idea for a workshop at next ICHL

[robert ratcliffe]

Proposal for a workshop at ICHL 20: Making explicit the mathematical  
basis of the Comparative Method



The comparative method is traditionally based on common-sense  
intuitive, reasoning about probabilites, involving statements of the  
type "Such and such a cross-linguistic pattern cannot possibly be due  
to chance."  Where we have abundant language data and large numbers of  
scholars qualified to evaluate it, as in Indo-European studies, this  
is fine as far as it goes.

But as we try to reach back farther into the remote past, non-chance  
similarities become harder and harder to distinguish from random  
noise, and it becomes increasingly difficult to judge whether a  
pattern which one linguist claims to see as historically significant  
really is so or is merely accidental.

This problem most obviously emerges in the case of proposed macro  
families. The Salmons and Joseph volume of a few years back to my mind  
showed one thing and one thing only. Except in cases where a proposal  
is obviously based on major errors in data handling, our current state  
of knowledge gives us no way to judge most proposed macro-families one  
way or the other. The fact that contradictory macro-groupings have  
been proposed and defended is also troubling to a neutral observer.

The problem also emerges when we try to reconstruct generally  
acknowledged families which are more diverse and less well documented  
than Indo-European, such as Afroasiatic. Several detailed proposals  
have been made.  But if we look closely we can see that they  
substantially contradict each other both in the proposed  
correspondences and proposed cognate sets. Here too different  
linguists see different patterns as significant, and the outside  
observer has little way to judge which if any is most plausible.

I think that in order to advance our knowledge beyond impasses of this  
type, it is necessary to make the mathematical  basis of the method  
explicit and objective. We need, for example, to be able to calculate  
a baseline of randomness for a given way of conducting a comparison,  
against which the significance of any pattern found can be evaluated.

In dealing with groupings like Afroasiatic, and much more so for  
proposed macro-groupings, it is difficult or impossible for a single  
individual to control all of the primary data. Collaboration is the  
obvious answer.  But this too can only work if we can agree upon clear  
and objective criteria for comparing and evaluating the data gathered  
by specialists in individual languages. I suspect that designing  
computer algorithms for conducting the actual comparison might be the  
best way forward.

I realize that Ringe and Kessler and others have made important  
strides in this direction. But there is a lot more to do. I would like  
for example to get beyond reliance on Swadesh lists to look more  
realistically at what comparatists actually do, and to get beyond  
evaluation of proposed families to look also at problems involving  
proposed cognate sets or proposed correspondences for widely accepted  
families that have so far resisted reconstruction.

These problems have been on my mind for a number of years. Is there  
anyone else in the world who is interested in them? If so would you be  
interested in participating a workshop on this at the next ICHL to be  
held in Osaka next July? Please contact me before Sept. 12.



Best Wishes,



Robert Ratcliffe

Arabic and Linguistics

Tokyo University of Foreign Studies



(alternative e-mail raml4405 at yahoo.com)





  
_______________________________________________
Histling-l mailing list
Histling-l at mailman.rice.edu
https://mailman.rice.edu/mailman/listinfo/histling-l