Phonological vowel length in American English due to (t-)flapping

Question

The following is a quote from a Wikipedia page on American English phonology and concerns flapping in American English:

The flapping of intervocalic /t/ and /d/ to alveolar tap [ɾ] before unstressed vowels (as in butter, party) and syllabic /l/ (bottle), as well as at the end of a word or morpheme before any vowel (what else, whatever). Thus, for most speakers, pairs such as ladder/latter, metal/medal, and coating/coding are pronounced the same. For many speakers, this merger is incomplete and does not occur after /aɪ/; these speakers tend to pronounce writer with [ʌɪ] and rider with [aɪ]. This is a form of Canadian raising but, unlike more extreme forms of that process, does not affect /aʊ/. In some areas and idiolects, a phonemic distinction between what elsewhere become homophones through this process is maintained by vowel lengthening in the vowel preceding the formerly voiced consonant, e.g., [ˈlæːɾɚ] for "ladder" as opposed to [ˈlæɾɚ] for "latter".

Q: Are there any studies on how many speakers have this vowel lengthening? Or can anyone confirm this based on anecdotal evidence?

It certainly makes sense that speakers might phonologise vowel lengthening before voiced plosives after the voicing distinction in alveolar plosives has been neutralised. But this is the first time I have read such a claim, so I would be glad if somebody can confirm or deny it.

Also see this previous question on a related topic, where question and answers did not specifically focus on the phonologisation of vowel length.

Answer 1

I don't have any references on this at the moment, but anecdotally I can say that I indeed display a version of this length distinction in my dialect. There are some issues that should be highlighted, however.

In general, the well-known vowel length alternation that is observed before voiced vs. voiceless consonants is the most common and the most robust when the consonants in question are coda consonants and when the syllable in question is phrase-final:

Case A: peas vs. peace or Say peas vs. Say peace

Many speakers also display this alternation phrase medially:

Case B: Say bad for me vs. Say bat for me

Finally, some length differences may be observed even when the consonant in question may not be analyzed as a coda:

Case C: ogre vs. ochre or buzz around vs. bus around

In Case C it may make sense to analyze the consonants as being ambisyllabic--that is, both belonging to the coda of the syllable to the left and the onset of the syllable to the right. Note that the syllable to the right is unstressed in all of the examples.

I produce all of the length alternations mentioned above, although my speech follows the general trend in that the length differences are greatest in Case A, less in Case B, and the least in Case C. Similar length alternations can be observed in some words that I realize with flaps:

coding vs. coating or Say bad again vs. Say bat again

Note though, that in all of the above cases, the segments that are realized as flaps are found at morpheme boundaries, meaning that in other contexts they may be realized as [d] or [t] (or [tʰ]). I don't make a reliable length distinction in ladder vs. latter, and it is likely because in my lexicon the medial consonants in these words are not /t/ and /d/ but rather just /ɾ/. I have read claims that some speakers do make this distinction (though I don't have references off the top of my head), but I am not one of those speakers.

It is interesting to note that, when the vowel in question is the diphthong /aɪ/ and the flap is at a morpheme boundary (as in rider vs. writer), I display both the length alternation as well as the vowel raising phenomenon discussed in the post mentioned by the OP, but the vowel in spider and cider is raised (and the same length as the first vowel in writer), which suggests that in my lexicon those words do not have underlying /d/ in them. They could therefore either be analyzed as having /t/ or just /ɾ/, as I mentioned above.