On some distinctions not to be conflated, in comp lx

[Alexis Manaster-Ramer]

----------------------------Original message----------------------------
It occurs to me that it might be useful to review some
elementary distinctions (based on a paper of mine which
has been languishing for some time) which have a tendency to be
conflated.

First, it is a simple thing to verify (by looking at the
relevant literature) that different linguistic families (I
now include only ones which are almost universally
accepted) were established in very different ways.
For example, the fact that Celtic is part of IE was
established on the basis of a comparison involving
a rather specific rule of correspondence between certain
morphological elements.  The unity of the Niger-Kordofanian
languages was also established by a comparison involving
(sets of) morphological elements but without the need for
rules of correspondence (some simple notion of similarity
was enough).  The unity of the Comecrudan languages was
established, as I have discussed earlier, by a comparison
of lexical and not morphological elements and without the
statement of a single explicit regular rule of correspondence
(so once again similarity was deemed sufficient to make
what Goddard called a "strong case"). And I believe that
the unity of Tai was established on the basis of lexical
comparisons but using rules of correspondence and not
merely notions of similarity.

Hence we can see that there are (at least) two
features (morphological vs. lexical and correspondence-based
vs. similarity-based) which are entirely independent of
each other (orthogonal alias cross-classifying in more
technical lingo), yielding (at least) four different
methods that have been used successfully and w/o
any objection on record (e.g., no one has ever objected to
the Comecrudan comparison on methodological grounds).

Keeping these distinctions straight may help unconfuse
some of the recent (and not-so-recent) confusions.
For example, Stefan Georg says that the method of
looking at phonetic similarities only works correctly
in the case of languages which are "obviously" related,
which he takes to be the case of the Tai languages to be.
But he fails to make the crucial distinction between
WHAT it is we are looking at to find similarities,
lists of words or rather morphological paradigms.

But as I said you can apply the method of looking for
phonetic similarities to morphology, and then you
find language families which certainly have not been
historically obvious and which in fact have only been
recognized within living memory.

This not to say that I accept his assertion that
comparing lexical items by the same method only works
in obvious cases.  I mean you can make this circular
by saying that any language relationship established
in this way is by definition "obvious".  This will not do.
'Obvious' has to mean something that has always been recognized
and never rejected by people working on language classification.
And I think that by this criterion Stefan is wrong.

Of course, the case of lexical comparison using rules
of correspondence is one where almost by definition
he canNOT be right.  Simply because if the relationship
was obvious why would anyone bother establishing the
correspondences. At any rate it seems to me that Ken
Hale's magisterial demonstration that the Pama-Nyungan
(did I spell this right?) languages are related to
the other languages of Australia relied on the discovery
of an amazing set of rules of correspondence which made
the two groups of language LOOK totally different in
places where in fact historically the relationship was
a close one.  Clearly no one can say that morphology was crucial
in this work (though grammatical morphemes may have been
mentioned), only the rules of correspondence.  Nor will
anyone say it was obvious.  It was to my mind one of the
greatest single feats of comparative linguistics of any age
and one which some still do not quite seem to get.

So we have at least four different methods, and I would
argue that actually there are additional distinctions that
have to be made, e.g., ones having to do with quantitative
issues (with a nod here to Johanna Nichols), such as how
many comparisons (in relative OR absolute terms) are actually
being made.  A lot of people seem to forget that there are
many cases of, especially extinct and poorly attested,
languages whose classification rests on a handful of items.

There are certainly Australian languages known to us only
from brief word lists which no one doubts are Australian.
There are, yes Virginia, there are Indo-European languages
whose Indo-Europeanness was discovered on the basis of
a handful word and/or morphemes simply because that's
all there was.  I don't how many IE-origin items we NOW
know for Messapic or Lydian or Lycian, but when they were
recognized (and I do not mean by one or two visionarieis
but by the whole field) as IE, the number was trivial (se
the index to Pokorny's IE etym. dict., for an easy
and instructive demonstration of this, though of course he
does not deal with most grammatical elements).  The poorly
attested Comecrudan languages are thus NOT an exception.
On the other hand, the relationship between Greek and
Sanskrit involves thousands of lexical AND morphological
comparisons of incredible intricacy and beauty.  But this
does not make Sanskrit MORE of an IE language than the
much less well attested and more divergent Lycian, say.

There is yet another set of distinctions that needs to be
observed in any serious discussion of this whole topic.
Namely, we must recognize that just saying "I have here
a rule of correspondence" does mean that you really have one.
Many of the correspondences (and reconstructions implied by
them) in many diferent areas of comp lx (I would single
out Kartvelian, Dravidian, Newman's Zuni-Penutian work,
much of the work on Uto-Aztecan, etc.) are not really so.
They are merely described as rules of correspondence but
are not treated as such in actual work, in which the
"rules" are ignored.

On the other hand, in addition to making more and more distinctions,
we must also recognize that the distinctions are not always
absolute.  For example, the distinction between a rule of
correspondence and a "mere" observation of similarity is
(or at least often is) one of degree.  Sometimes the rules
of correspondence are so obvious that one does not bother
stating them.  When Goddard compares the form kem 'woman'
across two or three Comecrudan languages, he obviously
assumes that k : k : k, m : m : m, etc.  At other times,
the similarities so-called are simply patterns of
correspondence which are not fully specified.

That is, if phoneme X in language 1 can correspond to
phonemes Y and Z in language 2 and we cannot tell (yet)
when Y and when Z is to be expected, that is not a
sign of the moral turpitude of the linguist who nevertheless
related the two languages.  After all, we know of precisely
such cases in all well-established language families,
with the best-known examples being from IE.  But whether
we say that this a case where we would LIKE to know the
rule or whether we say that we are simply observing
a similarity, the fact remains that we do not know
when Y and when Z are to be expected and yet we CAN
know that the relationship is valid.

Which brings up two more points.  One, "similarity"
is a misleading term, since sometimes the "similarities"
do not involve segments that are phonetically all that
similar.  The so-called similarities are rather patterns
which one notices which are not necessarily more
superficial but which are rather less precise (and less
easy to articulate) than the regular rules of correspondence
which we would prefer but cannot always have.

Two, as Eric Hamp pointed out twenty years ago, and
was right to point out, rules of correspondence are
often (usually, maybe even always) easier to discover
if there is some obvious superficial pattern (including
but not restricted to phonetic similarity).  The
relation of Armenian berem to Skt. bhara:mi 'I carry'
was noted long before that of Armenian erku to Skt.
dva was, and of course Armenian had long been recognized
as IE before the etymology of erku was established.

So, in practice, there are relationships between methods
that LOOK in theory quite distinct.

The moral? Simply that there is not a single comparative
method but a whole complex bunch of methodS and that
it would (if I may quote myself) behoove people to study the
methods and the history of the discipline before trying to
read other people out of the discipline, ruin their careers
and reputations, and stifle the serious examination of their
proposals.  For which one of us can be free in our intellectual
quest and safe in our chosen life if even someone of the
standing of Joseph Greenberg can openly, and without any
challenge from the audience, have most of his life's work
peremptorily dismissed as "not historical linguistics" by
people in positions of considerable influence and sometime
power who appoint themselves censor and whom we as a profession
confirm in that position?

AMR