Otdfbt

I was reading Wikipedia, and it maintains that it's unusual for a language to have a voiceless-voiced-breathy distinction (without a voiceless aspirated), but that the Sanskrit 4-way distinction is less weird. I've also heard of the glottalic theory from other sources, which is supposed to make PIE's phonology less weird.

I'm surprised that voiceless-voiced-breathy is so weird because it seems like the natural result of a merger in an aspirated-tenuis-voiced-breathy distinction. As I understand it, the difference between aspirated, tenuis, and voiced is mainly voicing onset time with tenuis being between the other two. Isn't it natural that tenuis should merge with one of the other ones? (maybe losing aspiration afterward?) Breathy voiced is so different it makes sense to me that it would stay different.

I suppose it's just an empirical fact that this distinction is uncommon, but does anyone know why it's so uncommon? (It's not just because so many languages are descended Proto-Indic is it?)

It is plausible to claim that the traditional IE system t d dʰ is rare, but it's not well-established. Very often, we don't actually know what the phonetic properties of the sounds of a language are, either because the author describing the language doesn't say anything specific and leaves it to his transcription system to imply what the phonetic properties are, or the author makes claims that aren't and can't be substantiated phonetically. The letters t, d might be used conventionally for [tʰ, t'; ɗ, dʱ, d̤].

If you state the claim in more general terms and not in terms of "exactly these phonetic values", you would probably be correct in saying that systems with more phonatory contrasts among the voiced stops than among the voiceless stops are rare (this implies at least three phonatory stop types). The common cases with three stop series is two voiceless and one voiced, or else three voiceless (in which case it is aspirated, unaspirated and ejective). The functional explanation for this asymmetry is that varieties of voiceless stops are easier to differentiate, because they have higher supraglottal pressure during closure which results in a higher-amplitude release burst, and that burst is perceptually more salient compared to the closure (which is just silence), so it's easier to tell what kind of consonant you heard. If a language were to have two kinds of voiced stops, there would be a tendency to eliminate one of the types because it's harder to hear. Sociolinguistic / areal features cannot be disregarded. Indic languages have hung onto their voicing distinctions, and Sindhi has managed to add a series (implosives).

To add on to all the excellent information already provided:

PIE had some very particular rules about the shape of roots. Most PIE roots consisted of two consonants or clusters, which an ablaut vowel was inserted in between. For example, the root *p-d- means "walk", while *d-ḱs- means "right" (as in the direction).

However, there was an extra constraint:

• No root can contain two plain-voiced stops


• No root can contain both a voiceless stop and an aspirated-voiced stop

This is very very weird. Voiced stops tend to all pattern together, so plain-voiced stops and aspirated-voiced stops acting so different is unexpected.

The glottalic theory makes this a lot cleaner:

• No root can contain two ejectives


• Pulmonic stops must agree in voicing

Both of these are very reasonable-looking rules with equivalents in many languages.

I coudn't say about your argument for naturalness of one particular loss of one member of a four way distinction: voiced non-aspirate/voiceless non-aspirate/voiced aspirate/voiceless aspirate, but the loss of any one of the four would violate a putative principle of phonological systems. And that is the requirement that features represent independently controllable aspects of articulation, in the language of SPE (The Sound Pattern of English). In Speech Sounds and Features, Gunnar Fant calls this orthogonality.

Much earlier, in 1669, the principle was invoked by William Holder in Elements of Speech, when he argued for the existence in English of the velar nasal eng on the grounds that there were velars and nasals already.

It is probably important here to distinguish between what is expected in human languages generally, on the one hand, and what is expected in particular language phoneme systems.

There are two things that are weird about the reconstructed stops. The first feature, the three series unvoiced, voiced, and voiced aspirated, already has been discussed in the other two answers.

The second feature is the strange frequency distribution of the bilabial stops, out of *p, *b, and *bʰ—the voiced bilabial stop *b is the rarest (or even absent). A lack of /p/ would be natural and not surprising, but a lack of /b/ is typologically rare. This leads some researchers to the postulation of a series of ejective stops for Proto-Indogermanic (see Glottalic theory on Wikipedia), but this has another difficulty: Not a single descendant of Proto-Indogermanic has preserved the ejectives, their occurrence in Armenian is secondary and due to areal effects.

The basic issue about the traditional system *t *d *dh is that it is extremely odd to have a series like *dh which combines voice and aspiration, when just aspiration would be enough for contrast with *t and *d. That's the core issue. Old Indian had a fourth series *th that was voiceless and aspirated, where the series *dh made sense. When this fourth series was removed from PIE, then the three series *t *d *dh no longer stand on their feet. Besides the glottalic theory is only a partial remedy to the issue.