The Scottish Vowel Length Rule and Low-Level Lengthening
Experimental evidence
As we have already seen, there are very few experimental investigations of vowel length in Scots and SSE. Leaving aside the limited and problematic study of McClure (1977), the only systematic experimental work on the supposed effects of SVLR is reported in two papers (Agutter 1988a, b) which embody an attack on the unity of SVLR and its restriction to Scots dialects and SSE. I shall briefly outline Agutter's investigation below, before proposing an alternative analysis of her data which corroborates the existence of SVLR as a productive but Scots specific process.

Agutter obtained data from two male and two female SSE speakers, all middle class and from Edinburgh, and from two RP speakers, one male and one female, each from a different part of the UK. All were university students aged between eighteen and twenty-three. Each in formant, recorded individually, produced a number of English mono-syllables in an invariant frame sentence `I say WORD sometimes.' The vowels tested were /ai i/, which Agutter assumes should undergo SVLR if there is such a process; /au/, which as noted earlier has an unclear status with respect to SVLR; /ɔ/, which Agutter asserts is consistently long; and /ɪ/, which is consistently short. These vowels appeared in the two sets of contexts in (1).
(1) 

Spectrograms of the monosyllables were produced, and vowel durations calculated by hand to the nearest centisecond. Weighted average values for each vowel for all speakers of each accent and for each context were then calculated, by multiplying average lengths per vowel per informant by the ratio 13.0/A, where 13.0 is an arbitrary average vowel length and A is the overall average vowel length for that informant. This weighting process is intended to allow a more meaningful comparison of the two accent groups by reducing the potentially distorting effect of idiolectal variation, which might be particularly severe given the small sample size. However, Agutter's weighting procedure may not be entirely valid for her results, since the technique used involves an assumption that any variation found will be normally distributed. Given that SVLR, as an accent-specific process affecting only certain vowels in certain contexts, would contravene this expectation and produce a skewed distribution, weighting might in fact mask exactly the variation Agutter is testing for.

There are other difficulties with Agutter's experiment. For instance, it is unclear how representative the informants were of their respective populations: in particular, SSE speakers are variably influenced by RP, and Agutter does not tell us whether her SSE informants had non-Scots characteristics like the /æ/ ~ /ɑ/, /ɒ/ ~ /ɔ/ and /ʊ/ ~ /u/ oppositions.

Furthermore, the distribution of informants across accent groups is unbalanced, with four SSE and two RP speakers, making further statistical testing difficult. Certain contexts were also unavailable for investigation due to accidental gaps in the English lexis; no monosyllables were found for /ɔ/ before /r ð s f/, /au/ before /v b p f/ or /i/ before /b/. The relative unfamiliarity of some of the words used, such as mouthe, gawp and dowd, is reflected in a number of gaps in the data, resulting from unusable tokens. The use of nonsense syllables would have solved the first problem, but whether it would have alleviated or exacerbated the second is debatable.

Agutter sees her results as inconsistent with a formulation of SVLR as Scots-specific, since they suggest that all vowels tested, for speakers of both groups, lengthened before all voiced consonants, albeit with consistently slightly greater duration in SVLR long environments. From these findings, Agutter concludes that `SVLR is too restrictive in the set of contexts which it designates as long contexts in Scots' (1988b: 16); more radically, she argues that `the context-dependent vowel length encapsulated in SVLR is not and perhaps never was Scots-specific' (1988b: 20).

We considered evidence above that SVLR was introduced historically only into Scots, and I shall show below that it is still restricted to Scots and SSE today. Agutter ascribes all the vowel length variation in her results to a single process; if this process is SVLR, then it must apply before all voiced consonants, and in RP. I believe that a more enlightening account of Agutter's data can be given if we assume that two overlapping processes are at work: SVLR in Scots and SSE, and a pandialectal low-level phonetic lengthening rule operating before all voiced consonants.