One Step at a Time

[Muke Tever]

>===== Original Message From Larry Trask <larryt at cogs.susx.ac.uk> =====

>> Are you saying there will be no systematic correspondences to be found
>> among those "half-a-dozen languages"?

>Yes; that's what I'm saying.  Or, to be more precise, there will be no
>systematic correspondences pointing to a common ancestor.  There will be
>all sorts of seemingly common elements, seemingly connected in strange and
>puzzling ways, but those common elements will not pattern in the way that
>we call 'systematic correspondences'.  If we try to apply the comparative
>method to the data, we will get some exceedingly strange results, but we
>will not get any proto-language at all.  We cannot possibly get a
>proto-language, because there wasn't one.

Suppose the parent of a "half-a-dozen languages" was a language that had two
different classes of verbs:  those from a language X that conjugated one way
and those from a language Y that conjugated another way.  (I think there was a
Greek dialect of Arabic that had a feature like this mentioned on the list
earlier?)

The child languages would either:
- retain this odd feature and both classes
- reduce to one class, analogizing or dropping the forms of the other
- lose both classes

(Even in those that followed the latter two paths, irregularities--relics of
the proto-system--might still survive.)

Why shouldn't the comparative method be able to reconstruct the verbs into a
"class 1" and "class 2" based on the evidence of the daughter languages,
similar to the way that, say, gender of PIE words is reconstructed?

Why shouldn't the comparative method be able to take these two classes, and
discover that "class 1" forms correspond regularly to cognate forms in
language X and its relatives, and that "class 2" forms correspond regularly to
cognate forms in language Y and its relatives?

If X, Y, and relatives didn't survive, other clues might lead to disparate
input hypotheses.  Of course, Ockham's Razor might cause these to be
disregarded unless there are other clues, such as differing input
phonologies("Why can 'class 1' verbs have *[?] or *[H9] in them, while 'class
2' verbs never do so?").

It may be implausible for such a language and resulting family to arise, but I
don't see how the comparative method wouldn't be able to handle it should such
a situation appear.

      *Muke!