The last several weeks has witnessed a flurry of excitement (and not a little criticism) over the publication of Caleb Everett’s work arguing for a correlation (and a couple of potential causal links) between ejective series of obstruents and altitude above sea-level. Many of these reactions have been negative, faulting the work for inadequate sampling (here), lack of statistical significance (here), or more generally a failure to account for spurious correlations (here; most amusingly in Mark Liberman’s chart showing a correlation of English quantifiers and American real-estate listings (here)).
Rather than rehearsing many of these objections, I’d like to focus on one particular area of interest relevant to Everett’s claims about which I have some expertise, the Caucasus, because I think it illustrates in miniature many of the problems linguists have trying to tease apart issues of areal features, typological probability and genetic phylogeny. More than any other of the other putative clusters of glottalization in Everett’s article, it is also a region with one of the longest documented histories and most deeply researched archaeological records in the world and, as it turns out, is pretty mountainous to boot, and so throws some of the implications of the article into sharper relief than elsewhere. I will have to go into some detail first though.
It is widely known that the three autochthonous families of the Caucasus — Abkhaz-Adyghean, Nakh-Daghestanian, and Kartvelian — all stand out from all neighboring large language families for the distinctness of their phonological profiles (including rara like lateral affricates and phonemically pharyngealized vowels) and for glottalization in particular. However, despite the best efforts of many highly talented historical comparativists like Arnold Chikobava and Sergei Starostin, no one has ever made a convincing case that any of these families are related, much less all three of them (indeed Starostin’s controversial Etymological Dictionary of North Caucasian does not attempt to include Kartvelian). But if such similarities do not come about by genetic retention — a fact that stands in Everett’s favor — then it must come about through some other means, such as contact, convergent evolution or quite possibly chance.
Chance is of course never appealing to academics seeking interesting generalizations, and there is no shortage of people who have tried to argue that even if the Caucasian families do not constitute a ‘deep’ language phylum, then the region could still constitute a kind of Sprachbund. It is undoubted that many features have been borrowed across language boundaries, including specifically glottalization. Ossetian’s origins as the last remnant of the once vast Steppe Iranian language is well-known, but under the influence of Georgian and Svan, Ossetian has clearly not only borrowed hundreds of words from Kartvelian languages both with and without glottalized consonants (zok’o ‘mushroom’ < Geo. სოკო /sok’o/ ‘id.’; zwar ‘cross, sanctuary’ > Geo. ჯვარი /jvar-i/ ‘cross’) but has also largely remodeled its case-morphology system along a Georgian model, losing its inherited fusional Indo-European case system and replacing them with an agglutinating plural suffix plus case suffix. Ossetian has also borrowed other features typical of Kartvelian languages and not found in related IE languages, such as a restriction of complementizers to a (more or less) preverbal position (according to David Erschler). Another non-indigenous language, the dialect of Armenian spoken in Georgia, has also reportedly adopted glottalized realizations for its series of plain voiceless stops, at least as one possible realization.
What is true of Ossetian and Armenian is also true many other languages in the region. Klimov (Kavkazkie jazyki, 1965) used the presence of glottalized obstruents as one of the five criteria for a Caucasian Sprachbund, along with a pharyngeal/epiglottal place of articulation, agglutinative morphology, agreement with both subject and object and ergativity. Klimov’s thesis was controversial, in part because he couched this specific view about the Caucasus within his larger theory of stadialism in which languages evolve unidirectionally from one alignment type to another (e.g. active > ergative > accusative), a view which has since been largely discredited by almost all linguists. Tuite (1998), in his precis about the notion of a Caucasian Sprachbund, argued that Klimov had essentially abstracted too much away from important details which might invite skepticism about this idea. For example, Klimov assumed that Kartvelian languages are basically ergative, a view which most Kartvelologists now reject (Georgian and Svan are closer to active languages like Guarani and Hidatsa). Tuite argued that the Caucasus might be considered a phonological Sprachbund, but not much more than that.
Chirikba (2008) is the most recent attempt to advocate for a Caucasian Sprachbund. In his article, he argued for at least three dozen different features that supposedly characterize the Caucasus as a unit distinct from surrounding regions. A while back I decided to test this by taking Chirikba’s criteria and setting them against various languages from major branches of each of the indigenous families: Georgian, Svan, Abkhaz, Kabardian, Chechen, Batsbi, Lezgian, Khwarshi, and Tsakhur. (See charts. Not all fields are filled in nor all features shown, in part because available grammars are often either silent or ambiguous on particular questions. No doubt, they could be further filled in as I come across more publications or more works are published.) What this mini-sample demonstrates though is that indeed there are very broad similarities across the three families, but often because Chirikba defined the features rather vaguely, or because the features themselves are actually quite common cross-linguistically. For example, it is not clear exactly how he defined ‘rich’ or ‘sibilant’, so with a good amount of wiggle-room one could find all sorts of ‘rich’ sibilant systems anywhere in the world, and that therefore at least some Caucasian languages would qualify. In some cases, Chirikba actually cited false data, as when he argued that the Georgian არ მეჭმება /ar meč’meba/ is a potential form meaning ‘I cannot eat’, when in fact it really means ‘I don’t feel like eating’.
What such debates illustrate of course is that, even among specialists, both the data themselves and the analysis based on those data are not always universally agreed upon and that the accuracy of their use among nonspecialists depends upon very specific definitions of features and correlations.
If the question of identifying features is problematic, in the case of the Caucasus, there are specific reasons to think that the other side of this particular correlation with mountains is also weak. This is because the history of some of the families leads us to think that at least some of them may not have originated in their current locations and/or in mountainous environments, in Everett’s sense of the word. Within Kartvelian, for example, there are toponyms in Ajaria along the Black Sea coast, which is neither particularly mountainous nor particularly close to historically attested Abkhaz-Adyghean territory, that suggest some kind of Abkhaz-Adyghean substrate in very remote antiquity; if Diakonov’s claim that the ancient Anatolian language Hattic (only sparsely attested in records from the imperial libraries of the Hittite Empire) is related to Abkhaz-Adyghean proves true, this fact would have even stronger resonance. Wolfgang Schulze, Victor Friedman, and other specialists of Nakh-Daghestanian have suggested that the two branches of this family were once spoken south of the Greater Caucasus mountain range, in what is now the lowlands of Azerbaijan, Armenia or northwestern Iran, and that one group of speakers moved north of the mountains at some unknown date, and then at a later equally unknown date the other branch also moved north, resulting in the major cleft between the two branches. That is, Nakh-Daghestanian languages are doubly yet separately residual in Nichols’ terms. (Another Anatolian connection of Diakonov’s, that of Hurrian and Urartian with ND, is no longer wildly held but if true would also highlight the residual nature of ND.) Of the three families, Kartvelian is mostly likely to be truly autochthonous to the region, but even for it, its point of greatest dialectal diversity would appear to be in western lowlands of Georgia where Megrelian, Georgian, and Svan all meet. (A more easterly origin has been mooted to explain substrate phenomena in Kurdish and Iranian languages (Don Stilo p.c.)). Although none of these facts are certain beyond a reasonable doubt, their very possibility highlights the dangers of just assuming that a given language family acquired its characteristics in one geographical location, near mountains or any other topographical feature.
To return to Everett’s work, this long digression about the history of typology and areal features of the Caucasus highlights a basic difficulty in how linguists go about identifying a feature as a feature. Typologists are necessarily dependent on specialists in fields about which they themselves might know little, and are therefore subject to the danger of abstracting away from disputes and controversies which are then filtered into more generalist literature in which those controversies can become completely obscured. This danger can be mitigated by careful selection of samples and by trying to stick to areas about which authors have some expertise. However, and this is a crucial point which has been made by many others in the last weeks, even if you can be sure about your data, correlation does not necessarily imply causation.
In short, while attempts like Everett’s to find broad-based generalizations between language structures and geographical features are welcome, they must be founded in rigorous dissection of details before they can be considered anything more than tentative. When they are, such as Johanna Nichols’ work on spread zones vs. residual zones, or her other work on the vertical polylingualism characteristic of the north Caucasus, they can greatly enrich our understanding of the spatial distribution of language and language contact.