Given two Swadesh lists of two languages, we can calculate a "coefficient of relatedness" by counting the proportion of cognades. Given two languages, L1 and L2, let's call C(L1,L2) to this coefficient, which will be a number between 0 and 1. Now, I was attempting to calculate C(PIE,L)* for every (or at least a bunch of representatives ones sampling every major branch) modern living indo-european language.

It is extremely tedious, but potentially doable with a computer, of course, but I was guessing that probably this is out there somewhere. Where could I find such an info?

Also, even if not all these numbers are available, I am very much interested in the following question: For which L's is the the coefficient C(PIE,L) largest? And for which languages would it be smallest? I am not sure either what would be "large" and what would be "small" in this context, so perhaps someone could comment on that too.

(*Note: In this context, substitue "being cognades" for "the word in L descend from the word in PIE of the same sloth in the Swadesh list")

  • Generally speaking, those languages which least deserve to be called separate languages rather than dialects. E.g. my dialect of English and General Montana English.
    – user6726
    Jun 23, 2020 at 19:19
  • To clarify, you're only interested in modern, living Indo-European languages?
    – Draconis
    Jun 23, 2020 at 19:29
  • For words like "see" you'll have synonyms to choose from. A cognate might exist as a hapax, or there might be no perfect translation from one to another, and sometimes there might be a match only because of loans or calques. So the list will be biased by the assumptions made. Severely, these difficulties are inherent in the reconstruction, so you would be effectively testing your assumptions against those underlying the reconstruction. This is egragious because synonymous roots reconstructed from disparate branches exist (cp. aqua, water, ...) and there's no PIE speaker to take a survey.
    – vectory
    Jun 23, 2020 at 20:29
  • Yet, I'm sure somebody's done it before. Swadesh lists are available in wiktionary from various user pages for example. But, as said, I think that's useless. And grammatical features like morphology count for a lot in typology, so Icelandic, Lithuanian for living languages. Not sure if I would rule out Satem languages all together, not my field. Hittite is of course deemed most archaic, though it would look least related because of all the shared innovations in the other branches, as is commonly assumed and thus reflected in those reconstructions where Hittite evidence exists. Good Q, actually
    – vectory
    Jun 23, 2020 at 20:34
  • 1
    I don't know about calculating a 'coefficient of relatedness' but SIL has for a very long time had methods for taking a table of distances (showing cognacies across multiple languages) and using that to generate a relatedness tree. This shows relative relatedness rather than an absolute number. There are plenty of examples out there, eg this from 1974 Jul 24, 2020 at 0:38

1 Answer 1


This is becoming too long for a comment ...

For pointers towards Swadesh lists see this question and its answers Database of Swadesh lists — unfortunately I don't see a list for Proto-Indogermanic or other reconstructed languages at first sight. You will have to annotate "cognateness" manually in any case, so you probably better go through the dictionaries and prepare it yourself. Be aware of potential duplicates (e.g., the two words for "water"), synonyms, and homophones. It is not as easy as it may look like.

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.