Alternative analyses
Harris (1994)
Harris considers four dialects: system A is rhotic; B has linking but not intrusive [r]; C is the typical non-rhotic type with both linking and intrusive [r]; and D is the Southern US variety with [r] in red, bread and variably in very, but generally in neither linking nor intrusive contexts. We shall focus on his treatment of systems B and C.

Harris (1994) presents an element-based phonology, within a principles and-parameters framework grounded on ideas of phonological licensing. Harris assumes that approximant [ɹ] is composed of two elements, coronal R and neutral vocalic @ (the latter corresponding to the `cold vowel' vo of Kaye, Lowenstamm and Vergnaud 1985, 1990); a tap would contain only R.

In non-rhotic varieties, Harris assumes that no underlying /r/ remains in non-alternating forms like party, harp, but that alternating fear, soar contain a final floating /r/ (1). These varieties have also innovated the Non-Rhoticity Condition, which licenses R only in onsets.
(1)

The two elements of the constellation are not realized together: instead, @ is incorporated into a preceding nucleus, lengthening and lowering the pre-existing vowel, or appearing as a schwa offglide: the centring diphthongs are therefore derived segments for Harris (which, incidentally, rules his approach out immediately for my version of Lexical Phonology: since centring diphthongs appear in underived environments like fear, hair, poor, they must be present underlyingly). This absorption of @ also accounts for the restricted set of vowels, /ɑ: ɔ: ə/, which can precede linking and intrusive [ɹ] synchronically. As for the R element, if a vowel follows floating /r/, it brings with it an empty onset, onto which R docks: it is then licensed and can be pronounced. If no vowel follows, R stays floating. Whereas McCarthy (1991) proposed effectively both insertion and deletion processes, Harris therefore combines deletion with a variety of underspecification, and argues that this account holds for dialects with linking [r] only, or both linking and intrusive [r]; the latter simply have floating R in more forms, in spa as well as spar, for example. The difference between his dialects B and C `is thus purely a matter of lexical incidence' (1994: 250).

In representational terms, Harris's analysis of [ɾ] as R versus [ɹ]as@ plus R means the change from trill to tap to approximant in the history of English, which is generally described as weakening, translates into elemental complication. Furthermore, since the @ will realign regardless of the fate of R, cases where R is licensed by a following empty onset should produce taps, while most realizations are in fact approximants (see Foulkes (1997) for data from Newcastle and Derby): in that case, another source of @ must presumably be found to add to the relocated R. More conceptually, it is not obvious why a constraint like the Non Rhoticity Condition should suddenly appear in the grammar of a particular variety (see also Broadbent 1991). One might argue that such an analysis is simpler than one invoking an [r]-Insertion rule in some varieties and an /r/-Deletion process in others; but notions of simplicity are notoriously subjective, and issues of the universality of constraints and their motivation would have to be addressed in detail before a valid comparison could be made. Finally, as with Scobbie's and Donegan's work, there is the familiar question of how a gradual, sporadic, analogical extension of final [r] could lead to the extremely consistent appearance of intrusive [r], even in novel or foreign words.

What, then, of the restricted set of vowels which precede /r/ in non-rhotic varieties? As we have seen, Harris proposes that the @ element influences the quality and quantity of preceding vowels, and argues that this is preferable to an arbitrary statement that [r] happens to surface only after /ɑ: ɔ: ə/. I shall return to the general issue of arbitrariness later, but note for the moment that the insertion analysis is not unidirectional either. That is, it accepts both that /r/ affects the vowels, and that /ɑ: ɔ: ə/ determine the presence of [r], but regards the first conditioning factor as diachronic and the second as synchronic. In historical terms, /r/, and specifically its weakening and eventual deletion in codas, did affect the preceding vowels; synchronically, because of the intervening rule inversion, this translates into the conditioning of [r] by a particular set of vowels.

Harris (1994: 252±3) provides one interesting historical argument, noting that his analysis and mine make different predictions about relative chronology:
Rule inversion firmly implies that the intrusive C-system pattern is an off-shoot of an older non-intrusive B. By contrast, the floating-r account is entirely neutral on the question of historical precedence. It would be entirely consistent with the latter analysis if intrusive r arose independently of etymological linking r. In fact ... it would not be surprising to discover cases of intrusive r in rhotic dialects.

To take the orthoepical evidence first, Harris claims that Walker (1791) provides the first indication of deletion of coda /r/, and of linking [r], but that `A generation before John Walker's description of smooth versus rough r, Thomas Sheridan was castigating Londoners for inserting r after the final -vowel of words such as and (sic).' This suggestion that intrusive [r] may predate the linking type clearly goes against all predictions made by the rule inversion account. However, let us examine Sheridan's comment more closely. It occurs in his Lectures on Elocution, and is part of a passage dealing with `vice[s] in the cockney pronunciation' (Sheridan 1762: 34). His main objection is to `the changing the sound of the last syllable of words ending in ow, where ever it is not sounded like a dipthong, but like a simple o, (which is always the case when the last syllable is unaccented) into er ± as feller for fellow - beller, holler, foller, winder, ± for bellow, hollow, follow, window. As also adding the letter r to all proper names ending in a unaccented, as Belindar, Dorindar, for Belinda, Dorinda' (ibid.). We can draw several conclusions from this, and as it turns out, none are detrimental to [r]-Insertion. First, it might seem chronologically significant that Sheridan mentions only intrusive, and not linking [r]; however, we have already seen (6.3.1 above) that, in his Dictionary and Grammars (see, for instance, Sheridan 1780, 1781), Sheridan's intentions are prescriptivist rather than descriptivist, and he deliberately archaizes the pronunciations he recommends: his claims that is never silent must be interpreted in this light. It is also notable that, although he does not allude to linking [r] in the Dictionary or Grammars, intrusive [r] does not figure in these works either. On the other hand, in the Lectures on Elocution, Sheridan is anecdotally selecting certain features of `vulgar speech' and warning his audience against them. He may simply have hit upon intrusive rather than linking [r] first - perhaps as an Irishman himself, and therefore presumably originally a rhotic speaker, the sin of commission impressed itself on him more forcibly than the sin of omission whereby etymological [r] was dropped before pauses and consonants. Rhotic speakers today still seem particularly sensitive to intrusive [r]. Alternatively, he may mention intrusive [r] because it is a relatively recent innovation, just gaining ground in the speech community, as opposed to linking [r] which may already have been fairly well embedded. This is not pure speculation: the orthoepical works surveyed in 6.3.1 above indicated that clear evidence for /r/-Deletion, and for linking [r], dates from the early eighteenth century - a generation before Sheridan, in fact.

We might, then, taking Harris's part, object that Sheridan's evidence indicates a broader distribution of intrusive [r] than in current non-rhotic varieties, supporting its independence from linking [r]. That is, when Sheridan writes `winder, Dorinder', how are we to know that these were restricted to prevocalic position? The answer is that, although Sheridan may not make this plain, others do: Ellis (1849: 37), for instance, comments that `An r is very often inserted by Londoners after ɑ [= [ə] AMSM], a', ɔ., when a vowel follows; thus ``the lawr of the land, Jemimar Ann, Sarahr Evans.'' This has given rise to the idea, that the Londoners pronounced law, Sarah as lawr, Sarahr, which is not the case.' Sheridan's contemporary Elphinston (1787; quoted by Jespersen 1909: 370) also explicitly connects linking with intrusive [r]: after discussing [r]-loss, he comments that `Dhe same cauz (febel vocallity in dhe end) haz made Grocenes [i.e. vulgarity] assume r in (dhe colloquial) idear and windowr, for idea and window.' Furthermore, the hypothesis that intrusive [r] was innovated independently of and earlier than linking [r] is not borne out by the present-day distribution of vowels. If intrusive [r] were earlier, and if /r/ is floating R plus nuclear @, with the latter affecting the quality of all preceding vowels, why is there a centring diphthong and prevocalic [r] in beer is, purist, but not in seeing or viewing? This contextual restriction means intrusive [r] must be connected with, and even be an offshoot of a prior linking type.

Of course, the weaker form of Harris's argument, which holds that evidence for linking and intrusive [r] begins to appear roughly contemporaneously, is not problematic: rule inversion, and therefore intrusion, would be predicted to be a rapid, if not immediate response to deletion for some speakers ± and recall that weakening and sporadic deletion had been going on since the mid-seventeenth century. There is also a theory internal argument from Lexical Phonology here. My assumption that underlying representations for underived forms are typically identical to their lexical representations will be violated in varieties with linking [r] only. I have also argued that the constraints of LP, if applied rigorously, might be of diachronic rather than synchronic relevance: that is, their violation might motivate some historical restitution; and this might be precisely such a case. If some historical development, like the loss of coda [r], means underlying representations conflict with the usual conditions on them, they might not persist for long ± such a stage might be predicted to be transient. It is interesting in this connection that varieties with linking [r] only are rare today, and that evidence for intrusion comes hot on the heels of evidence for [r]-loss.

Harris also argues that intrusive [r] is not restricted to non-rhotic dialects, as conventionally assumed, but that it appears also in `some present-day conservative rhotic dialects ... [as] demonstrated by spellings such as for ' (1994: 253), and presumably also hyperrhotic pronunciations like china[ɹ], banana[ɹ], [kɑ:ɹki] khaki. Again, Harris (1994, fn. 40) assumes that rhotic speakers have innovated underlying /r/ in the relevant forms. As for the motivation for this sort of development, Harris argues for a progressive disfavoring of final schwa in English; this disfavoring began with the widespread loss of final schwa in Middle English, but different strategies have had to come into play in more modern times, following the relatively recent borrowing of a fairly extensive set of words including sofa, comma, Laura, Sheena, banana, china, vanilla, America. In both non-rhotic and rhotic varieties, in Harris's view, the principal strategy is the internalization of underlying /r/, floating or not.

However, we can maintain rule inversion in non-rhotic dialects, and still account for the rather different phenomenon of hyperrhoticity in rhotic dialects in a variety of ways. For instance, final [ɑ: ɔ: ə] may attract [r] in rhotic British accents because the majority of British English speakers are non-rhotic, and therefore such vowels will frequently be heard with intrusive [r]. That is, rhotic speakers may mislearn words by assuming underlying /r/ because they hear alternating [r] ~ ∅ from non rhotic speakers: this sporadic, analogical extension corresponds to the analysis Harris, Scobbie, Donegan and others have proposed for non rhotic varieties. But the characterization of extraneous [r] for rhotic speakers as a purely lexical development need not extend to non-rhotic dialects. First, as noted above, [r]-Intrusion in non-rhotic varieties is typically much more regular and systematic, and seems best suited to a rule-governed account. Second, some evidence for our mislearning hypothesis comes from America, where, conversely, the majority of speakers and the principal standard variety, General American, are rhotic, and where this has had parallel but opposite effects on the non rhotic minority, including the well-known re-rhoticization of New York City (see Labov (1972) and 6.6 below). Finally, clear evidence for regarding intrusive [r] and hyperrhoticity as separate phenomena comes from their distribution: intrusive [r], as we know, occurs only intervocalically, and after a restricted set of vowels. Hyperrhotic [r] follows the same set of vowels, reflecting its possible origin in imitation of non-rhotic speakers, but is not restricted to intervocalic position: Wells (1982) cites khaki, camouflage with [r] between [ɑ:] and C, while rhotic speakers may have categorical final [r] in schwa words like comma, idea, Anna, Laura.

An alternative, dialect-internal explanation for words with final schwa is outlined in Hughes and Trudgill (1979: 32). This hypothesis relies on the fact that a number of words with final schwa were borrowed into English either while [r] was being lost non-prevocalically, or thereafter. These words are generally assimilated to other phonological classes, perhaps because of disfavoring of final schwas of the type suggested by Harris, with different options being exercised in different dialects. In non-rhotic varieties, schwa-final forms were treated in the same way as those with historical /ər/, and therefore, after rule inversion, developed an [r] ~ ∅ alternation controlled by [r]-Insertion. In rhotic dialects, where the retention of final [r] meant there were no schwa-final words, strategies varied. For instance, in Southampton, comma, banana, vanilla words adopted the pattern of butter, letter, with categorical final [r], giving [bənɑ:nər], [vənɪlər]; this accounts for West Country hyperrhoticity. In Bristol, on the other hand, comma, banana, vanilla seem to have been assimilated to the pattern of apple, bottle, with the other English liquid, giving intrusive [l].
In short, neither Harris's historical evidence nor his dialectal data support his contention that linking and intrusive [r], in varieties with both, are independent. It is worth noting finally Harris's (1994: 254) criticism of the notion that intrusive-r dialects are the direct descendants of a non-rhotic B-type system which lacks the phenomenon. The germ of this idea seems to be buried in a prescriptive myth, according to which non-standard dialects are deviant outgrowths from a central standard stem whose phonology somehow faithfully mirrors the orthography.

This is an extraordinary statement, and seems to reduce to the assertion that it is prescriptivist to hypothesise for present-day varieties with both linking and intrusive [r] a historical ancestor with linking [r] only. My [r]-insertion account does make this assumption, but purely diachronically: I have presented contemporary evidence showing that a linking-only stage existed, albeit relatively briefly. Even this was only an intermediate stage: all non-rhotic dialects are, on a longer historical view, derived from some rhotic ancestor. Indeed, in rhotic dialects, the phonology of [r] `faithfully ... mirrors the orthography'. But surely this is not a prescriptivist statement, partly because it simply reflects a historical fact, and partly because, by a typical quirk of linguistic fate, rhotic varieties of English, at least in Britain, are now typically non-standard ± as is the appearance of intrusive [r], which is now becoming the non-rhotic norm and impressing itself successfully on RP.

Perhaps the best gloss we can put on Harris's statement is that different varieties should not necessarily be derived synchronically from a common core. It is precisely because I do hold to this view, and because of the conditions on my model of Lexical Phonology, that I must of necessity propose [r]-Insertion in varieties with linking and intrusive [r], and [r]-Deletion in dialects with linking only (thus, ironically, Harris analyses these two types of variety as much more similar than I do). The arguments against underspecification above rested on similar grounds, namely that underspecification permits common underlying systems for different varieties where they are not warranted. But ruling out underspecification is one good reason why I could not in any case incorporate Harris's floating /r/ analysis into my Lexical Phonology. This is not underspecification in its usual sense: in the normal case, features are absent and filled in during the derivation, whereas here, the floating elements are present from the start, but are attached (optionally for R, obligatorily for @) later. Nonetheless, there is absence of attachment in the early stages; and more worryingly, if R is not adjacent to an empty onset, it is left floating, and not realized on the surface; yet it is not deleted. This is perhaps a mirror image of underspecification (present but unwanted material, rather than absent and required material), but the problems it gives rise to are likely to be the same. Similarly, Giegerich's (in press) analysis of [ɹ] and schwa as arising in different contexts from the same empty underlying melody will not be pursued here, as my prohibition on under specification equally rules it out.

مواضيع ذات صلة

Retrospect and p[r]ospect

Lexical Phonology and English /r/

Modelling the past

Using the past to explain the present

Synchronic arbitrariness and diachronic transparency The problem

Alternative analyses McCarthy (1991)

Alternative analyses Scobbie (1992), Donegan (1993)

Alternative analyses

Rule inversion and [r]-Insertion

Non-rhotic /r/: an insertion analysis An orthoepical interlude

English /r/: a brief outline

English /r/ Introduction