AN ANATOLIAN SCRIPT MYSTERY.

An Ask MetaFilter question that I’m hoping some LH reader can help with:

I found some stone tablets written in a strange alphabet amongst a bunch of graves from different eras at the city museum of Tire, Turkey. The guy working the desk at the museum didn’t know what they were. […] The museum had gravestones from many different eras of the city’s history — Roman, Selçuk, the Beylik period, Byzantine and Ottoman graves, and also some Armenian writing and some Jewish gravestones (seen in the first picture). As far as I can tell, it’s none of these. It seems that all the stones there were collected from around the area. 2 of the stones had this strange alphabet; here are some pictures:
The first
Close-up
The second stone
Does anyone have any idea what they might be?

There are various guesses in the comment thread—an archaic Semitic language, Khutsuri, and Nomyte’s suggestion, Armenian (cf. this image); so far, I’d say that’s the most plausible, but everybody’s just guessing, and I’m hoping somebody who actually recognizes it will weigh in. (Just ignore the stones with Hebrew lettering in the first image; there’s no reason to think they’re relevant.)

Comments

  1. Oops sorry, I see that people already got that far. I should have read the comments more closely before racing off on a search of my own. I just get too excited about this stuff. Gotta stay calm.

  2. I was going to say, they looked like Armenian lowercase letters but some of them are slightly weird. Then I saw AG’s Omniglot link for Georgian and that makes a lot better sense.

  3. David Marjanović says:

    Completeness requires me to mention Caucasian Albanian.

  4. A commenter in the AskMeFi thread says she’s sure it’s Armenian and will try to provide a translation.

  5. David Marjanović says:

    The “Liberman on suppletion” thread is closed, so…

    Jennifer 8 Lee, here’s an item with her byline:

    “By Jennifer 8. Lee.” o_O

    Piotr Gąsiorowski’s blog Language Evolution

    Oh! He has a blog now! I didn’t know that. There goes the rest of the day. 🙂 I know him from the Dinosaur Mailing List, though he hasn’t posted there in ages.
    Interesting to learn he considers his [f] optional.

    il est italo-compatible

    ROTFL!

    Anyway, he’s a rash choice since he is now going to be endlessly accused of taking one side or the other in the Argentinian civil war between two bunches of murderous thugs.

    War? What war? He’s accused of being too close to the military dictatorship during the Dirty War, which consisted of people being disappeared.

    dysphemism (had to look up the word, I confess)

    That surprises me, because I know it only in the context of historical Romance linguistics – the full frontal attack on the core vocabulary of Latin, which led to the curious situation that there’s no reconstructible Proto-Romance word for “head”, and Latin caput only survives in metaphorical meanings (like French chef).

    as a consequence of sound changes in the transition from Latin to early Romance many of the shorter forms were simply too ambiguous (thus, EO must have become homophonous with the reflex of Classical EGO in much of the Empire). As a result VADERE was used

    Oh. That makes sense.

    Romanian, in its splendid isolation, came up with a much more straightforward solution to the problem and wholly dumped IRE and replaced it with a single new verb, A MERGE, whose original meaning was “to dive”.

    Huh.

  6. David: I beg to differ. Reconstructing a Proto-Romance word for head would be easy. The data (drumroll, please!):
    1-Romanian CAP “head”, CAPUL “The head”.
    2-Spanish CABEZA “head”; since -EZA is a productive suffix in Spanish the segmentation CAB-EZA is scarcely arbitrary.
    3-French CHEF has preserved its older meaning in COUVRE-CHEF “Hat, anything worn on one’s head”.
    Analysis of the sound changes:
    A-Initial /k/ in Spanish (CAB-) and Romanian (CAP), followed by /a/, does correspond to the palatal fricative in French (CHEF) (cf. Spanish CANTAR/French CHANTER/Romanian CÂNTA(RE): the shift to  in Romanian is due to the following nasal).
    B-The vowel: stressed /a/ in Romanian and Spanish often corresponds to a French mid-low or mid-high vowel: cf. Spanish MAR/French MER.
    C-The /p/-/b/-/f/ equation: a little trickier. Romanian CAPUL “The head” is synchronically segmentable as CAP-UL, but other forms (FRATE, “brother”, FRATE-LE “The brother) would suggest that the original segmentation was CAPU-L.
    In which case (originally) intervocalic /p/ in Romanian would indeed match an intervocalic /b-v/ in Spanish, which other etymologies would confirm (Romanian LUP “wolf”, LUPUL “The wolf”, Spanish LOBO).
    D-French NEUF/Spanish NUEVO and NUEVE would establish both that French lost its final vowels and has /f/ word-finally corresponding to Spanish intervocalic /b-v/.
    Conclusions:
    1-CHEF and CAB(eza) and CAP(u) exhibit a perfect match for *every* phoneme, share a common meaning (if you include COUVRE-CHEF, above), and are geographically at peripheral locations of Romance-speaking Europe, making a common innovation unlikely.
    2-A reconstructed form */kapu/ or /kapo/ “head” for Proto-Romance would seem easily reconstructed. Indeed, writing a horror story in Proto-Romance, involving a singing wolf’s head and nine brothers by the sea, will be your next assignment.
    3-Next week, kids, we’ll take the above Proto-Romance form and compare it to English HEAD and other Germanic forms, and on that basis deduce Grimm’s law!
    4-Bonus question, kids: Since long-range comparison typically seems never to involve finding regular sound laws or morphologically analyzing the material, and involves taking considerable liberties with the semantics, how seriously should these comparisons be taken?

  7. Correction: under B, I should have written that stressed /a/ often corresponds to a French mid-high or mid-low FRONT vowel (Note to self: first coffee, THEN comments).

  8. “By Jennifer 8. Lee.”
    A friend of mine is named “B____ F M________”, where the “F” does not stand for anything (apparently his parents could not agree on a middle name, but their choices both began with “F”). The nameplate on his office read “B____ F. M________”, with an abbreviation point. When I tweaked him on this, he said that “F.” was an abbreviation of “F”.

  9. David Marjanović says:

    Romanian CAP “head”

    *tomato running down face*
    [ˈsɪkˑˈtʀansɪtˈg̊loːɐʀɪaˈmʊnd̥ɪ].

    COUVRE-CHEF “Hat, anything worn on one’s head”

    Huh. I didn’t even know that.

    long-range comparison typically seems never to involve finding regular sound laws or morphologically analyzing the material

    What do you mean by “long-range comparison”, and is it “typically” or “never”?

    apparently his parents could not agree on a middle name, but their choices both began with “F”

    That’s famously the case with Harry S Truman: his grandfathers were Solomon and Shipp. I failed to ask David S Berman (Carnegie Museum of Natural History, Pittsburgh) if he’s named after Harry S Truman.

  10. Mass vocabulary comparison should be taken very seriously in the context of discovery, when what one needs is hypotheses. In the context of justification, where one needs to prune those hypotheses, it has no value.
    Not all long-range comparison relies on mass vocabulary comparison: Dene-Yeniseian does not.

  11. marie-lucie says:

    Etienne: The root CAP is also found in the Latin derivatives CAPILLUS ‘hair (of the head)’ and (later) CAPPELLU- ‘hat’, which led by regular rules to French cheveu and chapeau respectively (-p- > -v- but -pp- > -p-, among other rules).
    long-range comparison typically seems never to involve finding regular sound laws or morphologically analyzing the material
    That is because most of those engaging in long-range comparison don’t seem to be aware of what they should be doing (and neither do most of their harshest critics). The second part is especially neglected by both long-rangers and their critics.
    JC: Mass vocabulary comparison should be taken very seriously in the context of discovery, when what one needs is hypotheses. In the context of justification, where one needs to prune those hypotheses, it has no value.
    Yes. MC is one tool, useful at the preliminary stage, but too blunt and imprecise to be used for detailed work.
    Not all long-range comparison relies on mass vocabulary comparison: Dene-Yeniseian does not.
    That’s because those who are working on it do know what they are doing. But they are a rarity.

  12. marie-lucie says:

    DM: What do you mean by “long-range comparison”
    This means comparison (with a view to proto-language reconstruction) between languages which show similarities but have not been “proven” (as far as that can be done) to be related to each other. This term describes research in “Nostratic”, “Eurasiatic” and a number of other proposed large-scale groupings which have not been generally accepted by linguists. “Indo-European” research would have been considered an example of it at its beginnings (but the term was not used then), before the membership of the group was established through detailed comparison of morphology and vocabulary, leading to the (largely successful) attempts at reconstructing the common ancestor.

  13. I think a lot of historical linguists working on non-IE families, frustrated by their lack of the “easy” comparisons available in IE because of the extensive evidence, react by either ignoring the whole mass of scientific procedure developed for IE or pretending it somehow doesn’t apply to their families. But the hard truth is that unless you can pin down convincing, consistent sound laws and morphological relationships, all you have is a hypothesis, which may be interesting and suggestive but will never be accepted as true by careful linguists.

  14. David Marjanović says:

    This means

    I know; I wanted to know specifically what Étienne understands by it in this context.
    I agree that mass comparison may be good for generating hypotheses, but is incapable of testing them. That’s why, other than their occasional coauthors, only Greenberg and Ruhlen ever seem to have used it – and they have indeed made cautionary statements about it.

    I think a lot of historical linguists working on non-IE families, frustrated by their lack of the “easy” comparisons available in IE because of the extensive evidence, react by either ignoring the whole mass of scientific procedure developed for IE or pretending it somehow doesn’t apply to their families.

    That’s not my impression. Austronesian in particular is now very well understood all the way down – arguably better than IE, because in IE there’s almost no research on how the almost universally accepted branches Italic, Germanic, Albanian, Celtic, Balto-Slavic, Indo-Iranian, Armenian and Greek are related to each other, while for AN, there are complete, testable tree proposals.
    It gets worse when lots of poorly documented languages have to be taken into account, cases where too few field linguists have ventured into semideserts where the air has elevated lead content. The existence of the Afroasiatic (Afrasian, Afrasan [!]) is so blindingly obvious* that it’s finally no longer doubted (it was, by some, for a long time!); but the three existing attempts to reconstruct Proto-AA have problems (1 (pdf); 2) to the point that in one of them you can see the way Arabic dictionaries are structured.
    * Never mind the Omotic family, which is, guessed it, poorly documented at present.
    Perhaps similarly, the one reconstruction of Proto-Nilo-Saharan has obvious Arabic loanwords in it. Oopsie.
    Sino-Tibetan is a mess, due both to the sheer number of relevant languages (hundreds) and the poor documentation of many of them.

  15. marie-lucie says:

    David: I agree with Etienne.
    Austronesian in particular is now very well understood all the way down – arguably better than IE
    Austronesian is a relatively young family – that helps.

  16. Austronesian in particular is now very well understood all the way down
    I didn’t say “all,” I said “a lot.” Respect to the Austronesianists, but I was thinking of the Nostraticists, Proto-Worldists, et hoc genus omne.

  17. Rodger C says:

    Well, as the Catalans say, “En cap cap cap el que cap en aquest cap.” (In no head fits what fits in this head.)

  18. Rodger C.: Yes, and bring in Catalan CAP (the noun!) into the equation, and the above sound laws would lead you to make predictions: that the Catalan cognate of French CHANTER/Spanish CANTAR/Romanian CÂNTA(RE) will preserve initial /k/ intact (it does: CANTAR), that the cognate of Romanian LUP/Spanish LOBO will also end in /p/ (LLOP, what a stunning coincidence!)
    The fact that Catalan initial /lj/ (LL) regularly corresponds to /l/ in French, Spanish and Romanian (LLENGUA/LANGUE/LENGUA/LIMBA, and LLOP/LOUP/LOBO/LUP) must be another coincidence. Or, as some “scholars” might have put it, the “wolf” word is onomatopeic! I mean, the stressed /o/ or /u/ is obviously an imitation of a wolf’s cry.
    (I am barely exaggerating here).
    There is a point I am trying to make here: in comparisons like this, quality is what matters, not quantity. I would maintain that demonstrating the genetic relationship of the Romance languages (with no reference to pre-modern data) could be done with fewer than twenty cognate sets.
    Let us take what I have already shown, notably the fact that Spanish final /o/ corresponds to zero in Romanian (LOBO/LUP). Compare verbal morphology in both languages: the fact that Spanish CANT-O corresponds to Romanian CÂNT (“I sing”), exhibiting the same phonological match which was found in the other vocabulary items, makes the possibility of loanword diffusion unlikely: person-marking morphology is very rarely if ever borrowed. Add a few more instances of morphology corresponding with the same sound laws operating (i.e. Spanish adjectives inflected for the masculine singular ending in /o/ versus zero in Romanian) and the demonstration of genetic relationship would be to my mind bulletproof.
    Long-range comparison (Marie-Lucie summed up exactly what I meant by the term above) seems, on the other hand, typically to favor quantity: lots of words that vaguely look alike, but without any kind of recurring phonological regularity (i.e., phoneme A in language 1 corresponding to phoneme B in language 2), and seemingly no awareness as to what is liable to be borrowed from another language and what isn’t, as to what is diachronically stable and what isn’t.
    As for morphological analysis: when it isn’t ignored it is done so rigidly that it does more harm than good. A certain “scholar” (DE CUYO NOMBRE NO QUIERO ACORDARME, as a certain Spanish writer once put it) once said that you must analyze your material morphologically before comparing anything: lexeme with lexeme, grammatical morpheme with grammatical morpheme.
    Of course, the fact that morphological boundaries change diachronically seems to escape said scholar: on the basis of the segmentation of Romanian CAPUL “the head” as CAP-UL (synchronically impeccable) an attempt at comparing -UL with other Romance masculine singular definite articles (Spanish EL, Italian IL) would yield real problems, as no other Romanian (initial or not) /u/ matches Spanish /e/ or Italian /i/.
    You’ll forgive me if a “method” that demonstrably makes comparison more difficult is something I cannot even pretend to take seriously.

  19. But, of course, beware the French and Italian superstrates in Romanian, which can be very misleading to the novice comparative linguist. By the same token, it would be easy to set up sound-laws showing that Modern French and Modern English are close relatives.
    Here are the first two sentences of the UDHR (“All human beings are born free and equal in dignity and rights. They are endowed with reason and conscience and should act towards one another in a spirit of brotherhood”) in different flavors of Romanian, from Wikipedia:
    The normal translation, with Latin/Romance loanwords in italics: “Toate ființele umane se nasc libere și egale în demnitate și în drepturi. Ele sunt înzestrate cu rațiune și conștiință și trebuie să se comporte unele față de altele în spiritul fraternității.”
    Rewritten to use Slavic loanwords instead, also marked in italics: “Toate ființele omenești se nasc slobode și deopotrivă în destoinicie și în drepturi. Ele sunt înzestrate cu înțelegere și cuget și trebuie să se poarte unele față de altele în duh de frățietate.”
    Rewritten again to use native words only: “Toate ființele omenești se nasc nesupuse și asemenea în prețuire și în drepturi. Ele sunt înzestrate cu înțelegere și cuget și se cuvine să se poarte unele față de altele după firea frăției.”
    The root drept- < Latin directum particularly stands out to me: direct is also present in the language, but is borrowed from French

  20. marie-lucie says:

    Etienne: Spanish LOBO, Rumanian LUP, etc. In Occitan too “wolf” is LUP.
    Having heard wolves in the wild, I don’t find the sound of English “wolf” to be onomatopeic! Perhaps if it meant “dog”, or some races of dogs …
    it would be easy to set up sound-laws showing that Modern French and Modern English are close relatives.
    Greenberg actually made this very point. Yes, this is what can happen when you have very old borrowings which look like they are true cognates (each descended directly from a common ancestor) and you draw conclusions based on that mistaken belief. What shows that French and English are not close relatives in spite of having a lot of common vocabulary with even many regular sound correspondences is that the basic morphology in the two languages (especially in verbs) is built under very distinct principles. Personally, I have a lot of faith in morphology, especially those parts of morphology that can be demonstrated to be archaic within a language ( unproductive, irregular, etc). It is easy to see that English has borrowed a lot of French words, including verbs, but not the complex way of making verb forms, similar to the patterns in Latin, Spanish, Italian etc. Instead English has kept verb patterns related to those of German, Dutch, and Scandinavian. In an earlier period French borrowed many Germanic words (from the Franks especially), but Germanic verb-formation has never been introduced into any version of French. Meillet (one of the most famous Indo-Europeanists) said that the only thing he needed to prove that a language belonged to the Indo-European family were the forms of the verb ‘to be’, which are based on several roots and are in most cases highly irregular (not a list of a few dozen “basic” words, for instance).

  21. marie-lucie says:

    Sorry, it’s JC who mentioned French and English.

  22. Trond Engen says:

    And the Dene-Yeniseyan connection is almost pure morphology. The cognates are there, in a growing number (or at least it was last time I looked), but without the morphological match they are too far apart in sound and sense to really stand out from a mere wordlist.

  23. David: I would say that in Austronesian there are way too many tree proposals, and little consensus on testability. The first-order groups (genera) are obvious enough, but above that chaos reigns. There isn’t even agreement on how many top-level groups there are: five, ten, twelve? Granted, the data is radically insufficient, as all but one of those groups are spoken only on Taiwan, and only in moribund languages. But even if we step down to the Malayo-Polynesian level, then (Oceanic excepted) we have a vast mess of groups below it with languages full of contact effects across group boundaries. And so on all the way down to the genera.
    None of this, however, is the result of failure to apply the comparative method properly; it’s just that it’s tough to apply and tougher to interpret. Furthermore, Austronesian is full of undoubted sound-changes that would be called bizarrely unlikely anywhere else on Earth, like *y > p, *l > ŋg, and the famous *t > k, which is barely two centuries old in Hawai’an (and does not affect Ni’ihau dialect, which is isolated) and is still in progress in Samoan. There are dozens of examples of *t > k elsewhere in AN, but only two or three in other families).
    m-l: PAN is dated to about 6000 years ago, the same as PIE. (I’ve always liked the semantic consonance of the names “PIE” and “PAN”.) Austronesian settlement on Taiwan is rather older by the archaeological evidence, but it was apparently quite a while before the common language broke up into groups with known descendants.

  24. Might this be Armazuli script?
    Apparently a unique local form of Aramaic writing was used in Georgia before the Georgian alphabet. There are theories that it could be amongst the inspirations for the original Asomtavruli Georgian alphabet, the form which bears a superficial resemblance to the Armenian alphabet.
    I’m a bit of a Georgiaphile but I hadn’t heard of this until reviewing the Talk page of the Wikipedia article on the Georgian alphabet.
    Some Google searching turned up an excellent image of a bilingual inscription if you scroll down this conversation thread at forum.ge.
    (Brace yourself for a hint of nationalist ranting always in the mix when the topic is the origin of the Georgian alphabet.)

  25. The current word on *t > k. It turns out that even on Ni’ihau, written Hawai’ian (in the form of the Bible) has affected local speech, so that /k/ is H and /t/ is L in a highly restricted sort of diglossia. This is the opposite of what’s happened on Samoa, where /t/ is used in writing and when talking to chiefs and elders, but not to equals.

  26. marie-lucie says:

    JC: Thanks for the details on PAN. I am leery of those absolute dates such as “6000 years ago” or such. This date was chosen for PIE on some presumably reliable basis and has since been considered as the limit of proto-language reconstruction for a large language family. I am not sure that PAN is not dated that far by analogy with the presumed date for PIE rather than according to independent criteria.
    Postulated “bizarrely unlikely” (indeed) changes such as *y > p, *l > ŋg, etc do not at all inspire confidence that the reconstruction is correct! I have found (elsewhere) that many times a correspondence between two sounds in different languages is labelled as a change from one to the other, rather than both being the result of a different series of changes from a common ancestral phoneme.
    As for *k > t, it is much easier to understand: those languages have a very limited consonantal inventory and when plosives become reduced to labial p and a single other contrasting plosive, that other one pretty much has to be either t or k. I am not surprised that in some of the languages both are used, but in different social contexts.

  27. Hat: I think part of the problem is that most scholars today who work with non-Indo-European languages have no idea how messy Indo-European data is, because they lack any first-hand experience with Indo-European: what they have been given in their textbooks is a tiny subset of existing data, specially chosen to exemplify perfect regularity.
    So of course they are puzzled and panic when they realize that the real linguistic data they collected themselves isn’t like this at all.
    Consider that most were taught that Sir William Jones recognized that Sanskrit and a half dozen other ancient languages must be “sprung from some common source”, but few if any know that Sir William Jones failed to perceive that Modern Indo-Aryan languages are later, changed forms of Sanskrit.
    John Cowan: like Marie-Lucie I strongly believe in the value of inflectional morphology. This is because it would quickly allow us to show that the French element in English must be borrowed, as Marie-Lucie pointed out above.
    We could likewise separate the native from the borrowed Romance elements of Romanian using the same methodology: take that word DREPT you are fond of. Comparison with Spanish DERECHO (same Spanish final /o/ -Romanian final zero correspondence here) points to some consonant cluster. Other word pairs show the same sound law (cf. Spanish LECHE/Romanian LAPTE) Now, regular correspondences can be due to inherited or borrowed material.
    *But*, crucially here, we find a shared irregularity in inflectional morphology in both languages which shows the same correspondence: Spanish HECHO, the *irregular* past participle of HACER, matches (archaic, dialectal) Romanian FAPT, the (likewise irregular) past participle of A FACE. Tellingly, while many Romanian words containing the cluster /kt/ have look-alikes in other Romance languages which likewise have /kt/, this /kt/-/kt/ correspondence is never found in shared inflectional morphology. That would be enough to indicate that Romanian PT matching Spanish CH is inherited, and /kt/ matching other /kt/ clusters in other Romance languages must be borrowed.
    Marie-Lucie: Meillet had trouble making up his mind: while he indeed once claimed that the forms of the copula (French EST/SONT, German IST/SIND…) in different Indo-European languages would be enough to prove their genetic relatedness, elsewhere he wrote that without written data from older Indo-European languages only we could never prove that the Modern languages are related, although we might suspect it.
    And I quite agree with you on the difference between a sound correspondence and a sound change: and it may surprise some to learn that a /k/ to /t/ correspondence exists in Europe: Vulgar Latin *CIPULLA “onion” was borrowed into Basque as TIPULA!
    It was Hugo Schuchardt who suggested that this was because the initial consonant of the Latin word was no longer /k/ but was realized as a palatal stop (IPA: /c/) and not (yet) as an affricate, and that Basque speakers, not having a /c/ in their phonology, but having a /t/ and a /k/, reproduced this Vulgar Latin /c/ with a /t/.
    I strongly suspect something similar was at work in those Austronesian languages where /k/ changes to /t/ or vice-versa.

  28. Hat: I think part of the problem is that most scholars today who work with non-Indo-European languages have no idea how messy Indo-European data is, because they lack any first-hand experience with Indo-European: what they have been given in their textbooks is a tiny subset of existing data, specially chosen to exemplify perfect regularity.
    So of course they are puzzled and panic when they realize that the real linguistic data they collected themselves isn’t like this at all.
    An excellent point that had never occurred to me.

  29. I don’t think there’s any doubt about /l/ > /ŋg/ (written g) in Rennellese, a Polynesian Outlier language. Such well-established pan-Polynesian words as *lima > gima ‘five’, *fale > hage ‘house’, and *tolu > togu ‘three’, where the reconstructed form is Proto-Nuclear Polynesian (very close to Samoan, except for the loss of PNP glottal stop), seem very firm, and there are plenty more. Indeed, Rennelese /l/ appears only in an adstratum of non-Polynesian words, and is described as a palatalized dental lateral, with allophones closes to /ð/. Admittedly, some such chain as l > r > γ > g > ŋg could account for it. More later when I’ve finished reading this paper on weird sound shifts, by the same author (Brust) as the paper linked above.

  30. As for *t > k (not the other way about) it has happened at least twenty independent times (counting diffusion across different languages/dialects as a single event), and in different ways. In Samoan, it is part of a general backing trend that has k > ʔ, and ŋ merging into n; in Hawaiian, however, it is ŋ that has merged forward into n! Even in the consonant-poor Polynesian languages, a three-way distinction in plosives is maintained, whether labial/coronal/velar or labial/velar/uvular, so it is not a matter of just filling up oral space, nor is it a matter of affrication. And there is no question of a chain here: Samoans switch from /t/ to /k/ and back in the course of single conversations.

  31. More later when I’ve finished reading this paper on weird sound shifts
    That looks like an extremely interesting paper; I’ll be curious to hear what marie-lucie thinks of it.

  32. Trond Engen says:

    some such chain as l > r > γ > g > ŋg
    Or it could have happened by way of a laminal n.

  33. John Cowan, Trond: or by a shift from /l/ to /lj/ to /nj/ to a nasalized /j/ and thence to Rennellese G. Some transplanted Gallo-Italian dialects of Sicily have a retroflex /d/ as their reflex of Latin initial /l/ (via some well-understood intermediate steps), so the Proto-Polynesian-Rennellese sound correspondence isn’t quite as exotic as it seems.
    John Cowan: minor typo: it’s “Blust”, not “Brust”.

  34. marie-lucie says:

    Etienne: I think part of the problem is that most scholars today who work with non-Indo-European languages have no idea how messy Indo-European data is, because they lack any first-hand experience with Indo-European: what they have been given in their textbooks is a tiny subset of existing data, specially chosen to exemplify perfect regularity.
    I agree. Most people who have taken a course in historical linguistics, or even in Indo-European, have seen (usually short) columns of matching words, nicely arranged to show regularity. In addition, they have learned “laws” discovered by others (Grimm, etc), which (as in many other cases) seem to be obvious after the fact, once the relevant examples have been assembled (sometimes over many years) by scholars dealing with primary data. For instance, Grimm’s and Verner’s laws about the evolution of some Germanic consonants are usually taught together, but there is a half-century or so between their dates of publication. So when some modern scholars are faced with having to organize primary data themselves, they are at a loss except when dealing with very closely related languages (local equivalents of eg the Romance or Germanic families in terms of resemblances both in structures and individual words), especially if those languages do not have a written tradition.
    A well-known scholar of Amerindian languages (whose name I shall not quote) wrote something like “A language family is either obvious, with a rich skein of correspondences … or it is forever unknowable”. This could have been said in the very early days of Indo-European linguistics, by pessimistic linguists doubting the possibility of reconstructing PIE, but it should not be acceptable today. From a scientific point of view, this “all or nothing, now or never” attitude is not defensible either: it leaves no room for gradually accumulating evidence and building on partial generalizations and crossfertilizations. Greenberg rightly denounced this attitude, but what he proposed to do instead was not convincing.

  35. marie-lucie says:

    Etienne: the value of inflectional morphology
    Usually morphological comparison deals with individual morphemes (such as pronouns or inflectional affixes), but I also try to pay attention to general morphological structure: for instance, vowel ablaut (as in English sing/sang/sung/song and similar alternations in German) is much more important in Germanic that in Romance (although there can be vestiges in Latin), and Semitic typically intertwines vowels and consonants in intricate manipulations based on the typical CCC root. Of course, sharing a type of structure is not ipso facto evidence of relationship (and the opposite is also true), but it can be one argument among others when trying to build up a case for or against a given hypothesis of genetic relationship.

  36. The trouble with these lengthy chains, says Blust (thanks for the correction, Etienne), is that there isn’t a shred of evidence for them. The only reason anyone assumes them is the assumption that all, instead of just most, sound shifts are phonetically or phonologically natural, and so we contort our view of the evidence to try make the story simpler. One of the other cases Blust discusses is intervocalic fortition in Kiput (a non-Polynesian but Austronesian language of Northern Sarawak), and the hoops people have jumped through to explain how it is that Kiput /f/, /c/, and /k/ voiced to /v/, /j/, and /g/, but intervocalic /t/:/d/ and /p/:/b/ contrasts remained intact. Intervocalic fortition rather than lenition is weird enough, but affecting only three of five possible consonants? Sometimes the theoretically conservative thing is just to list the exceptions to your theory and move on.

  37. marie-lucie says:

    JC: lengthy chains … not a shred of evidence
    In a language family with many closely related languages, especially languages for which there are texts from various periods, there is often evidence of “chains” of changes in which the sounds (usually consonants) of different languages “move” along the chain and stop at different points. Obvious examples exist in the various outcomes of Latin non-initial consonants in the Romance languages, as in Lat patre- ‘father’, Italian padre, Spanish padre now [paðre], but French père where there is no longer any trace of the original dental stop. This series of examples is just one among many similar ones.
    Here it seems that the dental consonants in the various languages gradually “reached” different positions on the weakening chain t > d > ð > zero, and since series of similarly defined consonants often “work” together and are affected in the same way we would expect similar changes for the other consonant series (such as labial and velar), but such changes do not always affect all places of articulation equally: French has lost almost all of the Latin intervocalic stops (plosives), but not the labial ones, as in Lat lupa ‘she-wolf’, Sp loba now [loβa], Fr louve where the chain is p > b > β or v, as compared with the dental chain in Lat rota ‘wheel’, Sp rueda now [rweða], but Fr roue (and similar changes in the velar chain).
    I have not read the Blust article yet, but I don’t find it unusual to find that the change in Kiput has not affected all consonant series but has left the labial and dental stop contrasts unchanged. Should the language survive into the next centuries, we would expect that eventually all stops would be affected.

  38. marie-lucie says:

    JC: I would also consider the Kiput changes instances of lenition rather than fortition.

  39. When does devoicing count as lenition? The lenizhon patterns I know are frixathivization (from gemminnatted stopp to plain stop to frixathiv to debuccalihation to zero) and voizing (from voiceless stop to voizd stop to voizd consinuant to awoxiwant to zero). As for vortitionn, it normally occurs initially or finally or both, but not interfocalically.
    (See “The Final, Complete, Authoritative List of Self-Describing Linguistic Expressions”.)

  40. marie-lucie says:

    JC: When does devoicing count as lenition?
    Not too often according to everything you and I have learned. I todally agree with you:

    The lenizhon patterns I know are frixathivization (from gemminnatted stopp to plain stop to frixathiv to debuccalihation to zero) and voizing (from voiceless stop to voizd stop to voizd consinuant to awoxiwant to zero). As for vortitionn, it normally occurs initially or finally or both, but not interfocalically.

    But I don’t see any devoicing in the data, instead I see voicing, which as you say, usually counts as lenition. You quoted Blust, who apparently quoted someone else [recursion allowed here], and somewhere there must have been an error in the transmission:

    One of the other cases Blust discusses is intervocalic fortition in Kiput ……. Kiput /f/, /c/, and /k/ voiced to /v/, /j/, and /g/, but intervocalic /t/:/d/ and /p/:/b/ contrasts remained intact. (italics added)

    Alternately, those other linguists Blust refers to may have a strange idea of the way lenis and fortis consonants are defined. Or there could have been instances of fortition in other data, besides the leniting examples here, and whoever copied them got them mixed up in the process of editing or even just typing. In any case, the data as shown here are not instances of fortition within the generally accepted meaning of the term, but rather of lenition.

  41. Arrgh!
    Yes, I was completely wrong to say “Kiput /f/, /c/, and /k/ voiced to /v/, /j/, and /g/”. What happened is that /v/, /j/, and /g/ devoiced intervocalically to /f/, /c/, and /k/. Note that /j/ here is the voiced palatal stop.
    ** facepalm **
    ** open mouth, insert foot, close mouth **
    ** headdesk **
    Anyway, here are examples of Proto-North-Sarawak to Kiput fortition: *dua > duvih > dufih ‘two’, *pawat > pavat > pafiet ‘fruit bat’, *tajem > tacem ‘blowpipe poison’, *tuju? > tuceu ‘seven’, *agem > akem ‘hand, foot, leg’, *pager > pagel > pakel ‘fence’. Note that e = schwa.

  42. marie-lucie says:

    JC, you are human after all!
    I think that the Kiput fortition rather than lenition is probably linked to stress placement: it would make sense if stress is on the second syllable, following the fortiting (?) consonants. The Romance cases of lenition all occur after stress (eg [róta > rwéda] ‘wheel’), and I think the Kiput fortition is more likely to have happened before stress (which is probably predictable on the second or final syllable in these languages, therefore not indicated in the transcription), thus probably *agém > akém ‘hand, foot, leg’. Compare English aspiration in stops: it occurs not only initially but also intervocalically before stress, as in pén but also upón and appéal, all with [pʰ], but not in ópen or háppy where /p/ occurs after the stressed syllable and does not aspirate).

  43. The only mystery is /g/ then. The other one are fricatives and affricates where a separate rule can indeed be conceived to govern the,.

  44. I would be extremely surprised if the stress in Kiput was anywhere but on the penult (counted by morae, not syllables, but it is all one here). Note that in the full chain dua > duwa > duva > duvi > duvih > dufih ‘two’ (I abbreviated this chain above), the protoform certainly has penultimate stress, as in Malay, and if there were a stress shift it would surely have been mentioned, perhaps at the stage duva > duvi. Note the ongoing fortition of which -f- is only the extremum.
    I was also wrong to call /c/, /j/ palatal stops: they are in fact palatal affricates, as minus273 implies.

  45. marie-lucie says:

    minus: You (like many others) assume that the fortition “rule” applied to each set of consonants as a group, rather than individually in stages in a chain where g > k could be followed later by d > t and b > p, just like Romance p > b > v stopped in French at /v/ but t > d > ð > zero stopped at /d/ in Italian but went the whole way to zero in French. Actually, in rapidly spoken colloquial French /v/ often disappears between vowels, as in “vous savez” pronounced [vusae], or “qu’est-ce que vous voulez” [kɛskəuule], thus completing the set of lenitions in the manner predicted by the chains.
    Chain changes are not currently very popular as descriptions partly because they are very difficult to explain in terms of distinctive features, something which to my mind is an argument against the usefulness of distinctive features rather than against chains.

  46. marie-lucie says:

    JC: I was also wrong to call /c/, /j/ palatal stops: they are in fact palatal affricates, as minus273 implies.
    In many languages, such affricates behave like stops in terms of what types of rules they are subject to, and are often listed among stops in phonemic inventories. Often they derive from stops historically speaking, as with the Latin velar stops c [k] and g [g] which became affricates before /e/ and /i/ and are still affricates in Italian.

  47. David Marjanović says:

    I’ll read the paper on weird sound changes ASAP!

    I was thinking of the Nostraticists, Proto-Worldists, et hoc genus omne

    The Proto-Worlders use only “mess comparison”, as far as I know. The Nostraticists don’t, they use the comparative method, complete with morphological comparison and attempts to discover regular sound correspondences. On my harddisk I have a paper about the Proto-Nostratic first person singular marker *q, another about why the PIE second person singular verb ending is *-s rather than the expected *-t (hint: the first-person ending, in the dual or somewhere, is *-H2 rather than the expected *-k), and another that tries to establish regular sound correspondences between the “laryngeals” of PIE and Afro-Asiatic. I guess I should blog about them as “soon” as I finally get around to starting a blog… but, anyway, at least two of them must be openly accessible somewhere online.
    The same holds for Dené-Caucasian, with the difference that Proto-DC is reconstructed as polysynthetic, so there’s more morphology to compare and reconstruct than for the apparently rather Japanese-style PN.

    There is a point I am trying to make here: in comparisons like this, quality is what matters, not quantity.

    Quantity, however, is an important test of quality. Without quantity, you can’t find out if anything corresponds regularly, i. e. with statistical significance. One word won’t allow you to tell that Catalan initial ll regularly corresponds to other Romance initial l; the more examples of this you find, the more robust your hypothesis becomes.
    That’s why disproportionately little can be done with very poorly attested languages (Thracian, Dacian, Illyrian, that kind of thing).
    It’s also why Grimm’s and Verner’s laws are universally accepted now, but Kluge’s is still in dispute – there are few enough examples of it that some (many? most?) think they could all be onomatopoietic. Indeed, there’s not even an article on it in the German Wikipedia.
    Finally, quantity – comparison with more languages – is what shows that the /o/ and /u/ of all those “wolf” words isn’t onomatopoietic: in PIE it simply wasn’t there, the word had a syllabic /l/ instead.

    But, of course, beware the French and Italian superstrates in Romanian, which can be very misleading to the novice comparative linguist. By the same token, it would be easy to set up sound-laws showing that Modern French and Modern English are close relatives.

    Reminds me of the fairly recent paper that shows that Bai isn’t the closest known relative of Chinese – the supposed cognates are just another layer or three of loans.

    Meillet (one of the most famous Indo-Europeanists) said that the only thing he needed to prove that a language belonged to the Indo-European family

    *sigh* “Prove”. “Prove”. For decades now, practically only creationists and historical linguists have used that word. Proof is for math, formal logic, and booze; in science, keep the strength of your belief proportional to the strength of the evidence. Rant over.

    I would say that in Austronesian there are way too many tree proposals, and little consensus on testability. The first-order groups (genera) are obvious enough, but above that chaos reigns.

    Oh. I had no idea.

    I’ve always liked the semantic consonance of the names “PIE” and “PAN”.

    🙂

    Hat: I think part of the problem is that most scholars today who work with non-Indo-European languages have no idea how messy Indo-European data is, because they lack any first-hand experience with Indo-European: what they have been given in their textbooks is a tiny subset of existing data, specially chosen to exemplify perfect regularity.
    So of course they are puzzled and panic when they realize that the real linguistic data they collected themselves isn’t like this at all.

    The other side of this could be that the Moscow School people have plenty of experience in IE, go on to write etymological dictionaries of Altaic or (North) Caucasian that are as comprehensive as they can make it, and are then criticized for including plenty of comparanda that are attested in few subbranches or have unexplained irregularities. Yet, there is no full explanation of the /h/ or the /i/ in Ancient Greek hippos “horse”, or at least there wasn’t when Don Ringe blogged about it four years ago.

    “A language family is either obvious, with a rich skein of correspondences … or it is forever unknowable”

    Contrast Allan Bomhard (whose work has its own potential problems, but never mind):

    One final point: How “scientific”, how “rigorous” is our discipline when all of the strictest methodologies and some of the most brilliant minds can only solve the EASY problems. Families like Indo-European and Uralic are fairly transparent. Even educated laymen can see the relationships here — indeed this is how it all got started (Sir William Jones). Are our methodologies so flimsy and is our imagination so impoverished that everything falls apart once we reach an arbitrary threshold of 5,000 years B.C. (which just happens to coincide with the most commonly proposed date for Proto-Indo-European)? Where would we be if Biology [sic], for example, were similarly constrained? We can (and must) do better.

    vowel ablaut (as in English sing/sang/sung/song and similar alternations in German)

    Great opportunity for a digression: sing/sang/sung line up perfectly with German singen/sang/gesungen*, but “song” is Gesang, and there’s no form with o. Now, I’m sure it’s understood in detail what happened here, but I’m also sure it took decades to figure out**, and I’m sure it helped a lot that English and German have been written for well over 1000 years now.
    * Footnote to the digression: this would be harder to see if only Upper German dialects, like mine, were attested. In those, the passé simple has died out, and the subjunctive of “sing” has been regularized, so we could only match 2 out of the 3 with English, and the third would be “whoops, missing data, how convenient”.
    ** A short look at Wikipedia shows that the history of the English vowels is frightfully complicated. The German ones are easier, at least as long as you stay in the standard, but not completely trivial either.

    What happened is that /v/, /j/, and /g/ devoiced intervocalically to /f/, /c/, and /k/. Note that /j/ here is the voiced palatal stop.

    …So, were there preexisting /f/, /c/ and /k/, or did /v/, /ɟ/ and /g/ simply drift into unoccupied space?
    …uh. I said I’ll read the paper. Never mind, then.
    Anyway, intervocalic devoicing has a precedent. The High German consonant shift turned /β ~ v/, /d/ ([ð]?), /ɣ/ into voiceless [b̥], [t], [g̊] in all positions, and /z/ seems to have been gotten rid of somehow during this process, too (as far as I can tell, it was randomly redistributed between /s/, always voiceless, and /r/). The long consonants /bː/ and /gː/ were completely fortified, too*. When it was all over, Sufficiently High German had no more aspirated consonants, no more voiced obstruents, and no more long lenes. See, folks, that’s how to do laziness right. =8-)
    * Somehow that’s never mentioned in the sources I’ve seen, though I do have to caution that they don’t range up much higher in technicality than Wikipedia. For /gː/ to /kː/, there are lots of examples like bridgeBrücke, midgeMücke, ridgeRücken (usually “back”, but still “ridge” in geography) and so on. For /bː/ to /pː/, I’ve only found two yet: Rippe “rib” (19th century poetic Low German Ribbe “breastplate”), Sippe “clan” (obsolete) – sib- as in sibling.

  48. David Marjanović says:

    they are very difficult to explain in terms of distinctive features

    What does that mean?

  49. David Marjanović says:

    there are few enough examples of it that some (many? most?) think they could all be onomatopoietic

    Uh, sorry, generally expressive, not necessarily specifically onomatopoietic.

  50. marie-lucie says:

    DM: Thanks for the many points. A few comments:
    Prove: I know that the word is not the most suitable, but I was writing fast. I am not sure what words Meillet used in French. I agree with you here: in science, keep the strength of your belief proportional to the strength of the evidence. We can rarely “prove” anything, we can only accumulate support for a hypothesis.
    Quality vs quantity:
    Without quantity, you can’t find out if anything corresponds regularly, i. e. with statistical significance. One word won’t allow you to tell that Catalan initial ll regularly corresponds to other Romance initial l; the more examples of this you find, the more robust your hypothesis becomes.
    I agree. This is why I am very dissatisfied with people looking at dozens if not hundreds of languages, who limit themselves to Swadesh lists of 100 “most common” words. These lists are generally reliable for very closely related languages, where you find lots ot cognates, but with more distantly related languages you might find more or less realistic correspondences (eg p : f or l : r) in several of the words, but only one or two (at best) instances of less obvious correspondences, which a linguist will probably discard or even not notice because of their apparent implausibility either phonetically or semantically.
    ablaut: English sing/sang/sung/song and similar alternations in German
    Note that I said “similar”, not identical. Most English alternations have fewer than four such forms (and some have none), but the principle of vowel alternations for morphological purposes is what I was focusing on, not the exact details or the derivation from Proto-Germanic.
    – Bomhard quote: I am placing this last because I think it deserves repetition, although I too have reservations about B’s work:

    One final point: How “scientific”, how “rigorous” is our discipline when all of the strictest methodologies and some of the most brilliant minds can only solve the EASY problems. Families like Indo-European and Uralic are fairly transparent. Even educated laymen can see the relationships here — indeed this is how it all got started (Sir William Jones). Are our methodologies so flimsy and is our imagination so impoverished that everything falls apart once we reach an arbitrary threshold of 5,000 years B.C. (which just happens to coincide with the most commonly proposed date for Proto-Indo-European)? Where would we be if Biology [sic], for example, were similarly constrained? We can (and must) do better.

    Many people claim to follow “the most rigorous principles of the comparative method”, but they mean they are following the letter of the method rather than its spirit. The great pioneers of the CM were allowing themselves to ask “what if?” and to follow the consequences through rigorous methods, something that seems to be anathema to many of their modern “mainstream” followers, who are content to be technicians, not thinkers.

  51. marie-lucie says:

    m-l: they [chain-shifts] are very difficult to explain in terms of distinctive features
    I mean that usually the individual steps in the chain of sound changes are easy, eg p > b = [- voice] > [+ voice], but with a chain, at least one different feature change is involved with each step, eg for the following step b > v = [- continuant] > [+ continuant], and there is no logical reason why these particular features are changing or why the changes are occurring in this particular order. Nor is there any explanation why the various changes subsumed under “lenition” or “fortition” (older, traditional concepts unrelated to that of distinctive features) may eventually lead to the complete disappearance of a sound (totally negating all of its features at once).
    (I stopped worrying about distinctive features years ago, so it is possible that some of my technical terms and concepts are outdated).

  52. David Marjanović says:

    So, the paper says on p. 244:

    It is true that pre-Kiput had no intervocalic f or c, and apparently had very few examples of intervocalic k, and that this differs from the labial and dental obstruents, where voicing contrasts *p : *b and *t : *d were found in intervocalic position.

    That’s good enough for me: */v/, */d͡ʒ/ and */g/ devoiced because they could, and /p/ and /t/ didn’t because they carried too much functional load. In contrast, in High German, /p t k/ got far out of the way of the devoicing /v d ɣ/ by turning into [fː sː xː].
    Table 6, incidentally, says that Proto-North Sarawak had a plosive system like PIE with voiceless, voiced, and voiced aspirated plosives, and the development of PNS prenasalized voiceless plosives to Kiput long plosives is reminiscent of the correspondence of a nasal infix in Beja to lengthening of the medial consonant elsewhere in Afroasiatic.
    Finally, I’m surprised that the “figure” is called a figure and not a table…

  53. David Marjanović says:

    Thanks, m-l.

  54. marie-lucie says:

    order of changes in chain-shifts
    I think I expressed myself poorly in my previous paragraph responding to David: while a chain-shift evidences a “scale of phonological strength” such as p > b > v where [p] is the strongest element and [v] the weakest in a labial series, weakening changes (instances of lenition) must begin with a weak element, not the strongest on, otherwise the result would merge [p] with [b], then [b] with [v], etc, so that all the consonants in a series would end up as one (and potentially zero). Inversely, fortition (or reinforcement) must begin with the strongest element.
    These concepts are useful in historical linguistics but (to my knowledge) have not extended to current phonological theory.

  55. marie-lucie says:

    Correction: Etienne: Spanish LOBO, Rumanian LUP, etc. – m-l: In Occitan too “wolf” is LUP
    It is true that Oc ‘wolf’ is [lup], but this is not how the word is commonly spelled. According to the French-derived older spelling it is loup (written as in French, but sounding the final [p]), and in the graphie occitane approximating the troubadours’ spelling it is written lop.

  56. David Marjanović says:

    Ah. Thanks again, m-l. 🙂

  57. Kluge’s law seems to be the victim of reverse consilience: it explains a great deal, but there are so many weak arguments against it that conservative IEists just ignore it: no non-smoke without non-fire, or something.

  58. David Marjanović says:

    Could be.

    I updated the Wikipedia article a few months ago, but got stalled somewhere in the middle and haven’t dared look at it since then. 🙂

  59. David Marjanović says:

    Correction to what I said about the High German consonant shift almost 2 years ago:

    and /z/ seems to have been gotten rid of somehow during this process, too (as far as I can tell, it was randomly redistributed between /s/, always voiceless, and /r/)

    /z/ was lost much earlier, and always merged into /r/; that’s a common West Germanic innovation. I fell for the many cases where different Verner variants were preserved in different languages, like English hare : German Hase or English freeze, froze, frozen : German frieren, fror, gefroren.

Speak Your Mind

*