How many laryngeals?

From: Piotr Gasiorowski
Message: 4812
Date: 2000-11-22

At the time de Saussure reconstructed the "sonant coefficients" (1879), reconstructed PIE vocalism was still very different from what we're familiar with -- very Sanskrit-like, in fact. To be sure, syllabic nasals and liquids had just been invented by Brugmann and Osthoff, and Brugmann's idea that there were two or more a's in the protolanguage (Saussure's a1, a2 and A, corresponding to modern *e, *o, *a) was beginning to sink in. The real quantum leap -- the discovery of the Law of Palatals, worked out independently by several linguists during the 1870s -- was to be announced at roughly the same time as Saussure's thesis. Saussure posited only two sonant coefficients -- A and O (*h2 and *h3, in modern terms) -- and the proposed derivations involving them included the crucial changes a1A > Â, a1O > Ô (i.e., *eh2 > *a:, *eh3 > *o:).
 
The following year Hermann Möller corrected the most glaring omission in Saussure's system by adding a third sonant coefficient to account for the ablaut of roots with long *e:. He also attempted to reconstruct the phonetics of the "coefficients" by relating them to Semitic laryngeals. Möller's triad (together with the misleading term "laryngeal") was in principle accepted by Kurylowicz in his famous 1927 publication. In this way a "laryngeal theory" featuring *h1, *h2 and *h3 became part of the Standard Model of IE ablaut. Departures from it have been common, which is hardly surprising given the nature of the evidence, which leaves much room for speculation. There are some hardline Brugmannians (such as Oswald Szemerényi) who reject all indirect evidence for the laryngeals and are content with just a single *h (accepted under duress, as it were, since Anatolian aitches can't be explained away); there are, on the other hand, aitch-splitters who follow André Martinet in multiplying laryngeals on the faintest shadow of justification. The three-term inventory represents a central position, favoured by most IEists. Why?
 
First, no doubt, because it is believed that "laryngeal colours" correspond to non-high vowel qualities in PIE in a one-to-one fashion. This is especially evident if a given author uses mnemonic symbols like (*E, *A, *O) rather than aitches with indices. This approach treats the laryngeals as consonantal vectors of the primarily vocalic features (a.k.a. "timbres" or "colours") [front], [low] and [round]. Another common approach aligns the three laryngeals (interpreted as dorsal fricatives) with the three series of dorsal stops: *x^, *x and *xW parallel to *k^, *k and *kW. Of course the two approaches can be combined into a unified, attractively symmetrical system:
 
       palatovelar  velar  labiovelar
 
plain       k^       k       kW
voiced      g^       g       gW
aspirated   g^H      gH      gWH
fricative   x^       x       xW
 
high        i/j              u/w
mid         e                o
low                  a
 
The pattern looks very nice indeed, but it's worth keeping in mind that we humans are instinctive pattern-matchers and tantalising symmetry is one of the most common mirages resulting from self-delusion. There are a couple of problems with this particular splendid symmetry. First, the colouring properties of the "laryngeals": *h1 does not change an adjacent *o into *e or impart its "colour" to neighbouring segments in any other way. There is therefore no compelling evidence for reconstructing *h1 as a distinctively [front] sound. Secondly, the behaviour of these sounds in Hittite: most IEists agree that *h1 has no consonantal reflex in Anatolian, while both *h2 and *h3 are reflectes as h (presumably a fricative in the velar/uvular range); at any rate, neither the postulated instances of *h1 > Hittite h nor the hotly debated evidence for the early loss of *h3 in Anatolian look very convincing. It seems that *h1 was lost rather early. It also seems that *h2 and *h3 merged in Hittite, so they must have constituted a natural class, probably in opposition to *h1. A possible additional problem is the skewed frequency of occurrence: *h2 is by far the most frequently occurring laryngeal, while plain velars are significantly rarer than other dorsal stops in reconstructions that employ the three series.
 
This has lead many scholars to speculate that *h1 was a weaker consonant than the other two "laryngeals". It is often reconstructed as a glottal stop [?], while the other two are classed together as spirants -- e.g. *h2 = [x] and *h3 = [xW], or *h2 = [h] and h3 = [GW] (voiced labiovelar fricative), or the like. "Colourless" *h1 can also be reconstructed as a breathy or breathy-voiced approximant [h], in which case *h2 and *h3 should be fricatives with the desired colouring properties. The most likely fricative to cause the lowering and retraction of an adjacent [e] is uvular [X], so it's tempting to assume that *h2 = [X] and *h3 = [XW]. The pattern is now different:
 
            dorsal    labiodorsal   glottal
 
plain         k          kW
voiced        g          gW
aspirated     gH         gWH
fricative     X          XW
approximant              w            h
 
I use [dorsal] as a cover term for a range of articulations including velars and uvulars (uvular fricatives commonly pattern together with velar stops). This interpretation has the advantage of not forcing all the "laryngeals" to be aligned with stop articulations -- very convenient if you are happy (as I am) with a two-way protosystem of dorsal consonants (*k and *kW, but no phonemic *k^). Also, any system with aspirated or breathy-voiced stops can on typological grounds be expected to include a phonemic "aspirate" /h/.
 
Now the final problem: what evidence do we really have for the phonemic contrast between *h2 and *h3?
 
Anatolian does not show spellings like hu- for putative *h3; e.g. Luwian has hawis for *h3owis 'sheep', Hittite has haran- 'eagle' < *h3or-on- and hastai- 'bones' < *h3ost-. To put it briefly, reconstructed *h3 seems to behave differently from labiovelar stops.
 
How can we tell, for a given form, whether attested *o- or *o: stands for underlying *h3o-/*oh3 < **h3e-/**eh3 rather than *o accompanied by one of the other laryngeals? Why, for example, do we posit *h3owis = /XWewi-/ rather than *h2owis = /Xowi-/ (Hittite h rules out *h1)? The argument that *h2 would have coloured *o as well as *e doesn't work: there are a number of secure counterexamples like *h2aiwo-/*h2oju-, *h2aus-/*h2ous-, *kreuh2as-/*kruh2o(:)s-, *h2ag-/*h2og-, etc. Even if we accept that any initial *a- (in Brugmannian terms) should be reconstructed as *h2a- < **h2e-, *o- may in principle go back to *h1o-, *h2o- or *h3o-. Of course it may be argued that any *o(:) appearing in a context where the e-grade can be confidently predicted is likely to have resulted from colouring by *h3. The right evidence, however, is not that easy to pin down. Present and preterite verb stems could be used; here the best examples I can think of are *deh3- 'give', *gneh3- 'know', and *gWjeh3- 'live'. I'll discuss *deh3- below (since it's usually the first and often the only sure example of root-final *h3 cited in handbooks), but if anyone feels discussing the other two or any other material, I'm always game.
 
Greek prothetic vowels, triple schwa reflexes and different vowel colours in *RH (Ro: vs. Ra:) are subject to so much dialectal variation and analogical levelling that even a careful analysis does not extract convincing evidence for *h3 as different from *h2. Beekes (1969) was more optimistic than I am here, but the "triple representation" has since been reanalysed e.g. by Lindemann (1982) and the view that it is a Greek innovation has gained much support.
 
Martinet's "laryngeal fission" *-eh3-o- > *-eh2wo- > *-a:wo- looks good on paper, but it can be shown that intrusive *w may appear in hiatus after *o: of any origin, e.g. *sto:h2-eje- > *stavi- 'to place' in Slavic (the causative of *stah2- 'stand').
 
If *do: is not *doh3- /deh3-/, what else could it be? The "o" timbre occurs in the present tense (Gk. di-do:-mi, arch. Lith. duo-mi) and deverbal nouns (Gk. do:ron, Slavic darU, Latin do:num 'gift'), and in the Greek and Indo-Iranian aorist as well (Gk. edo:ka, Skt. ada:t). One must remember, however, that verbs with persistent o-grade occur quite frequently; some of them probably formed a separate class in PIE, formally similar to the IE perfect and reflected in Hittite as the "-hi" presents (cf. Hittite mallai : Lith málti, Gothic malan, Latin molo:, etc.). We also have residual ablauting formations elsewhere (most notably, Tocharian subjunctives and Indo-Iranian aorist passives in -i like apa:di 'he fell') to demonstrate that the pattern associated with the perfect was once applied more widely, and that in fact there is nothing specifically "perfect" about it. A possible original paradigm would have looked like this:
 
*doh1-h2a(i)
*doh1-th2a(i)
*doh1-e(i)
 
These forms match the conjugation of Hittite da:- 'take' (usually assumed to be cognate to non-Anatolian *do: despite the semantic difference; note the absence of h in the Hittite root):
 
dahhi
datti
da:i
 
After the falling together of conjugational types in non-Anatolian IE new present and aorist formations appeared, based on underlying *do:- < *doh1-, e.g. preterite *doh1-e > *do: remodelled as *(e)do:-t.
 
Piotr