level (of which I am aware that the former one is clearly stronger than the latter two), I can think of one possible argument against this postulation. We would have to assume that within the Anatolian language family a development of * However, as we have seen above as well, this would match the situation in the Turkic language family, in which a fricativization of q to χ has taken place in several branches independently as well (e.g. Tuvan, Khalaj, cf. Johanson 1998: 99-100).21 To my mind, this counter-argument therefore does not outweigh the arguments in favor of postulating that *h2 was a uvular stop rather than a fricative.

Conclusions regarding *h2
We have seen that, typologically, it is difficult to understand how Lycian and Carian [k] could have been the outcome of *h2 if the latter sound originally was a uvular fricative. Instead, Lyc./Car. [k] is much better explained from a uvular stop, which can also account for the uvular fricative as found in Hittite and Luwian. Moreover, there are additional arguments to be given in favour of such a reconstruction, especially the fact that *h2 yields a fortis (long) consonant in Hittite and CLuwian. All in all, I want to propose the following values for *h2 in the different language stages:22 21 Moreover, another branch in which the development of *q to *χ must then have taken place is Classic Proto-Indo-European, where *χ subsequently developed into a pharyngeal fricative (see also section 1). 22 Note that

The lenis variant of *h2 in Anatolian
As is well known, Proto-Anatolian knew several lenition rules, according to which original fortis consonants were lenited in certain intervocalic positions, namely (1) after a long accented vowel, and (2) when standing in between two unaccented vowels in a posttonic position.23 Also *h2 was subject to these lenition rules: the lenited variant of *h2 is in Hittite and CLuwian spelled as a single -ḫ-, and in Lycian as g (its value in Carian is not yet clear Indo-European Linguistics 6 (2018) 69-94 As we see, the renderings of word-initial ḫ-and word-internal fortis -ḫḫ-are taken together here (first column), because they usually are the same. They moreover contrast with the rendering of word-internal lenis -ḫ-, which is therefore given separately (second column). In the third column, Simon lists the other post-velar phonemes of the respective languages, which were not used to render the Luwian laryngeals, and which therefore can be used as an argument for determining what kind of value they did not have.
Despite his very clear presentation of these data, Simon's discussion of the material is less lucid, containing several incomprehensible steps (some of which even contradict his own findings!). This leads him to state that the distinction between Luw. fortis -ḫḫ-and lenis -ḫ-must have been one between a fricative and a stop, and he therefore assumes that Luw. fortis -ḫḫ-was a voiceless uvular fricative [χ], whereas lenis -ḫ-was a voiceless uvular stop [q] (Simon 2014: 888). This is quite impossible, however: the only language in which lenis -ḫ-is rendered as a stop is Egyptian, where we find a [g], even though this language would have a [q] at its disposal to render the value of -ḫ-if this really were a [q]. The fact that in Egyptian not [q], but [g] is used instead, rather indicates that Hitt./Luw. lenis -ḫ-cannot have been a [q].
The one feature that Aramaic, Egyptian and Ugaritic share in their rendering of Luw. lenis -ḫ-is voice: in Aramaic we find a voiced pharyngeal fricative [ʕ], in Egyptian a voiced velar stop [g], and in Ugaritic a voiced fricative that was either velar, [ɣ], or uvular, [ʁ]. It therefore seems fully cogent that lenis -ḫwas a voiced sound. Since fortis -ḫḫ-by all means must have been a voiceless uvular fricative [χ], it is only logical that lenis -ḫ-was its voiced counterpart: a voiced uvular fricative [ʁ]. The fact that in Egyptian this sound was rendered as a voiced velar stop can then be explained by the fact that this language did not have a voiced uvular fricative at its disposal.24 24 Simon explicitly denies the possibility that lenis -ḫ-could have been a voiced fricative, but his reasoning is flawed. He first claims that if the distinction between fortis -ḫḫ-and lenis -ḫ-were one in voice, we would have to assume either a pair , das, obwohl genau unbekannt ist, bestimmt keinen Frikativ darstellt". However, the ⟨q⟩ that is sometimes found in Assyrian and Aramaic renderings of Hittite and Luwian names is in fact used to denote fortis -ḫḫ-and not its lenis variant -ḫ-. Moreover, Simon himself has convincingly argued that this ⟨q⟩ must have been a (Samʾalic) Aramaic substitution of Hitt./Luw.
[χ] (a sound that Aramaic itself did not possess), and that the names containing this ⟨q⟩ were then taken over in Assyrian from this variety of Aramaic (Simon 2014: 887-888; cf. also footnote 3 above). Therefore, the use of a ⟨q⟩ in Assyrian and Aramaic for Indo-European Linguistics 6 (2018) 69-94

8.1
Voice or length? This is not the whole story, however. As we have seen above, in both Luwian and Hittite, the fortis -ḫḫ-was in fact a long voiceless uvular fricative, [χː]. The question thus arises: to what extent is this length relevant for the phonemic distinction between fortis -ḫḫ-= [χː] and lenis -ḫ-= [ʁ]?
An answer may be provided by the following observation. The name of the Luwian Storm God tarḫunt-is in Egyptian attested as trgt-, with the voiced stop [g], and in Ugaritic as trġ(n)d-, with the grapheme ġ that either represents a voiced velar fricative [ɣ] or a voiced uvular fricative [ʁ].25 Nevertheless, there can be no doubt that the ḫ in this form was fortis: the corresponding form in Hittite is spelled tar-uḫ-ḫa-an-t°, with geminate -ḫḫ-, and also Lyc. trqqñt-and Car. trqδ-clearly point to an original fortis consonant (albeit its labialized variant, from *h2u̯ ). Apparently, fortis [χː] could allophonically be voiced when standing next to a resonant. This implies that voice was not the underlying phonemic difference between fortis [χː] and lenis [ʁ]: it must have been length instead. I therefore assume that in Luwian and, by extension, in Hittite the phonemic values of fortis -ḫḫ-and lenis -ḫ-were /χː/ and /χ/, respectively, albeit that in intervocalic position the latter was allophonically voiced, [ʁ], and that also the former could be allophonically voiced when standing next to resonants, [ʁː].

8.2
The value of Lycian g A similar reasoning goes for the value of the outcome of the lenis variant of *h2 in Lycian. Although the sign with which this sound is written, g, is graphically identical to the Greek sign γ, which represents a voiced velar stop [g] (just as its labial and dental counterparts b and d are graphically identical to Greek β = [b] and δ = [d], respectively), it is generally not seen as representing a voiced stop, but rather as a voiced fricative, [ɣ] (likewise b = [β] and d = [ð]).26 This is based on the observation that voiced stops are in Lycian usually rather spelled with the signs for voiceless stops with a preceding nasalization: e.g. ñtarijeus = Δαρεῖος and idãχre = Ιδαγρος. In my opinion, this observation has another implication, namely that voice may not have been a phonemic feature: if fortis χ = [k] could allophonically be a voiced stop [g], then the phonemic distinction with its lenis counterpart g was possibly not determined by voice. This means rendering Hitt./Luw. fortis -ḫḫ-has no bearing on the phonetics of the Hitt./Luw. lenis -ḫ-, and thus does not preclude that the latter was a fricative, too. 25 Cf. that the voice quality of g may have been allophonic as well, leaving open the possibility that its basic value was a voiceless velar fricative /x/.27 Nevertheless, the fact that this sound is spelled with the sign g, which is equivalent to Greek γ, indicates that it was often voiced. Such allophonic voicing would be very likely in intervocalic position and probably also when the consonant is adjacent to resonants.

8.3
The outcome of lenited *h2: conclusions Although both the Hittite and Luwian value of the outcome of lenited *h2, which was /χ/ (with an allophonic voiced variant [ʁ]), and its Lycian counterpart [ɣ] (which may be an allophonic variant of an underlying voiceless /x/) were fricatives, I do not think that it is likely that their Proto-Anatolian predecessor was a fricative as well. In Lycian, the other fricatives, b and d, go back to lenis stops, and it therefore seems attractive to me that Lyc. g goes back to a stop as well. All in all, I reconstruct for both the fortis and the lenis outcome of *h2 in the Anatolian branch the following values (all of which could in certain environments be allophonically voiced): PAnat. */qː/ vs. */q/ Proto-Luwic */qː/ vs. */q/

The phonetics of *h3
It is generally assumed that *h2 and *h3 were phonetically close, but the exact relationship between the two sounds is unclear. Some scholars assume that *h3 Indo-European Linguistics 6 (2018) 69-94 is the labialized variant of *h2;28 some think that *h3 is the voiced / lenis variant of *h2;29 and others postulate both differences.30 The idea that *h3 may have been the labialized variant of *h2 is of course based on the fact that *h3 generally has an o-colouring effect on a neighbouring *e (i.e. backing and rounding), whereas *h2 has an a-colouring effect on a neighbouring *e (i.e. backing and lowering, but not rounding). This is by all means certainly a cogent argument.
The idea that *h3 may have been the voiced / lenis variant of *h2 is based on several considerations, which I will treat one by one: 1. The reduplicated present to the verbal root *peh3-'to drink' must be reconstructed as *pi-bh3-e/o-, with a *b: Skt. píbati, Lat. bibō (with assimilation of initial *p-to b-), Arm. əmpəm < *pimb-(with secondary nasalinfix), OIr. -ib, ebait. Structurally, we would expect this reduplicated present to rather have been *pi-ph3-e/o-, however, with a *p (structure *C1i-C1C2-e/o-). It is therefore generally assumed that a change of *-ph3-to *-bh3-has taken place, which would then indicate that *h3 was a voiced consonant, since this development can then be viewed as due to voice assimilation. Since *h2 does not cause voicing, but in some branches instead causes aspiration (e.g. -th2e > Skt. -tha (2sg.perf.ending)), it is assumed that *h3 thus was the voiced variant of *h2.
There are a few odd things, however. First, there is not a single other good example in which a voicing because of *h3 would have taken place. For instance, the other verbal root of a structure *Teh3-, viz. *ḱeh3-'to sharpen' , does not show any voicing at all (e.g. Skt. 3pl.pres. śiśanti < *ḱi-ḱh3-enti, ptc.midd. śiśāna-< *ḱi-ḱh3-mh1no-, ta-ptc. śita-< *ḱh3to-). Second, the voicing effect of *h3 would have taken place in Proto-Indo-European already (that is, at least the stage from which Indo-Iranian, Italo-Celtic and Armenian derive). This contrasts with the aspirating effect of *h2, which is found only in Indic, Armenian and Slavic, and thus may have been a satəm-development only.31 According to Kortlandt (e.g. 1996: 53), we should therefore postulate an alternative scenario to explain the presence of *b in *pi-bh3-e/o-. In his view, the original root actually was *beh3-(with a reduplicated present  (Tichy 2000: 31). 31 It is debated whether Greek shows aspiration because of *h2, cf. the discussion in De Decker 2011.
Indo-European Linguistics 6 (2018) 69-94 *bi-bh3-e/o-), in which, in pre-PIE times, *b (originally a glottalized stop) in word-initial position lost its glottalic feature and merged with *p-. This would then explain why in all forms of the verb we find word-initial *p-, but not in word-internal position in the reduplicated present *pibh3e/o-. If this scenario is correct,32 *-bh3-would have been the original cluster and would not derive from earlier *-ph3-through voice assimilation. This then removes the necessity for *h3 to have been a voiced consonant. 2. The outcome of word-initial *h3-in Anatolian is debated, and one can find different opinions in the literature, especially on the outcome of *h3-in Lycian. For instance, it has been claimed that in Lycian *h3-is lost, as would be clear from the following two examples:33 epirije-"to sell" and epenẽtijatte "acts as a salesman", both from *h3ep-.34 Interestingly, the corresponding Hittite forms have an initial ḫ-(ḫappirii̯ e/aand *ḫappinantii̯ aḫḫ-< *h3ep-). This differs from the outcome of *h2-in Lycian, which was χ-(e.g. χñtawa-'rule' < *h2ent-), and which corresponds to Hitt. ḫ-as well (e.g. Hitt. ḫant-< *h2ent-). According to Melchert (1994: 55), we therefore have to assume two different outcomes for *h3-and *h2in Proto-Anatolian. Since *h3-was lost in Lycian, but *h2-was retained as a consonant, he states that *h3-had yielded in Proto-Anatolian a phoneme that he notes down as */h/, and which was the lenis variant of *h2-, which yielded PAnat. */H/. In Melchert's view, the main phonetic distinction between the two was that */H/ < *h2-was voiceless, whereas */h/ < *h3was voiced (an assumption based on the idea that *h3 was a voiced consonant in PIE, cf. 1994: 47). However, both examples in favor of loss of word-initial *h3-in Lycian are unconvincing: epirije-probably does not mean 'to sell' at all, and therefore 32 Note that the "b-gap" is a generally recognized constraint for Proto-Indo-European, and that such a constraint can only be explained by a pre-PIE development of *b into a different sound (thus already Pedersen 1951: 10-16). However, the b-gap is not complete: we do find *b in word-internal position in e.g. *h2eb-l-'apple' , *ghrebH-(?) 'to grab' , *lembH-'to hang down limply' , *seib-'let flow' , *steib-'to make stiff' , and *uremb-'to turn' (although the reconstruction of some of these verbal roots may be debated). It therefore stands to reason to assume that the sound law that caused the rarity of *b in Proto-Indo-European was a conditioned one. In that sense, Kortlandt's postulation of a pre-PIE soundlaw of word-initial *b-> *p-and thus explaining *pi-bh3-e/o-from earlier *bi-bh3-e/o-, is perfectly in line with more general considerations and is not as ad hoc as it at first sight may seem. 33 A third alleged example, Lyc. ẽtre/i-'lower' < *h3ndhero-~Hitt. ḫantii̯ ara-"niedrig" as adduced by Oettinger (2001: 84-86) has in the meantime been withdrawn (Oettinger 2007). See : 103 for criticism on this example. 34 Kimball 1987;Melchert 1989: 43. Indo-European Linguistics 6 (2018) 69-94 does not need to be cognate to Hitt. ḫappirii̯ e/a-;35 and the translation and interpretation of epenẽtijatte is seemingly based on etymological considerations only, and therefore cannot be used as an argument.36 Instead, I have argued elsewhere (Kloekhorst 2006: 102-103) that initial *h3-rather seems to have yielded χ-in Lycian ( χawa-'sheep' < *h3ew-and χerẽi 'name of a dynasty' < *h3er-on-). If correct, this would mean that in word-initial position, *h3-in principle merged with the outcome of *h2-, and that for Proto-Anatolian no distinction between the outcomes of initial *h2-and *h3-needs to be assumed. 3. Melchert (2011) argues that the Hittite verb lāḫu-i 'to pour' should not be reconstructed as deriving from a stem *leh2-u-, as was usually done,37 but rather contains a root *leh3u-, which is identical to the root *leuh3-(with laryngeal metathesis) as attested in Gr. λοέω, λούω, Lat. lavō 'to wash' . Thus far, such a reconstruction was impossible, since it was generally assumed that, in Hittite, *h3 was lost in intervocalic position, cf. e.g. *dóh3-ei > dāi 'he takes' . However, Melchert, accepting my proposal that the sequence *-h2u̯ -yielded a Proto-Anatolian unitary 'labiolaryngeal' ,38 argues that also the sequence *-h3u̯ -may have yielded such a 'labiolaryngeal ' (2011: 129). Since in lāḫu-i we find single spelling of the -ḫ-, Melchert assumes that the outcome of *-h3u̯ -was in fact a lenis labiolaryngeal "*ɣw/*ʕw", which thus contrasts with the fortis labiolaryngeal that is the result of *-h2u̯ -(for which Melchert assumes a value "*ħw/*xw"). The etymology of *lāḫu-i as reflecting the root *leh3u-may then be used as an additional argument in favor of viewing *h3 as the lenis variant of *h2. Although I accept Melchert's proposal to derive Hitt. lāḫu-i from *leh3u-, in which the retention of *h3 as a consonant is due to an earlier development of the sequence *-h3u̯ -into a monophonemic 'labiolaryngeal' , I do not think that his conclusion that this 'labiolaryngeal' was always lenis is justified. As I have extensively argued elsewhere,39 there are clear indications that PIA short accented *ó had in Proto-Anatolian become a long vowel, */ṓ/, which, just as other long accented vowels, caused lenition of a following intervocalic fortis consonant. This rule can, amongst others, explain the class of ḫi-verbs in which the 3sg.pres.act. form shows a lenis stem-final consonant (e.g. āki 'he dies' , ḫāši 'she gives birth'), whereas 35 Rasmussen 1992: 56-59. 36 Kloekhorst 2006 Cf. e.g. Kloekhorst 2008b: 511-513. 38 Kloekhorst 2006 in the other forms of the paradigm the stem-final consonant is fortis (akkanzi 'they die' , ḫaššanzi 'they give birth'). Especially cases in which the stem-final consonant is -ḫ(ḫ)-(nāḫi / naḫḫanzi 'to fear' , zāḫi / zaḫḫanzi 'to hit, to beat') are telling: this *-ḫ(ḫ)-can only reflect *-h2-, which normally yields a fortis consonant intervocalically. There is therefore simply no other option than that in the preforms of 3sg.pres.act. *nóh2ei and *tióh2ei it was the *ó that caused the lenition (through PAnat. */ṓ/). Although in his 2011-article, Melchert calls this analysis of these verbs "entirely ad hoc" (2011: 128),40 already a year later, in 2012, he has to admit that nāḫi, etc. can only be explained by a lenition because of the preceding *ó (2012: 177-179).41 The recognition that *ó has a leniting effect on following fortis consonants, opens up the possibility that also the lenis character of -ḫ-in lāḫu-i is the result of lenition, since this form reflects *lóh3u-ei. It therefore cannot be used anymore as an argument that *h3 must be a lenis consonant. We see that none of the three arguments in favor of viewing *h3 as the lenis variant of *h2 is compelling: alternative scenarios can be provided to explain the state of affairs.
To my mind, there are however two arguments to be given that would indicate that *h3 is, just as *h2, originally a fortis consonant. 1. Besides the verb lāḫu-i 'to pour' , which shows a single spelled -ḫ-, Hittite knows the lexemes laḫḫu-'a vessel' and laḫḫura-'offering table' , which both show geminate spelling of the -ḫḫ-.42 Especially laḫḫu-'a vessel' is semantically close to lāḫu-i, but also laḫḫura-'offering table' may be connected to it, cf. the following context (KUB 9.31 ii 8-9): (8) n=a-at=ša-an GIS̆l a-aḫ-ḫu-ri šu-uḫ-ḫa-i nu me-na-aḫ-ḫa-an-da (9) GESTIN la-aḫ-ḫu-u-u̯ a-i 'He scatters them (broken pieces of thick-bread) on the laḫḫura-and pours wine over (them)' . Moreover, laḫḫura-is occasionally spelled lāḫura-, which can only be explained as due to influence from lāḫu-i, indicating that, at least synchronically, the Hittites saw a connection between the two lexemes. Vice versa, lāḫu-i does in NH texts occasionally occur spelled laḫḫu-(like in the context cited), which must be due to influence by laḫḫura-.
On the basis of these considerations I had proposed already earlier on (Kloekhorst 2008b: 513-514, at the time of which I still reconstructed a stem *leh2u-with *h2) that the stems laḫḫu-(in laḫḫu-'a vessel' and laḫḫura-'offering table') and lāḫu-(in lāḫu-i 'to pour') are etymologically related to each other, and that the former one goes back to an e-grade stem *léh2u-, whereas the latter reflects the o-grade stem *lóh2u-, in which the accented *ó (through PAnat. long *ṓ) causes lenition. Taking Melchert's attractive connection to Gr. λοέω, λούω, Lat. lavō 'to wash' < *leuh3-< *leh3u- (Melchert 2011) into account, we should now reconstruct these stems as follows: laḫḫu-< *léh3u-, and lāḫu-< *lóh3u-. The consequence of this is that the normal, unlenited intervocalic outcome of *h3 in this environment was a fortis consonant, and that we thus must assume that *h3 originally was fortis as well.43 2. According to LIV2: 679, the Hittite verb u̯ alḫ-zi 'to hit, to strike' can be connected to, amongst others, Gr. ἐάλων 'was killed' , which points to a root *uelh3-(with Gr. ἐάλων < *h1e-ulh3-eh1-). This was followed by myself (Kloekhorst 2008: 946) and Melchert (2011: 129), who explicitly cites this verb as an example in which word-medial *h3 was retained as a conso-43 An anonymous reviewer objects to this argumentation, for two reasons. First, in his/her view, "there is no chance at all that the two words [laḫḫura-and lāḫu-i] are related". To my mind, an etymological connection between these words is certainly possible, however, cf. the context cited. Second, (s)he states that "[t]he noun laḫḫuš-surely is derived from 'to pour' , but shows merely the well-known prehistoric devoicing before *s seen in akkuške-" (note that the reviewer cites the stem of the word for 'vessel' as "laḫḫuš-", implying an interpretation as a (neuter?) -uš-stem, whereas CHD L: 13 cites the stem as a common gender u-stem laḫḫu-, interpreting the attested form la-aḫ-ḫu-uš as a nom.sg. form in -š). However, the geminate found in akkuške/a-(imperf. of eku-zi / aku-'to drink') is not the result of devoicing before *s. If this were true, we would expect a "devoicing" in 2sg.pres. ekušši, 3sg.pret. ekuzzi, 2pl.pres. ekutteni, 2sg.pret. ekutta, 3sg.pret. ekutta, 3sg.imp. ekuddu, etc., as well. All these forms show that there was no such thing as a "prehistoric devoicing" of *gw(h) before whatever original voiceless consonant at all. The geminate as found in akkuške/a-< *h1gwh-ske/o-is due to a specific fortition of *gwh before the cluster *sk. Therefore, the value of laḫḫu-'a vessel' (nom.sg. laḫḫuš) simply cannot be dismissed: its semantic connection to lāḫu-i is beyond any doubt, and its geminate -ḫḫ-cannot be explained as due to the -š (whether this š is part of the stem or forms the nom.sg.c. ending). This form therefore proves that the lenis -ḫ-in lāḫu-i goes back to an original fortis -ḫḫ-, which means that *h3 originally was a fortis consonant.
Indo-European Linguistics 6 (2018) 69-94 nant. Interestingly, this verb often shows geminate spelling of the ḫ: part. u̯ a-al-aḫ-ḫa-an-t°, verbal noun u̯ a-al-aḫ-ḫu-u̯ a-ar, inf.I u̯ a-al-aḫ-ḫu-u̯ aan-zi, imperf. u̯ a-al-aḫ-ḫi-iš-ke/a-, imperf. u̯ a-al-aḫ-ḫa-an-na/i-.44 There can therefore be no doubt that this consonant is fortis, /ualχː-/, which implies that *h3 originally was a fortis consonant. On the basis of these arguments, we may conclude that *h3 originally was a fortis consonant. If this is correct, it means that both *h2 and *h3 originally were fortis consonants, and that the distinction between the two therefore cannot have been that the one was the lenis variant of the other.45 Instead, we then have to assume that the distinction between *h2 and *h3 was some other feature, and it therefore is attractive to assume that the latter was the labialized variant of the former. Since we have above determined that it is likely that, in Proto-Indo-Anatolian, *h2 was a long uvular stop */qː/, we may assume that at that stage *h3 was a long labialized uvular stop */qwː/. An extra argument in favor of this view is that we can now explain why, in Anatolian, word-internal *h3 was only retained when it stood before a *u̯ : in this position the labialization of *h3 = */qwː/ was neutralized, causing it to merge with its non-labialized counterpart *h2 = */qː/.
We thus can set up a relative chronology of the development of *h2 and *h3 in Anatolian (exemplified by Hittite and Lycian): Cf. Kloekhorst 2008: 945-946. 45 If correct, it means that the reduplicated present *pi-bh3-e/o-should be explained as proposed by Kortlandt, namely that it reflects an earlier *bi-bh3-e/o-and that the root for 'to drink' in fact was *beh3-, originally. 46 I assume that the phonologization of the colouring of *e to *o when adjacent to *h3 caused the loss of the labialization, which results in the merger of *h2 and *h3.
Indo-European Linguistics 6 (2018) 69-94 The fact that in intervocalic position, *h3 seems to have been lost in Anatolian (e.g. *dóh3ei > Hitt. dāi 'he takes'), means that this consonant was in this environment probably lenited at an earlier stage. The fact that *h2 yields a consonantal outcome in this environment (e.g. *nóh2ei > nāḫi), however, shows that *h2 and *h3 did not develop in a parallel fashion. This should not surprise us too much. From other languages we know that labialized consonants can be lenited earlier than their non-labialized counterpart. Compare, for instance, the fact that PIE *(ǵ)-yields Lat. g-, whereas its labialized counterpart *gw-> Lat. v-, with loss of its buccal part. We may therefore assume that also *h3 = */qwː/ was in some environments lenited earlier than its non-labialized counterpart *h2 = */qː/, which would yield a relative chronology along the following lines (with pre-PAnat. (1) representing the stage before colouring of adjacent vowels, and pre-PAnat. (2)  > *[VħːV] > */VʔV/ > VV or VĀ similar development may also be envisaged for other environments in which *h3 did not yield a consonant in Hittite, but was ultimately lost, e.g. *Vh3C. But in this environment also *h2 was ultimately lost, and we may therefore assume a similar development: Moreover, to my mind it is quite possible that in these environments these first lenitions (*/qː/ > *[χː] and */qwː/ > *[χwː]) had taken place in Proto-Indo-Anatolian already, and that at that moment in time, both *h2 and *h3 had two allophones, namely *[qː] and *[χː], and *[qwː] and *[χwː], respectively. In fact, such an assumption is necessary anyway since the phonotactic behaviour of some PIA roots containing laryngeals demand that these were fricatives rather than stops. For instance, as was argued by Schindler (1975: 265-266), Indo-European Linguistics 6 (2018) 69-94 the distribution within neuter s-stems between nom.-acc.sg. forms of the type *CéCH-s (e.g. *kréuh2-s) and of the type *CéC-os (e.g. *men-os) << *CeC-s is best explained by assuming that in *CéCH-s the laryngeals could already be syllabified at a pre-PIA level, which precluded the replacement of the suffix *-s by its syllabic allomorph *-os.47 For *h2 and *h3 this implies that in this environment they had already early on undergone an allophonic fricativization to *[χː] and *[χwː], respectively.

Conclusions
We may conclude that Anatolian provides several arguments that indicate that *h2 was a long voiceless uvular stop *[qː] at the Proto-Indo-Anatolian level, as well as at the Proto-Anatolian and Proto-Luwic stages, having retained its stop quality into the Luwic languages Lycian and Carian, where (in some environments) it yielded a velar stop [k] (or a labiovelar stop [kw] when originally standing before *u̯ ). In the case of *h3, arguments can be given that this consonant was the labialized counterpart of *h2, i.e. *[qwː]. Although *[qː] and *[qwː] were the basic values of *h2 and *h3, it is likely that already in PIA they had in some environments allophonically become fricativized to *[χː] and *[χwː], respectively. It remains a task for the future to determine the exact details for the distribution between the plosive and fricative allophones of these two phonemes, but we may assume that in environments where Lycian and Carian show [k] (or [kw] when originally before *-u̯ -), the PIA renderings of *h2 and *h3 must have been *[qː] and *[qwː], respectively.