The argument against this is phonological. If we add to our phoneme inventory the rising diphthongs in these words, we shall have also to add those of weave, wet, whack, suave, war, woman, woo, work and yeast, Yiddish, yet, yam, yarn, yawn, York, yearn, i.e. more than the number of simple vowels we have in our inventory; not to mention additional triphthongs that we shall have to recognize in words such as way, woe, wine, wow, weird, yea, yoke, yikes, yowl, yeah. But these nonsyllabic u̯ and i̯ pattern like consonants (being at the margins of syllables), so it is clearly better to recognize just the two semivowels w and j, and to analyse all the polyphthongs just mentioned as /wV, jV/. A semivowel (or ‘glide’, if you prefer) is articulated like a vowel but patterns like a consonant. We no longer attempt to distinguish on the phonetic level between nonsyllabic [i̯, u̯] and [j, w].
You may be familiar with an indelicate limerick (search here, for example) in which Australia and dahlia (BrE for this flower, with eɪ as the stressed vowel) are made to rhyme with failure. Are these good rhymes? Is dahlia ˈdeɪljə an exact rhyme with failure ˈfeɪljə? Well, yes and no. The possible difference between them is not a matter of i̯ as against j, but rather of our awareness of the varisyllabicity in dahlia as against its absence in failure. We know that dahlia can optionally be said with three syllables (by some of us, at least), while failure can only have two.
(I was perhaps being too sweeping the other day when I suggested that this whole matter was a question of BrE vs AmE; but it is striking that Kenyon-Knott and Merriam-Webster do not allow for -eɪl.i.ə in Australia, while LPD and EPD do.) Then what about millennia compared with tenure? Peter Roach’s CPD has these as non-rhymes (mɪˈlen.i.ə and ˈten.jəʳ), just as in LPD I have mɪ ˈlen i‿ə and ˈten jə. Merriam-Webster, too, shows the difference here. So do Kenyon and Knott — though at Virginia K&K give -ˈdʒɪnjə but also add ‘esp. New England’ -ˈdʒɪnɪə. At this word ODP, by the way, gives for BrE only vəˈdʒɪnɪə(r) and for AmE only vərˈdʒɪnjə.)
What about a word like happier? It is clear that it can on occasion be pronounced as a disyllable. So if so pronounced, is ˈhæpjə the correct way to transcribe it? And for various, ˈveərjəs? What about DJ’s valuing ˈvæljwɪŋ? Is genuine truly ˈdʒenjwɪn?
We have seen that the reason why we hesitate to regard these as the underlying (lexical-entry, articulatory-target) representations is our awareness that in each case there is the possibility of a syllabic (= vowel) pronunciation in place of the putative semivowel. There are other possible reasons, too.
- Given that we get noticeable devoicing of j after p in words like pure, why do we not get similar devoicing in happier? (Or perhaps we do?)
- If the sequence -rj- is so awkward in garrulous, virulent, glomerula that we tend to avoid it by dropping the j, why does the same not apply in disyllabic various, barrier, glorious and so on? There is even disyllabic Istria, which must be ˈɪstrjə.
- In some kinds of AmE the sequence -lj- in words such as William, million, failure can get reduced to -jj-. Does this happen in volleying and jollier? If not, why not?
- The ‘semivowel’ solution leads to our treating valuing and genuine as containing sequences of semivowels, -jw-, something otherwise unattested in English and universally unusual.
In supplying a pronunciation entry for any word containing a possible w or j plus vowel, then, we have to ask ourselves: can this semivowel alternatively be pronounced as a syllabic vowel? If yes, then we take it as u, i; if not, then as just j, w. We may not always agree on the answer. As we have seen, British and American lexicographers disagree in the case of Australia. When I transcribed Daniel as ˈdæniəl the other day, I didn’t stop to consider the issue, though if you now ask me I would confirm yes, I can pronounce this name as a trisyllable. But some of those who commented obviously can’t. Anyhow, I also entered it in LPD as ˈdæni‿əl (the possible compression is predictable from context). On the other hand, I entered million in LPD as ˈmɪl jən, only to receive a complaint from one user that I ought to have allowed for a trisyllabic version and entered it as ˈmɪl i‿ən. (So in the current edition I give both.) You can't win them all.