focus-finality

I’ve been thinking again about the example I quoted on Friday, I fee[lʲ] ill vs. I may not look ill, but I do fee[ɫ] ill.

The only reasonable hypothesis I can come up with is that the ‘boundary’ that triggers the dark l is the end of a focus domain.

Marked up for intonation, we have
    I ˈmay not \/look ill, | but I ˈdo \/feel ill ||
or perhaps
    I ˈmay not \/look ill, | but I ˈdo \feel ill ||

In pragmatics, focus is the foregrounding of part of the message — typically what is new or contrastive or otherwise important — and the backgrounding of everything else. It is realized in intonation as the presence of accented syllables, i.e. syllables that are not only rhythmically but also intonationally prominent (pitch-prominent). Identifying the beginning of each intonational ‘focus phrase’ or ‘focus domain’ is sometimes tricky, but identifying the end is easy: it ends at the end of the word in which the nuclear accent appears (= the last accented syllable in the intonation phrase).

So in this example the focus domain in the first IP extends from may to look. The focus domain in the second IP is do feel. My hypothesis is that here the l in feel, being focus-final, is made dark even despite its prevocalic environment.

This is not a matter of being ‘prepausal’. In the example given, the phrase feel ill is not interrupted by any pause or intonation boundary.

Of course it would also be possible to place an intonation boundary and possible pause there.
    You ˈmay not be able to \/see anything unusual, | but I ˈdo \/feel | \ill ||

As we know, any lateral that is prepausal is dark. This applies whether or not it is also focus-final.
    You’re \i[ɫ] — (You’re wrong!) I’m /not i[ɫ]

I tried some more examples, and introspecting I think they work.
    This is ˈNell \Anderson. (nelʲ)
    ˈPlease /welcome | ˈJames /Anderson | and his ˈwife \Nell Anderson. (neɫ)

    There’s a ˈsmall \error. (smɔːlʲ)
    ˈNot a \/big error, | but a \small error. (smɔːɫ)

The big question, which I’ve never asked myself before, is this: are there any other instances of allophonic variation triggered by focus-final position? (The idea that this might apply to linking r, floated in the discussion on Friday’s blog, seems to me to be a non-starter.)