Feature correlates of the length and tenseness contrast in the low vowel /a/

Question

My ultimate goal is to be able to predict from external factors whether and what kind of vowel quality correlates an /a–a:/ contrast might have in a language, and specifically to determine which on is more likely to be front or back.

• It's common knowledge that vowel length contrasts are typically augmented, most commonly with tenseness (μμ > [+tense]). This in turn has consequences on the vowel's realisation and possibly further phonetic differentiation.

• Basic physiology suggests that the long /a:/ will be more open, and thus more back than the short one, and this seems to be supported e.g. by Lombard: front when short, back when long; also Parisian French at the turn of the last century, and modern Quebecois (with rounding and even diphthongisation). But there are opposite examples like Hungarian, with its /ɒ~ɔ/ vs /a:/. In both these cases no tensess seems to be involved (at least synchronically: I'm clueless about the latter's historical phonology). Proto-Slavic had the same: whether or not there was systemic vowel tenseness I cannot say, but two reduced vowels there were.

• In languages with vowel reduction, with or without a length contrast, the shorter an /a/ gets the less back it becomes, and a drift towards /æ~ɛ/ is possible (e.g. historic reduction in Latin). Conversely, in a final unchecked syllable Oscan exhibits backing, with /a > ɔ/ (presumably: it's transcribed Ú, like /o/, despite the letter O being unused in the writing system) but not elsewhere. It seems reduction might have something to do with the presence of a phonetically non-back /a/: Latvian has extremely short, sometimes barely audible short unstressed vowels, yet they're not phonologically reduced as in Russian, nor lax, but only subject to physical articulatory constraints. It's tempting to link this to its highly consistent maximally open /ɑ/.

• It starts to get interesting when tenseness (or is it tongue root?) comes into play. Dutch has a front tense /a~æ:/ vs. back lax /ɑ/, but apparently some dialects reverse this (Wiki). German has short /a/ vs. long /ɑ:/, neither of which is ostensibly lax, but seems to have a back short lax [ɑ] in open unstressed syllables (Wiki). In Irish, the long /a:/ but not the short one exhibits regional variation in being canonically either back or front. In Danish, /a:/ but not /a/ is front [+tense] [æ̝ ~ ɛ̞] (comment thread). In Middle English, the tensing and raising of both the short and the long low vowel to a low-mid one suggests that both were non-back (Kiparsky below).

• So far, it looks like the tensing (or +ATR) of the low vowel /a/ may be associated with either fronting or backing, but the latter seemingly more rarely, perhaps only if it's [+back]. Kiparsky in Joseph & Janda (2003), citing himself (1974), gives a diagram where æ and ɔ₂ (lower than the [-tense] ɔ₁) are the [+tense] pairs to a and ɒ. So if a system doesn't allow the [+tense] ɐ, one of these two are likely to be chosen as the tense allophone - and I'm interested in the factors that determine which one it is.

• So, how does the length/tensenss contrast in the vowel /a/ correlate with its phonetic realisation? Have some universal phonological correspondences and factors been found that determine which one is likely to be phonetically back or front? Is it possible to look at the phonology of a language and predict the phonetic realisation of its low vowel? Or to look at its phonetic realisation and learn something about its phonology? Can one determine either the phonology or the phonetics by looking at two states of a phonological system before and after its restructuring (eg. lack of differentiation of the long and short /a/ in Romance?).

• Albeit my grasp on this is mainly intuitive, I welcome technical descriptions of any depth, autosegmental, particle phonology, the more the merrier.

Answer 1

You cannot look at the phonology of a language and predict the phonetic realization of its low vowel. The horse is the phonetics and the cart is the phonology: you start with knowledge of how a vowel is pronounced, which determines the final phonological analysis – if the vowel is [a] then the phonology produces a front vowel, if the vowel is [ɑ] then the phonology produces a back vowel. You might be able to argue that [a] derives from /ɑ/ (or the opposite) based on some phonological pattern fact, but these letter choices are fundamentally phonetic claims, and when you make contrary claims like saying that [a] is really /ɑ/, you have to justify the added step (why not just say that [a] is /a/?). However, you can "predict" (say in advance of experience) that the phonetic vowel [a] most likely derives from phonological /a/, in the production sense, likewise you predict that [ɑ] probably derives from /ɑ/.

Another kind of prediction is historical prediction, where the question would be "Can you predict the (distant) future phonetic realization of a low vowel, based on ___ facts". So, could you have predicted backing and rounding of PIE *ā to Germanic ō? Could you have predicted the Attic Greek vowel change ā→ɛ̄? Since these changes happened millenia ago, "prediction" is inapplicable. You might however predict an outcome for future English, based on statistical patterns as recorded in the modern language by socio-phoneticians. The future elimination of the vowel distinction [a] "Don" and [ɔ] "Dawn" or however you want to transcribe them is predicted, and the probability is good that [a] will move further back and [ɔ] will only from and unround a little. This is because we can observe these tendencies right now, and see that they are spreading. This kind of prediction is a bit of a cheat, because it is based on direct observation of phonetics (the horse), and a conclusion that these patterns of variation will get phonologized (the cart) in a particular way in the future. (Well, they are solidly phonologized in many dialects, what remains is the expansion of this process to bastions of resistance like Chicago).

There is a perceptual basic for a weak prediction that applies to all languages, that there really should only be one low vowel. F1 and F1 are close enough together for low vowels that it is perceptually challenging to have a plethora of them which exploit tiny differences in formant frequency. We know that having front unrounded and back rounded vowels is a better perceptual strategy that having front rounded and back unrounded vowels. This perceptual factor lets us weakly predict that a language with [a æ ɐ ɑ ɒ] is going to change some of these vowels by neutralizing within the set or by raising.