SVLR and diphthongs
The question of whether the historical SVLR affected diphthongs or only monophthongs is not one which is usually asked: the alternation of [Λi] ~ [a:i] in tide ~ tied, featuring the only qualitative distinction in the SVLR set of vowels, has become a diagnostic of the modern process, and any attempt to exclude it from the historical version might seem perverse. I would, nonetheless, like to suggest that the historical SVLR was in fact restricted to monophthongs.

Part of the discussion here necessarily pre-empts aspects, on the synchronic SVLR, without the fuller discussion which will follow there. It seems that, in SSE and Modern Scots dialects, two of the three true diphthongs, the vowels of bout/now and noise/boy, are unaffected by SVLR. The former, for some varieties, is consistently long /a:u/; for others, it is consistently short /Λu/; and bear in mind that in Scots, this diphthong is marginal at best, given the failure of Middle English /u:/ to diphthongize in the GVS. The /ɔ:i/ vowel would appear to be consistently long.

What, then, of the final true diphthong? Turning to Scots dialects first, recall that there is not in fact one diphthong in this general area: there are two, namely the [Λi] of way, pay ( Middle English /ai/), and the [Λi] ~ [a:i] of tide ~ tied ( Middle English /i:/). The former is uniformly short in Scots, and is not affected by synchronic SVLR; the latter indubitably is. To distinguish the two, I shall propose underlying /Λi/ for the pay, way vowel, which does not vary in quality, and /ai/ for the tide ~ tied one, which does.

There are several reasons for thinking that none of these diphthongs was involved in the historical length neutralization described by SVLR. None are conclusive, but together, they are indicative of a historical restriction of the process to monophthongs. First, there is the quality difference itself: why should the diphthong be the only vowel to do more than lengthen by SVLR? Without SVLR, we can still derive the [Λi] ~ [a:i] alternation, since we have already seen that the GVS diphthongization of /i:/ must be assumed to have undergone an initial, partial shift to [Λi] before lowering later. I am here following an undeveloped suggestion from Johnston (1997a: 94), whereby `A diphthong of this type would tend to have the longest V1 in the most sonorant environment: this would also favour a peripheral realization. Now, the association with Aitken's Law might not be original.'

That is, lowering (giving greater peripherality) would be most likely where greatest lengthening was favored; for Johnston, that means perhaps only in final position. He therefore proposes a phonemic split between the vowels of his lexical sets BITE and TRY, and there are indeed cases where such a split would seem to be justified: some Scots (and SSE) speakers have [Λi] in words like cider, spider, idle, pilot, title, while others have long [a:i] (and not all words in the list need have the same vowel, for any one speaker). Milroy (1995) reports similar variant forms from his investigation of [Λi]/[ei] versus [ai] in Newcastle: [ai] appeared in SVLR short environments such as pride, site, cycle, predominantly in middle-class speakers, while working-class speakers tended to produce [Λi]/[ei] in forms like tied, higher, drive, sky, buy, my. I shall return to this marginal contrast below. But equally, there are many cases where the SVLR generalization holds robustly for this diphthong in Modern Scots and SSE; whatever the historical development of [Λi] ~ [a:i] might have been, it has now been incorporated into the synchronic SVLR.

We shall see that no current version of SVLR, including the one presented here, can give a formal synchronic description of the input vowels and the effect of the process in a way which includes [Λi] ~ [a:i], while also explaining why no other diphthong is involved. However, if the diphthong, like other diphthongs, was excluded from the process historically, but unlike other diphthongs, had its variant realizations reinterpreted as resulting from SVLR later, we should perhaps not be surprised by its apparent failure to fit the SVLR specifications in full.

There is one final aspect of the behavior of [Λi] ~ [a:i] which perhaps becomes clearer under this analysis. This is the similarity of the Scots/ SSE alternation with the North American process known as Canadian Raising (Chambers 1973, Gregg 1973), an alternation of [Λi]/[əi] and [Λu]/ [əu] before voiceless consonants, with [ai] and [au] elsewhere. Donegan (1993: 121), however, notes that this pattern `is not limited to Canadian dialects, nor is it necessarily a raising': she shares with Gregg (1973) the view that the process is more likely to reflect lengthening and lowering before voiced consonants and boundaries than shortening and raising in voiceless contexts. In that case, lowering in Canadian varieties (and in Maryland, Minnesota, Virginia, Michigan, Georgia and the sundry other places it has been reported) could be seen as an extension of SVLR to an extra set of lengthening environments, namely voiced stops.

This, however, raises more questions than it solves. Why should only diphthongs be affected, when all Scottish versions of SVLR include at least some monophthongs? One might say that the diphthongal alternation, involving quality as well as quantity, was more easily perceptible; but that hardly counts for /au/, which is also affected in the North American cases, but which does not undergo SVLR in Scotland. On the other hand, if we see this as a separate change affecting Scots and the North American varieties, the differing inputs are unsurprising, since in Scots at the time of GVS the /au/ diphthong was extremely marginal and almost unattested in final position, where lengthening may have taken hold. The different environments of these lengthening processes also follow naturally: in Canadian and other North American varieties, as we shall see later, vowels already lengthen finally and before voiced consonants, and the diphthongal alternations are only perceived as different because quality as well as quantity is involved. But only in Scots, and in SSE as it developed, was there a further process with more radical phonological consequences, the Scottish Vowel Length Rule, for the diphthongal change to collapse with. I therefore assume in what follows that the historical SVLR was restricted to monophthongs. I also propose underlying /Λi/ in Scots pay, way, versus lengthenable /ai/.