Proto-Indo-European * a

There are around sixty Indo-European roots that are (sometimes) reconstructed with a vowel * a in the scholarly literature that otherwise fully embraces the laryngeal theory. This number is extremely low compared to the number of morphemes in which the vowels that are traditionally reconstructed as * e and * o are found. This marginal status of the vowel * a is typologically odd and has led some scholars to deny the existence of a vowel * a in Proto-Indo-European or in a precursor of Proto-Indo-European.This paper discusses the comparative evidence for the reconstruction of Proto-Indo-European * a . Itconcludes that thereisinsufficient evidence for thereconstructionof * a foranystage of the proto-language.


Introduction
How many vowels did Proto-Indo-European (PIE) have? This is a question that became pertinent after the discovery of the laryngeals. It became clear that for the vast majority of cases in which Indo-Europeanists used to reconstruct *a and *ā, these vowels stood next to *h2 and could be reinterpreted as positional variants of the vowel *e. As a consequence, *a and *ā are now no longer reconstructed for any suffixes or endings (Rasmussen 1989: 261, Beekes 1991,1 and the number of nominal and verbal roots in which they are reconstructed Indo-European Linguistics 7 (2019) 122-163 is very limited. The Lexikon der Indogermanischen Verben (LIV2) lists only 23 verbal roots that are reconstructed with a vowel *a out of a total of over 1200 roots.2 Also, in most of the roots for which *a and *ā are still reconstructed, these vowels do not take part in ablaut, unlike the more frequent vowels *e and *o. The rarity of *a and the fact that it rarely if ever takes part in ablaut has led scholars to question the very existence of a phoneme *a in early or late Proto-Indo-European. While Kuryłowicz (1956: 193) still stated that "nous hésitons encore, en face d'étymologies comme *kaiko-, *daiu̯ er-, *kȃnku-, *kȃso(n)-, *sauso-, ghans-, à considérer comme définitive la preuve de l' origine post-indoeuropéenne (méridionale) du vocalisme ă", others have since concluded that the evidence in favour of reconstructing PIE *a was insufficient (Lubotsky 1989, Beekes 1991: 238, Smoczyński 2006: 85f., Kloekhorst 2008. The most elaborate treatment of the problem is that by Lubotsky, who provided a detailed discussion of a number of words often reconstructed with *a and concluded that none of them warrant the reconstruction of a PIE phoneme /a/. Beekes (1991: 238) agreed, stating that "I consider it as one of the most important insights provided by the laryngeal theory that PIE had no phoneme *a." A less radical view distinguishes between early PIE phonetic [a] and late PIE phonemic /a/ (NIIL xix, Kümmel 2012: 306). The meaning of "late PIE" in this context appears to be the latest reconstructable common ancestor of all Indo-European branches, including Anatolian.
The aim of the present paper is to re-evaluate the evidence for phonemic *a and *ā in PIE and, if there is any evidence supporting the reconstruction of these phonemes, to decide whether they must be attributed to a specific stage of PIE. It will be argued that the emergence of phonemic *a and *ā can be dated after the dissolution of the latest common ancestor of the non-Anatolian Indo-European languages. Before we discuss the data, a few words about typology are in order.
The "peripheral status [of *a] in the vowel system is typologically very odd" (Sihler 1995: 45). It may seem that this problem can be resolved by assuming that any *e that was adjacent to *h2 had already been interpreted as a separate pronk Indo-European Linguistics 7 (2019) 122-163 phoneme *a by speakers of late Proto-Indo-European, as a result of, or leading to, the borrowing of non-Indo-European words with a new phoneme *a. This scenario is also implied by the notations *h2a and *ah2 instead of *h2e and *eh2 employed by Eichner (1988) and a number of other scholars. Assuming, for the moment, that this is correct, it could be hypothesized that all words in which *a cannot be interpreted as a positional variant of *e or *o are recent borrowings that entered PIE when the colouring of *e to *a next to *h2 became or had become phonemic. This scenario is possible for some words for which reconstructions with *a have been proposed in the literature, like Lat. faber 'artisan' or cānus 'white, grey' , but much less so for nouns like Lat. nāris 'nose' , which belongs to basic vocabulary, and verbs like Gr. ἀΐω 'to perceive' , φαγεῖν 'to eat, consume' and αἴνυμαι 'to take, seize' . These words do not appear to be recent borrowings and likely belonged to the PIE vocabulary already before laryngeal colouring of *e had taken place. If they are reconstructed with *a, such a phoneme would have been very marginal in the early Proto-Indo-European phonological system, which would, as Sihler observed, be very odd from a typological perspective.3 A PIE system without *a would have had just four vowels *e, *o, *i, *u (or two, if one counts *i and *u as allophones of *i̯ and *u̯ ). Such a system is typologically less spectacular, especially if one takes into account that *e and *o are just labels. The phonetics of these vowels were almost certainly not mid [e] and [o]. For possible phonetic interpretations see Martinet (1972: 304), Villar (1993), Kortlandt (2010: 37) and Kümmel (2012: 306 ff.). The various phonetic realizations of vowels in around 50 languages that have been claimed to have a four-vowel system have been collected by Hitch (2017: 17, 27-29). Hitch distinguishes the following four types, of which type 4c comes closest to a reconstructed Proto-Indo-European without *a: Proto-Indo-European. The reconstructed phonological system of Proto-Indo-European is, however, ultimately not determined by typology, but by the comparative method. The comparative evidence will therefore take centre stage in the following discussion. It is of course impossible to disprove that PIE had the phonemes *a and *ā, but what we can do is see to what extent the data demand a phoneme *a or *ā. In other words: are *a or *ā part of the minimal set of phonemes of the proto-language from which all plausibly inherited forms can be explained? Their reconstruction is warranted if there are any etyma, but preferably more than a few, that are more likely than not to be of Proto-Indo-European origin and that show correspondences that cannot be explained using the tools otherwise available to us. Whether an etymology is solid enough to be used as evidence is of course to some extent subjective. For the purposes of this paper, full weight will be attached to etyma that are securely attested in two, but preferably more, branches that are not adjacent and for which the etymology does not require special pleading to account for formal or semantic peculiarities. Etymologies that do not fall into this category, e.g. because the semantics of the etymology are not straightforward, because the etymon is limited to a specific geographic area or because the etymology requires the assumption of otherwise unusual ablaut patterns or controversial sound laws, can be discarded as evidence, because the correctness of the etymology in that case partly depends on the existence of a PIE phoneme *a or *ā and the argument would become circular.
Below, we will reassess those reconstructions with *a that are found in literature that otherwise fully embraces the laryngeal theory. I have tried to be as complete as possible, but cannot rule out that I have missed some proposed reconstructions containing *a or *ā. The list of potential PIE roots with the vowel *a consists of around sixty items. Over half of the items on the list also occur in languages outside Europe (Anatolian, Tocharian, Indo-Iranian) and are à priori unlikely to be local post-Proto-Indo-European borrowings from some non-Indo-European language. This is important, because since Kuryłowicz (1956: 194) it has often been claimed that *a is widespread in words that were borrowed from one or more European substratum languages. Eighteen items on the list have a reflex in Anatolian. They are of special interest because they could serve to show that Proto-Indo-European had a phoneme *a even before the Anatolian branch split off or that it acquired *a after the split.
I will ignore roots that used to be reconstructed with an initial vowel and are now generally reconstructed with an initial laryngeal in accordance with Benveniste's ideas about Indo-European root structure, even if direct comparative evidence for a laryngeal is scarce or lacking, e.g. *h2eǵ-/*h2aǵ-'to drive' pronk Indo-European Linguistics 7 (2019) 122-163 and *h2eidh-/*h2aidh-'to kindle' instead of *aǵ-and *aidh-(cf. also the long ī-of Skt. ijate 'drives' < *h2i-h2ǵ-). Such reconstructions are a direct consequence of the laryngeal theory, not of the hypothesis that Proto-Indo-European did not have a phoneme /a/.

2.1.2
*(ḱ)u̯ as-'to kiss' (Eichner 1988: 33, LIV2, Melchert 2016a, Ringe 2017 Hitt. kuu̯ ašzi, Gr. κυνέω 'to kiss' . The vowel *a is reconstructed on the basis of the fact that Hitt. kuu̯ ašzi is a mi-verb, which makes a reconstruction with an ograde, i.e. *(ḱ)uos-, unlikely. However, a reconstruction with a-grade is equally unsatisfactory, because the full grade normally reflected in mi-verbs is an egrade. The Hittite verb could alternatively be cognate with OHG kussen 'to kiss' < *(ǵ)us-or, as preferred by Puhvel (1997: 312), an onomatopoetic word formed in a similar fashion to the Greek and Germanic verbs. Kloekhorst (2008: 506) observed that the Hittite verb is consistently spelled with a geminate -šš-, pointing to an earlier consonant cluster. He argued that, if the Hittite verb is indeed cognate with Greek κυνέω, they could reflect *ḱu-en-s-and *ḱu-ne-s-respectively. In his 2014 book on Hittite accent, however, Kloekhorst withdrew this reconstruction, because the attestation ku-u̯ a-a-aš-zi (KBo 30.101 iii 12) shows plene spelling, which, if taken at face value, makes the connection with Greek κυνέω impossible (2014: 286f., with fn. 1069). He proposed a new reconstruction *Ku(e)h3s-, which could also be reflected in Skt. cuṣati 'to suck, smack' . Summarizing, the prehistory of the Hittite verb remains uncertain, but there is no basis for the reconstruction of a proto-form with *a.

2.2.3
*atta 'dad' (Ringe 2017: 170) Hitt. atta-, Lat. atta, Gr. (voc.) ἄττα, Goth. atta, OCS otьcь, Alb. at. This word is a nursery term that cannot be used for the reconstruction of PIE phonology. This is confirmed by the fact that the word contains a geminate *-tt-, whereas the normal PIE lexicon never contains geminates. It is uncertain Indo-European Linguistics 7 (2019) 122-163 whether the Hittite word should be seen as directly cognate with the non-Anatolian forms, because similar Anatolian words like Hitt. anna-'mother' and HLuw. tata/i-'father' used to have o-vocalism in view of Lyc. ẽni and tedi.

2.3
Doubtful Anatolian etymologies In the following cases, the etymologies of the Anatolian words are too uncertain to be used as evidence for the reconstruction of the phonology of the protolanguage: 2.3.1 *Halbh-'white' (Weiss 2009: 41, Ringe 2017: 12) Hitt. alpaš 'cloud' , Lat. albus 'white' , Gr. ἀλφός 'dull white leprosy' . The etymology of the Hittite word is clearly based on the superficial formal resemblance to Latin albus, because the meanings of the words are rather far apart. Etymologizing the word for cloud as 'the white one' does not account for the fact that alpa-is predominantly associated with rain and thunder (Puhvel 1984: 38). The etymology obviously cannot be used as evidence for the reconstruction of the PIE phonemic system.
According to Melchert, the context "calls for 'confer/bestow upon, grant' ". However, other translations seem to be possible as well, including 'establish, fix' or 'make come true, realize' , in which case we would be dealing with a nufactitive of PIE *h2er-'to join' (see 2.2.2, cf. Hitt. āra-'right, proper(ly)') instead. Yakubovich (2017: 12, fn. 12) translates 'accord' , with the same etymology. In spite of Poetto's (1997) attempts to establish the meaning of the verb ārlanuwa-, which occurs twice in a single passage, as 'to bestow' , this interpretation remains uncertain due to the scant attestation of the verb, the fact that ārlanuwa-is used with the preverb anda-'in(to)' and the fact that the alleged Hieroglyphic equivalent 49a*-nu-wa/i-ha is used alongside pi-a/i(a)-ha 'I gave' , which makes it unlikely that 49a*-nu-wa/i-ha also meant 'I gave, bestowed' . Yakubovich's (2017) translation of ārlanuwa-as 'to replace, relocate' makes more sense. As long as there are uncertainties about the meanings of these words, their etymologies remain speculative and cannot be used in the present discussion.

2.3.3
*(h1)arg-u-'to plead a case' (Weiss 2009: 41) Lat. arguere 'to show, declare, accuse' , Hitt. arkuu̯ ae-zi 'to make a plea' . The primary meaning of Latin arguere is without doubt 'to make clear' , cf. argūtus 'clear, bright' , which is why it is traditionally connected with ToB ārkwi, Gr. ἀργύφεος 'white' , Skt. árjuna-'white, silver-coloured' etc. The meaning 'to show, declare' of the Latin verb appears to be a relatively recent development in view of the fact that the derivative argūtus preserves the older meaning 'clear, bright' . This renders the connection with the Hittite verb very uncertain. The latter would have to be an independent derivative from 'white, bright' , with a parallel development to 'clear' and subsequent derivation and further development to 'to declare' and finally to 'to make a plea' . This would be a remarkable coincidence. I think that the etymology is a mirage. Melchert (1998: 50) and Kloekhorst (2008: 205) maintained the etymology, however. Kloekhorst reconstructed *h2orǵ-u-i̯ e/o-for the Hittite verb, with loss of initial *h2 before *o, but the o-grade would be unexpected.

2.3.7
*aul-'tube' (Kimball 1994) Hitt. auli-'throat' or 'carotid artery' (?), Gr. αὐλός 'pipe, flute' , αὐλών 'hollow, defile, channel, pipe' , OPr. aulis 'shinbone' , Lith. aulỹs 'beehive' , aũlas 'leg of a boot' , Ru. úlej 'beehive' . The Hittite word denotes not only an organ, but also 'blood sacrifice' and 'sacrificial animal' . Kühne (1986) argued that there are two passages in which auli-denotes an organ associated with animal sacrifice that betray the meaning 'throat' or perhaps 'carotid artery' . On the basis of this interpretation, he proposed the etymological connection with Gr. αὐλός etc. (1986: 114). In fact, Kühne's translation is rather doubtful. The two crucial passages are the following: pronk Indo-European Linguistics 7 (2019) 122-163 1. ("When the cook prepares the sheep for sacrifice") UZUauliš šii̯ ezi (KBo 29.72 ii 13-14 and KBo 14.96 ii 11-12). 2. nu=ššan L Ú EN É-TIM ŠA UDU.ŠIR ŠA GUD.MAḪ=i̯ a aulii̯ a GÍR ZA-BAR-it QĀTAM dāi. "The lord of the house places on the auli-of the ram and the bull his hand with a bronze knife" (KBo 15.33 iii 10-13). The first example probably means "the auli-spurts [blood]", with an omitted object ēšḫar (Kühne 1986: 101, referring to ēšḫar šii̯ ati 'blood spurted' in KBo 3.16 verso 6-14). If this is correct, it is preferable to translate auli-as 'sacrificial animal' , which is a meaning that is well-established for auli-(cf. Kühne 1986: 107). In the second example, we are dealing with a ritual that takes place before the actual sacrifice. Although it is conceivable that the lord of the house places his hand on the place where the animal will be cut, this need not be the case. Because there are other attestations of auli-referring to an organ that cannot be the throat or the carotid artery, as Kühne (1986: 103-105) himself admits, there is actually little reason to think that in this passage it does mean throat or carotid artery. The etymological connection between Hittite auli-'blood sacrifice, sacrificial animal; some organ' and Gr. αὐλός etc. should thus be given up. The non-Anatolian words can be reconstructed as PIE *h2eul-o-.

*a in core Indo-European
None of the etyma with a secure Anatolian cognate turn out to contain a reflex of PIE *a. The following cases are etyma with (potential) cognates in Indo-Iranian and/or Tocharian, but not in Anatolian. These words cannot be borrowings from European substrate languages, but they could have entered Indo-European after the Anatolian branch split off.

3.4
*gras- (Sihler 1995: 153) Skt. gras-'to devour' , Gr. γράω 'to gnaw, eat' . The reconstruction with *a is based on the alleged connection with Latin grāmen 'grass' < *gras-men-. The Latin word is alternatively connected with Goth. gras 'grass' (see de Vaan 2008: 269 f. and Kroonen 2013: 187 for a discussion). Without the Latin word, the Vedic and Greek verbs have been reconstructed as *gres-, *grs- (LIV2). A reconstruction *grns-would, however, provide a better explanation for the Greek forms (van Beek 2013: 253). The Sanskrit superlative grásiṣṭha-, taken by LIV2 to refute *grn̥ s-, must be a recent formation in any case. Because there is an alternative pronk Indo-European Linguistics 7 (2019) 122-163 etymology for the Latin word that is at least equally plausible, we cannot use it as evidence for PIE *a.
A third scenario would be that the short reflexes of in ἅζομαι, ἅγιος, ἁγνός, yajñá-, iaiiūnus and iaiientāre reflect *(H)ieh2ǵ-with late Proto-Indo-European loss of the laryngeal due to the so-called "Wetter-Regel". This rule, which goes back to Schindler but was first argued for in print by Peters (1999), states that laryngeals were regularly lost before an occlusive and a resonant or glide, i.e. *CVHTR/I̯ -> *CVTR/I̯ -. Unlike in the scenario that operates with Lubotsky's law, almost all the attested forms of Sanskrit yaj-would have to have a secondary short vowel. Also, the amount of counter-examples against the "Wetter-Regel" speaks against the idea that the rule would have operated in Proto-Indo-European, and there seems to be insufficient evidence to suggest Indo-European Linguistics 7 (2019) 122-163 that the rule operated at a younger date within Greek, Indo-Iranian or Italic (Hackstein 2002: 226f., Müller 2007: 134ff., Zair 2012. For the present discussion, it suffices to observe that a reconstruction *Hiag-is warranted only if one rejects Lubotsky's law, secondary e/a-ablaut in Greek and the "Wetter-Regel".

3.16
*mad-'to be(come) wet' (Mayrhofer 1986 : 170, LIV2, NIIL 455 f.) The reconstruction with *a is based on the a-vocalism of Gr. μαδάω 'to loose hair' , Lat. madeō 'to be wet, soaked' , Skt. mad-, YAv. maδ-'to become intoxicated' and OIr. maidim 'to break, burst, gush' . Beekes (1988: 29) reconstructed *mh2d-, with regular vocalization to *mad-in Greek and Italo-Celtic (see 3.23 below on Gr. ἄστυ), while Lubotsky (1981: 135f.) listed the Indo-Iranian words as possible examples of loss of a laryngeal before a media followed by another consonant (Lubotsky's law, see 3.8). The Greek verb is often translated as 'to be moist' , but it only has this meaning in connection with a disease in fig-trees (Theophrastus) and as a medical term describing wounds. Its original meaning appears to have been 'to shed' , mainly of hair, which makes the connection with the Latin and Indo-Iranian verbs for 'to be soaked, to be intoxicated' unattractive. The connection with OIr. maidim 'to break, burst, gush' remains theoretically possible, but is not secure enough to play a role in the discussion about PIE phonology. Harðarson (1995) proposed a connection between μαδάω and μεστός 'satiated' , but this etymology has nothing to recommend itself.

*math-(Narten 1960, Oettinger 2017)
The reconstruction with -a-is based on the alleged connection between Skt. mathnati 'to rob, snatch away' and the Greek theonym Προμαθεύς, which was proposed with due caution by Narten (1960: 135, fn. 40) and accepted by Watkins (1995: 256, fn. 3). The Greek name would originally have meant 'fire-robber' in accordance with the myth about Prometheus stealing fire from the gods. The semantics of this etymology are of course appealing, but the formal side is difficult. Most importantly, there is insufficient evidence to support the sound law *th2 > Greek θ that is required for this etymology (cf. De Decker 2011). Also, Προcannot be a reflex of the word for 'fire' , PIE *peh2ur. It could eventually go back to a preverb *pro-, but this is very speculative as the required underlying verb is unattested in Greek. Vedic pra-math-hardly supports the etymology, because there is no indication within Indo-Aryan that pra-math-is an old formation. Moreover, there is a better etymology available for Skt. math-, viz. the traditional connection with the Skt. root manth-'to stir' < PIE *m(e)ntH-. The two verbal roots are synchronically distinct in Vedic, but the etymological connection between them is supported by the semantics of the Tocharian and Slavic cognates: ToB mänt-'to stir, destroy' (cf. Adams 2013: 486 f.), Cr. mésti 'to mix, disturb, trouble' . Skt. mathnati 'to rob, snatch away' can be explained from an earlier 'to disturb, destroy' , or even more directly from 'to stir' if it refers to a quick movement of the hand, as in English to whisk away 'to snatch away' . The forms with a preverb that Narten discusses, like vi-math-'to tear apart, pull back and forth' , all fit this semantic development. The zero-grade root *mn̥ tHis expected in the nā-present, cf. Skt. skabhnati to skambh-'to support' . The aorist mathīt and perfect mamatha must be inner-Indo-Iranian or inner-Indo-Aryan innovations based on this present. Προμαθεύς, like his brother Ἐπιμαθεύς, thus remains without etymology. It may well be a substrate name, like e.g. Ὀδυσ(σ)εύς and Ἀχιλ(λ)εύς with the same suffix (cf. Beekes 2014: 161-163; the attempts by e.g. Bader (1999: 44) and Nikolaev (2007) to provide Ἀχιλ(λ)εύς with pronk Indo-European Linguistics 7 (2019) 122-163 an Indo-European etymology require ad hoc solutions to account for the formal problems). If Προμαθεύς does consist of Indo-European elements, which now seems unlikely, there is no need to assume that any of those elements contained PIE *ā.

*a limited to the European branches of Indo-European (Italo-Celtic, Balto-Slavic, Germanic, Greek, Armenian, Albanian)
The following cases have been or could be claimed to contain PIE *a. A list of words containing *a that are likely to be borrowings from one or more European substrate languages is presented at the end of this paper.

*bhar(s)dh-o/ah2-'beard' (NIIL 4f.)
Lat. barba, OHG bart, OPr. bordus, Lith. barzdа, OCS brada. There are several irregularities, viz. Lat. b-instead of expected *f-and the sibilant in East Baltic. There is no consensus on the age or potential secondary origin of *s, which is only attested in part of the East Baltic cognates (cf. NIIL 5, Kregždys 2004). The Germanic forms can reflect either *bhardh-or *bharsdh-. The Slavic cognates cannot reflect *bharsdh-eh2, because the sibilant would have been preserved before a dental occlusive, cf. OCS prьstъ 'finger' < *pirsto-. Lat. barba probably cannot derive from a form with *s in view of turdus 'thrush' < *trsdh-, cf. Lith. strãzdas 'thrush ' . Van Beek (2013: 240, fn. 947) suggested that the word for 'beard' is a derivative from the PIE verbal root *bherdh-seen in Gr. πέρθω 'to raze, pillage, cut off' under the assumption that its original meaning was 'to shear, lop' and that the original meaning of the word for 'beard' was '(hair)cut' . However, as van Beek himself observes, this etymology leaves Latin barba (and the East Baltic forms with -z-) unexplained. Kroonen (2011Kroonen ( : 149-151, 2013 argued that the Latin and Balto-Slavic words are borrowings from Germanic *barzda 'beard; edge, brim' , which would be cognate with OHG brart 'edge' < *brazda and brort 'spear, edge' < *bruzda. This scenario does not account for the -b-of Lat. barba. The assumption that the Germanic word was borrowed into East Baltic before *-z-was rhotacized also poses chronological problems. It seems most likely to me that we are dealing with independent borrowings of a word for 'beard' from an unknown adstrate or substrate language (thus Schrijver 1991: 488, Derksen 2015. In any case, the word provides no evidence for PIE *a.

*(ḱ)agh-'to grasp, enclose' (LIV2: "auch *(ḱ)h2egh-wäre möglich")
Lat. caulae 'railing or lattice barrier' , MW kae 'hedge, fence' , ON hagi 'pasture' , OHG hag 'hedge, enclosure' . The noun for 'enclosure, hedge, fence' might be a derivative from the verbal root reflected in Umb. ku-kehes 2/3sg.fut. 'to take/get' , MW kehy 'to receive' , as suggested by, e.g. LIV2, but this does not seem obvious to me. Combinations of a tenuis and a media aspirata are not normally found in inherited Indo-European roots, which makes it likely that the word for 'enclosure' was borrowed from a non-Indo-European language of Europe (de Vaan 2008: 123) and thus provides no information about PIE phonology. Gr. κάμπτω 'to bend' , Lith. kum̃pti 'to bend' , kam̃pas 'corner, angle, handle' , Latv. kampis 'curved piece of wood, hook' , OCS kǫtъ 'corner' < *k(o/a)mp-to-. The appurtenance of Lat. campus 'field' and Goth. hamfs 'maimed' is less certain because of their meanings. The most plausible cognates are limited to two branches, Greek and Balto-Slavic, both spoken in the centre of the Indo-European world. In spite of the obvious restrictions this puts on the value of the etymon in the present discussion, the connection remains rather attractive. A reconstruction *kh2emp-, which Beekes (2010: 632) mentions with due scepticism, should have produced initial *x-in Slavic, while a reconstruction *keh2mp-would not account for the non-acute intonation of the Baltic words. Kroonen (2013: 207, 257) reconstructs *kep-with a nasal present *ke-n-p-under the assumption that the nasal in Greek is "secondary". A possible scenario would be to assume an ablauting paradigm *kemp-, *km̥ p-(= Lith. kum̃p-) > *kap->> Gr. καμπ-. A parallel case of levelling of this type is found in χανδάνω 'to hold' < *ghn̥ d-, *ghend-. I find this scenario more attractive than a reconstruction *kamp-, *km̥ p-or the idea that we are dealing with borrowings from the same or similar substrate languages (thus Beekes 2010: 632).

4.10
*knauk- (Strunk 1993) According to Strunk, Lat. (non) nauci/nauco '(not a) bit, straw, dime' is cognate with nux 'nut' < *knuk-. This is based on the meaning 'shell' that is attributed to nauci/nauco by grammarians and glossators. Surprisingly, this proposed meaning is found later than Plautus's admission that he does not know the actual meaning of nauci (Most. 1042), and the explanations of the word as 'shell' often specifically refer to nucis 'nuts' (Strunk 1993: 426). One therefore gets the impression that the interpretation of nauci/nauco as 'shell' is due to folk etymology. I am not convinced by Strunk's etymology, but even it is correct, it does not go back to a PIE form with *a, because the word for 'nut' cannot be reconstructed for Proto-Indo-European (cf. de Vaan 2008: 420 f.).

*lap-'to lap, lick' (Kroonen 2013: 327, labelled as "European")
Lat. lambō, Gr. λάπτω, OE lapian, OHG laffan, Lith. lapènti, Ru. lópat' 'to gobble' , Arm. lap'em. As de Vaan (2008: 324) observed, the root could also be reconstructed with *h2. It is generally assumed that the verb is in origin onomatopoetic. The reconstruction is therefore difficult, as secondary distortions are to be expected. Cf., e.g., Arm. lap'em with unetymological -p'-(or < *-ps-?). In any case, I consider it a possibility that there was a (late) PIE root *lep-, with Lat. lambō < *lp-n-like pandō 'to spread out' < *pt-n-and Gr. λάπτω < *lp-ie/owith the expected zero-grade in a ie/o-present. An e-grade of the root is found in Latv. lepêt 'to slurp, gobble' . The other Balto-Slavic forms and perhaps Germanic would reflect an o-grade *lop-. Guus Kroonen points out to me that the Germanic forms go back to an ōn-verb *lap(p)ōn-, where one would expect zero-grade of the root (cf. Kroonen 2012: 275). The vocalism of *lap(p)ōn-therefore points to *lHP-n-, unless it was created secondarily on the basis of the long vowel preterit, cf. OHG laffan, pret. luof. The alternative to a reconstruction *lep-is *lh2bh-/*labh-, cf. Gr. λαφύσσω 'to gulp down' , in which case the Balto-Slavic forms cannot be directly related. There are no compelling reasons to reconstruct PIE *a in this etymon.

*masd-'mast' (NIIL 463)
OHG mast 'stick, pole, mast' < *mo/asto-or *mo/asdo-, MIr. maide 'stick, pole, staff' < *ma(s)d(h)i̯ o-. It cannot be ruled out that Germanic borrowed the word from Celtic or vice versa. Lat. mālus 'pole, mast' is also often connected with these words, in which case it might reflect *masdo-with l < *d. Another cognate is perhaps Slavic *mostъ 'wooden paving placed over a stream or marshy land' < *mo/astu-. It seems less likely that the Slavic word was borrowed from Germanic in view of its meaning (Pronk-Tiethoff 2013: 199) and the fact that it appears to have been a u-stem (cf. Ru. loc.sg. mostú, Sln. gen.sg. mostȗ, nom.pl. mostọ vi, etc.). If the Germanic, Middle Irish and Latin words are inherited, they must go back to a proto-form *masd-or *mh2sd- (de Vaan 2008: 361). It is, however, conceivable that we are ultimately dealing with post-PIE borrowings from a non-Indo-European source.

4.15
*radh-'to shine' (Schaffner 2010) Lat. radius 'ray of light, spoke' , ON rǫðull 'radiant circle' , OE rador 'ether, sky' . Schaffner reconstructs a verbal root, but the attested forms are all nouns. The connection of these words with the name of the Vedic demon rāhu-that Schaffner proposes is conjectural and can only be entertained if it has been shown that the etymology is formally possible. The Italo-Germanic connection, on the other hand, is conceivable, but weakened by the fact that it is limited to two branches. An alternative reconstruction to Schaffner's would be *HrHdh-. For the vocalization of the second laryngeal cf. Lat. ratiō 'reason' , Goth. raþjo 'account' < *h2rh̥ 1-ti-. Assuming that the etymology is correct, there is no compelling reason to prefer a reconstruction *Hradh-over *HrHdh-.  Schrijver (1991: 431) argued that Lat. scabō may go back to *skebh-, ablauting with scobis 'filings' < *skobh-. However, Schrijver's sound law *ke > Lat. ca remains uncertain due to a number of counterexamples (Meiser 2006: 82f.). De Vaan (2008: 541) also reconstructs the root as *skebh-, with Lat. scabfrom the zero-grade *skbh-, which received an epenthetic vowel in the adjective scaber 'rough' < *sk ə bh-ro-(with expected zero-grade) as in quadru-'four-' < *kw ə t-ru-. If Gr. σκάπτω is cognate (which is not altogether certain, cf. Beekes 2010: 1342), it would have to reflect a form with a nasal infix: *sk-n-bh-. If the reconstruction *sk(e)bh-is correct, the root may also be reflected indirectly by the synonymous *skrebh-'to scratch, scrape' (OE sceorfan, Ru. skrestí, Latv. skrabt), with initial *skr-due to contamination with *(s)ker-'to cut, shave, scratch' (Gr. κείρω, ON skera, Arm. kcerem etc.). The alternative reconstruction *skh2ebh-offered by LIV2 and NIIL looks somewhat unusual, but cannot be ruled out. A reconstruction *skeh2bh-would not account for Lat. scobis 'filings' , Lith. 3pres. skãb(i)a 'to carve' or Ru. skóbel' 'scraper' . There are no compelling reasons to prefer a reconstruction *skabh-over the alternative reconstructions.
The conclusion that Proto-Indo-European did not have a phoneme *a or *ā also has implications for our reconstruction of the evolution of the vowel systems of the individual branches of Indo-European. The fact that the European branches of Indo-European borrowed extensively from one or more contact languages that had a vowel /a/ raises the possibility that these contacts played a role in the introduction of /a/ in the phonemic system of Italo-Celtic, Greek and Balto-Slavic (cf. Kortlandt forthc.), i.e. those branches which created two new vowels *a and *ā that were distinct from the reflexes of PIE *e, *o, *ē and *ō. In Greek, the rise of /a/ was preceded by the phonemic merger of *e with the prop vowel that had developed in consonant clusters containing a laryngeal, e.g. *ph2tēr > *[ph2 ə tēr] > *ph2etēr > πατήρ. In Tocharian, the sequences *h2e (> PToch. *a) and *eh2 (> *ā > PToch. *o) and vocalized laryngeals (> PToch. *a) also remained distinct from PIE *e, *o, *ē and *ō.9 However, the fronting of PIE *o to PToch. *ae is perhaps easier to understand if there was no short vowel *a yet before this fronting took place. In Germanic and Indo-Iranian, the PIE system with only two open vowels was preserved for some time and the position of a central open vowel was eventually occupied by PIE *o and PIE *e and *o respectively. In Germanic, the vowel /a/ in loanwords was borrowed with the timbre of the reflex of PIE *o, which was apparently an open vowel at the time of borrowing. The more open allophones of *e in the position after *h2 and *h3 also merged with this open vowel. In Indo-Iranian, /a/ and /ā/ developed regularly from PIE *e, *o, *ē and *ō and later also from the reflex of the syllabic nasals. In Anatolian, PIE *eh2 merged with the reflex of *ō, which was apparently an open vowel, into *ā. PIE *h2e became *ḫa, of which the vocalic part remained distinct from short *o.
The brief overview above shows that there is nothing in the evolution of the vowel systems of the individual branches of Indo-European that precludes the reconstruction of only two open vowels for the proto-language: one front vowel, approximately [ae], and one back vowel, approximately [ɒ].