L in the middle of a word: dark l or light l?

Solution 1:

In Southern Standard British English (RP), /l/ is always dark unless followed be a vowel (sound). When followed by a vowel it's always clear. In the middle of the words delete delegate and silly, the /l/ is followed by a vowel and will therefore be clear.

In a word like alright where /l/ is followed by a consonant, it will be dark. We use clear /l/ in the same environments where we would pronounce an /r/. We use dark /l/ where r would be silent.

Note that word boundaries do not affect this rule.

• call Ben

• call Ana

• or Ben

• for Anna

The /l/ will be dark in the Ben sentence above. It will be clear in the Anna one where it is prevocalic (appears before a vowel). In the same way, there is no [r] in the Ben sentence, but there is in the Anna one.

Descriptions which say that clear /l/ occurs in syllable onsets and dark /l/ occurs in the coda at the end of the syllable, assume a theory of liaison. This means that the /l/ in call Anna has moved from the end of the call syllable, to form the onset of the following one in Anna.

Solution 2:

In the American English I'm familiar with, and as Mitch says above in a comment, light versus dark l depends on where the syllable boundary is. In the onset of a syllable (before the vowel, i.e.), it's light. In the offset (after the vowel, i.e.), it's dark.

Of course, to interpret this rule, we need to know where the syllables are. Medially between vowels, l is in the syllable of a following stressed vowel, reckoning stresses as in SPE, but before a following unstressed vowel, l goes in the preceding syllable. So for this case of intervocalic l, the syllable theory says that l will be light before a stressed vowel and dark before an unstressed vowel.

For the words mentioned in the question, this gives light l in "delete" because the second syllable is stressed. It gives dark l in "delegate", because the second syllable is unstressed. It gives dark l in "silly", because the second syllable is unstressed. (Not everyone agrees that the second syllable of "silly" is unstressed, but that is the SPE description, and I agree with it.)

These syllabifications for medial intervocalic consonants are supported by a number of other cases in English where consonants have fortis varieties in syllable onset and lenis varieties in syllable offset. Fortis r is rounded; lenis r is not. Fortis alveolar stops are not flapped; lenis ones are. Fortis [j], yod, is a palato-alveolar; lenis [j] is palatal. Fortis syllable initial obstruent stops are aspirated, lenis obstruent stops are not. And there are other such cases.

I would pronounced the doubled l versions mentioned in the question only when sounding words out, syllable by syllable, but never in normal speech.

American English does not have linking, except in a few set phrases, where an invervocalic word-final consonant will be shifted into the syllable of the following word-initial vowel, so word-final l will always be dark.

Dark l is articulated with backing of the tongue body -- in SPE features, it is [+back], and that makes it uvularized. Authorities generally describe it as velarized, for some unaccountable reason. A velarized sound is [+high,+back]; that is, it involves raising of the tongue body, as well as backing. I find there is no such raising after non-high back vowels -- otherwise, I am not sure about the [+high] part.

Dark l, at least the uvularized variety, is often vocalized; i.e., it has no tongue-tip contact with the top of the mouth. I have not covered the case of l in medial consonant clusters.

Solution 3:

There is variation between accents. Some speakers might have non-dark [l] in all of these; some speakers might have non-dark [l] in delete and dark [ɫ] in delegate and silly (I think I fall into this category); and some speakers might even have "dark" [ɫ] in all of these (I think only speakers who use "dark l" in all contexts, including word-initially, would show this pattern of word-medial "dark l" usage).

"An amphichronic approach to English syllabification" (Ricardo Bermúdez-Otero, 2013) describes some relevant data and analyses.

Mr Beelik’s paradox (Bermúdez-Otero 2011: 2038-9)

§29 Kahn’s prediction:
Kahn predicts that, unless the segmental context is different, a liquid will have the same allophonic realization in word-final prevocalic position (e.g. seal in) and in foot-medial intervocalic position (e.g. Sealey)

§30 [...] The Midwestern US dialect described by Sproat and Fujimura (1993), henceforth ‘the SP dialect’.

• Midwestern speakers ➩ canonical (Kahnian) pattern of /t/-flapping
• /l/-darkening data:
  Beelik /l/ ambisyllabic by Coda Capture ➩ [l] i.e. clear l (coronal lead)
  Beel equates /l/ ambisyllabic by Onset Capture ➩ [ɫ] i.e. dark l (coronal lag)

∴ In the SP dialect, Kahn’s syllabification works for /t/ but not for /l/
(already noticed by Sproat and Fujimura 1993: 308)

[...]

The cyclic solution to Mr Beelik’s paradox (Bermúdez-Otero 2007b: §18-§20, 2011: 2039)

§34 English syllabification is onset-maximal at all levels (see §5 above). [Therefore, word-final consonants are in the coda at the word level, but they are detached from the coda and resyllabified into the onset at the phrase level when followed by a vowel-initial word.]

[...]

§46 The dialectal signature of rule generalization:

• In the SP dialect, stop lenition applies in weak positions in the foot, [_Σ …V́…__…], whereas /l/-darkening is still confined to weak positions in the syllable, __ _σ]

  However, there do exist other English dialects where /l/-darkening is more advanced, having become a foot-based process:

  e.g. ye[ɫ]ow, vi[ɫ]age.

  For American English, see e.g. Olive et al. (1993: 366), Hayes (2000: 95-96);
  for British English, see Carter and Local (2003, 2007).

There is also a blog post by John Wells that describes some interesting morphology-based contrasts observed by Christian Uffmann in certain British English dialects that might be related to the darkness of the /l/ (from what I understand, Bermúdez-Otero would account for these by saying that the /l/ is in the coda at the word level for some of the words): "newly minimal" (2012).

The kingly ruler, ˈruːlə, is taken as transparently bimorphemic, rule#(e)r, so retains the back uː of rule; but the measuring ruler, ˈryːlə, has lost touch with its origins and is taken as an unanalysable unit, with a corresponding clear l and fronted vowel yː.

This parallels the “GOAT split” that I described for London English in Accents of English (p. 312–313) and which gives us non-rhyming goal#ie vs. slow#ly.

[...] we may have a similar development in, for example, feeling vs. Ealing, failing vs. railing(s), where again the first of each pair is morphologically transparent, leading to an epenthetic schwa (“Breaking”) and a dark l as in feel and fail, whereas the second is seen as morphologically indivisible, so has a clearer l and no schwa.

Peter Shor's answer to "Why are “fun” and “hulk” phonetically transcribed with the same vowel but pronounced differently?" provides some interesting similar examples from an American English speaker: Shor says that he pronounces color and sully with non-dark [l], but duller, dully, gully with dark [ɫ]. In a comment, he adds that although he does not have two distinct pronunciations for the distinct meanings of the word ruler, he does pronounce cooler the adjective differently from cooler the noun.