Ask LLOG: Alternative standard pronunciations?

In an email, Coby Lubliner observed that many English words have more than one standard pronunciation. He asked whether there has been any academic discussion of this general issue, and noted that he isn't familiar with any other language with as much variety in its standard pronunciations.

From his email:

What I am referring to involves mainly loanwords, both fully naturalized ones and those that are still perceived as foreign, but there are some native words (those going back to OE) as well, e.g. (n)either, roof and others.

In naturalized loanwords, as in these examples, it's mainly a matter of vowels. In stressed vowels, it's whether it's "long" or "short", as in fecund, privacy, hologram, culinary and hundreds of others. In unstressed vowels, whether it's given its full value or reduced, as in fatality, financial, direct etc.

Consonant ambivalence is relatively rare (Celtic).

But there is also a matter of stress. With Anglo-Latin words (borrowed directly from Latin, not via French) there is a tradition that however the vowels may have shifted, the original stress is kept. But this is often replaced by a default stress for unfamiliar words, which is that (1) consonant-ending words of three or more syllables are proparoxytone (e.g. acumen, tinnitus), (2) vowel-ending words (other than ending in -y) are paroxytone (e.g. patina).

This stress variation is especially common with non-naturalized loanwords (those still perceived as foreign), including foreign place and personal names, and seems to depends on one's relative familiarity with the original language (not necessarily knowing the language but exposure to people speaking it), with words like, for example, paprika, mandala, basmati either proparoxytone as in the original or paroxytone by default.

There are certainly academic discussion of various different aspects of this issue, for example whether loanword stress in English retains the original patterns or adapts to local quasi-generalizations — see e.g. Ellen Broselow's 2009 chapter "Stress adaptation in loanword phonology: perception and learnability". And there are certainly academic discussions of variable adaptation in loanword vowels and consonants in English.

But I think the broader issue is how important standardization is to the culture (that sees itself as) in charge of a given language,  how univocal and official the resulting standards are, and how those (putative) standards are enforced, if they're enforced at all. Every major language has regional, social, temporal, and idiosyncratic variation. Is there one central authority that chooses among the variants? Or does the "standard" only emerge from social convergence rather than from an official committee? And are multiple speech communities and/or multiple official committees involved? Whatever the standardization mechanisms are, how committed are they to the idea that each word should have just one pronunciation?

As a non-English case where many words have multiple "standard" pronunciations, consider Japanese accentuation. Different regional and social dialects have different accent systems, as well as different treatments of cognate words. Even speakers of "standard" Tokyo-region Japanese often disagree among themselves about the accentuation of a given word, and as a result the NHK broadcasting system has for many years produced an accent dictionary, which announcers are supposed to rely on in order to create consistency in pronunciation.

For a different sort of situation, we could look at Italian, where there are multiple regional varieties forming either a dialect continuum or a set of different languages, depending on how you want to look at it — but where there are also locally-accented varieties of the Tuscan standard, differing in things like the existence of geminate consonants. Or Spanish, where a similar situation has coalesced into different "standard" varieties in different countries. Or Portuguese, or French, or Turkish, or …