*Degrees* of evidential value

[ECOLING at aol.com]

Larry Trask writes, concerning Pat Ryan's example /ama/ 'mother':

>No.  Their ubiquity is *precisely* the reason why they should be excluded.
>After all, in most of the languages on the planet, the word for a cat-noise
>is something like <miau>, but this ubiquity is
>not an argument for pushing the word back to Pre-Proto-Everything.
>Rather, it is a compelling argument for disregarding the word altogether,
>on grounds of *motivated* independent creation.

As in all discussions of sound symbolism and expressives,
as for anything else,
we do not have absolutely perfect knowledge of what causality is.

But we do have the ability to partly solve the *simultaneous equations*
(to use a metaphor or analogy) of these two kinds of factors.

As long as *not all* languages have words for a 'cat-noise'
which are *equally* similar to our English <miao>,
we may be able to make some partial, statistical distinctions,
and we may find that the different types of imitative words
for a cat noise fall into groups which *partially* correlate
with known language families and genetic relationships.
They would then be weak, or very weak (take your pick)
evidence.

Now switch from "cat-noise' to the name of the animal, 'cat'.
There should be a difference of *degree*,
words for 'cat' should show more variety, less uniformity,
than words for 'cat-noise'
(though some words for 'cat' will also relate to <miao>).

***

Now to the generalization from such specific contrast
as 'cat-noise' vs. 'cat':

For *each meaning* which we can identify,
and for which we can gather vocabulary,
we should be able to develop a measure of the *degree*
to which it shows arbitrariness across the world's languages,
roughly the number of different sound forms which occur
in words with that meaning, or better, the degree of difference
between such alternative sound forms for the same meaning,
weighted by independent occurrences (measured in a more
sophisticated manner, which information-theory specialists
can define for us).

Then we can *weight* each concept according to those results,
giving greater weight to those which are more arbitrary.
There is no need to have absolute lines of inclusion and exclusion.
Even expressives and sound-symbolic words
can be of *some small* evidential value in estimating likelihood of
a language A being more closely related to language B
than to language C, or the reverse.

I think almost anyone would grant that,
given A equally close to B and to C on every other measure
developed from vocabulary which is at the extreme of not having
any expressive nuance, we might
heuristically consider the value of expressive vocabulary as well.
Since there is a gradient between non-expressive
and expressive vocabulary, not a sharp line, there is a large
amount of evidence we can use, without throwing it away.

Lloyd Anderson