I've been recently thinking about a problem concerning Proto-Indo-European (PIE) that I probably should have dealt with a long time ago. It involves an issue that I had been having regarding "word-final voicing" in the Late IE period (i.e. the stages of Pre-IE following Syncope up to PIE itself). I just thought of a nifty idea that brings Szemerenyi's Law aboard for an intellectual thrillride while simultaneously explaining why the thematic vowel behaves as though the nominative singular *-s was voiced /-z/!
Here's the idea that I'm exploring. Perhaps, the "word-final voicing" that I'm perceiving in Pre-IE wasn't word-final at all. Afterall, phonemes don't normally voice in word-final positions at all. It's so rare that the only example of it I've found is in Lezghian but I'm not even sure whether this is even a valid example. It's a typological dilemma. So one way to get rid of this problem is to suppose that the voicing occured word-medially instead. However, this then could only have occured before Syncope. That means that the voicing of the nominative singular *-s, the ablative singular *-od and the pronominal inanimate *-d would then have occurred in Mid IE (MIE).
What does that have to do with Szemerenyi? Well... let's explore this. Let's take the ablative singular *-ód which I've traced back to MIE *-áta [-'atə]. Now, let's say that when the slow reduction of most unaccented vowels to short schwa was taking place (i.e. the event I call Reduction immediately preceding Syncope), compensatory length *was already* being transferred to the previous vowel if accented. So this first gives us *[-'a.tᵊ] with a slightly lengthened accented vowel. This is where the fun comes in: Let's now say that this added length also became perceived by MIE speakers as a sign of a following voiced consonant. The association between longer vowels and the presence of voice in the following phoneme is a commonplace phenomenon, even occurring in English. So now we have *[-'a.d̰ᵊ]. With Syncope, we end up with *-ád̰ [-'a.d̰].
Now that we got that covered. Let's explore the nominative singular which should go back to the postposed deictic *-sa in MIE. If we take an MIE stem declined in the nominative singular like napáta-sa [nə'patəsə] 'grandson', lengthening during Reduction would first produce *[nə'pa.tᵊsᵊ]. Let's now say that the nominative ending temporarily survives Syncope via the Suffix Resistance exception, producing *[ne'pa.tsᵊ]. Then, immediately following Syncope, the remaining vowel of the nominative ending is "clipped" while leaving behind even more compensatory lengthening. Since the preceding vowel is already half-lengthened, it lengthens even more to a long vowel. Hence we get the result we're looking for in early Late IE: *nepāts.
Nifty, nej? I have to still think about this to make sure that everything is consistant logically, but it's getting me excited so far. (And yes, I'm still at Bar Italia as per my last entry, hahahaha!)
(Jun 9 2008) I forgot the IPA wavy diacritic underneath the *d denoting creakiness (as in MIE ablative ending *-ad̰ above). This doesn't affect what I wrote; it's just more accurate since plain MIE *d becomes PIE *dʰ (although I personally now believe that this stop only became "breathy" after the fragmentation of the PIE-speaking community had already occurred).