Monday, 14 December 2009

Bryn Terfel

Yesterday on BBC TV’s hymn-singing programme Songs of Praise one of the guest artistes was the distinguished opera singer Bryn Terfel. I nearly fell off my seat on hearing the presenter pronounce his name correctly, i.e. as he would himself. Then I realized the reason: the presenter was not some Englishman but Aled Jones, who is also, like Bryn himself, a native speaker of Welsh.
The non-Welsh find it difficult to get their heads around the idea that in Welsh spelling the letter f stands for v, not for f (despite the model of English of). So the singer’s surname is properly ˈtɛrvɛl. But most people think it must be ˈtɜːfl.
We see the same thing in the name Dafydd, the Welsh for ‘David’. In Welsh it is ˈdavɪð. By some further confusion the English often spell it wrong, too, as Daffyd, which fits better with the mispronunciation ˈdæfɪd.
The singer’s first name, Bryn, is fine as brɪn. Strictly speaking, though, in the northern variety of Welsh which he and Aled speak this is a central rather than a front vowel: it is a short lax ɨ rather than a short lax i. Ideally we could represent it as , although this symbol is not recognized by the IPA. The Wikipedia page on Welsh phonology writes it ɨ̞.
Here is the vowel formant chart from the Wikipedia article, attributed to Martin Ball’s article ‘Phonetics for phonology’ in Welsh Phonology: Selected Readings, ed. M. J. Ball and G. E. Jones, Cardiff, U. of Wales Press, 1984. You will notice that the placement of ɑ (the long open vowel, as in tad ‘father’) looks wildly out of line. I checked the original 1984 article, and I find that it does not contain this diagram although it does give the formant values on which it is based. I can only suppose that the values given for ɑ — 800 Hz for F1 and 1350 Hz for F2, both much higher than one would expect — may be wrong.


  1. I find people will always believe the evidence of their eyes over that of their ears. No matter how many times you explain to them that f=[v], they find it absurdly hard to accept. After all it actually says f right there on the page, it can't really be [v], that's a whole different letter. They might be more broadminded about vowels, but f is f, there can't be a different "pronunciation" of such a straightforward consonant. I have often wondered why v fell out of favour -- quite recently in fact, and (according to T Arwyn Watkins in 1993) it is still used in Patagonian Welsh. It's one of the little flaws in the generally true statement that Welsh orthography is blessedly straightforward for the learner.

  2. What would be the best approximation using non-rhotic, English English phonetics/phonotactics, which is what most BBC announcers will have?


    The name Terfel is not easy for most of us to get right, because we don't have a trilled [r], are non-rhotic anyway, and would be inclined to reduce or omit the final ɛ, as we do with names like Handel and Mendel.

  3. I'd be quite happy with the anglicization ˈtɜːvl. Slightly better: ˈteəvl.

  4. Surely it's easy enough to check the formants? Just start up Praat and say "Praat".

  5. I'm still chuckling at the image of you falling off your seat. I note you changed the idiom from 'chair', though for TV it may have been an armchair, and can't help wondering whether it should have been 'off' that you changed to 'out of'. That is the image I find myself chuckling at, anyway.

    I must say it would have had some such effect on me too, but it's certainly no thanks to the BBC. Why when they get Llandewi Brefi spelt and mostly pronounced OK is the only gay in the village usually Daffyd and pronounced that way?

  6. ˈtɜːvl is probably what I would say if I didn't think about it.

    I've often wondered about how people in Wales pronounce their names (in practice, I mean, not the actual orthographic rules), bearing in mind the two different phonologies - e.g. non-Welsh speakers with Welsh names ("Rhydian"), Welsh-speakers with non-Welsh names ("Jones"), and those with mixed names ("Llunos Childs"). Does it vary according to code-switching, or do Welsh-speakers simply have a combined inventory of sounds which they will deploy in either language?

  7. It looks to me as though the F1 and F2 values for [ɑ] are double what they should be.

  8. Harry Campbell: Say "f as in of" and they usually get it. As for the spelling, Welsh makes it too hard to tell when an indeterminate u/v form is a u and when it's a v. Consequently, Middle Welsh spellings like auall 'apple tree' become much more readable as afall.

    English spelling handles the u/v for you for the most part: generally, an ambiguous letter is v only before a vowel. Consequently, very few English words end in v: Slav and a couple of very recent abbreviations like guv, lav,, and perv are about it, and even the last is often seen with final e, as in love,prove, have.

    The standard romanization of Sindarin, Tolkien's Welsh-flavored Elvish language, uses f, v in most positions, but ph, f finally, apparently to avoid the "uncouthness" of final v to the anglophone eye. Ph is also used (as in Welsh) to indicate soft mutation of p, and (not as in Welsh, as far as I know) to represent geminated /f/, which generally arises from sound change at a morpheme boundary (Tolkien's example is ephel 'fence' < et-pel. (None of these distinctions are made in the native scripts, only in romanization.)

  9. I'm the one who created this chart for Wikipedia. I too was bothered by the F2 of [ɑ] (the F2 of [a] looks about right relative to [ɛ]) but that's what the source says, so that's what I put.

  10. From memory, I think in some of his writing on the topic Martin Ball makes clear he's using IPA characters in a more phonological sense in Welsh and he does mention the fact that the long a is quite front in the NW Welsh he was describing. To me this is part of the problem of the IPA symbols having to bear phonetic and phonological interpretations (combined with the problems inherent in the wikipediaisation of knowledge). People here are assuming a strict phonetic interpretation which is not correct in this case. It's just the sort of problem countless beginning students have with the tradition of using ʌ to represent the STRUT vowel in English.

  11. The amusing thing is that it would jar a lot less to Welsh-speakers if non-rhotic English types went for "terefel" rather than "tỳf(f)l" -- at least inserting an extra syllable would result in its having the right letters in it.

    But it is the hardest thing about the Welsh language for the English to accept that "f" and not "r" is the weakest consonant in Welsh (e.g. (y)mo(f)yn ("to want"), nesa(f) ("next"), ca(fa)s e(f) ("he got" (if you're from Gwent)), arna(f) i ("upon me")). And come to think of it, it is weird that it's such a basic consonant that is very weak in Welsh.

  12. Hi everyone - I'm marooned in the frozen north at the moment (visiting the University of Vermont), but when I get back to Lafayette I will check (if I can find the material!!) those pesky formants.

    Martin Ball

  13. @Paul: It's nothing to do with how we transcribe this vowel. It's that the values given are physiologically just about impossible. In any case, I know that Welsh long a: ("ɑ") is pretty similar in quality to Welsh short a, so would have similar F1 and F2 values.
    @Martin: Thanks in advance.

  14. Hi John,

    Sorry, I'd gone for the idea of [ɑ] looking out of place (i.e. fronter than [a]), rather than concentrating on the actual formant values of [ɑ] in this chart, independently of the other vowels. The values for [ɑ] in the chart do tally with Martin's chapter referenced above (it's from Table 9 on page 30 in my edition). Subject A in Table 10.3 (p.137) of Ball & Williams (2001) "Welsh Phonetics" (preface by one J.C. Wells!) has identical formant values except for [i] and [ɪ], which have lower values in the later work. I'm assuming it's the same subject, so perhaps Martin reanalysed the data, or an error crept in somewhere. But in any case the [ɑ] is the same in both texts. Martin and Briony concentrate on a discussion of [i] and [ɨ] but it's clear both from his/their data and the older data they report that /ɑ/ (i.e. the long vowel) can have higher F1 and higher F2 than /a/ (i.e. the short vowel) in north-western varieties. I suppose it should be no surprise that a long vowel is more peripheral than a short one it is related to. My initial impression is that an F1 of 800 Hz is pretty high for a male speaker, but perhaps not impossible.

    Another thing to note is that the difference between these two open vowels is exaggerated by the Wikipedia chart not having logarithmic axes. If we use an ERB-rate transformation, for example, the difference between the F1 of [a] and [ɑ] here is 0.95, whereas the F1 difference between [e] and [ɛ] (which looks the same on the chart) is 1.12 (18% bigger).

    Must go back to doing what I should be doing now, rather than interesting stuff like this! Thanks for the post, John.

  15. David Marjanović30 December 2009 at 00:10

    Ideally we could represent it as ᵻ, although this symbol is not recognized by the IPA. The Wikipedia page on Welsh phonology writes it ɨ̞.

    Why not simply ɘ?

    Incidentally, [ɘ] seems to be the most common realization of the Polish y these days, except where it's [ɛ] as in the 1st person plural verb ending -my.