Glottalic vs. Classic PIE: an asymmetry
I've been building the Indo-European Isoglosses explorer, an online database that — among other things — displays the phonological inventories of the various IE languages and traces the PIE origin of each individual phoneme.
The project has a section that displays the PIE phonological inventory itself. Simple enough, right? Except: which PIE inventory?
There are at least two serious options on the table. The classical model (itself a family of slightly different proposals, but broadly consistent) gives you the familiar system with voiced stops, voiced aspirates, voiceless stops, and — depending on who you ask — voiceless aspirates. Then there's the glottalic model, independently developed in the early 1970s by Gamkrelidze & Ivanov and by Hopper, which reinterprets the whole stop system: the classical voiced stops become ejectives, the voiced aspirates become plain voiced stops, voiceless stops remain as such, while voiceless aspirates get demoted to allophones (whatever that means).
I wanted to build a toggle that lets the user switch between the two. My first instinct was simple: just swap the symbols and relabel the distinctive features and their values: *b becomes *pˀ, and voiced becomes ejective, done. No symbol should move to a different cell, and the table should not undergo a restructuring.
But it turns out the two systems are not congruent in that sense — they can't be converted into each other by a simple renaming of features.
Let's consider a fragment of the PIE phonological inventory:
Now let's try to switch the symbols and substitute the labels. Applying the substitution gives the following: p→p, b→pˀ, bʰ→b and Voice± → Ejective±, Aspirate± → Voice±. The outcome should be the following:
But wait, is the new *b ejective as well? It is not: in both versions of the theory, it represents the plain voiced series.
Then, perhaps, should the table have had the following form?
This solves the inconsistency, but violates my initial constraint that no symbol should move across the table. Indeed, the new *b replaces the classic *bʰ, and should remain in the same cell.
Try as you might, no rearrangement of cells and feature labels will make it work.
The truth is that my constraint is simply untenable: the glottalic theory assumes one ternary feature, rather than two binary ones. The only meaningful way of presenting the data is the following (even if it is far from being clear how such a three-valued feature should be named):
Note that the authors never mention this fact and, in general, pay little attention to the formal correctness of their phonological inventories (why am I not surprised?).