Public views of minority languages as communication or symbol: the case of Gaelic in Scotland

Two social roles for language have been distinguished by Edwards – the communicative and the symbolic. Using data from a survey of public attitudes to Gaelic in Scotland, the article investigates the extent to which people's view of language may be characterised as relating to these roles. Respondents were grouped, using statistical cluster analysis, according to their views of the communicative and symbolic roles of language. Indicators of membership of the resulting clusters were then used as explanatory variables in linear regression models to assess the relative importance of the communicative or symbolic view of Gaelic in explaining variation in attitudes to policy issues concerning Gaelic. Both sets of views of language were independently associated with attitudes to nearly all aspects of policy, but the view of Gaelic as symbol was mostly more strongly associated with attitudes to policy than the communicative view.

of Bòrd na Gàidhlig, in the words of the 2005 legislation, was to secure 'the status of the Gaelic language as an official language of Scotland commanding equal respect to the English language' (Marten 2009, 317;McLeod 2006;Oliver 2005). At the same time, the symbolic cultural role of Gaelic is much greater than the number of people able to speak it might indicate (Macdonald 1997(Macdonald , 1999MacKinnon 1990;Paterson et al. 2014;Withers 1984), and Bòrd na Gàidhlig (2007, 8) has claimed that 'the Gaelic language is a unique part of Scotland's national heritage'. Thus the communicative and symbolic views of language are present in Scottish discussions, and investigating their significance in connection with public attitudes to policy and about the future of Gaelic might cast light on how these theoretical debates are interpreted in everyday political practice.
Previous survey research (Scottish Government 2010;MacKinnon 1981), though informative about attitudes to particular aspects of Gaelic policy, has not sought to set attitudes in the context of a wider theoretical framework such as we are investigating here. Ethnographic research on Gaelic (Oliver 2005;McEwan-Fujita 2010, 2011 has found a recognition of the symbolic importance of Gaelic amongst those familiar with the languagestudents in schools where there is a Gaelic-medium stream, and adults learning the language. Oliver (2005, 3), following Chapman (1978), describes Gaelic as having been 'symbolically appropriated' as a part of identity, and reports that the young people whom he interviewedwhether or not they themselves could speak the language could recognise it as denoting a sense of belonging to a community. McEwan-Fujita (2011) noted the existence of two discourses of Gaelic language revitalisation which treat the language as a symbol of identitythe sense of 'Gaelic as ancient ', and, following MacDonald (1997), its place as 'a national language of Scotland'. These symbolic uses might be 'external' to language as a mode of communication, as Chapman (1978,131) argues when he describes symbolic appropriation as a process 'in which Gaelic culture, language and life have become the focus of sentiments and associations not intrinsic to an autonomous Gaelic life, but required by the external discourse of the English language'. Oliver (2005,(22)(23) contrasts symbolism with what he calls the instrumental role, in the sense simply of being able to speak Gaelic and using it 'pragmatically' to communicate.
Following from this theoretical, policy and research background, the paper aims to better understand the relative influence of the communicative and symbolic social roles of language in public attitudes to policy for Gaelic in Scotland.

Data and methods Data
The data were collected as part of the Scottish Social Attitudes Survey of 2012 (UK Data Service 2013). The survey aimed to provide a representative sample of adults aged 18 or over who were living in Scotland. The sampling frame was the Postcode Address File, a list of postal delivery points. The survey used a two-stage cluster sample, the clusters being 87 postcode sectors. In urban areas, they were selected with probability proportional to the number of addresses, and in rural areas the probability was proportional to twice the number of addresses. The sectors were stratified according to the Scottish Government urban-rural classification, by region and by percentage of household heads in non-manual occupations (Socio-Economic Groups 1-6 and 13, taken from the 2001 Census). Within each sector, 28 addresses were selected randomly. At addresses, interviewers randomly selected one adult. One thousand two hundred and twenty-nine interviews were conducted, a response rate of 54%. The achieved sample was weighted to match the age-and-sex structure of the population as estimated by the General Register Office for Scotland, and also to allow for the sample design. The weighted sample had a distribution of highest educational attainment that was close to that found by the much larger Scottish Household Survey of 2011. As in the 2011 Census, around 1% of the sample were Gaelic speakers.
The questionnaire in the survey covered many topics as well as attitudes to Gaelic. It was administered by face-to-face interviewing, along with a computer-based selfcompletion option for questions that were particularly sensitive. For the questions on Gaelic, respondents were offered the option of conducting the interview in Gaelic, but none chose to do so.
The variables included in the analysis reported here were of three kinds: those taken to indicate positions on the communicative and the symbolic roles of language; a selection of opinions about issues in the public debate about Gaelic in Scotland; and questions supplying demographic information.
Views about the social roles of language Views on language as communication. For the view which sees language as primarily communication, we record respondents' view of the right to use Gaelic in five domainswhen dealing with their local councils (responsible for administering local services such as education and social care), when appearing as a witness in a law court, when engaging with the health services, when speaking at a public meeting on a local issue and when writing to their bank. In connection with each social domain, respondents were asked whether 'Gaelic speakers should have the right to use Gaelic, regardless of whether they can also speak English'. The responses categories were: . Should have the right to use Gaelic wherever they live in Scotland.
. Should have the right to use Gaelic if they live somewhere Gaelic is spoken. . Should not have the right to use Gaelic in this situation.
In the analysis, these responses were scored 1, 2 and 3.
Views on language as symbol. The second framework is the symbolic role of language, in which it is related to heritage and identity. This was represented by four variables: respondents' views of the importance of Gaelic to Scottish heritage, to the heritage of the Highlands and Islands, and to their own heritage, and their views on how important the capacity to speak Gaelic is for 'being truly Scottish'. All were recorded on a five-point scale from 'very important' to 'not at all important', scored 1-5.
Respondents with missing data on any of the five communicative rights variables or on any of the four heritage and identity variables were omitted, leaving data from 1144 respondents for analysis. We describe in the next section how we used these two sets of variables to represent views about the social roles of language along the two dimensions of communication and symbolism.

Current issues in Gaelic
The aspects of language policy that are investigated here concern current issues in Gaelic policy as reflected, for example, in the topics which the Scottish Government chose to investigate in its 2010 survey (Scottish Government 2010). One set of policy issues, which related to using and learning Gaelic, was: . whether road signs or written public information should be bilingual in Gaelic and English (response categories 'across Scotland', 'where Gaelic is spoken' and nowhere); . whether parents should have a right to place their child in Gaelic-medium education (same response categories); . whether the respondent would place a child of their own in Gaelic-medium education (5-point scale from 'very likely' to 'very unlikely'); . whether Gaelic should be a compulsory subject in schools for all children aged between 5 and 16 (5-point scale from 'strongly agree' to 'strongly disagree'); . whether learning 'languages such as French is more useful than learning Gaelic' (same response categories); . whether government spending on promoting the use of Gaelic was too much, about right or too little. (Respondents were told that current annual expenditure was about £24m, or £4.60 for each person in Scotland.) The other set of policy issues related to the future of Gaelic: . whether Gaelic-medium education or Gaelic television is essential to ensuring that the language continues to be spoken (5-point scale from 'strongly agree' to 'strongly disagree'); . which of several groups has the main responsibility for whether Gaelic continues to be used in Scotland (parents who speak Gaelic, local communities, nursery schools and schools, the government, churches, the media and other); . preference for how many speakers of Gaelic there will be in 50 years' time (more, the same or fewer).

Demographic information
Information was used on respondents' sex, age, social class (UK National Statistics Socio-Economic Classification), highest educational attainment and geographical area. Geographical area was defined as five broad regions, 1 and was intended to allow for the possibility that views might be affected by whether people live in an area where Gaelic is widely used or is prominent in local policy.

Methods
The analysis was in two stagescluster analysis and linear regression. Cluster analysis was used to identify respondents with similar views on the questions about language as communication or on the questions about language as symbol. Preliminary principal components analysis suggested that the five variables recording views about the right to use Gaelic in specified domains constituted a single dimension that explained 75% of the variation; its eigenvalue (3.7) was very much greater than the next largest (0.40). 2 Similarly, for the variables about language as symbol, there was a single dominant principal component, explaining 56% of the variation, with eigenvalue 2.3 (the next largest being 0.86). The clustering technique used squared Euclidean values as the measure of distance, and the Ward method of defining the distance between clusters (Krzanowski 2000, ch. 3). 3 The analysis was carried out in the statistical computing environment R using the package hclust.
The main statistical analysis then investigated the relative importance of the two sets of clusters in explaining people's attitudes to the policy issues summarised above. This analysis was done by means of linear regression, which seeks to explain the variation in a particular variable in terms of specified explanatory factors. Here, we are seeking to explain the variation in attitudes to each of the policy questions noted above in terms of specified explanatory factors, which here are mainly the variables measuring membership of the clusters. Attitudes on the issues were reduced to sets of dichotomies (as shown in the tables). The relative importance of the two sets of clusters was assessed by 'Type II' mean squares (rather than sum of squares, to take account of the fact that there is a different number of clusters in the two classifications). For each set of clusters, the Type II mean square is defined to be the unique contribution which these clusters make to explaining the dependent variable.

Cluster analysis
Gaelic as communication In the agglomeration schedule from the cluster analysis, starting with 1144 clusters and a closest distance of 711, the next distances over which clusters were formed were 423, 176, 170, 56, 53, then decreasing gradually by small amounts. Thus the reduction of distance was 288, 247, 6, 114, 3. This suggested three clusters might be enough to summarise positions on these variables unless the somewhat larger 114 might indicate the need for four. For three clusters, Table 1 shows the size of the clusters and the distributions with respect to the variables used in constructing the clusters. A reasonable characterisation of the three clusters might then be, in ascending order of support for the right to use Gaelic: Communication cluster 1 (36%): almost no one supports Gaelic language rights across Scotland, and the support for Gaelic rights in Gaelic areas varies by social domain.
Communication cluster 2 (41%): nearly all support Gaelic language rights in Gaelic areas, and the support for Gaelic rights across Scotland varies by social domain.
Further analysis experimenting with four clusters resulted merely in a split of cluster 2. In one sub-cluster, everyone favoured the right to communicate in Gaelic solely in Gaelic areas for each social domain. In the other sub-cluster, some favoured such rights everywhere in Scotland. Since that description could also be said to apply to cluster 2 in Table 1 it was decided, for reasons of parsimony, to retain only the three communication clusters shown in that table.

Gaelic as symbol
The agglometration schedule started with 1144 clusters and a closest distance of 731; the next distances over which clusters were formed were 515, 249, 169, 120, 112 and 98, and thereafter in steps of about 20 or less. Thus the reduction of distance was 216, 266, 80, 49, 7 and 14. The small reduction of distance after 49 suggests five clusters. Table 2 shows their size, and their distribution with respect to the variables used in constructing the clusters. A summary description of the views of people in the symbolism clusters might be, again in ascending order of feeling that Gaelic is important: Symbolism cluster 1 (14%): few people see Gaelic as important to heritage or identity (whether of Scotland, of the Highlands and Islands or of themselves). Symbolism cluster 2 (13%): Gaelic is perceived to be important to the heritage of the Highlands and Islands, but not otherwise important. Symbolism cluster 3 (43%): Gaelic is perceived to be important to the heritage of Scotland, and of the Highlands and Islands, but is not important to the respondent personally, and speaking Gaelic is not perceived to be an important attribute of being Scottish. Symbolism cluster 4 (17%): Gaelic is important to the respondent's own heritage, and to the heritage of the Highlands and Islands and of Scotland, but speaking Gaelic is not perceived to be important to being Scottish. Symbolism cluster 5 (13%): Gaelic is important to the heritage of the Highlands and Islands and of Scotland, and speaking it is important to being Scottish. A majority (63%) of this group regard Gaelic as important to their own heritage.
Comparing the two classificationsof language as communication and language as cultural symbolallows us to assess how the theoretical perspectives outlined in the first section relate to each other empirically. There is overlap between the two classifications (Table 3). For example, the symbolism clusters 3, 4 and 5, in which people regard Gaelic as important to the whole of Scotland, also show greater proportions favouring the right to communicate in Gaelic throughout Scotland (communication cluster 3) than do the other symbolism clusters. Nevertheless, there is also much discrepancy between the two sets of clusters, suggesting that views of the communicative and symbolic role of Gaelic cannot be inferred from each other. For example, in the strongest symbolism cluster 5, no more than about one-third of people (36%) are in the strongest communication cluster 3, which shows that it is possible to regard Gaelic as a symbolically strong part of Scottish identity without supporting the right for Gaelic speakers to communicate in Gaelic throughout Scotland. The lower part of the table, showing the percentages by row, suggests even more clearly that the correspondence between the communicative and symbolic classifications is weak. The modal category on symbolism is consistently cluster 3. There is a weak tendency for people who are sceptical of Gaelic as communication (in communication cluster 1) to be more likely to be in symbolism clusters 1 or 2 than people in the other communication clusters, and thus to be less willing to regard Gaelic as being symbolically important outwith the Highlands and Islands. The opposite is true of those who favour the right to communicate in Gaelic everywhere (communication cluster 3), but even they have a very clear majority not in the most strongly pro-Gaelic symbolism cluster 5. Thus the empirical evidence reflects the distinction between the symbolic and communicative dimensions of language outlined by Edwards (2009). Although they overlap, a respondent's position on one dimension cannot be reliably predicted from the position on the other.

Demographic characteristics of the clusters
Membership of the two sets of clusters did not vary much by sex, age, social class, highest educational attainment and region. For the communication clusters, the strongest variation was by age and region. Young people (aged under 45) were more likely than older people to be in cluster 3 which is most favourable to the right to communicate in Gaelic. Residents in the Gaelic areas were more likely than residents in other areas to be in the clusters least favourable to the right to communicate in Gaelic, perhaps reflecting through local experience an awareness of the difficulty of implementing rights to communicate in a lesser used language within public services. There was even less association of the symbolism clusters with any demographic factors, the only clear pattern being that there was a weak tendency for people in lower social classes or with low educational attainment to be less likely to be in symbolism cluster 3 (where Gaelic is believed to be important to the heritage of the Highlands and Islands and of Scotland), and a corresponding greater tendency to be in the symbolism clusters 1 and 2 that are least favourable to Gaelic. Insofar as these views about the role of Gaelic were not strongly differentiated demographically, they may be said to be characteristics of Scottish political culture as a whole.

Gaelic policies
We can now investigate whether there is a relationship between people's views of language as communication or symbol and their attitude to policies to support Gaelic. The left-hand part of Table 4 shows that the communication clusters clearly distinguish among different levels of support for increasing the prominence of Gaelic in Scotland. For example, people in communication cluster 3 (which favours language rights for Gaelic speakers across Scotland) tend to favour prominence for Gaelic across Scotlandon signs (83%), in public information (82%) and in Gaelic-medium education (82%). A majority of this cluster (58%) also believes that learning Gaelic should be a compulsory element of schooling between ages 5 and 16. By contrast, people in communication cluster 2 (which favours language rights for Gaelic speakers where Gaelic is spoken) tend to favour these policies in Gaelic areas rather than everywhere. For example, 96% of communication cluster 2 are in favour of bilingual public information in Gaelic areas, but only 47% are in favour of that throughout Scotland. People in communication clusters 2 and 3 are less likely to think that the government spends too much on Gaelic than people in communication cluster 1. Nevertheless, despite these patterns, we must also note that an interpretation based on language as communication cannot be the whole explanation since there are minorities in each communication cluster who have a different opinion from the majority of their cluster: for example, 17% of people even in cluster 3 do not favour signs throughout Scotland, and 53% in cluster 2 do favour such signs.
Part of the explanation of the spread of attitudes within the communication clusters is found when we turn to look at the symbolism clusters (right-hand part of Table 4). There is again broad consistency between the characterisation of the cluster and attitudes to policies. Thus people in symbolism cluster 2 (which sees Gaelic as being important to the heritage of the Highlands and Islands, but not in other respects) were much more likely to favour bilingual signs or information in the areas in which Gaelic is spoken than across the whole of Scotland, a difference that diminishes as acceptance of a symbolic role for Gaelic increases through clusters 3, 4 and 5. Yet, as with the communication clusters, there are minorities who divergesuch as the 15-30% in symbol cluster 2 who favour bilingual signs, bilingual public information or access to Gaelic-medium education throughout Scotland. Table 5 shows, similarly, attitudes to the future of Gaelic. Again, there is a broad tendency for attitudes to vary among the clusters in ways that are consistent with the meaning of the clusters. In the left-hand part of Table 5, for example, around 80% in the most pro-Gaelic communication cluster 3 believe that Gaelic-medium education and Gaelic broadcasting are important for the future of the language, whereas only around 60% in the least pro-Gaelic communication cluster 1 take this position. Attributing responsibility for the future of Gaelic to parents or communities declines from communication cluster 1 to communications cluster 3, whereas the gradient is in the opposite direction for attributing responsibility to education or the government: this suggests that holding strong views about the right to communicate (cluster 3) is associated with a belief that public-policy action, as opposed to individual or voluntary local action, is required to ensure that right. In the right-hand part of Table 5, the symbolism cluster 5 that sees the widest symbolic importance for Gaelic is more strongly politicised than the other symbolism clusters, in the sense that it has the highest proportion of people who believe in the importance of education and broadcasting (over 90%), and the highest proportions who believe in public action (schools and government).
Which of these two ways of summarising people's views about Gaelicas communication or as symbolbetter explains their attitudes to specific issues in policy for Gaelic? We assess this question by means of linear regression, in which we model attitudes on the questions in the section above in terms of membership of the clusters. Table 6 shows the explanatory power of the two sets of clusters, using the type II mean squares corresponding to each set of clusters (as explained in the Data and Methods section). Because the interest here is to compare the overall explanatory power of each of the two sets of clusters (communication or symbolism), the regression coefficients are not shown: they followed the pattern shown by the percentages in earlier tables. The relative sizes of the mean squares shown in Table 6 were similar when statistical controls for sex, age, social class, educational attainment and region were included, and so are not shown here.
On most policy issues, both sets of clusters are independently associated with respondents' attitudes: there are very few non-significant results in Table 6. Thus neither type of view of Gaelic exhaustively explains people's attitude to policy. Moreover, neither set of clusters is uniformally more powerful than the other in explaining attitudes to policy issues. Nevertheless, on most topics the symbolism clusters matter more than the communication clusters. For example, in explaining variation in the likelihood of placing a child in Gaelic-medium education, the mean square in Table 6 associated with the symbolism clusters (8.1) is more than three times greater than the mean square associated with the communication clusters (2.5). A similar ratio (6.7 to 2.2) was found after the demographic controls were added. The only clear exceptions to the tendency for the symbolism clusters to be more strongly associated with people's views of Gaelic policy are on rights of access to Gaelicmedium education, and on the use of Gaelic in signs and in public information: for example, in explaining variation in the opinion that there should be a right of access to Gaelic-medium education throughout Scotland, the mean square associated with the symbolism clusters in Table 6 is 3.4, whereas the mean square for the communication clusters is more than four times greater, being 16.5; again, a similar ratio (2.8 to 11.3) was seen with the controls. These exceptions might tend to be interpreted as concerning opportunities to communicate through the medium of Gaelic when receiving public services, and the other issues might tend to be interpreted as being about safeguarding the future of Gaelic, or about passing it on through the education system.

Conclusions
This analysis has provided statistical evidence relating to public views of the Gaelic language in Scotland. It has shown that people may be coherently grouped according to their views about the right for Gaelic speakers to communicate using Gaelic and, Note: All results are statistically significant at the 0.01 level except those indicated by WS (weakly significantat 0.05 level) or NS (not significant at 0.10 level). GME, Gaelic-medium education.
separately, according to their views about the place of Gaelic as a symbol of personal, regional or national heritage and identity. Both sets of views about the role of language are associated with attitudes to specific policy or other issues affecting Gaelic, but views about symbolism generally explain more of the variation in attitudes on these issues than do views about communicative rights, except where the policy issue itself might be interpreted as being directly about Gaelic speakers' communicating through the medium of Gaelic when receiving public services. Such evidence shows that specific issues may be linked in the public mind to the communicative or symbolic social roles of languages. The advantage of developing the analysis in terms of statistical clusters is that they each comprise a group of people who share particular combinations of views about the social role of Gaelic. Thus, for instance, we know that the symbolism cluster 3 is internally coherent, in the sense that the respondents in the cluster believed that Gaelic is important to the cultural heritage of the Highlands and Islands, and of Scotland, but did not believe Gaelic to be important to their own cultural heritage, or to being Scottish. Because these clusters are fairly large, we have reasonably firm grounding in evidence for the coherence of the views that we have detected. The clustering approach allows us to ask whether people who share a set of views about the social role of Gaelic tend also to share attitudes on specific issues. A different statistical approach would have been to have regressed opinions about policy on the separate variables that were used to detect the clusters. That approach, though informative about the association with these individual variables, would not have been able to say anything about the effects of holding a distinct combination of the features measured by the clusters.
We can then infer in conclusion from our evidence that Edwards's distinction between the 'communicative' and the 'symbolic' roles of language is empirically valuable. From the cluster analysis we have seen that public views of Gaelic may reasonably be interpreted in these termsas the right for Gaelic speakers to communicate in Gaelic and as the position of Gaelic in cultural heritage and identity. The symbolic or cultural role is recognised by citizens generally, a conclusion which goes beyond those of Oliver (2005) and McEwan-Fujita (2010), who reported a recognition of the symbolic role of Gaelic among people in close proximity to Gaelic. Moreover, the aspect of the communicative role which we have examined here goes beyond the merely instrumental to become a matter of recognising the rights of Gaelic speakers to use Gaelic even in circumstances where communication would not require the use of Gaelic. Thus a dichotomous contrast between symbolic and instrumental views of a language may have to be refined in the light of the role which individual rights might play. Consistent with the conclusions reached by the writers on Irish and Welsh which we noted earlier in the paper, the symbolism clusters were stronger than the communication clusters as explanations of variation in attitudes to most areas of policy. Nevertheless, both sets of clusters showed an independent, statistically significant association with nearly all the dependent variables, even after controlling for demographic factors. Thus the case of Scottish Gaelic adds to the general cogency of the distinction between communication and symbolism, and we would tentatively suggest that the statistical methods which we have adopted here may be a useful way of exploring these ideas in other specific contexts.

Funding
The research was funded by the UK Economic and Social Research Council [grant number ES/J003352/1], the Scottish Government, Bòrd na Gàidhlig and Soillse.
2. Principal component analysis assesses whether several variables are so closely related to each other that they might be plausibly combined into a single variable. This is assessed by whether there is a dominant eigenvalue, which is a measure of how much of all the variation in the set of variables may be captured by a single summary variable. When the largest eigenvalue is much larger than the next eigenvalue, it is reasonable to combine the variables (Krzanowski 2000, ch. 2). 3. Cluster analysis forms larger and larger clusters by grouping together cases that are similar to each other in terms of their responses to the variables that are included (here, the views about rights or the views about symbolism). The similarity between two cases was measured here as the sum of the squares of the differences between corresponding values on these variables, a measure known as 'squared Euclidean distance'. The agglomeration schedule records this process of successive grouping together of cases. It is recommended in the literature to stop merging clusters in this way when the distance over which the merger would take place becomes small (Krzanowski 2000, ch. 3).