An interesting recent discovery: a couple of papers from Peter Norquest and Sean Downey, on enriching the consonant inventory of Proto-Austronesian. This is not a completely novel idea to pursue, but this time evidence comes from correlating widely separated branches of Malayo-Polynesian (as opposed to just fiddling with the Formosan languages).
The first paper proposes adding three new phonemes entirely, and extending the range where some other previously known consonant distinctions would have remained. The second shows that these details find a good amount of direct and indirect evidence also from Tai-Kadai — as further support for the hypothesis (emerging theory?) that TK is a branch of Austronesian. Some lexical isoglosses extending to the Austroasiatic languages also seem to show some evidence for these contrasts. Apparently a third similar paper is forthcoming as well.
This is already good and noteworthy news for the field of historical linguistics in general. There are further implications still, though. One of the contrasts involved (*t versus *C) has previously been used as a part of the definition of Malayo-Polynesian; while a couple of them are only attested in MP. This means that a separate Formosan branch could now be defined, supported by about as many phonological isoglosses as Malayo-Polynesian is as well (as for lexical evidence, the question now becomes if Formosan-only items are innovations or retentions). S&D also locate an alternate option for the barycenter of the diversity of the family, in Borneo. And though they don’t say quite as much: this would moreover even seem to substantially weaken the idea that the vast MP group is a separate branch of Austronesian at all, and not simply an areal entity.
N&D’s phonetic assignments for their new segments I find far from unproblematic however. They propose splitting Proto-Austronesian *p into *p and *f; *k into *k and *g; and *l into *l and *ɭ. In support of the retroflex value of the latter, they also reanalyze the previously known *C and *j as *ʈ and *ɖ. Also, a previous Proto-Austronesian *g has already been reconstructed; this they propose to reinterpret as *ɢ. Pretty much all of this seems dubious.
- “*f” is reflected as *p in the vast majority of Austronesian languages. This is not a natural direction of development. More typical is *p > *f, or starting from *f, voicing or lenition towards /v/ or /h/. Most of the exception developments still are exactly these, though, or simply /p/ vs. /f/.
The supposed *f > *p has been attested in some cases of loanword adaptation (e.g. Proto-Germanic to Proto-Finnic) and I figure such a change could plausibly be associated with a substrate population adopting Austronesian. But other options may be available as well (some ideas of mine in a moment). Since no detailed argumentation is provided in the papers either way, I would excercise caution and so far label this distinction as *p₁ vs. *p₂.
- “*g” has a similar problem. It is true that *g is frequently lost from consonant inventories without any particular motivation; but the natural direction of such a development is not > *k. Again, the exception cases look quite *g-like, but the evidence of the mainstream representation for the reconstruction still shouldn’t be ignored. Let’s call this distinction *k₁ vs. *k₂, similarly.
- The shunt for dealing with pre-existing *g looks even worse. /ɢ/ is cross-linguistically an unstable sound, prone to lenition to its fricative counterpart /ʁ/. Yet this would have mostly remained stable across Austronesian, while their *g would have widely devoiced and sometimes lenited? If for some reason we ‘d like to reconstruct both as voiced stops, even the opposite assignments might work better…
- “*ɭ” is typically distinguished from *l by turning up as a rhotic instead of a lateral. This suggests to me not so much a retroflex as some sort of an intermediate alveolar liquid; e.g. a tap like *ɾ? The main argument presented by S&D is that several Formosan languages reflect this as /ɭ/, but this does not strike me as convincing since the same languages also show *l > /ɭ/. Retroflexion can be entirely naturally innovated too, starting from a plain *l. I’ll keep at neutral notation and call the distinction here *l₁ vs. *l₂.
- The reinterpretation of *C as “*ʈ” seems to be rooted in a single language, Dhao, where it turns up as /ɖ/. Some other exception case languages have plain /d/, and it might be tempting to assume this was the former value in Dhao as well — if not for how (judging from their copious tables of example words) the language has also both a separate /d/ < *d, and even /ɗ/ < their “*ɖ”. Perhaps the retroflexion, then, is old here after all. Reconstructing a voiceless stop though only seems to be based on how this consonant merges with *t in most Malayo-Polynesian languages… and it’s quite unclear why would this consonant have spontaneously acquired voicing in most of the exception cases? And then there are the Formosan languages, on whose evidence a value *ts has been reconstructed previously. Again, *t₁ vs. *t₂ could be used as a more neutral notation until the phonological problems have been addressed.
- Some phonetical arguments, though less direct ones, are presented for the reinterpretation of *j as *ɖ. This segment is reflected reasonably frequently as an implosive /ɗ/, and S&D remind of a connection between implosion and retroflexion. This is true — but it works both ways. S&D appeal to a change *ɖ > *ɗ in the Cushitic language Honi, but we could just as well use the known change of the Proto-Cushitic coronal emphatic stop *ṭ to /ɖ/ in languages such as Somali, which probably proceeded thru a stage *ɗ. (In fact, I am now suspicious on if the Honi case even is an actual innovation, or perhaps instead an archaism?)
Aside from the direct /ɗ/ reflexes, there seems to be some indirect evidence that could be seen as being favor of *ɗ in particular, too. Some Formosan languages reflect *d₂ as /n/. S&D’s explanation by referring to some limited evidence for nasal/sibilant correspondences, suggesting a possible *ɳ behind these (since when do nasals have any tendency to assibilate?), and then suggesting *ɖ > *ɳ > /n/ in these Formosan languages sounds incredibly ad hoc. Instead, *ɗ > *n would work as a unproblematic single-step explanation, rooted in the sonorant-like nature of implosives.
So uncertainty remains here too, and so *d₁ vs. *d₂ it shall be here too.
S&D mention a single Formosan language, Puyuma, where their entire assumed retroflex series /ʈ ɖ ɭ/ indeed turns up as such. It seems kind of random to base the sound values of a reconstruction on a single language, though. If we’re going to start picking “key” languages: the evidence N&D bring up from the Bornean language Dohoi would actually suggest the distinctions *p₁ : *p₂, *t₁ : *t₂ and *k₁ : *k₂ to all have been instead based on phonation! These are reflected, medially, as /p t k/ (series 1) vs. /ʰp ʰt ʰk/ (series 2). And, recapping again, the other exception languages mostly reflect *p₂ as zero, /f/, or /β/; *t₂ as /d/; and *k₂ as varyingly zero, /ʔ/, /g/, or /ɣ/. All this seems to suggest a series of relatively weak obstruents, that were lenited in some languages but merged with the plain stops in others.
What would remain puzzling in this kind of an interpretation is how the lenited reflexes seem to have frequently “bypassed” the regular voiced stops. Should one e.g. perhaps assume that these were the original voiced stops, while the traditional “voiced” series used to be something else entirely (e.g., prenasalized stops), and they only settled in their current place following an elimination of the previous voiced series? I smell a Glottalic Theory-style controversy looming on the horizon here.
For *t₂ in particular, there is its reflex as *ts in several Formosan groups, and as has been done ’til now IIUC, using this as the starting point might still be feasible. This could have developed into *θ in Malayo-Polynesian — a notoriously unstable sound that could have easily later merged into *t; or, with intervening voicing (as in High German & Dutch), into *d. The lack of any branches of MP with *θ > *f or *θ > *s might then seem odd, but less so than positing a “retroflex” that does not behave very retroflexibly at all.
The Tai-Kadai evidence also gets very interesting here. According to Norquest, in Northern and Central/Southwestern Tai, *t₂ surfaces as a consonant cluster: *hr- in the former, *tʰr in the latter. Sure, a breaking *ʈ > *tr would not be entirely out of the question. But again, typologically we should expect the inverse instead. A cluster *tr could also easily spirantize to yield *ts, either by devoicing of the rhotic, or by epenthesis similar to the American English change /tr/ > /tʃr/. Seems like as parsimonious a value to start from as any, then?
Taking a further leap of speculation: this approach might be applicable to the “other rhotic” *R of Proto-Austronesian as well. N&D assign this to a uvular series, but Northern Tai again reflects *hr- here. What if this is a hint that the scattered Austronesian *ɣ-like reflexes indicate a former velar component of a cluster (*k₂r?), instead of a French/German-like drift from *ʀ to *ʁ to *ɣ…?
(Yes, I’ll be the first to grant a weakness of these last two ideas: both *C = *t₂ and *R = *r₂ occur in all positions in Austronesian roots, i.e. initially, medially and finally. Which clusters tend not to do; and adjusting the reconstruction to *rt / *rk finally also wouldn’t help, since then we’d expect to see contrasts *tr ≠ *rt and *kr ≠ *rk turning up word-medially.)
At any rate: I hope to have made by now the point that identifying a new correspondence and identifying an original sound value for any particular correspondence are two very different tasks…