Accent change in Glaswegian: A sociophonetic investigation
(a) The grant
Dates: January 1999 - April 2000; 1 year with 4 month extension.
Personnel: Dr Jane Stuart-Smith, principal applicant (1 day/wk throughout, including during maternity leave), Dr Fiona Tweedie, co-applicant (1 day/wk 7 months), Claire Timmins, research assistant (full-time); collaborators: Dr A. Wrench, Dr J. Scobbie, Queen Margaret University College, Edinburgh.
Site: Departments of English Language, and Statistics, Glasgow University
- To provide a quantitative and qualitative account of accent change with respect to consonant pronunciation in Glaswegian.
- By considering data from a Scottish context, to extend current theoretical models of the mechanisms of accent change, and thereby, of language change in general.
- To extend the use of acoustic analysis as a research tool in sociolinguistic research, with particular focus on consonantal variables.
- To contribute to our theoretical understanding of the phonetic processes of phonological change.
- To extend the statistical techniques available for the analysis of accent shift data.
(c) Research activity
- An existing speech dataset was used (Table 1, Figure 1; Stuart-Smith 1999).
- 11 consonantal variables were investigated:
(th), e.g. tooth; (dh), e.g. brother; (l), e.g. milk; (t), e.g. butter; (x), e.g. loch; (hw), e.g. whine; (r), e.g. car; (s), e.g. icy; (k), e.g. lock; (w), e.g. wine; (r2), e.g. red (Annexe 4).
- Auditory and instrumental acoustic analysis used existing phonetic techniques, but following Docherty and Foulkes (1999) extended these for investigating sociolinguistic variation in consonants.
- Individual analysis of each variable used exploratory and hypothesis-testing techniques: cluster analysis; log-ratio linear analysis.
- Combined analysis of all variables used t-tests and multivariate exploratory techniques: cluster analysis; principal components analysis.
The auditory transcription of some variables was more difficult than anticipated. Similarly, the statistical analysis was more complex than originally thought.
- preparation of data
- auditory transcription
- instrumental acoustic analysis (with A. Wrench and J. Scobbie)
- statistical analysis
- interpretation/writing up
Changes in project
(d) Conclusions and achievements
All objectives were met.
1. Our comprehensive and detailed quantitative and qualitative account of Glaswegian consonants reveals change in 7 out of 8 variables (Stuart-Smith, Timmins, Tweedie forthcoming).
2. Early sociolinguistic work (e.g. Labov 1972, Trudgill 1972) showed that notions of 'prestige' (overt/covert) and 'identity' are useful in explaining accent change. Current models which particularly consider diffusion and levelling processes (e.g. Kerswill and Williams 1999; Watt and Milroy 1999) emphasize 'social networks' and 'mobility' (geographical and social). It is assumed that weak network ties with mobility will promote change; close network ties with lack of mobility will promote conformity and inhibit change (L.Milroy 1980; J.Milroy 1992).
Overall, the Glaswegian data show that while the least mobile adolescents (working-class) show high linguistic conformity, they lead all changes; the more mobile adolescents (middle-class) show few changes (Figures 2-5, 9-11; Tables 2-4).
Thus it seems that:
- mobility with weak ties, in conjunction with identity, can inhibit change:
- Regular contact of middle-class adolescents with speakers from England may help maintain Scottish features to promote Scottish identity (Glauser 1994).
- lack of mobility with strong ties, in conjunction with covert prestige, can promote change: Working-class adolescents consistently distance themselves from the regional standard (middle-class), by using both local and non-local non-standard features. Their lack of social and geographical mobility promotes change to maintain the covert prestige of the Glaswegian vernacular, albeit in a changed form (Cheshire et al 1999; Foulkes and Docherty 2000).
- network history is crucial: the present close network structure masks a recent history of network fragmentation, largely due to the impact of rehousing policies which peaked during the 1950s and 1960s (Rae 1974; Horsey 1990; Reed 1999). This may account for the less consistent patterns of working-class adults and their reduction of traditional 'Scottish' features (Macafee 1983 after L.Milroy 1980). Thus repeated network fragmentation and imposed mobility may have provoked changes, which are now compounded by the introduction of innovations and accelerated by the maintenance of covert prestige within the reformed close network structure.
- Our findings from the Scottish context confirm 'mobility' and 'networks', together with 'prestige' and 'identity', as essential concepts in explaining accent change, but show that the way in which they interact to induce or inhibit change depends crucially on the context.
3. Qualitative and quantitative instrumental acoustic analysis on a range of the consonant variables
- confirmed our detailed auditory analysis, but sometimes in a complex manner: e.g. the auditory transcription of (s) identified a small but significant degree of variation correlating with gender (Figure 10b). Acoustic analysis confirmed gender patterning but across all variation (Figure 6)
- revealed fine-grained acoustic variation in correlation with social factors (Lawson and Stuart-Smith 1999; Figure 6). Thus we confirm that the claims made for (t) by Foulkes and Docherty (1999) may be extended to consonants in general.
4. Our research allows a contribution to theoretical understanding of the phonetic processes of phonological change in three respects:
- spread of variants across phonetic environments In Glasgow the introduction of innovations depends on the existing pattern of non-standard variation, and typically exploits gaps: e.g. [v] for /dh/ is found predominantly in word-final position, where the local non-standard [R] is unlikely to occur.
- complexity of consonant change (Socio)linguistic accounts of consonant change tend to operate with only a few variants (Docherty et al 1997). Our auditory and acoustic data confirm that in general a wide range of phonetic variation is involved in consonant change (Figure 7).
- phonetically intermediate variants We identified variants that were auditorily and acoustically 'intermediate' (e.g. for (x), Figure 8). We confirm that phonetically intermediate variants can occur for consonants as well as vowels (e.g. Trudgill 1986).
5. Several statistical techniques have been applied to this data. The constraints on the percentage figures have been addressed by the use of compositional data analysis (Aitchison 1986; Tweedie and Frischer 1999). Exploratory analysis was carried out using techniques such as cluster analysis (Aitchison 1983), applied for the first time to accent shift data, which allowed the natural groupings within the data to be investigated. Formal analysis was carried out with compositional data analysis and the application of log-ratio linear modelling. In addition, log-contrast principal components analysis (Aitchison 1983, 1986) was used to analyse the data as a whole.
The key findings are:
- Many features of consonant pronunciation in Glaswegian are changing, most noticeably in working-class adolescents
- Diffusion of non-local, non-standard [f] [v] for (th) (dh) is found alongside increased usage of local non-standard variants. L-vocalization is also found (Figure 9).
- T-glottalling, a long-standing non-standard feature of Glaswegian, is vigorous (Figure 10a; Stuart-Smith 1999a).
- Traditional Scottish (x) (hw) and postvocalic (r) are undergoing complex changes, probably loss (Figure 11).
- Working-class adolescents use both local and non-local non-standard variants to distance themselves from the regional standard of middle-class adults.
- Working-class adolescents are highly consistent in their speech patterns.
- Gender does not often influence patterning, and only consistently that of S-retraction (Figures 6, 10b).
- Middle-class adults tend to use traditional Scottish (hw) and postvocalic (r) more than their working-class counterparts.
- Consonants generally show very wide variation.
- Phonetically intermediate variants for consonants are attested.
- Instrumental acoustic analysis reveals that speakers produce very fine differences in consonant pronunciation which distinguish them from others socially.
- The use of compositional data analysis is a clear alternative to the VARBRUL method (e.g. Labov 1966, Cedergren and Sankoff 1974) which is not able to deal with the large number of variants that were found in this data. The complex structure of variation between social groupings would also have been very difficult to analyse within the VARBRUL system.
- Our results are contemporary with studies on urban accents across the British Isles (e.g. Foulkes and Docherty 1999). Glasgow is participating in the diffusion of consonant features ('youth norms', Williams and Kerswill 1999), and also shows the loss of some traditional consonant features. While the use of local non-standard variants may overlap with standard English variants (e.g. [k w V] for (x hw r)), there is no evidence that these are due to levelling processes, i.e. the reduction of regionally-marked features during dialect contact (Annexe 5; Dyer 2000). Diffusion of non-local features is not compromising local non-standard variation.
- The extensive variation found in consonant production supports the reconstruction of complex consonant change in historical phonology (e.g. Stuart-Smith 1996).
- The complex sociophonetic variation identified is problematic for theoretical phonology as currently conceived (Docherty and Foulkes forthcoming). This supports increasing calls to revise the fundamental conception of lexical representation within phonology (Coleman forthcoming; Scobbie et al 2000).
- With the exception of VARBRUL, there is relatively little statistical analysis carried out in sociolinguistic studies. This study demonstrates the application of a variety of novel statistical techniques that can be used for sociophonetic data. The use of multivariate techniques to consider the full pattern of variability opens up a new aspect of sociolinguistic research.
- This is the first comprehensive quantitative sociolinguistic work on Scotland's largest city for over 25 years.
- We demonstrate that consideration of all variation is essential for modelling accent change.
- Our combination of linguistics and statistics allowed us to explore successfully alternative methodologies.
- The widespread media interest in our findings shows the relevance of our work to the general public.
- We commend our researcher, Claire Timmins, who provided a consistently high standard of work.
- All forms of analysis undertaken proved to be complex, probably more than anticipated.
- The 'linguistic' side of the project called for two different analyses of the same data - auditory and acoustic. We gained useful information from combining the two. It would have been more satisfactory if both types of analysis had been used for all consonant variables. We recommend that future work considers fewer variables but in both manners.
- Where the amount of data was small, less demanding statistical methods could be applied. Further data collection with these particular variables in mind is recommended.
- This research was truly 'interdisciplinary' in that the analysis of the data was carried out across the rigours of two disciplines, (socio)linguistics and statistics. This interaction was sometimes challenging, but ultimately rewarding, as our results demonstrate.
(e) Publications and dissemination
Papers from the project have been/will be presented at the following conferences:
- 7th Phonology Workshop, Manchester University, 13-15 May, 1999
- British Association of Academic Phoneticians' Colloquium, Glasgow University, 3-6 April, 2000
- Sociolinguistics Symposium 2000, UWE, Bristol, 27-29 April, 2000
- 7th Conference on Laboratory Phonology, MPI, Nijmegen, 29 June-1 July, 2000
- Variation Is EveryWhere 2000, Essex University, 14-16 September, 2000
- International Quantitative Linguistics Conference, Prague, 24-26 August 2000.
Forthcoming report: Accent change in Glaswegian: A sociophonetic investigation, full report on Leverhulme Trust grant no. F/179/AX
- 'Accent change in Glaswegian', Language Variation and Change
- 'Sex and gender differences in /s/ in Glaswegian', Journal of the Acoustical Society of America
- 'Should phonologists trust their data? An empirical phonological study of Glaswegian', Phonology
- 'A sociophonetic investigation of the 'Scottish' consonants (/x/ and /hw/) in Glaswegian', Journal of Sociolinguistics
- 'Speech patterns in Glasgow', Journal of the Royal Statistical Society Series A: Statistics in Society
- Letting the data speak for themselves: The Glasgow speech project (collaborative book)
The findings will be presented to participating and interested schools during the autumn of 2000, and will be incorporated into teacher-training materials used by the principal applicant.
The project and its results have generated considerable interest in the local and national media (Annexe 6). A press summary has been released by Glasgow University.
(f) Future research plans in this field
- external factors in accent change: does the media play a role?
The diffusion of typically southern English features in the speech of the least mobile informants poses an apparent problem for models of change based on accommodation during face-to-face interaction. Recent reference to these features as 'youth norms' raises the possibility of the media as a catalyst in their rapid spread. However, to date there is no evidence to support or reject the role of the media in accent change. Our next research project will tackle this question.
- statistical methods for quantitative sociolinguistics
Here we have looked at exploratory techniques for compositional data analysis as well as log-ratio linear analysis. Other techniques are available and should be applied to this type of data. In addition, methodological issues should be addressed by comparing our results with those that would be obtained from using the VARBRUL logistic regression method, as well as MCMC methods for multinomial probit models.