Principled Comparative Method

[ECOLING at aol.com]

Here is an example of where the "Comparative Method" can be strengthened.
The matter of "recurring sound correspondences".
The standard presentation is pretty close to what Sean Crist just posted
(quoted at the end).

Now let's consider things as a matter of degrees, not a sharp cutoff in what
is and what is not reconstructible.

As the relation between two languages or language families becomes more
distant,
the number of examples of any sound CORRESPONDENCE become
fewer and fewer.  This is a matter of degree.  The reasoning must be
more sophisticated.  We then have more cases in which multiple sound changes
apply to the same word, and we might have cases in which
(to use the completely artificial and hypothetical example)

a)  there are many examples of PIE *d corresponding to PU *z
     (or even IE /d/ corresponding to Uralic /z/, using particular languages).
b)  there are 5 examples of PIE /d/ corresponding to /z/
c)  there are 2 examples of IE /d/ corresponding to Uralic /z/
d)  there is only 1 example of IE /d/ corresponding to Uralic /z/
e)  there are no examples of IE /d/ corresponding to Uralic /z/

EVEN IN CASE (E), IT STILL MIGHT BE CORRECT TO ARGUE
that IE /d/ "corresponds to" Uralic /z/.

The reason is that "recurring sound correspondences"
is actually NOT a cornerstone of the "Comparative Method",
if we mean by Comparative Method what it is in ideal practice,
that is the reconstruction of the series of changes by which
some ancestor language gave rise to descendent languages.
(Calvert Watkins is at least one prominent Indoeuropeanist
who has emphasized this principle.)

Rather, "recurring sound correspondences" is a short-cut
which works in the EASY cases.

What IS a cornerstone of the comparative method is
a PLAUSIBLE sequence of changes from ancestral language
to descendant languages or language families.

So if we had case (d)
and we also had case (d'),
one example of IE /t/ corresponding to Uralic /s/,
we might have more evidence for saying that /d/~/z/ was real.

Or if we had case (d) before a following /u/,
and we also had case (d'')
ONE example of IE /d/ corresponding to Uralic /z^/ before /i/,
(interpret the ^ as a hachek = wedge, upside down)
we would again have more evidence for saying that /d/~/z/ was real,
in this case because we would assume that /z/ became /z^/ before /i/.

And if lucky we might have case (d') above with following /u/,
and also case (d'''),
one example of IE /t/ corresponding to Uralic /s^/ before /i/.

With all four of these examples, we would have a plausible system,
and would in fact have four examples reflecting the same sound changes,
perhaps actually only a single sound change (namely
"affricate apical stops before high vowels, with the result being
the fricative of the same voicing and "reflecting" the vowel quality)
but we need not have more than one example of any particular correspondence
on the surface.

Of course multiple examples of the same surface results DO
make us more confident of our findings.  The point above is that
examples which are not identical on the surface can ALSO make
us regard the hypothesis of a particular sound change as more likely
valid than would be the case if we had only one of the examples
(d, d', d'', d''').  In extreme cases, I would think we could even
confidently hypothesize a sound change whose immediately results
remain NOWHERE visible on the surface, case (e) above.

The plausiblity of the highly specific changes hypothesized above
is sufficiently demonstrated by the occurrence of something like it
in modern Japanese, where we have in the syllabary

sa se s^i so su
ta te c^i to tsu
za ze j^i zo zu
da de j^i do dzu~zu.

Of course as the number of sound changes hypothesized
rises, we still do need an increasing number of examples,
since we don't want to be in the situation of having a new
hypothesis just to excuse each new example of words in
two languages we wish to claim decend from a common proto-form.
We just do not need to have "recurring sound correspondences"
recurring identically in their surface results.

This kind of situation can be seen in real cases
of language reconstruction, and it needs to be catalogued in standard
manuals of the comparative method, so that the short-cut of
"recurring sound correspondences" can be seen for just what it is,
a short-cut in the easy cases.

In fact, all well-trained specialists in comparative reconstruction
DO use more sophisticated tools than mere superficial recurring
sound correspondences.  Their task is easier with more recurrences,
but it is not limited to them, certainly not in principle.

What is considered "plausible" in the more difficult cases cannot be
simply the whim of the individual researcher about what is a "plausible"
vs. not a "plausible" sound change, on a priori grounds.  It must be
based on evidence of some kind.  We cannot assume our manuals
contain examples of all the possible sound changes,
we can only assume that known sound changes ARE plausible,
and that as-yet-unknown ones will always exist.
For beginning linguists, the substitution of /l/ and /s^/ comes as
a surprise, until they learn that these two segments have some strong
acoustic similarities.
Learning what constitutes a "plausible" sound change is a lifelong task.

Since the number of sound changes between a proto-language and its
descendants does increase with time depth,
the number of recurring sound correspondences will decrease,
and we must make increasing use of the more sophisticated tools.

The real task of extending the Comparative Method to deeper time
depths is to make explicit more of these sophisticated tools,
and to CREATE more such tools by discovering what ways of
handling the data are robust across what kinds of intervening changes.

One of them I am quite certain is to develop both articulatory and
acoustic "spaces", relative "distances" between different articulations
and different acoustic effects, so that when attempting to judge
likelihood of cognacy of pairs of words, we can judge similarity
by degrees, not by yes/no dichotomies.  (These will partly depend
on the general typology of the sound systems of the languages
concerned, they will not be completely universal, but they will
also not be completely idiosyncratic.)

So for example, the sound change w>k does not seem "obvious",
yet if we add y>c^ or even y>t, or even if we don't, it is not the
most distant pair of sounds /w/ and /k/ do share some things
(and we do have /erku/~/duo/~"two" with r~d~t and k~w~null before /u/).
We need a measure of DEGREE of similarity between such known
pairs of words, such that Albanian and English, or Armenian and English,
can be found to be related languages.
Throwing up our hands, as some do, and saying
"any sound can change into any other sound" past a certain time depth,
is not helpful.  At the extreme, in an absolute sense, it is quite possibly
true,
but that is the reaction of those who do not yet have the tools to handle
relative-degrees-of-plausibility in many dimensions simultaneously,
rather than simple yes/no decisions or even degrees of plausibility
in a tiny number of dimensions.

Instead of always using the easy method (recurring sound
correspondences), let's discover the more sophisticated tools that
work in the harder cases.

Who is doing that now?

I would like to see forums which can highlight and reward such
productive work.  How about testing methods which can work
to establish PIE based on English and Albanian (see below)?

Lloyd Anderson
Ecological Linguistics

Below are the quotations which quite adequately represent the
normal presentation of the Comparative Method.

>Suppose for the sake of
>argument that there is in fact a genetic relationship between
>Proto-Indo-European and Proto-Uralic (to take one possible example). Even
>if this were true, and given the evidence which we actually do have at
>hand, would we expect to be able to _show_ that there is such a relation?

>Here, the problem _is_ like trying to reconstruct PIE on the basis of only
>modern English and Albanian.

[Lloyd:  So to establish more empirical validities, we need to test
methods to find which highly general methods (not tailored to the specifics
of the English and Albanian case) would give a positive answer
to the question of relation of English and Albanian.  We might have
almost no recurring sound correspondences (though I suspect there are some),
and a large number of correspondences which result from successive
applications
of more than one sound change in different contexts, so not obviously
intuitive on the surface.]

>Even tho external evidence might lead us
>to
>guess that there could be a relation between PIE and PU, the cognations
>are so obscured by millenia of sound changes, loss of old lexical items,
>the noise of new lexical items, etc., that it might well be impossible
>to
>show a genetic relationship on the basis of the available information.

>Suppose, for example, that PIE *d corresponds to PU *z (I don't know that
>there is a *z in PU; I'm making this up).  Even if there were originally
>dozens of words showing this correspondence, it might well be that case
>that only one such pair exists between the small subsets of the original
>lexicons of PIE and PU which we can reconstruct.  Given the fundamental
>assumptions of the Comparative Method, you _can't_ show that the two words
>are cognate when you've only got one example of the correspondence.

>You need multiple examples of sound correspondences to be able to conduct
>the Comparative Method at all; and when the the cognates become as
>rarefied as they are this time-depth, the likelihood of having access to
>an adequate number of examples to work out the relevant sound changes
>becomes proportionately smaller, eventually reaching what for practical
>purposes is an impossibility.

[Lloyd:  The following clause, taken as an absolute, is false:

>You need multiple examples of sound correspondences to be able to conduct
>the Comparative Method at all; ...

It is true only when "sound changes" is substituted for "sound
correspondences",
yet then it cannot be as quickly seen whether we have that in a given
situation.]