Darwin-L Message Log 8:15 (April 1994)

Academic Discussion on the History and Theory of the Historical Sciences

This is one message from the Archives of Darwin-L (1993–1997), a professional discussion group on the history and theory of the historical sciences.

Note: Additional publications on evolution and the historical sciences by the Darwin-L list owner are available on SSRN.

<8:15>From BENEDICT@VAX.CS.HSCSYR.EDU  Thu Apr  7 08:35:34 1994

Date: Thu, 07 Apr 1994 10:33:14 -0500 (EST)
Subject: Distance method for linguistics
To: darwin-l@ukanaix.cc.ukans.edu

  I've been thinking about distance methods for areas other than biology, and
I've come up with the following idea that might work in linguistics. If
someone wants to (or has) tried it, post a response.
  The goal is to get a mass comparison of two languages without being bogged
down by grammar, vocabulary and phonetics. That's not quite possible, because
without some consideration you'll be comparing arbitrary segments of human
speech. In biology, physical chemistry takes care of aligning "same" thing in
different organisms, and we need a linguistic equivalent to physical
chemistry. So here's the proposal.
  First, construct a lot of simple sentences (child speech) that talk about
fairly universal things - environmental constants like night, day, sun, rain,
clouds; human constants - basic anatomy, gender, physiology; social constants
- biological family relationships, generic activities like singing and
talking, and whatever else you decide. Part of the experiment will be to
decide how much you need. Examples:  Two men are sleeping.  I am thirsty.  My
mother gives me drink.  His wife is singing.
  Second, you need to record (in a constant voice and rate) each of the
sentences in a fixed order. It might be helpful to use a computer to generate
the records, to minimize variation due to voice and timing.
  Third, now do spectral correlations between the different recordings. The
correlation coefficients are the distance values. A correlation between
phonemes might also do the job without the requirement for constant voice and
rate, if you can devise an a priori value to measure the differnce between
phonemes. And decide on a "universal" system of phonemes (hope this is the
right word - I mean a basic unit of sound in speech).

Note that this proposal detects two types of change without distinguishing
between them - phoneme shifts and grammar shifts. But it does yeild a mass
comparison. An interesting subquestion would be - how extensive should the
sentence list be to give meaningful results.

If I were doing this, I'd start with something like Romance languages and if
it produces anything reasonable, see how different things have to be before it
"blows up" if it ever does.  But I'm not a linguist, so the idea is free for
anyone who wants to give a try (if it's great, I'd appreaciate at least
partial credit - if it's a disaster, forget I had anything to do with it).

Paul DeBenedictis
SUNY Health Science Center at Syracuse

Your Amazon purchases help support this website. Thank you!

© RJO 1995–2022