Doing historical linguistics (part 2)

[John Hewson]

----------------------------Original message----------------------------
On Fri, 13 Nov 1998, Larry Trask wrote:
 
> On Thu, 12 Nov 1998, John Hewson wrote:
>
> >     Actually there is an algorithm that a Systems Analyst and I
> > developed to produce my _Computer-generated Dictionary of
> > Proto-Algonquian_ (Ottawa: National Museums of Canada, 1993). It may be
> > stated in a single sentence, as on p.iv of the above: From the data of the
> > daughter languages generate all possible protoforms, then sort
> > alphabetically, and examine all sets of identical protoforms collocated by
> > the sort.
>
> Very interesting, but I'm amazed.  What particularly catches my
> attention is the expression "all possible protoforms".  How on earth can
> this label be fleshed out?  How do you know that your program generates
> all possible protoforms?
 
Let me try to answer with a simple example. An /s/ in Fox can come from 6
different Proto-Algonkian sources. The program generates all 6 from every
/s/ in a Fox word. Let us call this the productivity of Fox /s/. The total
of all possible protoforms for a Fox word is based on the productivity of
all the phonemes in the word.
 
Since Algonkian words often contain several morphs, the
number of possible protoforms can be very high. In the first batch we did,
the average number of possible protoforms for every item of input data was
over 20. Given the length of Algonkian words we were able to simplify by
deleting the vowels and working from the consonant frameworks only.
Otherwise the program would probably have bombed, even on the mainframe on
which we were working at that time. Lowe and Mazaudon, by contrast, worked
on monosyllables that had tone.
 
Such systems are for the reconstruction of the vocabulary once all the
correspondences and reflexes of the relevant daughter languages have been
worked out. It saves the worker the donkey work of pounding through
dictionaries looking for items that may not exist. It cannot replace the
linguist, it is merely a tool to aid the linguist. Algorithms may operate
mechanically, but they are not created mechanically. To create them
requires imagination, and to use them requires a professional level of
understanding of the comparative method.
 
 
John Hewson, FRSC                               tel: (709)737-8131
Henrietta Harvey Professor of Linguistics       fax: (709)737-4000
Memorial University of Newfoundland
St. John's NF, CANADA A1B 3X9