Dialect contact and distinctiveness: The social meaning of language variation in an island community

In this paper, we analyse linguistic variables which are well-established in British English, the vowels in the TRAP and BATH lexical sets. We demonstrate that the social meanings of these variables are both historically substantiated and locally-elaborated. Our data is taken from the speech of individuals living on the Isles of Scilly, a group of islands off the south-west coast of England. Our initial analysis shows that TRAP and BATH variants found on the islands are linked to contact with Standard English English, on the one hand, and the nearest neighbouring variety of Cornish English, on the other. The general distribution of variants is shown to reflect educational differences amongst our speakers. However, two case studies show speakers using forms atypical of their education type in order to position themselves in interactionally-dynamic ways. This reveals how speakers exploit the multidimensional meanings of linguistic variants to reflect and construct local practices and alignments.


INTRODUCTION
The search for social meaning has been variably prioritised in variationist research, in accordance with the research objectives of individual studies. While it is tempting to see one form of analytical practice as preferable over another, it is important to remember that different kinds of variationist research bring their own benefits and drawbacks. For instance, studies in what has been described as the third wave of variationist research (Eckert 2012) provide rich accounts of social meaning potentials, but their tendency to focus on the meanings associated with variables in a particular and specific set of interactional moments means that they sometimes lose sight of the long-term processes of language variation and change which give linguistic variables history. This can lead to researchers missing what Agha (2003: 247) calls the 'sociohistorical continuities in referential practices' that variables carry with them. Likewise, whilst more traditional language variation and change studies capture the trajectories of variables, they often fail to address Eckert's (2008: 462) observation that 'variables that historically come to distinguish geographic dialect can take on interactional meanings based on local ideology'that is to say, they lose sight of subtle local and contextual factors which transform the social meaning of a variable when it enters into a different repertoire.
In this paper, we show that it is possible to provide a more nuanced account of the social meanings associated with linguistic features by drawing upon a range of variationist techniques. We do this by focusing on language variation in a speech community on the Isles of Scilly, a small group of islands off the south-west coast of England. In particular, we examine some of the social meanings associated with two related variables, the vowels in the TRAP and BATH lexical sets. 2 Our analysis demonstrates how the social meanings of these linguistic features are inextricably linked to the specific social history of the speech community in which they are found. This is not to say similar processes of language variation and change are not shared by different speech communities; we begin by observing very familiar patterns of dialect contact and education effects in our data. We show: (1) that the kind of TRAP and BATH variants found on the Isles of Scilly are linked to contact with Standard English English, on the one hand, and the nearest neighbouring variety of Cornish English, on the other; and (2) that the general distribution of variants reflects educational differences amongst our speakers. However, we go on to show that the social meanings associated with TRAP and BATH can be ideologically reconstrued to reflect more local oppositions. Using two case studies, we consider how TRAP and BATH variation occurs in relation to topic and discourse positioning. This part of our analysis reveals that speakers are capable of using forms which are atypical of their education type when constructing social meanings linked to locally-specific practices and alignments.
Our analysis also exposes the limitations of superficially identifying vernacular variants as 'local' and more standard variants as 'non-local'. Social meanings rarely exist in the kind of uni-dimensional space that this implies, and we demonstrate this by showing how non-localisable forms can be used to embody particular social personae which do not necessarily conflict with a speaker's orientation to place. In this regard, our paper also contributes to recent debates about the importance of properly contextualising place and space in variationist research (See Eckert 2004;Johnstone 2004;Britain 2009aBritain , 2009bBeal 2010; Montgomery and Moore forthcoming).

THE RESEARCH LOCATION: THE ISLES OF SCILLY
The Isles of Scilly (pronounced ['sɪli]) are a group of islands situated twentyeight miles off the south-western tip of the English mainland ( Figure 1). Their population in the latest census was 2203 (Office for National Statistics 2011).
Seventy-five percent of the population live on St. Mary's, the largest of Scilly's five inhabited islands. St. Mary's is the only island to have an airport, a secondary school and key facilities such as a supermarket and banks. It is also where the passenger ferry and freight ship dock. The other islands (known as the 'off-islands') are Tresco, St. Martin's, Bryher and St. Agnes. Small passenger launches and (increasingly) jet boats link these islands to St. Mary's. The islands' major industry is tourism, which provides over 85 percent of the islands' income (The Isles of Scilly Council 2005: 14). There is some farming and fishing on the islands, with favourable weather conditions supporting flower farming in particular.
The islands have a particularly interesting history of governance. They were leased from the British crown by Cornish peer, Sir Francis Goldolphin, in 1571. It is generally believed that Scilly's current indigenous population was founded at this time, with the islands effectively repopulated by their new governor (Banfield 1888: 43-45;Bowley 1964: 69). Borlase (1756: 84), describes the islanders as 'all newcomers', and Mumford (1967: 62) observes that, under Godolphin, '[l]and was divided up into plots and allotted to tenants; people began to come to the islands; employment was found'. The islands' lease continued in the Godolphin line until 1834 when it was taken over by Hertfordshire landowner, Augustus Smith. The lease remained in Smith's family until 1920, at which point all but one island -Trescoreverted to the Duchy of Cornwall (a title currently held by Prince Charles). Today, the Duchy own most of the freehold on the islands, with the exception of St. Mary's main town, Hugh Town, which was sold off in 1949. The onset of Augustus Smith's governorship of the islands in 1834 effected significant social change. Novelist and playwright, Wilkie Collins, who visited the islands in the mid-nineteenth century, noted that Smith 'succoured, reformed, and taught [his tenantry]; and there is now, probably, no place in England where the direr hardships of poverty are so little known as in the Scilly Islands' (Collins 1861: 93). Smith effected various improvements to the islands, including the building of new roads and the development of shipping and piloting opportunities (Uren 1907: 67). However, Smith was an uncompromising governor: 'the ne'er-do-weels were dispatched to the mainland; sons were not allowed to remain at home on the farm if there was not sufficient work for them; schools were opened, and education made compulsory long before it was so in England' (Mothersole 1914: 48).
This early access to education, and the 'dispatching of the ne'er-do-weels' gave some Scillonians a sense of privilege which, they believed, differentiated themboth socially and linguisticallyfrom their nearest neighbours on the Cornish mainland. This can be seen in Extract 1, written in the Gentleman's Magazine in 1888 by Scillonian in exile, Frank Banfield. In this quotation, Scillonian English is juxtaposed with varieties of Cornish English and aligned with more standard forms of English:

Extract 1
There have been immigrants from Cornwall . . . but the Godolphin settlers having had the upper hand for so many generations have, impressed their own correcter locution and more Eastern English of inheritance and education upon the population. (Banfield 1888: 45) Scilly has always had 'immigrants from Cornwall'. For instance, the 1901 census shows that 30.4 percent of Scilly's population of 1877 people were born in places other than Scilly and, of these incomers, 48.3 percent (276 people) were from Cornwall (data from The Isles of Scilly Museum 2007). Nonetheless, other historical commentary made by visitors to Scilly supports Banfield's view that the Scillonian dialect is not as broad as that of its Cornish neighbours. Extracts 2, 3 and 4 exemplify how Scillonian English was characterised by island visitors in the eighteenth, nineteenth and twentieth centuries respectively.

Extract 2
The language spoken in Scilly is a mixture of the West Country dialect, with the common English; the islanders in general speak much better English than a stranger would expect to find in their detached situation. (Troutbeck 1794: 168) Extract 3 . . . not only is English spoken here, but spoken with a purity of accent, and intelligent discrimination of diction, which I remember in no other part of the English dominion. (Lewes 1860: 205) Extract 4 . . . even the English they speak lacks a marked peculiarity . . . Yet one can trace in it, not unnaturally, a tinge of Cornish dialect and dialectal usage; and a collection of island wordsnever yet made so far as I know, though there are several still in usemight reveal the islanders' linguistic kinship with Cornwall. (Grigson 1948: 20) There is no significant linguistic work to substantiate this metalinguistic commentary. Relying on the word of 'the Proprietor of the Isles', who said he did 'not know of any part of the British Isles in which "the Queen's English" is less murdered', Alexander Ellis (1890: 41) determined that 'no attention . . . need be paid to [Scilly]' when compiling his account of early English pronunciation. Likely as a consequence of this observation, the islands did not feature in the twentieth century Survey of English Dialects (Orton 1962) either. However, historian and archaeologist, Charles Thomas, produced A Glossary of the Spoken English in the Isles of Scilly in 1979, in which he comments on the phonology of the dialect. He claims that Scilly's English was 'scarcely removed from Standard (southern) English' and 'notably different from the group of English dialects encountered in Devon and Cornwall' (Thomas 1979: 109). This demonstrates that claims about the 'purity' of Scillonian English continued well into the twentieth century.
This commentary suggests that Scillonian English is a variety with little local flavour, that is to say, one which contains little to interest the traditional dialectologist. Given the historical context, this is, perhaps, unsurprising. Although we might expect the language of a locale to share similarities with the varieties of its closest neighbours, it is important to remember that Scilly is separated from its closest neighbour by 28 miles of ocean. Furthermore, Scilly's history suggests disruption in the native population across time (with something of a tabula rasa situation on Scilly in the sixteenth century). This would also reduce the possibility of continuity in the variety spoken on the islands. As Trudgill (2004: 23) has observed in his work on colonial Englishes, where there is dialect mixture, with no indigenous input, levelling across the available dialects is likely to occur. In these cases, socially and linguistically marked forms are lost, and this may explain why new levelled varieties are often considered to be 'better' or 'purer' than other, older, varieties of English. However, in the absence of systematic linguistic description of Scillonian English, it is not possible to separate the ideologies about the Scillonian variety from the realities of Scillonian speech and its influences. Even Thomas' (1979) account, which is the most empirical of the literature on Scillonian English, relies upon dubious sources (it lists eight 'oral informants', of which three are not native to Scilly), and is somewhat ambiguous in its claims (it states that Scillonian English is akin to RP, but later concedes that, if Scilly's dialect is like Cornish English, it is more similar to the eastern areas of Cornwall than the western areas).
In accounting for the social meanings associated with the Scillonian variety of English, we first provide a systematic account of how the variety relates to the other varieties most closely associated with it in the historical metalinguistic commentary: Cornish English and Standard English English. Our aim in investigating these associations is to fully understand the trajectory of Scillonian English and to comprehend the ideological implications of any dialect contact with these cited varieties. In the next section, we outline how we undertook this analysis.

THE DATA
The TRAP and BATH lexical sets are ideal variables for exploring the relationship between Scillonian English, and Cornish English and Standard English English (StEE). This is because the vowels in these lexical sets pattern differently in Cornish English than they do in more standard varieties of English.
Although the precise trajectory of these lexical sets has been debated (see Lass 1976: Chapter 4;MacMahon 1998;Beal 1999Beal : 105-111, 2004Piercy 2010: 9-24), Table 1 illustrates what is generally accepted to be the development of TRAP and BATH in StEE and Cornish English. It shows that TRAP and BATH have been distinguished by both duration and vowel quality since the eighteenth century in StEE. However, traditional varieties of Cornish English (like many other rural Southern English English varieties) only shows a split by duration and not by vowel quality. Wakelin (1975Wakelin ( , 1986, using data from the Survey of English Dialects, observes that the changes in Cornish English were fossilised at the stage at which English was introduced into the Cornish area following the loss of the Cornish language. In order to establish the TRAP/BATH pattern in Scillonian English, we compared the oldest Scillonian data available with contemporaneous data from Cornish English and StEE. Table 2 shows our sample. The Scillonian speakers were drawn from the Isles of Scilly Museum's Oral History Archive. This is a series of recordings made by local people interviewing other local people and recordings date from the 1970s onward (the archive can be searched online here: www.hrionline.ac.uk/scillyvoices). The purpose of the archive is to record the experiences of Scillonians, and the informants are identified by museum volunteers on the basis of their 'Scillonian character' (a vague criterion, but one which includes consideration of Scillonian heritage, community roles, and how well-known someone is within the community). We sampled six of the oldest men born on St. Mary's, Scilly's largest island. We focused on the men to enable us to compare these recordings with the 1960s interviews with the Cornish Survey of English Dialects (Orton and Wakelin 1968) participants, who were born at the end of the nineteenth century, and were all male. (For an account of female speakers' use of these variables, see Moore and Carter forthcoming.) 3 As Table 2 shows, we divided the Survey of English Dialect locations into east and west Cornwall, according to their geographical location. These locations are shown in Figure 2. Notice that we also split our Scillonian speakers according to a locallyrelevant social difference, that is, education type. Prior to 1966, Scilly did not have its own secondary school, consequently, children were either sent away to boarding school between the ages of 11 and 16, or they remained in the allage school on St. Mary's until the age of 14. 4 It is important to stress that this education difference does not reflect different orientations to the islands. All of the Scillonians sampled talked passionately about their home in their interviews. Our mainland-educated speakers may have had a period away from the islands, but they considered themselves to be no less Scillonian than the Scilly-educated Scillonians. Furthermore, their inclusion in the archive suggests that other locals did not consider them to lack 'Scillonian character' status either.
The StEE recordings are drawn from a set of recordings held by the British Library (http://sounds.bl.uk/accents-and-dialects). These are a series of recorded scripted interactions between StEE speakers produced by Linguaphone in order to educate non-native speakers of English. Given their purpose, it can be assumed that these recordings exemplify very formal StEE from speakers who were born around the same time as our other informants. Whilst the style of the recordings is somewhat different from the Scillonian and Cornish samples (both of which are conversational in style), they nonetheless serve to provide an illustration of model StEE that British speakers were exposed to in the twentieth century. We sampled all seven of the men found in these recordings, all of whom were associated with UCL's Linguistics department when the recordings were made in around the 1930s.
In order to analyse TRAP and BATH for these speakers, we extracted formant data, sampling formant tracks every five milliseconds through each vowel, with LPC order set to appropriate values for each speaker, in each case checked by visual inspection of spectrograms. This allowed us to obtain a range of descriptive statistics for the F1 and F2 trajectories in each vowel; given that we deal with monophthongs in this paper we report only on the median value for each formant in the vowel. (As is now standard in much sociophonetics work, we focus on F1 and F2 in our analysis to consider vowel quality, given that F1 has been found to correlate with vowel height, and F2 with how front or back a vowel issee Ladefoged 1982, amongst others; moreover, the quality of these old recordings was sufficient to have some confidence in measuring the lower formants although above the F2 region the signal was often attenuated.) We also measured the duration of each vowel. We transformed durations into a logarithmic domain to account for the fact that hearers seem to perceive durations as ratios rather than as absolute amounts. Logarithmicallytransformed durations also approximate the normal distribution more closely, so this kind of transformation also made sense statistically. We also transformed our formants into the domain of the equivalent rectangular bandwidth (Glasberg and Moore 1990), as a step towards speaker normalisation. For clarity of presentation, we provide axis notation in our figures in Hertz for spectral data and milliseconds for temporal data.
We coded for a number of phonological environment factors, namely the position of the syllable in question within each utterance (initial, final or somewhere in the middle of the utterance), the number of syllables in the word, whether the rhyme of the syllable was open or closed, the manner of articulation and the voicing of the following consonant, and whether the syllable seemed to be carrying sentence stress. We also coded for lexical versus grammatical words.
To analyse our results, we used two forms of statistical modelling to test the patterns suggested by the raw data: random forest variable importance measures, and mixed effects modelling. Mixed-effects regression modelling allows us to specify the individual speaker as a random effect, thus allowing us to draw conclusions which generalise beyond the speakers in our corpus. Having speaker as part of the random effects structure of models also provides a degree of normalisation across speakers.
However, the unbalanced nature of a relatively small corpus of recordings of speech gathered for purposes other than linguistic analysis means that regression models are not always possible to construct and, when they are, their robustness is questionable. For this reason, we also incorporate conditional-inference regression trees, and random forests (Strobl et al. 2008(Strobl et al. , 2009a(Strobl et al. , 2009b. The benefits of using this type of analysis on linguistic data (which may include small archive corpora such as ours) has already been reported (Tagliamonte and Baayen 2012). Random forests are random in the sense that randomly-chosen subsets of the data and randomlychosen subsets of the predictors are used over a number of iterations of the decision tree algorithm and the effect of each predictor can then be compared. Because repeated subsets of data and variables are chosen, alternative splits can emerge if the number of iterations is high enough. We start from a base of 500 trees per forest, increasing that as necessary if the results are not robust (where we interpret robustness to mean the same ordering of variable importance across ten random forests).
For the random forests reported in this paper we plot the mean values across ten random forests as bar charts. The numerical values associated with each predictor are not comparable across different analyses (Strobl, Malley and Tutz 2009a); what matters is the relative ranking of predictors, represented in the lengths of the bars (with longer bars reflecting more important predictors in any individual analysis). 5 In the next section, we describe the outcome of this analysis. Figure 3 shows the densities of the distributions of acoustic parameters. The top row shows two-dimensional densities for the first and second formants. TRAP vowels are shown in grey and BATH vowels in black. The contour plots show where the greatest co-occurrence of realisations exists. The second row of the figure shows two-dimensional densities for duration. The peaks indicate the distribution of different durations. Again, TRAP vowels are shown in grey and BATH vowels in black. The raw data in Figure 3 suggests that TRAP and BATH are split by quality in StEE. The two lexical sets show no overlap in F1 or F2. Of the other varieties, mainland-educated Scillonian English (meScE) appears to have the next most innovative pattern: there is overlap of TRAP and BATH, but the BATH vowels show a tendency to have lower F2 values than the TRAP vowels. East Cornish English (eCE) also shows a tendency towards this quality distinction, although there is much more overlap between the TRAP and BATH values. On the other hand, both the west Cornish English (wCE) and the Scilly-educated Scillonian English (SeScE) dialects appear to show no distinction in quality for TRAP and BATH; the raw data shows complete overlap between TRAP and BATH values (although the SeScE data appears to have a wider F1 range than the wCE data). The second row of Figure 3 suggests that all dialects have a duration difference between TRAP and BATH, with the latter lexical set typically having longer vowel durations.

RESULTS: THE RELATIONSHIP OF SCILLONIAN ENGLISH TO CORNISH ENGLISH AND STANDARD ENGLISH ENGLISH
Random forest variable importance measures shed more light upon this data and, importantly, enable us to evaluate the effects of linguistic factors on our results. Figures 4 and 5 show the relative importance of the factors predicting F1 and F2, respectively, in TRAP and BATH. Figure 4 shows that all of the expected linguistic factors are important in predicting the F1 values of TRAP and BATH tokens. Nonetheless, this figure also shows that 'speaker group' (the dialect in which a token is uttered) is the most important factor in predicting F1. Lexical set is also significant, but less so. 'Speaker group' may be more important than 'lexical set' because, as Figure 3 shows, whilst different dialect types have different F1 values, only the StEE speakers show marked F1 differences between the two lexical sets. Figure 5 shows that the expected linguistic factors are also important in predicting the F2 values of TRAP and BATH tokens, but, in this figure, both 'speaker group' and 'lexical set' are stronger predictors of F2 than the linguistic factors. The dual significance of 'speaker group' and 'lexical set' is likely because, as Figure 3 indicates, different groups have different F2 values from one another, and several groups seem to exhibit some use of F2 to distinguish TRAP and BATH.

Figure 3:
The densities of the distributions of acoustic parameters for TRAP and BATH, across the following varieties of English: Scilly-educated Scillonian English (SeScE); mainland-educated Scillonian English (meScE); west Cornish English (wCE); east Cornish English (eCE); and Standard English English (StEE) Figure 6 shows the relative importance of the factors predicting duration in TRAP and BATH. As with Figures 4 and 5, it shows that the expected linguistic factors do indeed affect duration. Note, though, that lexical set (whether a token is labelled as TRAP or BATH) is still a significant predicator of duration, despite the inclusion of linguistic factors in this model. That is to say, there is general correspondence between the duration of a token and the token's lexical set, and this holds even when phonological environment is included in a statistical model. However, Figure 6 also shows that the second strongest effect on the predicton of duration in TRAP and BATH is 'speaker group'. This suggests that there may be differences in how dialect groups use duration to distinguish TRAP and BATH. This is harder to discern in the raw data, as Figure 3 suggests that there is a remarkable consistency in using duration to distinguish TRAP and BATH across dialect groups. Of course, it could be that this raw data is skewed by the interaction of linguistic factors.
Further mixed effects modelling (incorporating both word and speaker as random factors) confirms that there seems to be a genuine difference in the use of duration across groups. Table 3 shows the outcomes of this kind of modelling. The mixed effects modelling shows that F1, F2 and duration are all important for StEE, but F2 appears to be the most important effect. This means that, as expected, StEE speakers differentiate TRAP and BATH by vowel quality and duration. On the other hand, duration is the only significant effect for east  Cornish and Scilly-educated Scillonian speakers (although random forests results for individuals in the eCE group do show some effect of F2, or vowel quality). West Cornish speakers do not show any significant effects, but this may be a consequence of inadequate data for this dialect type, where tokens were more sparse (random forests results for individuals in the wCE group suggests that duration is a predictor of TRAP and BATH for this group). Intriguingly, duration seems to disappear as a significant effect in the data from the mainland-educated Scillonians' group, and only F2 is significant. That is to say, the meScE speakers seem to differentiate TRAP and BATH by vowel quality, but not by duration. This is interesting, given that duration is the only way in which Scilly-educated speakers seem to differentiate TRAP and BATHwe return to this below. These results suggest that Scillonian English has a complicated relationship with Cornish English and StEE. The first thing to note is that there are two competing TRAP/BATH patterns amongst our group of Scillonians. Those educated exclusively on Scilly seem to pattern most closely with the Cornish English speakers. This is because they predominantly distinguish TRAP and BATH by duration, and the vowels in both lexical sets typically have a front quality. It is difficult to discern whether their vowel qualities are most like east Cornwall or west Cornwall speakers, given the F1 range of these speakers, as shown in Figure 3 (note also, that these speakers are capable of very open TRAP vowelsa matter we return to in section 5). This suggests that the Scilly-educated Scillonians have an archaic TRAP/BATH pattern, which we will refer to in subsequent discussion as 'Scillonian Pattern 1'. Whatever variation existed on the islands when it was repopulated in the late sixteenth century, the predominant form of the day seems to have been maintained, and perhaps also supported by ongoing contact with Cornish English speakers via migration into the islands (as discussed in section 2). On the other hand, the mainland-educated Scillonians seem to pattern more like StEE, as evidenced by their low F2 scores for BATH: they distinguish TRAP and BATH predominantly by quality, with BATH vowels more retracted than TRAP vowels. We will refer to this in subsequent discussion as 'Scillonian Pattern 2'. This pattern undoubtedly reflects the norms that our mainland-educated speakers encountered in their mainland boarding schools (see note 4), but it may also reflect ongoing contact with StEE speakers on the islands themselves via governing staff and Duchy of Cornwall management employees. However, it should be noted that, whilst there are some similarities with StEE, their TRAP/BATH patterns do not mimic StEE precisely.
What we find most intriguing about the mainland-educated speakers' data is that they have maintained a pattern which is distinctive from the Scilly-educated speakers across their lifespans (none of the speakers was younger than 60 when he was interviewed, and the only significant period of time away from the islands experienced by these speakers occurred during their schooling). We are also intrigued by the finding that our mainland-educated speakers are less likely to differentiate TRAP and BATH by duration than our StEE speakers, given that this is the main way our Scilly-educated speakers distinguish TRAP and BATH. This leads us to believe that the distinctiveness of the Scilly-educated and the mainlandeducated TRAP/BATH vowels is not just a consequence of dialect contact; it is also supported by island-internal factors and local forms of differentiation. We explore this possibility in the next section, where we consider the social meanings associated with TRAP and BATH on the islands.

THE SOCIAL MEANINGS OF TRAP AND BATH IN SCILLONIAN ENGLISH
Building on Silverstein's (2003) use of the concept of 'indexicality', Eckert (2008) introduced the notion of the 'indexical field' to define the ideological landscape of meaning for particular linguistic items. The indexical field relates a particular indexical value (that is to say, a social association or meaning) with indexical values of higher or lower orders. Values are activated in an indexical field when there is an ideological link between a particular linguistic item and a specific form of social meaning. These social meanings may be at a range of semiotic levels. Drawing on this literature, Moore and Podesva (2009) refer to three distinct forms of social meaning, 'social type', 'persona' and 'stance', and we follow that practice here: • 'social type' refers to demographic and regional categories of speakers; • 'personae' refers to local social categories (with particular communityrelevant associations); and • 'stance' refers to fleeting forms of positioning or affect, which are activated within the context of a particular interaction.
Our analysis thus far suggests that there may be an association between the two Scillonian patterns for TRAP and BATH and the two varieties most frequently cited in the historical metalinguistic commentary about the variety: Cornish English and StEE. The metalinguistic commentary we encountered in section 2 suggested that these frequency correlations reflect meaning potentials that exist within the community, given the tendency to cite these varieties in talk about Scillonian English. These associations were also found in responses to a questionnaire circulated by Scillonian schoolchildren in a knowledgeexchange project conducted by Emma Moore with the local school in 2008. For example, see the comment in Extract 5.
Extract 5: 26-year-old IT technician, lived on Scilly for five and a half years 'Two groups of accents seem to exist here: -Cornish -"Rather Posh" . . . Although many accents are apparent from other regions of the UK.' Our correlational analysis also revealed links to education type. Evidence to suggest that there are ideological associations between education type, prestige and styles of talk can be found in recent interviews conducted with Scillonian descendants of speakers in the museum's archive. For example, Extract 6 is taken from an interview conducted by an island interviewer (I), talking to the son of one of our mainland-educated speakers (M), and his island-born wife (F). Although these speakers do not make claims about the educated-associations of StEE speech, it is clear that they associate not having an accent (which, for non-linguists, typically means having a regionally-unmarked form of speech, such as StEE) with being educated on the mainland, and with being 'posh'.
Extract 6  Considering these results, we might imagine two distinct (and, at this point, simplistic) indexical fields for the TRAP and BATH patterns we noted earlier: Scillonian Pattern 1 (as typified by our Scilly-educated Scillonians), containing the social-type value 'Cornish'; and Scillonian Pattern 2 (as typified by our mainland-educated Scillonians), containing the social-type value 'Educated'. The types of meanings in these fields reflect the ideological baggage that variants of TRAP and BATH carry in the wider geographical area beyond the Isles of Scilly. For instance, in studying a change in progress in /a/ in Dorset (also in the south-west of England) where a more StEE-like pattern is emerging, Piercy (2010: 252) observes that the variation in realisations in Dorset may be 'attributable to social class difference and their correlates e.g. socially mobile versus non-mobile speakers, educational attainment and the perceived necessity to use standard versus regional dialects'. This suggests that, in the broader south-west area, forms of TRAP and BATH that differ from StEE are linked to less prestigious education-types and/or whatever regional dialect is relevant to the speakers studied. These ideological links would be easily transmitted via dialect contact within and beyond the Isles of Scilly.
These social meanings operate at a level of 'social type'they are linked to broad demographic categories of speakers. We now turn to the question of whether there are any social meanings that are more specific to the Scillonian context. There are clues to the range of local personae in the interviews themselves. Table 4 shows the distributions of topics by speakers in the interviews studied, with the three Scilly-educated Scillonians at the top of the table and the three mainland-educated Scillonians at the bottom. The shading shows the three speakers who talk most about the topic shown in each column (unless rates are zero for more than three speakers, in which case all those with rates above zero are shaded). The figures were calculated by allocating topics to speakers' turns and then recording the duration of each turn. Percentage of each topic per speaker was calculated by dividing the duration of talk about a specific topic by the total duration of that speaker's talk and multiplying by 100.
Recall from section 2 that the purpose of the interview was to reminisce about Scillonian life. The most or second most frequent topic of conversation for the mainland-educated speakers is their management responsibilities. These might be responsibilities to do with their own businesses, or authoritative roles such as being local councillors or magistrates. The Scilly-educated Scillonians, on the other hand, are more likely to talk about local history (that is events and occurrences on the islands, such as the World Wars, their roles in voluntary organisations such as the lifeboat service, or shipwrecks) and/or their employment on the sea. The distribution of these topics reflects what is known about the roles and responsibilities of these islanders, as shown in Table 5, and, most likely, reflects the social activities (and associated practices) which have figured significantly in the course of their lives, as expressed through their talk. 7  Eckert (2001: 123) has argued that there is a relationship between 'the events within the interview and the events that are embedded in everyday situations in which meaning gets made'. That is to say, considering what people talk about, and how they talk about it may provide indications of how linguistic features become associated with local personae. Table 5 seems to suggest that variation in TRAP and BATH on Scilly does not just correlate with the social types discussed above, but that it may also link to personae types. As such we might expand the indexical field for Scillonian Pattern 1 to include local personae such as 'seafarer' or 'workman'; and the indexical field for Scillonian Pattern 2 to include local personae such as 'authoritarian' or 'manager'.
The existence of these localised meanings suggest that, whilst speakers' TRAP and BATH forms correlate with their education type, if other social factors are interactionally relevant, we might expect speakers to use forms which are atypical for their educational type. We have two case studies which suggest that this is the case. The first comes from one of our Scilly-educated speakers, and the second from a mainland-educated speaker. Table 4 shows that one of our Scilly-educated speakers, Charlie, has a tendency to talk about management responsibilities. Charlie was the only one of our Scilly-educated speakers who had also served on the islands' council and the discussion of this role makes up his talk about management. When we analysed the interview in which he discussed his council responsibilities, we found two main distinctions in the content of his talk. Charlie tended to talk about management responsibilities (the council), and local history. Talk about the latter either involved telling stories about island characters, often from his childhood and involving some kind of mischief ('character reminiscence'), or recalling facts about island places or businesses, such as their locations or management (we refer to this as 'neutral reminiscence', due to the factual nature of this kind of talk). Charlie only had 18 tokens of BATH in this interview and, as most of these tokens occurred in relation to 'character reminiscence', there was insufficient data to discern any patterns in the distribution of the variants in this lexical set. However, Figure 7 shows the distribution of Charlie's TRAP vowels according to the categories of 'character reminiscence', 'neutral reminiscence' and 'council'. Figure 7 suggests a split in the data for 'character reminiscence' and 'council', with the vowels categorised as the former clustering lower in the Figure 7: The distribution of Charlie's TRAP vowels according to topic vowel space, and vowels categorised as the latter clustering higher and fronter in the vowel space. Vowels categorised as 'neutral reminiscence' do not show any discernible pattern (we return to this below). Of course, it is possible that linguistic factors explain the oppositional patterning of the 'character reminiscence' and 'council' tokens. Random forest variable importance measures suggested that both F1 and F2 were significant predictors of topic soto test for interactions with linguistic factors -Figures 8 and 9 show a Conditional Variable Importance (mean of ten 500-tree forests) 0.005 0.010 0.015 Figure 9: A random forest analysis of the relative importance of the factors predicting F2 in TRAP and BATH for Charlie's data, categorised according to the topics 'character reminiscence' and 'council' (■ = significant; □ = not significant) random forest analysis of the relative importance of the factors predicting F1 and F2, respectively, in TRAP and BATH for the data categorised as 'character reminiscence' and 'council'. Figure 8 shows that linguistic factors do, indeed, affect vowel heightonly the context of following voicing is not significant. Duration is a particularly important predictor (no doubt because longer vowels allow time for more open articulations), and following nasal environments are also significant, as is the word itself. However, notice that topic (that is, whether a word is categorised as a 'character reminiscence' or 'council') is also significant, albeit less so than the relevant linguistic factors. Figure 9 shows that topic is also significant in predicting F2in fact, only word is a more significant predictor. Duration may have some effect but a following nasal or voiced context is not significant in predicting F2. Figures 8 and 9 suggest that topic has a robust effect on the vowels Charlie produces in this interview. When he is telling stories about himself or local characters like him, he tends to use more open vowels, and when he is talking about the council, he tends to use more close and more front vowels. As mentioned at the end of section 4, the raw data from all of our speakers shown in Figure 3 suggests that only Scilly-educated Scillonians have very open vowels. Thus, it may be that Charlie is using more of these especially SeScE-like vowels when he is talking in a playful way about his upbringing on Scilly.
Previous sociolinguistic research has demonstrated the importance of topic to style shifting (for instance, Rickford and McNair-Knox 1994;Love and Walker 2013), so finding this kind of patterning in Charlie's data is hardly surprising. However, why is it that there is no obvious patterning for the vowels categorised as 'neutral reminiscences'? We would argue that this is because Charlie's style shifting is not just occasioned by the topics of his talk. Work in third wave sociolinguistics has suggested that the use of a linguistic variant is motivated primarily by the stances taken by virtue of using a particular variant (Ochs 1992;Kiesling 1998;Kirkham 2013). Stance has been construed in many different ways in sociolinguistics (see Jaffe 2009, for a review), although most definitions involve some kind of evaluation. At its simplest, it can be described as 'the attitude or position that a person takes in conversation' (Kiesling 2009: 193). Consequently, we might interpret stance shifting as occasioned by the content of the talk itself, but also by a speaker's orientation towards or alignment with the content of the talk.
Unfortunately, as Agha (2005: 53) has observed, the multiplicity of alignments individuals make in interaction far exceeds our ability to identify them in posthoc analysis. This means that it is difficult to decipher how Charlie's stances towards places and businesses on Scilly might have affected his talk. However, our second case study provides evidence that stance and alignment do have a role to play in style shifting. This case study focuses on one of our mainland-educated speakers, Victor.
Our focus on Victor's data comes from a desire to explain the anomalous cluster of fronted BATH vowels shown in the results for our mainland-educated Scillonians in Figure 3. This cluster is made up of four tokens, three of which occur in a section of talk in which Victor is reminiscing about the Isles of Scilly Steamship Company and listing the company's captains. This company runs the ferry service between the islands and the mainland, and Victor had various management roles in this company over the course of his lifetime. The extract of talk, given in 7, is marked up for when Victor is talking about particular captains. Extract 7 also shows where the topic shifts towards the end of this section when Victor begins talking about company managers, rather than captains (at line 113). Words containing a TRAP vowel are in small capitals and bolded, and words containing a BATH vowel are in small capitals, italicised and underlined.
Extract 7: Victor [Cpt. Macbeth] 1 Oh yes, the first one was CAPTAIN MACBETH. 2 I used to be a BAD sailor and he used to 3 come down in the saloon, I was always 4 stretched out, to talk to me. He spoke such 5 broad scotch, I HAD a job to UNDERSTAND him 6 when I was feeling well and I was saying yes 7 and no all the time but I didn't know what I 8 was saying yes and no about! [Cpt. Rowan] 9 And he HAD a mate CAPTAIN Bob Rowan . . . 10 which when he retired, we made him skipper. 11 ACTUALLY he didn't HAVE his CAPTAIN'S 12 ticket, but we knew he was such a good MAN 13 we sent him up to Liverpool at the company's 14 expense to take his ticket. Well he failed. 15 So we sent him up again and he PASSED! And 16 he proved . . . well he was there for 17 many, many years and he really was a MAN of 18 the sea and he looked the part! He MARRIED a 19 local girl and I remember I was up in London 20 one time and I went to Covent Garden early 21 in the morning and came BACK and I bumped 22 into CAPTAIN Rowan and his wife, they'd been 23 MARRIED the previous day here in the 24 islands. I HAD a CHAT with them and in 25 CHATTING to me they got on the wrong bus and 26 instead of going to Liverpool they found 27 themselves in Warrington. But then AFTER a 28 while . . .. [Cpt. Bailey] 29 Let me see, who was the next skipper? We HAD A closer look at the discourse in Extract 7 reveals that Victor does not evaluate all of the Steamship Company's captains in the same way. The majority of the captains are evaluated positively; for instance, Captain Macbeth is depicted as a kind man who looked out for the seasick Victor (lines 1-4); Captain Rowan was 'a good man' (line 12); Captain Williams was 'a very, very good skipper, a disciplinarian' (lines 36-37); Captain Bailey was 'a jolly good chap' (lines 73); and Captain Morris is 'doing a very good job now . . . very efficient' (lines [91][92][93][94]. However, two captains -Captains Sanford and Atkinsreceive more negative evaluations. Captain Sanford is marked out as being Cornish (from Hayleline 40) and he is introduced as 'a sick man' (line 42) who 'couldn't be bothered with discipline' (lines 42-43)indeed, he allowed discipline to get 'rather slack' (lines 43-44). Captain Atkins was the only Scillonian to captain one of the larger Steamship Company vessels, and he was educated entirely on Scilly. He is described in ways which imply that he was rather unambitious -'unfortunately he gave up the sea' (lines 95-96), following 'persuasion' from his wife (line 104). Later in the extract, Victor goes on to note how Captain Atkins went on to be appointed a manager of the Steamship Company, but he is at pains to point out that Captain Atkins did not make it to being a managing directora position reserved for another of Victor's mainland-educated peers (Jim) and his own mainland-educated son (lines 113-117).
There are only five BATH tokens in Extract 7 but, as mentioned above, three are from the anomalous cluster shown in Figure 3. Two of these three occur in talk about Captain Sanford (lines 56 and 86), and the third occurs in a narrative segment which concludes with Captain Sanford's death (line 78). There are no BATH vowels in talk about Captain Atkins, so it is not possible to consider how this lexical set patterns in talk about him. However, there are several TRAP vowels in talk about both of these captains. Figure 10 shows all of the TRAP vowels in this extract. Vowels in grey are those uttered in talk about Captain Sanford and Captain Atkins, and vowels in black are those uttered in talk about other members of the Steamship Company. Figure 10 suggests that TRAP vowels which occur in talk about Captains Sanford and Atkins tend to be more open.
Again, it could be that these results are a consequence of linguistic factors. Random forest variable importance measures suggested that only F1 was a significant predictor of the vowel space distinction between talk about captains. Consequently, Figure 11 shows the factors predicting F1 values in the extract of talk given in Extract 7. Figure 11 shows that linguistic factors do predict the F1 values of the TRAP vowels in Extract 7. As with Charlie's data, duration is the most significant predictor. Voicing of the following context is not significant, but a following nasal is important, as is the word itself. Notice, though, that the kind of Steamship Company employees the talk is about is also a significant predictor, as shown by the 'Captains' bar in Figure 11. This suggests that F1 values are significantly different in talk about Captains Sanford and Atkins than in talk about other Steamship Company employees, and that this effect is independent of phonological environment.
Topic is clearly not a sufficient factor to explain why talk about Captains Sanford and Atkins occasions more open TRAP vowels, as Victor is talking about the same topicthe Steamship Companythroughout this segment of talk. However, the way in which Victor evaluates these men, or his stance towards them, may beat least, there seems to be a correlation with more typically mainland-educated vowels (TRAP vowels which are more close and BATH vowels which are more retracted) and depictions of authority, discipline, ambition and institutional status. On the other hand, Victor uses more vowels typical of a Scilly-educated speaker (TRAP vowels which are more open, and BATH vowels which are more front), when distancing himself from those whoin his opiniondon't achieve this kind of status. The case studies involving Charlie and Victor suggest that the social meanings associated with the distinctive patterns of TRAP and BATH are not limited to social types or personae; they also extend to more fleeting expressions of stance, positioning and affect. Taking all of the social meanings uncovered in this section, we propose the indexical fields shown in Figure 12.
Notice that Figure 12 shows social meanings that are associated with demographic social types (shown in italicised upper case), local personae (shown in regular upper case) and more fleeting stances (shown in lower case), and that social meaning seems to operate at all of these indexical levels. 8 In this way, our paper provides support for the existence of multiple, connected, and ideologically-mediated 'levels' of social meaning discussed elsewhere (Bucholtz and Hall 2005;Johnstone, Andrus and Danielson 2006;Coupland 2007;Johnstone and Kiesling 2008;Moore and Podesva 2009;Snell 2010).
It is important to note that the meaning potentials given in Figure 12 are not intended to comprise an exhaustive list, rather an illustration of the kind of meanings that we have been able to uncover in the course of our analysis. No doubt other meanings are possible, based on ideological connections with the types of meanings we have identified. Also, we do not make claims about the hierarchical status of different levels of meaning; which meanings are activated in a specific context will depend upon how interlocutors engage with the ideologies that facilitate links between social meanings and linguistic items. As  Moore and Podesva (2009: 479) note, 'at any one time . . . meanings may be discrete and in competition with one another . . . depending upon who . . . hears [a linguistic item] and the ideological lens through which it is filtered.' Retrieving these levels of meaning required different variationist techniques. In the next section, we consider what these results suggest about the ways in which social meanings operate, and the ways in which we examine the social meanings of linguistic features.

DISCUSSION AND CONCLUSION
We have demonstrated how linguistic forms can take on complicated social meanings that span a range of semiotic levels. However, it is important to note that, despite the multiple social meanings we have identified, the speakers we have analysed all share one thing: they are all Scillonians who are passionate about their home island and who are recognised and acknowledged as true island locals (see discussion of data collection procedures in section 3). Previous sociolinguistic studies have tended to focus on individuals' use of local vernacular forms to make claims about local and regional identity. This is most clearly exemplified by Labov's (1963) seminal Martha's Vineyard study. In identifying the social meaning of centralised diphthongs on the island, Labov (1963: 36) notes that the immediate meaning of this feature is 'Vineyarder'. When a man says [reɪt] or [hɐʊs], he is unconsciously establishing the fact that he belongs to the island: that he is one of the natives to whom the island really belongs.
Of course, this is evident from Labov's meticulous analysis but, as Eckert (2008: 462) notes, the English-descent fishermen's use of a traditional, local feature is not just about them claiming to be Vineyarders but about them 'making a claim about what a Vineyarder is.' There are other ethnicities on the island (as Labov discusses) and it is likely that they feel their own kinds of loyalty to the islands, but their different social histories may lead to different experiences of 'ownership' and, ultimately, differential semiotic marking of 'Vineyarder' status. None of this makes them any less 'Vineyarders'it just offers up a different perspective on what a 'Vineyarder' is. Similarly, the different TRAP/BATH patterns that we see on the Isles of Scilly and their associations with different personae and stances suggest that there are competing ways to embody a Scillonian persona on the islands. Whilst it is tempting to see vernacular variants as 'local' and more standard variants as 'non-local', our results suggest that linguistic forms can take on more complicated social meanings that are not so easily categorised. For instance, in generally using Scillonian Pattern 2, a more standard-like pattern (and distancing themselves from individuals who don't), it is highly unlikely that mainland-educated Scillonians like Victor are trying to be mainlanders or to reject their local identity. For one thing, the context of the interview itself focuses them upon what they consider to be the important aspects of Scillonian life. Rather than rejecting a local identity, we would argue that they are identifying with one type of historically-relevant Scillonian personaone designed by governors like Augustus Smith who strove to make Scillonians educated, aspirational and distinct from the neighbouring Cornish (see section 2). Mainland-educated Scillonians may be using features which might fail to interest the traditional dialectologist (or even the traditional sociolinguist) but their language use is deeply embedded in local history and local ideologies. It is no less so than that of the Scilly-educated speakers, who embody an alternative Scillonian persona-type.
The sociolinguistic opposition between the different Scillonian speaker groups reminds us of the recursivity of social meanings. Irvine and Gal (2000: 38) define fractal recursivity as 'the projection of an opposition, salient at some level of a relationship, onto some other level.' We are intrigued by the fact that the speakers in our analysis were born at the turn of the twentieth century as the islands went from the traditional governor system to a more democratic arrangement (see the discussion in section 2). This timing may play a role in explaining why differentiation in TRAP and BATH is so marked for this cohort of speakers. The mainland-educated Scillonians returned from their schooling and assumed the positions of responsibility that gave them access to the forms of power once more tightly controlled by the governing elite. Given the iconic status of TRAP and BATH in British English, the distinctive patterning of these lexical sets on Scilly may be supported by the shifting of authority from an external governor system to an internal hierarchical system. In assuming a governing role, the mainland-educated Scillonians may have come to emulate the practices (social and linguistic) associated with those in power. In this way, the opposition between 'governing elite' and 'islanders' may have been recursively projected onto an intragroup distinction between 'mainlandeducated Scillonian' and 'Scilly-educated Scillonian'.
This discussion highlights the importance of attending to the specific trajectory of the speech community under analysis. In this paper, we have demonstrated that the social meaning of linguistic features is inextricably linked to histories of dialect contact, local oppositions and the momentby-moment interactions that take place between speakers. We have done this by, first, examining how the Scillonian data relates to the varieties with which the islands are most closely linked, and, secondly, by considering how these patterns of variation are operationalised in the community itself. This has revealed competing local personae. As sociolinguists, we need to be alert to the ways in which social meanings may be determined by the peculiarities of a specific geographical locale. Britain (2009aBritain ( , 2009b has argued that 'place' has been under-theorised in sociolinguistics, where the tendency has been to focus on correlating language with the boundaries on a map, rather than correlating language with the ways in which a range of speakers may experience what is only superficially one 'space'. By using a range of techniques on our data and being open to the co-occurrence of multiple and completing social meanings, we hope to have demonstrated that we are better able to capture the multidimensionality of language variation. NOTES 1. We are grateful to the AHRC for the research grant which made this work possible (Scilly Voices: Language Variation and Distinctiveness on the Isles of Scilly, AH/I026243/1) and to the Isles of Scilly Museum for their assistance on this project. We also owe our thanks to Dave Britain, Jen Smith and Sam Kirkham for helpful feedback during the writing of this article, and to the editors and two anonymous reviewers who enabled us to improve the paper further with their suggestions for revision. All errors are, of course, our own. 2. Previous studies have noted the limitations of social meaning studies which consider one linguistic variable only, pointing out that social meaning is often constructed synergistically using a range of linguistic variables (for an example of this, see Sharma 2011). However, other studies have demonstrated that it is possible to trace the contributions that single variables make to social meanings (see for example, Campbell-Kibler [2008] on word-final (-ing); or Podesva, Jamsu and Callier [forthcoming], on released [t]). By focusing on TRAP and BATH, we do not wish to imply that these variables work in isolation to create social meanings in our speakers' talk. Rather, we offer our analysis as an illustration of the contribution that these specific variables make to what is, undoubtedly, a much more dynamic overall style. 3. The limitations of only considering male data mean that our analysis reflects the social meanings which typically occur in styles that are associated with men. It should not be assumed that these social meanings operate and apply in the same way in female styles. Moore and Carter (forthcoming) explore the differences between male and female speakers' use of TRAP and BATH. We show that the variants of these features are less clearly differentiated in the speech of women, and suggest that this is, in part, a consequence of the different social and cultural pressures which operate on speakers of the generation studied here. 4. As can be inferred from Table 5, children were most often sent to Truro School in Cornwall, although one of our speakers, Bill, went to St. Erbyn's School in Penzance, Cornwall. Both of these schools were academically selective, feepaying institutions which took day-students and boarders. Cheshire and Trudgill (1989: 95), amongst others, consider boarding schools to be one of the main places in which Standard English is cultivated. Consequently, we can expect our mainland-educated speakers to have experienced ideological pressure to conform to non-localisable StEE forms whilst at these mainland boarding schools. 5. One outcome of the algorithm is that sometimes predictors appear to have negative importance. Since predictors can only ever have positive importance or no importance at all, a negative result must be a consequence of the random noise in the procedure. Since this amount of (negative) deviation from zero must be noise, a similar positive deviation from zero cannot reliably be separated from noise and all predictors which have an absolute value no higher than the largest negative value can be regarded as not significant. In our analyses we count as significant only those predictors which are significant by this measure in all ten of our iterations. 6. Square brackets indicate overlapping speech, and (LAUGHS) indicates laughter. 7. The names in this table, and all subsequent names used in data extracts, are pseudonyms. 8. Note that some of the 'personae' types given in Figure 12 correspond directly with occupations (such as 'manager'). This may lead to questions about the necessity of evoking the term 'persona' when the speaker characterisation is so closely aligned to something akin to a demographic category. However, it is important to stress that personae refer to the embodiment of a set of social and linguistic practices, rather than the simple reflection of a social role or occupation. A speaker does not speak in a certain way by virtue of being a manager, but as a consequence of embodying a 'manager style'. This is seen most clearly in Charlie's case study. Charlie could be characterised as a 'manager' because of his counsellor role, but he doesn't always adopt a manager persona. This persona is only adopted when a certain style ('manager') is evoked in, and by, his discourse. Similarly, it could be argued that some of the stances in Figure 12 simply reflect some of the status/solidarity orientations found in attitude research. It is certainly the case that stances often reflect speakers' general attitudes. However, in labelling something as a stance, we highlight the transitional nature of the positioning. Attitudes suggest a more durable disposition in which there is continuity between a speaker and an alignment. The stances adopted by Victor in the case study offered here reflect how he positions himself relative to certain individuals in the context of this particular piece of discourse. It is likely that, given another topicsuch as Scillonian heritage, for instance, rather than the decorum of Captainshe may have expressed an alternative (and, possibly, more positive) stance towards the Scillonian-born Captain Atkins.