On Out of Eurasia and linguistic time depth

So here’s the hypothetical (as developed previously). Suppose modern humans have been hanging out at least somewhere around Eurasia already for 100, perhaps 200, maybe as much as 300 millennia, instead of merely 50–70. Should any of our views on the history of language(s) be affected?

A basic immediate result is that this substantially increases the time depth available for the language families around. This includes not just the known and proposed ones — but also the undetected ones that anthropology and genetics tells us will have to exist. As established before, an Out of Africa theory of modern human origins demands that ~all languages of outside Africa ought to go back to a single common ancestor about 70,000 years ago, since the Near East creates a natural bottleneck for early, pre-naval migrations. An alternative Out of Eurasia however does no such thing. It does suggest the existence of a few linguistically unconfirmed macrofamilies like African, Amerind or Australian, but these do not need to go back to any especially closely related Eurasian ancestors. These in turn do not have to be especially closely related to any modern Eurasian language families either.

If so, a failure to detect any relationship between even geographically relatively close-by families such as Sino-Tibetan and Indo-European, or Semitic/Afrasian and Sumerian, does not have to mean that the Comparative Method is therefore likely to run out of steam after 10 or even 20 millennia. Maybe the effective limit is much deeper, but it is also the case that these kind of “patently unrelated” languages really have been separate from one another ever since the Lower Paleolithic.

Eurasia houses also the great majority of the world’s well-studied language families. (In this context I would count also Austronesian as a “Eurasian family”, given its homeland in Taiwan, or maybe adjacent continental China. [4]) Those elsewhere have been documented and reconstructed on average much more scantily. A few blazing successes such as Bantu or Algonquian also suggest that there are many more results left to be claimed. It is therefore only in Eurasia that we can really with decent confidence claim that plenty of the language families appear to be unrelated or only vaguely related. Macrofamilies elsewhere in the world, such as Amerind or Australian, cannot be decreed invalid just on the basis of the so far poor results for macrofamilies across Eurasia!

It is every so often also claimed that linguists working with African languages in particular would tend more towards “lumping”, while linguists working with Eurasian languages would tend more towards “splitting”. I don’t think this is fair, except perhaps if we define “lumping” and “splitting” purely by the size of the language families involved. If African macrofamilies appear to have about as much evidence for them as Eurasian macrofamilies do now, when the languages of Africa are far less documented and researched, then I think we can expect the evidence base to keep growing. Over the 21st century I expect further solid reconstructions and new perimeters to be reached.

On a historiographical note, I would also like to briefly note that while “lumping” is often blamed on Joseph Greenberg and his alleged uncritical followers, almost all of his macrofamilies had been proposed or at least explored already earlier on. He may have brought in some new annexations and new unwarranted confidence, but the concept of “macrofamilies” has in principle nothing to do with Greenberg’s barely-method of mass comparison. [5]

I expect also one cross-linguistically important result to eventually emerge from ongoing research on the major African language families in particular. Besides big, these are likely to be fairly old… Hence, if we have one day a detailed reconstruction of Proto-Niger-Congo or Proto-Macro-Sudanic (“Nilo-Saharan” at its widest I do think is a wastebasket taxon), and their development to a few fairly distant languages, this will be able to show us what a linguistic relationship that’s 10,000+ years of age really looks like! I am not confident that this would have to turn out as minimal as is often claimed.

It is an unfortunately common notion that a linguistic relationships of 10,000+ years of age would have to be undetectable. This seems to come from two main sources, both of them IMO fallacious. The first is naive extrapolation from examining relationships maybe halfway or third as old. Proto-languages of this age we can reconstruct, but only partially. E.g. in terms of lexicon: in any language that is still spoken, tens of thousands of words can be attested, while in multi-millennia-old proto-languages only some hundreds can be reconstructed. So should we not assume that further two, three or four time periods equally long should squeeze the available evidence down to definitely nothing? Well, maybe — if you’re willing to bite the two bullets that (1) glottochronology works at least on long enough time periods, but (2) there is no core lexicon or grammar, and everything is equally likely to change. Without the first, punctuated equilibrium models will allow for the possibility that some languages may have remained, at times, mostly stable for millennia. Without the second, you will have to admit that languages’ core features actually remain stable much longer than their peripheral features, [6] and they will likely allow reconstruction efforts well beyond what naive extrapolation suggests.

The second error is based on the essentially winged age estimates for Eurasian macrofamilies. This is already internally incoherent, though. If a maybe-family like Nostratic is proposed to be in the age range of 10,000 years, and if the Nostratic proposal is too weak to be accepted… then it does not follow that all language families in this age range will be too weak to be accepted: rather, this means that the age proposal for Nostratic is, together with the family itself, also too weak to be accepted. What does not exist, cannot be dated either. (Did Cthulhu invent cephalopody before or after the molluscs did?)

One possible compromise for this would be to treat (again, e.g.) Nostratic as a language family that can be accepted through archeological / cultural-anthropological / genetic arguments, but not linguistic ones. I don’t think anyone really believes in this exactly, though. I only ever see arguments to the effect that, if Nostratic exists, based on cultural & genetic distance, then it will have to be at least 10,000 years old (and this much is easy to agree with). But this is only a lower bound! It gives us no evidence about an upper bound. Since language shift and cultural convergence exist, there is no linear or even monotonic relationship between linguistic distance and cultural-genetic distance. After a few language shifts and centuries of convergence, the modern citizens of Haparanda and Tornio are just about indistinguishable by cultural and genetic distance. Yet by linguistic distance, one side still remains Indo-European, the other Uralic.

Finally, most of the various ideas above can be pooled together as a provocative new hypothesis for thinking about the Eurasian “maybe-families” and their place in the general context of “macro-comparison”: the vague resemblances we see in e.g. Nostratic could be indicative of what language relationship after 100,000 years looks like. That is, it might turn that the reason macro-comparison has been thought to be mostly fruitless is that, the entire time, most linguists have been actually beating their heads against the hardest such problems around! Once we have Indo-European-level (Uralic-level? [7]) documentation of most languages involved, maybe not just units like Atlantic-Congo or Trans-New-Guinea, currently established on fairly meagre evidence, but also even much older and larger units, say Niger-Saharan or Southern Amerind, will turn out to be relatively feasible to establish and reconstruct. Only future work will tell for sure. [8]

(To be continued still…)

[4] As suggested by Blench in various draft papers. Also the today quite well-emerging Austro-Tai theory would fit in with this: given probable continental relatives as well, Taiwan may simply constitute a residual zone populated by Austronesian groups driven off of or extirpated on the mainland by the Chinese.
[5] Even mass comparison can be probably steelmanned, but that’d be a topic for another time.
[6] Cf. the fact that radioactive isotopes’ half-lives come as anything between billions of years and some fractions of picoseconds.
[7] It seems possible that Uralic is actually the best-documented large language family out there, if “large” is defined to cover both diversity and time depth. There’s a grammar for everything (even if from the 1800s for some languages), and at least one extensive multidialectal dictionary for almost everything (usually more; oldest unsuperceded ones are from around 1940; biggest omission is probably Veps); multiple etymological dictionaries and historical grammars for every language with a longer written tradition (well, all three of them), and a bunch of them even for minority languages. Indo-European definitely sports the best-documented individual languages out there, but the family-wide average is killed hard by modern Indo-Iranian languages, which AFAICT have been essentially deemed “comparatively useless” to document due to Sanskrit and Avestan being available. Semitic fares somewhat similarly (weak points: Ethiosemitic, Modern South Arabian). On comprehensiveness + diversity, several other (once again usually Eurasian) families like Japonic and Turkic are doing well or getting there, but all of these are younger. On comprehensiveness + time depth the only contender I can think of is perhaps Kartvelian, which does poorly on internal diversity.
[8] Provided that people will not write out even the prospect of such future work by unsubstantiated assertions about the Comparative Method “stopping working” after some random number of millennia. But, I also think it doesn’t really matter if this belief is being held by some people currently… It will be a good idea anyway to get a decent reconstruction of Benue-Congo before trying to reconstruct Atlantic-Congo or the whole Niger-Congo, and also to get a decent reconstruction of Bantoid or Edoid before trying to reconstruct Benue-Congo. And if I am right about families like NC being able to eventually provide us with explicit examples of languages being demonstrably related over 10,000+ years, then the skeptics are not going to be denying any of the results happening along the way; they’re going to be gradually retracting their supposed time limit, until it turns out to be deep enough that it can be no longer used to flatly deny other similarly far-away but less obvious results like Dene-Yeniseian either.

Tagged with: , , , , ,
Posted in Methodology
25 comments on “On Out of Eurasia and linguistic time depth
  1. David Marjanović says:

    The silliest thing about the supposed time limit of the comparative method is that Afro-Asiatic is widely accepted to be older than 10,000 years (12,000 is a figure I’ve seen). Sure, its morphology makes it a special case, but it does have common basic vocabulary, and I don’t think there’s been an attempt to explain just why the morphology would be unusually stable.

    • John Cowan says:

      I suspect that people would really really like to say that AA is only 5-6,000 years old too. But unfortunately for them we have Akkadian and Egyptian of almost that age, obviously related and yet already deeply diverse. So it’s more like “Awright, awright, we’ll double the time for AA, but the rest of yiz can take a hike. You get 6000 years [the traditional Biblical age of the earth] and that’s it.”

  2. Actually the supposed 10,000 years limit on the comparative method is a self-fulfilling prophecy: students are taught that any reconstruction on this time-depth is impossible and as a consequence the only people who try to do anything are proponents of Greenberg-style mass comparison, who simply do not know the comparative method.

    • j. says:

      Though before real reconstruction it’s hard to say how old various language families are exactly. I would be a bit surprized but not shocked if, for example, maximally wide versions of Sino-Tibetan or North Caucasian turned out to be in this range. This kind of a dating could be conceivable even for some smaller families; say Kartvelian, if someone were to work out an ancient layer of loanwords between Karto-Zan and Svan, vs. an even older dating for the truly native material.

  3. Howl says:

    This is a great blog post! It also touches on the issues I struggle with when it comes to Nostratic. I don’t buy the ‘spread of agriculture’ theory of Bomhard. Any linguistic expansion of a language needs speakers of that language. And those speakers must leave at least some trace in the genetic history. I don’t see anything from the Middle East reaching the Eskimos, for example. When you look at the genetic history of Eurasia, it is really hard to fit in something like Nostratic.

    If you go for the natural 50ky BP, then Nostratic must be the proto-language of not only *every* Eurasian language (including the North-Caucasian ones, Austronesian, Sino-Tibetan, Austroasiatic, etc) but also of all the languages in the Americas.

    After that there is the European/Asian split, which makes it almost impossible that there was one language family that is ancestral to both Afroasiatic (almost pure European + Basal + African DNA) and Japanese/Korean (almost pure East-Asian DNA).

    A pruned Eurasiatic fares a little better. Assuming this came from the Ancient North Eurasians (ANE; old Siberians), it could have formed around 15k-12k BP. This is the time when the Eastern European Hunter Gatherers (EHG; predecessors of IE and Uralic) and Neosiberians (predecessors of Turkic, Tungusic and Mongolic) formed. There are some problems, though. Neosiberians only have about 5%-10% ANE DNA. The rest is East-Asian. Japanese and Korean fare more poorly, as they are almost 100% East-Asian. But a Eurasiatic from an East-Asian origin runs into the problem that EHG has almost no East-Asian DNA.

    Maybe that a more detailed genetic history with more DNA samples from the paleolithic could provide a ‘Nostratic’ signal. But it will be very subtle, and open to different interpretations. Even the not-so-subtle Yamnaya genetic signal of Indo-European still leaves us with open questions.

    • John Cowan says:

      Any linguistic expansion of a language needs speakers of that language. And those speakers must leave at least some trace in the genetic history.

      Not necessarily. There is very little modern European DNA in Haitians, for example (a mean of 8%, with about 10% having none at all, and it’s likely that the light-skinned upper and middle class were oversampled here) yet all of them speak a modern European language. Similarly, the Republic of Ireland has had very little English DNA contribution in recent centuries (vs. about 20% Irish contribution in the UK), yet all its people speak English.

      • j. says:

        This can work for a few exceptions here or there at the edges of a larger expansion. Most native speakers of English are regardless of European ancestry, most IE speakers have some Yamnaya ancestry, etc. A young Middle Eastern Nostratic would basically require spreading by language shift from something else among the future speakers of Afrasian, I suppose (but then who else would these people have been? pre-Kartvelians??)

        I seem to recall a proposal of exactly this type for language shift early in the history of Sinitic in particular, with Sino-Tibetan brought from the southwest by “southern East Asians” but then adopted by “northern East Asians” alongside agriculture… This doesn’t help much with the still deeper East Asian / West & Central Asian split, though.

  4. David Marjanović says:

    I’ve been wondering which language families to associate with the Caucasian Hunter-Gatherers/Iranian Neolithic Farmers. Perhaps a linguistically weird “southwestern Eurasiatic” assemblage: Kartvelian, Dravidian, and IE – after all, the potential speakers of Anatolian had no EHG ancestry. But perhaps rather North Caucasian and Burushaski. Or both, requiring some language shift at some point…

    • …or neither, given that languages and whole families tend to die out. Where are the modern descendants of Hurro-Urartian, Sumerian, Hattic, Elamite, Tyrrhenian?

      • David Marjanović says:

        Of course. CHG ancestry is so widespread that it’s perhaps more likely that an associated language family has survived, but that’s not the only possibility.

  5. Y says:

    Let me be contrarian here. I don’t really see any argument that older relationships (say > 8000 y) are discernible. You are saying that thet could be, and that perhaps linguists have a bias leading them to rely on circular arguments, but there’s not one example of such a relationship which is secure in both its existence and its antiquity.

    The idea underlying glottochronology is that there’s a steady attrition of discernible cognates. Such an attrition is driven by semantic change: hound becomes specialized (later perhaps to the point of disappearance) and replaced by dog. However, another factor in language unrelatability are sound changes which affect the language as a whole. WIth more time, more changes to a sound system are likely. This would make two languages less recognizable as genetically related, unless the changes can be undone through careful reconstruction. I suggest that such systemic changes are more responsible than word-by-word semantic changes for the diffculty of finding deep relationships.

    When you’re talking about “what language relationship after 100,000 years looks like”, remember that these are exponential changes. If German and Hindi are 6,000 years apart, take these changes to the 17th power and you’ll get what languages 100,000 years apart look like, at best. That doesn’t seem plausible to me.

    As to African lumping and splitting, I see the opposite of what you say. With more documentation and study, African phyla previously accepted as at least working hypotheses are being split. The consensus now is that “Khoisan” is three to five language families with no demonstrable genetic relationship, and Niger-Congo is on weaker grounds than it has been in the past.

    • David Marjanović says:

      Grottoclonology* is a dead horse linguists can stop breaking their toes on. Molecular dating got over the assumption of a molecular clock long ago**; the methods used in the recent Nature papers on the phylogeny and age of Indo-European and now Sino-Tibetan do not assume that any two words are replaced at the same rate, or that any word is replaced at a constant rate throughout an entire tree.

      (…They do, however, assume correlated rates along each internode: if one word speeds up, so do all others proportionally. In biology, that assumption makes them prone to certain kinds of errors when applied to morphological as opposed to molecular data. In linguistics, it might fail when confronted e.g. with the wave of dysphemism that swept early Romance and selectively attacked the basic vocabulary while leaving less basic words alone.)

      * © Justin B. Rye
      ** 25 years is a very long time in molecular phylogenetics.

      • Y says:

        I menat the qualitative idea of glottochronology, that is, the concept of the ongoing attrition of recognizable cognates. Swadesh’s assumption of a constant clock was no more than a crude first approximation, to be sure.

    • j. says:

      remember that these are exponential changes

      No, I don’t remember that, if you re-read you’ll see I explicitly deny treating this as a given.

      I don’t think we have yet established reasonable bounds for rates of linguistic change, and I would think that actually for many types this should be treated as proportional to the total population and/or its density — meaning that the languages of small and poorly connected bands of hunter-gatherers would be expected to show a slower rate of development on some fronts. Again, this gives you exponential decay on long enough timescales, but does not guarantee you can straightforwardly extrapolate from known rates of change in agricultural societies. “Long enough” might as well end up being in the timescale of 100,000 rather than 10,000 years in a number of cases. (I think we can straighforwardly rule out timescales of millions of years, though, otherwise we’d all still be speaking barely dialectified Classical Babelic.) This is the point: “say > 8,000 y” is a fabricated number and not an actually known limit.

      Additionally even in the presence of constant change, I would also think sound change isn’t actually long-term exponential as much as long-term attracting/cyclic: something looking like *t > *d > *dʰ > *tʰ > *θ > *t > *tt > *ttʲ > *čč > *č > *š > *s > *ɬ > *t > … is substantially more likely than anything looking like *t >> *f >> *ŋ >> *ü >> *ve >>… (even if you can find a handful of languages here and there that show major “bucking” developments like *t > /f/ or *ü > /ve/). The main things making long-range phonology hard are leniting developments where correspondences drop off to zero (though even this is constrained by maintaining intelligibility), plus derivational morphology introducing pseudo-correspondences. — A good concept I’ve recently picked up from an Egyptological paper is that of “pillar consonants”: unusually stable segments that can be expected to show mostly trivial correspondences, e.g. the front nasals m, n. German /naːmə/ and Hindi /naːm/ don’t just show regular correspondence (as also in e.g. /noɪ/ ~ /nəjaː/ and /maʊs/ ~ /muːs/), they show persistent identity.

      I suspect cyclic semantic change is also a thing to some extent, though we may not have yet detected this in detail. Verbs like ‘cut’ or ‘tie’ seem like promising candidates.

      Lastly I think you might be making a third common mistake: I am not (yet) talking about comparing straight off the bat languages A and B that have separated 10,000+ years ago, I’m talking about the possibility of reconstructing a large and old family step-by-step, where A and B separated 2000 years ago, proto-AB and proto-CD separated 2000 years further back, these separated from proto-EFGH 2000 years still further, etc. so that every single “reconstruction step” remains shallow enough that we should be able to reconstruct the history involved in some detail. This won’t be possible always; even if Khoisan weren’t abandoned, there are just too few languages in there to possibly yield this kind of a setup. What remains of Greenberg’s other African families are however all still pretty big units after pruning. Within Niger-Congo, the Benue-Congo core looks as solid as ever, and also the extended Atlantic-Congo may be moving forward now that it’s common knowledge that “Atlantic” is not a single branch that should be reconstructed as a first step. Similarly, while comparing German and Hindi in isolation results in a lot of noise, allowing for more data and then reconstructing Proto-Germanic and Proto-Indic first will help a lot.

      • Y says:

        My skepticism regarding future long range connections is not entirely based on glottochronology. Rather, it is based on the fact that no established families have established archaeological or historical ages past the high thousands. The oldest families listed in the calibration list of the ASJP paper are Benue-Congo (6500 y), IE (5500), and Pama-Nyungan (4500). I have seen Proto-Afro-Asiatic and Proto-Australian dated to 10,000 or so ybp, but haven’t seen any concrete extra-linguistic support for these dates. Are there any other examples?

        Put another way: there are dozens or hundreds of language families which have have already been established using linguistic methods. Are there any of them which you think could be demonstrated to be much older than the 6,500 yo Benue-Congo family?

        Sound change: My issue is with non-reversible phonological changes, like sound mergers or segment deletions. Things like accretion and fossilization of nominal class markers or manner suffixes also act sporadically, and across the lexicon. Several rounds of these over the millennia could obscure genetic relationships more thoroughly than per-word semantic shifts and innovations.

        I appreciate your point about demonstrating and reconstructing proto-languages in stages. However, for now I remain pessimistic. The reconstruction of PIE and PU have relied on a great deal of material from many languages. Such reconstructions often rely on the varied histories of daughter branches, each of which may be conservative in a different aspect, and put together they may preserve a large number of hints regarding the protolanguage. Suppose now we go from reasonably constructed PIE and PU to a putative common ancestor: it will have to be reconstructed from only two daughter languages, and a lot less will be certain about it.

        • j. says:

          Things like accretion and fossilization of nominal class markers or manner suffixes also act sporadically, and across the lexicon. Several rounds of these over the millennia could obscure genetic relationships more thoroughly than per-word semantic shifts and innovations.

          For sure. This is basically the proposed root structure / reconstructibility correlation: in families with both prefixation and suffixation, root identity is more likely to end up obfuscated due to phonological change (we could maybe add ablaut-plus-apocope a la IE zero grade as a third such mechanism). If this happens at all depends however on if the language’s grammar allows e.g. noun class markers in the first place, secondly also on if they do accrete or not. These kind of morphosyntactic features are themselves quite a bit more stable than their particular realizations. Benue-Congo is workable precisely because its class markers have remained productive, while if straggling groups like the Kordofanian bits end up being relatable after all, then that probably requires working out some amount of class marker fossilization processes.

          (This is, incidentally, a good heuristic argument against Ural-Altaic or Uralo-Siberian etc.; these families all have exceedingly clear root-initial word structure, and if these really were not too distantly related, they should have relatively clear cognates. Instead looking for them comes up not much better than comparison with phonologically more difficult families like IE, Kartvelian or Chukotko-Kamchatkan, suggesting that either the family tree is more complicated or we’re drawing up mostly chance resemblances.)

          I quite firmly agree that just keeping hammering at some popular hypotheses like Indo-Uralic is the wrong way to go about long-range reconstruction; we need some good clues first on what is likely to survive for long periods (under what morphosyntactic, phonological etc. conditions) and therefore able to be reliable evidence also in cases where we have longer single-branch development to every daughter group.

          Which then brings me back to your first question: no, we don’t yet have clear dates over 7000 years for anything — but, we also don’t have clear dates at all for several usually accepted families. This is one of the most important points I wanted to get at in this post! We should wrap up investigating that first of all before making proclamations about what is possible or impossible. Otherwise you’re not really saying anything new, are you?

          To bring the point home clearly enough, maybe I should do a quick run-over of some particular families in need of more work in a separate post entirely.

        • David Marjanović says:

          Suppose now we go from reasonably constructed PIE and PU to a putative common ancestor: it will have to be reconstructed from only two daughter languages, and a lot less will be certain about it.

          In this case we are – potentially – lucky in that much of what ties IE and U together hints at wider relationships. At minimum, I would compare IE, U and Yukagir.

    • There is no basis for the numbers like 10000 or 8000 years other than gut feeling. I strongly advise anyone who is interested in this issue to read A. Manaster Ramer’s paper “Some Uses and Abuses of Mathematics in Linguistics”, specifically the part starting from p. 111.
      This does not mean that I am very optimistic about finding super-deep relationships. Rather I would say that 1) the temporal limit depends on so many variables that it must be different in each particular case, and 2) we’ll never know the limit for any such case unless we try to approach it.

      • Y says:

        I am not married to glottochronology, particularly the simplistic Swadeshian version, but it does happen that the estimate for the maximum time limit that it predicts, that 7000 years or so, coincides with the oldest language families which are confidently identified and dated.

        • David Marjanović says:

          My impression of the language families that are confidently identified and dated is that they are simply the easiest problems. Many others are solvable, but haven’t been solved because that simply requires a lot more work at all levels. Does Sino-Tibetan have any identifiable relatives? Welp, to investigate this question thoroughly and rigorously, we’ll first have to figure out the internal phylogeny of ST and reconstruct its protolanguage. And before we can do that in earnest, we first have to describe a few hundred languages that are currently known to science only in superficial ways. Then we have to reconstruct a lot of branch protolanguages; so far we have Proto-Lolo-Burmese, and apparently that’s pretty much it. And then some genius or team of geniuses has to put a hundred dictionaries and grammars on a desk(top) and compare them all, after somehow securing funding (remember that the funding for the Lexikon der indogermanischen Nomina was cut).

          I’m quite optimistic about what is possible. It’s a lot more than what is feasible right now.

          • I agree with Y. above that the loss of common inherited material across languages is exponential. Since unlike in DNA there is no lexical or morphological component that needs to remain stable for the languages to remain functional, it follows that there will necessarily be a limit after which there will be no way to distinguish between chance resemblances, borrowing and related material (it may be that 10000 BP is too low in some cases, but nevertheless there must be a limit). As pointed out by Mikhail, this limit depends on many factors: in my opinion, it should be much lower in the case of isolating languages. For example, while I do believe that hmong-mien is probably genetically related to AN (and hence to Kra-Dai and potentially ST, see https://panchr.hypotheses.org/2262), the lack of morphology in this family means that distinguishing between contact and inheritance may be a moot question.

            • j. says:

              There’s surely an eventual limit after which too much has been lost, but speaking of exponential loss is a stronger claim, equal to claiming that at least certain types of loss proceeds at a stable rate. This is what radioisotope decay does (you can straightforwardly divide the decay rate per millennium by 365,250 to gain the decay rate per day), but I don’t think this holds for loss of linguistic material. Words and structures do not spontaneously fissionate, they get replaced by a large number of various means which have variable rates, both amongst themselves, depending on lexical and structural conditions, and depending on sociolinguistic conditions.

              I’m not sure if by “much lower” you mean “more recent” or “more remote”. (My guess would be that for isolating languages, replacement-via-derivation and phonetic erosion are lesser problems, but replacement-via-synonymy could be more so, and which effect wins over in any given case is not clear.)

              • Howl says:

                “you can straightforwardly divide the decay rate per millennium by 365,250 to gain the decay rate per day”.

                This is only true as an approximation. Exponential decay is NOT linear. If an atom has a decay rate of 50% per year, it does not have a decay rate of 100% per two years (but actually 75% per two years).

                • j. says:

                  Right, sorry, inexact terminology: divide the the logarithm of the decay rate over a period to gain the same for another period.

              • Y says:

                To clarify, what I’d meant before by “exponential” was more a qualitative statement than a quantitative one. Of course lexical change does not follow an exact clock the way radioactive decay does. Swadeshian decay dis an approximation. And yet qualitatively and subjectively, a 6,000 y.o. language looks so much older than one 3,000 y.o., in the same way as they would if language decay, by some measure, was exactly exponential.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Enter your email address to follow this blog and receive notifications of new posts by email.

%d bloggers like this: