Evidence Schmevidence
Normally, in historical phonology, we would prefer to have at least some dozen or so examples for the development of a given proto-phoneme X in language A before claiming that Y is the regular outcome (and, perhaps, irregularly sometimes Z, etc.). If we’re considering a minor conditional development, asking for similar numbers would be maybe unreasonable: as long as no counterexamples pop up and the sound change being assumed makes real phonetic and/or phonological sense, a handful of examples will still perhaps do. Three examples is probably still the minimum generally acceptable, two is pushing it. And as an adage of historical linguistics puts it, one example is no examples: there are no grounds to be talking about a “regular” development if it does not occur at least twice in the data.
— Nostraticist, popping in: “Hold my beer and watch this”
I am indexing A. Bomhard’s dictionary, and it seems that for him the minimum number of examples required to assert a sound change is apparently zero. He has things like entire table rows for reflexes of *ɣ and *tɬʼ word-medially, even though he does not actually reconstruct any examples of Nostratic roots ending in these…
These are still not completely pulled out of thin air. They seem to be generated by assumptions about general place-of-articulation sound changes (e.g. lateral affricates become velars in PIE) combined with general manner-of-articulation sound changes (e.g. ejectives become the “plain voiced” series in PIE). Also, in Bomhard’s view several languages have distinct reflexes for some consonants initially and medially (e.g. Nostratic voiced stops become in Proto-Eskimo initial voiceless stops vs. medial voiced fricatives), and he then goes on to list separate medial reflexes for all consonants in all languages… although some other languages like PIE do not seem to have any initial/medial variation going on. Presto, we now have all the reason needed to assert even completely unevidenced sound changes all over the place.
This kind of logic, with “multiplying” chainshifts together, can work often enough. If I tell you that in Germanic *p *t *ǵ *ǵʰ become *f *þ *k *g, then you can probably guess that also *ḱ > *x… It’s still far from generally reliable however; if I also tell you that in Tocharian *t *dʰ *g *gʷ become /t t k kʷ/, you’d guess that also *d > /t/ — and yet in this case, the actual answer is *d > /ts/.
