07 October 2014

Two Is Company, Four Is a Party

Neuter nouns with the suffix *-wr̥/*-w(e)n- are relatively rare in most branches of Indo-European. The only group where they can be found in great numbers is Anatolian. In Hittite, the suffix productively  formed verbal nouns (names of actions), but there are also examples of nouns that had  become independent lexical units, no longer bound to a particular verb paradigm. They had usually acquired a concrete meaning (referring to a thing or substance rather than an abstraction). One of such nouns is Hitt. pahhur/pahhuen‘fire’, evidently an ancient word, preserved in many branches of the family and showing evidence of archaic vowel alternations and mobile stress: nom/acc.sg. *páh₂wr̥, gen.sg. *ph₂wéns, etc. It may be etymologically connected with the verb *pah₂- ‘guard, protect’, but it’s doubtful if even the speakers of Hittite were still aware of any such connection: the semantic distance between the verb and its derivative was already too great.

Outside Anatolian, the suffix does not play any major role. The nouns that contain it are scattered remnants of a Proto-Indo-European pattern of word-formation. Their attestation is very uneven. They are quite well represented in Sanskrit and Greek, but only isolated examples are found elsewhere (the ‘fire’ word, which became part of Indo-European basic vocabulary sufficiently early, is exceptionally well attested). Here are a few typical *-wr̥/*-w(e)n- nouns evidently connected with known verb roots:

  1. *h₂árh₃-wr̥, gen. *h₂r̥h₃-wén-s  ‘arable land’ (root *h₂arh₃- ‘till, plough’);
  2. *snéh₁-wr̥, gen. *sn̥h₁-wén-s ‘string, sinew’ (root *(s)neh₁- ‘spin, twist’);
  3. *séǵʰ-wr̥, gen. *sǵʰ-wén-s ‘steadfastness’ (root *seǵʰ- ‘conquer, take possession of; hold, own’);
  4. *h₁éd-wr̥, gen. *h₁d-wén-s ‘food’ (root *h₁ed- ‘eat’).

Their reflexes in the historically documented languages rarely display the whole range of vowel, consonant and stress variations, most of which were levelled out analogically in prehistoric times. Still, these alternations are reconstructible thanks to the fact that different fragments of the pattern have been preserved in different languages. They can be reassembled into a complete picture like the pieces of a jigsaw puzzle or the disarticulated skeleton of a fossil animal.

Got wheels?
A four-wheeled toy from the Cucuteni-Trypillian culture;
the early fourth millennium BC.
Neuters of this kind formed collectives by inserting a lengthened *ō into the suffix. The collective of a count noun denotes simply a set of objects (a collective plural), while the collective of a mass noun like ‘fire’ denotes a particular quantity or sample of the thing in question (‘a fire, a burning mass’). This became one of the derivational mechanisms by which Indo-European mass nouns could be transformed into count nouns. The accent was commonly shifted to the suffix in the process, causing the reduction of the root vowel: *páh₂wōr (collective) > *ph₂wṓr > *pwṓr (a countable neuter with its own case forms such as gen.sg. *p(h₂)un-és). Still later, the distiction between the original mass noun and its collective could be blurred and abandoned, the younger form ousting the older and serving in both functions (‘fire’ or ‘a fire’). The archaic Proto-Indo-European form *páh₂wr̥ is unambiguously preserved only in Anatolian, while the remaining Indo-European languages show reflexes of *pwṓr or its further modified descendants.

Now we can view the reconstruction *kʷét-wr̥ in this light. Supposing it was derived from our hypothetical verb root *kʷet- ‘group into pairs’, the original meaning of *kʷétwr̥ (as a nomen actionis) would be something like ‘pairing’, and its collective *kʷétwōr would mean ‘a particular result of pairing, a complete set organised into pairs’. In the Proto-Indo-European world, there were many “natural” sets of things conceptualised as consisting of two pairs: human hands and feet; fore and rear legs of animals; the wheels of a wagon; the four directions, whether cardinal (east and west, north and south) or relative (forward and backwards, left and right); paired organs of perception (two eyes and two ears). This could have provided sufficient motivation for treating ‘4’ as the prototypical case of an “even collective”. An interesting parallel can be seen in the “fraternal” numeral systems widespread in Amazonia. In the languages that employ them, the numeral ‘4’ is derived from an expression meaning ‘each has a brother/companion/spouse’. At a more primitive stage, preserved in the Dâw language, there are only three “exact” lexical numerals, ‘1’, ‘2’, and ‘3’. The values from 4 to 10 are described as ‘even’ (‘has a brother’) or ‘odd’ (‘has no brother’). The precise value can’t be expressed linguistically, but the words ‘even’ and ‘odd’ can be supplemented by clarifying hand gestures:
Dâw speakers indicate ‘four’ by holding the fingers of one hand separated into two blocks; for ‘five’, they add the thumb; for ‘six’, they place the second thumb against the first to make a third pair; and so on until for ‘ten’ all fingers are grouped into five pairs, the thumbs together.
[Epps 2006: 265]
Once established as a concrete numeral (rather than part of an even-odd tally system), *kʷétwōr (or *kʷətwṓr) was interpreted as an ordinary neuter plural, and – like the numerals ‘1’, ‘2’, and ‘3’ – formally an adjective, inflected not only for case but also for gender. This resulted in the analogical creation of the animate plural in *-wor-es (and the periphrastic feminine ‘four females’, soon univerbated and phonetically mutilated in the process). Note that if the adjective had been formed directly from the verbal noun *kʷétwr̥/*kʷ(ə)twén-, its animate plural would probably have ended up as *kʷet-won-es. In addition to the Greek and Vedic words for ‘fat’, already discussed, compare Greek peîrar (gen. -atos) ‘boundary’ < *pér-wr̥/*pr̥-w(e)n- versus the Homeric adjective a-peírōn (animate) ‘boundless, endless’ < *n̥-per-wōn.

All this suggests that the word *kʷétwr̥ (coll. *kʷétwōr) was transparently derived from a verb root and adopted as a cardinal numeral at a rather late date, perhaps in “Core Indo-European” (the non-Anatolian part of the family) rather than in Proto-Indo-European proper. It is a well-known fact that Anatolian has a different word for ‘4’, *meju- (Hittite meu-/meyau-, Luwian māwa-). Since the jury is still out on whether Hittite kutruwa(n)- ‘witness’ has anything to do with the numeral ‘4’*), we should seriously consider the possibility that the familiar reconstruction *kʷetwores is not Proto-Indo-European at all but represents a “dialectal” innovation which replaced its older synonym in the common ancestor of Tocharian and the extant branches of the family.

If this were a journal article rather than a blog post, I would now be obliged to account for every puzzling irregularity in the branch-specific reflexes of *kʷetwores and its variants. I will spare my visitors such excruciating details, but if anyone is really interested in discussing them, welcome to the Comments section.

And now back to other matters – next time.

*) A witness in court could be denoted as ‘the fourth man’ (beside the two contracting parties and the judge).


Epps, Patience. 2006. “Growing a numeral system: The historical development of numerals in an Amazonian language family”. Diachronica 23(2): 259-288. [a preprint version is available here]


  1. [I am going to address some of the points made by Douglas G. Kilday here. They are more closely connected with the proposals made above, so I have decided to redirect the discussion to the Comments space of this post. First, the curious vocalism of the Slavic cardinal numeral '4'.]

    Douglas: I regret the early closure of this thread. I was eagerly awaiting the non-laryngeal explanation of the acute in Proto-Slavic *c^etýre.

    One important type of non-laryngeal acute is that found in Balto-Slavic vr̥ddhied derivatives. In roots ending in VR a morphologically lengthened vowel gets acuted before a vowel-initial suffix. I find the evidence paraded by Miguel Carrasquer Vidal convincing, and his phonological explanation vastly preferable to other solutions. It explains why we find the acute in cases like *sla̋va ‘fame, glory’, where the root is evidently aniṭ (laryngealless), cf. *sloves- ‘word’ < *ḱlewes-. The question is only why the vowel in the Slavic word for ‘4’ became lengthened: such a lengthening is conspicuously absent from Baltic *ketur-. Let’s first observe that Baltic has only one relict of the original consonantal stem: the accusative (masculine), Lith. keturì < *keturins < *kʷetur-m̥s. No nominative forms survive in any Baltic language. In Slavic, the same accusative (but with a mysteriously lengthened and acuted vowel) is the source of the nom./acc. (f./n.) *četyri, but in addition we have the nom. (m.) *četyre for expected *četvore < *kʷetwores (the strong form of the stem).

    Neither Baltic nor Slavic has any direct reflexes of the old neuter *kʷetwōr. Let’s try to “reconstruct forward” the hypothetical Slavic outcome. Scanty as the evidence is, one would expect the same sonorant loss, circumflex accent and vowel raising as in the types represented by OCS kamy ‘stone’ and mati ‘mother’ (nasal and rhotic stems, respectively):

    *kʷetwōr > *ketwō̃ > *ketwū > *ketū > PSl. *čety

    This would have led to excessive variation in the paradigm of ‘4’: *četvor-e/*četur-ī/*čety, had the development been regular. My suggestion is that already in pre-Proto-Slavic the short *u of the weak cases was lengthened on the analogy of the neuter form; eventually the nom.pl. masc. adopted the same vocalism. A similar substitution took place in a few parallel cases (I’ll keep the examples to myself for the time being – it’s something I intend to publish). This lengthened medial *ū acquired an acute intonation as predicted by Miguel. By Late Proto-Slavic times this vowel was generalised in all the forms of the cardinal number, and the old neuter form had been replaced by *četyri.

    1. Aaargh! How much I miss an Edit button! A self-correction:


      should be:


  2. P.S. I forgot the reference:

    Carrasquer Vidal, Miguel. 2013. "Balto-Slavic long vowels". Baltistica 48 (2): 205-217.

  3. Douglas: De Saussure's Effect would have deleted *h1 from /o/-grade, *kWeth1-wor- > syllabic *kWet-h1wor- > *kWet-wor-.

    The *o appearing in posttonic suffixes has a different origin from apophonic *o in roots, and may not behave in quite the same way with respect to the Saussure effect. Anyway, as shown by Andrew Byrd, the Saussure effect operates only in sonorant environments. Word-medially, it happens between a liquid and a nasal: H > /oR_N.

    Douglas: Original heteroclitic morphology of 'four' is very unlikely due to the complete absence of /n/-forms. No such absence is found with 'fire' and 'water' even though no non-Anatolian language preserves the original declension.

    But 'fire' and 'water' are part of basic vocabulary. They are used orders of magnitude more frequently than words with meanings like 'pairing', and are much more resistant to lexical replacement. Despite of that, the n-forms of 'fire' survived only marginally, and only thanks to the fact that the original locative in *-wen(i) was not completely abandoned (the best witness is Germanic). Note that even Greek, which has retained many heteroclitic stems, has only -r- in 'fire'.

    And in the case of '4' we are not really talking about the heteroclitic noun *kʷét-wr̥, but about its collective which became a numeral, severing all ties with its etymological source. As a numeral, it never had anything else but *-r.

    1. Correction and apology: I cited Byrd from memory and I misrepresented his views. The version of the Saussure effect that he subscribes to seems to be:

      H > /{oR_$, $_Ro}. I'm not sure what Andy, or anyone else who has discussed the issue, would say of the effect of posttonic *o -- they always focus on the root vowel. The collective suffix *h₂ is lost after sonorants (with the compensatory lengthening of the vowel preceding the sonorant), but this loss, treated as part of Szemerényi's Law, does not depend on the colour of the vowel.

  4. I fail to see how Miguel's paper justifies getting *c^etýre by substitution of the final vowel from *c^ety, which would be circumflex (*ketû < *ketwô < *kWetwo:r). Miguel's explanation of 'salt' and 'cow' has circumflexes regularly arising from the root-final resonants, and these were maintained when the root-nouns were transferred to the /i/-stems.

    Miguel rightly rejects long grade in *bé´rme, 'burden' and must follow Derksen in positing a set.-extension to *bHer-. This is not ad hoc as shown by Lat. _praefericulum_ and Grk. _phéretron_ (from *bHerh1-). The root *k^leu- also has set.-behavior in Grk. _klûthi_ 'hear thou!' and Gmc. *xlu:da- 'easily heard, loud' (which shows that Dybo's Law did not apply before obstruents). Both Skt. _bhr.-_ and _s'ru-_ are among the eight verbs which must omit the connecting-vowel -i- with consonant-initial terminations in the perfect. It seems that *-h1, whatever it meant ('distinctly, separately' vel sim.?) was a fairly rare root-extension, but pre-Vedic Skt. still had to distinguish between the reflexes of *bHer- and *bHerh1-, *k^leu- and *k^leuh1-, etc. With the eight verbs, it was necessary to keep the set.- and anit.-forms separate until the former became obsolete as independent verbs.

    Skt. _pa:vakah._ 'fire' (Upan. etc.) probably reflects PIE *pah2wn.-kó- whose form appears in Vedic _pa:vaká-_ 'clean, pure, bright', evidently confused with *pava:ká- 'id.' (indicated by metrics, KEWA 2:264) from _puná:ti_, _pávate_ 'cleans, purifies'. The WGmc 'funk' word meaning 'tinder, kindling, spark' etc. (OHG _funcho_, MLG _funke_, Eng. dial. _funk_) appears to be a weak noun based on Gmc. *funkka- < PIE *ph2un-k-nó-, with Upper Ger. dial. _Fanke_, _Vanke_ from secondary /o/-grade by analogy.

    Andrew Byrd and the others seem to have published no objection to de Saussure's Effect occurring in the second part of a compound word, which is what I consider 'four' to be. Given the syllabication *kWet-h1wor-, deletion of *h1 is parallel to that of *h3 in one of the original examples, Grk. _loigós_ 'ruin, plague, reduction of many to few' against _olígos_ 'few'.

    1. What I suggest is the generalisation of length, not of the intonation. The newly lengthened vowel would have become acute in a medial syllable in accordance with Miguel's rule.

      The acute in *slava can hardly be due to a laryngeal following the semivowel. We need a process which makes a lengthened vowel acute without a tautosyllabic laryngeal, and that's what Miguel's solutioin offers.

      I have retracted my objection to the operation of Saussure's effect in this position, but it doesn't really matter for explaining the reflexes of '4'.

  5. I have a question of no real relevance that I was wondering if you had a quick answer to put down - my not-a-professional-being means I'm not in too good of shape to answer my own question sufficiently at the moment.

    I was reading about correspondences between Japanese and Chinese and happened to think about PIE soon enough after reading some minutia on the origin of Mandarin's and the Japanese shift /p/ > /ɸ/ > /h/. I also thought of some labializing effect of the /β/ phoneme in some varieties of Spanish and the /v~u/ in Romani.

    I was wondering if the apparent weirdness of PIE's labial system at the time (the distribution of *p *b and *bh) could be due to one of the labials softening into *h3 (hence it's "labializing" qualities) at a pre-PIE stage? Possibly either *p > *ɸ with *b > p or *b > *β.

    I think a good dismissal would be serious and especially chaotic violations of phonotactics. I unfortunately don't have a good dictionary with me, and the ones I normally look at are down, at least on my end.

    1. It has been argued that PIE *m and *w are more frequent in roots than one would expect, so one (if not both) of them may have absorbed the original occurrences of *b. They also pattern distributionally with the obstruents: the clusters *mr-, *ml-, *mj-, *wr-, *wl-, *wj- are permitted word-initially, as opposed to *nr-, *nj-, *nw-, *jl- etc. The usual explanation is that the pre-PIE mediae had some unusual articulatory feature (lenis implosives?) making them akin to sonorants. I wouldn't exclude some plosive sources for the PIE "laryngeals", but what makes me doubt pre-PIE *b (or the like) > *h₃ is its root-initial occurrence before *m, as in *h₃meiǵʰ- 'piss, sprinkle' -- an environment that is particularly hostile to labial stops.

    2. P.S. The (basically unidirectional) historical trajectory /p, pʰ/ > /f, ɸ/ > /h/ > zero is of course common cross-linguistically. Celtic, Spanish and Gascon, Armenian, and Japanese are familiar examples, but Paul Foulkes lists plenty of other examples from numerous language families in this article:

      Foulkes, Paul. 1997. Historical laboratory phonology -- investigating /p/ > /f/ > /h/ changes. Language and Speech 40(3), 249-276.

    3. I was wondering if the apparent weirdness of PIE's labial system at the time (the distribution of *p *b and *bh) could be due to one of the labials softening into *h3 (hence it's "labializing" qualities) at a pre-PIE stage?

      I'm reminded of the Moscow School Nostraticist correspondences, where PIE *p *b bʰ come from Proto-Nostratic *pʼ *pʰ *b, respectively; of the latter three, [pʰ] is the most likely one to disappear.

      what makes me doubt pre-PIE *b (or the like) > *h₃ is its root-initial occurrence before *m, as in *h₃meiǵʰ- 'piss, sprinkle' -- an environment that is particularly hostile to labial stops.

      How common is this, actually? I'd like to know what you think of this paper; I don't have access, but this earlier one reconstructs a PIE root with initial *ts- and proposes that this cluster arose by compounding (an unidentifiable noun root ending in *-t followed by a verb root with *s-). If so, how common is *h₃m-?

      On the other hand, of course, if *h₃- was some kind of rounded labial approximant, how did it stay distinct from *w-, which notably had no effect on surrounding vowels (or the queen's wedding would be the quoon's wadding, as someone put it on a Wikipedia talk page)?

    4. I'm not so sure the process described by Piotr was always unidireccional. For example, if as I think, IE *ped- 'foot' is related to Macro-Caucasian *Hutʼ- (reconstructed from Burushaski and East Caucasian data), it would mean *Hʷ > *p.

    5. I think that's a pretty good reason not to consider this word a loan, actually, but rather a chance resemblance. It's a pretty basic word that I wouldn't expect to be borrowed in the first place, too...

    6. Sorry, I meant these words are long-rate cognates, not a Wanderwort like e.g. *h₁ek´w-o- 'horse'.

      On the other hand, Latin catēna 'chain' is an Etruscan loanword which in turn would have been borrowed from a satem reflex of IE *kʷet-. Surely also belong here caterva 'crown, troop', Oscan kateraamu, caterahamo ‘(that) we form a troop’ (2nd pres. subj.) and Middle Irish cethern ‘crown, troop’.

      Apparently, Lithuanian kek(e)tà 'detachment, flock' would be a reduplicate form of this lexeme, with long-range cognates in Uralic *kakta ~ *kæktæ ‘two’, Altaic *gàgtà ‘one of a pair’. In my opinion, the IE numeral *Hok´te-h₃(u) '8', a fossilized dual corresponding (either by cognacy or borrrowing) to Kartvelian '4' (=2x2) would also belong here. The disparity of sound correspondences would indicate *kʷet- and *Hok´te-h₃(u) originated in different proto-languages/lexicon layers (Kurganic and pre-Kurganic), something which is quite common in IE

      So I guess the original pre-Kurganic '4' shifted to '8' by some reason and a new numeral was coined from the lexeme *kʷet-, of Kurganic origin.

      On the other hand, Anatolian '4' would be related to Etruscan mach '5', possibly from a Macro-Caucasian word for 'hand', so Kassian's proposed long-range cognates would be wrong.

    7. David: How common is this, actually?

      Not very common. In fact, I can't think of any other verb in *h₃m- (though, on the other hand, both he 'piss' root and its initial laryngeal are secure). But there is an even stronger argument against *h₃ < *b: the third laryngeal can often be found in the same root with a media despite the well-known co-occurrence restriction. Examples include some of the most common IE roots, such as *doh₃- 'give', *ǵnoh₃- 'know', *h₃od- 'stink', *h₃neid- 'abuse', *h₃reǵ- 'stretch, rule', gʷelh₃- 'want', and *gʷerh₃- 'devour'.

      I'd like to know what you think of this paper...

      Guus has kindly sent me his articl, and I'm going to read it. If you'd like a copy, just e-mail me: gpiotr{at}wa.amu.edu.pl.

    8. Octavià Alexandre: Apparently, Lithuanian kek(e)tà 'detachment, flock' would be a reduplicate form of this lexeme, with long-range cognates in Uralic *kakta ~ *kæktæ ‘two’...

      The Lith. word, which can also mean 'cluster (of berries, nuts, flowers)', can hardly be separated from kẽkė, which means the same but has no t.

    9. Sorry, I meant these words are long-rate cognates

      ...Are you... sure you've figured out the Borean sound correspondences well enough to tell 1) that PIE *p corresponds to Proto-Macro-Caucasian *Hʷ, 2) that this correspondence goes back to Proto-Borean *Hʷ rather than *p, and finally 3) that the Moscow School is wrong in deriving PIE *p from Proto-Nostratic *pʼ?

      If so, why haven't you published? If you need help writing up a paper and sending an extended abstract to Nature, do let me know – this is not sarcasm.

      Guus has kindly sent me his articl, and I'm going to read it. If you'd like a copy

      Yes, please! I'll drop you an e-mail.

    10. David, I don't think entities such as "PIE", "Proto-Borean" or "Proto-Nostratic" were exactly what mainstream linguists think, neither their genealogical trees are correct.

      That said, I've got the impression labialized consonants were reduced to labial stops in some Eurasiatic (aka "Borean") languages, including part of IE and Altaic.

    11. ...So, basically, your answer to my first paragraph is "yes"?

    12. Yes, but I'm afraid Moscow School's taxonomy is very confusing and misleading. To begin with, the deepest node in chronological terms is Macro-Caucasian, from which (following Kerns) presumably sprang the real Nostratic to the south (Taurus-Zagros mountains) and Eurasiatic to the north.

      So if C represents a voiceless consonant (either fricative or plosive), then Macro-Caucasian *Cʷ > Eurasiatic > IE *p. But by no means all the IE lexicon is of Eurasiatic origin. For example, IE *h₁ek´w-o- 'horse' is a Caucasian Wanderwort (where IE *k´w ~ *ʃw) from a lexeme which originally designated a Pleistocene ungulate (cfr. Kartvelian *eʃw 'pig' from Nostratic). As expected, the Eurasiatic outputs have *pʰ, reflected as such in Altaic as well as in e.g. IE *(w)eper-o- 'boar' and *kapro- 'male goat' (probably there are more than just one Eurasiatic language embedded in the IE fabric).

    13. Sorry for the long silence. My offer stands, except I'll probably be very busy to the end of January.

    14. I think voicelesss fricatives and their corresponding aspirated stops can be regarded as isomorphic obstruents, with some languages preferring the former and others preferring the latter. So it doesn't matter which ones came first (as in the tale of the chicken and the egg) as long as they're interchangeable. Likewise, voiced fricatives and "voiced aspirated" stops are also isomorphic.

  6. Oops! I meant the IE word for 'horse' is a Wanderwort shared with Caucasian and Sumerian (where it means 'donkey') and ultimately derived from a "Nostratic" lexeme which originally designated wild ungulates. Sorry for the off-topic.

    On the other hand, Latin catēna 'chain' is an Etruscan loanword which in turn would have been originated in a satem reflex of IE *kʷet-. It looks like the ancestor of Etruscan was in contact with IE-satem languages, and more specifically, Baltic.

    1. ...Do you have any evidence for your second paragraph other than this one word?

    2. Yes, I have. Latin caesius 'light blue' is an Etruscan loanword with correspondences in caisie-, caisie-, ceisi(e)-, caiz-na, ceiz-na. The Etruscan forms can be linked in turn to Lithuanian gaĩsa-s ‘glow, redness in the sky’, Latvian gàiss ‘air, wheather’, gàišs ‘bright, clear’, gaisma ‘light’.

      Also Etruscan marth 'bride' has correspondences in Lithuanian martì ‘bride, daughter-in-law’, Latvian mā̀rša ‘brother's wife’.

    3. Also from a satem reflex of IE *kʷer- 'to make' (Sanskrit karóti, kr̥ɳóti), we've got Etruscan car- 'to build, to make', ceriχu- 'to build', as well as Leopontic karite '(he) made' and Cisalpine Gaulish karnitu '(he) built'.

    4. You are comparing CVC roots in which the consonants are only roughly similar, the vowels "ne font rien", and the morphological context is conveniently ignored. How do you want to convince anyone that the matches aren't accidental?

    5. Let's take a look at it from a different angle. We've got two similar verbs in Lepontic and Cisalpine Gaulish with no convincing Celtic etymology but which can nicely be explained as loanwords from an extinct IE-satem language. However, in this case I'm not sure if Etruscan acted either as an intermediate or an independent receptor.

      In the case of verbs morphology plays a secondary role, contrarily to nouns. For example, Latin catēna is most likely an Etruscan loanword because of the suffix -na, presumably a collective.

      But as regarding "accidental matches" meeting the above criteria, you'll find plenty of them in Pokorny's and other IE dictionaries. Take for example Latin manus and Greek márē, both meaning 'hand'.

    6. Why no convincing Celtic etymology? I'm not sure about the Lepontic verb but Cisalp.Gaul. KarniTu is rather obviously denominal: *karne-je/o-, from Proto-Celtic *karno- 'heap of stones, landmark, burial mound' (see English cairn, borrowed from Gaelic). There's zero evidence of any connecton with *kʷer- 'cut, shape, make' or with anything Etruscan.

      Since *-no-/*-nah₂ is an extremely common IE suffix, there's nothing compellingly Etruscan about catēna. BTW, why do you think Etruscan -na forms collectives? The less known a language, the easier it is to spin imaginative yarns about it.

      I don't understand your point about Gk. márē. If it's an authentic word at all, it can hardly be related to Lat. manu.

    7. In my opinion, the purported connection between karnitu and *karno- is rather weak, and also the verb isn't attested elsewhere in Celtic apart from Lepontic. This would indicate a loanword from a non-Celtic language, most likely an IE-satem one giving the striking Sanskrit parallels.

      On the other hand, Etruscan -na would be an attributive suffix rather than a collective, and it also appears in other borrowings to such as persona 'mask' < Etruscan *phersu-na, from phersu 'masked character'. In fact, Latin has a lot of Etruscan loanwords, reflecting the deep influence of the Etruscan culture upon ancient Rome.

      The puroprted relationship between Greek márē and Latin manus is in Pokorny's.

    8. BTW, Transalpine Gaulish has also a bunch of non-Celtic loanwords (also with no convincing Celtic etymologies) which I've investigated myself, although this is another matter. :-)

    9. The puroprted relationship between Greek márē and Latin manus is in Pokorny's

      By modern standards, half of the stuff in the Wörterbuch can't be taken seriously (and the rest would have to be practically rewritten to bring it in line with the current state of the field).

    10. In my opinion, the purported connection between karnitu and *karno- is rather weak, and also the verb isn't attested elsewhere in Celtic apart from Lepontic. This would indicate a loanword from a non-Celtic language, most likely an IE-satem one giving the striking Sanskrit parallels.

      Or, y'know, it's unattested in the generally poorly attested Gaulish, and happened to be lost in Insular Celtic.

      I would rather not move people over thousands of km based on a total of four CVC roots.

      By modern standards, half of the stuff in the Wörterbuch can't be taken seriously

      Pokorny consistently erred on the side of inclusion, including everything that might conceivably go back to PIE or some other IE stage – and that's with not using laryngeals in almost any reconstructions.

    11. Sorry for the delay.

      It isn't exactly Gaulish, but precisely Cisalpine Gaulish. And there's no direct relationship with Sanskrit or Indo-Iranian other than a similarity in verbal morphemes, which of course are different in Etruscan.

      By Occam's Razor, the source could be the same Baltoid (i.e. Baltic-like) substrate language identified by Coromines (who called it "Sorothaptic") and Villar (although they apparently conflated it with a different Italoid substrate), to which I'd attribute other loanwords to (Transalpine) Gaulish.

  7. Very glad I read this. I think you're right.