generally the same in all positions in Ml. and Mn. W., even local usage agreeing; except in shortened words § 51 vi.
ii. But in diphthongs many changes must have taken place. As a “vowel before a vowel” was long then as now, trṓ-ï must have had a long ō, so that, when first contracted, it was still long; it remains long in Montgomeryshire; thus the short o in trŏi̯ is probably late. Similarly short ĕi for e-i, ău for a-u, ŏu for o-u. Other diphthongs also probably differ, and we can infer nothing as to Ml. W. quantity in diphthongs from the Mn. W. pronunciation.
§ 56. i. The quantity of a vowel in British determines its quality in Welsh; but its quantity in Welsh depends, as we have seen, on the consonantal elements which follow it in the syllable.
ii. A short accented vowel in Brit. or Latin ollowed by a single consonant was lengthened in Welsh; thus Brit. *tălos gave tâl, § 51 iv Note, *rŏtā (cognate with Lat. rŏta) gave rhōd, Lat. sŏnus gave sôn, etc. This took place after the change in the quality of long vowels, for while original ā gives aw § 71, long ā lengthened from ă remains â. It also took place after the reduction of pp, tt, cc into ff, th, ch, for the latter are treated as single consonants for this purpose; thus Lat. saccus became *saχos with single χ, which gives sach (≡ sāχ) in Welsh. Long vowels remained long, as in pûr from Lat. pūrus. On the other hand, a vowel originally long was shortened before two consonants; thus the ō of Lat. fōrma became ǖ, which was shortened in the Welsh ffŭrf. Hence the general rule § 51 i, which probably goes back to Early Welsh and beyond; for the lengthening of short vowels originated at the time of the loss of the ending, and is due to compensation for that loss.
iii. There is no reason to suppose that this lengthening took place only in monosyllables. Thus O. W. litan ‘wide’ (: Gaul. litanos in Κογκο-λιτανος, Smertu-litanus, etc., Ir. lethan) was probably sounded *llỿ-dā́n, while guinlann was doubtless *gwinl(l)ắnn. In Ml. W. when the ultima became unaccented this distinction was lost, the a of llýdān being shortened, § 49, and the nn of gwín-llann being simplified, § 27 ii. The rule forbidding the rhyming of such a pair was handed down from the older period, and is given in r.g. 1136; such a rhyme is called trwm ac ysgawn ‘heavy [with 2 consonants] and light [with one]’. But the bard’s ear no longer detected any difference in the unaccented ultima; he is therefore instructed to add a syllable to find out whether the syllable is “heavy” or “light”: kallonneu (ll ≡ l‑l) is given as an example to show that the on(n) of kallon [sic] is “heavy”, and amkaneu to show that the an of amkan is “light”. The Early Ml. bards avoid trwm ac ysgawn; but in the first poem in b.b., where the rhyme is ‑ann, several forms in ‑an occur, as imuan 1 (: gwanaf ‘I wound’), darogan 7 (: canaf ‘I sing’), which shows that
