It's a mixed bag.
Frequent, common words that are long tend to be pronounced quickly. Quick pronunciations tend to lose distinctions in consonants and vowels, they tend to be clipped or the consonants weakened. "Fast" and "easy to pronounce" are usually the same.
This is an old generalization. Bybee's notion of token-based phonology (which means a token-based kind of language acquisition) provides a way of implementing this kind of change.
This doesn't work with some very short, grammatically essential words, IIRC. If you need them, you don't reduce them beyond some minimum.
Oddball words and words that signal high-style tend to be preserved. In many cases we find ways to make "formal" mean "more syllables" by adding additional endings or finding larger expressions to merge them with.
That's at the word level.
At the level of the sound system, mergers happen at the expense of distinctions (this is one of Bartoli's norms and while it's a universal tendency it obviously isn't the whole story). For example, t/d merge in flapping. /a/ and "open o" (Don, dawn) are pretty much merged in American English.
It's also the case that consonants more often undergo lenition--weakening, in a sense--than strengthening or "fortition." Not all consonants are equally subject to lenition. Consider how AAVE "bad" and "bat" are often pronounced with an unreleased or even absent final consonant. Spanish /d/ is probably best seen as an interdental voiced fricative.
= new reply since forum marked as read
