Proto-Austronesian Phonology

[----]

After a small discussion in the Bizarre Sound Changes thread, I've been plagued with [even more] constant thinking about the historical phonology of Austronesian languages. There are at least a couple more people on this board who are apparently interested in this topic other than myself and I'd really like to discuss this more, so I'm making this thread.

The main topic of discussion that came up is this: what is the value of *j in Blust's reconstruction of PAn? What about *R (and naturally *r as well)? My current opinion after all this thinking is this:

*j was a fricative, either /z/ /ð/ or something else alveolar.
*R was an approximant or was on the way to it at the earliest reconstructable time.

Here's why I say it's this way:
+approximants are prone to elision (explaining why *R disappears completely in so many languages)
+*j as /z/ has neatness benefits: it gives us an /s z/ voicing pair
+it provides a simple explanation for the velar reflexes of *j in some languages: rather than an independent shift to a velar POA, it just merged with *R in most languages which then velarized as expected.

Any thoughts?

[Tropylium]

A recent proposal of Laurent Sagart's is that "*j" should be interpreted as *ɲ. (See also: presentation.)

[Salmoneus]

I'm afraid I don't know more about Austronesian than Blust, so I can't help you much.

However, I would say:

- if you're positing *j > *R > *g, you'ld expect *j>*g only in languages where *R > *g, yes? So how come in Toba Batak *R > *r, but *j > *g? More generally, I don't see the reflexes of *j and *R tending to be the same in that many languages?

- as you say, approximants are weak sounds that are easily lost. So why would, say, an alveolar approximant turn into strong, hard sounds like *s (repeatedly!), or the voiceless lateral? I think Blust's idea of a trill seems to make more sense to me in that respect. A trill is also probably more likely to shift to uvular position (to give *g) than an approximant is, but could still be weakened to an approximant where necessary.

- *j as /z/ looks possible, but not overwhelmingly appealing. It would require changes like /z/ > /?/, /z/ > /g/, /z/ > /d/, /z/ > /tS/, /z/ > /X/.

- more generally, though, AN has clearly shown some weird sound changes in general, and several phonemes have almost all possible outcomes. It may simply not be possible to reconstruct their original values. It's also just possible that we completely misunderstand things, having missed something important. For instance, in getting to Oceanic we have to assume a whole raft of unprovoked sporadic changes: massive sporadic prenasalisation, including in initial position, and then the innovation of a whole new set of labio-labiovelars (and possibly velar labiovelars too), and then some sporadic vowel changes as well.

[Salmoneus]

A recent proposal of Laurent Sagart's is that "*j" should be interpreted as *ɲ. (See also: presentation.)

That looks more attractive. But still, sound changes like /ɲ/ > /s/ and /ɲ/ > /t/ aren't exactly intuitive!

The suggestion is also a bit weak in its arguments, isn't it?

- I don't see "assumes no migration in the whole of history" as a good thing!
- in particular, do we really see the mountainous centre of Taiwan as a cosmopolitan place of (many totally different) innovations, with the lowlands as the periphery? Yes, the lowlands are further apart, but for a culture confident in boats they were probably culturally closer together than tribes in different mountain valleys would have been? don't know much about prehistoric Taiwan, i'm afraid. The periphery/centre distinction also assumes no migration and more importantly ignores the rest of the world. That is, they assume wild innovation a) in the mountainous centre of the island, and b) among the peripheral population who left the island by boat. That doesn't seem to add up, to me!

- we have to assume palatalisation only of one coronal (and whatever *N was!), not others, and only before *iV, not before plain *i, even though this *i is considered to be a vowel in hiatus, not /j/

- we have to assume then that *niV and *NiV undergo the exact same palatalisation in two completely unrelated branches (Malayo-Polynesian and a single language of Southern Tsouic), while not even happening in other Tsouic languages (*niV > /s/ or /h/ in Tsou?)

- we have to assume that Kanakanabu backed /n/ to a velar before /i/? (OK, not impossible... palatalisation, and then merger of back nasals - Kanakanabu has reduced its inventory a lot)

- we have to assume that the /i/ in the hiatus is then lost in all branches simultaneously, except where it is sporadically retained?

- we have to assume that, eg, Atayal turned /dz/ > /g/?

- Blust believes *N is a lateral, not a nasal. Presumably we're not meant to believe *liV > *ñV? So we're to assume that *N was always nasal, but then why does *N keep spontaneously turning into laterals in Taiwan? Specifically, it turns into a lateral in eight of the nine Formosan branches.

- the only solid bit of evidence seems to be that some languages with *ñ in PMP have *n, not *N in Taiwan. But this isn't a killer blow: you could just assume that those words ought to have *n, but that *n > *ñ in a later process in PMP.


So there's some attractive reasoning in that argument, but it's not exactly without puzzles of its own...

[----]

- if you're positing *j > *R > *g, you'ld expect *j>*g only in languages where *R > *g, yes? So how come in Toba Batak *R > *r, but *j > *g? More generally, I don't see the reflexes of *j and *R tending to be the same in that many languages?

Good point with Toba Batak, but let me clarify: I'm not saying *j and *R merge in some middle-proto-austronesian period, I'm saying that they may have merged in several languages independently or in groups (because of the similarity of their hypothetical values in this system)

- as you say, approximants are weak sounds that are easily lost. So why would, say, an alveolar approximant turn into strong, hard sounds like *s (repeatedly!), or the voiceless lateral? I think Blust's idea of a trill seems to make more sense to me in that respect. A trill is also probably more likely to shift to uvular position (to give *g) than an approximant is, but could still be weakened to an approximant where necessary.

I'm not really sure how to deal with this.

- *j as /z/ looks possible, but not overwhelmingly appealing. It would require changes like /z/ > /?/, /z/ > /g/, /z/ > /d/, /z/ > /tS/, /z/ > /X/.

I don't think z > d is really that hard to accept and I think there's plenty of historical evidence that there were medial stages in those other changes. For instance, > g, X can both be explained if we accept that *j and *R merged in some cases. /z/ > /?/ would only have to be accepted in positions where extreme consonant reduction is totally uncontroversial, such as word finally or before consonants. I can't find any examples of *j > tS so I'm not really sure what you're referring to.

- more generally, though, AN has clearly shown some weird sound changes in general, and several phonemes have almost all possible outcomes. It may simply not be possible to reconstruct their original values. It's also just possible that we completely misunderstand things, having missed something important. For instance, in getting to Oceanic we have to assume a whole raft of unprovoked sporadic changes: massive sporadic prenasalisation, including in initial position, and then the innovation of a whole new set of labio-labiovelars (and possibly velar labiovelars too), and then some sporadic vowel changes as well.

I'm aware of the weird sound changes in austronesian but I don't think that makes this such a futile quest. At the very least now we have a place to collect dialogue on these things. Also: the prenasalization and innovation of velarized labials aren't unprovoked at all; it's commonly accepted in Oceanic historical linguistics that the prenazalization in the initial position of many nouns is a result of pressure from the article *na which had been losing its semantic weight for some time. As for other positions these aren't hard to deal with at all because spontaneous prenasalization occurs in many other Austronesian languages as well.

most of the velarized labials were allophonic and remain so even in modern oceanic languages. The few 'spontaneous' cases appear almost exclusively before low or back vowels where velarization is a fairly intuitive change. Anyway, the velarized labial series was pretty clearly marginal at the proto stage; very few instances of them can be reconstructed as distinct phonemes, and several authors have even reconstructed the series as incomplete (without *pw). Some languages have many instances of these phonemes but in all of those cases they're obviously later developments, necessitated by the gradual merging of many other phonemes; e.g. the persistent fronting of vowels in general.

A recent proposal of Laurent Sagart's is that "*j" should be interpreted as *ɲ. (See also: presentation.)

His justification based on geography is tempting but ultimately I find this pretty unconvincing: I don't find it any more parsimonious than my hypothesis, and *ɲ > g is exactly the kind of change he says we should be suspicious of. He claims it is to fill the void left by the previous change g > k in Atayal but again *j as /z/ can explain this just as easily, with much less controversial medial points.

I am willing to throw out *ñ per Blust tho

[Tropylium]

- as you say, approximants are weak sounds that are easily lost. So why would, say, an alveolar approximant turn into strong, hard sounds like *s (repeatedly!),

Being easily lost doesn't mean it's the only thing they do. Perhaps the most common other sound change affecting /j/ is fortition to something in the /ʒ/ ~ /ɟ/ region, and from here, you can easily get to /s/.

A recent proposal of Laurent Sagart's is that "*j" should be interpreted as *ɲ. (See also: presentation.)

That looks more attractive. But still, sound changes like /ɲ/ > /s/ and /ɲ/ > /t/ aren't exactly intuitive!

This looks like a Fully General Counterargument that doesn't actually allow distinguishing between different proposals. Whenever there's a set of reflexes that's as wide open as here, you can always single out something and point at it and go "but X > Y is not intuitive!". Of course it's not; there would be no discussion to have if everything were intuitive.

On other other hand, if you read closely, you'll notice Sagart is actually positing [ɲʑ] and not [ɲ], which he's trying to sweep under the rug as allophony or whatever. But that seems to be relevant in explaining all the obstruent reflexes.

- we have to assume palatalisation only of one coronal (and whatever *N was!), not others, and only before *iV, not before plain *i, even though this *i is considered to be a vowel in hiatus, not /j/

There seems to be an implicit assumption that this *i in hiatus first turned into *j. AFAIK PAN did not have CjV sequences?

- Blust believes *N is a lateral, not a nasal. Presumably we're not meant to believe *liV > *ñV?

Sure we are. Relative chronology is a thing: we can assume *l > *n, then *niV > *ñV.

[Salmoneus]

This looks like a Fully General Counterargument that doesn't actually allow distinguishing between different proposals. Whenever there's a set of reflexes that's as wide open as here, you can always single out something and point at it and go "but X > Y is not intuitive!". Of course it's not; there would be no discussion to have if everything were intuitive.

That seems to be making the theory entirely unfalsifiable, even conceptually. If you won't accept the relative improbabilities of sound changes as evidence, what possible evidence could you have for or against positing this realisation rather than that? How is it meaningful to argue "it can't have been X, because it's unlikely that X would change to Y, so it must have been W. I know that it's unlikely for W to change to Y either, but hell, you can always single out SOMETHING improbable, that's no argument against the theory!" - that's having it both ways at once!

On other other hand, if you read closely, you'll notice Sagart is actually positing [ɲʑ] and not [ɲ], which he's trying to sweep under the rug as allophony or whatever. But that seems to be relevant in explaining all the obstruent reflexes.

Sure, and if you, say, assume [Ng_j], that makes it easier to get nasal reflexes. But it's just pushing part of the required sound change back a stage... [and the best option would then be to assume something like [Nglrnsztkj], that could change easily to any modern realisation just by cluster-simplification...

- we have to assume palatalisation only of one coronal (and whatever *N was!), not others, and only before *iV, not before plain *i, even though this *i is considered to be a vowel in hiatus, not /j/

There seems to be an implicit assumption that this *i in hiatus first turned into *j. AFAIK PAN did not have CjV sequences?

No - but did PMP either?

- Blust believes *N is a lateral, not a nasal. Presumably we're not meant to believe *liV > *ñV?

Sure we are. Relative chronology is a thing: we can assume *l > *n, then *niV > *ñV.

Yes, but going back to falsifiability - if we can use as many intermediate sound changes as we want, I don't see how it remains meaningful to argue about precise original realisations at all.

[Tropylium]

This looks like a Fully General Counterargument that doesn't actually allow distinguishing between different proposals. Whenever there's a set of reflexes that's as wide open as here, you can always single out something and point at it and go "but X > Y is not intuitive!". Of course it's not; there would be no discussion to have if everything were intuitive.

That seems to be making the theory entirely unfalsifiable, even conceptually. If you won't accept the relative improbabilities of sound changes as evidence, what possible evidence could you have for or against positing this realisation rather than that?

Relative improbabilities assessed in the bulk would be just fine, but just saying "these two reflexes look nonintuitive" without any comment on if the others have been now explained better or worse does not seem to be doing that.

(Also note that he does not propose /ɲ/ > /s/ and /ɲ/ > /t/ as single sound changes, he proposes them as reflexes, i.e. likely with multiple intermediate stages.)

if we can use as many intermediate sound changes as we want, I don't see how it remains meaningful to argue about precise original realisations at all.

Maybe not. But what we can probably do is pin things down to likely phonological values within the larger system. Adding a new separate palatalized velar slot would seem to be clearly less parsimonious than assuming that there was a consonant that occupied the already implied palatal nasal slot.

[Zhen Lin]

On other other hand, if you read closely, you'll notice Sagart is actually positing [ɲʑ] and not [ɲ], which he's trying to sweep under the rug as allophony or whatever. But that seems to be relevant in explaining all the obstruent reflexes.

That sounds rather suspiciously like Middle Chinese.

[Nortaneous]

So it does.

That's Sagart as in Baxter-Sagart reconstruction.