How are the varying boundaries of phonemes across idiolects and dialects objectively described?

Question

I am taking a language I know nothing about as an example.

In Vietnamese, I am told “hello” is *Xin chào”, pronounced |sin tʃaw|, with a mid tone like saying “aaah” at the doctor, and a falling tone like saying “huh”.

How much more objectively precise and accurate can this description be, and can it be notated? How?

For example: how do each of the phonemes get articulated specifically by Vietnamese speakers as compared to their supposed analogies (like “s”) for English speakers? Is the acoustic graph different? Is there a notation to try to express the complete, “real” sonic quality (and distributedness) of linguistic sounds?

http://sealang.net/sala/archives/pdf8/vu1982phonetic.pdf

https://phoible.github.io/conventions/

Is there a phonetic notation that comes as close as possible to a human readable acoustic spectrograph, perhaps?

Or a language indicative of such precise physiological specifications or dynamics that there would be a readable difference between the “s” of various languages and dialects, like between English and Vietnamese, or two speakers of English?

How precise has or can the notation be or become?

Answer 1

I want to deny that it is possible because it is bad practice, but it is somewhat possible. First, you have to absorb the International Phonetic Alphabet in all of its detail. Second, you have to focus on what, exactly, those symbols mean, for example [a] is an "open front unrounded vowel". This is an articulatory description, which is frequently not open to ordinary inspection, but you can also consult various phonetics books and articles to see how to see the mostly-unseeable. However, the IPA page gives you reference performances of categories. Therefore you can follow the reasoning that "if it sounds most like [a], it is an open front unrounded vowel".

The next step is rather hard. I substitute a different example, "hoặc những chú", since I have a reasonable belief that this wasn't edited to destroy information (omitting diacritics that specify prounciation). The problem here is that we must struggle to figure out how this is pronounced, which is the basis for creating an IPA transcription. You might get lucky and ask a passer-by who speaks Vietnamese to say this, then using your acquired expert skills in converting language sounds to IPA letters, you can proffer a transcription. If your transcription is good enough, there is a chance that any well-trained linguist can pronounce what you wrote, and it would be perceivably similar to the original model.

The IPA has a limited range of symbolic categories, but it also has "adjustment diacritics" which allow you to nudge things a bit, for example "retracted relative to the standard position defined as 'front'". There are no standard performance models, so this is a difficult judgement. This allows you to gain a bit more precision, but not much. Such transcriptions are an extremely coarse-grained device for expressing phonetic facts, and are not designed to convey phonetic measurements, it is designed to convey phonological categories which are used in talking about actually-measured properties of sound or articulation. Actual numeric measurements are how we obtain precision.

Here's a synopsis of the fundamental limit on the method of symbolic writing. To start with, switch to English. How do you phonetically transcribe "how"? That varies extensively according to dialect. In Derry English it's something like "hoy". We can focus on American English, and reduce the problem a little but, but you still get [hæə, hæw, haw]. The more specific you are as to dialect and idiolect, the more meaningful it would be to ask a "spectrographic" question (note that you can't ever make a spectrogram of "how, in English", you can only make a spectrogram of a single utterance). It is trivial to see differences in the actual acoustic output of two utterances of a word from the same person. Yet we want to say that these are "the same thing", which tells you that a certain level of over-precision is undesirable.

Symbolic methods (transcriptions) are based primarily on an extensive program of training (getting the various vowel differences and internalizing them by multiple exposure) and careful auditory judgments, which can tell you that "ð" in Danish is clearly different from "ð" in English, and [c] in North Saami is clearly different from [c] in Hungarian. This is where the nudge diacritics may be useful so that Danish ð might be transcribed as [ð̠˕ˠ]. Hawrami (Kurdish) has a similar yet slightly different consonant, which might have to be transcribed the same way, given the desideratum of encoding sound properties in the transcription in a standard way, thus there are limits to the precision that one can gain.

Answer 2

How much more objectively precise and accurate can this description be, and can it be notated? How?

The International Phonetic Alphabet has a variety of diacritics that can be added to its glyphs to convey what is often called "phonetic detail" (and there are additional diacritics used by speech pathologists to describe "disordered speech"). There are a couple of key concepts necessary to understand what "phonetic detail" means:

1. Representing a dynamic, continuous speech signal with discrete symbols on paper necessarily involves some loss of information. No transcription system can capture all the details.
2. What counts as "phonetic detail" depends on the language. For many languages, the difference between [s] made with an lamino-alveolar constriction and [s] made with a apico-dental constriction is a difference that doesn't carry any linguistic meaning (though it might signal something sociolinguistic, if certain subcommunities prefer one variant over another). In other languages, the s̪ versus s̻ distinction is contrastive (i.e., phonological, not "phonetic detail").

For example: how do each of the phonemes get articulated specifically by Vietnamese speakers as compared to their supposed analogies (like “s”) for English speakers?

Depending on in what way the Vietnamese [s] is different from your comparison group's [s], you might or might not find a suitable IPA diacritic to mark the difference.

Is the acoustic graph different?

Every token of [s] spoken by one individual is going to be different, so naturally if comparing two [s] tokens from two different speakers you should expect them to be different, and mutatis mutandis two [s] tokens from two speakers of two different native languages are also going to be different. What matters to most phoneticians/phonologists is whether the way in which they differ is interesting or not.

Is there a notation to try to express the complete, “real” sonic quality (and distributedness) of linguistic sounds?

Not really. Good quality audio/video recordings are the gold standard if you really want to know what speech sounds like. There are various summary measurements that you can extract from a sound recording (or corpus of sound recordings) but what people choose to extract will always depend on what aspect(s) of the speech they are most interested in (i.e., if they are more interested in the language's tones, or its fricatives). Moreover, we know what the range of human hearing is, so we more or less know how good an audio recording needs to be (in terms of samples per second and bits per sample) in order to capture "all" of the detail that would have been perceivable if we'd been there in person to hear it.

Is there a phonetic notation that comes as close as possible to a human readable acoustic spectrograph, perhaps?

No.

Or a language indicative of such precise physiological specifications or dynamics that there would be a readable difference between the “s” of various languages and dialects, like between English and Vietnamese, or two speakers of English?

There are some helpful technologies for studying articulation like ultrasound tongue imaging, x-ray articulography, electromagnetic articulography, static palatography, electroglottograms, etc. With these tools (combined with good audio/video recordings), you might be able to say something like "Speaker X makes the [s] sound with a lamino-dental constriction, slight lip rounding, and a central tongue groove with side bracing; the lip rounding decreases across the duration of the fricative." You could also include in some numbers or graphs of acoustic measurements, like of the duration or spectral characteristics of their [s]. But you will never be able to come close to capturing everything that is in the utterance.

Again: it's always a matter of deciding which details are interesting / relevant and measuring / describing those details.