So, not to sound like a crank, but... I find a lot of details about reconstructed PIE a little hard to believe

[dɮ the phoneme]

[This is a moderately edited version of a question I posted on the linguistics StackExchange some time ago and didn't get any kind of satisfactory answer to, so I thought I'd post it here]

I am certainly not the first person to notice that a number of features of reconstructed PIE are typologically very strange. The most famous of these is the voiceless/voiced/voiced aspirated distinction among the stops, which is of course the motivation for glottalic theory (though I must admit, I don't really see how it's meant to help). Another famously strange feature of PIE is its vowel system: /i/ and /u/ basically pattern with the nasals/liquids, and apparent /a ā/ can be entirely explained as actually originating from /eh2/ or /h2e/, so the only vowels which are actually necessary to reconstruct are /e ē o ō/. Such a vowel system would surely be extremely unstable, and I find it hard to see how it could realistically even develop in the first place. Less frequently pointed out oddities include the proposed fricative inventory (based on common phonetic values assigned to the laryngeals) and the syllable structure. What kind of fricative inventory is /s ɣʷ χ h/ or /s ʕ ʕʷ/ (depending on your preferred theory)? Why can some fricatives, such as /h2/, be syllabic, while others such as /s/ cannot? The particulars here just seem completely all over the place.

I'm not actually trying to contest the validity of the reconstruction, I understand that it is generally very well substantiated. I'm just trying to get a better grasp on how I'm meant to conceptualize the reconstruction of PIE within the framework of modern linguistics typology. Have these oddities been addressed by any Indo-Europeanists or language typologists? Is there a proposed explanation? Or am I just wrong about what is and isn't a "reasonable" feature for a language to have?

[Anyway, yeah, I didn't get any particularly insightful answers over there, so I was hoping someone here could help me out]

[Pabappa]

some scholars reconstruct a few roots with syllabic /s/, though I dont have them handy right now and all i know is that one of them means breast and begins with /pst/. There may have been some syllabic /s/ in the zero grades of other words, as well.

Some scholars (not necessarily the same as above) would also tell you that your inventory of /e ē o ō/ is too large, and that the true vowel inventory was just /e o/, or just /e/, and there is one scholar in the 1950s who put forth a proposal that some early stage of PIE had no vowels at all, not even a schwa, nor any stress accent.

I think the Wikipedia article on the glottalic theory is an excellent presentation of both the strengths and weaknesses of the theory. I've read it through several times, at first to learn more, but nowadays because it makes entertaining reading to see the pieces come into place and then fall apart again.

https://en.wikipedia.org/wiki/Glottalic_theory

[Nortaneous]

I am certainly not the first person to notice that a number of features of reconstructed PIE are typologically very strange. The most famous of these is the voiceless/voiced/voiced aspirated distinction among the stops, which is of course the motivation for glottalic theory (though I must admit, I don't really see how it's meant to help).

Original glottalic theory was *T *D *Dh = T Tʼ D, which is much less typologically unusual and is arguably expected on areal grounds. But it's hard to defend. The most recent revision instead has *D being an implosive or nonexplosive series, which doesn't add much either.

One problem with the glottalic theory that I haven't seen pointed out much is that the typologically common system is actually Th Tʼ D, at which point you don't need ejectives. I don't know if there's a good reason for reconstructing *T instead of *Th with secondary deaspiration in Indic and Armenian. (Earlier PIE was reconstructed with a separate, fourth *Th series, but this was shown to be secondary.)

It seems like *T and *Dh formed a binary opposition that excluded *D, which looks kind of like the relation between voicing and tone in Yabem - so it's possible that these come from a tonal opposition on the vowel that was otherwise lost. However, if you're reconstructing vowel features anyway, the vowel feature could just as well be breathy voice. I think Roland Pooth suggested something like this at one point, but without the Yabem comparison. (edit: Kortlandt mentioned Yabem)

Another famously strange feature of PIE is its vowel system: /i/ and /u/ basically pattern with the nasals/liquids

cf. Salishan

and apparent /a ā/ can be entirely explained as actually originating from /eh2/ or /h2e/

Mostly, not entirely, unless you're committed to *a being a sign of a loanword into Brugmannian PIE.

so the only vowels which are actually necessary to reconstruct are /e ē o ō/. Such a vowel system would surely be extremely unstable, and I find it hard to see how it could realistically even develop in the first place.

cf. Arapaho, but there are a number of other proposed solutions. Roland Pooth prefers *e *ē = /a ā/ with raising blocked by *h2; others have suggested *e *o = /ɛ ɑ/. I think *e *o = /ə a/ with phonetically fronted schwa as in NWC, and maybe phonetically backed and rounded /a/ as in Hungarian, seems like a reasonable guess (and explains the kwetwores rule); fronting can be Brugmannian, and *a can be some other low vowel. I'm not too concerned about reconstructing a vowel system that was unstable due to either too many or too few low vowels - that also describes English.

Less frequently pointed out oddities include the proposed fricative inventory (based on common phonetic values assigned to the laryngeals) and the syllable structure. What kind of fricative inventory is /s ɣʷ χ h/ or /s ʕ ʕʷ/ (depending on your preferred theory)?

I prefer neither. Kloekhorst argued against assuming voicing for *h3 - I agree. /s x ħ h/ seems reasonable and areally expected.

Why can some fricatives, such as /h2/, be syllabic, while others such as /s/ cannot? The particulars here just seem completely all over the place.

I don't know anything about this but maybe it's secondary - there's a lot of weird vowel reduction at least in Latin.

Is there a proposed explanation?

many

[Ephraim]

Another famously strange feature of PIE is its vowel system: /i/ and /u/ basically pattern with the nasals/liquids, […], so the only vowels which are actually necessary to reconstruct are /e ē o ō/. Such a vowel system would surely be extremely unstable, and I find it hard to see how it could realistically even develop in the first place.

I think this is a case where a phonological analysis that makes sense language-internally is not necessarily the most useful for typology, and external comparison in general.

Within PIE, it may make sense to treat */i/ and */u/ as merely non-syllabic allophones of */j/ and */w/ rather than the other way around (due to morphological behaviour etc).

However, if you’re doing vowel system typology in the style of that thread from the old board, and if you’re concerned about the stability of the inventory over time, then PIE must, at least in my opinion, be treated as having both */i/ and */u/. Not having a phonological contrast between syllabic /i u/ and non-syllabic /j w/ is probably not that unusual cross-linguistically (even having morphological alternations between the two), but for the purpose of vowel system typology, it’s probably not that interesting whether a particular vowel contrast with a particular consonant, the important question is whether they contrast with other vowels in the same environment. Just looking at the short vowels of PIE, most or all modern reconstructions at least include a four-way contrast between *CiC, *CuC, *CeC and *CoC. While it may be possible to analyse the close vowels as positional allophones of semivowels, they still occupy the close part of the vowel space, so there isn’t really room for any "true vowel" to raise in order to make more efficient use of the vowel space.

Now, the case of *a in PIE is another question…

[KathTheDragon]

PIE certainly had a phonetic contrast between *e *a *o, since laryngeal colouring preceded Anatolian's departure from the family, and hence was a feature of PIE. The convention of not writing it is both anachronistic and for morphonological reasons. Pre-PIE, on the other hand, if native PIE *a always comes from laryngeal colouring, almost certainly had *e and *o as low vowels, probably *æ and *ɑ (or *ɑ̄, which lets you further reconstruct *a *ā). This does support seeing *h₂ as just blocking the raising *æ > *e, incidentally.

[Nortaneous]

PIE certainly had a phonetic contrast between *e *a *o, since laryngeal colouring preceded Anatolian's departure from the family, and hence was a feature of PIE. The convention of not writing it is both anachronistic and for morphonological reasons. Pre-PIE, on the other hand, if native PIE *a always comes from laryngeal colouring, almost certainly had *e and *o as low vowels, probably *æ and *ɑ (or *ɑ̄, which lets you further reconstruct *a *ā). This does support seeing *h₂ as just blocking the raising *æ > *e, incidentally.

Doesn't *e look like it was schwa-like? The kʷetwóres rule makes more sense if it was *kʷətwáres than *kʷætwɑ́ræs.

[KathTheDragon]

PIE certainly had a phonetic contrast between *e *a *o, since laryngeal colouring preceded Anatolian's departure from the family, and hence was a feature of PIE. The convention of not writing it is both anachronistic and for morphonological reasons. Pre-PIE, on the other hand, if native PIE *a always comes from laryngeal colouring, almost certainly had *e and *o as low vowels, probably *æ and *ɑ (or *ɑ̄, which lets you further reconstruct *a *ā). This does support seeing *h₂ as just blocking the raising *æ > *e, incidentally.

Doesn't *e look like it was schwa-like? The kʷetwóres rule makes more sense if it was *kʷətwáres than *kʷætwɑ́ræs.

Or it could be *kʷatwā́ras.

[WeepingElf]

Original glottalic theory was *T *D *Dh = T Tʼ D, which is much less typologically unusual and is arguably expected on areal grounds. But it's hard to defend. The most recent revision instead has *D being an implosive or nonexplosive series, which doesn't add much either.

One problem with the glottalic theory that I haven't seen pointed out much is that the typologically common system is actually Th Tʼ D, at which point you don't need ejectives. I don't know if there's a good reason for reconstructing *T instead of *Th with secondary deaspiration in Indic and Armenian. (Earlier PIE was reconstructed with a separate, fourth *Th series, but this was shown to be secondary.)

It seems like *T and *Dh formed a binary opposition that excluded *D, which looks kind of like the relation between voicing and tone in Yabem - so it's possible that these come from a tonal opposition on the vowel that was otherwise lost. However, if you're reconstructing vowel features anyway, the vowel feature could just as well be breathy voice. I think Roland Pooth suggested something like this at one point, but without the Yabem comparison. (edit: Kortlandt mentioned Yabem)

My personal alternative to the glottalic theory is that the *T set was aspirated; everything else as in the standard reconstruction. Thus, the root structure constraints become a combination of a voicing assimilation rule for [+breath] stops (the only part of the system with contrastive voicing) and a dissimilation rule for [-breath] stops. Of course, a *Th-D-Dh system without *T is unstable, and in fact, no IE language has preserved it. Germanic and Armenian closed the *T gap by devoicing *D; while about half of the family deaspirated both *Th and *Dh, resulting in a *D/*Dh merger. The weak spot in this is that Italic, Greek and Indo-Aryan require a deaspiration of *Th but not of *Dh, which looks bizarre, though the deaspiration of *Th would be motivated by the *T gap and the non-deaspiration of *Dh by the feature [±breath] being contrastive here. Hence, and because nobody better qualified than me seems to have proposed it yet, I am still not convinced by this idea!

[dɮ the phoneme]

Since we're all talking about the stops, and bearing in mind that this sort of thing is completely out of my depth: has anyone suggested Th T~D Dh for the stop series? As in... something like modern English + a breathy voiced series? Grimm's law seems more reasonable if T was actually Th, and it's easy for me to imagine really any set of mergers taking place within this system.

[WeepingElf]

Since we're all talking about the stops, and bearing in mind that this sort of thing is completely out of my depth: has anyone suggested Th T~D Dh for the stop series? As in... something like modern English + a breathy voiced series? Grimm's law seems more reasonable if T was actually Th, and it's easy for me to imagine really any set of mergers taking place within this system.

Which looks like a reasonable improvement of my *Th-D-Dh model. Thanks!

[alice]

Is it possible to come up with a reasonable reconstruction of PIE which is *not* typologically weird in any way?

[Moose-tache]

The PIE vowel system looks stupid, but this is a symptom of poor communication, not poor reconstruction. Imagine you’re doing a science experiment, and you need to keep track of your significant figures. Adding 4.000 and 4.000 gives you 8.000, but adding 4 and 4 and getting 8.000 would be blatant misrepresentation of the precision of your data. When we reconstruct vowel ranges for proto-languages, our precision decreases, but typographically we use the same IPA characters. When we say PIE had vowels /e/ and /o/, all that means is that the daughter languages have vowels that could originate in a system where a front vowel and a back vowel are distinguished. No linguist would confidently say that these phonemes map exactly to [e] and [o]. It’s possible that they were realized [æ] and [ɑ], and raised after the development of /a/. It’s also possible that PIE had more vowels that cannot be reconstructed because they merged with other vowels at some point. Also, the idea that /i/ and /u/ were not proper vowels is not accepted as irrefutable by all linguists. Some forms, like the IE word for “son,” can only be reconstructed with vocalic /u/, and so it is pure speculation to say that this is just a sneaky /eu/ in zero grade. Bottom line, what we can reconstruct does not have to be a plausible vowel inventory any more than a nautical painting seen through a fence needs to show a plausible version of a sailboat.

[Pabappa]

Is it possible to come up with a reasonable reconstruction of PIE which is *not* typologically weird in any way?

sure, if by typology you just mean the phonology .... the mid-1800s traditionalists are few in number these days, but theyre still out there .... no laryngeals, which means that sequences like ph₃ must be phonemic, so therefore aspiration is found on both series of stops. And since there are no laryngeals, there must be an independent vowel /a/, which gets us back to the typical /a e i o u/ system, and schwa is used to take care of the laryngeals that occur between consonants.

[KathTheDragon]

Is it possible to come up with a reasonable reconstruction of PIE which is *not* typologically weird in any way?

I don't know if you'd consider it "not weird", but I personally find the Cao Bang theory (Early PIE *t *ɗ *d > areally *t d *dʰ) quite attractive.

[Rounin Ryuuji]

Looking at the Indo-European languages we have now, and the ways in which their earliest attested descendants had diverged, I would've expected the original language must be something rather "strange"; odd languages, even if they're rare, do sometimes exist, so I don't see any reason to suppose Proto-Indo-European might not have been one.

[WeepingElf]

Is it possible to come up with a reasonable reconstruction of PIE which is *not* typologically weird in any way?

Just how many languages are not typologically weird in any way? Almost every language shows some typological peculiarity...

[Talskubilos]

Looking at the Indo-European languages we have now, and the ways in which their earliest attested descendants had diverged, I would've expected the original language must be something rather "strange"; odd languages, even if they're rare, do sometimes exist, so I don't see any reason to suppose Proto-Indo-European might not have been one.

PIE isn't anything like a real language, but a theoretical construct, i.e. a kind of conlang.

[WeepingElf]

Looking at the Indo-European languages we have now, and the ways in which their earliest attested descendants had diverged, I would've expected the original language must be something rather "strange"; odd languages, even if they're rare, do sometimes exist, so I don't see any reason to suppose Proto-Indo-European might not have been one.

PIE isn't anything like a real language, but a theoretical construct, i.e. a kind of conlang.

Fair. It is a model of a language which is not known first-hand. There are probably features from different dialects and time stages in it, which means that it probably was never and nowhere spoken precisely the way we reconstruct it. But it probably gets quite close to what kind of language the IE languages descend from. Yet, the fact that the reconstruction was changed several times since it was first done tells us that we should expect further changes to it as linguists find out more about the IE languages and their relationships - if they did not get it right, say, 100 years ago, why should they have got it right now?

[alice]

Equally: is it just reconstructed PIE, or do other reconstructed proto-languages seem typologically odd in several ways, like proto-Uralic, proto-Semitic, or proto-Vasco-Caucasian?

[Talskubilos]

Fair. It is a model of a language which is not known first-hand. There are probably features from different dialects and time stages in it, which means that it probably was never and nowhere spoken precisely the way we reconstruct it. But it probably gets quite close to what kind of language the IE languages descend from. Yet, the fact that the reconstruction was changed several times since it was first done tells us that we should expect further changes to it as linguists find out more about the IE languages and their relationships - if they did not get it right, say, 100 years ago, why should they have got it right now?

The thing is the classical genealogical tree model is always a simplification, because it doesn't take into account lateral relationships (i.e. substrate/adstrate), and in the case of the IE family, a huge one. Thus the reconstructed PIE is a kind of "Frankestein" whose materials come from several sources.