Learn English – British pronunciation of privacy

I think it's over-optimistic, and probably not very possible, to try to derive a formula to come up with the right choice in all situations. In short, I doubt the usefulness of the proposed rule.

In my (American) experience, both pronunciations for anti/semi/hemi are used interchangeably, and probably inconsistently even in an individual's lexicon.

On the other hand, I feel like "quasi" ends in "zee" more often than "zai" or "sai."

For Mobile/Fragile/Docile, I find myself using a short i sound that makes them sound like "mohbuL" "frazhuL", "DossuL". I don't doubt for a second though that there are going to be Americans who also pronounce it with the "ile" sound.

NOTE: This is off the top of my head; there may be aspects of this I've overlooked, so I would welcome any correction anybody wants to supply.

You're missing a simpler way of understanding this, because you're working off letters (which should properly be enclosed in ‹› rather than []) instead of sounds (specifically, phonemes, which should properly be enclosed in // rather than []).

Here are the rules:

1. Your two suffixes are what linguists call archiphonemes, which are sound-types "realized" as different phonemes in different sound contexts. We may designate them as /S/ and /T/, and categorize them as

   /S/ - (dental/alveolar) sibilant, realized as either voiceless /s/ or voiced /z/
   /T/ - dental stop, realized as either voiceless /t/ or voiced /d/

2. How the archiphoneme is realized depends on the sound which ends the base word and therefore immediately precedes the suffix:

   • If that sound is voiced, the suffix takes its voiced realization, and
   • if that sound is voiceless, the suffix takes its voiceless realization, EXCEPT THAT
   • if that sound is of the same category as the suffix, an unstressed vowel is inserted between the base and the suffix so you can hear the two sounds as distinct. Since all vowels are by definition voiced, a voiced sound now precedes the suffix, and the suffix takes its voiced realization. The vowel itself may be realized as anything in the approximate range [ɛ] – [ɪ], usually 'reduced' to [ə] – [ᵻ].

As far as I know, these rules are invariant in Standard English; but when both a plural and a possessive /S/ are applied (e.g., the Joneses' house) one may be suppressed in speech, I believe many Scots dialects realize /D/ fairly consistently as /ɪt/, and there may be other variations I've overlooked or I'm not familiar with.

Similar rules govern prefixes; but since nowadays most prefixes are of Latin or Greek origin, the rules are mostly derived from practices in those languages.