LINGUIST List 14.3453

Fri Dec 12 2003

Disc: Re: Are new language classifications necessary?

Editor for this issue: Sarah Murray <sarahlinguistlist.org>

Directory

1. Martintmrtn5, Re: Are new language classifications necessary?

Dear readers,

This is in response to Yuri Tambovtsev's invitation (Linguist 14.1914)
to other linguists to share their opinions on Angela Marcantonio's new
book 'The Uralic Language Family: Facts, Myths, and Statistics'. 

I found Johanna Laakso's review on the internet, and Ante Aikio's
review on the Linguist List. They are very negative, even more
negative than my own opinion. I agree with Laakso that some of
Marcantonio's arguments are not really impor- tant in determining if
these groups are related to each other. But the main strength of her
book, in my opinion, is her recognition that we need to apply
statistical methods to determine if languages are related. She uses
the statistical method introduced by D. Ringe. It is quite
complicated. 

So first I thought that method would be superior to lexicostatistics,
but later I realized it can have the same weaknesses as
lexicostatistics, like different rates of change, and difficulty of
determining if words are cognate, or sufficiently similar. Though at
least it does not have the problem of determining which synonym to
use. Anyway, her result, for lexicon, was that the Uralic groups,
like Samoyed, Balto- Finnic, Permian, Ob-Ugric, and Saami are just
slightly more similar to each other than what would be expected with
randomly generated words. 

So consequently, since I found her results so surprising, I decided to
try the simple lexicostatistical method. I compared Finnish,
Hungarian, Moksha Mordvin, and Selkup. I found that Finnish,
Hungarian, and Moksha Mordvin are closer, using this method, to each
other, than different branches of Indo- European to each other, and
Selkup is about the same distance from the other 3 as different
branches of Indo-European from each other. 

So why did Marcantonio get such different results with Ringe's method?
One reason seems to be, as Laakso and Aikio pointed out, that when
Marcantonio matched words, and a consonant in one language
corresponded with zero in another, she counted is as consonants not
matching. I think Laakso and Aikio are right, that the consonants
should be matched. For example, Fin. suoni 'vein' is clearly cognate
with Hung. in 'sinew'. She counted it as a match of one consonant
only, but I would count it as 2 consonants matching. 

Also I think she made far too much of an adjustment (on Table 5.3,
page 142) for cognates with different meanings in different
languages. When the meanings are clearly related, the words are
clearly cognate, so there is no need to make such a big adjustment. I
just got an email from Aikio pointing out, that Marcantonio's claim,
that the large number of languages in some low-level branches of
Uralic dramatically increases the probability of false matches (page
142), cannot be correct, because the languages in these subbranches
are so closely related, that the effect would be minimal. So now I
think Aikio is right. 

I am surprised though, that Laakso and Aikio think the control case
equivalent to random words, is not documented in Marcantonio's
book. It seems to me that the tables on pages 142-143 provide that
control case.

-Thomas Martin