Re: Armenian, Sumerian, Burushaski, and Turkic languages

On May 25, 8:20 pm, Nathan Sanders <nsand...@xxxxxxxxxxxx> wrote:
In article <1180073501.701765.282...@xxxxxxxxxxxxxxxxxxxxxxxxxxx>,

Darkstar <darkstar...@xxxxxxxx> wrote:
On May 25, 4:46 am, Nathan Sanders <nsand...@xxxxxxxxxxxx> wrote:

I mean "Where language 1 has phone X in environments E and F, language
2 systematically has phones Y and Z respectively".

By your tally, you might not be able to prove the relation between
English and Spanish.

Of course you can, because it's been done. Hint: you don't start by
comparing English directly to Spanish. You compare English to the
other Germanic languages to derive Proto-Germanic, and you compare
Spanish to the other Romance languages to derive Proto-Romance. Then
you compare the proto-languages together.

Trying to derive PIE from English and Spanish would be an epic
nightmare!

It's good to adjust for the margin of recent phonological deviations
by going to a family proto-state. But you're making it look as if
stand-alone isolates couldn't be compared at all.
If English and Spanish had no family realtives, would we never prove
they are related??

Have you tried this sort of experiment with your
students?

Yes, using the method outlined above.

Besides, you're forgetting that I'm using Swadesh lists, which are
highly stable,

There are well-known problems with blind adherence to the Swadesh
lists.

Even if the Swadesh lists were totally viable, with the random
tolerance for superficial correspondence you're using, you're
predicted to classify up to a third of *any* list as cognate between
any two randomly selected languages.

Here's the simplified mathematics: you are only concerned with one or
two consonants in the word. To be generous, let's say you're
consistently looking at the first two consonants and require them both
to be similar between languages. You count consonants as being
similar if they have the same major place of articulation (so p, b, m,
f, and v all correspond). There are essentially only three major
places of articulation (labial, coronal, and dorsal), so each word on
the Swadesh list fits into one of 9 (3x3) categories, which means any
given word has roughly a 1-in-9 chance of being in the same category
of phonetic similarity in two languages (this simplistically assumes
the categories are the same size...). Now, since you often only
consider one consonant, you can have as large as a 1-in-3 chance of
finding similarity that arises randomly.

That is, just on articulatory similarity alone, pure dumb luck causes
your method to have 11-33% *accidental* similarity between languages
that absolutely nothing at all to do with real genetic relationships.

It's a long story. You make it sound as if the Swadesh lists were
tottaly unsuitable for anything. But that greatly depends on *how* you
count the potential cognates. Swadesh can't be blamed if a
comparativist is not able to perform phonological analisys. So the 33%
you've mentioned one can only get if one is tottaly impartial or
indescriminate.

Normally, the method yields about 7-15% of background noise (if the
comparativist is careless enough). Anything above that can be seen as
non-zero correlation.

I've prepared a link with Yoruba-English-Spanish Swadesh lists. There
are a few adjustments and refinements to the classical procedure.

http://www.geocities.com/indo_european_geography/Swadesh_lists.htm

Some comment:

1) In practice, there are more places of articulation than "just 3".
You're forgetting vowels, semivowels, glottals as well as differences
in vibrants, spirants, affricates, etc which all sound different to a
discriminate observer. No one's analysis is as dumb as just taking
vowles, labials, coronals and dorsals in the first syllable. (Though,
even that is supposed to give 25% of random chance coincidence, not
33%.)

2) Moreover, I always adjust for lexical stability. Some lexemes are
particularly stable (I, not, this, that, thou, mother, water, foot,
fauna/flora, numerals), while others can be slangy, compound and
semantically vague (fat, good, kill, cloud, rain). For instance,
"rain" is "falling water" in Turkic. "Tear" is "eyewater" in many
Altaic and many SEA languages. But "water" itself is semantically
stable, because it's not similar to any other things, it's just a
generic name for "something wet". On the contrary, "kill" can be
easily substituted by "croak", "ice", "take out", "beat down",
"shoot", "murder", etc.
Fauna/flora names are also quite stable. Such words as "moss",
"goose", "birch" have very few synonyms (in fact, none) and they are
rarely used so they don't get a chance to change, hence Russian "moh",
"gus'", "bereza" and Spanish "musgo", "ganso", "abedul".
The high stability of numerals can also be demonstrated (by refering
to Rosenfelder's page).
For this reason, I make ordered subgrouping within the main list.

2) The phonological stability and the proto-phoneme stability must
also be taken into consideration. "Cinco-five" case is so distinct
because phonetic monsters like *qwinqwe are highly unstable. But "dos-
too" is similar because pure initial coronals rarely dissipate. If
proto-language had many affricates, clusters and vowels, we couldn't
expect much correspondence. But if it had almost pure ptk-structure,
it can survive much longer. For this reason, if you cannot reconstruct
ptk-type phonemes, but have some vague correspondence in vowels,
spirants, and glotal stops, there's your big red flag.

3) Prothetic phenomena and other types of inclusion are rare.
Therefore, Yoruba "awa" and English "we" might not be cognate, even if
they look so. Long dropouts must be accounted for. Therefore, Yoruba
"eniayn" and English "all" don't look too similar.

4) There is supposed to be some typological similarity in grammar,
otherwise you won't be able to reconstruct grammatical morphemes. For
instance, the absence of classifiers in English indicates that it
would be difficult to relate English to Niger-Kongo. Of course,
grammatical similarity alone proves nothing. But the absence thereof
may cause difficulties when doing comparison.

Etc.

[...]

What you have is, roughly, "Where language 1 has phone X, language 2
randomly has V, W, Y, or Z".

I say, "Where language 1 has phone X, language 2 randomly has Y1, Y2,
or Y3, where X is phonetically related to Y"

That right there is a problem. Many common sound changes
(debuccalization, syncope, apocope, epenthesis, metathesis, rhotacism)
or sequences of sound changes (palatalization of k to tS/ts followed
by simplification to S/s (as in many IE languages); acquisition of a
secondary place of articulation and subsequent loss of the primary
place of articulation (as with Polish *l > w); frication of t to T
followed by a shift in place from T to f due to acoustic similarity
(as in Rotuman); etc.) need not result in obviously related sounds.

In light of the vast range of possibilities that actually occur in
real sound changes even over relatively short periods of time, the
fact that your method can get so many results just with (place of
articulation) similarity should be a big red flag that you're not
doing something right.

Just think for a second about how overwhelming this can actually be:
we know of sound changes that can cause n to correspond to N, N to g,
g to k, k to ts, ts to s, s to z, z to r, r to l, l to w, w to b, b to
p, p to f, f to h, h to ?, ? to t, t to d, d to dZ, dZ to Z, Z to j...
essentially, any consonant in one language could in theory be related
to any other consonant in a different language, given the proper
sequence of sound changes. The classic example of how this can happen
in real languages is the pairs of real cognates English "horn" and
Hindi "singa", or English "five" and Spanish "cinco", which do not
look like cognates at all.

What you're forgetting is that for infinite changes, you will need
*infinite* amount of time. So for a phone to move ten steps away, it
will take ten temporal units, which in practice might be as long as 10
thousand years. In a finite amount of time, sounds don't change
infinitely.

[...]

Finding correspondences
is only part of the challenge; the next (and most important) step is
finding the systematic sound changes that create those correspondences.

Okay. I agree. I do have some laws, they're just not sufficiently
elaborated.

.

Relevant Pages

  • Re: Release a spanish version of Delphi
    ... It is probably right however for DevCo to examine the idea of localizing ... Delphi and the other development tools into languages other than English, ... For the Spanish translation, on the costs side there is the localization ...
    (borland.public.delphi.non-technical)
  • Re: O.T: Linguistics
    ... In English, it comes from Arabic, then Old ... The only languages which I'm aware of off-hand which have more basic ... Mayan languages, though working things out is complicated by the fact ... Spanish colour terms which are, by now, largely nativised). ...
    (talk.origins)
  • Re: OT Completely!!! Spanish? Que?
    ... ATMs, on-line banking, phone help lines, ad nauseum give us ... choice of English or Spanish. ... original languages so they can read scripture in its original text. ...
    (rec.woodworking)
  • Re: Off-Topic, Sharon Stone
    ... Yet you can learn foreign languages in a week? ... I don't find Spanish and Italian to be much different than Portuguese, ... honest I've never heard a Portuguese that can speak English as fluently ...
    (rec.sport.soccer)
  • Re: How many languages dont have "until"
    ... there theories why these sound changes are happening in English? ... At least the English glottal stop ... Sweden-Swedish are tonal accent languages, but Finland-Swedish is not, ... aspiration is dropped in the US, one theory might perhaps be that it is ...
    (sci.lang)