Here is the setext version of this lesson.
The question of word-level stress in Welsh is exactly what I did my PhD on (finished in 1983). The main experiment was published as: "Pitch and duration in Welsh stress perception: the implications for intonation", Journal of Phonetics (1985), vol. 13, pp. 381-406. I'll summarise below:
I made several recordings of native Welsh speakers, both in the studio and in the field (literally! at the National Eisteddfod at Machynlleth in 1981). I made sound spectrograms of the speech, and measured the duration of vowels and consonants from these, as well as the intensity and fundamental frequency (F0) of vowels. I carried out a statistical analysis of these measurements, and found that there was a statistically significant lengthening of a consonant after a stressed vowel. In contrast to stress in English, German, etc., there was no consistent lengthening of stressed vowels, and no significantly greater intensity or higher F0 on stressed vowels. In this respect, Welsh seemed to break all the rules (about the acoustic realisation of stress) that had been formulated by non-Welsh-speaking phoneticians.
I then carried out a perceptual experiment to check the perceptual (and hence linguistic) validity of these speech-production findings. I used the word- pair "ymladd" (to fight) / "ymla+dd" (to tire oneself out). The first has lexical stress on the penult, the second on the ultima. this was the only stress-related minimal pair I could find in Welsh. I recorded a native speaker pronouncing these words, then used a waveform editing package to alter the duration of the /m/ in stages for each word. For each of the several word tokens thus produced, I used LPC synthesis to resynthesise the token, and overlaid a new, synthesised F0 pattern. I prepared a tape with these stimuli in random order and played it to ten native Welsh speakers. The results of analysing their responses confirm that the cue to perceived lexical stress was a complex interaction of the pitch pattern and the duration of the /m/. In several cases, a stimulus derived from "ymladd" was perceived as "ymla+dd", and vice versa, given appropriate /m/ duration and F0 pattern. So perceived lexical stress in Welsh depends to some extent on the duration of the post-stress consonant, as well as on the intonational patterns of Welsh.
I also carried out a study of rhythm in my recorded utterances, and found that if the penult were counted as the stressed syllable, then there was more of a tendency to isochrony than if the ultima were counted as stressed (isochrony is the tendency for stressed syllables to recur at regular interavals, however many unstressed syllables intervene - i.e., rhythmic regularity). So the penult functions as the keystone of the rhythmic unit in Welsh.
There's an interesting footnote in a thesis by J. Martin Rees ("Aspects of Welsh intonation", 1977, Univ. of Edinburgh). He observes that several non- Welsh-speaking English speakers, when played samples of Welsh words and asked to locate the stress, consistently chose the ultima, which is more acoustically prominent than the (stressed) penult. I tried this out in a limited way, and got the same general result. It's because stress in English is a matter of longer duration, greater intensity and (usually) higher F0, none of which necessarily occur on stressed penults in Welsh (but they do occur on unstressed ultimas, and on stressed monosyllables, which are a different story - i.e., they never lost their stress in the Old Welsh Accent Shift, as there was nowhere for the stress to shift to).
The internal structure of the syllable is often described in terms of the "Sonority Hierarchy", whereby the least sonorous sounds occur towards the outside of the syllable, and the most sonorous sounds occur in the centre of the syllable. A summarised sonority scale (and references to published work) can be found in R. Hogg and C.B. McCully, "Metrical Phonology: A Coursebook" (1987) Cambridge: Cambridge University Press. The scale (given on p. 33) is as follows (most sonorous elements first, with examples):
| low vowels (/a/), mid vowels (/e/), high vowels (/i,u/), "flaps" (/r/),|
laterals (/l/), nasals (/m,n,ng/), voiced fricatives (/v,dh,z/), voiceless
fricatives (/f,th,s/), voiced stops (/b,d,g/), voiceless stops (/p,t,k/).
This works for most syllables in Welsh: e.g., camp, celf, gair, beirdd. It's also true of English, among many other languages. But the syllables with (potential) epenthetic vowels in Welsh break this rule, so that (unless the epenthetic vowel is inserted) the outermost segment is more sonorous than its neighbour. A new (here, epenthetic) vowel causes a new syllable, hence a reset of the segment position.
The question might be asked: If -dr in "lleidr" splits because you cannot end a syllable with it, what is the mechanism for it to hold together in "lladron"?
The answer is that the second vowel forms the nucleus of a new syllable, and the /r/ syllabifies with this instead of with the first vowel, so it's no longer in a sonority relationship with the /d/, hence there's no longer any problem. (Either that, or both the /d/ and the /r/ syllabify with the second vowel: in this case, the sonority values of the segments follow the rules (most sonorous segment nearest the vowel) so again there's no problem).
Nick Kibre <email@example.com> wrote that:
Geraint Jones <geraint.jones@OXFORD.WOLFSON.AC.UK> responded:From various grammars I have gathered that there is a vowel-copying rule in Welsh which changes words like aml, pobl, cefn into amal, pobol, cefen
Nick Kibre further responded:Are there really rules that determine what intrusive vowels get used? Since they depend on accent (or do I mean dialect?) the rules will have to vary too. Because the intrusive vowel is always unaccented it is sometime hard to tell, but I think that I always put a dark "y" in aml, not an "a"; also an "y" in "lleidr", whereas I would bet that someone from Bangor would put an obvious "a" in it, and someone from Corwen an "i" (except that I may have the Corwen and Bangor accents the wrong way around...).
This rule, so far as I have been able to glean, affects only etymological monosyllables ending in consonant clusters with final -r, -l, -n. The copied vowel then makes the word behave just like a historical two-syllable word; hence the first syllable of llyfyr has schwa and the second the clear-y sound, as though the word had been disyllabic llyfyr in the first place. Similarly, in the southern dialects (and this is basically a southern thing) that get long vowels in open stress penultimate syllables, these new monosyllables get lengthened penultimates, e.g., [ke:ven] for cefn. (Schwa is never long; so the rule would not generally apply to llyfyr, though I don't know what would happen in schwaless Pembrokeshire dialects. You might get two clear y's with the first longer.)It's possible that I got the impression that there is a vowel copying rule by comparing "aml" as pronounced in an "a"-inserting dialect with "llyfr" as pronounced in a "y"-inserting dialect. Another issue is that, even if there is a copying rule, certain words with apparently copied vowels, like "pobol" have become fairly standard (I am assuming so since they are written this way) & may be used in dialects with no such rule. My only solid source on the copying issue is an (annoyingly short) section in Pedersen's Vergleichende Grammatik, which has very few examples.
Anyway, even if the vowel inserted is not a copy, I'm still curious as to whether it causes w & dark-y to become schwa-y.