Skip to navigation – Site map

HomeAnglophonia30Intra-speaker phonetic micro-vari...

Intra-speaker phonetic micro-variation, and its relationship to phonetic and phonological change

Florent Chevalier


This paper looks at intra-speaker phonetic micro-variation patterns in four dyads, recorded in the 1980s in the working-class community of Glasgow. We present the importance of this kind of interaction-based variation in sociolinguistic studies, since short-term speech accommodation is considered to be one of the mechanisms for community-level sound change over time. We consider two examples of sound change which have taken place in vernacular Glaswegian English, one phonetic (vowel quality) and one phonological (vowel quantity). We predict that intra-speaker variation in quality and timing alternations for /i/ and /ʉ/ during conversations will reflect the trajectory of real-time sound change, and that this variation will relate to convergence towards the speakers closest to the future norm. Our results do not validate these hypotheses; however, they highlight the relevance of unresolved issues about the very nature of speech accommodation and the role played by social factors such as speaker age.

Top of page

Author's notes

We would like to thank the Glasgow University Laboratory of Phonetics for granting access to the Sounds of the City corpus (Leverhulme Trust-funded project RPG-142), as well as the TGIR Huma-Num consortium for providing RStudio Server facilities.

Full text

1. Background

1.1 Intra-speaker variability in sociolinguistics

1 Over the last decades, the variationist sociolinguistics framework has become one of the most popular theoretical approaches in studying and accounting for language change. Since the works of Weinreich and his successor Labov, this framework has laid the foundations of modern studies of sound change; it has also reconciled variation in a language variety with the social structure of the community of practise, asserting strongly the unbreakable bond between a language and its speakers.

2 Linguistic variation is at the heart of the sociolinguistic way of thought, since it relies on opposing linguistic behaviours depending on the speakers’ social characteristics. This approach has allowed Labov to identify patterns of language change: for instance, contrasting language use by age of speaker allows us to recreate a diachronic perspective and to visualise sound change in progress through the apparent time paradigm. In his Principles of Linguistic Change (1994, 2001), Labov also presents the role of gender or urban hierarchy in the adoption of a new phonetic feature.

3 Although variation is the cornerstone of the sociolinguistic approach, variability has mostly been kept out of this theoretical and methodological framework (Cukor-Avila and Bailey 2013). By variability, we mean the tiny fluctuations in the phonetic productions of a specific sound by a given speaker, including within one utterance or interaction. This kind of variability is well established: we know we pronounce things differently, sometimes due to context, sometimes not. However, this micro-variation has mostly been ignored and treated as a nuisance (Foulkes and Docherty 1999).

4 This explains the methods commonly used by researchers in the field: quantitative methods are indeed an asset to deal with large samples (including high counts of occurrences by the same speaker), but they inherently present important degrees of variability for each phonetic feature. As a consequence, sociolinguists usually try to control and cancel this intra-speaker variability, and the use of aggregates is common practice in order to draw conclusions from one virtual average phonetic representation per speaker. This is in a way going back to Saussure’s vision that by nature, synchronic data is incompatible with variation, and that no observable variation can ever be significant (1916) – hence the need for this synchronically observable variation to be cancelled out in order to follow the apparent-time paradigm, i.e. to see an artificial construction of change in progress. Albeit crucial in the sociolinguistic approach, this paradigm remains artificial in the sense that instead of live-tracking variation across real time, it merely consists in comparing speakers of different ages, and therefore relies on the incorrect assumption that these speakers’ idiolect have been stable since their early years.

1.2 Variability of variation?

5 This tendency to avoid dealing with intra-speaker variability in sociolinguistics is surprising. After all, it is possible that this short-term micro-variation is linked to long-term community-level sound change. Decades of sociolinguistics studies have demonstrated the mechanism of diffusion of a given sound change. This means that once an ongoing change is spotted in the community, we can track its spread in that community, and predict its adoption by the different social groups which compose the community. However, the question of the very origin of sound change remains unanswered, and the mechanism responsible for phonetic innovation keeps on being debated. One of the theories which has been positively received is that of Trudgill (1986), who suggested that inter-speaker linguistic accommodation could be the driving force behind the first step of sound change.

6 In Trudgill’s words, accommodation refers to the Communication Accommodation Theory (formerly Speech Accommodation Theory) formulated by Giles (see Giles et al. 1991). Giles’s accommodation theory predicts that people tend to adapt to each other when communicating. It is expected that people will either converge towards each other (i.e. become more similar to one another), or diverge from each other (i.e. become more dissimilar). Although other patterns of accommodation strategies, such as complementarity or speech maintenance, have been progressively included in the theory, most studies up to date have considered convergence to be the most common behaviour. The motives behind this specific kind of inter-speaker adaptation and variation continue to be investigated (Toma 2014).

7 If accommodation is the driving force behind innovation, then the actuation of sound change lies in this short-term micro-variation, as speakers undergo tiny phonetic adjustments while they accommodate to each other. Given that part of the typical sociolinguistic methodological paradigm, described in 1.1, consists in cancelling out individual variation, then phonetic innovations can only be noted once they are adopted by enough members of the community of study. These methods are thus doomed to ignore any kind of micro-diachronic variation which may derive from speech accommodation. It remains possible that the very mechanism of long-term sound change might be short-term speaker-specific variability, and this variability would in turn represent sociolinguistically meaningful variation.

1.3 Research questions

8If it is the case that those two phenomena of variation are related, then we may expect short-term variation to reflect sound change – and vice-versa. In other words, if and when two speakers converge in their pronunciation while interacting with each other, the direction of the variation within minutes may be expected to be similar to the direction of a community-level sound change. This is easy enough to apprehend when discussing phonetic change over time: the articulatory adjustments (e.g. vowel quality, fricatives’ centre of gravity) resulting from speech accommodation within minutes would correspond to the direction of that given phonetic change over decades. However, trying to observe a change in progress in such a narrow window of time means that phonological changes such as quantity opposition patterns would have to be observed through the prism of phonetic variation as well. In this paper, we will consider the articulatory characteristics of vowel realisations over the course of four 40-minute long conversations, and observe how individual phonetic micro-variation in those recordings relates to two sound changes in progress within the community (namely the working-class community of Glasgow):

9- Changes in vowel height and fronting (vowel quality), which constitute phonetic changes
- Changes in vowel length alternation pattern, known as the Scottish Vowel Length Rule (vowel quantity), which constitute a partly phonological change (Scobbie et al. 1999).

This will be an opportunity to see whether short-term variation can be studied in the same way for phonetic and phonological variables.

2. Method

2.1 Corpus

10As mentioned above, we based our study on vernacular Glaswegian English, i.e. the variety spoken by the working-class communities in Glasgow. We used the Sounds of the City corpus, a private collection of 142 recordings of Glaswegian English from the 1970s onwards (Stuart-Smith et al. 2015b). Table 1 shows the real and apparent-time structure of the corpus, with three generations of speakers for each decade of recording, offering an apparent-time perspective of a century, with the oldest speakers born in the 1890s and the youngest in the 1990s. This corpus totals approximately 60 hours of spontaneous speech produced in diverse contexts (peer to peer conversations, oral history and sociolinguistic interviews, etc.). All recordings have been forced-aligned using HTK and are now accessible for academic researchers through a LaBB-CAT interface, which allows for orthographic and phonemic searches within the recordings (Fromont and Hay 2012). Sounds of the City has now been integrated within the SPADE project (SPeech Across Dialects of English – see Sonderegger et al. in press).

11Table 1: Structure of the Sounds of the City corpus

Speaker age


67-90 years old

(Decade of Birth)


40-55 years old

(Decade of Birth)


10-17 years old

(Decade of Birth)

Decade of recording


70-O (1890s)

70-M (1920s)

70-Y (1960s)


80-O (1900s)

80-M (1930s)

80-Y (1970s)


90-O (1910s)

90-M (1940s)

90-Y (1980s)


00-O (1920s)

00-M (1950s)

00-Y (1990s)

12 Prior studies using this corpus have already contributed to the description of Glaswegian English and to its variation over time. Most sounds have been found to be changing, in their timing characteristics (e.g. Stuart-Smith et al. 2015a’s study on Voice Onset Time, or Rathcke et al. 2016 study on the Scottish Vowel Length Rule) as well as in their quality: loss of rhoticity (Lawson et al. 2014), vowel fronting and lowering (José and Stuart-Smith 2014). It appears those sounds have been changing gradually over time; this has been regarded as dialect-internal changes in Glaswegian English, given the very low influence of other varieties of English in Glasgow (Chevalier 2019).

13 In the present study, we have chosen to focus on two characteristics: vowel quality and vowel quantity. Vowel quality is a phonetic variable which includes vowel height (F1, measured in Hz) and vowel fronting (F2, measured in Hz). José and Stuart-Smith (2014) found that all Glaswegian vowels have been moving in the vowel space over time; three vowels in particular have undergone systemic changes in their articulatory characteristics; with the BOOT vowel lowering, GOAT rising, and LOT rising and moving to the back of the mouth.

14 Vowel quantity hereby refers to the patterns of vowel length alternation specific to the varieties of English spoken in Scotland, and to some extent in Ulster and in the northern English town of Berwick; this quantity opposition patterns is known as the Scottish Vowel Length Rule (SVLR), or Aitken’s Law. It differs from the pattern in use in all other varieties of English, commonly called voicing effect or low-level lengthening, which contrasts short and long vowels depending on the voicing of the next segment: a vowel followed by a voiceless consonant will be short and a vowel followed by a voiced segment will be long (House and Fairbanks 1953). In Scottish English, some vowels are long only before /r/, voiced fricatives and morpheme boundaries, and short in all other morphophonemic contexts (Aitken 1981). Aitken also notes that different dialects have different sets of vowels impacted by the SVLR, comprised of some or all of the following vowels: /i ʉ e o a ɔ aɪ/. This quantity opposition pattern is described as quasi-phonemic in the sense that the timing pattern induces minimal pairs such as brood ~ brewed or need ~ kneed (Wells 1982). Previous work by Rathcke and Stuart-Smith (2015) and Chevalier (2019) has documented the ongoing weakening of the SVLR in Glaswegian English. Both studies compared the duration of the /i/ and /ʉ/ vowels in recordings from the 1970s and the 2000s, and found that although the opposition between short and long vowels still remains, vowels in a lengthening context have become much shorter over time.

2.2 Segment selection and labelling

15 Previous studies have demonstrated how those variables have changed, and they have shown what the common realisation of vowels was like in the 1970s and then in the 2000s. If we assume that what happens within minutes reflects what happens over time, then we can predict the direction of short-term intra-speaker variation in between the start and end point of the maximal diachronic window provided by the corpus – namely between the 1970s and the 2000s in real time, or between speakers born in the 1890s and 1990s. The present study focuses on an intermediate point: we look at four different recordings from the 1980s from a series of oral history interviews. Each recording lasts about 40 minutes and consists in a one-to-one conversation, in which the interviewee belongs to the oldest age group (80-O, born in the 1900s) and the interviewer to the youngest (80-Y, born in the 1970s). Our sample comprises two young speakers, one female, one male; for each interviewer, we have selected two recordings, one for each sex of interviewee.

16 For these speakers in the four recordings, all lexically stressed instances of /i u/ were extracted through a LaBB-CAT search. Those vowels were selected because /i u/ have been shown to be changing both in quality and quantity: indeed, as mentioned above, the BOOT vowel (both FOOT and GOOSE from Wells’ lexical sets in Scottish English) is lowering in Glaswegian English, while the FLEECE vowel has become more fronted (Stuart-Smith et al. 2017). /i/ and /ʉ/ are also the two monophthongs for which the SVLR is active in Glasgow, and the weakening of SVLR-induced lengthening has been demonstrated for both vowels (Rathcke and Stuart-Smith 2015, Chevalier 2019). Vowel length was obtained automatically through the HTK time alignment; F1 and F2 values consist of the average of measurements taken at 25, 50 and 75% of each token through a Praat script (Boersma and Weenink 2001).

17 The initial search results (n = 4709) were then pruned in the following manner:
• All segments shorter than 50ms were excluded from the sample, given their likelihood to be either reduced or poorly time-aligned; this is consistent with previous work on the same corpus (Tanner et al. 2020);
• All words in the dataset were then checked to withdraw all instances of stumbling, onset of unfinished words, interjections, and errors in the phonemic transcription;
• For all the remaining tokens, outliers for F1 and F2 were withdrawn after in 1.5IRQ test per vowel and speaker, given Praat was likely to have selected the wrong formant track for these tokens without the selected range (Solanki 2017).

18 The final dataset comprises 1905 tokens, whose distribution per vowel and speaker is shown in Table 2. It must be noted that given the nature of the interaction, there are substantially more tokens for the interviewee (80-O) than for the interviewer (80-Y).

19Table 2: Distribution of tokens per speaker

































20 Following that, all tokens were coded for several factors:
• Predicted realisation in SVLR based on the following segment: short or long;
• Place of articulation of the next segment, for its influence on F2: coronal, dorsal, labial, vowel, none; /r/ and /l/ were coded separately given the changes in liquids in Glasgow;
• Potentially reduced words (e.g. could, would, you, be, me, she), given these may be reduced from /ʉ/ to schwa or from /i:/ to /i/
• Phrasal position: final or non-final, for impact of prosody on vowel lengthening (see Rathcke et al. 2016).

2.3 Statistical analysis

21 Dynamic micro-variation in duration and quality for these tokens was monitored using Generalized Additive Mixed Modelling (see Winter and Wieling 2016, Sóskuthy 2017), as GAMMs allow for obtaining smooths over time while still controlling for factors likely to influence the dependent variable. As specified in 3.1 and 3.2, one model was run for each pair of speakers and for each variable (F1 for each vowel, F2 for each vowel, duration for SVLR-short vowels, duration for SVLR-long vowels). Statistical analysis was conducted in R (v4.0.2) with the bam() function from the mgcv package; plotting used the plot_smooth() function contained within the itsadug package. Significance of the difference between the two speakers smooths was also tested with the plot_diff() function from the same package.

22 It must be said that the choice was made not to normalise formant data, unlike in previous work on the same corpus. Indeed, although vowel normalisation is common practice in sociolinguistics (Labov et al. 2006), we wanted to retain as much of the individual differences in vowel articulation as possible.

3. Results

3.1 Vowel quality

23 The model used for vowel height and frontness was as follows:

bam(F1 or F2 ~ Speaker + Following point of articulation +
s(Word, bs = "re") + s(Time, by = Speaker))

Speaker and following segment’s point of articulation were treated as fixed effects, while word was included as a random factor. The last part of the model formula provides one smooth per speaker, ordered along a Time x-axis, offering a linear perspective of time elapsed in the conversation. A total of sixteen models were run, four for each pair of speakers: two models were run per level of formant (F1 and F2), one for each vowel (/i/ and /ʉ/).

24As mentioned in 2.2, we know from previous work that the FLEECE vowel has been becoming more fronted over time. This means that F2 has been rising over time; we can then expect F2 to be rising during the conversation, and convergence to happen towards the speaker with the highest F2 to begin with. There has been no reported change in F1, which is then expected to be stable.

25 Results for the FLEECE vowel are shown in Figure 1 below. In this series of plots and the following, the blue smooth represents the interviewer (younger speaker) and the red smooth represents the interviewee (older speaker). Overall, we can see that the blue smooths have a slightly more serrated trajectory, while red smooths are less wiggly, which means that the older speakers are more consistent in their trajectories of variation. There is relatively small overlap between speaker’s trajectories and confidence intervals, meaning speakers within each pair differ from each other in their realisations of /i/. For F1, the tendency is for the formant value to decrease during the conversation; this is the reverse in pair 2 (YFOM), in which both speakers show a continuous rise in F1. This means most speakers pronounce /i/ higher and higher during the conversation. There is no obvious tendency for F2: formant values are lowering over the course of the interaction for pairs 2 and 4; however, in pairs 1 and 3, they are lowering for the youngest speaker, while rising for the interviewee.

26 We predicted convergence to happen towards the speakers with the highest F2, but no such trend is visible. This could perhaps be partly the case for F2 in pair 4: testing the significance of the difference between speakers’ trajectories using plot_smooth revealed that the two speakers in that pair were statistically different from each other during the first seven minutes of the conversation. During this time, they both converged towards each other, the older speaker shifting towards a less fronted articulation of /i/ while the interviewer’s instances of the same vowel were becoming more fronted. When they reached the same level of frontness, they continued converging towards a more fronted articulation.

Figure 1: intra-speaker variation in F1 (top row) and F2 (bottom row) for the FLEECE vowel, for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for formant values are on the y-axis, time elapsed in the conversation is on the x-axis. The blue smooth represents the interviewer (younger speaker), the red smooth represents the interviewee (older speaker).

Figure 1: intra-speaker variation in F1 (top row) and F2 (bottom row) for the FLEECE vowel, for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for formant values are on the y-axis, time elapsed in the conversation is on the x-axis. The blue smooth represents the interviewer (younger speaker), the red smooth represents the interviewee (older speaker).

27 Let us now turn to the BOOT vowel, which we know has been lowering over time. This means that F1 has been rising over time; we can then expect F1 to be rising during the conversation, and convergence to happen towards the speaker with the highest F1 to begin with. There has been no reported change in F2, which is then expected to be stable. Results are shown in Figure 2.

28 Similarly to trajectories and confidence intervals for /i/, there is very little overlap between speakers for F1. However, F2 is fully overlapping for the first three pairs, meaning in those pairs, /ʉ/ is pronounced with a similar degree of frontness by both speakers. Only members of the last pair do not overlap. F1 is lowering across the conversation for pair 1 while rising for the other three pairs. F2 is lowering for the first two pairs, while rising for the last two.

29 Our prediction that convergence would be towards the speaker with the highest F1 at the beginning of the interaction is once again not a straightforwardly visible pattern in our results. In pair 1, convergence actually takes place towards the speaker with the lowest F1. In pair 2, the younger speaker does not have a consistent shift, while the older speaker progressively articulates /ʉ/ lower. Pair 3 is full divergence. And although both speakers in pair 4 rise in F2, they become less similar to each other during the first minutes of the conversation.

Figure 2: intra-speaker variation in F1 (top row) and F2 (bottom row) for the BOOT vowel, for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for formant values are on the y-axis, time elapsed in the conversation is on the x-axis.

Figure 2: intra-speaker variation in F1 (top row) and F2 (bottom row) for the BOOT vowel, for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for formant values are on the y-axis, time elapsed in the conversation is on the x-axis.

3.2 Vowel quantity

30 The model used for vowel length (SVLR) was as follows:

bam(Segment duration ~ Speaker + Position + Reduced +
s(Word, bs = "re") + s(Time, by = Speaker))

Speaker, phrase position (final or non-final), and likeliness to be a weaker vowel (yes or no) were treated as fixed effects, while word was included as a random factor. The last part of the model formula provides one smooth per speaker, ordered along a Time x-axis, offering a linear perspective of time elapsed in the conversation. A total of eight models were run, two for each pair of speakers: one for vowels in SVLR-short contexts, one for vowels in SVLR-long contexts. Considering previous work by Rathcke and Stuart-Smith (2015) and Chevalier (2019) showed that there was no difference between /i/ and /ʉ/ in terms of quantity oppositions, both vowels were included in the same model, unlike for vowel quality analysis.

Figure 3: intra-speaker variation in vowel length for vowels in SVLR-short contexts (top row) and SVLR-long contexts (bottom row), for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for vowel duration are on the y-axis, time elapsed in the conversation is on the x-axis.

Figure 3: intra-speaker variation in vowel length for vowels in SVLR-short contexts (top row) and SVLR-long contexts (bottom row), for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for vowel duration are on the y-axis, time elapsed in the conversation is on the x-axis.

31 Those authors have also demonstrated that vowels in SVLR-short contexts have not changed in duration, while vowels in SVLR-long contexts have been shortening. We could then expect vowels in short contexts to remain stable in duration across the course of the interaction, and vowels in long contexts to become shorter as time elapses. This means that in long contexts, convergence should happen towards the speaker with the shortest vowels to begin with.

32Results are included in Figure 3. For vowels in SVLR-short contexts, speakers within pairs overlap in their trajectories and confidence intervals, which means they do not differ from each other in vowel duration in these contexts. There is also an important overlap between speakers for vowel length in lengthening contexts. This is not the case in pair 2, in which the speakers continuously and gradually diverge from each other (one shortening, one lengthening) across the whole conversation; the difference between them is significant past the fifth minute of the recording. For vowels in SVLR-long contexts, the tendency during interactions is surprisingly to lengthen long vowels. This doesn’t hold true for the first pair (both speakers shortening), and is only true for the older speaker in pair 2, as the younger speaker shortens her vowels.

33We predicted convergence to happen towards the speaker with the shortest vowels in SVLR-long contexts. This may be the pattern for pair 1, given both speakers converge towards each other, one lengthening and one shortening, in order to both shorten in the end. However, this is clearly not the trajectory in pair 2 (absolute divergence, one going each way), nor in pairs 3 and 4, where convergence is mostly towards the speaker with the longest vowels.

3.3 Summary

34 Our predictions and results are summarised in Table 3. None of the predictions have been confirmed in our results. The current study does not allow us to describe any further the relationship between individual micro-variation and community-level sound change, since the variables we studied did not shift in the same direction within minutes as they did in the long term at the same period. Furthermore, we are unable to comment on the role accommodation plays in sound change, given inter-speaker variation in our dataset did not demonstrably involve convergence towards the speaker the closest to the future norm.

35Table 3: Summary of findings




FLEECE vowel – F2

F2 rising during conversations
Convergence towards highest F2


BOOT vowel – F1

F1 rising during conversations
Convergence towards highest F1


Vowel quantity in SVLR-long contexts

Vowels shortening during conversations
Convergence towards shortest vowels


36However, several elements may have impaired our analysis. The main issue comes with the restricted number of tokens we could use reliably in the analysis: our dataset included relatively few tokens, especially for younger speakers, making statistical analysis less reliable given the number of factors included. Additionally, the complete overlap for vowel quantity is surprising; it may be explained by the method used for corpus alignment. The transcriptions were indeed time-aligned with the recordings using HTK with frames of 10ms in length: this means the alignment offers very little granularity, as segments are 50ms, 60ms, 70ms… Vowel duration measurements as extracted from LaBB-CAT do not provide the best opportunity to study small levels of variability for this phonological feature; investigating micro-variation in vowel quantity patterns further would require hand-correction of time boundaries in the alignment for more precise and granular measurements.

37It is thus inconclusive whether the same methods of monitoring tiny phonetic variation can be used to account for both phonetic changes and phonological changes. On one hand, hand-corrected measurements would be necessary to extend this study further; on the other hand, more data would be necessary to allow more reliable statistical analysis.

3.4 Discussion: age and accommodation behaviours

38 Our results raise questions about age and accommodation behaviours: for all the variables we considered in the present study, older speakers appeared more stable in their phonetic behaviour throughout conversations. If this unlikeliness to vary can be generalised to all speakers belonging to this generation, this suggests most of the accommodation-related shifts will come from the youngest speakers in mixed-generation interactions. This is in line with earlier work by Coupland et al. (1988), among others.

39 The two younger speakers in this dataset had each been recorded with several older speakers, and for the present study, we selected one interlocutor of each gender for each interviewee. We move on to asking how similar the two young speakers behave linguistically in different conversations. The female interviewee is very consistent – there is no striking difference in her measurements between the two recordings (YFOF and YFOM), which suggests she does not accommodate to her interlocutor prior to the interaction. On the contrary, she does seem to accommodate to the other speaker during the interaction when there is a substantial difference in their speech (e.g. her downward shift in F1 for BOOT). On the other hand, the young male speaker appears to have different ways of articulating /i/ (both in height and frontness) and lengthening vowels in SVLR-long contexts, and this is the case from the very beginning of the interaction (i.e. his starting point for those variables is not the same in recording YMOM and recording YMOF). For those variables, his initial phonetic behaviour is each time very close to his interviewee’s pronunciation. This is a strong sign of accommodation; however, this suggests that speech accommodation does not actually take place during the recording, but prior to the discussion, implying it is more likely motivated by linguistic prejudice on his interlocutor’s speech rather than by phonetic imitation.

40This echoes one of the main debates about accommodation, namely the very nature of its mechanism: is it the mirroring of a phonetic stimulus, or proactive adaptation to the interlocutor’s speech based on non-speech elements? Bell (1984) devised what he called the “audience design model”, to account for intraspeaker variation depending on the speaker’s audience. This audience includes not only the speaker’s addressee(s), but also their auditor(s) and their overhearer(s). If a speaker accommodates to the last two, it can only be through a non-individual linguistic factor and therefore cannot be phonetic convergence per se, meaning that non-speech elements do play a role in the way speakers adapt to their audience. Consequently, Bell considers that the most likely process of interpersonal accommodation consists in “speakers assess[ing] the personal characteristics of their addressees, and design[ing] their style to suit it”. Given our results, we can apply Bell’s thinking to our data, and we can regard the young male speaker’s phonetic behaviour as accommodation based on non-linguistic features. In other words, this speaker would have evaluated the social characteristics of his addressee (in particular age and gender), to then adopt a way of speaking that corresponded to what he expected from the addressee; it is thus convergence towards what the speaker “mistakenly assumes will be the addressee’s speech on the basis of non-speech attributes”. This suggests that the study of linguistic convergence might benefit from greater consideration of the audience’s social characteristics in addition to individual speech variables.

Top of page


Aitken, Adam J. “The Scottish Vowel Length Rule.” So meny people, longages and tonges: Philological essays in Scots and Mediaeval English, presented to Angus McIntosh. Edited by Michael Benskin and Michael L. Samuels. Edinburgh: The Middle English Dialect Project, 1981: 131-157.

Bell, Allan. “Language Style as Audience Design.” Language in Society 13:2 (1984): 145-204.

Boersma, Paul and David Weenink. “PRAAT, a system for doing phonetics by computer.” Glot International 5 (2001): 341-345.

Chevalier, Florent. “On sound change and gender: the case of vowel length variation in Scottish English.” Anglophonia [Online] 27 (2019).

Coupland, Nikolas et al. “Accommodating the Elderly: Invoking and Extending a Theory.” Language in Society 17:1 (1988): 1-41.

Cukor-Avila, Patricia and Guy Bailey. “Real Time and Apparent Time.” The Handbook of Language Variation and Change, Second Edition. Edited by Jack Chambers and Natalie Schilling. Boston: Wiley Blackwell, 2013: 239-262.

Foulkes, Paul and Gerard Docherty. Urban Voices. London: Arnold, 1999.

Fromont, Robert and Jennifer Hay. “LaBB-CAT: An annotation store.” Proceedings of the Australasian Language Technology Workshop (2012): 113-117.

Giles, Howard et al. “Accommodation Theory: Communication, Context, and Consequence.” Contexts of Accommodation: Developments in Applied Sociolinguistics Edited by Howard Giles et al. New York, NY: Cambridge University Press, 1991: 1-68.

House, Arthur S. and Grant Fairbanks. “The influence of consonant environment upon the secondary acoustical characteristics of vowels.” Journal of Acoustical Society of America 25 (1953): 105-113.

José, Brian and Jane Stuart-Smith. “Methodological issues in a real-time study of Glaswegian vowels: Automation and comparability.” LSA Annual Meeting Extended Abstracts (2014).

Labov, William. Principles of linguistic change. Vol. 1: Internal factors. Oxford: Blackwell, 1994.

Labov, William. Principles of linguistic change. Vol. 2: Social factors. Oxford: Blackwell, 2001.

Labov, William et al. The Atlas of North American English: Phonology, Phonetics, and Sound Change. A Multimedia Reference Tool. Berlin: Mouton de Gruyter, 2006.

Lawson Eleanor et al. “A socio-articulatory study of Scottish rhoticity.” Sociolinguistics in Scotland. Edited by Robert Lawson. London: Palgrave Macmillan, 2014: 53-78.

Rathcke, Tamara and Jane Stuart-Smith. “On the Tail of the Scottish Vowel Length Rule in Glasgow.” Language and Speech 58 (2015): 1-27.

Rathcke, Tamara et al. “What is the fate of the Scottish Vowel Length Rule in Glasgow?” Phonetik und Phonologie 12. Ludwig-Maximilians-Universität München, 14 October 2016. Conference presentation.

Saussure, Ferdinand. Cours de linguistique générale. Edited by Charles Bally and Alert Sechehaye. Lausanne – Paris: Payot, 1916.

Scobbie, James et al. “Morphemes, Phonetics and Lexical Items: The Case of the Scottish Vowel Length Rule.” Proceedings of the XIVth International Congress of Phonetic Sciences, vol. 2 (1999): 1617-1620.

Solanki, Vijay. Brains in dialogue: investigating accommodation in live conversational speech for both speech and EEG data. 2017. University of Glasgow, PhD thesis.

Sonderegger, Morgan et al. “Managing data for integrated speech corpus analysis in SPeech Across Dialects of English (SPADE).” Open Handbook of Linguistic Data Management. Edited by Andrea Berez-Kroeker et al. Cambridge: MIT press, in press.

Sóskuthy, Marton. “Generalised Additive Mixed Models for dynamic analysis in linguistics: A practical introduction.” (2017). arXiv:1703.05339

Stuart-Smith, Jane et al. “The private life of stops: VOT in a real-time corpus of spontaneous Glaswegian”. Laboratory Phonology, 6.3-4 (2015a): 505-549.

Stuart-Smith, Jane et al. “Using big-data techniques to tackle real-time change: The voicing contrast across 100 years of real- and apparent-time in Glaswegian.” Celebrating 10 years of research on language change in real time. University of Copenhagen, 30 September 2015. Conference presentation (2015b).

Stuart-Smith, Jane et al. “Changing sounds in a changing city: an acoustic phonetic investigation of real-time change over a century of Glaswegian.” Language and a Sense of Place: Studies in Language and Region. Edited by Chris Montgomery and Emma Moore, Cambridge University Press: Cambridge, 2017: 38-64.

Tanner, James et al. “Towards ‘English’ phonetics: variability in the pre-consonantal voicing effect across English dialects and speakers.” Frontiers in Artificial Intelligence 3:38 (2020).

Toma, Catalina L. “Towards Conceptual Convergence: An Examination of Interpersonal Adaptation.” Communication Quarterly 62:2 (2014): 155-178.

Trudgill, Peter. Dialects in contact. Oxford: Blackwell, 1986.

Wells, John. C. Accents of English 2: The British Isles. Cambridge: Cambridge University Press, 1982.

Winter, Bodo and Martijn Wieling. “How to analyze linguistic change using mixed models, Growth Curve Analysis and Generalized Additive Modeling.” Journal of Language Evolution 1:1 (2016): 7-18.

Top of page

List of illustrations

Title Figure 1: intra-speaker variation in F1 (top row) and F2 (bottom row) for the FLEECE vowel, for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for formant values are on the y-axis, time elapsed in the conversation is on the x-axis. The blue smooth represents the interviewer (younger speaker), the red smooth represents the interviewee (older speaker).
File image/png, 106k
Title Figure 2: intra-speaker variation in F1 (top row) and F2 (bottom row) for the BOOT vowel, for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for formant values are on the y-axis, time elapsed in the conversation is on the x-axis.
File image/png, 95k
Title Figure 3: intra-speaker variation in vowel length for vowels in SVLR-short contexts (top row) and SVLR-long contexts (bottom row), for the pairs YFOF – YFOM – YMOM – YMOF. Estimates for vowel duration are on the y-axis, time elapsed in the conversation is on the x-axis.
File image/png, 95k
Top of page


Electronic reference

Florent Chevalier, Intra-speaker phonetic micro-variation, and its relationship to phonetic and phonological changeAnglophonia [Online], 30 | 2020, Online since 20 December 2020, connection on 27 January 2022. URL:; DOI:

Top of page

About the author

Florent Chevalier

FoReLLIS, Université de Poitiers

By this author

Top of page


Licence Creative Commons
Anglophonia – French Journal of English Linguistics est mis à disposition selon les termes de la licence Creative Commons Attribution - Pas d'Utilisation Commerciale - Pas de Modification 4.0 International.

Top of page
Search OpenEdition Search

You will be redirected to OpenEdition Search