Re: Armenian, Sumerian, Burushaski, and Turkic languages

In article <1180365098.014677.199440@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
Darkstar <darkstar100@xxxxxxxx> wrote:

First of all, I'll have to thank you (and Peter T. Daniels) for
indicating the importance of finding regular changes. My only
objection is that you seem to be setting the standards too high

I didn't set the standards.

- no
regular correspondences have ever been found within some of the well-
established (?) families, like Afro-Asiatic, for instance.

Nonsense. There are certainly plenty of known regular correspondences
within Afro-Asiatic. Do you honestly think that scholars look at
Berber and Arabic and say "oh, they look the same", and leave it at
that?

If I give
you (say) Dahalo, would you be able to indicate in what way it's
regularly related to Hebrew?

If I knew the relevant facts, then yes. If two languages have been
shown to be related, then that means (some subset of) their common
ancestor has been reconstructed and (some of) the sound changes in
their history have been identified.

Even in the main Omo-Cushitic group there
might be many inconsistencies. And the main IEL's are not entirely
unproblemetic, either.

Note that due to sporadic language change (borrowing, analogy, etc.),
regularity can be masked. That doesn't mean the regularity isn't
there (though it does make it harder to identify).

But if you say that finding regular laws is absolutely crucial, I'll
probably have to agree. The only question is how regular these laws
are supposed to be? Can you provide a perfect example of "good laws"?

Read an introductory textbook on historical linguistics. Trask's and
Campbell's are both excellent and reasonably priced (~ US$30 each).

If you use the old IE laws for this matter, it just might turn out
that many generally-accepted families simply don't live up to this
standard, and by this reasoning will have to discard Afro-Asiatic,
Austro-Asiatic, Austronesian, let alone Altaic, or most Amerind
families.

What IS the standard that this kind of work is supposed to comply to?

There is no set standard. You publish your theory, and people either
accept it or don't. Further work may or may not be done and new data
may or may not come to light that support or detract from your
analysis.

Suppose, I give you some laws, and you or someone else points out a
few errors and debunks the whole idea just by saying "these
correspondences are not regular enough, period". So which ones are,
then?

Well, certainly one example of a correspondence is not enough. A
hundred is probably enough. Somewhere in between is the boundary, but
the boundary is not a perfect delineation between a proven analysis
and an unproven one. It's fuzzy, and different people will have
different perceptions of what is "enough" (and the boundary will be
different depending on what else is going on in the languages).

If you must have a strict, mathematical answer, you could always
construct a computer model of the languages in question and determine
how likely it is that the correspondences you found could have arisen
by chance. If there is only about a 5% chance or lower that what you
have found could have been an accident, then you're usually considered
safe. Given how lax your methods are (no systematicity, no regular
sound changes, large variance in what counts as "similar"), you're
going to be well over 5%.

Another metric you could use is relative to other analyses. You only
need to be better than all competing analyses to survive. Since the
currently accepted analysis of Armenian is significantly better than
anything you have come up with, you are probably barking up the wrong
tree by including Armenian.

You claim erku, alto, and iki as cognates. You have a vowel
correspondence set of e~a~i. In other words, you don't care what the
vowel is, as long as it exists.

My counter-argumentation for this particular example will be as
follows:

(1) Vowels changes are quite rapid and unpredictable, but I do take

This is not a linguistic universal (cf. Polynesian), so you cannot
just assume it a priori.

Even in language families with lots of vowel changes, they are not
necessarily rapid nor are they truly unpredictable. You simply don't
find spontaneous unconditioned changes like a > ü. What you do find
are (step-by-step) sound changes constrained by phonetic naturalness.

them into consideration in these few instances (at least partially).

(2) There seem to be no other languages in Eurasia which have
reconstructed VCCV-structure for "two", and those who do have are
located in Central Asia (the above mentioned).

Your suggestion that this is somehow relevant and important means that
you believe either that the modern languages have all inherited a VCCV
form or that they have a shared innovation that created a VCCV from
something else. Are you really prepared to defend either of those
claims? Do you have any other evidence whatsoever besides this one
word?

And of course, since Armenian "two" is strongly believed not to have
come from a VCCV shape to begin with (since it can be derived by
regular systematic sound changes from a PIE form that was not
vowel-initial), it should not be included in this analysis, until you
can come up with an analysis of Armenian's history that is better than
what is currently accepted.

Since there is a law of spirantization in modern Turkic, it could be
natural to suppose that it was acting already on the Proto-Tukic
level, so the final reconstruction as **/irkI/ < */ishkI/ seems
plausible.

Do you have any other evidence besides this one word for this
analysis? Where else in Turkic do you find S > r?

(3b) /altV/ in Burushaski (the exact ending depends on class)
Therefore, we have "erku"-"altV"-*"ichkI/irkI", which may be quite
regular, except for the /t/ in Burushaski, which can possibly be
explained as something like *arku > *arthV > *altu conversion in Proto-
Burushaski.

Do you have any other evidence besides this one word for this
analysis? Where else in Burushaski do you find k > t and r > l?

Note, that I'm not using any rare or unimaginable sound changes in
this example.

That's not your problem. Your problem is that you come up with sound
changes to explain *one word at a time*, with no eye for
systematicity. Sound changes are regular, affecting *ALL* words that
they could apply to. If you posit an unconditioned change of S to r
for Turkic, then *every* instance of S in Turkic should change to r,
and you'd need a damn good explanation for why there are still some
instances of S floating around.

You have labials randomly corresponding to other labial: m~m~b, m~b~m,
v~b~m, b~b~b, p~b~u.

Ditto for dorsals: k~g~x, k~q~q, k~kh~q.

But bilabials-to-bilabials and dorsals-to-dorsals seem to be regular
to me for the purposes of a quick-and-dirty analysis!

"Quick and dirty" is not scientific.

Especially
considering the fact that /q/-to-/k/-to-/x/ and /b/-to-/m/-to-/p/
transition is a natural event in Turkic,

Turkish has a full contrast between b, p, and m in most positions.

which are known to be highly
convergent in most other respects! Presumably,

You presume too much and analyze too little.

there were just 3 or 4
phonemes: */q/ for a weak dorsal, /k/ for a strong one, and */B/ for a
weak bilabial.

What do "weak" and "strong" mean?

Voiced/unvoiced opposition of these phonemes was
irrelevant in Proto-Turkic.

What is your source for this claim? Everything I have seen on
Proto-Turkic has posited both p, t, k and b, d, g.

In addition, you allow for semantic variation (father/man/person,
blood/meat, sun/fire/burn, etc.), which exponentially increases the
chances of finding accidental similarity: with your lax criteria of
(a) unpatterned similarity within (being generous optimistic) 5
similarity categories and (b) semantic variation up to (being
generously conservative) 2 meanings deep, any given word has a chance
of having an apparent cognate of 1-(1-1/5)^2, or about 36%.

I allow for semantic variation only when it's semantically
appropriate. I do not compare "water" to "fire". Take any good
etymological dictionary of any language, there's lots of ALLOWABLE
semantic variation.

Of course. Semantic drift is a perfectly valid type of language
change. But allowing for semantic drift (which you *have* to allow
for) increases the chances of finding accidental similarities, making
regularity and systematicity all the more important.

In reality, you have only about 3 similarity categories and allow
semantic variation up to three meanings deep. This results in a
chance of similarity of 1-(1-1/3)^3, or about 70%!

That's an exaggeration.

It's one of the extreme ends of your method, yes. But given how you
are operating, for some words (like "man"), there is about a 70%
chance that you find a cognate in any randomly selected language,
because you are allowing yourself to look at not just words for man,
but also father and human, and you are only considering similarity
when the first consonant matches in place of articulation.

And of course, there are vastly more possible semantically related
concepts, which ultimately means that, if you look hard enough, you
are essentially guaranteed to find a pair of words with man-related
meaning that look "similar" in any two languages you look at.

Just consider what's available in English for "man": human, father,
leader, chief, adult, person, king, and dude are all only one or two
semantic steps away from "man". That gives us the ability to count as
apparent cognates any word with a comparable meaning that begins with
any of m, h, f, l, tS, Ø, p, k, or d, or any phonetically similar
sounds (b, v, r, dZ, S, Z, s, z, g, t)!

The demonstratives (this/that) easily rise and fall (e.g., in
Romance), and some languages don't distinguish between "this" and
"that" with single morphemes (e.g., French has ceci and cela, which
are transparently bimorphemic: ce=this/that, ci=here, and la=there).

You might be right about "this/that". Though, the problems you've
mentioned do not preclude the reconstruction of *to-/ta in PIE.

Of course, because we have systematic correspondences that help sort
through the mess (and an abundance of written records not often found
for other languages).

Nothing is absolutely stable. It's just that the grammatical items
exhibit more semantic stability.

But the words you claim are stable have changed in just a few thousand
years in the Romance languages, so why would you suppose without
evidence that they remained stable for a longer period of time in a
different language grouping?

As for "water", you're wrong here too. There are at least two
distinct IE cognate sets for "water": PIE wed (English water, etc.)
and PIE akw (Latin aqua, etc.).

That is just one well-known exception for "water". Any others in any
other languages?

I have no idea. Without studying the languages in question, I
wouldn't want to make claims about whether their word for water was
stable or unstable. There's no reason to suppose either way without
evidence, because both are possible.

If
proto-language had many affricates, clusters and vowels, we couldn't
expect much correspondence. But if it had almost pure ptk-structure,

"Pure ptk-structure" is pretty rare in the world's language families.
Off hand, I can't even think of a major language family that doesn't
have at least one member language with doubly-articulated consonants.

There seems to be some kind of misunderstanding of my argumentation. I
only said that the a proto-language that has unstable phonemes will
yield more divergent offsprings, then a proto-language with a simple
and robust ptk-type of structure (consider Japanese as an example of
such).

Japanese has affricates.

Aside from the fact that you often get details wrong, I notice that
you have the tendency to point to a single example as somehow
exemplifying a general rule. Even if Japanese had a "ptk-structure"
(which it doesn't) and a stable history (which it doesn't) does not
mean that "ptk-structured" languages are generally stable!

3) Prothetic phenomena and other types of inclusion are rare.

Hardly! I'd be rather surprised if there was any language in the
world that had no epenthesis at all in its history.

You're applying the same faulty logic that you accused me of. "Any
language with all of its phonemes in all of its history"! Yes, /h-/
may indeed be on and off sometimes in some languages (I've already
mentioned the Yasin dialect of Burushaski). But GENERALLY and
STATISTICALLY cases of pure random inclusion of some random phonemes

I'm not talking about "random" sound changes! I'm, as always, talking
about *systematic* sound changes.

are rare. For this reason, explanations of some kind of epenthetic -k-
in "duos" with a simultaneous loss of such statistically stable
phoneme as the Anlaut /d-/ to explain the origins of "erku" may not
work.

The sound changes deriving Armenian from PIE are not random or
arbitrary. They are regular across the language and phonetically
natural.

I said nothing about infinity. You only need a few attested sound
changes to go all over the map of the consonant system.

Suppose your proto-sound is *t. Then one daughter language could have
the development t > k (like Hawaiian), while another could have t > ts
(like German).

From these, you could have further developments such as k > h (like
Germanic) and ts > s (like Pipil), resulting in a possible
correspondence set of t~k~h~ts~s after a maximum of two sound changes
per language.

Now, just add in a few more changes, and the diversity of possible
correspondents is staggering. Consider the sound changes h > O
(Romance), O > v or j (Slavic), O > ? (German), s > r (Latin), and r >
l (Samoan). This gives us the 11-member correspondence set
t~k~h~O~v~j~?~ts~s~r~l, all from no more than four sound changes in
any one language.

If this sort of reasoning were right, no proto-language reconstruction
would be possible. You seem to be overdoing it.

Proto-language reconstruction is possible with *SYSTEMATIC*
correspondences. When you ignore systematicity, the wide variety of
possible outcomes is what makes the analysis impossible (or more
accurately, indistinguishable from random chance).

The point is that you can't just rely on similarity alone, because
even for real language change, similarity is not necessarily the
result. But when they are dissimilar, they are *systematically*
dissimilar.

Four sound changes in one language's history is a miniscule number in
the time range needed to find a common ancestor for Armenian and
Turkish.

I use only NATURAL, ALLOWABLE or WELL-EXPLAINED phonological changes,
such /b/ being related to /m/ and /p/ already in Proto-Turkic.

But you don't use them *systematically*.

Consider Old Turkic /biz/, Yakut /bihi-gi/, Khakas /pes/, Salar /
piser/ but West Yughur /mIs/ for "we". I do not use just ANY changes.

But from word to word, you do just use "any" change. For one word,
you use a m~m~b correspondence, and for another, you use m~b~m. It's
completely random which set you use. All that matters to you is that
you have some labial corresponding to some other labial, with no
concern that you're implying multiple sound changes on the basis of
individual words.

False. Sound changes do not happen in regularly-spaced time
intervals. You can easily get sweeping changes in just a few
centuries (the Great Vowel Shift in English, the period of
disintegration in Slavic, etc.).

Aha. Then how many of these "catastrophic", sudden events do we have
in the world language history?

Who knows? But you can't pretend they don't exist, especially with
the time scale and divergent languages you're dealing with.

Nathan

--
Nathan Sanders
Linguistics Program
Williams College
http://wso.williams.edu/~nsanders/
.