Length, tenseness and English vowel systems
Let us begin with RP and Scots/SSE. In RP, as we have already established, there are six pairs of vowels, as listed in (1), the members of which are distinguished partly by length (with the vowels on the left consistently longer), and partly by quality, or tenseness (with long vowels more peripheral and frequently also diphthongized).
(1) 

In SSE and Scots dialects, this dual distinction of quality and quantity is not operative. The /ɑ:/ ~ /æ/, /ɔ:/ ~ /ɒ/ and /u:/ ~ /ʊ/ oppositions are entirely lacking, and members of the remaining pairs, /i/ ~ /ɪ/, /e/ ~ /ε/ and /o/ ~ /Λ/, are distinguished primarily by quality, quantity being predict able by the Scottish Vowel Length Rule. If we accept that [+ tense] is the crucial specification for SVLR input vowels, it follows that Scots and SSE vowels are underlyingly, contrastively [+ tense] or [- tense]. In a full-entry theory of the lexicon, all vowels will also be short (or singly attached, in autosegmental terms); and in varieties where SVLR affects only vowels of a certain height, there will also be underlyingly long vowels in the system. Nonetheless, the crucial dichotomizing feature for the Scots and SSE vowel system is tenseness.

We have already noted that RP vowels fall into two classes, long tense and short lax: the question is how these are to be analyzed at the underlying level. Three approaches have been adopted in the literature: in SPE, [± tense] is the underlyingly relevant feature, and a late redundancy rule links [+ tense] with [+ long] and [- tense] with [- long]; Halle and Mohanan(1985) take the opposite point of view, with length bisecting the underlying system and tenseness introduced subsequently; and Halle (1977) proposes that both length and tenseness are specified underlyingly.

There are arguments against at least the first two analyses. If only [± tense] is underlyingly contrastive, the stress rules will have to be sensitive to tenseness rather than length, going against the tendency to make prosodic processes responsive to prosodic features. There is also the historical problem that, assuming [± tense] was introduced into the vowel system in Middle English, as I argued on the basis of Open Syllable Lengthening, we know of a process which might allow this feature to supplant length as underlyingly relevant in SSE, namely the SVLR, but have no equivalent in the history of RP. We might then prefer to assume underlying length only; however, Halle and Mohanan (1985) require various additional lexical tensing rules (they propose, for instance, independent but mysteriously nearly identical pairs of rules like Stem-Final Lengthening and Stem-Final Tensing, or Prevocalic Tensing and Prevocalic Lengthening), and encounter certain derivational problems unless [± tense] is introduced extremely early in the derivation by redundancy rule. In that case, it might just as well be specified under lyingly, which is Halle's (1977) position. His assumption that [± tense] and length are independently specified underlyingly largely follows from his adoption of long-tense and long-lax low vowels for American English; but even without such vowels, we might wish to consider tenseness and length as independent because they are so frequently manipulated separately across dialects and across time: see Giegerich (1992: 3.7), for instance, who argues that length distinctions in Australian English do not correlate straightforwardly with tenseness.

Our conclusion, then, is either that length alone is underlyingly distinctive in RP, or that both length and [± tense] are. This situation differs clearly from that of Scots and SSE, at least in varieties where SVLR has been fully implemented: here, only [± tense] is distinctive and length, which was contrastive historically, has been neutralized by the SVLR. Length will have only one underlying value, whereas both length and tenseness have both feature specifications available at the underlying level in RP. If each variety is analyzed on its own terms, we therefore produce an underlying dialect difference.

Our next question is whether, in a Lexical Phonology, this analysis could be revised to derive the two dialects from the same underlying system. The usual course of action in the SGP tradition would involve reanalyzing SSE, which has innovated in this case, in accordance with RP, and the crucial step would therefore be to make length underlyingly contrastive in SSE. The simplest way to achieve this is to assume that no rule inversion took place in the history of SVLR, making the synchronic process, like its antecedent sound change, a bipartite length neutralizing process as in (2).
(2) 
 

If we decide that [± tense] should be underlyingly distinctive in RP and SSE, subrule (2b) can be modified to refer to short tense vowels; if not, some other way must be found of excluding /ɪ ε Λ/.

Of course, the only motivation for this analysis is the desire to derive RP and SSE from a common set of underlying features: it would not be preferred if each dialect were considered in its own right, from either a synchronic or a diachronic point of view. Furthermore, the proposal of a system with length underlyingly distinctive, but neutralized on the surface, creates obvious problems of learnability: a child acquiring SSE will be required to divide her vocabulary along synchronically opaque lines by reversing the historical SVLR in order to internalize vowels of the appropriate length in lexical items at the underlying level. For instance, the learner will hear words like [fat] fat and [het] hate, with surface short vowels, and have to decide whether these have underlyingly long vowels which shorten by SVLR or short vowels which remain short in this short context, and conversely words like [fa:r] and [he:r] with surface long vowels, which might reflect underlyingly long vowels with length retained, or lengthened short vowels. The learner clearly cannot obtain the necessary information from observation. Furthermore, if SVLR operates on Level 1, then the Derived Environment Condition will not permit it to lengthen or shorten vowels in underived environments; and the vast majority of forms involved (including fat, hate, far and hair) will be underived. Even if SVLR applies on Level 1 only in environments specifically derived on that stratum, underived applications on Level 2 are generally disfavoured from a learnability point of view by the Alternation Condition.

A Lexical Phonology without underspecification therefore performs relatively realistically with respect to the analysis of dialect variation, permitting and perhaps even requiring limited differentiation at the underlying level. The distinctions established here will also extend in various ways to other varieties of English. Even within Scots there is a difference between the core, central Scots dialects where SVLR applies throughout the tense vowel system, and where only [tense] is relevant underlyingly, and those where some vowels are consistently long on the surface, and therefore at the underlying level too. Similarly, Lindsey (1990) argues for an underlying distinction between RP and GenAm, based on the treatment of loanwords. He points out that GenAm speakers typically assign to the stressed syllables of loanwords the vowels of hod (= balm), hayed, heed, hoed and who'd, corresponding to ortho graphic a e i o u respectively, while RP speakers typically interpret the same orthographic representations as the had, head, hid, hod, hood vowels, albeit with more exceptions and a variable additional process of lengthening in open syllables. Lindsey argues that the motivation for these different strategies, `given cross-dialectal uniformity of spelling, must be sought in differing phonological representations' (1990: 108). He proposes that, universally, languages will prefer to assign unmarked or default feature values to loans. Following SPE markedness conventions, this would predict that languages with their underlying vowel systems structured as tense versus lax will assign [+ tense] values, which is what we find in GenAm. On the other hand, since shortness is unmarked relative to length, vowel systems dichotomized according to length will have short vowels assigned to loans, as is the tendency in RP. Lindsey speculates that the inconsistencies in RP may in fact indicate that both [± tense] and length are underlyingly relevant: since both tenseness and shortness are unmarked, there is scope for a markedness clash, which might account for the more variable behavior of RP speakers. Lindsey also provides further evidence for the more central role of length in RP: for instance, `the difference between the duration of long-tense vowels and short-lax vowels is greater in RP than in American dialects' (1990: 113), while a comparison of the inherent vowel duration specifications for American and British speech synthesis systems reveals an increased long/short ratio for RP. Again, collapsing this underlying distinction simply to unify the system for the two varieties would lose Lindsey's explanation for the different treatment of loans.