I have read some old works on lexicostatistics and glottochronology, like Swadesh's original articles or this work, where using Swadesh's basic assumptions, the author obtains a temporal estimation for the separation between two mexican languages. However, I am aware that later the theory received a lot of criticism for its too radically simplified assumptions, which led to some absurd result (a summary can be found here). These assumptions basically boil down to:
(1) Assumme a core of relatively estable words.
(2)Assumme the "replacement" rate in the core of stable words is approximately constant across languages.

My question is: What is the status of lexicostatistics/glottochronology in current research in linguistics? Is it completely disregarded, or are there more modern improvements that make this theory useful? I ask as a complete stranger to linguistics. A different but closely related question: In Glottochronology, starting with Swadesh's lists, as far as I know, all that is taken into account are the proportion of cognates. However, given known cognates, could the ammount phonetic changes be used too to estimate the temporal distance? This, if possible, could be one of the improvements that I mentioned above. I ask this motivated partly by this article, in which they study models for the probability of sound change and their role in reconstruction of proto-languages, but they do not mention anything about Glottochronology.

  • 1
    taking phonological changes into account would requiring weighting by a measure of distance. Besides it has been not very successful so far, as far as I can tell from those results from the last century that were summarily rejected, which I would take as a meassure of advances. There are still computational models being worked on, I'm sure, but these involve a lot more than the glottochronologic formula. Methods from Genetic analysis have been employed. At that, one problem is that language isn't made from four certain base nucleotides.
    – vectory
    Commented Dec 5, 2020 at 20:29
  • Yes, a distance between words would be needed, of course (only to be applicated betweend pairs of cognades, I guess). I guess from empirical data one could construct tables to record which sound changes are more common/more likely and how long do they take to "settle"? Intuitively speaking, for example, I guess a word-distance would assing a greater weight to a change t -> k than to the change t -> d. Is something on that line ever done?
    – Qwertuy
    Commented Dec 5, 2020 at 20:54

3 Answers 3


(This was going to be a comment but it got too long. I hope it is useful nonetheless.)

I can speak for Austroasiatic linguistics (a fairly large family with a small core of researchers actively working on reconstruction). The consensus in the field is that glottochronological estimates are outdated and not particularly useful. In current presentations, Bayesian computational phylogenetic analyses are often referred to as useful benchmarks instead. They are often used to support other analyses (e.g. my archaeological evidence / genetic evidence / phonological reconstruction suggests that speakers of these two languages lived here at time n and then diverged by time n+1, and here's my computational phylogenetic analysis which roughly agrees). One advantage of these analyses is that they can incorporate known facts, such as the fact that two languages had definitely diverged by a certain date. They can also incorporate assumptions about differing rates of change in different sub-branches, and much else.

I assume things are similar in other language families. The modelling papers usually get a lot of press when they come out. The last big one that interested me was Sagart et al 2019 in PNAS, on Sino-Tibetan.

  • Thank you for the comments and the very interesting reference. I don't know anything about these Bayesian analyses. Could you give some more references about the method or add some details about it?
    – Qwertuy
    Commented Dec 7, 2020 at 13:08
  • 1
    @QwertuyI don't know of specifically introductory material, but this paper goes through the process in detail: escholarship.org/content/qt9cr00162/qt9cr00162.pdf
    – legatrix
    Commented Dec 7, 2020 at 13:33
  • Looks great, thanks. I will need time to read that in detail.
    – Qwertuy
    Commented Dec 7, 2020 at 18:20

There are several approaches of these word-lists.
Some people like J. Greenberg deemed them practical, but attached no extra value to them.
Some others assign them more methodological values than just being practical.
Among present-day comparatists, only the Moscow School still "believes" in that kind of word-lists and calculations. Most linguists normally consider them close to worthless.
As a rule, my own opinion is that human intelligence and hard work cannot be replaced by automated processes.
Apart from the fact that "stability" can only be a relative feature, and apart from the known fact that rates of change and replacement are highly unstable, (Icelandic is fairly stable, Armenian is full of Iranian words, etc), another problem is that languages often have more than one word to express a given concept or realia.
Besides, inside a given family, dialects and languages often have different words, all of which may in fact be inherited.
So people who use word-lists are compelled to choose among all that complexity only one word, in so doing they are creating artificially simple word-lists.
It can also be noted that these word-lists as they exist do not necessarily include the words that are most interesting from a comparative point of view. For example, none includes the word "lip", which is nevertheless highly revealing.

  • I think everybody would agree with your statement that "human intelligence and hard work cannot be replaced by automated processes." As I saw it, lexicostatistics/glottochronology was always seen as as complementary technique (never as a substitution) to the comparative method. Cognade recognition still demands, and always will, a huge "manual" effort. The interesting criticism in your last paraghraps could be easily overcome by expanding the word-lists or some other similar modifications, don't you think? I am more interested in the fundamental limitations of the method.
    – Qwertuy
    Commented Dec 7, 2020 at 12:57

One of the core assumptions of lexicostatistics, the constant word replacement rate, was questioned be Daniel Nettle (1999) who suggests different rates of language evolution depending on the size of the speech community. Language in small (and small means really small here) speech communities changes faster than in larger ones.

  • 2
    And there are all sorts of other contributing variables as well. It's pretty clearly exploded.
    – jlawler
    Commented Dec 6, 2020 at 0:13
  • Thank you for this very interesting reference! Do you know of any appealing response to this criticism from the point of view of some more up-to-date lexicostatistics?
    – Qwertuy
    Commented Dec 7, 2020 at 12:59

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.