Correlation and causality: ejectives

Spurious correlations

“Recent studies have been uncovering some surprising links between cultural traits. For example, between chocolate consumption and the number of Nobel laureates a country produces, between the number of phonemes in a language and distance from East Africa, between a language’s tense system and the propensity to save money, between the quality of the sounds of a language with the amount of extra-marital sex and genetic influences on political outlooks.”

This is how Seán Roberts  and James Winters introduce their recent article: Linguistic diversity and traffic accidents: lessons from statistical studies of cultural traits (PLOS ONE, 2013, accessed August 2013), reviewed and discussed in Language Log.

The authors argue that the recent proliferation of digital databases of cultural and linguistic data, together with new statistical techniques may underestimate the probability of finding spurious correlations between cultural traits, showing that this kind of approach can find links between such unlikely factors as traffic accidents, levels of extra-marital sex, political collectivism and linguistic diversity (Fig. 1). This suggests to them that spurious correlations, due to historical descent, geographic diffusion or increased noise-to-signal ratios in large datasets, are much more likely than some studies admit.

Figure 1
Chains of spurious correlations from the article by Roberts and Winter (their Figure 3).
Statistical links have been reported between these cultural traits.
Authors’ names refer to studies. The “Results” label refers to results reviewed by Roberts and Winter.

High altitudes favour ejectives?

Another recent article demonstrated a correlation between high altitude and languages with phonemic ejective consonants:

Everett, Caleb (2013), Evidence for Direct Geographic Influences on Linguistic Sounds: The Case of Ejectives, PLOS ONE, 2013 (accessed August 2013), reviewed and discussed on Language Log on 14 June 2013.

Everett argues (1) that ejective sounds might be facilitated at higher elevations due to the associated decrease in ambient air pressure, reducing the physiological effort required for the compression of air in the pharyngeal cavity, and (2) that ejective sounds may help to mitigate rates of water vapor loss through exhaled air, both explications demonstrating how a reduction of ambient air density could promote the usage of ejective phonemes in a given language. He concludes that the results reveal the direct influence of a geographic factor on the basic sound inventories of human languages.

Everett’s findings were criticised by Roberts on 13 June 2013 in an evolutionary linguistics blogpost: Altitude and ejectives – hypotheses up in the air. Roberts found the correlation to be typically spurious after applying further tests as safeguards, as Roberts and Winters had proposed in their article. Some of Roberts’ test results were favourable, others unfavourable. Roberts’ point is that however favourable, the statistics do not prove the postulated explanation, they justify continued research.

Everett’s definition of high altitude areas was “major regions greater than 1500 m in altitude, plus land within 200 km of such a region”. For example, Everett reports that 57 of 92 (62%) languages with ejectives are located within areas thus defined, while 80 of 92 (87%) languages with ejectives are located within 500 km of a region exceeding 1500m. That doesn’t mean the language is spoken at 1500m. Far from it. It could mean sea level if that was within 200km.

Ejectives in Kartvelian languages

For example, take the the four Kartvelian languages spoken in Georgia, which is undeniably mountainous with several peaks rising to 4500 or 5000m:

  • Georgian (about 4.5 million speakers throughout the country),
  • Mingrelian (some half million speakers, mostly in the coastal provinces Abkhazia, from which many were displaced 1991-92, and Samegrelo),
  • Svan and Laz (that have a few thousand speakers each).

Wherever you are, from the Black Sea to the cities or to the mountain valleys, you are always within 200 km of the 1500m altitude, or indeed from the highest summits around 5000m (see Fig. 2).


Figure 2. Map of Georgia, showing the places mentioned in the text, and two additional towns Kutaisi and Gori. The highest mountains are along the northern frontier (top), there are also lesser mountains in the south. The distance scale shows 200km. Satellite map from

But few Georgians live at the 1500m level. Ten percent of the area of Georgia is either alpine (snow and glaciers) or subalpine; forests cover 40% of the country, mainly in foothills and mountains at 500-2000m; the remainder is lowland; half the population of nearly 5 million live in the 10 largest towns, ranging from Poti at sea level to the capital Tbilisi at 380-770m altitude, most of the remainder living in other towns and countryside in between. So even if Everett’s explanation turns out to be correct in the future, very few Georgians seem to get any aerodynamic advantage from it, whether they are adults producing their ejectives in daily speech or infants acquiring them.

One possible harder interpretation of Everett’s explanation is that high altitude was conducive to the original emergence of Kartvelian ejectives, assuming a community living exclusively near or above the 2000m treeline, descending to the lowlands later, bringing their newly evolved ejectives with them. Whether this story agrees with local stone and bronze age archeology is another matter. The earliest surviving Georgian writing dates from the 5th century AD, when the four languages were already differentiated. The whole country was known to ancient Greeks and Romans as fully inhabited, so this story of the evolution of Kartvelian ejectives has to take us back to a distant age before the languages separated, some 4-5000 years ago, to reconstructed proto-Kartvelian for which phonemic ejective consonants are also postulated (Gamkrelidze, Tamaz, 1966, A typology of common Kartvelian, Language 42, 69-83).

Anyway, life in the denser air of the lowlands during the intervening millenia hasn’t threatened the Kartvelian ejective phonemic contrasts.

Figure 3.
Professor Gamkrelidze, lecturing on the phonology of consonant systems
Photo: Wikipedia

To add a little spice to this story, Gamkrelidze (Kartvelian and Indoeuropean: a typological comparison of reconstructed linguistics systems, Bulletin of the the Georgian National Academy of Sciences 2, 154-161, 2008) finds similarities between between Proto-Kartvelian and Proto-Indo-European that raise “new questions concerning the historical connections of the Kartvelian languages with Indo-European”, possibly a common origin or contact. Invoking the Glottalic Theory, which proposes ejectives rather than the voiced stops of traditional reconstructions of Proto-Indo-European (In defense of ejectives for Proto-Indo-European, Bulletin of the Georgian National Academy of Sciences 4, 167-178, 2010) he creates a vision of Kartvelian inheriting or loaning its ejectives from even farther back. The glottalic theory is controversial, but Gamkrelidze’s proposal points to an origin for Kartvelian ejectives around the time of the last ice age at the latest.  However, the orthodox view of the origins of Proto-Kartvelian is that no-one knows.

English glottal stops

Everett’s definition of high altitude would exclude the United Kingdom. Yet there have been discussions on various blogs about similarities between ejectives and the preglottalized stops of English, for example in John Wells’ blog on 8 December 2011, Ejectives in English, and additional sites linked from there.  Wells pointed out that no-one really knew how preglottalization was distributed across English dialects, but thought it was at least as prevalent in Northern English. Owen McCarthy has recently studied ejective variants of velar stops by adolescents in a Glasgow high school (2011, MPhil thesis at the University of Glasgow). Reviewing the literature on glottal and ejective variants of English stops, McCarthy was surprised to find mainly anecdotal remarks rather than research.  He concludes his study by observing that ejectives certainly have been around for a long time (in English) and they can be a very common allophone for /k/.

Most Northern English is spoken within 200km of Scafell Pike (978m, pronounced scaw fell), a region that also includes coast. While we’re in northern England, I wonder if the Venerable Bede (distinguished Anglo-Saxon theologian, historian and translator 672-731 near Durham) had preglottalized stops? Or were they brought over by the Danes? Or the Normans? Now there’s a pinch of the same spice here. One supporter of the Glottalic Theory, Frederik Kortlandt, has suggested that English glottal stops and the Jutland Danish vestjysk stød (preglottalized /p t k/ in defined positions) are surviving traces from that same glottal series proposed for Proto-Indo-European (How old is the English glottal stop, 1997, North-Western European Language Evolution (NOWELE) 31/32, 175-179, ms version linked from here; Vestjysk stød, Icelandic preaspiration, and Proto-Indo-European glottalic stops, in Languages and Cultures: Studies in Honor of Edgar C. Polomé, Berlin, Mouton, 1988, 353-357, online here).

English glottal stops and Proto-Indo-European? Imagine it. Something that is never talked about in nice company might turn out to have the same noble ancestry as the most desirable vowels. And cousins of Kartvelian ejectives.

I acquired my southern preglottalized stops in the 1930s in Kent, gathering cockles at low tide on the mudflats of the Thames Estuary, and the most elevated ground in the county is 260m. Neighbouring Essex rises to 147m, E Sussex to 248m, and Surrey to a giddy 295m. To find our nearest 1500m heights we’d have to cross France to the Juras on the Swiss borders, which is more than 200km away. But then, our stops tend to be humbly glottal, seldom majestically ejective.

Postscript 8 Sep 2013

Thomas Wier (Ejectives, altitude, and the Caucasus as a linguistic area, 2013) reviews the typology and area features of Caucasian languages, showing that disputed migrations and homelands of languages can obscure any link with the mountains. He demonstrates 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”.

See also the blogpost by Harald Hammarstroem, that preceded Wier’s post.

Martin Lewis and Asya Pereltsvaig (Ejectives, high altitudes, and grandiose linguistic hypotheses, 2013) recall more examples of spurious linguistic correlations and associated causal explanations, and question Everett’s interpretation of the distribution of ejectives. As I did above, they emphasize that these languages occur close to, rather than in, areas of high altitude, and in many cases they are in the lowlands. They also point out that in known cases of languages evolving ejectives, they were derived by contact or by borrowing vocabulary.

©Sidney Wood and SWPhonetics, 1994-2013
This entry was posted in Articulation, Caucasia, Caucasus, Consonants, Ejectives, English, Methods and tagged , , , , . Bookmark the permalink.