--- In cybalist@..., "Piotr Gasiorowski" <gpiotr@...> wrote:
>
> ----- Original Message -----
> From: tgpedersen@...
> To: cybalist@...
> Sent: Thursday, October 18, 2001 12:22 PM
> Subject: Re: Which Manansala? (was [tied] a(i)s-)
>
> >>>> I think it is a good article, but it has some flaws.<<<<
>
> It's a *very* good article, IMO, and the flaws are imaginary (see
below). From your mention of Chinese/Quechua I gather you've read
Rosenfelder's other essay on "Reconstructing Proto-World with the
tools you probably have at home" as well. I recommend it very warmly
to everybody on this list.
>
> http://zompist.com/proto.html
>
> >>>> 1) He wants all comparisons to be done on languages of similar
age. That would rule out all comparisons of Lithuanian with Old Greek
and Sanskrit. It would have to be done on New Greek and Hindi etc
instead. That would invalidate a good deal of IE research.<<<<
>
> Rosenfelder says no such thing. You've been (mis)reading between
the lines. What he says is this:
>
> "It's quite naive to compare individual Semitic languages with
modern Cuzqueño dialect. On the Semitic side proto-Semitic or proto-
Afro-Asiatic should be used; and on the Quechua side, reconstructed
proto-Quechua. We also know some words in an even earlier form; for
instance qocha is related to Aymara qota -- which looks even less
like the proposed cognate gubshu."
>
> Proto-Quechua and Proto-Semitic are not even approximately of the
same age. What the above means is that *if possible*, the earliest
attested or reconstructible forms should be used for comparison to
eliminate recently accumulated noise. This is what we normally do. If
a Proto-Slavic word is securely reconstructed, there is no need to
cite its modern Bulgarian, Ukrainian or Slovene reflexes (except when
some details of the reconstruction have to be discussed or
justified). That would only darken the picture or oblige the author
to add footnotes explaining language-specific changes (which are
familiar to most of the interested readers anyway). Modern Greek
forms are not used if Ancient Greek forms are available, and the same
holds for French or Portuguese vs. Latin, Hindi or Panjabi vs. Old
Indo-Aryan, etc. Branch protolanguages are already "pre-processed"
for more effective comparison.
Thank you for the lecture. As I explained earlier, the reason why I
haven't inserted reconstructed Proto-Austronesian is that I haven't
got around to it yet.
>
> >>>> 2) He does run through a supposed non-valid list of
comparisons between Quechua and Chinese, but he does not provide a
counter-example of running through a supposed valid list of eg. pairs
of words from two IE languages. If he had, his method would be in
> trouble, since he does not take the context of phonemes into
consideration. Therefore, if eg. one of the IE languages palatalizes
before front vowels, he would have no option but to register this as
slack, although we know for a fact that the palatalisation is
> completely conditioned by the following vowel. Thus, if any context-
sensitive rule has applied in the time the two IE languages have been
separate, the outcome of his theory will tend to lead you to believe
that the similarity between the two languages is a mirage.<<<<
>
> Standard comparative analysis is what should be used in such cases.
It's purpose is precisely to detect hidden context-sensitive
correspondences. If you apply it to related languages, you get a
historical reconstruction. If you apply it to Chinese and Quechua,
you get nothing. Rosenfelder does not advocate any
alternative "theory" of his own. What he shows is that a "method"
based on counting eye-catching resemblances cannot be a reliable
procedure at all.
Standard comparative analysis does nothing of the kind. It does not
help you set up rules; you still do that by intuition. What it does
is to provide a means for checking that the rules you have set up do
not lead to faulty results (by checking theit outcome with actual
data.
>
> >>>> The drift of the article (by opposition) is that recognized-as-
related languages will have a low prior probability of similarity (so
that the result, if they are, is significant).<<<<
>
> No, Rosenfelder only claims that the notion of "similarity" is too
informal, impressionistic and subjective to be of real use. Closely
related languages will of course be very "similar" in most cases, but
what's decisive for determining relationship at any historical
distance is consistent correspondences, not lexical lookalikes.
>
> >>>> Also, there is an implicit challenge in your posting: come up
with a set of rules that will reduce the slack in comparing
Austronesian and IE, so that the prior probability of them matching
is low. But as I have shown, if those rules are context-sensitive,
they can't reduce the slack. The whole theory would have to revised
to take context-sensitive rules into account.<<<<
>
> What theory? Sorry, I don't quite follow.
Apparently not.
Example: languages A and B. Vocabularies (five words, let's wlg,
without loss of generality, assume one-to-one semantic correspondence:
A B
c^i ki
c^e ke
ka ka
ko ko
ku ku
We propose one rule: In language A velar stops are palatalized before
front vowels. I think you'll agree this looks OK.
However:
In order to get a set of words into Marc Rosenfelder's Procrustean
theory, we'll have to set up correspondences. In this case:
A B
k, c^ k
Which is where, I assume, you would protest vehemently, that I am
introducing slack where none was before. Not so, I'm just following
Rosenfelder's example.
In other words, what we think of as a set of words that conform to
our linguistic ideas, will, after having gone through this hasty
scheme of Rosenfelder's come out no better than if they had been:
A B
ki ki
c^e ke
ka ka
c^o ko
c^u ku
with correspondences
A B
k, c^ k
in other words, the same. In his zeal to create a Medusa's head to
scare off abominable amateurs, Rosenfelder has created a monster that
will turn any language comparison into stone, including IE (that is,
excluding pairs of languages, the path between which consists of
context-free rules only).
And BTW, you may point out for the umpteenth time that no, no,
Rosenfelder shows that you can't use "similarity" in linguistics. But
I agree, he does; unfortunately in the process he has shown
that "proper" linguistics is impossible too. That's why I called for
a modification of his theory.
> Piotr
Torsten