zompist bboard

http://phys.org/news/2013-06-altitude-a ... poken.html

This actually seems plausible to me — could someone with more formal linguistic knowledge speak to whether this seems like a reasonable conclusion? I wonder if we'll begin to see other correlations with geography, though I think altitude would probably be the strongest factor (considering how it changes air density).

One question I have that I do not see explicitly answered:

How does the proportion of ejective languages near high elevations compare to the proportion of all languages near high elevations? It says 87% of languages with ejectives are near a high-elevation region, but if someone told me 87% of all languages were near high elevations I wouldn't be surprised in the slightest. That is because mountainous regions are well known to have disproportionate linguistic diversity, so the number of languages in such a region is often much higher than for a comparably sized region of lowlands nearby. Sometimes spectacularly higher, as in New Guinea.

Oh Rad, you're hard to please...

There's a nice map AND a link to the original article included in the Phys.org article: http://www.plosone.org/article/info%3Ad ... ne.0065275.

The raw data of the sample appear to be as follows:

With high elevation defined as within 200 km of 1500m+ regions
57 languages with ejectives, high elevation
35 languages with ejectives, low elevation
96 languages without ejectives, high elevation
379 languages without ejectives, low elevation

With high elevation defined as within 500 km of 1500m+ regions
80 languages with ejectives, high elevation
12 languages with ejectives, low elevation
202 languages without ejectives, high elevation
273 languages without ejectives, low elevation

Make of this what you will. It''s probably best to look at it this way: at high elevation (definition 1) 37% of the languages have ejectives, against 8% of the languages at low elevation. With definition 2, these numbers are 28% and 4%. Those numbers seem quite impressive to me (but of course further statistical analysis, and probably analysis of the WALS sampling method is needed).

I'd take it with a generous helping of salt. Radius's point is very good, and I'd add two things:

1) Surely we should look at families, not individual languages? E.g. Indo-European, Austronesian, and Bantu are three families that happen not to have many ejectives, but they cover an enormous area of the globe. Diversity in languages is just not randomly scattered across the map. Huge areas are covered by the descendants of languages that expanded wildly in the last 2000 years. To put it another way, the fact that languages in most of Europe, or in most of sub-Saharan Africa, are similar in whatever feature is due to political reasons.

2) They're looking at about 600 languages, or 10% of the total. That would be fine if that 600 was a good random sample-- but there are good reasons to think it's not. The less-studied languages are also the most likely to be weird. Some of the most diverse regions (e.g. the Amazon) are the hardest to get good information on.

There are lots of myths in Japan that people in the cold mountains of the north like to open their mouths for less time, and that makes them shorten a lot of words. Not the same thing of course but this reminded me of that.

Discussion on Language Log. They seem to take it at least somewhat serious.

Actually I'd say that they're skeptical, just very polite. They brought up the same objections Radius and I did, and added another one:

(2) The astronomically large number of such possible relationships guarantees that many of them should exhibit a strong pair-wise connection by chance, even if all of the distributions were statistically independent;

That is, there are so many possible variables, and so many languages, that some weird correlations may be found, and hold up to preliminary analysis, completely by chance. Liberman also mentions the "file drawer effect"-- i.e., our window on the data mining process is skewed, as correlations that don't pan out are ignored.