7.1 Introduction

It is generally assumed that we know which language is which and that consequently we can give a name to each one. But let us take this question of what English is by considering a few simple answers to see if they are sufficient. Firstly, we may say that English is the native language of those born in England. This is obviously too narrow because many speakers in Scotland, Wales, Ireland, Canada, the United States, Australia, New Zealand and South Africa have English as their native language. So, we cannot equate a language with a country. English also has the status of a second language, a lingua franca, in many countries. This is not primarily to enable the speakers to communicate with native English speakers, but with compatriots who have a native language other than their own. This situation obtains in parts of Africa; for instance, in Kenya English is the language of those in higher education, Kiswahili is the language of Kenyan nationality and the many ethnic languages are local, tribal markers, for example, Luo, Kikuyu, Kalenjin. In addition, English serves as an (if not the) international means of communication in many areas of commerce and travel, for instance. It is the lingua franca of most of the world. We can see from this that English has long outgrown its parochial functions of everyday life in the British Isles.

In an attempt to deal with some of the issues I have just mentioned, writers have tried to define a language with a combination of social and political factors, and in some cases have added linguistic considerations such as mutual comprehensibility in an attempt to deal with the problem of variation. (See Romaine, 1982 and Dorian, 1982 on the problems of defining a speech community; see also Fasold, (p.121) 1984, on nations and languages. Recent studies of this issue, such as Barbour & Carmichael, 2000, place the arguments clearly in the political arena.) If English is associated with a particular country, then it is politics and social considerations that determine the boundaries, not linguistic structures, and mutual comprehensibility does not help in defining a language. Since in the British Isles the dialect continuum does not cover two separate sovereign states, I will look at such an example furnished by what are usually considered varieties of German (see also Boase-Beier & Lodge, 2003). In this case we find that many of the varieties found in Germany are mutually unintelligible, as much as English and Dutch are. The fact that they are closely related languages does not mean that their speakers can each understand one another. Let us take a speaker from the German side of the Dutch-German border and one from Bavaria. If they are speakers of the local dialects, they will understand one another only with the greatest difficulty. In some respects the Plattdeutsch speaker from the north has more in common linguistically speaking with an English speaker, and certainly a speaker of Dutch, than with a Bavarian. For example, the former will have initial [p] and [t] like in English, where the latter has [pf] and [ts], as in pund versus Pfund and tien versus zehn, respectively. Despite the fact that they live in the same political entity, Germany, pay the same central taxes, owe allegiance to the same flag, would serve in the Bundeswehr, if they wished to do military service, they do not seem to speak the same language. Furthermore, northern speakers will be able to understand their Dutch neighbours far better than they can understand their Bavarian compatriots. There is an important sense in which the north German and the Dutch speakers speak the same language. From a linguistic point of view their national allegiance is irrelevant. Of course, they are each taught a different standard language in school, but this, too, is a political and social matter not a linguistic one. The picture we end up with, if we look at geographical variation in language, is of a dialect continuum, a slowly changing set of partially overlapping linguistic systems, which at the extremities may be very different indeed.

From a linguistic point of view it would be convenient if linguists could define a language by its structural characteristics. This is implicit in the notion of a panlectal grammar (Bailey, 1973): one language has one grammar, so New York English, Birmingham English and East African English are varieties of the same language. This is high-level linguistic sameness. Any variation in form is considered to (p.122) be relatively superficial and can be accounted for by differences in the rule system relating the stored lexical forms of the native speakers to their realizations in phonetic substance. For instance, English rhotic and non-rhotic accents can be seen as minor variations in the substance. A panlectal approach will propose a single underlying, phonological form for each such word and the phonetic forms will be derived from it by different rules. In this case the rhotic forms are taken as underlying and a rule deleting /r/ after tautosyllabic vowels will account for the non-rhotic versions. This kind of approach is often implicit in discussions of varieties, as in the example of Carr's exercise on English /r/ in his coursebook on phonology (1993a: 41–2, answer: 307). Basically, this relies on two (ordered) rules, as in (7.1) and (7.2).

(7.1) /r/-deletion

/r/ → Ø / V ___ (C)$

(7.2) /r/-realization

/r/ → [ɹ]

In (7.3) we can see how the operation of the extra rule of /r/-deletion produces the difference between rhotic and non-rhotic accents. So rhotic speakers and non-rhotic speakers order their rules differently. (It is not always clear whether both types of speaker are assumed to have both rules; rhotic speakers either order (7.2) first, so that (7.1) applies vacuously, or do not have (7.1) in their grammar.)

(7.3) /kar/ /farm/ /kar/ /farm/

/r/-deletion -- -- [ka] [fam] (non-rhotic)

/r/-realization [kaɹ] [faɹm] -- -- (rhotic)

Lowland Scots and RP are thus related to one another by a rule of coda /r/-deletion. This assumes that speakers of RP know where Lowland Scots speakers have lexical coda /r/. There is, however, plenty of evidence to show that this is not the case; in fact, non-rhotic English speakers generally cannot predict rhotic forms accurately. Trudgill 1983: 8–30) shows how relatively poor native speakers of English are at recognizing forms of their own language that they do not themselves use, and, in a separate paper, looks at a number of linguistic features of British pop singers in the 1960s and 1970s, among them rhoticity (1983: 141–60). Hypercorrect forms such as [əɹ bӕtʃələɹ bɔΙ] a bachelor boy and [sɔːɹ ðεm] saw them are frequent (see 1983: 149) in the singers' attempts to sound American. If (p.123) the grammar represents a native speaker's knowledge, which enables him or her to predict all forms of his or her system and produce them when needed, then how are we to explain these anomalous forms? They occur too often to be explained away as slips of the tongue. Rather they are mistaken analyses of a target variety on the basis of the speaker's own system. The rule that non-rhotic speakers invent is something like the following: ‘Insert [ɹ] after the vowels [a ε ɔ ə].’

Of course, such a dichotomous presentation of accents as rhotic or non-rhotic overlooks the fact that there is often a cline from some obstruent realization (for example, a trill or tap) through a range of possible articulations to no apparent phonetic presence. For instance, one of the informants from Edinburgh for Lodge 1984 : 82–5) has the following forms: [əwə⌠] aware, [Mεər₀] whether, [püəɹ] poor, [koət] court, [bΛdz₀] birds, [pɔʃΛ] posher and [fΛst] first, with varying ‘degrees of rhoticity’. So, phonetic implementation of the lexical forms is going to be less straightforward than presented in (7.1) and (7.2). Similarly, in Lancashire, where some urban areas have rhoticity, there is variation between speakers as to whether they have a coda retroflex approximant, or a general retroflex posture (Honikman's, 1964 articulatory setting), for example, in [bɜɻɳɭἓ] or [bɜɳɭἓ] Burnley. (Note that the latter has no ‘compensatory lengthening’ of the vowel phase.)

Chomsky (1980: 117–20) concludes that the notion of a language (that is, ‘English’, ‘French’, ‘German’) is of little use to linguists, who should concentrate on grammars. If we follow this line of reasoning, it means that we end up with a set of grammars some of which overlap fairly closely, others of which do not, but all need to be described separately. It could be suggested that comprehension will take place in those cases where overlap is greatest. A large part of this overlap will be in the lexical forms of the varieties in question, but, to my knowledge, no one has tried to correlate the amount of overlap with the degree of comprehension, though Trudgill 1983: 29–30) makes some tentative suggestions in this regard. However, comprehensibility is not necessarily commensurate with linguistic overlap. For example, most broad speakers from the north of England have no difficulty in understanding RP-speaking newsreaders on the radio and television, but the systems do not overlap very much at the phonological level, especially as regards phonetic implementation. And the RP speaker may have more difficulty understanding a northerner than the other way round. This has to do with currency and exposure to some extent: RP is heard fairly frequently in the media, broad northern varieties (p.124) less so, though since roughly the mid-1990s fewer RP speakers are heard on British television. However, the extent to which accents in the media affect individual speakers is far from clear. Whereas a Bolton (Lancashire) accent may be heard in popular programmes from time to time (for example, the comedian Peter Kay or the late demolition and steam traction expert Fred Dibnah), we do not find newsreaders who are obviously from Bolton, or from Norfolk, or from Birmingham, even though there are plenty of ‘regional’ varieties to be heard in news broadcasts, in particular Scottish, Irish and Welsh. In other words, which accents are heard in the media and the programmes in which they are heard are controlled and socially determined, and it is largely a matter of fashion. The formerly ubiquitous RP accent of television and radio announcers has given way not to the locally determined, indigenous varieties, but to a carefully selected set of regional accents. In the 1960s, in the wake of The Beatles, the Liverpudlian accent was the hallmark of the age, but it has disappeared from the media scene, and its populist position has been filled by the supposedly anti-establishment, so-called Estuary English. But the key to linguistic variation and change is more likely to be everyday face-to-face interaction, not experience via the media, which might encourage copying as opposed to interaction. (For a discussion of this issue, see Trudgill, 1986: 40–1.)

The implication of the panlectal approach to the phonology of a language is that all speakers of English have knowledge as native speakers as to where coda /r/ can occur in all varieties, which would mean that they could be rhotic or non-rhotic at will. Trudgill 1983, as presented above) has demonstrated clearly that native speakers cannot recognize, let alone predict and produce correctly, all varieties of English. One further point needs noting in respect of rhoticity, however. There are clear signs that in the United States and in Britain many speakers produce ‘hyperrhotic’ forms such as [Өɔət] thought (see, for instance, Trudgill, 1986: 71–8, for a discussion, and a recent paper by Krämer, 2008). Of course, the explanation for this spread could well be that since rhotic forms are prestigious in America at least, hypercorrect forms take hold more easily than in England, where the prestige accent is non-rhotic. In other cases hyperrhoticity can be seen as an indication of dialect loss, that is, speakers no longer know the details of their traditional local accent or dialect. Whether the hyperrhotic accents become established as models for acquisition remains to be seen. But the evidence against panlectal grammars does not rest solely on rhoticity. That it is not just a matter of rhoticity can (p.125) be seen in other cases of hypercorrection which are a matter of phonological contrast as opposed to the phonotactic distinction involved in rhoticity. In Lodge 1984: 15) I refer to those speakers in the north of England who try to copy RP speech with its /℧/-/Λ/ distinction as in put, butcher and putt, come, respectively. Most of them produce a vocoid articulation of the following kind: [ə] ~ , but they use it in both lexical sets, that is, in put, butcher, putt and come. What they are actually doing is using their own underlying phonological system, which has no distinction, but changing the phonetic realization to sound more like the target realization of RP /Λ/.

Even polylectal grammars, which cover a restricted number of varieties, cannot be defined other than by linguistic criteria. That is to say, one single grammar may account for a number of different realizations with just minor rule variation, often at the level of realization, but they do not coincide with social groupings. Trudgill (1974) attempts to describe all the varieties of Norwich with one system (diasystem), but this ignores the fact that, say, most young speakers from the city do not know the distinction between /ε:/ as in gate and /æI/ as in gait, used only by the very oldest inhabitants andrural speakers in Norfolk (see Trudgill, 1974, and Lodge, 2001). That they may recognize individual forms as ‘the way grandad speaks’, for instance, is no more knowledge of the system than knowing that chat is French for ‘cat’ and chien French for ‘dog’ indicates an ability to speak French.

The picture I presented above of a Dutch-German dialect continuum is handled by positing a set of overlapping linguistic systems, which at each end may be very different, as different as standard Dutch and standard German. The same applies to linguistic varieties of English on mainland Britain. Dialects merge into one another, changes occur subtly over a few miles. It is only convenience that leads us to label them ‘Norfolk dialect’, ‘a Yorkshire accent’, ‘Scouse’. Such labels are no more accurate from a linguistic point of view than the language labels I discussed above. ‘Norfolk dialect’ belongs to the group with an /℧/-/Λ/ distinction, but then what of those speakers in the far west of the county, in the Fens bordering Cambridgeshire and Lincolnshire, who vary between [℧] and [Λ] in many words, or those who have only [℧]? Aren't they ‘real’ Norfolk speakers? (Recall our discussion of the defining characteristics of the standard English king of hearts in Chapter 1.) The fact of the matter is that there are no clear-cut linguistic boundaries that stop at the county line; the systems merge and interact in interesting ways, and variation, even (p.126) in one and the same speaker, is the order of the day. Most speakers (except for those who lead very isolated lives) have a number of overlapping systems at their disposal, which they use on different social occasions. Needless to say, British English and American English do not form a dialect continuum in the way varieties on mainland Britain do. Time and space have brought about changes in each which were unconnected. Of course, with the increase in worldwide communication these geographical obstacles become less and less significant. A thousand years ago the Fens were, indeed, an area that was difficult to travel through and so constituted a real barrier. Today, however, they are not significant in these terms. Distance, too, is less important today with our road, rail and air transport. Communications between, say, Manchester and Liverpool are easy and that has an effect on linguistic interaction. In some cases connections between major cities may be more effective than between a city and its surrounding area, for example, between Manchester and nearby Pennine villages. But this does not mean that panlectal or polylectal grammars are justifiable. I shall return below to a consideration of polylectal grammars and the way in which the abstraction of phonological forms may be applicable to this particular issue.

7.2 Levels of Variation

Let us consider now the kinds of phonological variation that take place. The following types of variation relate to different aspects of phonology, as discussed by Petyt 1980, all exemplified from English:

1. 1. contrastive differences: could and cud have contrasting vowels or are homophonous. This involves a different number of vowel contrasts in the two types of accent;

2. 2. ‘allophonic’ differences: [Ɂ] may be a realizational variant of /t/ or of /p t k/. Many accents of English have the glottal stop as a realization of /t/ only, but others, such as London varieties and some Norfolk ones, have it as a contextually determined realization of any of the so-called voiceless stops, for example, Cockney [k^haɁ ətsəΙ] cup of tea, [leɁɐ] letter, [a l aiɁ jɐ] I like you; Norfolk [p^hΙiɁϮ] people, [sΙi jəɁmaրə] see you tomorrow, [lΛΙɁ jʉ:] like you;

3. 3. realizational differences: /æ/ (‘short a’) may be realized as [æ] (RP), [a] (Manchester), or [a] (Belfast);

4. 4. phonotactic differences: coda /r/ is permitted (rhotic accents) or is not permitted (non-rhotic);

5. (p.127) 5.lexical incidence differences: path, class and dance have the same vowel as start or as man depending on the accent.

Hypercorrection is important evidence for lack of systemic overlap. It is important to note that there are two different phenomena referred to as hypercorrection: Labov 1966 uses it to refer to the ‘cross-over effect’ of a group of speakers using a variable form more than the group of speakers they aspire to copy, as in the case of coda /r/ in New York; alternatively, it can be used to refer to the use of forms that belong to no one's native phonological system, as in the use of pronunciations such as [AfɹIkə] Africa, based on an incorrect extension of the backing and lengthening of /æ/ in many southern accents of British English and RP from forms such as after, craft, laugh. Both types are the over-application of a phonological feature, but in the first type it is within the bounds of the grammatical system. The second type is more significant from a phonological point of view in that it reflects a lack of knowledge of the target system in the same way as L2-learners extend their L1 system to cover areas of the grammar of L2 of which they are unsure. It could be argued that learning someone else's accent is very similar to learning someone else's language, the difference being that most of the lexis and syntax will be the same in the first case, but not in the second. This is part of the degrees of overlap between systems.

In the next two sections I want to consider interpretations of sameness or difference at phonological (section 7.3) and lexical levels (section 7.4). The lexical examples involve (mis)interpretation of the realizations.

7.3 Mergers

Since we are dealing with language change when we consider the sociolinguistic effects of variation, one of the issues related to systemic change is that of mergers of phonological distinctions. We ask, have these two formerly distinct phonological units merged into one? In other words, are they the same unit now? (Note that the northern English accents with no /℧/-/Λ/ distinction have not undergone a merger; from a historical perspective Middle English /℧/ split into two in the more southerly dialects.)

The first example I should like to discuss is the phenomenon of /f/-/θ/ variation in British English accents. In some accents /f/ and /θ/ may genuinely have fallen together as /f/, but in others there is (p.128) evidence that this is not the case. So, (7.4) would represent realizations of a merged /f/, but compare these realizations with those in (7.5), potential evidence of non-merger.

1. (7.4) [pæɁ fΙŋks] Pat thinks

   [lIz fIŋks] Liz thinks

2. (7.5) [pæɁ fΙŋks] Pat thinks

   [lΙz sΙŋks] Liz thinks

In the last example the sibilant assimilation found in speakers who use [θ] occurs in the same environment in the speakers under consideration. On the other hand, such a speaker would not assimilate any /f/ in these circumstances any more than speakers who use the forms in (7.4) or those with the /f/-/θ/ distinction would do, as, for example, in (7.6).

(7.6) [pæɁ faΙnz] Pat finds

[lΙz faΙnz] Liz finds

*[lΙz saΙnz] Liz finds

Clearly, though, it would be odd for any analyst to say that for such speakers /θ/ has two allophones [f] and [s] (but no [θ]), if fully specified phonemes were assumed. On the other hand, an underspecified phonological representation, even a segmental one, could manage to cover the three different types of speaker: (1) those with alternating [f ~ s], which I will symbolize /f/, and /f/, (2) those with just /f/ and (3) those with /f/ and /θ/. My suggestion in Lodge 1992: 29) is that /θ/ is represented as [voiceless] and [dental] underlyingly, but this cannot work for speakers who assimilate onset /θ/ to a preceding coda /s/, since place of articulation changes. Whereas speakers of type (3) could have the place feature [coronal] specified, interpreted differently by phonetic implementation according to the environment (see section 6.4.4 above on the phonetic implementation of [coronal]), this cannot be the case for type (1) speakers, since they have labiodental realizations. It would be possible to alter the lexically specified features for /F/ for such speakers to just [oral]. There would have to be a predictive statement to the effect that lexical [oral] implied [voiceless] and [fricative], as in (7.7).

(7.7) [oral] → [voiceless], [fricative]

The place feature would either be [alveolar] attached at ambisyllabic coda/onset level in the assimilating cases (see section 6.2 above), and (p.129) [oral] would predict [labiodental] in the other cases. No phonological distinction needs to be made between /f/ and /θ/ for speakers of type (2). So the three grammars differ slightly from one another: type (2) has only /f/ in the phonological representations; type (3) has /θ/ specified as [coronal] and type (1) has a distinctive unit /F/, represented as just [oral].

I now want to consider some vowel distinctions from English, which are variable in East Anglia. Trudgill & Foxcroft (1978) discuss the beer/bear distinction and its merger in various parts of East Anglia. The realizations for a large number of speakers in both lexical sets are in the region of [ε:], and is considered to be a merger. This is reflected even in local dialect poetry and other written representations of dialect forms. Trudgill & Foxcroft (1978: 75–7) wanted to see if there were generational differences in the maintenance of this merger and whether London speech, which has the distinction and was spreading from the south-west, was having any effect. For whatever reasons, it became apparent that for some young speakers the two lexical sets have passed through each other's phonetic space, as it were, so that the vowel of the beer-set was realized in the vicinity of [ε:] and the vowel of the bare-set was [e:] with a closer articulation. Whether it is appropriate to talk about merger and de-merger in cases like this is not at all clear, but we can see that the notion of phonological sameness has to be treated very carefully. In the case of East Anglia we have some speakers with the distinction and some without. Those who have no distinction often produce hypercorrect forms, especially when interacting with non-East Anglian speakers, such as [tIə] instead of [tε:] in tear (up the paper). This is clearly a case of accents in contact and interacting; a discussion of this complex phenomenon goes beyond the scope of the present book, but see Trudgill 1986 for the discussion of a number of examples.

The other East Anglian vowel distinction I want to consider is by now restricted to rural Norfolk speakers, but in phonetic terms it can be considered in relation to the vocoid articulations discussed in the previous paragraph. The Middle English distinction between made and maid is still maintained by some, mostly older speakers with the realizations [mεːd] and [mæId], respectively. If we use phonetic similarity as a criterion of phonological sameness, then why not consider made as belonging to the same lexical set as beer/bare? Since these vowels come from very different historical origins, ME /a:/ and ME /Vr/, respectively, analysts are not tempted to suggest such a solution (for example, it is not part of Trudgill's, 1974 proposals). The (p.130) different historical origins have resulted in different distributions and behaviour. Words of the make-set always end in a consonant and so to that extent belong to the so-called short vowels-that is, those that can only occur in closed syllables, irrespective of their actual duration (see section 5.7 above, and the examples in (5.10)). On the other hand, words with historical coda /r/ have linking [ɹ] before another vowel. Of course, it could be argued that since vowels in the make-set and the /Vr/-vowels are in complementary distribution, there is still no evidence that they have not merged. However, there is sociolinguistic evidence that is important and relevant to our discussion. Speakers who have the made/maid distinction are aware that it is a local, ‘in-group’ linguistic feature. Outsiders, whether from Norfolk or further afield, do not have this distinction, so the former group change their linguistic behaviour and use [æI] in the make-set, merging made and maid, as most other English speakers have done. This sociolinguistic sensitivity does not occur with words of the beer/bare-set, and no hypercorrection occurs either, which indicates a clear separation of the two lexical sets, despite the overlap of the [ε:] realizations.

One issue that is thrown up by the merger of made and maid in Norfolk is how mergers come about. We may be aware of realizations of phonological units spread around a particular focus of phonetic space. This is easy to imagine with vocoid articulations, since there is no contact between the tongue surface and the upper part of the mouth, but it applies equally to places and manners of contact. Such realizations can move gradually from one location to another. (For a discussion in some detail of the effects of this on phonological categories, see Silverman, 2006, especially Chapters 5 and 6.) Gradual movement through phonetic space is what Trudgill & Foxcroft (1978) study in relation to the realizations of particular vowel distinctions. On the other hand, it is equally possible for systemic change to come about through contact between the speakers of two different systems, so that some speakers take on the forms of a system not originally their own. This could account for the young East Anglian speakers taking on the London-based distinction of the beer and bare sets, or the replacement of traditional Norfolk [ε:] in made by another regional, but non-distinctive variant realization [æ Ι]. Replacement is a more abrupt development than merger. The latter is system-internal, the former is the result of the interaction of different varieties.

Some mergers, apparently purely phonetic in nature, are accepted as such in historical terms, but there are plenty of other instances, such as those discussed above, that need clear functional criteria to (p.131) determine whether a merger has taken place. Examples of mergers from the history of English include the following:

/e:/ versus /ε:/ in meet/meat

/a:/ versus /ai/ in name/day

/x/-loss leading to mergers in might/mite

though, as we have already seen in one case, there is dialect variation in respect of all of these; for example, the retention of the meet/meat distinction in parts of the Lake District, and the loss of /x/ later in the northern dialects leading to the maintenance of distinctions such as port with [Ɔʊ] and thought with [Ɔ℧] (see, for example, Lodge, 1973).

What I want to consider now are a few examples of lexical sameness from different dialects of English. The lexical interpretation of some variant forms of English relies on the imposition of a panlectal set of lexical items based on the written standard.

7.4 Lexical mergers and standardization

Sameness of linguistic form is sometimes assumed in cases where historical processes have obscured the origins of the assumed variants. Four examples from English will suffice to show slightly different ways in which false identifications can occur. Firstly, the modern English present participial suffixes -ing and -in' are typically related to one another via ‘g-dropping’ (for example, Chomsky & Halle, 1968: 85, and Labov, 1972: 240). This is clearly a shorthand, based on orthographic convention: ing versus in', for a complicated historical development. Lass (1992: 144–6) and Fischer (1992: 250–6) present details of the way in which two separate suffixes became confused and reinterpreted. The phonological and phonetic aspects of this ‘merger’ are of particular interest. The Old English participial suffix was -end(e), which lost its final [d] during the early Middle English period. A variable vocoid quality is quite likely, given the modern accent variation for the suffix. The deverbal noun suffix was -ung; this form was used in a number of constructions, including the analytical progressive of the verb, formed with to be + on + gerund. The vowel of the suffix changed its quality to [Ι] and the preposition became a schwa; cf. dialectal and archaic standard he's a-coming. So, originally, the -ing suffix was nominal. Given their use in two different forms of the progressive and the phonological changes they were subject to, in particular vowel reduction, the scene was set for the confusion and (p.132) conflation of the two suffixes [In] and [Iŋg]. The distinction between the verbal and nominal uses, respectively, was gradually lost (see, however, claims that it has not been lost in all English dialects in Houston, 1985). By the late Middle English period [In], orthographically in/yn, can be found for the gerund alongside the -ing form, for example, in the Paston letters, as in (7.8), from Warrington (1956: 255), from a letter by Agnes Paston to her husband William written before 1440.

(7.8) I sende yow gode tydyngges of þe comyng and brynggyn hoom of þe Gentylwomman

So, rather than actually merging, the two suffixes were used as alternatives for a while. As the standardized written form of the language developed, the -ing form was chosen as the ‘proper’ form. In speech, on the other hand, -in' continued to be used. In modern dialects (with the exception of those discussed by Houston, 1985) speakers either use [In] exclusively or [In] and [Iŋ] variably. At some stage, probably during the nineteenth century, the spelling -in' was adopted as the standard way of representing the non-standard form.

What the label g-dropping fails to acknowledge is that this alternation is functional not phonological in origin. Indeed, it is difficult to see how [Iŋg] could possibly become [In] in unstressed syllables by the removal of the stop articulation. The loss of the final stop leaves a homorganic nasal, hence modern English [Iŋ]. This is the problem with the SPE analysis of [ŋ] as /ng/, so that removal of the /g/ leaves /n/= [n], as a description of the historical process. (Of course, if one assumes that both /n/ and /ŋ/ are phonemes, then ‘g-dropping’ can only refer to the spelling.) The point is that, whereas an abstract analysis of [ŋ] as /ng/ in all cases allows us to claim that the [-In] forms are accounted for by ordering /g/-deletion before assimilation, this makes no sense as a description of the historical process, since historical change operates in its initial stages on phonetic forms, not on abstract phonological ones. In other words, the pronunciation of /-ng/ was always [-ŋg], so loss of the final stop gives us the standard and southern forms in [-ŋ]. This appears to be yet another instance of the confusion of levels.

We might be tempted to say that there is phonological-as opposed to morphological-evidence that the -in' is related to stress, since we find it in other words that are not participles or gerunds in unstressed syllables: somethin', anythin', nothin', mornin', evenin'. This would then be a historical development of Middle English [-Iŋg] in all such cases. Of course, it should be pointed out that during the Middle (p.133) English period final voiced stops after nasals were unstable, and both /b/ and /g/ disappeared in this position (the latter not in parts of the north of England) and /d/ did in the case of -end. In some accents today /d/ still exhibits this instability, especially before another consonant (see Lodge, 1984). However, in each case the resultant nasal is homorganic with the lost stop, for example, [læm] lamb and [kIŋ] king. But, as I have just pointed out, the resultant [-n]-form is difficultto account for as a loss of [g] in the non-participial unstressed -(th)ing forms, unless they are analogical forms based on the participial form. But, finally, there is no reason why participial -in' and -thin' endings should be treated as the same; a polysystemic approach treats them differently, as they are from different parts of the grammar (whether analogy is involved or not).

The next example is very similar: them/'em. The former is a Scandinavian borrowing along with they, and the latter is the Middle English reflex of the Old English hem. Indeed, in Chaucer's writings they is the subject form and hem the object form. Once again, the standardization process makes them the correct form and 'em is seen as the spoken version of it, though the /ð/-deletion rule is word-specific; for instance, there is no equivalent *[Də℧nt eI] don't they? etc. On the basis of examples like these, -ing and them, it would be possible to argue that there are two grammars in this respect, one based on speaking, one on writing, and that when speaking some people switch between the two. There will also be speakers who do not use the standard written forms in speech. So, rather than trying to produce a single grammar of English to account for this sort of variation, I am proposing that speakers grammar-switch under various circumstances. Code-switching is well attested in bilinguals, so why not in monolinguals at this grammatical level? In fact, we might want to describe all such switching, whether bilingual, multilingual or monolingual, as grammar-switching. In cases such as these from English there is no need to have recourse to the notion of merger.

The two final examples are simply instances of lexical misinterpretation with no suggestion of a merger. The first is once again from Norfolk dialect. In this, unlike in most dialects of English, there is a form, [ət/əɁ], used as the unstressed object form in cases such as (7.9)-(7.12), where I have used the standard orthographic representation it.

1. (7.9) [gΙv əɁ] give it

2. (7.10) [hΙt əɁ] hit it

3. (p.134) (7.11) [Ιz əɁ] is it?

4. (7.12) [εnt əɁ] ain't it?

It also triggers smoothing (cf. section 5.7 above), as in (7.13), where the monophthongal realization indicates lexical /ʉ/ followed by /ə/.

(7.13) [dзːɁ] do it.

All such cases are interpreted as it (cf. Trudgill, 1974).

On the other hand, the subject form is that, as in That's raining, that is, that's right, and no form of it is used in these cases. The realizations of that vary somewhat, depending on stress, for example, [ðæɁ/ðaɁ/aɁ/ɐɁ], and that's may be [as]/[as] (cf. Trudgill, 1974), as in (7.14).

(7.14) [ɐ(Ɂ)s ƐvəsənʌΙs] That's ever so nice.

It is only the standard written form that makes us identify [ət/əɁ] as it, as in the preceding orthographic representations; there is no reason not to interpret it as the object form of that, which can lose its initial consonant even in subject position. In this case we are not dealing with grammar-switching as much as a misinterpretation of the object form of the pronoun via the standard language (and most other English dialects).

The final example is a dialectal form that is difficult to interpret: the definite article in many northern English varieties. Typically it is realized as [Ɂ] (cf. Wells, 1982; Lodge, 1984), as in (7.15)-(7.17).

1. (7.15) [ΙðɁgaːdn]] in the garden

2. (7.16) [gΙvəsɁbɹ℧um] Give us the broom.

3. (7.17) [avjəgaɁɁtaƐm] Have you got the time?

Note that (7.17) is different from (7.18) in that the glottal closure is longer in the former.

(7.18) [avjəgaɁtaƐm] Have you got time?

It also occurs utterance-initially as a closed glottis onset to a voiced stop, as in (7.19).

(7.19) [Ɂb℧sΙzk℧mΙn] The bus is coming.

In Middle English there were two forms of the definite article, one with initial [t], one with [θ], the former written with t, the other with th (or a thorn, þ, as in (7.8) above). The former is interpreted as an assimilated form. (The latter was later voiced in unstressed position (p.135) like other [θ]-initial words.) In standard English the reflexes of the te form have disappeared through phonological simplifications, for example, at te last 〉 atte last 〉 at last. In the north, however, the glottal stop form is presumably the descendant of the [t]-initial form, though the development is not necessarily straightforward, as pointed out in Jones (2002). An interesting development found to the west of the Pennines, mostly in Lancashire, is the complementary distribution of the [Ɂ] and [θ] variants such that the pattern is: [Ɂ] + consonant, as in the examples above, and [θ] + vowel, as in (7.20) and (7.21).

1. (7.20) [θƐ:ΙəϮ] the aerial

2. (7.21) [θAspΙtϮ] the hospital

And there may even be both realizations together in some contexts:

1. (7.22) [wΙɁθƐːɹIəϮ] with the aerial

2. (7.23) [InɁθaspΙtϮ] in the hospital

For some older speakers the distribution can even be [Ɂ] + consonant, [t] + vowel.

In addition to this dialectal form, many speakers use standard English [ðI]/[ðə] forms. For example, in Lodge (1984: 36) speaker N makes the following utterances in close succession:

(7.24) [dæ℧nɁsƐllə] down the cellar

[teːksθƐːɹəϮ] takes the aerial

[wƐːjəgƐɁeƐəϮ] where you get the aerial

(The lengthened lateral in cellar is not a normal part of his phonology.) Here again we can talk of grammar-switching between local and standard grammars. Rupp & Page-Verhoeff (2005) claim that there is a semantic/pragmatic difference between the use of the and what is often referred to as Definite Article Reduction (DAR)-that is, the forms being discussed in this section. The categories they propose are hard to apply in the examples recorded for Lodge 1984, but if all speakers who use both the and DAR do make such distinctions, then both forms are part of the same grammar. I have little data on this point to decide one way or the other. Further investigations are needed. (For a much more detailed phonetic investigation of the phenomenon, see Jones, 2007.)

7.5 Abstractness and polylectal grammars

An aspect of abstractness in phonology that was not discussed in Chapter 5, but can be considered here, is its relevance to panlectal (p.136) grammars. This is a matter of high-level sameness or difference. Even though notions such as ‘the English language’, ‘the French language’ underlie a lot of published work in linguistics, we have set them aside as unhelpful linguistic concepts. On the other hand, we can investigate the possibilities presented by abstraction of having lexical entry forms that cover several linguistic varieties. This is certainly implicit in Fudge (1969) in which he uses totally abstract phonological elements to describe English syllable structure. For instance, he says (1969: 269–70):

The inclusion of postvocalic r (places 4 and 5 [in the syllable; KRL]) must not be taken as implying that the scheme does not apply to ‘r-less’ dialects: D3 is an abstract element which in some dialects (notably RP) may often have no realization of its own, but which will, so to speak, contribute to the realization of the preceding vowel.

This suggests that abstraction, with little or no phonetic content in phonological structures, makes panlectal, or, at least, polylectal grammars easier to devise, though such an approach ignores the problem of native-speaker knowledge discussed above. The question will be whether differences of realization are to be accounted for at the lexical level or elsewhere, and what that ‘elsewhere’ might be. It certainly seems attractive to be able to account for all vocalic variation in English in such abstract terms, giving a lot more work for the phonetic implementation component. In Fudge's proposal even vowel + coda /r/ realizations will be handled in this way. So, using IPA symbols rather than Fudge's original formulae, /Ɛr/, as in serve, for instance, would be implemented as [Ɛɹ] in some rhotic accents, [ɜːɻ] in others and [ɜ:] in non-rhotic ones. But the question still remains, how far can one go in connecting different accents of a language? Eventually there will be phonological representations for vowels that very few people recognize, for example, /Ɛ:/ from Middle English /a:/ in make that we discussed in section 7.3 above, or reflexes of Middle English /x/ making vowel distinctions between port and thought, wait and weight in many northern varieties, also referred to above. Such differences will have consequences for storage forms of the same lexical item in different accents. So extreme abstractness of the kind proposed by Fudge does not solve the ‘one language’ issue any more than the derivational approach of SPE with its abstract phonological forms (including /x/).

As we have already pointed out, if we want to capture native speaker knowledge with our grammars, then a grammar can only (p.137) account for a limited number of varieties. This does not mean that polylectal grammars work better in the sense in which they are normally applied. That is, a polylectal account takes a geographically or socially prescribed area, the home of a ‘speech community’, as its focus-for example, Norwich, as described by Trudgill 1974. But since this also cuts across native speaker knowledge in that very often different generations have (slightly) different systems (compare the informants in Stockport, for example, in Lodge, 1984), we need to describe the linguistic overlap of the relevant systems as well as the differences.

Of the types of phonological variation given above in section 7.2 only some can be handled by means of underspecification and systemic overlap. Type (1), contrastive differences, (for example, could versus cud) and type (4), phonotactic differences, (for example, rhotic/non-rhotic) furnish evidence of different systems, whereas type (3), realizational differences, can be treated under sameness of underlying unit. For example, / æ /, realized as [æ] (RP), [a] (Manchester) or [A] (Belfast) can be treated as the same vowel by specifying it as [low] in lexical entries; phonetic implementation will interpret it as [front], [central] or [back] accordingly. Type (2), ‘allophonic’ differences, are not handled in the same way in a polysystemic approach, as we saw in Chapter 6, so coda [Ɂ] may or may not be associated with onset /t/ or /p t k/ depending on the rest of the system. The example I gave in Chapter 2 (see example (2.2)) would suggest that there are two accent types: those that have alternating glottal stop and oral stop(s), and those that do not. Systemic overlap is not appropriate in this case. Similarly, type (5), lexical incidence differences, since they are distributional, may involve different systems. Whereas economics with initial /Ii/ or /e/, or scone with /o℧/ or /D/, are minimally different and individually determined, speakers with /a/ in path, grass, France etc. cannot predict properly the distribution in those accents which have /a/ in such words. This is not a matter of meaningful contrast, because accents with /a/ in such words also have an /a/-type vowel (usually realized as [a:]), as in park, calm.

As an example of how system overlap can be treated in abstract terms, irrespective of geographical origin, I will take English three-consonant onsets. The basic pattern is as in (7.25).

(7.25) /s/ + voiceless stop + approximant

The category approximant covers /j w r l/; note that the approximant in [j℧u] is classified here as a consonant. The syllable structure of such

(p.138) onsets is as in (7.26). Since C₂C₃ mimic the phonotactic constraints of two-consonant onsets, the /s/ is interpreted as dependent on their structure.

I will concentrate on those clusters that involve one of the first three consonants, not /1/. If we take their phonological representations in single consonant onsets to be as in (7.27) (from Lodge, 1992: 29), then we can see whether they apply in three-consonant ones as well.

(7.27) j w r

[round]

[front] [central]

In the case of [-w]-onsets the attachment of [round] must be at the highest onset level (or higher, if the nuclear vocoid articulation is also round), as in (7.28).

(The other two possibilities /spw/ and /stw/ do not occur; the former is ruled out by a constraint on [bilabial] over two consonant places and the latter simply is not found in any words, although /tw-/ is legitimate as a two-consonant onset.)

At onset level, [front] and [central] are also attached, as in (7.29), in which I give all the possible distinctive onsets with /j/ and /r/.

These structures are sufficiently distinct from one another and any other onset types to make them capable of interpretation by the phonetic implementation component. In earlier discussions in Chapter 6 predictable features were supplied for all layers for the phonetic implementation to interpret, but in the case of English three-consonant onsets there is no need to do this because of the restricted number of features that can apply. So any structure of the kind in (7.29) will be interpreted as starting with a voiceless oral fricative with the appropriate lip position and resonance. This will change to a stop articulation at the specified place, or as alveolar by default, moving to the approximation at the end of the onset. The only predictive statements we need to make relate to the place of the approximation, as in (7.30) and (7.31).

1. (7.30) [front] → S [dorsal]

2. (7.31) [round] → S [back], [dorsal]

Voicing starts at some point during the approximation, so a statement similar to the one given in (6.27) above is needed.

Variation in the realization of three-consonant onsets occurs in particular in terms of place of the approximation and place of the friction and closure. The whole of the articulation of the onset is affected by the kind of approximation used. The different kinds of phonetic implementation can be applied to the structures in (7.29), (p.140) so palato-alveolar, alveolo-palatal and post-alveolar points of contact are all referred back to a single underlying structure. In (7.32) I give a selection of possible realizations.

(7.32) [ʃtɹ-] [ʃtj-] [ʃtʃ-] [ʃkɹ-] [ʃkj-]

These are only some of the possibilities, and they are fairly broadly transcribed, but the point I am making here is that as far as three-consonant onsets in English are concerned a single phonological structure can cover a whole range of realizations.

Any exceptions to the constraints on onsets, such as /smj-/ in smew and /sfr-/ in sphragistics will have the unusual features specified in their lexical forms, as in (7.33).

Underspecification of lexical forms enables us to deal with exceptions in this straightforward way.

7.6 Variation and ponological forms

Throughout this book I have tried to determine the balance between phonetics and abstract phonology. Attention to phonetic detail is crucial before we can say for certain what aspects of the speech continuum are important to speakers and hearers. That they may not necessarily be the same for both parties has been pointed out by Cutler (1992), as we saw in section 4.1 above, but nevertheless the debate about the extent of the relevance of phonetic detail continues from various points of view (for example, Archangeli & Pulleyblank, 1994; Johnson & Mullennix, 1997; Local, Ogden & Temple, 2003; Scobbie, 2005a; Silverman, 2006; and the theme of the plenary session of the 16th Manchester Phonology Meeting, as reflected in Cohn, 2008 and Hawkins, 2008).

Silverman 2006: 114–87) discusses three approaches to the way (p.141) in which phonetic detail might be used in phonological systems: the relaxed constraints model, the prototype model and the exemplar model. All these models assume that phonological categories come from the actual experience a speaker (or group of speakers) has of the utterances around him or her. The relaxed constraints and the prototype models also assume that speakers have some kind of internalized targets for the sounds they hear and produce. In the former case there is a built-in tolerance of (a certain amount of) variation from the targets, a recognition of the nature of the human speech apparatus, which cannot be controlled with the same exact precision on each occasion of use. In the latter case there is an assumption that speakers have exact targets but are incapable of achieving them, so that all instances of utterances are ‘mistakes’. Speakers are, by their very nature, inaccurate articulators. But the problem with such an approach is that it does not tell us where the targets come from. How are they acquired? By listening to other, adult inaccurate speakers? The rôle of the prototype seems to be to support, or, perhaps, reflect, the segmental representations of citation forms accepted by most versions of phonology. And in what sense can we justify the notion of inaccuracy in speech, given the very nature of the speech organs? How ‘perfect’ can the operation of the speech organs be? On the other hand, the exemplar model does not assume some kind of preordained target, but, rather, that the phonological forms develop from within the range of utterances of any particular item a speaker hears. (For a discussion of probability matching in this regard, see Silverman, 2006: 120–54.) Speakers store all instances of utterance in their memory stacks, and the most recent instances that a speaker hears serve as the basis of new utterances. In this way change through one speaker's lifetime can be explained as well as generation-to-generation change. Of course, this process of change and stabilization has to take place within a community of speakers, that is, it must operate for the most part in face-to-face interaction, as was stressed above in section 7.1. If a speaker hears a Bolton accent on the television, she or he will not just pick up that form simply because it is at the top of the memory stack. Usable exemplars must be similar to ones already in the stack and in use. So [fз ː] fair, a Lancashire form, will not influence the speech of a Glaswegian, for example. Once again, we need to know what criteria of similarity to apply in such cases, and I have tried to give some answers in this book. I maintain they will mostly be functional criteria and some measure of phonetic similarity based on the phonetic implementation statements (p.142) of the particular system. Even if speakers do store their experienced utterances in the way the proponents of the exemplar model suggest, there will have to be some kind of phonological form to which these exemplars relate. Otherwise, how do speakers know they are the same linguistic item?