Dirk: > On Sun, 15 Oct 2000, And Rosta wrote: > > > Ray: > > [...] > > > As I said, I'm keeping an open mind on 'ambisyllabicity'. At present, I > > > find the ambisyllable analysis easier to swallow than ['h&p.i]. > > > > Yeah. We need to ask Dirk how he accounts for: > > > > Sal [saw] > > Sally [sali] > > > > in demotic SE Insular English, if /l/ isn't in an onset in "Sally". > > Wow. This is very nice! Is the /l/ in 'Sally' not velarized at all? That's right. > This would be good evidence for [s&.li]. For /l/ being in the onset of the following syllable, at least. (Cf. the way /r/ in nonrhotic accents must be in an onset, yet can follow a lax vowel, as in "carry".) > > Also: > > > > hoe [h@w] > > holy [h@wli] (/l/ only in 2nd syllable) > > whole, hole [hOw] > > wholly [hOwli] (/l/ ambisyllabic, triggering vowel > > allophone in 1st syllable) > > Hmmm. These are posers. So if I understand the forms you cite here, > the <w> in the first pair of examples is part of the tense vowel, > while in the second pair, it is the allophone of coda /l/; correct? No, actually. Well, that may be how to do it in the final analysis, but a comparison with RP suggests otherwise ("5" = dark L): hoe [h@w] holy [h@wli] (/l/ only in 2nd syllable) whole, hole [hOw5] wholly [hOwli] (/l/ ambisyllabic, triggering vowel allophone in 1st syllable) cf. nonconservative RP ("8" = barred u): coo [k8] coolie [k8li] cool [ku5] coolly [kuli] If /l/ precedes a vowel then it is clear (= in following onset). If /l/ follows a long vowel and doesn't precede a morpheme boundary then it is not coda of the preceding syllable and does not affect the realization of the preceding vowel. But if /l/ does precede a morpheme boundary then it is coda of the preceding syllable and does affect the realization of the preceding vowel. (For similar evidence see also "failure/dahlia" from another post of mine from today). > If > this is the case, then the behavior of the "ambisyllabic" /l/ is > strange; "half" of it becomes [w], while the other "half" remains [l]. > This would seem to constitute good evidence for the covert gemination > analysis of ambisyllabicity; there really are two parts to this /l/, > and they are subject to different phonological processes. Had you been interpreting the data correctly, this would be a sensible conclusion, except that I'd be inclined to see a pseudogeminate sequence of two /l/s rather than a geminate (= two structural positions sharing segmental content), because the two /l/s are subject to the different phonological processes. Note that some speakers do have "wholly" [hOw5li] and "coolly" [ku5li], while yet others (but possibly not RP?) have 'true geminate' [hOwl:i] and [kul:i]. --And.