Ejective and Implosive Theories of Proto-Indo-European Consonants

I’ve finally gotten around to reading Allan Bomhard’s 2016 paper on the ejective theory of PIE obstruents, which offers a renewed defence of the idea that the stops usually reconstructed as plain and voiced, like *d, actually go back to ejective stops, *t'. Since the subtitle of his paper is 'reigniting the dialog', I figured writing out my thoughts wouldn't hurt. Looking at the state of the discussion these days, especially in light of the many variants of ejective and glottalic theories that have piled up over the years, it seems to me that there are a few points that are worth restating (or at the very least writing through for my own benefit), especially on the phonetics and phonologies involved. None of what I have to say is really new, and there are important works by Barrack, Haider, Salmons, Weiss, and others that have definitely gone into this post (not to mention a number of grammars of particular languages with ejective and implosive consonants that I've looked at over the course of thinking about this problem), even though I didn't find a convenient point to mention them specifically. Maybe if I have the time and energy in the future I'll expand this and give it better references. I've divided this post into seven points, roughly leading into one another, but dealing with different aspects of phonology and reconstruction.

1) One of the more popular variants of ‘glottalic’ theory is the Leiden school’s approach (Kloekhorst 2016, pp. 232-235 gives a recent summary with extensive references). This is usually stated in phonetic, not phonological terms, as involving a ‘Proto-Indo-Anatolian’ series of ‘preglottalized stops’, phonetically voiceless and short: *[ˀt] (using the dentals for illustration). This should probably be classed phonologically as a type of ejective, /tʼ/. Assuming a bifurcated family tree, with Anatolian on one side and the rest of IE on the other, this approach posits a change to phonetically voiced preglottalized stops, *[ˀd]. Phonologically, such a stop would seem to belong to a category of consonants including implosives and creaky/laryngealized stops. Phonetically there is a continuum of realizations here, but phonologically there is, to my knowledge, never a contrast within this spectrum (see, among others, Lombardi 1991, Clements and Osu 2002, Ladefoged and Maddieson 1996, p. 53; Kehrein and Golston 2004, p. 330; for a classic phonetician’s discussion, see Ladefoged 1968, pp. 16-17). For PIE, I will refer to this group of sounds all as implosives as a cover term of convenience, and notate them as /ɗ/, without implying anything in particular in terms of either phonetic realization or featural structure (though 'voice' plus 'constricted' or 'glottalized' would be standard phonological characterizations). I think it’s important to think about this from a phonological perspective, because the putative evidence for preglottalization (all of which, it should be said, is disputed) comes largely from syllable codas, a position in which a preglottal allophone of an implosive might be expected. The main point here is that even if we accept the Leiden position, we probably should not make a habit of talking about ‘preglottalized’ stops in general, as if this were their phonemic specification or universal realization, since that may well involve an inappropriate generalization of one, possibly marked allophone to all positions.

2) Is there much serious defence today of ejectives specifically for the immediate ancestor of the majority of the IE languages, the way the original proposals of the 'glottalic' model envisioned (see, e.g., Gamkrelidze & Ivanov 1995, p. 45ff.)? Details vary, but most discussions now posit a change of earlier ejectives to something else, something voiced, as a common innovation to the precursor to many, most, or all IE languages before they fully diverged as separate branches. This is the argument of Bomhard (who sees this in 'Disintegrating PIE') and the Leiden school (who see it as a major innovation of non-Anatolian IE). Kümmel’s important 2012 paper proposes that ‘early PIE’ had implosive (phonetically, non-obstruent) stops. The popularity of this position is easy to understand. Studies like Fallon's work on ejectives have, even while ostensibly arguing for the diachronic possibility of voicing ejectives, demonstrated rather effectively that this change is not nearly usual enough that we should comfortably assume that a change of */t’/ to *[d] took place separately in six of the ten known major branches of Indo-European. Hence the desire among those envisioning original ejectives to suppose a shift at least to some intermediate stage like */t’/ to */ɗ/ (however this was realized phonetically), if not already to */d/, as a shared innovation. (Since shared innovations can spread across diversifying dialects, so the relevance of this idea to subgrouping and the phylogenetic structure of the IE family is limited.) While the details differ (and not just in superficial ways), some sort of (partial?) shared shift would seem to be a part most ‘glottalic’ approaches today, and should probably now be assumed as a standard part of  any ‘glottalic theory’ unless specified otherwise.

3) On a methodological note, rather than phonological one, this shared innovation is important, since one of the typological features meant to be explained by an appeal to ejectives, the rarity of ‘*b’, only works for ejectives proper (insofar as it is compelling; this argument is perhaps not quite as strong typologically as it’s sometimes made out to be), not for implosives (implosives favour labial articulations). Among other things, we should probably recognize that the points often cited together as evidence for a single ‘glottalic’ theory are in fact often pointing to different layers in a proposed sequence of development. Much of the Leiden school’s arguments for non-Anatolian PIE *[ˀd] work on a different level from the typological arguments for PIA *[ˀt] (which draw on cross-linguistic comparisons with ejectives, implying phonological */t’/). This is not an objection to these sorts of approaches, but methodologically I think it’s very important to recognize the chronological mismatch between these lines of argumentation.

4) The reconstruction of */ɗ/ widely in IE also has a further consequence for reconstruction: the reconstruction of */t’/ at any stage, rather than just /ɗ/ all the way back, now depends very crucially on one’s view of the importance of typological arguments specifically in explaining the rarity/lack of ‘*b’. As has been noted many, many times before, this gap could have other explanations, such as a late pre-PIE change altering */b/ or */ɓ/ to something else, like *w and/or *m: one could suggest that in pre-PIE */ɓ/ was very common (in keeping with typological norms), and its absence is simply a quirk of sound change. If I’ve read him correctly, Kümmel in fact does not suggest an earlier */t’/ stage (though I don’t think his arguments exclude such a thing either). This is part of why some have, quite rightly, suggested abandoning the term 'glottalic' theory entirely, and instead always being explicit: do we mean an ejective theory, an implosive theory, an ejective-to-implosive theory, or something else entirely? This kind of precision might be cumbersome, but the term 'glottalic' probably now evokes too many possible options to be all that useful.

5) The position of Germanic and Armenian as ‘relic areas’ should probably just be abandoned at this point. Under no reasonable view of the patterns of innovation within IE are these outlying dialects, and it hardly seems plausible to argue that they’d esecape any major change to an old ejective series. It’s having our cake and eating it too, if we propose that there was a shared innovation in order to get around the problem of general voicing in 'Inner' PIE (an implosive or voicing theory), while also invoking a supposedly simpler version of the Germanic and Armenian consonant shifts as evidence for an ejective theory of PIE. In any case, in terms of explanatory power, an ejective model really offers little compared to traditional accounts, which can explain these consonant shifts just fine.

6) Back to phonological typology, the most significant typological issue with the traditional reconstruct is and remains not the 'plain voiced' series, but the murmured ('breathy voice', 'voiced aspirated') stops like *dʰ, which I'm writing /d̤/ (/dʱ/ is an equivalent notational convention, as far as phonology is concerned; actually the traditional notation /dʰ/ is not really problematic, despite its absence from the IPA, since the relevant phonological feature, now often described as 'spread', is the same as in voiceless aspirates, /tʰ/), but it still looks like there’s little consensus on how to handle these between different ‘glottalic’ models. Kümmel relates the development of murmur specifically with the shift of implosives to plain voiced stops, in a kind of push-shift. Bomhard also sees murmur as arising in ‘disintegrating’ IE, and ancestral at least to Greek, Armenian, Italic, and Indo-Iranian. No matter what we do, we’re going to run into the same essential conflict that’s been at the heart of this debate for half a century now: do we avoid reconstructing */d̤/ (at the cost of having to posit some usual sound changes in many particular branches), or do we reconstruct */d̤/, and just accept that some stage of (post-)PIE had a typologically unusual system? Postponing the question to ‘late PIE’ doesn’t really make this tension go away, except insofar as a brief transitional system might perhaps be more tolerable than an age-old one. I'm not sure about other people, but for me it's the idea of abandoning wholly the murmured series that's the real sticking point with things like the Leiden model, rather than the 'glottalic' part of the model as such.

7) The whole 'problem' with the murmured stops arose because of the ‘removal’ of voiceless aspirates from the PIE phonemic system. Actually it's occasionally been suggested that the Neogrammarians were right, and voiceless aspirates like */tʰ/ were part of the PIE phoneme inventory. This might not end the debate about ejective or implosive consonants entirely, but it would put the whole question on a very different typological footing: in particular, it would make the assumption of the murmured series unproblematic. But most voiceless aspirates are so obviously secondary that there’s a natural reluctance to see them as phonemic. Maybe (this is a point I first heard made by Mark Hale) the more important question is: does this matter? The significance of the ‘phoneme’ depends a lot on precisely what theoretical phonological paradigm we’re using, and it’s a potentially defensible position to maintain that significant surface allophones are what we should really care about here. If we have sequences like */th₂e/ -> *[tʰa] as a real part of the phonological (not phonemic) structure of the language, then we might not face nearly as compelling a typological objection to 'voiced aspirates'. How to formalize this will vary. One possibility is that the assimilation process relies on adding the feature +SPREAD to the voiceless stop, incresaing the phonological salience of this feature, allowing its use in the underlying specification of voiced segments. The surface space would certainly be fleshed out with *[t, d, tʰ, dʱ=d̤]. Such sequences might not have been frequent, but they probably did not need to be: 'voiced aspirates' are much more common than voiceless in Sanskrit, for instance. Not every phonologist will like that kind of unbalanced feature specification, but it’s a potential criticism of phonological theory in general that perhaps too much attention has been paid to economy and symmetry at the underlying level, and not enough to surface or mid-level structures. Of course, even if one does accept this idea, its exact bearing on the ‘glottalic’ question is not straightforward. There's also a question of chronology. The clearest evidence for *[tʰ] allophones comes from Greek and Indo-Iranian, which are also two of the branches where there’s the greatest need for murmur, and which are often thought to be in some sort of inner-IE dialect area (not subgroup). Perhaps the rise of voiceless aspirates in an inner area could be a part of a dialectal rise of ‘voiced aspirates’.

As I said, none of these points are really new, and, this being a blog post, and other work staring at me accusingly as I take too long writing this already, I haven’t given all the references I should. There are also interesting suggestions, such as the possibility that murmur was not a segmental property, but a suprasegmental feature of roots, that I haven't touched on. These are just the main points that struck me as most in need of restating, given where literature on the subject seems to stand now.