The Two Ways to Say “Celtic”.

Stan Carey has a mental_floss post on the word Celtic that says just about everything that needs to be said on the subject. He explains that “The now-dominant pronunciation ‘Keltic’ is a modern innovation”:

We can see the shift by comparing Fowler’s original Dictionary of Modern English Usage with Robert Burchfield’s revised third edition. Here’s Fowler, 1926: “The spelling C-, & the pronunciation s-, are the established ones, & no useful purpose seems to be served by the substitution of k-.” Burchfield, 1996: “Except for the football club Celtic (in Glasgow), which is pronounced /’seltɪk/, both Celt and Celtic are pronounced with initial /k/ in standard English.”

Burchfield doesn’t mention the Boston Celtics, and that’s not his only oversimplification. Celtic may be pronounced either way in standard English—even if this bothers some people. A lot of antagonism over language use stems from misconceptions about correctness, such as the common belief that there can be only one correct form of a word (one meaning, spelling, pronunciation, etc.), and that variants are therefore wrong. […]

According to the Merriam-Webster Dictionary of English Usage, “the closer you get to circles substantively concerned with Celtic lore and languages, the more likely you are to hear \’kel-tik\”—though “Seltic” may be heard “at times from very well-educated speakers.” The American Heritage Dictionary elaborates:

Although many people pronounce this word with an initial (s) sound, an initial (k) sound is standard in historical, linguistic, and sociological contexts. Interestingly, the introduction of the (k) sound is a linguistic change started by scholars, contravening the historical development of the word.

English borrowed Celtic in the 17th century from French celtique, soft-c, and from Latin Celtae, also soft-c in Britain at the time (unlike Classical Latin, which used a hard c). Centuries later the pronunciation changed, because language, but it didn’t switch from “Seltic” to “Keltic”—it just added the variant, which then spread. So now we have two acceptable forms. (And two spellings: Keltic, though unusual, is a variant that recalls Greek Keltoi, “the Gauls.”)

As both a Celt and a linguist, Stan is in a good position to adjudicate the matter, and his conclusion is unimpeachable:

Claims about correctness in language can’t override the facts of usage, and the important fact here is that both pronunciations are standard and correct. […] Critics are entitled to dislike “Seltic” or “Keltic,” but they have no business saying either pronunciation is wrong. Because they’re both right.

Comments

  1. For me, it’s ‘seltic’ when talking about the basketball team, and ‘keltic’ in any other context.

  2. Eli Nelson says:

    The /k/ pronunciation is definitely an oddity in terms of modern spelling. I prefer /s/ for that reason. “Sceptic/skeptic” is kind of the opposite, in that everybody pronounces it one way, but there are two spellings that are both fairly frequent. The OED suggests it always had /sk/ in English, though. And then there’s “schedule,” which used to have /s/ (the equivalent word in French still does), but now has /sk/ or /ʃ/.

  3. Like Steven, I always use /k/ except when speaking of the basketball team. It is odd that AmEng didn’t take up the spelling Keltic as it did with skeptic – the former being confined to New Age and other marginal contexts.

  4. John Alvey says:

    Steve,

    As an /s/ purist and a grammar Nazi, I would argue that /k/ is an affectation, which is wrong. Who says kenter/kentre, kereal, kement, kentury, kellulose, keiling, kentimeter or kedar? Do you know anyone called Kelia, Kecil (Kekil?) or Kedric? I cannot think of another ce- word in English pronounced ke-, except perhaps for some obscure word of Greek origin. Case closed.

  5. There are a fair few soccer club in both parts of Ireland with “Celtic” in their name, and I suspect they all mimic the Glasgow club’s /s/. The Glasgow club’s Irish founder supposedly used /k/ himself.

  6. Sir JCass says:

    There’s also “celt”, a prehistoric tool:

    And on the tables every clime and age
    Jumbled together; celts and calumets,
    Claymore and snowshoe, toys in lava, fans…

    (Tennyson, The Princess)

    If the Wikipedia article is to be believed, this “celt” has an interesting etymology.:

    The term “celt” came about from what was very probably a copyist’s error in many medieval manuscript copies of Job 19:24 in the Latin Vulgate Bible, which became enshrined in the authoritative Sixto-Clementine printed edition of 1592; however the Codex Amiatinus, for example, does not contain the mistake In the passage: Stylo ferreo, et plumbi lamina, vel certe sculpantur in silice (from Job 19:24, “Let it indeed be carved with an iron pen on a plate of lead or in stone”), the certe (“indeed”) was spelled as celte by mistake, which would have to be the ablative of a non-existent third-declension noun celtes or celtis, the ablative case giving the sense “with/by a celt”. Neither the Hebrew text (בעט־ברזל ועפרת לעד בצור יחצבון) nor the Septuagint (εν γραφείωa σιδηρώ και μολίβδω η εν πέτραις εγγλυφήναι) has a word corresponding to either certe or celte.

  7. Certe in this context is probably not so much ‘indeed, certainly’ as ‘fixedly, forever’.

  8. If the Wikipedia article is to be believed, this “celt” has an interesting etymology

    Thanks for that, I love a good etymology-by-corruption story!

  9. I would never say Seltic except for the basketball team. This is certainly influenced by the unambiguous stop at the beginning of “Gallic” and “Gaelic.”

  10. I say “seltic” for the Boston basketball team and the Glasgow football team. Am I alone in following both?

  11. J. W. Brewer says:

    Does anyone pronounce “Celticist” as “Keltikist”?

  12. I presume the fact that Welsh ‘Celt’ and the various Gaelic versions have C- spelling and K- pronunciations (in accordance with the normal spelling rules in those languages) helps to encourage the K- option in English? Logically, of course, it should make no difference to English conventions given the direction of borrowing, but people may feel it is somehow more ‘authentic’.

  13. I seriously doubt whether more than the tiniest minority of English-speakers are aware of that feature of Welsh orthography, and it seems unlikely that that minority has much influence on English pronunciation.

  14. J. W. Brewer says:

    But if you think, as seems plausible, that the vogue for the k-pronunciation started in academic or elite or “enthusiast” circles and then was diffused more broadly, that sort of “say it the way the Welsh do” possibility seems a little less far-fetched.

  15. not so much ‘indeed, certainly’ as ‘fixedly, forever’.

    Indeed, the OED s.v. celt says that not only is the etymology celte < certe, but that this certe represents the Hebrew לעד la’ad ‘forever’ in the original. I don’t see anything in the LXX representing this concept, though.

  16. Chris McG says:

    I seriously doubt whether more than the tiniest minority of English-speakers are aware of that feature of Welsh orthography, and it seems unlikely that that minority has much influence on English pronunciation.

    Within the UK, Cymru being the Welsh name for Wales is pretty well-known (not least from the presence of Plaid Cymru in General Elections), and while I’ve heard it mispronounced in a good dozen different ways, I’ve never heard someone get the ‘c’ wrong

  17. @languagehat: I have heard “the Celtic languages don’t have a soft c” offered by Americans as an explanation for the /k/ in Celt. People over here probably have Irish in mind more than Welsh, of course.

  18. I say “seltic” for the Boston basketball team and the Glasgow football team. Am I alone in following both?

    Are you alone in (a) following [“supporting”, “rooting for”] both teams, or (b) following [“adhering to”] both pronunciations? The answer to (a) is that there must be many Boston Irish who do; the answer to (b) is that almost everybody does.

    There must be some people who (1) have never heard of any team called “Celtic[s]” and (2) use the modern-default pronunciation “Keltic” for the general adjective; any such person would probably use the same pronunciation on first encountering the name of the team, and probably accept their mistake once corrected.

    There must be some people who (1) have heard of precisely one team called “Celtic[s]” and (2) use the modern-default pronunciation “Keltic” for the general adjective; I don’t know whether their guess on first encountering the name of the other Celtic[s] would match (1) or (2).

    So when did “Ceilt” enter the Irish language? “Celsius” in Irish is pronounced as in English.

  19. @John Alvey, I do know a woman named “Ceri” pronounced “Keri”. I’ve known her since 1978, so it’s no innovation.

  20. There must be some people who (1) have never heard of any team called “Celtic[s]” and (2) use the modern-default pronunciation “Keltic” for the general adjective;

    I think your best bet for finding such people would be in Australia, where “Anglo-Celtic Australian” is a current term but no teams, as far as I know, are called the Celtics.

    There must be some people who (1) have heard of precisely one team called “Celtic[s]” and (2) use the modern-default pronunciation “Keltic” for the general adjective;

    This described me until a few years ago, when I learned about the Edinburgh team; but I learned about its pronunciation as soon as I learned of its existence, so I can’t honestly say what my intuition would have been.

  21. J. W. Brewer says:

    Proper names in the U.S. can and sometimes do deviate from all usually-standard orthographic conventions matching spelling to pronunciation, especially when someone or their parents has pursued a variant spelling. Googling suggests the existence of a non-zero number of people named “Ceitlyn,” which presumably has the same hard-c initial sound as the more popular spellings Caitlin etc etc etc. But looking through the SSA’s list of the thousand most popular names for 2014-born US babies, the only Ce-initial ones have the soft-c pronunciation you would expect: Cecilia, Celeste, Cecelia, Celine, and Celia. (The boys’ list has no Ce-initial entries.)

    The Irish version of Cecilia fwiw appears to be either Sisile (for the saint) or Sheila/Shelagh (as a more common given name), from which I guess we are to infer that the C had already shifted from /k/ to /s/ in Latin before the lexeme was borrowed and the Celts weren’t going to re-harden it for the sake of a spelling pronunciation.

  22. But if you think, as seems plausible, that the vogue for the k-pronunciation started in academic or elite or “enthusiast” circles and then was diffused more broadly, that sort of “say it the way the Welsh do” possibility seems a little less far-fetched.

    A good point.

    I have heard “the Celtic languages don’t have a soft c” offered by Americans as an explanation for the /k/ in Celt. People over here probably have Irish in mind more than Welsh, of course.

    Interesting! I agree with your second sentence, though.

  23. There are rules of spelling in English, and they shouldn’t be ignored without good reason. Ce spells S. English spelling is hard enough without people going out of their way to adopt irregular pronunciations. We don’t pronounce other words of Greek origin as if they were Greek.

    IMHO “keltic” is a shibboleth. It’s a hyper-correction. It’s become too widespread to get rid of so it has to be accepted as “standard,” but it’s an example of some people thinking that they’re smarter than they are.

  24. This seems like a good opportunity to mention a curiosity I came across when researching the life of Lord Kelvin. In his Baltimore Lectures, Kelvin mentioned a term cyboid symmetry that was used by the Scottish physicist/engineer W.J.M. Rankine. By way of explanation Kelvin said that one of Rankine’s enthusiasms was his insistence on pronouncing English words of Greek origin in a way that, according to him, more accurately reflected their original pronunciation.

    Kelvin observed: Rankine was the last writer to speak of cinematics instead of kinematics. Cyboid is a very good word, but I do not know that there is any need of introducing it instead of Cubic.

  25. The /d/ in London and the /l/ in Bristol are the results of hypercorrection, but none the worse for that, and John Emerson (wherever he is) would probably be pleased to know that his Dravidians were originally Damilians (cf. Tamil), with the /r/ and /v/ inserted into the Skt form by hypercorrection representing an attempt to reverse the simplification of Skt consonant clusters.

  26. James Kabala says:

    I wonder if soft-c pronunciation for the general term is more common in Boston or Glasgow or any other area where one of the Celtic(s) teams is popular. I am from New England myself and would generally use the soft-c pronunciation across the board.

  27. “The /d/ in London .. [is] a hypercorrection”
    Cite, please. The [conjectural] etymologies I can find have a d.

    “the /l/ in Bristol … [is] a hypercorrection”
    Completely different. “Bristol” has been in use since 1200, long before modern standardized spellings, and spelling it with an l doesn’t require you to learn a rule applicable to only one word in the whole language.

    People who say “keltic” seem to think that that’s the “authentic” way, as if “keltic” were the Celtic way to say Celtic. But in English, it’s the German way. If you think the German way is better then spell it Keltic. It’s just bizarre to use the French-origin spelling and the German-origin pronunciation.

  28. “The /d/ in London .. [is] a hypercorrection”
    Cite, please. The [conjectural] etymologies I can find have a d.

    Etymology is exactly where the hypercorrection comes in. The idea is (and I don’t know if it’s true) that the pronunciation was “Lunnen,” but because it came from Londinium it was hypercorrected to “Lunden.”

    It’s just bizarre to use the French-origin spelling and the German-origin pronunciation.

    How do you feel about the pronunciation of colonel?

  29. It was Londinium in Roman times all right, but was pronounced “Lunnon” historically, the /d/ having been lost. The /d/ was restored from the spelling at some point, though not in Scots. Sorry to be so vague about dates, but I can’t find a useful reference. The same story applies to all initial /h/ in English: lost, and then restored irregularly from the spelling, and not quite the same way in all varieties either, notably with herb.

    as if “keltic” were the Celtic way to say Celtic

    Well, /k/ is the Celtic way to say it now. Consider the names of the Celtic Congress: Ar C’hendalc’h Keltiek, An Guntelles Keltek, Yn Cohaglym Celtiagh, A’ Chòmhdhail Cheilteach, An Chomhdháil Cheilteach, Y Gyngres Geltaidd, all with /k/ (modulo initial mutations). None of these are native words, to be sure: all are originally from Greek, thanks to Edward Lhuyd/Llwyd. So English /k/-users are in pretty good company.

  30. Speaking of Celts, I just came across this article which claims that DNA analysis of Irish skeletal remains vindicates the “Celtic from the west” hypothesis advanced by John Koch – which appears to hinge on his controversial identification of Tartessian as Celtic.

  31. Celtic with an “s” in English is older than the loan-word Keltiek/Cheilteach/Celtiagh/etc in the Celtic languages. There’s nothing authentic about the English pronunciation “keltic.” It’s a 20th-century affectation.

    As for colonel, I’ve got nothing against the pronunciation. It’s the spelling that sucks. Colonel/kernel is a bizarre historical accident. But least it was an accident. “Keltic” is intentional.

    “Keltic” is not like the natural and regular evolution of pronunciation – like the American vocalization of voiceless consonants in some positions, or the loss of r, or the migration of vowel sounds – that’s “language happens.” But this was an intentional crime against spelling. It wouldn’t have bothered me if people had said, let’s spell it Keltic! But to say, let’s spell it Celtic and pretend it’s Keltic! – boy, that’s annoying.

    I think that people should try to avoid introducing new irregular pronunciations into English for no reason and you’re saying, hey, look, here’s one! and here’s another one! I know, there are a boatload of them. That’s no reason to go out of your way to make another one.

  32. people going out of their way to adopt irregular pronunciations… some people thinking that they’re smarter than they are… People who say “keltic” seem to think… nothing authentic about the English pronunciation “keltic”… let’s spell it Celtic and pretend it’s Keltic! – boy, that’s annoying…

    I say “keltic”, don’t think I’m smart for doing so, don’t have to go out of my way to do so, am not knowingly engaging in any form of pretense, and at the risk of annoying you further, I’ll tell you what I actually think, however your logic tells you I must think. I think that “keltic” is the way my teachers pronounced it from primary school onwards, the way my parents and siblings pronounced it, and my friends and colleagues, and people I hear on the radio and television… It’s what I’ve heard all my life. If I were to start saying “seltic”, it would be heard by my speech community as an eccentricity or an affectation. In light of this overwhelming influence of speech community, the one that matters most to me by far as a speaker, why should I care how my pronunciation measures up against some historical measure of authenticity?

    Of course, there may be some suitably qualified version of your complaint that seems somewhat more defensible on its face: you might be assuming that we’ll all understand you to be talking about Americans, for example, though I don’t know why you should assume such a thing in this forum. My point is this: in some very large proportion of cases, people pronounce words the way they do because that’s the way they hear them pronounced, and I would expect a person with some basic understanding of how language acquisition works to see the futility of imputing disreputable motives to people on no stronger grounds than the way they say certain words.

    As for “crime against spelling” – the hell with that. The spelling causes not a moment’s trouble to me nor anyone I know, any more than “Houston” does to you.

    When I find myself at the receiving end of a harangue about a usage that comes entirely naturally to me, and to the people around me, it gives me some inkling of what it must be like to be looked down on every time you open your mouth to speak your own language.

  33. Hear, hear!

  34. I don’t know the actual, historical origin of the “k” pronunciation. But I strongly suspect that its widespread adoption resulted from a new generation arising for whom it sounded cooler, for whom it was more fun to say. This, surely, is how pronunciation changes that succeed come about — changes in taste. To rail against them is the linguistic equivalent of railing against the miniskirt, or the hoodie. Good luck with that.

  35. Here’s a recent paper which examines Koch’s ideas without coming to a strong conclusion.

  36. Trond Engen says:

    There’s a difference between reading-pronunciation and hyper-correction. I’d sort Lunden under reading-pronunciations.

    I read Bloix’s railing against ‘Celtic’ pronounced with k not so much as a railing against ‘Celtic’ pronounced with k as a railing against ‘Keltic’ written with c. Spoken language is what it is, but how we represent it in writing is a choice. There are now two competing forms. They have different connotations and different socio-linguistic implications. Etymologically there’s a case to be made that these are independent borrowings. The result may be that one or the other wins, or it may be lexical specialization. In the latter case it will not be helpful to write both with c.

  37. AJP Crown says:

    The /d/ in London…Cite, please.

    My late uncle’s very old thatched cottage, built of cob (mud & straw), is on a patch of the Leicestershire village of Dunton Bassett called Little Lunnon or Lunnen. It’s said (by my uncle) that the name comes from it being the site where evacuees from the 1665 Great Plague lived.

  38. The woman named “Ceri” is English, not American.
    I said “Celtic” all my life with a soft c, until I began to study “Celtic” art, where my teacher used a /k/.
    I think (I may trip up) that I conventionally still say it with a soft c except in the context of art. Insular majuscule & minuscule scripts, interlace, key patterns, the Books of Kells, Lindisfarne, &c.–when I speak of those I use a hard c. In English. It all drives me a little crazy; because a lot of my teaching is in France in French, & there’s a lot of code-shifting. The only place the hard c pronunciation gets a look-in is in Brittany, & the Bretons write K when they mean “k” (we had an exhibition once at a place called “Ar Bed Keltiek” in Brest); I have never heard “celtique” pronounced with other than a soft c by a French person.
    But I stay with the anglophone shibboleths I was taught, in my area of study, in English. Largely, I think, so that my students won’t find themselves mocked when they travel.

  39. Good practices, but I can well imagine not easy to maintain.

  40. Sir JCass says:

    Wikipedia on the names of the Celts suggests the “K” pronunciation was pushed by Irish nationalists. It includes this hilarious quotation from 1857:

    “Of all the nations that have hitherto lived on the face of the earth, the English have the worst mode of pronouncing learned languages. This is admitted by the whole human race […] This poor meagre sordid language resembles nothing so much as the hissing of serpents or geese. […] The distinction which English writers are too stupid to notice, but which the Irish Grammarians are perpetually talking of, the distinction between broad and narrow vowels—governs the English language. […] If we follow the unwritten law of the English we shall pronounce (Celt) Selt but Cæsar would pronounce it, Kaylt. Thus the reader may take which pronunciation he pleases. He may follow the rule of the Latin or the rule of the English language, and in either case be right.”

  41. January First-of-May says:

    I say “kelt” in English because that’s how we say “Celt” in Russian (and I say “kedar” for essentially the same reason; many of the other words in John Alvey’s list start with /ts/ in my speech – partly also due to Russian pronunciation – though which particular ones do seems to vary from one time to another; a few years ago, it might well have been all but “cedar”, but now it’s irregularly normalizing). If referring to the Celts specifically, I would probably say “keltik” for “Celtic”, because, as already mentioned, “Celt” is “kelt”.

    I learned of the Glazgo Seltik football club (from my brother’s extensive football monologues) before I found out how it’s actually spelled (that is to say, Glasgow Celtic); though it’s not like there were any other plausible spellings 🙂 and, to the best of my knowledge, I’ve never heard of the Boston Celtics basketball team, in any pronunciation, before this thread, but I would’ve probably said “Seltiks”. I think I just don’t mentally associate the word “Celtic” as a team name with the Celts at all – to me it’s a different word that happens to be spelled similarly.
    I did, however, pronounce the other big British football club with non-final “ce” in its name with a K – Leikester City. (In case you wonder, I do say “City” correctly.) But now I’m more likely to say Leichester. (Yes, I know it’s actually Lester.)

  42. Lester Kitty–now that would be a great name for a sports team!

    And I say “Boston Keltics” to confound my Boston Irish friends.

  43. I think I just don’t mentally associate the word “Celtic” as a team name with the Celts at all – to me it’s a different word that happens to be spelled similarly.

    That’s probably the most sensible way to approach it.

  44. On another note, I wonder whether this would be a good thread to initiate a discussion of whether the “Celtic” languages are really a subfamily within Indo-European (like Slavic or Germanic) going back to a “common Celtic” or Ur-keltisch, or, as I’ve read some have suggested, just a bunch of late Indo-European dialects that share some common innovations through diffusion and parallel development, as well as having some individual features that distinguish them from one another. The recent article on Irish DNA findings suggests that the current population of Ireland may go back to around 2500 BCE–during the period and far too early, if I’m not mistaken, to be speaking something like an Ur-Keltisch. And–again if I’m not mistaken–there’s no archeological evidence of a “Celtic”-speaking population entering Ireland in the first millennium BCE, as used to be thought. And the notion of a Celtic “people” is a construct of 19th century racialism and Romantic nationalism, as discussed in the current issue of TLS. Of course, that’s very simplistic, but I’d be interested to read some thoughts by commenters who, unlike me, know something about this. David Marjanovic, I’m thinking of you in particular.

  45. I too would be curious to know what David M. has to say.

  46. J. W. Brewer says:

    How do you suppose this fellow pronounces it? “Harry is currently exploring the migration of the Celts from northern India from a musical point of view, and preparing music for a feature film. He is also developing a musical collective to explore the musical synthesis of Bollywood and the Australian Bush.” https://en.wikipedia.org/wiki/Harry_Williamson

  47. Sir JCass says:

    I wonder whether the “k” pronunciation of “Celtic” might have been part of a general trend towards more “authentic” pronunciation or spelling. You get this kind of thing in Leconte de Lisle’s poetry. Take these lines from “Le massacre de Mona” from Poèmes barbares, for example:

    Et ceux qui, dans les bois, portent la Serpe d’or,
    Ceux de Kambrie et ceux d’Érinn et ceux d’Armor.

    He also writes of “les Dieux Kymris”.

    Oscar Wilde calls John the Baptist “Iokanaan” in Salome. It’s a kind of “alienation effect” to make you see familiar things from a different angle. Leconte’s “Chiron” is “Khiron” (and “K” is even more “alien” in French). Likewise, some modern translators of Homer have changed “Achilles” to “Akhilleus”, “Hector” to “Hektor” etc.

  48. I don’t know about Celtic specifically, but the idea that the IE subbranches are the results of areal diffusion/convergence rather than true clades has been argued for by Andrew Garrett. Two relevant papers:

    http://www.linguistics.berkeley.edu/~garrett/BLS1999.pdf
    http://linguistics.berkeley.edu/~garrett/IEConvergence.pdf

    I think Piotr was going to discuss this theory on his blog at some point, but that doesn’t seem to have happened as of yet.

  49. Houston actually does cause me moderate amounts of trouble when I have to explain to tourists that it is pronounced with the MOUTH rather than the GOOSE vowel in connection with the NYC street, unlike the Texas city.

    I’d say that reading pronunciations are a subset of hypercorrections: they result when people assume that the spelling gives the true pronunciation and the actual pronunciation is a corruption, rather than seeing the spelling as an imperfect rendering of the pronunciation. While I’m at it, the term hypercorrection as a nominalization of hypercorrect strikes me as subtly wrong, suggesting an excessive degree of correction (which is not what is meant). I think hypercorrectness (as in political correctness) would have been a better technical term. But too late now.

    Both London and Lunnon are surnames, I forgot to mention. We can perhaps say that /s/eltic is now likewise preserved only in names, at least among the majority of its users.

    Boswell noted that Johnson pronounced heard as if spelled heered, and when Boswell asked why, Johnson replied that in all other cases the spelling ear was pronounced so, with heard the single exception, and he preferred not to have that exception. (This may have been a rationalization on Johnson’s part: it is known that he had a Staffordshire accent throughout his life, and apparently did not have the FOOT-STRUT split: Boswell records him saying poonsh for punch.)

  50. The few times that I’ve heard Brits speak of Houston, the city, it’s been with GOOSE.

    I think hypercorrectness (as in political correctness) would have been a better technical term. But too late now.

    But then we couldn’t use it in the countable sense.

  51. January First-of-May says:

    I’m not familiar with the vowel description system, but don’t “foot” and “goose” have the same vowel?
    I was sure that Houston in Texas is pronounced Huse-ton, same vowel as “huge” and “human” (or as “new”, come to think of it). I mean, the Russian transliteration Хьюстон had to come from somewhere – it’s not Хустон (which I would’ve expected with the vowel of “foot”, which, again, I believe to be the same as the vowel of “goose”) or Хаустон (which I would’ve expected with the vowel of “house”).

    Never heard of the New York street before. Probably wouldn’t again for a while (since I do not intend to travel anywhere near that continent anytime soon).

  52. “I think I just don’t mentally associate the word “Celtic” as a team name with the Celts at all – to me it’s a different word that happens to be spelled similarly.”

    Bostonians are very conscious of the fact that the team is named for their city’s Irish-American population. If they forget, they’re reminded by the team mascot, Lucky the Leprechaun. Origin of the name: “Boston is full of Irishmen. We’ll put them in green uniforms and call them the Boston Celtics!” http://www.nba.com/celtics/history/Name.html

    As for Glasgow’s Celtic F.C., it was founded by an Irish Catholic brother, its symbol is a shamrock, and its original fan base was Irish immigrants to Scotland. If “keltic” had been prevalent in the late 19th c, the team would be called “keltic.”

    Even today, Celtic is the Irish Catholic club – Protestants support Rangers.

    “At Rangers’ Ibrox Stadium, the Union Flag and Ulster banner are often displayed, whilst at Celtic Park, the Irish tricolour prevails.” https://en.wikipedia.org/wiki/Sectarianism_in_Glasgow

  53. I’m not familiar with the vowel description system, but don’t “foot” and “goose” have the same vowel?

    No, though it’s a hard distinction to learn if your native language is Russian; foot has /ʊ/ and goose has /uː/.

    I was sure that Houston in Texas is pronounced Huse-ton, same vowel as “huge” and “human” (or as “new”, come to think of it). I mean, the Russian transliteration Хьюстон had to come from somewhere

    And you were right! Хьюстон is an excellent representation, so long as you replace the harsh, throaty Russian Х with the mild, breathy English H.

  54. James Kabala says:

    Once well-established, a sports team name takes on an independent life of its own – e.g., how often does anyone reflect that the baseball team hated in New England has a name that originally meant “New England natives?”

    But I nonetheless think that Bloix is mostly right on “Celtics.” People might not picture an ancient Celt in their minds every time they mention the Celtics, any more than they picture a pair of red socks every time they mention the Red Sox, but I (respectfully) think that “I don’t even think of it as the same word” would strike most Celtics fans as a very odd thing to say.

  55. goose has /uː/

    Well, that’s what we were all taught, of course, but really hardly any English dialect these days has a /u:/ sound. In most varieties of English the GOOSE vowel is significantly fronted, more like [ʉː]. Why high back vowels move forward in language after language (Greek, French, Welsh…) is a question I’d really like to know the answer to.

  56. @John Cown: I don’t think that “hypercorretion” is formed from *”hypercorrect” at all. The progression would be “correct” > “correction” > “hypercorrection.”

  57. PS – thank you, Trond Engen.

  58. Eli Nelson says:

    @TR:

    Well, it often occurs at the same time as raising of /o/ or /ou̯/ to /u/, (this occured in Greek, French and Swedish/Norwegian) although I don’t know whether this is a push or a pull. Despite the spelling “u,” it seems Old Welsh /ʉ/ was actually a preservation of Early Common Brythonic /ʉː/ (Old and Middle Welsh), which Wikipedia says is from from the diphthongs /au/, /ou/, /oi/. It seems probable that [uː] was an intermediate stage (especially since there was no contrastive /uː/), but I haven’t seen anything that specifies this. So /ʉ/ may not have been completely back to start with in Welsh, although it’s definitely moved further forward over time, as it merged with /ɨ/ on the way to Modern Welsh (which I believe is fronted even further to /ɪ/ or /iː/ in some dialects). There is fronting of /u/ involved in the change from Proto-Celtic /uː/ to Early Common Brythonic /iː/.

  59. @TR: Well, I’d say that there are still a lot of North Americans (me included) for whom the quality hasn’t yet reached [ʉ]. It is a valid point that you raise, though. Japanese, interestingly, diverges from the classic 5-vowel paradigm not by fronting the [u] but by unrounding it.

    @Bloix: Then there’s the thorny issue of the Celticity of Scotland. Maps of “Celtic nations” invariably lump the whole thing in, but as far as I can tell the Lowland Scots seem pretty Germanic – at least no more Celtic than, say, Cumbria.

  60. Eli Nelson says:

    @Lazar: Isn’t Japanese /u/ also fronted compared to cardinal [u]? I think I’ve read that it is especially fronted after /j/ or palatalized consonants such as /ɕ/ (which can also be analyzed as the cluster /sj/). And it is labialized as well, just compressed labialization rather than endolabial labialization.

  61. Trond Engen says:

    Hypercorrection is a deed motivated by hypercorrectness.

    (… or that’s how ot feels to me when I see the two words together.)

  62. JFoM: Note that the /j/ in /ju/ is not considered part of the vowel, though historically it comes from either French /y/ or older English /ɪw/, /ɛw/. It has been lost in /rju/ everywhere, and in most people’s pronunciations in /sju/, /zju/. In American English, /tju/, /dju/, /nju/, /lju/ become /tu/, /du/, /nu/, /lu/.

    Brett: The OED reports hypercorrect from 1922 (Jespersen), hypercorrection only from the mid-1930s, and does claim the latter is derived from the former. There are two instances of hypercorrectness, both from the 1950s.

  63. It [the /j/ in /ju/] has been lost in /rju/ everywhere

    Not everywhere. Check out the pronunciation of rhubarb in one of the North of England samples here. I think Welsh English might also preserve this.

  64. People might not picture an ancient Celt in their minds every time they mention the Celtics, any more than they picture a pair of red socks every time they mention the Red Sox, but I (respectfully) think that “I don’t even think of it as the same word” would strike most Celtics fans as a very odd thing to say.

    Respectfully, I don’t care what most Celtics fans would think; my point was simply that for foreign learners like January First-of-May it is probably easier to separate the sports-team name mentally from the linguistic/ethnographic term in order to keep the pronunciations straight.

  65. @John Cowan: I think that OED entry has significant problems. At the very least, the dates of the OED citations for “hypercorrect” and “hypercorrection” are not very useful. A few minutes of Googling dates them both to the nineteenth century, and “hypercorrect” does seem to be older, although the earliest mentions do not appear to refer to specifically linguistic hypercorrectness. Moreover, all early uses I can find for “hypercorrection” appear to have nothing to do with the linguistic notion of “hypercorrect.” They refer to a process in the human eye. In fact, as I think back, I am pretty sure I knew this ophthalmological sense before I knew the linguistic sense of “hypercorrection,” even though the relevant definition is completely absent from the OED. In the ophthalmological sense, it is often hyphenated “hyper-correction,” which suggests that it may be a direct extension of the standard term “correction” in this context, without passing through the very “hypercorrect.” How the linguistic sense of “hypercorrect” (which shows up by the late nineteenth century, at the latest) and the ophthalmological “hypercorrection” (which seems to be of comparable age to linguistic “hypercorrect”) influenced linguistic “hypercorrection” may still be an open question.

  66. Trond Engen says:

    I don’t know about Celtic specifically, but the idea that the IE subbranches are the results of areal diffusion/convergence rather than true clades has been argued for by Andrew Garrett.

    Thanks for the links. Areal convergence and loss of intermediate dialects, as I read Garrett. I think the basic premises are generally accepted. One regularly see things like “Pre-Proto-Germanic, or the set of dialects that would later take part in the formation of the Germanic branch.”

    It”s also clear that dialect continuums can contain remarkable relics of earlier stages of the language while still maintaining the general sense of unity. As I recall it, the Novgorod birch bark letters show a blend of features that point to modern Northern Russian dialects, innovations that must have been lost again, and archaisms that were believed to be lost in common Slavic (or East Slavic). Elfdalian has kept nasal vowels lost after Proto-Norse but has also taken part in the Central Scandinavian vowel shift. The dialect of my father’s birthplace in Northern Norway retains a velar stop in åga “the river” with åg- < *ahWa- (some say the -g- is intrusive and coincidental, but then we ought to see it elsewhere). Etienne’s recent examples of Corsican and Sardinian vowel retentions would also fit in here. One way to look at this is that the initial bi- and trinary branchings of Romance happened in Corsica, then Sardinia, but a more sensible interpretation may be that a diagnostic feature in one context may be completely within normal variation in another.

  67. Man, that’s interesting stuff. I wish I’d known about it when I was a linguist in training.

  68. James Kabala says:

    Sorry – I didn’t realize that your statement only applied to foreign learners. (I knew the person to whom you were replying was a foreign learner, but I thought your statement was intended as general advice.)

    But I guess that brings me back to my original point – I think a lot of people, like me, and perhaps more commonly in New England (or Glasgow) than elsewhere, think of soft-c as the “regular” pronunciation and hard-c as the “ultra-precise academic” or dare-I-say “hypercorrect” pronunciation. When I read all the posts in this thread, I hear the soft c in my head; I would have to make a conscious effort to think or say the hard c. As the original post by Dr. Carey said, soft-c should be considered equally correct.

  69. Well, as a young New Englander I don’t share that perception. For me, hard c is the “regular” pronunciation, without any particular academic air to it, and soft c is exclusively for the team. When I hear the latter used in other contexts, it sounds odd and antiquated to me. (This is my unmediated feeling, of course; I agree that both pronunciations are fine.)

  70. David Marjanović says:

    I have never heard “celtique” pronounced with other than a soft c by a French person.

    Of course not. Their tolerance for such disconnects of spelling and pronunciation is much lower than among the English-speakers of the world.

    BTW, the reason why [k] is attributed to German is that German is fairly happy to borrow learned words directly from Ancient Greek without sending them through Latin first. Clearly, die Kelten are straight from Greek Keltoi, not from Latin Celtae.

    I too would be curious to know what David M. has to say.

    Archeology: “pots, not people”
    Ancient DNA: “people, not languages”

    In the case of Yamnaya, everything fits so well together that it’s safe to say the people of the Yamnaya culture spoke either PIE or a language on the stem of the non-Anatolian side or something very close to these. This case is very much the exception; elsewhere we generally just don’t have that much data.

    However, the Proto-Celtic language can be reconstructed without more than the usual share of problems; the Wikipedia article I’m linking to lists a lot of innovations from PIE, citing:

    Ranko Matasović (2009) Etymological Dictionary of Proto-Celtic. Leiden Indo-European Etymological Dictionary Series, 9. Brill Academic Publishers. ISBN 978-90-04-17336-1.

    Of course I haven’t read it, but it is cited a lot in recent IEist literature, always approvingly or mostly so. – Just yesterday, I read somewhere that it says the change from *p…kʷ to kʷ…kʷ cannot be a common Italo-Celtic innovation, because there are unspecified Italic languages which lacked it and apparently Celtic exceptions, too; I suppose this could mean that the name of the Hercynian forest could be Celtic after all.

    Хьюстон is an excellent representation, so long as you replace the harsh, throaty Russian Х with the mild, breathy English H.

    Most people really do say [hj] as you state; but the merged pronunciation [ç] (with the German ich-Laut, almost exactly the same as хь) is pretty common. A few people just say [j]; two of them are running for President right now and drawing ‘uge crowds, so suddenly it’s a ‘uge deal. 🙂

    People who use [ç] also use it in here.

    It”s also clear that dialect continuums can contain remarkable relics of earlier stages of the language while still maintaining the general sense of unity.

    Absolutely. And so, if you tried for example to reconstruct Proto-Indo-Iranian based on just Sanskrit and Avestan alone, you’d get very far, but you’d reconstruct a fricative for the correspondence of Skt ś : Av s, while the comparatively poorly known Nuristani languages, which are probably closer to Indic than to Iranian, retain [ts] to this day, strongly suggesting descent from a PIIr affricate.

  71. For me, the sports teams are uncomplicatedly /s/, and the rest is a mess — I think I started out using /s/ and then heard people saying /k/ and I didn’t know what was “right” so I just threw up my hands; I think I still mostly use /s/ but feel slightly guilty about it.

  72. David Marjanović says:

    Ancient DNA: “people, not languages”

    …so, if you find that today’s population of some place is mostly descended from the people that lived there a few thousand years ago, that doesn’t mean the same holds for their languages. The purest living descendants of the Yamnaya people, by a small margin, are the Estonians.

    The dialect of my father’s birthplace in Northern Norway retains a velar stop in åga “the river” with åg- < *ahWa- (some say the -g- is intrusive and coincidental, but then we ought to see it elsewhere).

    Interesting. Is it there in any other words with *hʷ or *ɣʷ or maybe just *w?

    Edit:

    /h/ is a (usually voiceless) fricative. The friction is normally glottal [h], but sometimes it is dorsal: palatal [ç] when near front vowels, velar [x] near back vowels. It can be voiced [ɦ ~ ʝ ~ ɣ] between two voiced sounds.[6]

    Maybe this [ɣ] was shifted to [g]?

  73. The /hj/ > /j/ change is characteristic of New York, Philadelphia, Dublin, Cork, and the whole of Norway.

  74. David Marjanović says:

    (Also characteristic of German accents in English, because [hj] is a bit hard to believe. 🙂 )

  75. Trond Engen says:

    Interesting. Is it there in any other words with *hʷ or *ɣʷ or maybe just *w?

    Not that I’m aware of. It may have survived in this particular form of this particular word because of its abundance in toponymy.

    Maybe this [ɣ] was shifted to [g]?

    I imagine something like that. Maybe influenced by the /h-/ of the definite suffix (later lost). The g is absent in all other forms.

    New York, Philadelphia, Dublin, Cork, and the whole of Norway

    Heh. But no, not completely. A couple of dialects have /hj/ > /ʃ/.

  76. Picky, picky, picky. AAVE speakers in NYC and Philly don’t say yuge, but I wouldn’t be suprised if yat-dialect folks in New Orleans did. (No connection with Old Slavonic vowels.)

  77. TR: Thanks for the two papers by Andrew Garrett. Interesting reading. He is right in principle but gives a misleading picture, I think: to me the case for the existence of a “Proto-Greek” language and of a “Proto-Celtic” language are very strong, whereas the case for there ever having existed a “Proto-Italic” strikes me as very, very weak (something I once pointed out here at Casa Hat, pointing to the work of Madison Beeler, as does Garrett).

    Bill W: the reason why I believe Proto-Celtic is likelier to have existed than Proto-Italic boils down to the fact that within a dialect continuum convergence is certainly possible, but it is never total. That is to say, if you have convergence taking place between distinct, genetically related varieties, the effects of this convergence will not as a rule eliminate all the differences which originally set apart the different varieties in question.

    Thus, in the Sabellic languages you find an ending -OM marking the active infinitive, and in Umbrian there is another ending -FI. Both endings are alien to Latin. Even more tellingly, -FI has a cognate in …Indo-Iranian. This and other differences between Latin-Faliscan on the one hand and Sabellic on the other suggests that originally each was a separate branch of Indo-European, each having lost certain Indo-European morphemes but not others, and that the similarities are due to later contact/convergence. This is also suggested by the fact that, as Garrett points out, the verb *system* of Latin and Sabellic are strikingly similar in terms of organization, but NOT in terms of shared morphemes. Such a situation, where you have shared grammar without shared morphemes, is very much like what you find in the Balkans, where various languages share grammatical features without having borrowed (most of their) grammatical morphemes, a state of affairs unanimously agreed to have come about, in the Balkans, through convergence.

    In Celtic, on the other hand, between the two well-attested branches (Goidelic and Brythonic) there are many differences, but none of them point to the two branches having once been separate branches of Indo-European. Let’s take the well-known loss of Proto-Indo-European */p/: this sort of phonological innovation certainly could have spread as an areal innovation. It could also have been a feature of Proto-Celtic. Which is the likelier scenario?

    Well, Brythonic, unlike Goidelic, shifts Proto-Indo-European */kw/ to */p/. This is seen as a very fundamental division between what is called P-Celtic and Q-Celtic. Fine. Could “P-Celtic” and “Q-Celtic” have each been a separate branch of Indo-European, whose common “Celticity” could be explained though convergence?

    The answer is no. Chronologically, because Proto-Indo-European */kw/ never yields zero in Brythonic, it is clear that the two sound changes must have taken place in a strict order: first, */p/ was lost in Celtic, and then and only then, */kw/ became /p/ in Brythonic. Thus, if we want to assume that Proto-Brythonic and Proto-Goidelic each arose as a separate branch of Indo-European, we also have to assume that each had preserved the distribution of Proto-Indo-European */p/ and */kw/ unscathed. And the problem is that as far as I know there is nothing in the phonological, lexical or morphological history of Celtic which would be more economically explained by postulating separate Proto-Brythonic and Proto-Goidelic daughter languages of Indo-European which subsequently converged towards a “Celtic” language type than by simply postulating a “Proto-Celtic” language.

    Now, things are very different in the case of “Continental Celtic” varieties: they are so poorly attested that I could well imagine that some might owe their “Celtic” features to convergence only.

  78. Also characteristic of German accents in English, because [hj] is a bit hard to believe.

    Well. Austrian/Bavarian accents, maybe. In northern Germany, where Chemie, Chinesisch begin with [ç], there should be no trouble with English /hj/. I myself say [ç] in /hju/-words, but not in /hi/- or /hI/-words like heed, here.

    Thus, in the Sabellic languages you find an ending -OM marking the active infinitive, and in Umbrian there is another ending -FI. Both endings are alien to Latin. Even more tellingly, -FI has a cognate in … Indo-Iranian. This […] suggests that originally each was a separate branch of Indo-European

    The evidence, as far as you give it, is perfectly consistent with -fi being a symplesiomorphy (shared primitive character), which therefore shows nothing at all about relatedness. Sharks have gills and trout have gills whereas cows do not, but trout are more closely related to cows than to sharks: having gills is symplesiomorphic.

  79. For me and, I think, a lot of other speakers, /hj/ is in the direction of [ç] but doesn’t quite get there. Canepari characterizes it as a voiceless palatal approximant – [ḩ] in his notation – which I think fits.

  80. > Also characteristic of German accents in English, because [hj] is a bit hard to believe.
    Well. Austrian/Bavarian accents, maybe. In northern Germany, where Chemie, Chinesisch begin with [ç], there should be no trouble with English /hj/. I myself say [ç] in /hju/-words, but not in /hi/- or /hI/-words like heed, here.

    My time in northern Germany ten years ago agrees with David (I think you’re at cross-purposes, John, he’s not saying they’re unable to pronounce [hj], he’s saying they find it hard to credit as a consonant cluster in English). I pointed out that I was using the Ich-Laut in the relevant words to a colleague, he said “huh” and merrily went on his way saying [juman] and [jutʃ].

  81. John Cowan: you’re right, I could have expressed that better: the point is that the Latin active infinitival ending (historically -/se/) is itself an innovation not found in Sabellic. Thus, comparing Latin DIC-E-RE and Oscan DEIK-OM, both “to say”, there is no way a proto-Italic form can be reconstructed which is distinct from the late Proto-Indo-European form: the suffixes are distinct, and the common root, */deik-/, is certainly not “Proto-Italic”: it can be found in other Indo-European languages (cf. Ancient Greek DEIK-EIN). The only common innovation shared by Latin and Sabellic involves the meaning, which shifted from “to point” to “to say”. But while this Italic meaning could be explained by convergence or by common ancestry (from a Proto-Italic language, distinct from Indo-European), the actual forms do not require that a separate Proto-Italic language be postulated.

    You’re right that, in principle, the presence of conservative features proves nothing one way or the other. But if conservative features are found here and there, none of which are shared across all of “Italic”, then ALL these conservative features must be assumed to have been part of Proto-Italic; but the more such conservative features there are (i.e. ones which the data require you to reconstruct), the more “Late Indo-European”-like this “Proto-Italic” will be. And to my mind an adequate reconstruction of the common ancestral language of Latin and of Sabellic languages is so much like Late Indo-European that I strongly suspect that a model whereby Sabellic and Latin were separate branches of Indo-European which subsequently converged would fit the data better.

  82. @Etienne: Of course, Garrett takes a similar position regarding Mycenaean and the other Greek varieties, arguing that the former is basically a Greek-leaning dialect of late nuclear PIE – and that the other IE branches would likely present us with the same problems if we had textual evidence of comparable antiquity. Would you go that far, or do you think Italic is weaker than the rest?

  83. Ancient Greek DEIK-EIN

    Actually deik-nu-nai, with the present suffix -nu- and (usually) athematic conjugation.

  84. John, he’s not saying they’re unable to pronounce [hj], he’s saying they find it hard to credit as a consonant cluster in English).

    Apparently a lot of English speakers agree, at least when the following vowel is anything other than /u/, which is why they had trouble with Comet Hyakutake. Actually that seems to be true for any consonant + /j/ cluster, so pronouncing Myanmar with two syllables is a similar problem.

  85. J. W. Brewer says:

    Yes re Keith Ivey’s point, as can e.g. be seen by tendency of Anglophones to pronounce “Kyoto” and “Tokyo” trisyllabically. I at one point in my life was unduly proud of my ability to do those “correctly” with the /kj/ combo, which was almost certainly merely a function of my having been first exposed to Japanese at age 8, before my baseline sense of what phonotactics were and weren’t cromulent was as difficult to overcome as it would have been at a greater age. Maybe I can do “Myanmar” more easily than most by analogy?

  86. David Marjanović says:

    In northern Germany, where Chemie, Chinesisch begin with [ç], there should be no trouble with English /hj/.

    Trouble is, they don’t know that [ç] is an option in English. The spelling, after all, doesn’t suggest it; and German [ç] isn’t /hj/, if only because /hj/ doesn’t exist at all.

    I think you’re at cross-purposes, John, he’s not saying they’re unable to pronounce [hj], he’s saying they find it hard to credit as a consonant cluster in English

    I’m saying they’re unable to imagine that anyone would go to the trouble of pronouncing [hj], even though the spelling suggests it quite clearly. “They can’t be serious about this”, they assume, and simplify the situation.

    Unexpected phonetics and phonotactics are a very common aspect of language teaching where it helps a lot if a teacher simply makes things explicit and states clearly that, yes, native speakers of the target language really do go to various unimaginable troubles. I mean, some students will think “whatever, I’ll never be able to pronounce this”, but most will try, and many will succeed pretty quickly; they just needed to get the idea, which didn’t occur to them by itself.

    tendency of Anglophones to pronounce “Kyoto” and “Tokyo” trisyllabically

    BTW, these are spelled Tokio and Kioto in German, and then there’s Kenia. All of these are pronounced with three syllables, as the misguided spelling suggests. What’s going on here is an early-/mid-20th-century mechanical replacement of foreign y by i.
    *sigh*

  87. David Marjanović says:

    Thus, in the Sabellic languages you find an ending -OM marking the active infinitive, and in Umbrian there is another ending -FI. Both endings are alien to Latin. Even more tellingly, -FI has a cognate in …Indo-Iranian. This and other differences between Latin-Faliscan on the one hand and Sabellic on the other suggests that originally each was a separate branch of Indo-European, each having lost certain Indo-European morphemes but not others, and that the similarities are due to later contact/convergence.

    But… is -om an innovation, or is it shared with other IE branches? If so, can a difference of meaning be reconstructed between it and the ancestor of -fi?

    Because, if -om and the Latin -se are both innovations, then nothing stands in the way of reconstructing a Proto-Italic language which formed active nominatives in the inherited *-fi.

  88. David Marjanović says:

    I don’t know about Celtic specifically, but the idea that the IE subbranches are the results of areal diffusion/convergence rather than true clades has been argued for by Andrew Garrett. Two relevant papers:

    […]

    I think Piotr was going to discuss this theory on his blog at some point, but that doesn’t seem to have happened as of yet.

    A bit of discussion, in which Piotr participated, did get going; it starts here and extends for a few comments. Mycenaean in particular is mentioned; in short, it’s Greek.

  89. Thanks, David, that’s what I was remembering and should have linked to.

    I’m not sure what you mean by the last four words of your comment; I don’t think anyone would say that Mycenaean isn’t Greek.

  90. January First-of-May says:

    That whole discussion of weird phonemes – and ich-laut in particular – reminded me of NikoZnate’s description of Ancient Egyptian phonology (specifically, the Ancient Egyptian of Akhenaten’s era, which the post called “Late Egyptian”).
    Back when I was active on AH.com (to be precise, about three years ago… I was active on AH.com until December 2015, and would’ve still been but abandoned it for a while and couldn’t find a free weekend to jump back in), I asked NikoZnate (not in the linked thread, IIRC, but in a separate thread which was only visible to logged-in users) whether the “Late Egyptian” phoneme he spelled as “sj” corresponded to Russian хь. He said that it’s not exactly that, but it’s close enough that I could still use that as the best approximation for a Russian speaker.

    BTW, these are spelled Tokio and Kioto in German, and then there’s Kenia. All of these are pronounced with three syllables, as the misguided spelling suggests.
    The Russian spellings are Токио, Киото, and (assuming the African country is meant) Кения, respectively. All of these are pronounced as spelled, that is to say, with three syllables, and in the latter case also with an extra /j/.
    I’ve been previously told that the correct way to pronounce the Japanese cities is as in Polivanov transliteration – Токё and Кёто, that is to say (give or take some consonant specificity), /tokʲo/ and /kʲoto/ (with palatalized consonants, as typical for Russian). Is there an actual /kj/ cluster involved? (I take a pride in trying to pronounce the consonant clusters in foreign words… which are often a result of transliteration in the first place, ironically.)

  91. BTW, these are spelled Tokio and Kioto in German, and then there’s Kenia. All of these are pronounced with three syllables, as the misguided spelling suggests.

    German doesn’t treat i as a semivowel in these cases? I’ve been saying things like [ˈʃpaːnjən], so I’d have expected [ˈtʰoːkjo] and [ˈkʰeːnja].

    I’m not sure what you mean by the last four words of your comment; I don’t think anyone would say that Mycenaean isn’t Greek.

    Garrett does, in the second paper you linked. “
It 
might
 not
 overstate 
the 
case 
to 
say 
that 
Mycenaean
 was 
a
 late NIE dialect with Greek vocabulary; a distinctively
 Greek phonological and inflectional profile was largely
 a 
development
 of 
post‑Mycenaean 
history.”

    @January First-of-May: To my understanding, that syllable in Japanese is [kʲjoː], with a glide but also a palatalization of the preceding consonant.

  92. @January First-of-May:
    Copy 境界 きゅうり きゃっかん キャベツ into the search box at БЯРС and have a listen – just click on the icon!

  93. David Marjanović says:

    German doesn’t treat i as a semivowel in these cases? I’ve been saying things like [ˈʃpaːnjən], so I’d have expected [ˈtʰoːkjo] and [ˈkʰeːnja].

    This only seems to happen in the least stressed positions: in front of unstressed e (further example: Bakterien [bak̚ˈtʰeːjɵn]), and in front of a stressed vowel (Kenianer [kʰeˈnjaːnɐ] – allowing the stressed vowel to be preceded by a phonemic consonant rather than a glottal stop).

    The reason I’m not completely sure right now is that i is never interpreted as /j/ south of the White-Sausage Equator. I actually do say [ˈʃpaːnɪɛn] with 3 syllables, [kenɪˈaːnɐ] with 4 syllables, and [b̥akˈteːɐ̯ʀɪɛn] with 4 syllables and the prevocalic realization of /r/. I cannot, however, remember having ever heard Tokio, Kioto, Kenia with /j/.

  94. In Ireland, ads for “Hyundai” pronounce it almost the same as “high and dry”.

  95. David Marjanović says:

    in the least stressed positions

    Actually, Christian counts. It can be reduced all the way to [ˈk͡χʉʃn̩].

    in front of a stressed vowel

    A further example is the talk show host Sabine Christiansen: [k͡χʉstˈjanzn̩] – I’d have said [ˈkʀɪstɪansn̩]…

  96. @mollymooly: Here in North America, everyone shamelessly pronounces it /ˈhʌndeɪ/. In fact they even made one ad telling us, “It’s Hyundai, like Sunday.”

  97. I pronounce it /ˈhʌndeɪ/ because that’s how it’s pronounced in (American) English; to pronounce it in a more “authentic” fashion would feel to me like saying “Paaah-RHEE” for Paris.

  98. When Hyundai started moving into the American market, they introduced more than just their motor vehicles. I had a friend who had a Hyundai computer (a PC clone, back when we called them that). What was notable was that, like a car, it required a key to turn on.

  99. TR: Thanks for the correction.

    Lazar: Yes, to my mind Italic is definitely weaker, MUCH weaker than the rest. To repeat my point, convergence between genetically related varieties is certainly possible, but as a rule it does not obliterate the differences which had originally existed between the varieties in question. In the case of Greek the discovery of Mycenean does indeed require that some features once considered “Proto-Greek” now be considered due to later convergence: but it most certainly does not thereby automatically follow that ALL the common features shared by Greek varieties must be due to convergence. And as far as I know nobody has shown that the existing divergences between Greek dialects can be better accounted for by assuming convergence between originally distinct Indo-European varieties rather than by descent from a hypothetical Proto-Greek language. The same point is true for the two living branches of Celtic.

    David: let’s assume Proto-Italic existed and had an infinitive marked by inherited Indo-European -FI. For this ending to be replaced by -SE/RE in Latin would certainly be possible. But without the older -FI ending even leaving a trace? This is less likely. And the matter is not confined to infinitives: it has long been noticed, for example, that Latin and the Sabellic languages share a “Perfect” tense, which historically in both instances is due to conflation of the Indo-European perfect and aorist tenses. But the Sabellic and Latin perfect-marking morphological systems are utterly unlike one another, and indeed the total number of shared (Latin-Sabellic) morphological innovations in marking the perfect is exactly zero. We could assume that Proto-Italic had a large number of perfect-marking bound morphemes, granted.

    But how likely is it that this unified Proto-Italic language would have yielded daughter languages (Latin and the Sabellic languages), each of which would have *just happened* to generalize *all* the innovative perfect-marking morphemes which the other language lost?

    A convergence model is to my mind much likelier: the Latin-Sabellic situation, where a perfect shares the same functions, all the while etymologically different morphemes are used (in Latin and in Sabellic) to mark said perfects, is very similar to a case such as the Balkan Sprachbund, where different languages (for example) use suffixed articles, which in the various Balkan languages are etymologically native and not borrowed.

  100. a Hyundai computer […] required a key to turn on

    Not an uncommon feature of all kinds of PCs back in the day, including the IBM PC AT (1984), which was my first Unix machine. I normally left the key in the lock, as I’m sure most users did. It was aftermarket equipment for the IBM PC (1981) and PC XT (1983).

  101. David Marjanović says:

    David: let’s assume Proto-Italic existed and had an infinitive marked by inherited Indo-European -FI. For this ending to be replaced by -SE/RE in Latin would certainly be possible. But without the older -FI ending even leaving a trace? This is less likely.

    Why? Wouldn’t there be strong pressure to regularize this?

    Any ancestor of Umbrian and Iranian pretty much must have been an ancestor of Germanic as well. I, for one, am not aware of a -b or -d suffix with infinitive or any similar meanings in Germanic.

    But the Sabellic and Latin perfect-marking morphological systems are utterly unlike one another, and indeed the total number of shared (Latin-Sabellic) morphological innovations in marking the perfect is exactly zero. We could assume that Proto-Italic had a large number of perfect-marking bound morphemes, granted.

    Or we could assume that Proto-Italic had a third system, and both Latin and Sabellic innovated on it independently.

    What would actually convince me would be a list of exclusive shared innovations of Latin or Sabellic with some non-Italic branch. Celtic appears to be a good place to look, as would be the confusion over whether Venetic was an unspectacular “Q-Italic” language or rather somehow intermediate between Italic and Celtic.

  102. There do exist a few clear morphological innovations that are shared across both putative branches of Italic, including the imperfect indicative, imperfect subjunctive, and gerund: in all of these cases the Latin morpheme has clear cognates in Sabellic. Etienne, do you see these too as due to diffusion?

  103. Henk Metselaar says:

    I’m reading a somewhat older book (1885) on ceramics which is there consistently spelled keramic. I guess it’s a related but opposite move because I can’t imagine anyone actually pronouncing it with /k/ in English, even though that’s the only possible pronunciation in my native Dutch. The spelling in Dutch is keramiek and keltisch, hard c only being allowed before u, o and a (the name Cees is perhaps an exception because it’s short for Cornelis. Anyway, all good rules need their exception).

  104. The OED1 (last modified 1889) lists keramic as an alternative spelling, with two quotations: one from 1850 and one from Mark Twain in 1880 (so very recent at the time). There’s no corresponding pronunciation given, but surely it was with /k/ and directly borrowed from Greek.

  105. For Hyundai, I find the common American pronunciation /ˈhʌndeɪ/ quite close to the Korean pronunciation, which is /hjəːndɛ/ in the prescriptive, conservative pronunciation, but /hjʌnde/ for most speakers today. It’s certainly much closer than pronunciations like /hiˈʊndaɪ/ or /ˈhaɪəndaɪ/ and others that you hear. I myself usually say /ˈhjʌndeɪ/ in English, even though I know /hj/ isn’t usual in English in this position.

    As in English, Korean /hj/ is also realized as something like [ç] except usually more as an approximant rather than as a fricative (like Canepari’s [ḩ]).

  106. There’s also leuk(a)emia, which is interesting because it nonetheless uses Latinate (a)e rather than ai, and because there are other medical terms in English with -c(a)emia.

  107. Trond Engen says:

    Since Mycenean is attested in the 15th century BC, Proto-Greek must have been spoken at least as early as 2000 BC. For all practical purposes, that’s Early Helladic Greece. I’ve never really thouhgt of what this might mean for the other branches, but a similar age of Proto-Celtic doesn’t seem unwarranted. That’s well before both La Tène and Hallstatt, around the beginning of the Urnfield culture, or even earlier. I don’t necessarily imagine all arms of Urnfield to be Celtic, but if Oscan-Umbrian were para-Celtic with a Sabinoid substrate, that would surely fit. Much would become clearer with a better understanding of the dynamics behind the emergence and spread of Urnfield.

  108. Trond Engen says:

    (I just had my interest in Urnfield renewed with this article, brought to me by Dmitry Pruss on Facebook.)

  109. Recognizing that languages are not bones or DNA or pottery, it nevertheless strikes me that, given Ireland’s insular and peripheral location, if (1) modern Irish are descendants of a population group that entered Ireland around the middle of the 3rd millenium BCE, as the recent evidence seems to imply, (2) there is no archeological record or other evidence of another group entering Ireland before the first or second millenium CE, and (3) Celtic is a true PNIE clade, then the population that entered Ireland in the mid-third millenium BCE must have already been speaking either Common Celtic or a daughter language. In other words, Common Celtic was already differentiated from PNIE at least as early as around 2500 BCE and perhaps even earlier. Does that seem too early?

    I guess it’s possible that a few subsequent arrivals from Britain could have spread a Celtic language throughout Ireland, but doesn’t that seem unlikely?

    I suppose we’ll never know for sure, but it’s intriguing to speculate.

  110. David: Of course Latin and Sabellic can each be assumed to have modified the morphology of the perfect (and other things) as they differentiated from Proto-Italic. But in such a fashion that ALL the innovations which must have separated Proto-Italic from Indo-European were entirely eliminated? That strikes me as extremely unlikely: Modern Romance or Modern Germanic languages have certainly innovated when you compare them to Latin or Proto-Germanic, but major innovations which differentiated the latter two languages from Indo-European remain visible in their daughter languages: All Romance varieties (minus pidgins and creoles) preserve a productive reflex of the Latin imperfect indicative, for instance, which is an innovation vis-vis Indo-European; all Germanic varieties (again, minus pidgins and creoles) preserve, in forming the past tense and the past participle, the Proto-Germanic distinction between strong and weak verbs. The list of shared innovations found in all daughter languages of Latin and Germanic is a long one: whereas the list of shared Proto-Italic innovations found in both Latin and Sabellic is very, very short.

    TR: You point to the indicative imperfect, subjunctive imperfect and gerund as being shared Proto-Italic innovations found in Latin and Sabellic. But at least one scholar claims that the Sabellic gerund was borrowed from Latin; as for the imperfect, it consists in Latin and Sabellic alike of an innovative form whose origin is suffixation of the (inherited) copula in both languages, and this strikes me as being as liable to be a case of areal convergence as (say) the spread of “have” perfects across Western Europe. As for the shared *-se- morpheme as a formant of the imperfect subjunctive, it is readily segmentable and thus could have been borrowed one way or the other (before rhotacism in Latin came into effect, naturally).

  111. Trond Engen says:

    2500 BC means Beaker. doesn’t it? It’s conceivably IE, but we’d need to explain a massive celticity extending from Ireland to Iberia and Bohemia at the advent of the Roman Empire, two and a half millenium later. I think that means a closeknit coastal Celtic area and massive back-migration to the Donau region. It’s not impossible, but we ought to see that too.

  112. Etienne, that’s a tenable position and is more or less what I expected you to say, but it does involve at least two cases of bound-morpheme borrowing (gerund/gerundive and imperfect subjunctive). Weiss (Hist. Gramm. Latin, 470) also lists as shared morphological innovations the 1sg pres. of “to be”, and the creation of a new 2pl. middle ending containing -m-. That seems like a lot of borrowing of basic verb morphology.

  113. Bill W.: Since the language of Ireland today is a Germanic one, you could use the same set of facts to show that Proto-Germanic was spoken in Ireland around -2500.

  114. January First-of-May says:

    @John Cowan: as far as I understand Bill W’s argument, the big part is that “there is no archeological record or other evidence of another group entering Ireland before the first or second millenium CE”, and the written record of assorted Old Irish goes back (nearly?) that far. There is (pretty much) no record of Germanic in Ireland prior to the second millenium CE, so your comparison doesn’t really work.
    I think this still isn’t enough, because Picts, but whatever.

  115. David Marjanović says:

    (2) there is no archeological record or other evidence of another group entering Ireland before the first or second millenium CE

    Is the archeological record of that time in Ireland good enough (and well enough understood today) that absence of evidence is evidence of absence?

  116. JFoM:

    My point is that it’s easy to draw doubtful conclusions from limited evidence. If all we knew was that the Irish population has been substantially the same for four or five thousand years, and that they speak a Germanic language today, then it would be reasonable to suppose that they had been speaking a Germanic language since their arrival. The only reason why we can reject this aboriginal-Germanic argument is that we have overwhelming evidence of a Celtic language spoken there quite recently. Since we have no evidence of any pre-Celtic language in Ireland worth mentioning, then it seems reasonable to suppose that there was none. But that is no more than an argumentum ad ignorantiam, which is sometimes the best we can get in science (Jimmy Carter never was a member of the American Nazi party), but is always subject to being shot down by external positive evidence (Proto-Celtic didn’t exist that long ago).

    The oldest Primitive Irish evidence is from about the 4C.

  117. Well, there’s evidence of humans in Ireland from as early as 12,500 BP, so I think it’s pretty certain that there were pre-Celtic languages there.

  118. I meant, of course, as of -2500.

  119. I wasn’t making an argument–I was just raising some questions which I find perplexing.

    We do have a good idea how Germanic languages got to Ireland–we have not just an archaeological, but a historical, record.

    And, honestly, I don’t want to be contentious. I’m just hoping that some of the participants here who are much more knowledgeable than I am–again, I’m not being ironic when I write this–might give me some helpful guidance.

    I recognize that absence of evidence is not evidence of absence, but it just puzzles me how Celtic languages made it into Ireland. Doesn’t the DNA evidence suggest that there was no population movement into Ireland between -2500 and at the earliest late antiquity? Of course there were languages spoken in Ireland before Celtic, but how did Celtic get there?

    I have no Irish ancestors myself–I don’t have an axe to grind here–and, once again, I’m respectful of everyone else’s views.

    Another question: is there any evidence at all that the Hallstadt and La Tene culture had anything to do with the “Celtic” languages? As far as I can tell, as applied to the “Celtic” languages and the populations that spoke these languages and speak them to day, is simply an artifact of 19th century Romantic nationalism. The recent TLS article makes this point emphatically.

    Thanks to all of you who responded to my query!

  120. I wasn’t making an argument–I was just raising some questions which I find perplexing.

    We do have a good idea how Germanic languages got to Ireland–we have not just an archaeological, but a historical, record.

    And, honestly, I don’t want to be contentious. I’m just hoping that some of the participants here who are much more knowledgeable than I am–again, I’m not being ironic when I write this–might give me some helpful guidance.’

    I recognize that absence of evidence is not evidence of absence, but it just puzzles me how Celtic languages made it into Ireland. Doesn’t the DNA evidence suggest that there was no population movement into Ireland between -2500 and at the earliest late antiquity?

    I have no Irish ancestors myself, and, once again, I’m respectful of everyone else’s views.

    Thanks to all of you who responded to my query!

  121. “The only reason why we can reject this aboriginal-Germanic argument is that we have overwhelming evidence of a Celtic language spoken there quite recently. ”

    John, I think we understand the historical processes that planted a Germanic language in Ireland. We don’t have a good idea how a Celtic language got there, but it’s strange that there doesn’t seem to be an archaeological record that might suggest how this occurred other than a population movement that happened circa -2500 or before.

    Again, I recognize that absence of evidence is not evidence of absence, but still, doesn’t it seem strange that there is no evidence?

  122. Doesn’t the DNA evidence suggest that there was no population movement into Ireland between -2500 and at the earliest late antiquity?

    Yes, but it also suggests that there was similarly no population movement in the 19-20C when English was taking over from Irish, and that in fact detectable population movement is not at all required for language replacement, though doubtless the replacement of pre-Celtic by Celtic took longer to complete.

  123. Trond Engen says:

    Bill W.: Another question: is there any evidence at all that the Hallstadt and La Tene culture had anything to do with the “Celtic” languages?

    I realize you meant to remove this question, but it might still be worth answering. The relation to late La Tène is obvious, almost by definition, since it’s the culture of regions that we know were Celtic when the Greeks and the Romans first described them. But how early? The sack of Rome by the Gauls in 390 BC makes it clear that there were Celts in northern Italy, confirming the celticness of the archaeological cultures there. The invasion of the Balkans from the late 4th century BC didn’t come from nowhere, so the upper Danube valley must have been Celtic. Pre-Grimm Celtic loanwords in Germanic, mainly in the realm of politics and warfare, show an early presence in Northern Europe. Taken together all these suppose that La Tène was Celtic from the outset at around 500 BC.

    La Tène is essentially Hallstatt with the addition of accumulations of Mediterranean trading goods and widespread “Celtic” art. It’s reasonable to assume that the Celtic languages were already there in Hallstatt, but I guess this is where Cunliffe, Koch et al. will differ. If, as they say, “Celtic” art came from the west, the increased trade might be the vector of both language and artistic tastes, giving prestige to the Celtic of the western fringe and spreading it all over the Hallstatt world. Also, while La Tène grew gradually out of Hallstatt in the west, I read that there’s evidence of a sudden, maybe violent, collapse in the east. It could be that western Hallstatt was celticized from the west through prestige, and those Celts either caused the collapse of the old Hallstatt core, or moved in to fill the power vacuum after the collapse. In this scenario core Hallstatt may have been “Illyrian”, “Venetic” or even “Italic”.

    Anyway, Hallstatt grew out of Urnfield, probably even out of the core region of Urnfield in Southern Germany. While Hallstatt seems to have been a largely peaceful and prosperous indigenous development, Urnfield was a period of upheaval and consolidation around a new political system with professional armies and fortifications. That certainly would be a good time for languages to gain prestige and spread, and Urnfield makes a decent fit with later Celtic settlement. Even if it’s eastern parts were something else, later celticized. However, the great 1200 BC battle of Tollense makes me think that this was the time when the Germanic tribes of Northern Europe would pick up new political and military terminology. And it did so from Celtic, not Venedic.

  124. Trond Engen says:

    “its”. Damn.

  125. January First-of-May says:

    Never heard of the battle of Tollense before – fascinating story!

  126. Trond Engen says:

    Me too. I’m probably going over board in my interpretations, but for me this has meant a whole new way to look at the beginning of the Urnfield culture.

  127. Actually, I was trying to add the question, not remove it, but I did something wrong.

    Anyway, the question I was posing was whether there’s any reason to link the central European “Celtic” archaeological assemblages to the so-called “Celtic: languages spoken in the British Isles, or whether alternatively, “Celtic” as applied to the languages in question is simply an artifact of 19th century historical philology. Why are Irish, Welsh, Manx, Cornish, etc., called “Celtic” languages?

    “Celtic peoples” or “Celtic nations” are constructs of 19th century romantic nationalism, aren’t they?

  128. Trond Engen says:

    Oh, OK: Yes, probably, at least in the sense that the term ‘Celtic’ is historizing. And the exotism of the glorious Celtic past was very much a Romantic construct, since the myths and arts attributed to it was just as much Common European motives from the Medieval Era or the Renaissance surviving in the outback. But the linguistic relationship between the modern Celtic languages and those Keltoi mentioned in the first sources is not in doubt. Even if not all dots are connected yet.

  129. “Celtic” as applied to the languages in question is simply an artifact of 19th century historical philology. Why are Irish, Welsh, Manx, Cornish, etc., called “Celtic” languages?

    What Trond said. All the Druidic cultural stuff is bullshit, but the linguistic relationship is rock-solid; Celtic languages are as obvious a group as Germanic. See Wikipedia for a list of common elements.

  130. David Marjanović says:

    we have no evidence of any pre-Celtic language in Ireland worth mentioning

    Some think there’s a big fat substrate in Irish and in the place names of Ireland. While it doesn’t seem to be discussed much, I haven’t encountered attempts to debunk it.

  131. Irish substrate, how?

  132. Eli Nelson says:

    @Languagehat:

    A lot of the common features listed in that Wikipedia article seem like they could be due to areal convergence rather than shared ancestry. For example, I thought I’d read the morphological consonant mutation systems in different languages have different effects and occur in different environments, and so are not inherited from a common ancestor. But there seem to be different views on this; I found this article that I am currently going through: https://www.uni-due.de/~lan300/13_Initial_Mutation_in_Celtic_(Hickey).pdf

    Similarities in the overall structure of the gender and verbal systems could also be areal. The most convincing evidence would lie in the shared sound changes and morphological forms. Based on Etienne’s earlier comments in this thread, I assume evidence like this exists, but I’m not sure what it consists of, aside from the loss of /p/ that he mentioned. From the Wikipedia page on Proto-Celtic, it looks like some other sound changes that might be distinctive for Celtic are the pattern of epenthetic vowel insertion for syllabic resonants, and the change of plosives to *x before other plosives or *s. But there doesn’t seem to be much distinctive morphological change between PIE and Proto-Celtic.

  133. Eli Nelson…No Celtic scholar claims that the initial mutations go back to Proto-Celtic, as these only arose after the fall of word-final syllables. Among other phonological innovations which Celtic scholars broadly agree go back to Proto-Celtic there is the merger of voiced aspirate and non-aspirate stops, the shift of *gw to *b, the merger of PIE long /e/ and long /i/ as /i/, the merger of PIE long /o/ and long /a/ as /a/, except word-finally, where long PIE /o/ is raised and merges with long /u/ instead. Except for the merger of long /o/ and long /a/ none of these changes is shared with Germanic or Italic, and all are pan-celtic.

    The morphological innovations are harder to identify, but that is because the better-attested Celtic languages, Old Irish and Middle Welsh, had lost their final syllables before the appearance of most written attestations and thus both had prehistorically lost much morphology and reshaped the remaining morphology a huge deal, making it difficult to reconstruct what Proto-Celtic morphology was like. But there are some striking shared innovations, such as the loss of a cognate of Latin EGO as the nominative of the first person singular pronoun and its replacement by a form */mē/. Again, quite alien to its Germanic or Italic neighbors.

  134. Eli Nelson says:

    @Etienne: Thanks, those are exactly the kind of things I wanted to know. I wasn’t sure about the initial mutations because the paper by Hickey that I posted a link to says at one point “Positing lenition at an early period helps to account for the appearance of mutations in all the Celtic languages as a common development and not a shared innovation. The latter view is difficult to sustain seeing how typologically unusual morphological mutation is and that it is present in all attested Celtic languages.” However, further on Hickey writes that “What is probably the case is that the weakening of consonants and the presence of external sandhi was part of the phonetic makeup of the Celtic languages in their common stage but that this was only phonologized after their split,” so he does not consider grammaticalized mutation to be a common inherited feature.

  135. Trond Engen says:

    Areal features… It occured to me that we discussed initial syllables in Celtic and Basque some time ago. I think it must be That Strange Muffled Utterance of this very date in 2013, starting from this comment by Etienne.

  136. Trond Engen says:
  137. That’s a fascinating story. (For those who, like me, need to know such things: Tollense is stressed on the second syllable: [tɔˈlɛnzə].)

  138. Trond Engen says:

    Yes, oddly enough. From the Slavic substrate, I suppose.

    I see that the finds are being catalogued under ‘Weltzin’ by the archaeologists, for the nearest village, so maybe it’s better named the Batlle of Weltzin.

    Also this from Die Welt (in German).

  139. “the linguistic relationship between the modern Celtic languages and those Keltoi mentioned in the first sources is not in doubt. Even if not all dots are connected yet.”

    That’s exactly my question. I’m persuaded that Common “Celtic” was a real language at some point in prehistory. But what evidence connects the Keltoi in eastern Europe with the attested “Celtic” languages? Why do we call these languages “Celtic”

    Once again, this is idle curiosity, not an attempt to debunk the connection.

    And where better can you satisfy your idle curiosity about ancient languages than on Language Hat?

  140. Well, Caesar, Livy, and Pliny applied the word Celtae to the inhabitants of Gaul (specifically Middle Gaul) and Spain, and it seems unlikely they would have done so if those peoples were fundamentally different from the Κελτοί of Herodotus and Polybius.

  141. David Marjanović says:

    Similarities in the overall structure of the gender and verbal systems could also be areal.

    Literally anything could be areal. In the absence of evidence that it was areal, it should be taken as genetic until further notice.

    (The analogous insight was the birth of phylogenetics as a science in biology.)

  142. I don’t know about that. Some features diffuse more easily than others. It’s sensible to argue that in the case of rarely-borrowed features like bound morphemes, genetic affiliation should be the null hypothesis — see my comments about Italic above — but I don’t see why that should be true in all cases.

  143. Eli Nelson says:

    @David Marjanović:

    Some kinds of things are more likely to be areal than others, though, and therefore weaker evidence for a genetic relationship. For example, the Wikipedia article lists “two grammatical genders” as a “family resemblance” between most of the Celtic languages. Even though this is currently true about the group, it doesn’t seem to be a feature inherited from Proto-Celtic, which had three genders. At most, the descendants of Proto-Celtic could have inherited features that predisposed their gender systems to develop towards a two-way contrast. Many of the other examples looked like this to me (for example, the “interplay between the subjunctive, future, imperfect and habitual” seems to mainly consist of secondary developments, more or less parallel in the different languages, as described in James Fife’s introduction to The Celtic Languages, edited by Martin Ball and Nicole Muller). However, it’s true this section of the Wikipedia article also lists some inherited morphological similarities, such as the impersonal form, for which the suffix -r seems to be shared across the Celtic languages.

  144. David Marjanović says:

    and therefore weaker evidence for a genetic relationship

    Yes, absolutely – but this shouldn’t be exaggerated to “weak evidence for something is proof against it”, as has often been done. Keep the strength of your belief proportionate to the strength of the evidence.

    the impersonal form, for which the suffix -r seems to be shared across the Celtic languages

    That, though, is usually considered shared with Italic (the Latin passive in -r), and often even extrapolated straight down to PIE because “mediopassive” forms in -r are also found in Tocharian and Anatolian. Ringe goes so far as to use this one feature as insurmountable proof that all other IE subbranches form a single branch which replaced this -r by -i!

  145. Eli Nelson: While it is true that Modern Celtic languages all have a two-gender system (masculine/feminine), I see no reason to even assume that there was anything about Proto-Celtic which predisposed it to lose the neuter.

    First of all, loss of the neuter (chiefly through merger with the masculine) and reduction of the gender system to a masculine/feminine opposition is widespread in Indo-European outside of Celtic: Romance, Baltic, and a number of Indo-Aryan languages (Hindi, Punjabi, Romani) have undergone the same change: hence it is certainly possible that Proto-Celtic had a neuter gender which was as well-established as the Latin or the Sanskrit neuter, and that Brythonic and Goidelic each lost the neuter as part of a common Indo-European “drift” which Romance and Hindi/Punjabi/Romani shared.

    Second of all, comparative and philological data show that the loss of the neuter in Brythonic and Goidelic took place at different historical periods: On the one hand Breton was transplanted from Britain to the Continent in the fifth century AD, and like its sisters in Britain (Cornish and Welsh) has a two-gender system: in none of the three languages do we find any real trace of the neuter, even in the oldest written records. This points to the Brythonic loss of the neuter having taken place at the latest in late Roman Imperial times. On the other hand the first extensive records of Old Irish date back to the ninth-eleventh centuries AD, and in this language (“Classical Old Irish”) the neuter is alive and kicking: its disappearance took place later, over the past thousand years.

    Third of all, Brythonic and Goidelic nominal morphology, during the relevant historical period, otherwise have little in common: Brythonic had lost noun declension and the dual number, seemingly in the days of Roman Britain, whereas Old Irish had preserved its inherited declension system incredibly well (including the dual) despite its loss of final syllables (as one linguist pointed out, taking mutation into consideration, Old Irish FER had as many distinctive forms as the Latin cognate VIR): tellingly, no Modern Gaelic variety (leaving aside instances of obsolescent varieties) has entirely lost declension and reduced its nominal morphology to a mere singular/plural opposition, the way Proto-Brythonic had. Because of this strong dissimilarity I am frankly skeptical of the claim made by one Celtic scholar that the loss of the neuter in Old Irish/Gaelic (which we can follow across centuries of written records) can teach us a great deal about the loss of the neuter in Brythonic.

    David: I agree with TR and Eli Nelson, some things are far more likely to be diffused than others. Language contact yields cases of shared structures far more readily than cases of borrowed grammatical morphemes: the “have”-perfects of Western Europe or the suffixed definite articles of the Balkans are good examples of this.

    And “Italic” is to my eyes striking in its paucity of shared bound morphemes which can be reconstructed back to “Proto-Italic” but not to Proto-Indo-European, as opposed to the shared grammatical structure. The shared innovation of a perfect created through merger of the aorist and perfect of Indo-European, but which lacks any undeniably Italic innovation in perfect-marking morphology, is an excellent example of this.

    Admittedly, sometimes there can be instances where through mutual contact languages each end up recruiting phonologically similar native morphemes to play similar functions. Thus, while English “to have” goes back to a Proto-Germanic */habjan/ and French “avoir” goes back to Latin HABERE, my students are always surprised when I tell them the two verbs do not go back to a common Indo-European verb. Yet the fact that Germanic recruited, to express possession as well as the past tense, a verb so similar in surface phonological form to the Latin/Early Romance well is something I strongly suspect is due to language contact.

    And I suspect, TR, that the same is true of the “Italic” second-person plural medio-passive ending: if all the Latin and Italic endings have in common is an element /m/ I suspect no common ending can be reconstructed back to Proto-Italic. Now, granted, we could assume some kind of unusual phonological or morphological change, but considering that the other person-marking suffixes seem to correspond quite regularly, it does seem AD HOC to do so.

    But if, instead, we assume that there never existed a Proto-Italic and that Sabellic and Latin are separate branches of Indo-European whose shared “Italic” features are due to contact, well, then a similar-looking “Italic” bound morpheme (second person plural medio-passive with an /m/, for instance) which cannot be reconstructed back to a “Proto-Italic” language is exactly the sort of thing which we would expect to find and which thus would not require any sort of AD HOC explanation.

  146. David Marjanović says:

    David: I agree with TR and Eli Nelson, some things are far more likely to be diffused than others. Language contact yields cases of shared structures far more readily than cases of borrowed grammatical morphemes: the “have”-perfects of Western Europe or the suffixed definite articles of the Balkans are good examples of this.

    Again, I’m not going to deny any of this. All I’m saying is that if we have weak evidence for a hypothesis and none against it, we still have to come down for it – however tentatively.

  147. Eli Nelson says:

    My point in saying the evidence in the Wikipedia article seemed weak was not to dispute the genetic relationship between the Celtic languages. I just wanted to learn what stronger evidence existed for that relationship. Etienne obliged by posting about the sound changes and striking inherited morphological innovations that are common to the Celtic languages.

  148. David: well, in this instance, there are two competing hypotheses: either the Latin-Sabellic shared similarities (when compared to other Indo-European languages) are due to both sharing a common ancestor, Proto-Italic, which is intermediate between them and Proto-Indo-European, or to their having influenced one another heavily, without it being necessary to postulate a “Proto-Italic” language. I believe the evidence strongly favors the second of the two possibilities, because the similarities between Latin and Sabellic languages involve shared structures with few if any shared (especially bound) morphemes (leaving aside, obviously, those which must go back to Proto-Indo-European). That is to say, similarities of a nature which elsewhere is indubitably due to convergence rather than common inheritance.

    I’ve explained why the perfect in Sabellic and Latin strikes me as a likelier case of convergence than of inheritance from a “Proto-Italic”: can you point to any shared Latin-Sabellic feature which seems to you more economically explicable through inheritance (from a language other than Indo-European, obviously) rather than convergence?

  149. What about the imperfect subjunctive in *-sē-? There’s no consensus as to its origin, but it’s shared between Latin and Sabellic and does not occur in other branches.

  150. TR: A fair question. Well, /sē/ is a readily segmentable morpheme and therefore could have been borrowed easily, and thus strikes me as equally likely to be a “Proto-Italic” feature as an areal feature. The same is true of the shared gerund, which as I wrote above one authority in fact sees as Latin in origin and borrowed into Sabellic.

    Contrast this with the Latin and Sabellic perfects: both are the result of a fusion of the Indo-European aorist and perfect, both Latin and Sabellic have several perfect-marking bound morphemes, and the overlap between Latin and Sabellic perfect-marking morphology is close to zero: there is no possibility of reconstructing a “Proto-Italic” morphology of the perfect. Consider the overlap between the bound morphemes marking the infinitive in Sabellic and those marking the infinitive in Latin: also zero. Consider the overlap between Latin and Sabellic in terms of shared innovations alien to the rest of Indo-European in nominal/pronominal declension: also zero. Consider the contrast between hic/haec/hoc in Latin and the Sabellic demonstrative stem *e(s)k(s)o- , which obviously cannot go back to a common “Proto-Italic” form. The same is true of the “shared” second person plural medio-passive ending with /m/ in Sabellic and Latin, which likewise seemingly cannot go back to a common “Proto-Italic” form.

    To my mind, in short, the weight of the evidence points to convergence between two originally separate branches of Indo-european. Now, it would be nice if this could be proved beyond any reasonable doubt, but as David wrote (quite correctly) above, the strength of the belief must correspond to the weight of the evidence.

  151. Etienne, that’s all very cogent, and I’m certainly less convinced of the unity of Italic now than I was coming into this thread. Still, I’m not so sure that *-sē- “could have been borrowed easily”. Borrowing of inflectional morphology does happen, of course, but it’s quite rare, and though it is aided (as you say) by phonological transparency, it is also impeded by semantic/functional opacity, which is the case of the imperfect subjunctive with its broad and contextually dependent array of meanings (in Latin anyway; there’s less evidence for its use in Sabellic).

    There are also apparent phonological shared innovations, and although these too could be cases of convergence and many are rather trivial, a few are unusual, such as the change of *dh > f.

    I suppose a problem for the Proto-Italic theory is to explain why, if Proto-Italic existed and was of comparable age to, say, Proto-Greek, the Italic languages are so much more different from each other than the Greek dialects.

  152. Trond Engen says:

    The unity of Greek in spite of dialect differences is really striking. At the time of Classical Greek the dialects must have diverged for at least as long as Slavic or Romance languages today. One reason for the difference may be that Greek dialects never stopped interacting, while Italic dialects came moving into Italy at different times, meanwhile having diverged beyond simple adjustment to new surface patterns.

  153. Trond Engen says:

    Bill W.: I think the linguistic evidence for the celticity of Eastern Celts consists of toponyms, recorded personal names, and apparently a few loanwords in Greek and Gothic. Non-linguistic evidence is correlation in time and space with La Tène type archaeology.

    I wonder why there aren’t (or if there are) Celtic loans in Albanian. Could be important.

  154. Albanian is all full of loanwords at every level: there are three strata of Latin loanwords alone. If there are Celtic loanwords, they are probably lost or buried. Nobody has a decently defensible theory about the place of Albanian in the IE tree, as we don’t know enough and we never will.

  155. Trond Engen says:

    Good point about loanwords being buried under layers of new loans. My idea was that if Albanian, as many believe, spent considerable time in more central parts of Balkan, the language would have been under direct Celtic influence, but you’re probably right that it would be hard to recover. Or I might devise a crackpot theory of Albanian as the lost Eastern branch of Celtic.

  156. Or I might devise a crackpot theory of Albanian as the lost Eastern branch of Celtic.

    Do it! Today’s the day!

  157. Trond Engen says:

    The Western Albion!

    Now I’m thinking that it might be worth considering for real since it would solve a lot of problems.

  158. Trond Engen says:

    Not a lot. But at least one or two. And that would simplify some others.

  159. Don’t sell yourself short. You’re going to revolutionize linguistics. And history.

  160. Trond Engen says:

    I’ll have to get on to it,, then. There’s only a few hours left before the time of the revolution is overcome.

  161. J.W. Brewer says:

    A fair amount of funding and weaponry for Kosovo independence came from the ethnic-Albanian diaspora in North America especially in the NYC area. I think they (plus the North American Irish-nationalist diaspora) might be sufficiently interested in this revolutionary new theory that Trond should feel like he has until midnight N.Y. time to finish up the details.

  162. gwenllian says:

    Speaking of the Irish and Kosovo, I remember a poster on some Irish forums dedicating a whole lot of time to proving the close connections between Irish and Serbian.

  163. “…a verb so similar in surface phonological form to the Latin/Early Romance well is something I strongly suspect is due to language contact.”

    Etienne, has that idea (habian nudging its semantics to match habēre) been published anywhere?

    It’s not such a rare mechanism, even intra-linguistically, but I haven’t seen it systematically discussed or even acknowledged anywhere (my favorite is English speakers interpreting puce as a shade of green, after puke.) For language contact, Zuckerman has written about Modern Hebrew repurposing old words to match the meaning of European words it resembles in sound, which is not all that different from the habēre story.

  164. gwenlllian says:

    I do know a woman named “Ceri” pronounced “Keri”. I’ve known her since 1978, so it’s no innovation.

    The woman named “Ceri” is English, not American.

    It’s a very common Welsh name. I was about to call it a unisex name, but behindthename says they’re actually two names – the feminine short for Ceridwen (“possibly from Welsh cyrrid “bent” or cerdd “poetry” combined with ven “woman” or gwen “white, fair, blessed”) and the masculine possibly derived from Welsh caru meaning “to love”.

    It’s unlikely to be the case in the UK, but there are probably quite a few Ceris out there whose parents didn’t have the Welsh name in mind at all. Parents fairly often give their child an established name purely by accident while trying to innovate. I recently read one of those smug lists making fun of “bogan baby names”, and it mentioned a Macsen whose parents couldn’t decide between Mackenzie and Jacksen.

  165. Some time ago I read an article about prohibited baby names in New Zealand. The list includes Justus – presumably encountered as a respelling of Justice, itself noted to be the most common disapproved name – in seeming unawareness of the fact that it was an established Latin given name in medieval Europe.

  166. Trond Engen says:

    Well into Mid-Atlantic time now, and the best thing I’ve been able to come up with is Tosk and Gegh being derived from *teut- and *gal- , and Tirana, Dürres, and Shkodër from *tirjon- “lands”, *duro:- “hard, and *skeito- “shield”, respectively.

  167. Works for me! Send it off to Speculative Grammarian, and tell Trey languagehat sent you.

  168. David Marjanović says:

    I’ve explained why the perfect in Sabellic and Latin strikes me as a likelier case of convergence than of inheritance from a “Proto-Italic”: can you point to any shared Latin-Sabellic feature which seems to you more economically explicable through inheritance (from a language other than Indo-European, obviously) rather than convergence?

    I don’t know comparative morphology well enough to present any examples from that field. But all the phonological stuff is more economically – that’s a low bar! – explained through inheritance. What comes to mind immediately is the system of changes of which the abovementioned *dʰ > f is a part: first, the aspirates are devoiced (like in Greek); then they become fricatives (like in Greek a thousand years later); then *sr merges into *θr; then and *xʷ merge into *f… it’s simpler to assume that this whole system happened on the way to a unitary Proto-Italic than to assume that each and every one of these individual changes spread through a Sprachbund and ran to completion before the next change came along.

    Notably, all the possibly neighboring IE branches (Celtic, Germanic, Albanian/Thracian/Dacian/Illyrian) underwent sound changes that would have interacted with these, but I’m not aware (for what that’s worth) of a supposedly Italic language that shows evidence for such phenomena.

  169. January First-of-May says:

    Parents fairly often give their child an established name purely by accident while trying to innovate.

    The name Гертруда (Gertruda), a well established name of, IIRC, German origin, was (or, at least, is said to have been) fairly popular in 1930s Soviet Union, because it also happened to be the obvious abbreviation for Герой Труда.

  170. Ha! I never get tired of those early Soviet names.

  171. I’ve been surprised to learn that there are even Vilenas and Oktyabrinas born after 1991.

  172. People are still being named after their grandparents, I suppose.

  173. Etienne says:

    David: actually, that whole set of changes is more readily accounted for by assuming diffusion than common inheritance.

    Let’s focus upon the *bʰ > f and *dʰ > f change: in Latin and Sabellic languages alike it affects word-initial *bʰ and *dʰ. But whereas the exact same change is also found word-internally in Sabellic, in Latin the word-internal reflexes are /b/ and /d/ respectively (in some word-internal environments *dʰ yields /b/ in Latin, a complication which for the purposes of the present discussion we may ignore).

    If you assume that “Proto-Italic” originally turned all its voiced aspirate stops into voiceless fricatives (but without the reflexes of *bʰ and *dʰ merging, at least word-internally, as they remain distinct in Latin), then you need to postulate a separate set of changes to turn them back into simple voiced stops word-internally in Latin. Plus another set of mergers in Sabellic.

    If, however, you assume instead that Proto-Italic had turned its aspirate stops into voiceless fricatives word-initially only, and preserved its aspirates word-internally, then you can indeed obtain the Latin reflexes by postulating simple de-aspiration: but to get the Sabellic forms you need to postulate a specifically Sabellic word-internal sound change whereby THE VERY SAME CHANGES which had taken place word-initially in “Proto-Italic” (*bʰ > f and *dʰ > f) also took place later in Sabellic. For the former change (*bʰ > f) that might be credible (it happened once in the history of Greek, so for it to have happened twice in the history of Sabellic is not impossible), but the latter change (*dʰ > f) is too idiosyncratic for this two-stage sound change hypothesis to work.

    It seems to me that a simpler assumption would be to assume that Sabellic and Latin were originally separate branches of Indo-European: the former preserved aspirate stops everywhere, and the latter preserved them word-initially only, de-aspirating them word-internally. Then, somewhere, the sound change in question (*bʰ > f and *dʰ > f) began and spread throughout Sabellic and Latin alike, much in the same fashion that uvular R spread throughout Western Europe.

  174. Etienne, isn’t the Sabellic lexicon closer to that of “Italic” (i.e. Latin) than to that of any other IE branch?

  175. Trond Engen says:

    Uvular R is a very good example of phonological diffusion.

    But I don’t think I understand what you mean here. Do you say it’s *dʰ > f with no intermediate stage(s)? That indeed looks idiosyncratic and, frankly, unbelievable, if it’s also a merger. I’ll take the thorny road…

    Latin:
    *bʰ…bʰ… > b…bʰ… > b…f…
    *dʰ…dʰ… > d…dʰ… (> d…þ…) > d…f…

    Sabellic:
    *bʰ…bʰ… > > f…f…
    *dʰ…dʰ… > > (þ…þ…) > f…f…

    That aside, I don’t see why these sequences force us to tear up Italic.The first change (in Latin) could be dialectal and the second universal. But it probably does take migration or dialect replacement for intermediate forms to disappear. Is the point that at this early stage, the initial divergence makes them separate branches by definition?

  176. Trond Engen says:

    What I mean is, whether or not Italic is a sprachbund or a true family with some archaic dialectal retentions may indeed hinge on e.g. shared lexicon. With the proto-languages as far back as 2000 BC or so we approach a stage of IE where the difference between a branch and a set of related dialects may be hard to define. Which, I suppose, is close to Andrew Garrett’s position.

  177. Trond Engen says:

    … while yours is that since being a language family takes a common origin (from an at-least-in-principle reconstructible proto-language) not shared by any language outside the family, any group of languages, however closely connected they were over however long time, and however many innovations they share, is a sprachbund by definition if there’s a divergence from the outset.

  178. Trond Engen says:

    … or maybe that’s mine, put in your mouth.

    Anyway, I have no problem seeing from your examples that there was considerable initial variation in Italic, apparently more than in Greek (although the dialectal soundlaws in Greek seem old too), but since deep archaisms may be retained within a dialect continuum, the better and more accurately you reconstruct a family, the more likely it is that you’ll eventually have to dismiss it because of some initial variation. So there has to be some additional bar that Italic fails and the others pass.

  179. Trond Engen says:

    One last thing before I hit the pillow. If I seem to be arguing in all directions, it’s because I’m arguing in all directions. Or trying to explore the argument in all directions. Models are handy, but it’s where they break down that things get interesting. And Italic is getting interesting.

  180. Eli Nelson says:

    Intermediate steps are generally postulated in the transition *bʰ > f and *dʰ > f, but it’s not necessary to derive the voiced intervocalic reflexes (such as Latin b and d, and Sabellic f, which was apparently voiced [v] or [β] in this position) from voiceless aspirate plosives (which first became voiceless fricatives everywhere, and were then subject to intervocalic voicing). That’s one scenario, but another view is that the voiced aspirate plosives developed to voiced fricatives first in all positions (e.g. *bʰ > β and *dʰ > ð), and then these fricatives devoiced word-initially. A third view proposed by Jane Stuart-Smith in Phonetics and Phonology: Sound Change in Italic (2004) is that in medial position, PIE voiced aspirate plosives developed phonetically in Proto-Italic to voiced fricatives/approximants (with no intermediate voiceless stages), while in initial position, they first devoiced to voiceless aspirate plosives, and then these voiceless aspirate plosives developed to voiceless fricatives.

  181. Eli Nelson says:

    @Etienne:

    If you assume that “Proto-Italic” originally turned all its voiced aspirate stops into voiceless fricatives (but without the reflexes of *bʰ and *dʰ merging, at least word-internally, as they remain distinct in Latin), then you need to postulate a separate set of changes to turn them back into simple voiced stops word-internally in Latin. Plus another set of mergers in Sabellic.

    If, however, you assume instead that Proto-Italic had turned its aspirate stops into voiceless fricatives word-initially only, and preserved its aspirates word-internally, then you can indeed obtain the Latin reflexes by postulating simple de-aspiration

    The development you describe in the first paragraph doesn’t seem very unusual to me, even though it is somewhat circuitous. German is a familiar example of a language where historically we have the changes *f > [v] > [b] and *θ > [ð] > [d]. And the cases where intervocalic *dʰ has a Latin reflex of b do suggest a possible fricative stage in the history of Latin, since [ð] is generally more susceptible to fronting to a labial place of articulation than /dʰ/ or /d/ are. The cluster du seems to have had a plosive rather than a fricative when it developed to b, so it’s not like a change of [d] > [b] requires a fricative intermediate step, but weak evidence is still evidence of some sort.

  182. [Latin] preserved [aspirates] word-initially only, de-aspirating them word-internally

    There has to be the additional complication that it did not deaspirate the velars (plain or palatalized), since these always give h, rather than g as would be expected by this account.

    Another difficulty is the fate of *zdʰ, which seems to have given Latin st. This is plausibly accounted for under the aspirate-devoicing theory: *zdʰ > *ztʰ > *stʰ > st. The deaspiration theory should predict that *zdʰ would fall together with *zd, which is not the case. That said, there are only a handful of etymologies involving *zdʰ, and they could conceivably be wrong, I suppose.

  183. From the book Eli linked to (p. 151) it looks like the *zdʰ > st etymologies have indeed been contested, and there’s another set of controversial etymologies in which *zdʰ does appear to fall together with *zd… It’s all far from clear.

  184. And bizarre sound changes do happen. A few of Blust’s examples have been discredited, and the rest can be accounted for by natural, but completely unattested, chain shifts, but why force ourselves to believe in such chains merely to save phonetic naturalness as an absolute? Surely it’s enough that it’s a statistical universal.

  185. David Marjanović says:

    Let’s focus upon the *bʰ > f and *dʰ > f change: in Latin and Sabellic languages alike it affects word-initial *bʰ and *dʰ. But whereas the exact same change is also found word-internally in Sabellic, in Latin the word-internal reflexes are /b/ and /d/ respectively (in some word-internal environments *dʰ yields /b/ in Latin, a complication which for the purposes of the present discussion we may ignore).

    If you assume that “Proto-Italic” originally turned all its voiced aspirate stops into voiceless fricatives (but without the reflexes of *bʰ and *dʰ merging, at least word-internally, as they remain distinct in Latin), then you need to postulate a separate set of changes to turn them back into simple voiced stops word-internally in Latin. Plus another set of mergers in Sabellic.

    I thought the textbook explanation was as follows: voiced aspirates > voiceless aspirates > voiceless fricatives in all positions on the way to Proto-Italic; then* intervocalic voicing of all fricatives, notably including *s; then, on the way to Latin**, the voiced fricatives merge into the voiced plosives, except for [z] which merges into r.

    I’ll have to read the book, though!

    * Or perhaps already in Proto-Italic, given that the voicing was allophonic, so that the lack of a special way to write [v] in the P-Italic alphabets doesn’t necessarily mean the sound was absent.
    ** I have no idea of Faliscan.

    German is a familiar example of a language where historically we have the changes *f > [v] > [b] and *θ > [ð] > [d].

    Not quite, no. I can’t think of an example where *f became b. The change from *b, which was [β] and/or [v] in most positions, to the voiceless (!) plosive b did happen, but it didn’t touch *f; *b and *f are distinguished about as well in High German as in Gothic.

    did become the voiceless plosive d; this sound change spread far beyond High German, all the way to the sea – where, unlike in HG, it manifested as a merger with the existing *d. I’m not sure how that happened. There are Early OHG documents that feature the spellings th and dh, and I suppose* the latter may indicate intervocalic and perhaps initial voicing… but turned into d in all positions, and this d is voiceless. Perhaps lenition happened, but didn’t go all the way to voicing, and the still voiceless fricative then became a voiceless plosive?

    * I’m not aware of any literature on this topic. This doesn’t mean there isn’t any, but it would still be surprising if such literature had never been cited in the works I have read.

    The Proto-Germanic and Proto-Northwest-Germanic *d, which was [ð] in most positions, became the voiced plosive [d] in Proto-West-Germanic, which went on to become the fortis voiceless plosive t in HG; it wasn’t merely devoiced. One would think that that’s a push shift, pushed by > d, but apparently that’s not the case; in OHG before the latter shift, there is spelling variation k/c~g for expected g and p~b for expected b, but a t is a t is a t. It is quite possible that all PWGmc. voiced plosives became fortes, and then most of the allophony was leveled out… I’ll stop here, this is getting long. 🙂

    For the sake of completeness, though, I’ll mention a fascinating little complication: Bahder’s law, which apparently turned P(NW)Gmc. voiced fricatives into voiceless ones before *m *n *l *r somewhere in the ancestry of High German. This sound change has approximately never been mentioned according to Google, probably because the examples are so few and present additional complications with interdialectal borrowing or something, but some of the examples are striking, like Schaufel “shovel” – note that this f is short (as if from PGmc. *f), not long (which would be from PGmc. *p), even in accents like mine that continue to distinguish long and short fricatives in such positions. If Guus Kroonen is right and Boden “bottom” belongs (from a paradigm with nom. sg. *budmēn, gen. sg. *buttaz…), then this change must have happened before the West Germanic change of *[ð] to *[d]… and that would mean this latter change must have spread areally and had not yet happened in the last common ancestor of all known West Germanic languages.

  186. Trond Engen says:

    Me:

    Latin:
    *bʰ…bʰ… > b…bʰ… > b…f…
    *dʰ…dʰ… > d…dʰ… (> d…þ…) > d…f…

    Sabellic:
    *bʰ…bʰ… > > f…f…
    *dʰ…dʰ… (> þ…þ…) > f…f…

    Late night horror show! I managed to turn that inside out. Anyway, my question is the same: Does the first step (the standard sequencing of which David laid out in more detail) make it a split by definition?

  187. Etienne says:

    Thanks for your comments, everyone!

    What most strikes me, though, is this: you need to assume more changes if you derive Latin and Sabellic from a “Proto-Italic” than if you assume the shared changes are areal in nature and that Latin and Sabellic were originally two separate branches of Indo-European.

    Word-internally, Latin turns Indo-European *bʰ into /b/ everywhere: postulating simple de-aspiration is easier than postulating a shift to a (voiceless?) fricative which turned back everywhere into a plosive. Indo-European *dʰ turns into /d/ word-internally too, and even in the minority of cases where it becomes /b/ instead an intermediate fricative stage needn’t be postulated: *dʰ becomes /b/ when it follows /u/, for example, and in this instance we could simply assume that the reflex of *dʰ was allophonically labialized before becoming /b/. For Indo-European *gwʰ, its Latin word-internal reflexes (/w/, except after /n/, where it is /gw/) are exactly the same as those of Indo-European *gw. Again, postulating de-aspiration seems simpler. In fairness, Proto-Indo-European *gʰ is weakened to /h/ in Latin in all positions.

    Nevertheless, in a majority of cases simple de-aspiration explains most word-internal reflexes of Indo-European voiced aspirates in Latin. For Sabellic, if you assume that no such de-aspiration took place, then word-initial and word-internal reflexes of Proto-Indo-European voiced aspirates alike can be accounted for with a single set of sound changes (*bʰ and *dʰ become /f/, *gʰ becomes /h/), which also accounts for the Latin forms…AFTER Latin had undergone some changes (de-aspiration) alien to Sabellic.

    David pointed out upthread that what would convince him of the non-existence of Italic as a genetic subgroup would be innovations shared between either Latin or Sabellic with some non-Italic branch of Indo-European. I cannot help but note that Latin, if it indeed de-aspirated its inherited voiced aspirate stops word-internally, seems to be intermediate between Sabellic and Germanic on the one hand (where the three-way contrast between stops was basically preserved) and Celtic (where voiced aspirates were de-aspirated everywhere) on the other.

    Indeed, the “Italo-Celtic” theory might best be called the “Celto-Latin” theory, because of the few isoglosses linking “Italic” and Celtic one is shared between Latin and Celtic to the exclusion of Sabellic: Sabellic has no reflex of the masculine/neuter genitive singular ending /i/ found in Latin and Celtic. Moreover, the assimilation of /p/…/kw/ to /kw/…/kw/, another Italo-Celtic isogloss, may or may not be shared by Sabellic. The problem is that Sabellic turns its labio-velar stops into bilabial stops. As a result, the derived form of “five” in Oscan (PUMPERIA) may go back either to an “Italo-Celtic”-type form */kwenkwe/ or the Proto-Indo-European form */penkwe/. We have no way of knowing.

  188. Etienne, again, aren’t there more cognates shared between Sabellic and Latin than with other languages?

  189. Etienne says:

    Y: We don’t know. First, our knowledge of all aspects of Sabellic languages (including the lexicon) is quite incomplete because of the limited amount of data. Second, leaving aside bilingual inscriptions Sabellic inscriptions were deciphered on the basis of (perceived) similarities to Latin, and it thus could be the case that undeciphered inscriptions simply contain fewer Latin cognate vocabulary than those inscriptions which we have deciphered.

  190. Eli Nelson says:

    @David Marjanović:

    Not quite, no. I can’t think of an example where *f became b. The change from *b, which was [β] and/or [v] in most positions, to the voiceless (!) plosive b did happen, but it didn’t touch *f; *b and *f are distinguished about as well in High German as in Gothic.

    Thanks for the correction! I should have checked before making that statement. I guess I was misled by all the words that end in a vowel followed by “b” in German where the English cognate ends in /f/… I wrongly assumed some of these were originally /f/.

  191. David Marjanović says:

    Indo-European *dʰ turns into /d/ word-internally too, and even in the minority of cases where it becomes /b/ instead an intermediate fricative stage needn’t be postulated:

    What about the strange change *sr > br, though? That seems to require one of just these fricatives: either *sr > *[θr] > *[fr] > *[vr] > br or *sr > *[zr] > *[ðr] > *[vr] > br.

    I haven’t finished reading the book, so I can’t yet say how convincing its case for Proto-Italic is.

    Sabellic has no reflex of the masculine/neuter genitive singular ending /i/ found in Latin and Celtic. Moreover, the assimilation of /p/…/kw/ to /kw/…/kw/, another Italo-Celtic isogloss, may or may not be shared by Sabellic.

    Both of these changes are not quite universal within Celtic either (contrary to what I have claimed here for years in the latter case).

  192. It thus could be the case that undeciphered inscriptions simply contain fewer Latin cognate vocabulary than those inscriptions which we have deciphered.

    Sounds like it would be quite a project trying to decipher the rest by lookiong for cognates in other branches. Has anyone tried doing anything like that?

    In the bilingual inscriptions alone (which IIRC are not negligible in number), is there a significant non-Latin cognate component?

  193. David Marjanović says:

    Sounds like it would be quite a project trying to decipher the rest by lookiong for cognates in other branches. Has anyone tried doing anything like that?

    There is, in any case, a handy compendium of Sabellic words that aren’t understood

  194. Surely there are more such words than múfqlóm.

  195. David Marjanović says:

    I mean the book, not the blog post about the book. Ceci n’est pas….

  196. Oh, right! Sorry, I haven’t been up for long, and the caffeine is still being assimilated…

  197. David: I will grant you that the identity in Latin of the reflexes of */s/ when followed by */r/ on the one hand and of *dʰ (when flanked by */r/, followed by /l/, or after /u/) on the other (/f/ word-initially, /b/ word-internally) does suggest that in these positions *s and *dʰ merged phonologically, and a fricative of some kind does seem likely.

    But does it thereby follow that all other word-internal aspirates which yielded stops in Latin must also have gone through a fricative stage? I don’t think so. All instances of word-internal */dʰ/ other than under the conditions listed above, and all instances of word-internal */bʰ/ and */gwʰ/, yielded unaspirated stops (in the case of the last phoneme the identity of its word-internal development with that of */gw/ makes de-aspiration likely). The only reason anyone postulates a fricative stage is because of the need to shoehorn Latin sound changes with those of Sabellic. Treat them as separate branches of Indo-European, though, and things become simpler.

    For instance, take the phonological split of */dʰ/ word-internally in Latin (in the environments listed above, it becomes /b/ word-inernally; otherwise it becomes /d/): this is alien to Sabellic, where all instances of word-internal */dʰ/ yield /f/. If we assume a common Proto-Italic parent language, distinct from Proto-Indo-European, then we’ve a problem.

    For, if the split in Latin of the reflexes of */dʰ/ is not Proto-Italic, then we are forced to the conclusion that word-internal */dʰ/ was still a separate phoneme in Proto-Italic: mark you, we could assume a Proto-Italic where aspirates had become fricatives word-initially and where word-internally the distribution of stops was Proto-Indo-European-like: but in that case you would have to postulate two separate sound changes of */dʰ/ to */f/: a “Proto-Italic” word-initial one, and a specifically Sabellic word-internal one.

    Or, conversely, we could assume that the split dates back to “Proto-Italic”: but since all instances of */dʰ/ (word-internal and word-initial alike), as well as of */bʰ/, yield /f/ in Sabellic, whereas simple voiced stops remain stops, the split in question must have yielded word-internal aspirates. If you assume a “Proto-Italic” where aspirate stops had become fricatives word-initially and remained unscathed word-internally, then you *again* need to assume two instances of the shift to fricatives: one word-initially, in “Proto-Italic” times, and the other word-internally, in Proto-Sabellic.

    To my mind Occam’s razor suggests the following solution: there never was a Proto-Italic. Prehistorically Latin de-aspirated its word-internal stops, with the exception of */gʰ/, which shifted to /h/, and of some instances of */dʰ/, whose reflex became identical with that of */s/ when followed by */r/. Meanwhile, in Sabellic, the distribution of aspirate stops remained what it had been in Proto-Indo-European. Within what we may call the “Italic language area” a single sound change spread, whereby */bʰ/ and */dʰ/ yielded /f/, via a fricative stage.

    Finally, regarding the assimilation of /p/…/kw/ to /kw/…/kw/: there is no reason to assume it must have taken place in Sabellic, but every reason to assume it took place everywhere in Celtic: Old Welsh PIMP “five”, or the Gaulish ordinal PINPETOS “fifth”, cannot go back to a form */penkwe(tos)/: the Proto-Celtic loss of Proto-Indo-European /p/ would have yielded *IMP, *INPETOS. The attested forms must go back to a form */kwenkwe(tos)/.

    As for the /i/-genitive, I myself once pointed out here at Casa Hat that it is not pan-Celtic: glad you agree.

  198. via a fricative stage

    Or else not. This business of sound-changes having to be “natural” may be a win when you are trying to decide between two different paths, but Occam says not to introduce paths at all when there is no evidence for them. Let a thousand /du/ > /rk/s bloom.

  199. I’m struggling to understand that line of reasoning. I mean, a leap and a gradual transition are both paths, aren’t they, so why not assume the more natural one one until there’s evidence for the less natural one? Surely it’s reasonable to suppose that there were intermediate steps in that famous Armenian example.

  200. The point about the *sr > br change isn’t just that it implies an intermediate interdental fricative stage, but that it provides independent evidence that those fricatives hardened to stops word-internally in Latin. If we didn’t know that, Etienne’s model could indeed be argued to be more parsimonious, but given that we do, it looks to me like everything falls into place if you simply assume that this last hardening change was Latin-specific but the previous changes were Italic.

  201. It’s reasonable, but it’s not necessary. I mean, when confronted with the evidence of Proto-Nuclear Polynesian /l/ regularly becoming /ŋg/ in Rennellese, we can invent various chains to get us from one to the other: this thread has l > r > γ > g > ŋg and l >lj > nj > j˜ > ŋg as alternative “natural” pathways — but really, three intermediate steps, whatever they may be, with no evidence for them? Occam is rapidly whirling even as I speak.

    Similarly, confronted with b > k in Long Terawan Berawan from the same paper, it can probably be accounted for as a two-stage result: an idiosyncratic shift b > g (itself quite unusual typologically) followed by a general fortition of intervocalic stops (which is less strange, see High German, but still strange). But in related Sa’ban, the shift g > p happens not only finally, where there is general devoicing, but also initially, where there isn’t, and does not occur intervocalically, meaning that there is no reason to think it didn’t happen all at once. Here’s Blust’s summary:

    It has already been noted that the line between speculation regarding possible multi-step sound changes which produce bizarre results and the acceptance of phonetically or phonologically unmotivated single-step changes is often difficult to draw. With *t > k we seem to be on firm ground in claiming, at least for some languages, that this was a single-step change; for cases like Proto-Nuclear Polynesian *l > Rennellese ŋg, on the other hand, the matter is far less clear-cut. Should a given reflex which implies a bizarre sound change be treated like *t > k, where there evidently was no intermediate step, or like *l > ŋg, where intermediate steps may have been present? Setting this problem aside, it seems clear that bizarre reflexes are far less numerous than historical developments which result from phonetically transparent processes of assimilation, lenition, fortition, and the like. Nonetheless, changes of the type highlighted in this paper form a distinct subset of apparent phonological innovations which is difficult to dismiss, and closer attention to this problem will undoubtedly turn up many other examples.

    Blust thinks that the “unnatural” sound changes must be the result of conscious manipulations, and cites the parallel of the Uisai dialect of Buin, in which all the feminines are masculine in other dialects and vice versa, a change which surely couldn’t have happened all by itself. Similarly, Iban has a great many words with Malay cognates that mean exactly the opposite, notably kampong, which in Malay means ‘cluster of buildings (> English compound ‘id.’), but in Iban means ‘forest’. But I prefer the Shit Happens theory, in which naturalness predicts which sound changes are likely to recur across languages, but has nothing to say about the ones which actually have occurred in a particular language. No one doubts, for example, that intervocalic voicing of stops is a “natural” change, but then we have to ask why it never occurs in English or German with the exception of AmE /t/. The Shit Happens (or Shit Doesn’t Happen, rather) theory accounts for this perfectly.

  202. George Gibbard says:

    > in which all the feminines are masculine in other dialects and vice versa, a change which surely couldn’t have happened all by itself.

    Another example is Laro, in the Nuba Mountains of Sudan. It has four noun classes (which I’m using in the sense of ‘genders’ — for each, one prefix characterizes the singular and another characterizes the plural, while another convention in African linguistics is to consider singular and plural different noun classes), call them A, B, C and D. While the closest relative Ebang/Heiban has more noun classes, which were reduced on the way to Laro, comparison shows clearly Laro must have at one point switched all class A nouns to class B and vice versa, and likewise for classes C and D.

    (Reference: Thilo C. Schadeberg. 1981. Die Geschichte der Nominalklassen des Laru (Kordofanisch). In Jungraithmayr, Herrmann (ed.), Berliner Afrikanistische Vorträge: XXI. Deutscher Orientalistentag, Berlin 24.-29. März 1980, 203-212. Berlin: Dietrich Reimer.)

    Obviously it must have started as a sort of secret language to keep out outsiders and eventually was nativized.

  203. George Gibbard says:

    Of course comparison with other related languages is what shows that Laro and not Ebang/Heiban switched its noun classes.

  204. Occam says not to introduce paths at all when there is no evidence for them. Let a thousand /du/ > /rk/s bloom.

    Your example blows your point out of the water. You can’t possibly believe that *du magically became rk with no intermediaries. Occam’s razor is a tool, not an omnipotent deity.

  205. John Cowan: whatever intermediate steps you posit between */dʰ/ and /f/, it is clear that this sound change is fairly unusual, and thus it is unlikely to have taken place more than once. As I have tried to show above, any “Proto-Italic” language you postulate requires you to assume that this sound change must have taken place twice in the history of Sabellic. If you assume Italic was a language area, and not a Proto-language, the difficulty vanishes.

    TR: I disagree. The *-/sr/- to -/br/- change, in Latin, is the ONLY context in which */s/ becomes a stop: intervocalically it is rhotacized, and in other positions remains /s/. By contrast, word-internal *bʰ and *dʰ always yield voiced stops in Latin, whether intervocalically, followed by /r/, or indeed in contact with any other consonant. Hence, if you wish to postulate a Latin “hardening of fricatives” the phonological environment wherein such fricatives were hardened differs greatly between */s/ on the one hand and the fricatives you assume existed as reflexes of Indo-European aspirates on the other.

  206. Etienne: The apparently limited environment of fricative hardening in the sr > br change is simply due to the fact that it was only before [r] that [s] changed to an interdental fricative in the first place. It’s not the environment of the change that is restricted, it’s the inputs to the change in this specific case (coming from [s]).

    If we agree that sr > br went through an interdental stage (whether [f] or [v]), then there are two ways of describing the subsequent sound change: either (a) those interdental fricatives hardened only before [r], or (b) they hardened generally (word-initial position excepted in both cases). Obviously in the scenario you propose, there’s no empirical difference between (a) and (b) since the interdentals only occur before [r]. But to the extent that “hardening of fricatives generally” strikes us as a more likely change than “hardening of fricatives only before [r]” (and we may disagree on that extent, of course), formulation (b) seems preferable, and if we opt for it then the aspirate changes, too, fall into place.

  207. Actually I should not have said “word-initial position excepted”, since the word-initial fricatives were voiceless and the hardening change targeted voiced fricatives. And “whether [f] or [v]” isn’t right either, but was meant as a shorthand for “however you get from [sr] to [vr] and thence to [br]” . Going to make my coffee now.

  208. Oh, I assumed that by “interdental” you meant /θ/ or /ð/.

  209. Yes, that is what I meant, which is why it made no sense to say “whether [f] or [v]”. But “whether [θ] or [ð]” wouldn’t necessarily be right either because it could conceivably have gone through both (sr > θr > ðr > vr > br).

    Whatever the exact trajectory of the change, the fact that the [s] of [sr] went to an interdental makes it more likely than not that there were already interdental fricatives in the language at that stage. It’s more probable that this [s] merged into an existing phoneme in that environment, it seems to me, than that it gave rise to a whole new (and relatively marked) phoneme in that environment only.

  210. the fact that the [s] of [sr] went to an interdental

    But a fact is what it is not. It is a hypothesis based on a priori grounds only: “we cannot conceive of [sr] > [br] in a single step, so it couldn’t have happened”. But the history of science is littered with such failures of nerve. If it were a fact, then the inference that an interdental fricative existed elsewhere in the language at that time would be more plausible, but as it isn’t, it ain’t; contrariwise.

  211. Well, this is why my earlier comment said “If we agree that sr > br went through an interdental stage…” If not, then of course this whole line of argument is irrelevant. Failure of nerve it may be, but I find it unimaginable that pre-Latin speakers one fine morning went from saying muliesris to saying muliebris with no intervening stages (how? by a vote in the assembly?), and the intervening stage that makes most sense is an interdental.

  212. But the history of science is littered with such failures of nerve.

    Perhaps you consider the reluctance of science to embrace the theory that the pyramids were built with alien help a failure of nerve? I’m not sure to what extent you’re being contrary for the sake of being contrary, but I still refuse to believe that you actually think *du woke up one day to find it had been transformed into a giant rk.

  213. No, of course that was only a joke (as the Maoist line, I should have thought, indicated). But what I am protesting against is the notion that only natural sound-changes are possible, and that it is a fact that an unnatural sound change happened through an unobserved intermediary.

    I personally am not sure when I transitioned from /kænt/ to /keɪnt/ or /bu/ to /bʉ/, but they definitely have happened in the course of the last 30 years or so. Granted, both are natural changes, but if someone told me as a teen that I would no longer be saying /kænt/ or /bu/ when I was an old fart, I would have considered them an idiot.

  214. I should have used some different nominalization than “the fact that”, and/or made it clear that it was to be understood as part of the apodosis of “if we agree…”. (Probably Guarani or some such language has a nominal-mood marker that can neatly express such things.)

  215. David Marjanović says:

    No, of course that was only a joke (as the Maoist line, I should have thought, indicated).

    That line isn’t even in the same comment…

    In places where each clan has its own language (like Arnhem Land, as I read just yesterday), when a clan splits, the new clans will often make deliberate changes to their speech to differentiate it from everyone else’s – in one documented case literally by a discussion in an assembly. I have trouble imagining that this was the sociolinguistic situation of central Italy in the 2nd millennium BCE.

    On the Great Armenian Puzzle, I recommend this paper.

  216. David Marjanović says:

    As for the /i/-genitive, I myself once pointed out here at Casa Hat that it is not pan-Celtic: glad you agree.

    I don’t agree or disagree; I was simply copying a statement about Ranko Matasović’s Proto-Celtic reconstruction from Wikipedia. The same holds for the lack of universality of *p…kʷ > *kʷ…kʷ: I have no idea in whichever Lepontic or Celtiberian this change might be lacking, or if perhaps even the Hercynian Forest is the only piece of evidence here.

    I haven’t been able to continue reading Stuart-Smith’s book lately; so far I agree that the *-sr- > -br- is strong evidence that [θ] or [ð] must have been part of the sound system of an ancestor of Latin.

  217. Where there is enough evidence, some of these odd sound changes owe their oddity precisely to the fact that they are the result of chaining several sound changes together: for Armenian , see Viredaz, The Great Armenian Puzzle Is Not Thorny; likewise Jacques, The sound change *s>n in Arapaho. Where you don’t have enough evidence for such a sequence of changes, it’s reasonable to speculate that such a chain existed, and hope to eventually elucidate its history. As these papers show, even when there’s enough evidence to figure these things out, linguists are often wrong on the first few tries.

    (Scooped by DM on Armenian!)

  218. David Marjanović says:

    even when there’s enough evidence to figure these things out

    In the case of Arapaho, some very helpful evidence comes from the most closely related language which is not only extinct, but just barely documented at all. Reading that paper was downright scary.

  219. Where you don’t have enough evidence for such a sequence of changes, it’s reasonable to speculate that such a chain existed, and hope to eventually elucidate its history.

    Moses < Middletown by our very own Ken Miner, via a brain-softeningly logical sequence of natural phonological changes. The only thing wrong with it is the misspelling of “vowol harmono”.

  220. I seem to remember Blust’s article, Must sound change be linguistically motivated?, came up here on LH once. It’s a catalog of weird sound changes in Austronesian, which Blust finds no explanation for. Austronesian historical linguistics is prone to turn sober scientists into mystics.

  221. He does have an explanation for them, which I don’t think much of (people make bizarre sound changes on purpose). But that’s no worse than postulating unobserved four-step shifts just to save naturalness in 100% of cases. Treating naturalness as a merely statistical universal that can be violated, but rarely, makes a lot more sense to me, and is what I’ve been saying throughout this thread.

    [insert Trond-style disclaimer here]

  222. I just don’t get it. I mean, of course if one has good evidence that a given change happened in a bizarre way, that’s fine, but you don’t get to leap on every undocumented change and say “Aha! Since there’s no proof it happened in a natural way, I’m going to assume it didn’t!”

  223. As David says, the historical sciences deal with unobservable events, and so have to use parsimony in order to figure out what’s most likely to happen. Which is the more parsimonious assumption: that four probable but unobservable steps took us from *n to /s/, or that a single improbable step did so? When it turns out that there are two four-step hypotheses, both involving only natural steps, the likelihood of either one being right drops further. Improbable does not mean ‘inconceivable’.

    All I am arguing is that the assumption that probable steps always, even in the absence of intermediary evidence, beat improbable steps is not good science. In semantics, we accept that catastrophic (in the mathematical sense) change happens: we do not postulate natural steps to take us from yelp ‘vaunt, boast’ to yelp ‘make a dog noise’, nor from beg the question ‘assume the matter at issue’ to ‘raise the question’. These changes don’t happen instantly, of course, but they do happen abruptly in the sense that there is no evidence for, and we do not postulate, semantically intermediate stages. Why should we automatically postulate them in phonology?

  224. David Marjanović says:

    He does have an explanation for them, which I don’t think much of (people make bizarre sound changes on purpose).

    They probably do sometimes in small communities in certain kinds of cultures. Even then, though, that probably only accounts for one step in a chain of 3 or more steps in the more thoroughly “unnatural” of Blust’s examples.

    Which is the more parsimonious assumption: that four probable but unobservable steps took us from *n to /s/, or that a single improbable step did so?

    Clearly the former. Not only because, in this particular case, this scenario also accounts for the evidence from another language, but also because really long chains of sound shifts happen all the time. (Historical linguists actually tend to present them as shorter by not reconstructing the more obvious intermediates.) In the abovementioned article on *dw > rk in Armenian, Viredaz presents the change from Latin [ɪ] to French [ŏɐ̯] about as follows (IPA transcription mine): [ɪ] > *[e] > *[eː] > [ei̯] > *[əi̯] > [ɔi̯] > [ɔɛ̯] > [ɔɐ̯] > [ŏɐ̯]. That’s eight changes in just 1500 years (one every 7.5 generations or so) with four attested intermediate states out of seven.

    (I would actually stretch out [ɔi̯] > [ɔɛ̯] > [ɔɐ̯] to [ɔi̯] > *[ɔɪ̯] > [ɔɛ̯] > *[ɔæ̯] > [ɔɐ̯], for ten changes in 1500 years or one every 6 generations, but at this level of detail it becomes completely arbitrary to decide which intermediate states are attested and which would have been attested in the same way as the previous or the next one.)

  225. That’s eight changes in just 1500 years (one every 7.5 generations or so) with four attested intermediate states out of seven.

    Yes, but they ARE attested. Which means we have several chains of the form (attested)-(unattested)-(attested). That’s very different from (reconstructed/unattested)-(unattested)-(unattested)-(attested).

  226. One issue to look at here is time depth – e.g, attested Armenian begins roughly a minimum of 4 millennia after the supposed break-up of PIE, so it looks much more likely that /dw-/ > /rk/ involves a chain than if we would be talking only about a century of two.

  227. David Marjanović says:

    Yes, but they ARE attested.

    That’s my point: they attest to the fact that long chains of sound changes happen.

  228. David Marjanović says:

    Reading on, I have learned that:

    1) Not only *-sr- and *-bʰr become -br- in Latin, but *-dʰr also does.

    2) The Faliscan reflex of *-gʰ- is /g/, without known exceptions. …Or, anyway, it is early k~q, late c; the invention of the letter g never made it to the Falerii. Other than this unique development, the Faliscan reflexes of the aspirates are much more like Sabellic than like Latin, arguing against the existence of a Q-Italic branch.

  229. they attest to the fact that long chains of sound changes happen

    Of course they do. Wait ten thousand years, and you can have any sound changes you want, and obviously not because at the five-thousand-year mark something changed abruptly before and after remaining constant.

    What I object to, and all that I object to, is postulating long and entirely hypothetical chains when the sole motivation for them is to make every change one that often happens. But I’m going to shut up now, because obviously this is not working.

  230. Somewhere in this is the difference between aggregate and individual probabilities of these hypotheses.

    Experience shows that it is not unlikely that a long sequence of unsurprising changes in the evolution of some language will turn out feed each other in such a manner that an a priori ‘surprising’ change appears in some words two thousand years later.

    Experience likewise shows that it rarely happens that the speakers of a language call a plenary meeting and decide that [dwo] shall henceforth be pronounced [erk’], just to pull a fast one on the Neogrammarians.

    In fact, judging by cases where independent evidence has elucidated the history of ‘surprising’ changes, it is probably safe to say that they are overwhelmingly more often of the first type.

    HOWEVER, and here’s the rub: This does not mean that any single, defined chain of changes you may posit has a higher probability of being the actual patch to the observed result than the plenary meeting model has. There may be a thousand, or a hundred thousand, ways of getting to the same result with different sequences of unsurprising changes, so all you can safely say is that ‘the actual development was probably a long chain of individual changes’.

    (However, given a lot of data on the differential probabilities of individual changes and how they can feed each other it might be possible to assign quite high probabilities to a statement of the type ‘this phoneme probably acquired this feature before it lost that other one’ because it will be true for a large proportion of the most probable paths. You can also try just arguing from intuition, it’s less work and might give just as good results).

  231. Well put.

    Sometimes you can eliminate change paths in this way: A > B (unattested) > C is impossible, because B merged with B’ at that or a later point, but the reflex of B’ today is not C but something else. Merges can’t unmerge, though splits can heal.

Speak Your Mind

*