English is known to have a phenomenon of "pre-fortis clipping": in certain contexts, vowel and sonorant phonemes before a fortis/voiceless consonant are realized with shorter duration than the same phonemes before a lenis/voiced consonant (or before no consonant).
English vowels are subject to pre-fortis clipping, then, when they are followed by a fortis consonant within the same syllable. The /f/’s in self, selfish /ˈself.ɪʃ/, and dolphin /ˈdɒlf.ɪn/ trigger clipping, but not those in shellfish /ˈʃel.fɪʃ/ or funfair /ˈfʌn.feə/. So do the /t/ in feet and the /ʧ/ in feature, but not the /p/ in fee-paying or the /k/ in tea-kettle. The vowel /æ/ undergoes pre-fortis clipping in lap, lamp, happy /ˈhæp.ɪ/, and hamper /ˈhæmp.ə/, but not in slab or clamber.
(Susan Ramsaran (ed.), Studies in the pronunciation of English, A commemorative volume in honour of A.C. Gimson (London and New York: Routledge, 1990), pages 76-86.)
This data is consistent with Wells's method of syllabification, where intervocalic consonants are assigned to the coda of a preceding syllable when it has greater stress than the following syllable. However, I don't believe that a consensus has yet formed in favor of Wells's system of syllabification. I've seen various post-1990 literature, such as "An amphichronic approach to English syllabification", by Ricardo Bermúdez-Otero (2013), that instead syllabifies such consonants into the onset.
Onset-favoring theories of English syllabification need to have a way to deal with the facts of pre-fortis clipping, and as far as I can see, a syllable-based account won't do the trick for those analyses.
Non-syllable-based formulations that I've found so far
"The foot as a segmental domain", by John Harris, 2000, suggests that the clipping rule is foot-based: "English pre-fortis clipping/pre-lenis drawling: duration of V is conditioned by co-pedal C" (p. 14).
This foot-based formulation accounts for all of the examples mentioned so far, but not for a "latex" example mentioned in Bermúdez-Otero 2013. Bermúdez-Otero says that in certain American English accents that have [ɾ] in words like "better", the word "latex" can be produced with a clipped vowel before a voiceless plosive [t] (not lenited [ɾ]). This poses a problem for certain analyses. Specifically, it excludes
a) any analysis that both restricts clipping to tautosyllabic sequences and predicts lenition for all coda /t/
b) any analysis that that both restricts clipping to foot-internal sequences and predicts lenition for all non-foot-initial /t/.
Bermúdez-Otero 2013 cites Bermúdez-Otero 2007, "Word-final prevocalic consonants in English: representation vs derivation". Here, the possible presence of clipping of the first vowel in latex without lenition of the /t/ is explained by the following formulation:
§24 A prosodic solution:
- The environment of stop lenition is defined by the foot.
- The environment of prefortis clip is defined by the prosodic word: a vowel undergoes clipping if followed within the same prosodic word by a fortis consonant that does not belong to a stronger syllable.
I find this description a little complex, but satisfactory for the most part (I'm not sure whether the "that does not belong to a stronger syllable" exception is necessary: do words like "tattoo" not show clipping?). Does anyone know of problems with this rule, or further refinements that have been made to it?