Non-rhotic /r/: an insertion analysis
An orthoepical interlude
It is generally assumed that non-rhotic accents of English became non rhotic via three sound changes, which are traditionally listed separately and sequentially (Wells 1982): these are Pre-/r/ Breaking, Pre-Schwa Laxing/Shortening and /r/-Deletion, and are shown in (1). In brief, a schwa is inserted between any vowel and [r]; the pre-existing long vowels shorten and lax before this new schwa; and finally the [r] drops.
(1) 

Wells (1982: 210ff.) assumes that these changes took place after about 1750, since their results can be observed in RP and other south-eastern English varieties; in the southern hemisphere extraterritorial Englishes of Australia, New Zealand and South Africa, where English was introduced after this period; and in the non-rhotic accents of the earliest North American settlements, but not in rhotic General American. Wells also notes that the relative chronology implied by (1) may not be appropriate: while Breaking must have preceded /r/-Deletion, Laxing could equally well have followed it. However, Wells's absolute chronology must also be cast into doubt, since there is evidence for at least Breaking from well before 1700, and precisely because the first American colonies are non-rhotic, and seem likely to have been at least partially so at the time of settlement. Breaking and /r/-Deletion were probably gradual changes which were under way, producing variants in the speech community, before 1700.

There is a considerable amount of orthoepical and other evidence to support this hypothesis. For instance, John Hart (1569) gives phonetic transcriptions of [feiër] `fire', [meier] `mire', [o'er] `oar', [piuër] `pure', [diër] `dear' and [hier] `here'; assuming that [ë] and [e] indicate something close to schwa, we can conclude that Breaking was at least an option by the late sixteenth century. Jespersen (1909: 318) notes parallel sixteenth century spellings like for earlier ; and similar spellings like ( OE ), ( French ) and ( ME ) have been maintained, indicating that Pre-/r/ Breaking must have been well established before the fixing of the modern spelling system. By the eighteenth century, evidence for Breaking is much more explicit and commonplace. For instance, Abraham Tucker (1773) provides a special symbol ʋ and notes that `it is commonly inserted between ``ē, ɪ̄, ō, ū'' and ``r'', as in ``there, beer, fire, more, poor, pure, our,'' which we pronounce ``theʋr, biʋr, fʋiʋr, moʋr, puʋr, ʋuʋr''' (1773: 14). It is quite clear that Tucker's ʋ is intended to correspond to schwa: he recognizes the vowel as a ubiquitous casual speech marker, noting that `there are none of the vowels but what are often changed into ``ʋ'' in common talk' (1773: 15); and he also identifies it with the schwa used in hesitations, observing that `we can draw it out to a great length upon particular occasions, as when the watchman calls ``past ten ʋ-ʋ-ʋ clock,'' or when a man hesitates till he hits upon some hard name, as ``This account was sent by Mr. ʋ-ʋ-ʋ Schlotzikoff, a Russian''' (Tucker 1773: 14). Tucker's observation that `This short ``ʋ'' is easiest pronounced of all the vowels ... and therefore is a great favorite with my country men, who tho not lazy are very averse to trouble, wishing to do as much work with as little pains as possible' (ibid.), also seems highly appropriate for schwa.

While the long high and mid monophthongs were developing into centring diphthongs before /r/, low /ɑ ɔ/ seem to have lengthened in the same context. The earliest evidence for this lengthening is probably from Cooper (1687), who observes that ɑ is short before word-final /r/, as in bar, car, tar, but long before /rC/, as in barge, carp, tart, while o is long before certain specific final clusters including /rn/ horn and /rt/ retort. This lengthening seems subsequently to have spread, as evidenced by Mather Flint (Kökeritz 1944). James Mather Flint is a fascinating character; born in the early years of the eighteenth century and brought up by his uncle in Newcastle, he spent most of his life in Paris after his fanatically Jacobite family ¯ed to France in 1717. Mather Flint became a Catholic priest, but also published (in 1740) a guide to English pronunciation for French speakers, in which he observes that /r/ ‘rend un peu longue la voyelle qui le précede’ (Kökeritz 1944: 41), giving examples like barb, guard, arm, yarn. Similarly, o is said to lengthen before rd, rk, rm, rn (Kökeritz 1944: 20). Mather Flint also notes the difference between words like name and those like care, chair, bear, where he observes that the vowel is `un peu ouvert' on account of the followingg /r/, thus providing an early description of modern [eɪ] versus the centring diphthong [εə]. The three pronouncing dictionaries from the 1760s and 1770s surveyed by Beal (1993) reveal that long [ɑ:] was primarily found before /rC/ clusters, but might also appear before final /r/ in certain lexical items, such as far, mar and tar. Finally, Walker (1791) indicates that /ɑ ɔ/ are categorically long before final /r/, pointing out in connection with a that `we seldom find the long sound of this letter in our language, except in monosyllables ending with r, as far, tar, mar, &c. and in the word father' (1791: 10). More generally, Walker notes that when a and o `come before double r, or single r, followed by a vowel, as in carry, marry, orator, horrid, arable, forage, &c. they are considerably shorter than when the r is the final letter of the word, or when it is succeeded by another consonant, as in arbour, car, mar, or, nor, for' (Walker 1791: 15). Walker equates this variant of o, `the long sound produced byrfinal, or followed by another consonant, as for - former' (1791: 22) with the vowel written ,as in laud.

The merger of earlier /ɪr/, /Λr/ and /εr/ as /з:r/ in bird, word, herd words also seems to have been under way by the eighteenth century; indeed, Jespersen (1909: 319) argues that /ɪr/ and /Λr/ had begun to coalesce by around 1600. Mather Flint (KoÈkeritz 1944: 72) again provides relevant evidence, including fir and fur in a list of pairs of words which sound identical but are spelt differently, and noting (KoÈkeritz 1944: 70) that the same vowel appears in herd, search, dirge, girl, earn, learn, disperse, rehearse, earth and birth. Although he does not explicitly discuss the quality of this vowel, he observes that the earn, learn vowel is long: this presumably generalizes to the other words in the class, and indicates that the merged reflex of Middle English /ɪr Λr εr/ had lengthened by the mid eighteenth century. Sheridan (1786: 28±9) also reports this merger as ongoing: he blames the actor-manager Garrick, `who, according to the Staffordshire custom, ... called gird gurd, birth burth, firm furm. Nay he did the same when the vowel e preceded the r, heard was hurd, earth urth, interr’d inturr’d, &c.' Sheridan notes with disapproval that `His example was followed by many of his imitators on the stage' (1786: 29). Finally, discussing i, Walker (1791: 15) notes that:

The letter r ... seems to have the same influence on this vowel, as it evidently had on a and o... the i, coming before either double r, or single r, followed by a vowel, preserves its pure, short sound, as in irritate, conspiracy, &c. but when r is followed by another consonant, or is a final letter of a word with the accent upon it, the i goes into a deeper and broader sound ... . So fir, a tree, is perfectly similar to the first syllable of ferment, though often corruptly pronounced like fur, a skin. Sir and stir are exactly pronounced as if written sur and stur.

This merger, along with the processes of lengthening, Pre-/r/ Breaking and Pre-Schwa Laxing, produced a restricted inventory of vowels before /r/; distributional restrictions of this sort are found in rhotic varieties, as witness the General American homophony of Mary, merry, marry, but are typically more extensive in non-rhotic dialects. The particular inventory of vowels preceding historical /r/ in RP and similar southern British English varieties is shown in (2). As (2) shows, historical /r/ in the ancestor of RP, following the vowel changes outlined above, came to stand after only five vowels: these are long low [ɑ: ɔ:]; [ε:], when optionally smoothed from [εə]; [з:], which we might regard as long schwa; and schwa itself, which may be the offglide of a centring diphthong. Further smoothing of [ʊə], [ɔə], [aɪə] and [aʊə], which is variable but spreading in all but the most conservative current RP, does not increase this inventory.
(2) 

If Pre-/r/ Breaking were indeed under way in the late sixteenth century, /r/-Deletion could have been in progress from the same period, accounting for the non-rhotic nature of the first, eastern American colonies. In fact, /r/-Deletion is arguably a misnomer: the [r] seems to have undergone a gradual, dialectally variable weakening change, which resulted in eventual loss in non-onset positions. Indeed, Lass (1993) argues that /r/-loss was implemented over around 500 years, and was manifested in two phases. The earliest was sporadic and restricted to particular lexical items, such as bass `fish' OE , ME , and worsted from , which is recorded as in 1450; in , an etymological spelling has been reintroduced. We find occasional rhymes and spellings indicating this item-specific loss from the fifteenth to the seventeenth centuries: thus, the fifteenth century Cely Papers have `morning', `parcel' and the inverse spelling `master', while in 1642, Lady Sussex writes `persons' (Lass 1993). Seventeenth century loans also indicate some loss of [r]: Spanish salva is borrowed as , but Dutch genever `gin' as .

However, we are more concerned with the second phase of /r/ Deletion, a general weakening and loss of [r] in non-onset positions. Jespersen (1909: 318) argues that Old English and Middle English /r/ was probably a trill in all positions: this seems to have weakened to a tap, then an approximant, and is finally subject to conditioned loss in non-rhotic varieties. The tap remains intervocalically, and occasionally initially, for many Scots and Irish speakers; but approximants are fairly consistent finally and pre-consonantally. As we have seen, the weakened approximant reflex is similarly retained intervocalically and initially in non-rhotic accents, while the zero alternant has been innovated before consonants and pauses. In other words, across all varieties of English, weaker realizations of historical /r/ are found in coda positions (the translation of the C/pause disjunct from (1) in syllabic terms), while stronger reflexes are maintained in onsets. Since pre-consonantal and pre-pausal positions have been shown to be prime lenition sites cross linguistically, and specifically in other studies of consonantal weakening in English (Leslie 1989, Harris and Kaye 1990, Harris 1994, Tollfree 1995), this distribution is exactly what we would expect.

Again, orthoepical evidence suggests that weakening of /r/ can be traced back to the mid-seventeenth century, given Ben Jonson's (1640: 47) comment that `R ... is sounded firme in the beginning of the words, and more liquid in the middle, and ends.' The actual loss of /r/, like pre-/r/ vowel lengthening, seems to have begun pre-consonantally, and spread to word-final position in cases where no vowel-initial word or suffix follows to sanction resyllabification of /r/ into an onset. Harris (1994: ch.5) claims that Walker (1791) provides the first evidence of loss, with his observation that London /r/ `is sometimes entirely sunk' (1791: 50). However, a range of earlier and more specific observations can also be found. Viëtor (1904) claims that Theodor Arnold in 1718 is the first to note /r/ loss, giving examples of pre-consonantal contexts such as mart, parlour and scarce in which /r/ is `mute'. Similarly, Jespersen (1909: 360) cites König, who in 1748 diagnosed mute /r/ in horse, parlour, partridge and thirsty, among others. As we have seen, Mather Flint in 1740 testifies to vowel lengthening before /r/; but he also tells his French readers that `dans plusieurs mots, l'r devant une consonne est fort adouci, presque muet' (KoÈkeritz 1944: 41). In fact, he italicizes r before consonants, explaining that `vous verrez souvent aussi l'r en Italique, les Anglois l'adoucissant beaucoup plus que les FrancËois & ne le prononc Ëant que treÁs foiblement, sur tout lorsqu'il est suivi d'une autre consonne' (Kökeritz 1944: 3).

Mather Flint italicizes final r only very sporadically: this may indicate that /r/ had not yet been lost in this position, although we can probably conclude that it was already very weak. On the other hand, the presence of a final schwa offglide or extra vowel length alone, or the variable retention of [r] prevocalically, may have interfered, so that Mather Flint may be transcribing a final /r/ which is only variably pronounced. The same goes for Tucker (1773), who, as we have seen, describes Pre-/r/ Breaking, using the special symbol ʋ for schwa. Tucker (1773: 35) claims that ʋ is easy to pronounce and consequently ubiquitous, but that the same cannot be said for /r/: `Upon rendering the end of the tongue limber, so that it will shake like a rag with the bellows, it will rattle out ``r'', but this requiring a strong stream of breath to perform, makes it the most laborious letter of all, and consequently as much out of our good graces as I said ``ʋ'' was in them.'

Tucker goes on to report that `you shall find people drop the ``r'' in ``fuz, patial, savants, wost ... backwad,'' and many other words, and whenever retained we speak it so gently that you scarce hear a single reverberation of the tongue' (1773: 35-6). Again, the fact that the example words all have preconsonantal /r/, and Tucker's representation of there as `theʋr', might indicate that [r] had not been lost word-finally at this period. Alternatively, Tucker may have used `theʋr' to indicate that [r] was variably pronounced (whether contextually, when a vowel followed, or depending on speech rate or other sociolinguistic factors).

Sheridan's assertion that `R ... has always the same sound, and is never silent' (1781: 34), along with Walker's (1791: 50) similar claim that /r/ `is never silent', may initially seem to constitute counter-evidence; Walker's statement in particular seems inconsistent with his description, on the same page, of London /r/ as `sometimes entirely sunk'. However, we must recall the tendency towards prescriptivism in many grammarians of the time, which means that ongoing changes are often denied, and supposedly `ideal', archaic pronunciations encouraged. In the Preface to his Dictionary (1780: 4), Sheridan admits that the spelling and pronunciation he records `scarce deviates from that used ... in Queen Anne's reign' (1702±14); he therefore seems unreliable as a reporter of actual contemporary usage, and indeed, his assertion that /r/ never deletes may tell us quite the opposite. Similarly, Kökeritz, in his commentary on Mather Flint's work (Kökeritz 1944: 155), interprets Walker's statement as indicating that `in the contemporary London dialect r had been silent for a long time, although elocutionists probably endeavoured to pronounce it'. Indeed, Walker himself seems to acknowledge the extreme weakness or loss of final and preconsonantal /r/ by suggesting that speakers aiming at `polite' usage may produce initial [r] in Rome, river, rage as forcibly as they wish `without producing any harshness to the ear'; `but bar, bard, card, hard &c. must have it nearly as soft as in London' (1791: 50). For Walker to recommend such a pronunciation, it must have been socially acceptable and therefore well established in the (developing) standard.

It seems, then, that /r/ in coda positions was weakening by the seventeenth century; this weakening went substantially further in some dialects than in others, and [r] dropped, having first conditioned certain diphthongisations, lengthenings and mergers in preceding vowels, from the eighteenth century onwards. The loss of [r] seems to have begun preconsonantally, and proceeded to word-final position, and Lass (1993) assumes it was again a gradual process, being completed during the nineteenth century. In some present-day varieties, this historical development has in a sense been arrested part-way, since in certain parts of Yorkshire and Lincolnshire (Wells 1982), /r/ has been lost medially when preconsonantal, but not finally. /r/-Deletion also increased the phonemic vowel inventory in non-rhotic dialects by removing the conditioning context for the centring diphthongs. These began as contextually determined allophones of /i: e: u: o:/, but became contrastive, although still defective in distribution, after the loss of postvocalic /r/. We shall return to these historical developments, and argue that they need not be seen as individual changes, but can instead be modelled as an integrated complex. For the moment, however, let us turn to the present-day non rhotic varieties they have created.

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 Harris (1994)

Alternative analyses McCarthy (1991)

Alternative analyses Scobbie (1992), Donegan (1993)

Alternative analyses

Rule inversion and [r]-Insertion

English /r/: a brief outline

English /r/ Introduction