The WALS chapter on consonant inventories shows that the distribution of inventory sizes across languages follows a normal curve, with average size inventories (22 ± 3 consonants) being the most frequent. This is approximately the number of consonants that Proto-Indo-European had. Not surprisingly, this is also the category in which most Indo-European languages (at least the ones charted) are classified.

But there are some exceptions. Lithuanian and Irish (Donegal), for example, were put into the "large" class (34 or more consonants). What happened to these languages that made them deviate from the most common size and from the size of their ancestor, PIE?

  • 6
    Well, if it follows a normal curve, then we would expect to find "some exceptions to both ends". Feb 7, 2012 at 17:49
  • 4
    Yes, but should we be satisfied with a purely statistical explanation? Or is it possible to go further and explain what caused those deviations? Feb 8, 2012 at 0:25
  • 1
    Best of my knowledge Ubykh is the language with the largest number of consonant [phonemes]s: 84! (Khoisan languages excepted). Feb 8, 2012 at 1:12
  • 1
    @MarkBeadles, thanks for your comment. I've edited the title, taking your observations into account. Hope it's clearer now. Feb 8, 2012 at 11:41
  • 2
    It is definitely an error. The same language is listed as having 30 consonants in UPSID (web.phonetik.uni-frankfurt.de/L/L2022.html), and Maddieson's source is the same in both WALS and UPSID. I'll inform the WALS administrators to see about getting it corrected.
    – user483
    Feb 9, 2012 at 4:48

2 Answers 2


This issue has recently caught the attention of typologists and historical linguists, with a recent publication by Atkinson (2011), where it is claimed not only that a language's phoneme inventory size correlates positively with its speaker population, but that population bottlenecks in the course of human expansion out of Africa 50-70k years ago are reflected in the phoneme inventory sizes of modern languages. According to Atkinson's proposal, population sizes should have some role in explaining the differences in numbers of consonants in modern IE languages.

The proposal is, to say the least, controversial among linguists. The current issue of the journal Linguistic Typology (2011, vol. 15, no.2) contains a number of critical evaluations of Atkinson's proposal. Most relevant to the present question is Ringe's commentary (pp. 223--231). Ringe catalogs the sound changes leading to the development of Proto-Indo-Iranian, Sanskrit, Proto-Germanic, Old English, Greek and Latin and makes three main conclusions about the gain and loss of contrasts in IE lineages: (1) new contrasts were usually not due to borrowing, (2) most new contrasts were due to a reorganization of syllable structure or prosodic structure, and (3) the overall trend is neither a loss or gain of phonemes, although individual languages vary.

Ringe contrasts IE with Oceanic, where most of the modern languages have similar phonotactics, and supposes that the reason for the progressive loss of contrasting segments as one goes deeper into the Oceanic phylogenetic tree is that the usual process for gaining segments, phonotactic changes, has for some reason not happened in most languages.

What I don't know is how Ringe's proposal plays out for other branches of IE, including Lithuanian. I don't know whether many new consonants in Lithuanian were due to reorganization of syllable or metrical structure in the proto-language, or more straightforward conditioned splits that did not alter the phonotactics by much, but it would be interesting to check out. (Possible starting points for reading would be Kortlandt 2008 as well as other contributions to the journal Baltistica, which appears to be open access) For bonus reading, see Francois 2005 for a discussion on how several Oceanic languages of North Vanuatu developed unusually large vowel systems, bearing in mind that most present-day Oceanic languages (along with POc) have small (5 or fewer qualities) vowel systems.

  • There's a discussion of the Atkinson proposal at LanguageLog along with a link to a paper by Hunley, Bowern and Healey testing Atkinson's proposal and coming up with very different results. Feb 24, 2012 at 12:27
  • @GastonÜmlaut thanks for the link. Hunley et al's paper has some very beautiful visualizations.
    – user483
    Feb 24, 2012 at 13:27

I think we should look at this problem in a different way. Change in language is to a considerable degree a random phenomenon and, for the sake of this answer, it can be likened to coin throwing. If you end with 51 heads and 49 tails, you don't ask what happened to throw number 51 that it landed on tails rather than heads. It just did, and all we can do is capture the general tendency for the results to distribute closely to half-half.

It's the same thing with consonant inventories: in every language a certain number of changes has happened which resulted in its inventory being the size it is. All we can do is capture the general tendency for languages to have ca. 22 consonants. If you ask what happened to a particular language, all that can be answered to this is a list of changes that happened in it and influenced the size of its inventory.

Atknison's 2011 idea is obviously wrong in every way but even if it weren't, it wouldn't have answered your question, either. It would have only captured a tendency.

So, there isn't a single thing that made some languages distinguish more consonants and others less. It's always the cumulative result of different changes that took place in the language, and we can observe that on average these changes more or less cancel each other out, i.e. in most cases the inventory is changed throughout ages but its size remains approximately the same.

It only seems that anything particular must have happened because that's the kind of impression statistic tendencies make on us. Most people who buy beer also buy chips. But one thing this doesn't mean is that someone who buys beer is also supposed to buy chips. More often than not, they will, and that's all there is to it.

  • The coin throwing example is ridiculous. If you have never asked yourself that question, you simply have no basis to answer. If you did, you would sinoly throw a hundred times more and improve the ratio down to perhaps 499/501. And this still wouldn't answer the question, because language has many more dimensiins, it is at least a dice, and as a topic a very loaded one. Following your line of argument, you said nothing of value, because it's just random letters assembled through chance.
    – vectory
    Mar 10, 2020 at 15:44
  • @vectory Kamil said nothing about language usage being random, only language evolution. And this answer is from 2012.
    – Draconis
    Mar 10, 2020 at 18:33
  • The answer is from 2012, but the argument is one from ignorance, which, apparently, never grows old.
    – vectory
    Mar 10, 2020 at 20:34
  • @vectory You're very confident for someone who misses the point so completely. The gist of what I said is that when you're faced with a series of random events, you can capture the general tendency but there's little point asking why a specific event happened the way it did. The rest of your comment follows from a fundamental misunderstanding of what I said, so it's way off-topic and I won't address it.
    – kamil-s
    Mar 11, 2020 at 16:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.