Phonological Merger Meets Resistance: Diffusion,Divergence,and Convergence in East Anglia

Abstract

The MOAN–MOWN merger (i.e., the merger of the Middle English sources of GOAT) is an ongoing sound change in Lowestoft English, a variety of East Anglian English spoken at the intersection of the region’s northern and southern sub-zones. This study examines the merger’s progress across three generations in apparent time, using 2615 tokens of spontaneous and word list speech—analyzed dynamically with generalized additive mixed models—alongside perception data from a binary forced-choice identification task. The analysis addresses how phonemic mergers diffuse into areas of stable resistance and whether the target vowel in Lowestoft assimilates to southeastern British varieties where moan and mown are merged under goat. Results suggest a gradual merger by approximation, progressing via incremental stages with different intergenerational targets of assimilation. Change is linked to demographic shifts following the collapse of Lowestoft’s fishing industry, but the merger’s diffusion differs from that of other mergers triggered by social change into areas of resistance, both in terms of mechanism and timeframe. The diffusion of goat fronting is reported, affecting both moan and mown where a contrast is maintained, without a loss of distinction in perception. This demonstrates how phonetic diffusion can operate, despite differences in phonological structure between donor and recipient varieties. However, mown lags behind moan, highlighting how local divergence can accompany long-term convergence toward supra-local norms.

1. Introduction

The Long Mid Mergers of the sixteenth/seventeenth centuries, which developed from the Great Vowel Shift, did not immediately take place in East Anglia, unlike in most other British English varieties (Wells 1982:193). This left the phonemic distinction between vowels descended from Middle English ǫ: /ɔ:/ (moan) and ou /ɔu/ (mown) intact, contrasting pairs such as road-rowed and doe-dough (East Anglia = /ʊu/-/ʌu/). The merger toward a single goat vowel has only been reported as complete in southern East Anglia since the 1970s (Trudgill & Foxcroft 1978), while the merger’s distribution in northern East Anglia remains unclear and may be the last holdout for a moan-mown contrast in Britain (Trudgill 2021:207). The present work examines the moan-mown merger for working-class speakers of Lowestoft English, a variety spoken in Britain’s easternmost town, located on the fringes of northern East Anglia.

Lowestoft can be classified—along with contiguous/nearby East Anglian English varieties, for example, Norwich (Labov, Yaeger & Steiner 1972; Trudgill 1974)—as an area of ‘stable resistance’ to the merger. This term is used by Labov, Ash, and Boberg (2006:65) who argue that such pockets of resistance develop when structural factors are strong enough to inhibit a merger’s spatial expansion, challenging the widely accepted assertion that mergers expand at the expense of distinctions (“Herzog’s Principle”; Labov 1994:35). In the Lowestoft case, resistance stems from moan undergoing a variable (and thus reversible) merger with goose (Labov, Yaeger & Steiner 1972; Trudgill 1974; Butcher 2021), increasing/maintaining the phonetic distance between moan and mown. Evidence of weakened resistance to the moan-mown merger in working-class Lowestoft English has emerged only recently (Butcher 2021), some five decades after Trudgill and Foxcroft (1978:72) reported the absence of a consistent merger among the working-class population. Our understanding of how mergers diffuse into areas of stable resistance remains limited, partly due to the rarity of such pockets of defiance, although they do seem to appear primarily in the context of geographically widespread mergers, such as cot-caught, which is now relevant to over half of North America’s Anglophone population (Labov, Ash & Boberg 2006:41).

This paper has two main aims. The first is to explore how mergers diffuse into areas of stable resistance, and the social/linguistic factors implicated in change. Two distinct patterns of diffusion are reported in the literature, namely (1) immediate community spread at the point of expansion/contact (e.g., the cot-caught merger on the Massachusetts-Rhode Island border; Johnson 2007), and (2) expansion by gradual sound change in the direction of merger, often affecting one phoneme (e.g., the cot-caught merger in Upstate New York, historically impeded by independent cot fronting as part of the Northern Cities Shift; Dinkin 2011). Lowestoft presents an opportunity to shed light on the mechanics and potential timelines for type-two merger diffusions. Further, this merger-meets-resistance scenario has yet to be examined in the context of British English.

The second aim is to ascertain the stage of the moan-mown merger and evaluate whether the phonetic realization of the resulting merged goat vowel in Lowestoft English assimilates or dissimilates to/from other varieties of Southern British English. The merger involves the loss of a distinction that is no longer present in other English varieties; speakers outside of Lowestoft English therefore have no underlying phonological knowledge of it. Without an external blueprint for how these phonemes should merge in the contemporary context, Lowestoft speakers may only rely on goat exemplars encountered via exposure to speakers with a phonological system that differs from their own. Theoretically, this ought to lead to a situation of convergence. However, some varieties resisting the diffusion of a phonological feature have been found to diverge from supra-local phonological norms (e.g., the diffusion of t-glottaling into Liverpool; Watson 2006). In these cases, hybrid outcomes may emerge, as documented by Britain (2005) for the diffusion of the moan-mown merger and goat fronting—an increasingly common phenomenon in southern England (Williams & Kerswill 1999; Kerswill & Williams 2005; Altendorf & Watt 2008)—into the East Anglian Fens. Britain finds that, although the distinction persists in production, fronting only affects the historical mown class. The application of goat fronting to the two-vowel East Anglian system illustrates how the phonetics of phonemes involved in mergers can be shaped by external models, resulting in both localized divergence (moan and mown are now phonetically more distant) and external convergence (mown is now phonetically more similar to external goat). Whether Lowestoft English behaves similarly remains an open question.

The present study draws on data from three generations of working-class Lowestoft speakers. Dynamic vowel trajectories taken from word list and spontaneous speech data are modeled using Generalized Additive Mixed Models (GAMMs). GAMM outputs, together with perception data from a binary forced-choice identification task, are used to identify the stage of merger and to address the main aims of the paper. This work aims to deepen our understanding of the specific mechanics underlying diphthongal mergers, which remain underexplored relative to monophthongal mergers. Theoretically, it offers insight into the role of phonetic variation in the early stages of merger, while methodologically, it highlights the importance of integrating both production and perception for identifying the stage of a sound change, as well as the value of incorporating dynamic measurements into studies of vowel mergers.

2. Background

Historical dialectological records from Ellis (1889), Wright (1905), and Kökeritz (1932) indicate that the phonemic distinction between the moan and mown classes remained fully intact in East Anglia until the early to mid-twentieth century. Evidence points to the merger’s proliferation northwards and eastwards of London in both a temporally and geographically gradual manner through East Anglia toward Lowestoft from the 1960s onwards. The first evidence in favor of this interpretation comes from the Survey of English Dialects (SED, Orton et al. 1962-1971), which documents the distinction as variable in southern East Anglia, while Trudgill and Foxcroft (1978) report its disappearance from the same area by the late 1970s, surviving only as a relic in certain moan words (e.g., go). They note that the distinction persists in northern East Anglia with considerable consistency at this time, while on the north-south border it is preserved in rural areas only and significantly diminished in urban settings, like Lowestoft.

Trudgill and Foxcroft’s Lowestoft data (1978:72) from the 1970s also reveal social stratification of the merger: 30 percent of middle-class speakers exhibit a consistent merger compared to 0 percent of working-class speakers, while 40 percent of the middle class show a variable merger, versus 23 percent of the working class. This signals the early stages of weakened resistance, a trend observed in more recent studies. For instance, Simm (2019:27) reports [əu] (36 percent) as the modal moan variant, followed by [əʊ̝] (31 percent), in her study of Great Yarmouth and Gorleston—approximately 7 miles north of Lowestoft—although her data exclude words with mown-type spellings. Ferragne and Pellegrino (2010:13) treat both moan and mown under the goat phoneme in their Lowestoft data, noting a “rather back” quality. ¹ While Butcher’s (2021) Lowestoft study treats the two phonemes separately, findings indicate that young speakers are either fully merged in production (word list speech), or in an advanced transitional stage, while older generations retain a full distinction.

Regarding the mechanism by which the merger is taking place, the SED strongly indicates “merger by approximation”—whereby two phonemes gradually approximate one another in phonetic space (Trudgill & Foxcroft 1978:72)—as evidenced by intermediate variants in both vowel classes. For example, in Kedington and Kersey (southern Suffolk SED locations), [o̜ʊ] appears alongside [ʌʊ], possibly indicating a transitional stage of moan unrounding. However, Trudgill and Foxcroft (1978) convey the situation as more complex. Middle-class Norwich speakers (northern East Anglia) produce intermediate vowels for both moan and mown, indicating merger by approximation, apparently driven by increased exposure to RP vowels. Butcher (2021:68) also discusses merger by approximation as a likely mechanism in Lowestoft, where intermediate forms are documented in word list speech. Conversely, Trudgill and Foxcroft (1978) report “merger by transfer”—merger on a word-by-word basis—toward mown among working-class speakers in southern East Anglia, apparently influenced by proximity to London and its working-class goat vowel. The authors essentially argue that the motivations for merger differ based on social proximity to prestige varieties for northern East Anglia, and a combination of social and geographic proximity to regional varieties for the south, ultimately leading to the implementation of different mechanisms.

Today’s context for linguistic change in Lowestoft is unique. The town was historically defined by its fishing industry, which can be traced as far back as the medieval period. At the turn of the twentieth century, fishing still accounted for 75 percent of the town’s employment (Brookfield, Gray & Hatchard 2005:66) and remained a way of life until after the First World War when the demand for fish fell, triggering the onset of the industry’s sharp decay. Today, the industry is in a “state of terminal decline” (Bond 2017:66), employing around 10 percent of the Lowestoft population (Brookfield, Gray & Hatchard 2005:65). The town has since undergone extensive economic diversification, re-establishing itself as a tourist destination (Chow Tat Sing 2017:1095), which has led to new contact.

Despite belonging administratively to the county of Suffolk, Trudgill (2021:57) positions Lowestoft within the northern subzone of “Core Linguistic East Anglia”—alongside all of Norfolk—arguing for a north-south divide that is both linguistic and cultural (Figure 1). Consequently, Lowestoft straddles a cultural, linguistic, and administrative boundary in a northeastern East Anglian location, which is geographically isolated from the rest of England (as its most eastern point) and where prototypically East Anglian linguistic features are more likely to be retained (Trudgill 2021:51). It is therefore uncertain whether Lowestoft will follow the diffusion patterns observed in other northern East Anglian communities, where ‘social’ and ‘geographic’ proximity to neighboring varieties is more straightforward conceptually.

Figure 1.

North and South Linguistic Zones of Core East Anglia

3. Methods

3.1. Production Methods

Thirty participants (ages 16-82, mean = 50.0, SD = 22.3) took part in sociolinguistic interviews (Labov 1966). All had lived in Lowestoft their entire lives, spent no more than one year away, and were from working-class backgrounds. Data were collected between 2018 and 2019 and recordings were carried out with a Zoom H4n Handy Recorder at a 16-bit 44.1 kHz sampling rate. Participants were evenly distributed across three age groups: young (16-34 years, born 1986-2005, n = 10), middle-aged (35-64 years, born 1957-1975, n = 10), and old (65+ years, born 1942-1951, n = 10). They were also split by gender (male, n = 15 and female, n = 15). Age stratification followed an emic approach (Eckert 1997), based on shared knowledge or experiences of Lowestoft’s fishing industry. The sixty-five and over group had direct industry ties, the thirty-five to sixty-four group recalled family involvement, and those under thirty-five, who were largely born after the industry’s decline, had only passive knowledge. I refer to these groupings as ‘generations’ herein.

Fifteen monosyllabic tokens per word class per speaker were drawn from spontaneous speech, all in stressed contexts without following approximants. The word list included fifteen instances of each target vowel (moan and mown), balanced across open and closed monosyllables, as goat F2 is typically higher in open syllables (Baranowski 2013; Warburton 2020). Seven sole and seven soul words (moan and mown with coda /l/) were also included and analyzed separately. Twenty monosyllabic fillers from other lexical sets were used, along with five fleece, goose, trap, foot, lot, and thought tokens for vowel normalization.

Recordings were transcribed orthographically in ELAN version 5.8 (2008) and force-aligned using Montreal Forced Aligner (McAuliffe, Socolof, Mihuc, Wagner & Sonderegger 2017). Post-hoc checks and manual correction took place to ensure boundaries were accurate. In line with recent research that takes a dynamic approach to vowel trajectory analysis (Kirkham, Nance, Littlewood, Lightfoot & Groarke 2019; Renwick & Stanley 2020; Cole & Strycharczuk 2022), F1 and F2 measurements were extracted from eleven points across each vowel token at equal intervals. This was achieved using a simple custom Praat script, with Linear Predictive Coding (LPC) parameters—for example, maximum formant value, number of formants—manually adjusted on a speaker-by-speaker and vowel-by-vowel basis, as is routine for sociophonetic analyses (Kendall & Fridland 2021:45). In addition to formant values, the script extracted preceding and following environments (phoneme and place) for each vowel token, plus vowel duration and word.

Vowel normalization was undertaken using the Modified Watt and Fabricius method (mW&F; Watt & Fabricius 2002), with 2615 tokens analyzed in total. This method was developed specifically with sociolinguistic studies of vowel variation and change in British English in mind (Watt & Fabricius 2002; Fabricius, Watt & Johnson 2009), and has been shown to perform comparably to its vowel-extrinsic competitor “Lobanov” in reducing inter-speaker variation without requiring a fully populated vowel space. In a broader comparison of twenty vowel normalization techniques, Flynn (2011) found mW&F to be the most effective for aligning vowel spaces across speakers of different ages and genders.

To distill production results for comparison with and modeling of perception data, participants were categorized as either ‘distinct,’ ‘transitional,’ or ‘merged’ based on their speech production patterns, in a similar vein to definitions used by Labov, Ash, and Boberg (2006:62) in the Atlas of North American English. Categorization was based on concordance (or lack thereof) between merger status in production, as assessed via both Pillai scores—a MANOVA-derived statistic measuring overlap between vowel distributions—and GAMMs, which predict differences between vowel trajectories, focusing on height, backness, and shape (see Wood 2017 for an overview). As each method captures different aspects of vowel data, this dual approach aims to achieve more accurate classifications, while evaluating the efficacy of both techniques.

Sample size for moan and mown tokens was balanced across all speakers and speech styles (fifteen each, per speaker, per style), as uneven distributions are known to reduce the accuracy of Pillai scores (Stanley & Sneller 2023). For each speaker, Pillai scores were calculated separately by speech style (spontaneous versus word list) and by measurement point (30 percent versus 70 percent of total vowel duration), with F1 and F2 values entered into a MANOVA in R (R Core Team 2025). A Pillai score of 1 indicates a distinction, while 0 indicates complete overlap. Following Austen (2020) and Bowie (2021), speakers were classified using condition-specific Pillai cut-offs at the p = .05 significance threshold, as illustrated in Table 1. In each condition, Pillai values below the cut-off indicate no statistically reliable separation (merged), while values above the cut-off indicate a reliable separation (distinct). Participants whose Pillai scores were not statistically significant for both vowel nucleus (30 percent of total duration) and offglide (70 percent) measurements were labeled as ‘merged.’ Those whose scores were significant for one measurement but not the other were labeled as ‘transitional,’ while those with significant scores for both were labeled ‘distinct.’

Table 1.

Pillai Cutoffs for moan–mown Merger/Distinction (30 Percent Nucleus, 70 Percent Offglide) by Style

Speech style	Nucleus	Offglide
Word list	0.14	0.30
Spontaneous	0.30	0.39

To determine labels via GAMMs, I followed Sóskuthy (2017:22), using model comparisons to assess whether vowel trajectories differ significantly. Each speaker’s data was separated between speech styles and input into a model including vowel as a parametric effect (ordered with contrast coding), a difference smooth (measurement by vowel), and a smooth over measurement. This was compared to a nested model with only measurement, using compareML(). A significant result indicated a difference between trajectories and therefore vowels, providing a clear basis for ‘distinct’ and ‘merged’ labels. If Pillai and GAMM classifications disagreed, participants were labeled ‘transitional,’ as were those whose labels differed across speech contexts.

All observed disagreements between the two methods (n = 6) involved cases where GAMMs classified a speaker as distinct, but Pillai scores indicated merger, suggesting spatial overlap alongside a residual distinction between trajectories not captured by the ANOVA that GAMMs are sensitive to. Purely as a representative example, Figure 2 shows static plots for moan and mown offglides in one such case. In this example, the speaker received Pillai scores of 0.11 (nucleus) and 0.01 (offglide), which are both below the Pillai threshold for merger.

Figure 2.

moan/mown Raw Vowel Plot at 30 Percent (Left) and 70 Percent (Right): Speaker With Mismatched GAMM–Pillai Labels

GAMM predictions suggest a distinction between moan and mown, most pronounced in the offglide, as shown in Figure 3. The GAMM F1 difference smooth (Figure 4) indicates a statistically significant difference along the entire trajectory, while the F2 smooth shows a significant difference across most of the trajectory, except for a brief region around 40 percent where the confidence interval crosses zero. The discrepancy between GAMM and Pillai results may arise in both cases because the Pillai trace summarizes overall separation across F1 and F2 without weighting formants where distinctions are concentrated. As shown in the static plots at 30 percent, the majority of mown tokens are concentrated above 600 Hz for F1, but fall largely within the distribution of moan. Similarly, at 70 percent of vowel duration, most mown tokens fall within the overall moan distribution, but moan includes outliers around 1800 Hz that could represent a distinction. Speakers showing this type of GAMM–Pillai disagreement were retained as ‘transitional’ to avoid privileging one method and to preserve the possibility that such cases reflect genuine cases of variable or near-merger—an interpretation explored in ongoing work.

Figure 3.

moan/mown GAMM Fits: Speaker With Mismatched GAMM–Pillai Labels (F1 Left; F2 Right)

Figure 4.

moan/mown GAMM Difference Smooths: Speaker With Mismatched GAMM–Pillai Labels (F1 Left; F2 Right)

It is important to point out that the categories described here, although precise in their construction, are only intended to separate the ‘most’ merged participants and the ‘least merged’ participants from the ones most likely to be in between and, in general terms, this categorization scheme—while rough and approximate—allowed for the systematic inclusion of speakers’ vowel production as a predictor of speaker perception. This is not to say that the scheme is perfect, more an abstraction for methodological convenience; further exploration and modification is encouraged in future research.

3.2. Production Data Trajectory Modeling

The datasets were analyzed using Generalized Additive Mixed Models (GAMMs), built in R and run in the mgcv package via the bam() function (Wood 2017). Separate models were fitted for F1 and F2 as dependent variables for moan-mown and sole-soul pairs, resulting in four GAMMs. In each model, vowel (moan-mown or sole-soul) was treated as ordered and contrast was applied. Following Stanley, Renwick, Kuiper, and Olsen (2021:401), three model types were constructed. Each full model included vowel, speech style, generation, and gender as a four-way interaction, both as smooth and parametric effects. A factor smooth, preceding place of articulation, accounted for shape variation across vowels; random intercepts were fitted for word and speaker; and a tensor product smooth over measurement number controlled for duration. Reference levels were set to ‘moan, spontaneous speech, old, male’ and ‘sole, spontaneous speech, old, male’ as the most phonetically divergent from ‘standard’ goat and goal vowels. Smooth complexity (k) was checked using gam.check() to avoid over/underfitting.

GAMM outputs are complex and lengthy (see discussion in Stanley, Renwick, Kuiper & Olsen 2021:401), thus they were evaluated in two ways. First, predicted trajectories were visualized with ggplot2, offering a clear assessment of differences in vowel height/backness and shape, while pinpointing at what point exactly along the vowel trajectories speakers are differentiating between classes of sounds. Second, a series of nested models was built by removing one predictor (generation, gender, or speech style) from the interaction. These were evaluated against the full model using compareML() from the itsadug package (van Rij, Wieling, Baayen & van Rijn 2020), with estimated degrees of freedom and p-values used to assess the predictor’s effect on vowel overlap. In all cases, the full four-way interaction indicated a significant effect of all included predictors, supporting their inclusion. Full model specifications and outputs are available in the Supplemental Materials online.

3.3. Perception Methods

A participant’s ability to categorize or label moan/mown phonemes was assessed through a binary forced-choice identification (ID) task in which participants were played a single audio stimulus from a seven-step vowel continuum and asked to label it as one of two categories presented to them orthographically. The continuum between moan and mown was synthesized between natural speech endpoints and spliced back into two different contexts (doe-dough, toed-towed) using Praat (Boersma and Weenink 2022). The full procedure for continua construction follows Stewart (2018), with endpoints based on the speech of a Lowestoft native who had the highest degree of distinction of all participants in an earlier pilot study, as determined by Pillai scores. Both continua were checked for naturalness by two phoneticians and adjusted as needed. Figure 5 shows the doe-dough continuum (top to bottom).

Figure 5.

Spectrograms Showing Seven-Step Vowel Continuum for Doe-Dough

In the task, each context was presented twice, and the order of responses was counterbalanced. Each continuum was therefore presented to participants four times (total number of trials = 28). Audio stimuli were randomized, and no fillers were used (following e.g., Drager 2010; Fridland & Kendall 2012). All responses were collected in PsychoPy v3.1.5 (Peirce et al. 2019) and participants wore Sennheiser HD 520 Headphones.

To quantify categorical perception, a category boundary width score was calculated for each condition and participant. Boundary width reflects the distance between 25 percent and 75 percent probability of perceiving stimuli as A or B. Wider boundaries indicate less categorical perception and greater likelihood of merger. Logistic regressions were fitted to each participant’s responses by condition, with continuum step as the fixed predictor. Boundaries were derived from model intercepts and step coefficients by calculating x when y is 0.75 in addition to x when y is 0.25. Mixed effects linear regression was then applied to boundary widths, with models refined via step-down procedures and AIC comparisons. Full models included generation, gender, minimal pair condition, that is, doe-dough or toad-towed, and merger status in production (‘distinct,’ ‘transitional,’ ‘merged’), as determined by performance in the word list and spontaneous speech tasks.

4. Results

I begin by describing any changes in vowel height, backness, and trajectory shape observed in apparent time. GAMM predictions for F1 and F2 are presented separately, as differences may vary in magnitude at different points along the trajectories for each dimension. Moreover, mergers may proceed unevenly, with one formant leading change while the other follows or remains stable. All F1 and F2 values given in the following discussion are mW&F normalized.

4.1. moan-mown F1 Trajectories

The F1 predictions for moan and mown (Figure 6) show that, for the youngest generation, both vowels have onsets at around 1.1-1.25 for both speech styles and across genders—a phonetically intermediate point between the clearly distinct onsets of moan and mown observed in the older generations’ spontaneous speech. This suggests both vowels are participating in change: moan’s nucleus is lowering in the vowel space and mown’s nucleus is raising. A gender difference is observed in the latter portion of the mown vowel, which looks to be raising in apparent time (lowering of F1 values) for females in the middle generation—but not males—when compared to the older generation. Similarly, the moan offset remains stable for these females, but lowers in the vowel space (increased F1 values) for males.

Figure 6.

Predicted GAMM F1 Trajectories (mW&F): moan and mown by Vowel, Generation, Gender, and Style

Overlapping confidence intervals for the youngest generation suggest that neither vowel height nor trajectory shape varies significantly, providing strong evidence of a merger along this plane. In fact, only minor differences in trajectory shape can be observed across the data in apparent time at all. This is with the exception of mown, which has a more dynamic, steep trajectory in word list speech than in spontaneous speech for both the old and middle generations.

4.2. moan-mown F2 Trajectories

The model predictions for moan and mown F2, plotted in Figure 7, show variably distinct trajectories across both speech styles for speakers of both genders in the oldest generation, with vowel frontness decreasing into the offglides. This pattern shifts dramatically for the youngest speakers, whose offglides instead front. The result is a striking inversion in trajectory shape: front-to-back in the oldest generation, back-to-front in the youngest. This is most evident in word list speech, where moan’s average offglide value shifts from <1 (old) to >1.50 (young), and mown from <0.85 (old) to >1.35 (young). Intriguingly, the middle generation does not follow either pattern. Their less dynamic trajectories reflect a variable phonetic distinction between moan and mown that is even greater in magnitude and consistency than that of the oldest generation. This divergence does not reduce phonetic distance between the vowels but in fact renders them more distinct.

Figure 7.

Predicted GAMM F2 Trajectories (mW&F): moan and mown by Vowel, Generation, Gender, and Style

The plots also reveal offglide fronting as a change in apparent time. The middle generation’s vowels begin in similar positions to the older group, but instead of retracting, their trajectories remain relatively static, resulting in the observed fronting trend. This may be a female-led change, initiated by the middle generation, as fronting of both vowels is most advanced for this group, while young females show the highest overlap in spontaneous speech. This speech register typically reflects vernacular norms and emerging changes (Labov 1966). Notably, both vowels may not have begun fronting at the same rate: moan’s endpoint is uniform across genders and styles for the middle generation (c. 1.25), while mown lags behind in all contexts except middle female spontaneous speech (our potential leaders). The moan trajectory for the middle females most closely resembles that seen in young female spontaneous speech, further supporting the idea that middle-generation females led this change.

The predicted trajectories and confidence intervals also offer insight into the stage of the merger. For the youngest generation, the two vowels behave similarly in both speech styles, but wider confidence intervals at endpoints may indicate individual variation. Some speakers may approximate or even overlap moan and mown F2 values, while others maintain a distinction via a fronted moan offglide. Although wider confidence intervals at edges are expected—since the model has less information to constrain estimates—this widening is noticeably greater than at the onset and may reflect genuine individual differences consistent with near-merger. Stylistic differences look to emerge among the youngest generation: offglide fronting is more pronounced in word list speech than in spontaneous speech, and trajectories are overall more dynamic. Such style-shifting can reflect broader community norms (Gordon 2013:252), and the exaggerated fronting in formal contexts may indicate that moan-mown/goat offglide fronting is now recognized as a community norm among younger speakers.

4.3. sole-soul F1 Trajectories

The GAMM predictions for sole and soul F1, shown in Figure 8, reveal only minor variation for height, shape, generations, or genders across styles. Most trajectories suggest a subtle raising of both vowels, beginning at around 1.25-1.35, which is more prominent in the spontaneous speech of the oldest females and their younger counterparts’ word list data. Only the sole trajectories of the oldest generation’s word list speech can be said to differ meaningfully, instead beginning with a lower F1 value of around 1.0, indicating a more raised tongue position. This group exhibits clear style-shifting, maintaining a clear sole-soul distinction along the F1 plane in word list speech, which is lost in spontaneous speech. This contrasts with the findings for moan-mown (non-coda-/l/ contexts), where older speakers consistently retained a distinction across styles.

Figure 8.

Predicted GAMM F1 Trajectories (mW&F): sole and soul by Vowel, Generation, Gender, and Style

This style-shifting pattern is not observed in the speech of the two younger generations, for whom both vowel trajectories exhibit either overlap or close approximation. The overlapping, but wide, confidence intervals may indicate merger for some but not others, or potentially variable merger. However, given that the oldest generation merges or nearly merges these two vowels in spontaneous speech, close approximation without full merger among the generations below would be somewhat unexpected. Overall, the (nearly) merged forms produced by the youngest generation mirror the F1 of soul in word list speech of the oldest generation in terms of shape and roughly in height, suggesting the merger may be occurring in the direction of soul.

4.4. sole-soul F2 Trajectories

The model estimates for the young and middle generations show a high degree of consistency across trajectory shapes and speech styles. For both groups, confidence intervals for sole and soul (Figure 9) remain narrow and overlap for most of their trajectories in both speech styles. This overlap suggests that the distinction between the two vowels has likely been lost, though there is perhaps some variation between individual speakers. Both vowels exhibit falling F2 values throughout their trajectories, reflecting a consistent movement toward the back of the vowel space. Significant variation is only observed for sole in the word list speech of the oldest generation. Here, the predicted trajectories are considerably more dynamic, with a higher degree of spectral change occurring across the vowel’s total duration. As a result, a distinction between sole and soul is maintained by the oldest speakers in word list speech; this distinction disappears in spontaneous speech. This mirrors the patterns observed for F1, but once again contrasts with the moan-mown findings, where the oldest generation maintained a full distinction across both speech contexts.

Figure 9.

Predicted GAMM F2 Trajectories (mW&F): sole and soul by Vowel, Generation, Gender, and Style

4.5. Description of Full Vowel Trajectories

Model predictions for all four vowels have been averaged and plotted to contextualize their relative positions within the wider vowel space (Figure 10). F1 and F2 measurements for fleece, goose, trap, thought, and lot monophthongs, taken at 50 percent of total duration, serve as a means of contextualizing the relative position of these vowels in the wider vowel space. When F1 and F2 are observed simultaneously across generations, the average GAMM trajectories confirm a change in apparent time toward a moan-mown merger. Both the old and middle generations maintain a distinction, while the youngest generation has an ongoing merger, with some speakers potentially retaining a minor distinction via F2. The merged goat target for the youngest group looks to approximate [ɐɪ ~ ɐʏ], beginning in a near-open central position and ending in a more peripheral/fronted position in word list speech than in spontaneous speech. Gender differences are overall quite minor, however.

Figure 10.

Average Predicted GAMM Trajectories for moan, mown, sole, and soul Vowels by Generation, Gender, and Style

For the oldest generation, moan begins in a close-mid central position and progresses to a close back position, neatly mirroring traditional [ʊu] variants found earlier in Norfolk (Trudgill 1974) and parts of Suffolk (Kökeritz 1932; Orton et al. 1962-1971). By comparison, the moan trajectory for the middle generation begins in a less raised position and, for the females, ends further forward in a near-close central position behind fleece. The males have shorter, less dynamic trajectories compared to the old females, which finish in a less peripheral position.

Overall, mown trajectories differ vastly across all generations. The oldest generation’s vowel approximates trap at the onset, bearing striking resemblance to the traditional Suffolk variant [aʊ] of the early twentieth century, rather than the [ʌu] variant heard in Norfolk and other parts of southern Suffolk (Kökeritz 1932; Trudgill 1974; Trudgill & Foxcroft 1978). This variant is more dynamic in word list speech than in spontaneous speech, traversing a greater portion of the vowel space and ending in a higher, more retracted position. The mown trajectories for the middle generation indicate a more centralized onset [ɐʊ], which represents a move away from the conservative form. For the males in this generation, mown is less peripheral/raised in its end position than for females, with overall shorter, less dynamic trajectories. Interestingly, despite the observed changes in the relative start/end positions of the vowel trajectories for different generations, their shapes remain somewhat stable. Similar findings are reported by Stanley, Renwick, Kuiper, and Olsen (2021) in their GAMMs-led evaluation concerning cross-generational changes to the back vowels in Southern American English.

Turning to the predicted trajectories for sole and soul (excluding word list speech for the oldest generation for now), onset positions are reasonably stable across generations and styles, beginning in an open-mid or back position that approximates lot/thought. Both vowels move toward more close, retracted positions regardless of generation, gender, or speech context, resulting in [ɔo ~ ɔʊ ~ ɒ:ʷ]-like trajectories. Minor differences appear in trajectory length and offglide endpoints, which vary by style: word list trajectories are generally longer, while middle-generation females show notably shorter trajectories in spontaneous speech.

When examining both F1 and F2 simultaneously, it is clear that the older group maintains a sole-soul distinction in word list speech, which disappears in spontaneous speech. All other contexts show substantial overlap between the two vowels, unlike moan-mown, where a distinction is retained. However, the sole–soul contrast is not phonetically identical to moan–mown for the contrast-maintaining group. This is driven by the outlying sole trajectory in the oldest generation’s word list speech, which begins in a close centralized position ahead of foot, much like moan. However, compared to moan, sole shifts into a more retracted position, likely due to allophonic conditioning from coda-/l/. Although sole diverges from earlier transcriptions ² —for example, Ellis <óoɐ> (~[o:ə]), Wright <ū> (~ [ʊ]), Kökeritz <ʊ:> (~[ʊ:]), and the SED [ʌu]—the Lowestoft vowel is not worlds apart in some senses, as will be discussed in Section 6.2.

Overall, the quality of the observed merged goat and goal forms in the youngest generation diverges from what is reported in south Suffolk (Trudgill & Foxcroft 1978)—a consistent intermediate form [ɵu] across all moan and mown items, except in coda-/l/ environments, where [ʌu] is reported—but nevertheless, a goat-goal split is still attested.

5. Perception Results

After classification into categories, two participants were labeled as ‘merged’ in speech production, eleven as ‘transitional,’ and seventeen as ‘distinct’ with respect to moan/mown classes (Table 2). A clear divide emerges, strongly correlated with generation. The group with a merger in speech production is exclusively formed of young participants, while the transitional group comprises the remaining young participants plus three middle-generation participants. All old participants retain a distinction perceptually.

Table 2.

Summary of Speech Production Categories Used for moan-mown ID Task

Merger status in speech production	Old	Middle	Young	Total
Distinct	10	7	0	17
Transitional	0	3	8	11
Merged	0	0	2	2

To assess whether listeners perceive categorical differences between moan and mown, or whether the two diphthongs share an overlapping target in the auditory space, a generalized linear regression model predicting category boundary width was built. The best fit model is given in Table 3 and includes both merger status in speech production (‘merged,’ ‘transitional,’ or ‘distinct’) and minimal pair condition as fixed effects. Collinearity between generation and merger status in speech production (merger status in speech production vif = 5.77, generation vif = 5.79) led to generation being removed from the model. Although the inclusion of minimal pair condition as a predictor improved overall fit, it did not affect the outcome variable (t = −0.95, p = .349).

Table 3.

Mixed Effects Linear Regression Predicting Boundary Width for moan-mown

Predictors	Est.	SE.	t	p
(Intercept)	1.88	0.43	4.42	<.001
Merger status in production: Transitional	2.59	0.62	4.21	<.001
Merger status in production: Merged	4.28	1.19	3.60	<.001
Minimal pair: Toad-towed	−0.34	0.36	−0.95	.349

The effect of merger status in speech production on boundary width is illustrated via the raw boundary width scores for the identification task, shown in Figure 11. The plot suggests that production largely matches perception for participants with a moan-mown distinction in production. The median boundary width (indicated by the horizontal black line) traverses 1.5 steps along the 7-point vowel continuum, meaning the perceptual distinction between the two vowels is narrowly defined. The model coefficients indicate that this group’s boundary width is, on average, 2.59 less than the group with a transitional moan-mown merger in speech production (t = 4.21, p < .001). Speakers with a full distinction in production therefore more closely approximate categorical perception of the moan and mown vowels. Estimated boundary width is greatest for participants who have a moan-mown merger in production which is, on average, 4.28 greater than the participant group with a distinction in production (t = 3.60, p < .001). The two participants with a merger in speech production are shown to attain the maximum boundary width score of 6 across both conditions. This means all continuum steps were classified as belonging to a single phoneme category, indicating absence of a category boundary and likely perceptual merger of moan and mown.

Figure 11.

Boundary Width for moan and mown by Merger Status in Speech Production

Although the median boundary width (6) for merged and transitional groups is identical, transitional participants show considerably more variation in how consistently they categorize the boundary between moan and mown, with boundaries spanning a wide range of values on the continuum (0.8-6). To further contextualize these findings: of the eleven transitional participants, only four had categorical perception of both toed–towed and doe–dough conditions, with the remainder varying or not exhibiting categorical perception. These results suggest that participants who produce a variable merger or imperceptible distinction in production generally experience greater uncertainty in their perceptual boundaries, as expected.

When the perception results are broken down by speaker and generation (Table 4), and combined with boundary width scores averaged across conditions, a generational divide emerges once more. The highlighted cells contain the seven speakers who exhibit categorical perception across both conditions (boundary width = 6). Six of these belong to the youngest generation, while the remaining young speakers all have an average boundary width of 4, indicating either inconsistency across conditions or near categorical perception. Only members of the middle and old generations have an average width of three or less, pointing to more stable categorical perception.

Table 4.

Speech Production and Perception Categories by Generation and Participant: Highlighted Cells Represent Categorical Perception

Participant	Generation	Production	Average boundary width
YF1	Young	Transitional	6
YF2	Young	Transitional	6
YF3	Young	Merged	6
YF4	Young	Transitional	4
YF5	Young	Merged	6
YM1	Young	Transitional	6
YM2	Young	Transitional	4
YM3	Young	Transitional	4
YM4	Young	Transitional	6
YM5	Young	Transitional	4
MF1	Middle	Transitional	6
MF2	Middle	Transitional	2
MF3	Middle	Distinct	3
MF4	Middle	Distinct	2
MF5	Middle	Distinct	0
MM1	Middle	Transitional	1
MM2	Middle	Distinct	0
MM3	Middle	Distinct	2
MM4	Middle	Distinct	2
MM5	Middle	Distinct	2
OF1	Old	Distinct	2
OF2	Old	Distinct	3
OF3	Old	Distinct	1
OF4	Old	Distinct	2
OF5	Old	Distinct	0
OM1	Old	Distinct	2
OM2	Old	Distinct	0
OM3	Old	Distinct	3
OM4	Old	Distinct	1
OM5	Old	Distinct	3

6. Discussion

6.1. Overall Findings

The findings indicate that working-class Lowestoft English is advancing toward a full phonemic merger of moan and mown in both speech perception and production. Dynamic vowel trajectories modeled via GAMMs show that young speakers exhibit either complete merger or close approximation of the two vowels in production, maintained by offglide fronting. Remarkably, the oldest generation produce the ‘original’ Suffolk mown variant [aʊ] described by Kökeritz (1932:67) in the early twentieth century, indicating a striking degree of spectral change for mown over three generations. The emerging goat vowel is a novel variant [ɐɪ ~ ɐʏ] that diverges from the quality of both original vowels, meaning both have been affected, as per Britain’s (2005) Fenland findings. This contrasts with Ferragne and Pellegrino’s (2010:13) description of Lowestoft’s goat vowel as having a back starting point and juxtaposes Trudgill and Foxcroft’s (1978:77) findings showing moan moving toward mown [ʌʊ] in Lowestoft.

Goat variants with raised onsets and extremely fronted offsets [ə̟ʏ ~ əʏ ~ əɪ] have already been documented in the likes of Reading, Milton Keynes, Aylesbury, and Essex (Williams & Kerswill 1999; Altendorf 2003; Kerswill & Williams 2005; Lewis 2022) and are argued to reflect the shared adoption of a novel variant, rather than dialect leveling (Cheshire, Fox, Kerswill & Torgersen 2008:463). Importantly, all of the southeastern locations mentioned here are contiguous with Greater London and, if not, are contiguous to a variety that is. The county of Essex is contiguous with Suffolk, Lowestoft’s administrative county, thus wave-based dialect diffusion by contact—as was earlier observed in southern East Anglia (Trudgill & Foxcroft 1978)—remains a plausible explanation for this shared adoption and ostensible convergence.

The corresponding perception data facilitated a richer evaluation of the merger’s progress. When categorizing moan-mown stimuli along a seven-step continuum, participants with distinct moan-mown vowels in production showed categorical perception of the distinction, evidencing a stable perception-production relationship, despite exposure to the incoming community norm of merger. Two young speakers—who perhaps, not coincidentally, are young females, despite gender not improving model fit—showed the opposite pattern: full merger in both perception and production.

If we take the timeline for mergers by approximation to be three to four generations (Labov 1994:323), it is plausible that the next generation will be the first to see the merger complete in both domains. The remaining eight young speakers did not have a merger in speech production and were instead labeled as transitional—confidence intervals for F2 were large for the offglides of both vowels across speech styles, with moan produced further forward in the vowel space than mown in word list speech. Their boundary width scores (4-6) suggest perceptual ambiguity that is consistent with near mergers, that is, they produce differences that they cannot perceive (Labov, Yaeger & Steiner 1972). These data support the argument that mergers occur in perception before production (Bowie 2001; Baranowski 2007; Pinget 2015). The phonetic differentiation of two phonemes via a slight difference in peripherality realized via the offglide has also been identified in other studies of East Anglian near mergers concerning diphthongs, such as moan-goose and near-square in Norwich and south Suffolk, respectively (Labov, Yaeger & Steiner 1972; Trudgill & Foxcroft 1978).

The oldest generation’s lack of participation in the merger may reflect continued moan-goose overlap in the Lowestoft community—a pattern which, according to Butcher’s (2021) Lowestoft findings, is limited to the oldest speakers. Such a merger structurally constrains the systems of those who have it, as mergers, logically, cannot progress while affected phonemes are involved in other changes (Labov, Ash & Boberg 2006:65). By contrast, the middle generation—particularly females—are strong candidates for having led this change, as the first group to alter both the phonetic realization of the vowels and produce transitional merger judgments. A female-led change by this generation in Lowestoft follows well-established principles of sound change that predict gender-based stratification (Labov 1994), specifically Labov’s Principle II, which suggests that women tend to lead changes from below, for example, merger by approximation.

While we cannot entirely rule out that the older generation initiated the change (since no acoustic data exists for speakers older than those in the current ‘oldest’ group), several points support this interpretation. Kökeritz’s (1932:192) descriptions from the early 1930s indicate that even in southern Suffolk—where the distinction was lost earlier—Middle English /ǫ/ was realized as <ʊ:> ([ʊ:]), and /ou/ as diphthongs ranging from <ɑʊ> to <oʊ> ([aʊ ~ oʊ]). These notations closely resemble realizations produced by the oldest Lowestoft speakers in this study, despite being geographically and generationally removed, suggesting retention of an older system. More broadly, the onset of other major sound changes over the past century has also been reported to begin with this generation, that is, those born in the late 1950s through the 1970s. Renwick, Stanley, Forrest, and Glass (2023) describe a ‘Gen X cliff’ in their work on Southern Vowel Shift reversal, which echoes earlier work by Dodsworth and Benton (2019) concerning southern vowel features.

Finally, the realization of moan and mown in pre-/l/ environments in Lowestoft reflects a broader change. Studies have shown that /l/ inhibits vowel fronting (e.g., Hall-Lew 2005; Johnson & Britain 2007; Fridland 2008; Holmes-Elliott 2015), resulting in a goat–goal split. The quality observed for merged goal in Lowestoft resembles [ɒʊʷ ~ ɒɣ] variants reported in London (Tollfree 1999:165) and [ɔ̝ʊ] variants in Milton Keynes and Reading (Williams & Kerswill 1999:143). Halfacre’s (2022) GAMMs analysis shows goal is becoming more dynamic even for southern RP speakers, due to increasing backness. However, /l/ darkening arrived in East Anglia only recently. Trudgill (2021:207) reports clear /l/ in all positions well into the twentieth century, and even by the late 1970s, vocalization was absent in Lowestoft (Trudgill 2021:209). In the current data, vocalization remains uncommon and mostly limited to the youngest generation, where darker /l/ variants are now widespread. ³

6.2. Mechanisms of Merger

The GAMM predictions strongly support merger by approximation for moan and mown and the gradual diffusion of a merger into this area of stable resistance. Evidence includes the resulting vowel’s phonetic dissimilarity from both original phonemes (Herold 1990) and the conceivable presence of near merger among younger speakers, which is only possible in situations of gradient merger (Labov, Yaeger & Steiner 1972). The findings align with the three to four generation timespan associated with mergers by approximation (Labov 1994:323), also paralleling earlier findings from Trudgill and Foxcroft (1978) in northern East Anglia—the sub-region Lowestoft is classified as belonging to both linguistically and culturally (Trudgill 2021:58).

There remains one issue preventing a stable conclusion regarding the merger’s mechanism. The oldest generation uphold a consistent moan-mown distinction across both speech styles, but sole-soul contrast is only upheld in word list speech, whereas the middle group upholds a moan-mown distinction across both styles, which is then collapsed in all pre-/l/ contexts. This gives the impression that sole and soul may have merged ahead of moan and mown, potentially suggesting two concurrent mechanisms: phonological transfer for sole-soul, evidenced by the lack of intermediate forms, and approximation for moan-mown. The simultaneous occurrence of two mechanisms is rare, with empirical cases limited to Dinkin’s (2016) cot-caught merger study in Upstate New York, Mellesmoen’s (2018) /æ/–/eɪ/ merger study in British Columbia English, and Thomas’ (2019:193) discussion of the bot-bought merger for Ohio speakers. However, if the sole-soul merger was truly driven by phonological transfer, the Lowestoft pattern would potentially be consistent across speech styles, reflecting the stable phonological restructuring seen with merger by lexical transfer (Trudgill & Foxcroft 1978). This is purely speculative, though, as knowledge of phonological transfer remains limited.

If we are to account for the observed style shifting in the word list speech of the oldest generation of Lowestoft speakers, it is important to consider that the seemingly divergent phonetic realization of sole is not without phonetic basis. Traditional moan variants in Lowestoft routinely exhibit [ʊ]-like onsets, and the sole vowel in this context shows a similar starting position. This could hint that the sole variant observed in word list speech reflects a conservative retention of the older [ʊu]-like moan diphthong in monitored, more formal speech—a pattern typical of a change in progress, with speakers exaggerating community norms in formal settings (Labov 1966).

Nonetheless, regardless of this stylistic variation, both sole and soul are susceptible to the influence of coda /l/, with their offglides taking a retracted position. What is therefore particularly interesting about the Lowestoft case is that the phonological merger of moan and mown in non-pre-/l/ contexts does not appear to be a prerequisite for the application of allophonic rules, instead yielding moan-goal and mown-soul splits for the oldest generation.

6.3. The Diffusion of the Merger

The direct diffusion of underlying structures between donor and recipient dialects is not expected. Labov (2007:349) maintains that only surface, or “observable,” elements of a language are subject to immediate change at the point of a sound change’s diffusion into a community. The merger in Lowestoft illustrates this well, as the phonetics of the two merging phonemes are being shaped by supra-local phonological systems beyond the community, which are also undergoing phonetic change (goat fronting), without leading to reorganization of underlying local structures. This latter point is reiterated by the perception findings, wherein 80 percent of the middle generation retained categorical perception of the two vowels, despite considerable changes to the phonetics of their moan-mown contrast. Arguably, therefore, the phonetic changes affecting the two vowels in this group do not reflect the point in a merger where production is catching up with a merger in perception.

The application of fronting to both moan and mown—usually affecting the goat phoneme in donor dialects but adapted to the two-phoneme scenario in the recipient (Lowestoft) system—suggests that superficial (phonetic) elements of diffusion affected the two vowels before a merger took place. That is, the merger appears to have followed a process of incremental sound change. Dinkin (2011:341) discusses how changes can occur in the ‘direction’ of merger, rather than a merger itself being the object of diffusion. In the current case, simultaneous fronting and centering of moan and mown by the middle generation—potential initiators of change—reflects a change in the ‘direction’ of merger, which is moving into the community (stage one) and enabling merger in later generations (stage two). Note that change in the ‘direction’ of merger does not imply literal approximation of the two phonemes, as the phonetic changes in stage one actually caused divergence of moan and mown along the front/back dimension. Stage one can justifiably be viewed as change in the direction of merger, as it serves to align the two vowels more closely—both perceptually and phonetically—with merged goat forms in external varieties already subject to fronting. Dinkin (2011) points to a similar scenario in Upstate New York, where gradual cot-backing is occurring ahead of the cot-caught merger, reflecting a non-teleological sound change. Such an interpretation might also explain similar patterns observed by Britain (2005) in the Fenlands, where mown’s fronting alongside retention of the distinction could reflect a change in the direction of merger, that is, stage one.

6.4. Why Might mown Lag Behind?

Fronting may have first affected Lowestoft English independently of contact, instead reflecting system-internal processes. Labov (1994:116) observes that goose fronting often precedes goat fronting, following Principle III of linguistic change: back vowels move to the front. This pattern is attested in other British varieties (Williams & Kerswill 1999; Kerswill & Williams 2005; Baranowski 2017). While Lowestoft looks to follow the goose-fronting trend (Butcher 2021), research into Principle III in action concerns only varieties that possess a merged goat phoneme. If goat fronting in Lowestoft were purely system-internal, both moan and mown would exhibit similar increases in frontness, reflecting their treatment as a single vowel class. However, the differing degrees of phonetic change, alongside retention of a distinction in both production and perception, suggests otherwise.

A good explanation for these patterns is grounded in moan’s phonetic composition. Trudgill and Foxcroft (1978:78) describe moan’s quality as “conspicuous,” noting it as one of the first features noticed by visitors from the Home Counties to East Anglia (Trudgill 1974:100). Traditionally realized as [ʊu], moan is more deviant than mown from more standard goat pronunciations, which generally have less close onsets. It was on this basis that Trudgill and Foxcroft (1978) anticipated the merger would continue in the direction of mown. The more robust fronting of moan compared to mown may reflect conscious adaptation of the more socially salient variant, potentially overshooting the RP-like [əʊ] offglide target when shifting from Lowestoft [ʊu], which is not observed for mown.

Hypercorrection is known to occur once one phoneme in a merger becomes subject to negative social evaluation, for example, the near-square merger in New Zealand English (Maclagan & Gordon 1996; Maclagan & Hay 2007). This interpretation is further supported by the Lowestoft data, wherein a speaker from the middle generation told me that “. . . moon and moan is something I have problems with . . .,” before spontaneously breaking into a rendition of Moon River, enunciating moon as [mʊun]. Eckert and Labov (2017:484) argue that mergers do not carry social meaning per se, with commentary typically focused on the pronunciation of just one member of the pair. This exact scenario is highlighted here by our Lowestoft speaker, where the high-back realizations of moan appear to be the focus of community-driven change.

However, it is plausible that the answer to why moan fronts ahead of mown, or rather why mown lags behind, is even simpler: reversal of the (near) merger between moan and goose, argued by Butcher (2021) and Trudgill (1988) to have triggered the moan-mown merger, means moan was already involved in change in the community, whereas MOWN was not. The fronting of moan ahead of mown represents a situation of localized divergence amid broader, long-term convergence, as was observed by Britain (2005) in the Fens. Such ‘compromised’ outcomes align varieties more closely with external dialects but in doing so produce a local system where allophones or pairs of a split diverge.

6.5. Social Impetus for Change

A final aim of this paper is to explore the social factors that determine how mergers diffuse into areas of stable resistance. The gradual progression of the Lowestoft merger aligns with Dinkin’s (2011) findings for the cot-caught merger in Upstate New York in terms of its mechanism and incremental trajectory of change. However, unlike Dinkin’s study, which lacked a clear social impetus, the Lowestoft merger may be linked to societal change, tied to population movements. The decline of the fishing industry (culminating in the 1980s), alongside mass closures in related industries, drastically altered Lowestoft’s economic and social complexion. By 1982, 16 percent of the working population in Waveney—the town’s administrative ward—received unemployment benefits (Waveney District Council 1984:26), and unemployment nearly doubled between 1990 and 1995 (Dyson & Mann 1995:14). Despite economic challenges, Lowestoft’s population grew by 26 percent from 1961 to 2011, contrasting with modest increases in larger regional centers like Norwich (2 percent) and Ipswich (3 percent), and a 2 percent decline in Great Yarmouth (Long 2005). This population growth is difficult to attribute to native population expansion.

Towns and cities that follow an ‘ageing manual labor’ trajectory—that is, those that move from a blue-collar population toward a large economically inactive, retiree population—are susceptible to declining young populations who favor out-migration (Patias, Rowe & Arribas-Bel 2022). The paradox of rising population amid economic downturn can instead be attributed to substantial inward migration, specifically from retirees and older second-home owners from higher-priced regions like London and the Southeast (Waveney District Council 1991:1). The Lowestoft area is particularly sensitive to the scale of net out-movement from London and other pressurized centers (Fielding 1992), and by 1980 Waveney had one of the highest rates of second home ownership in England (Long 2005:123). Outbound movements in Lowestoft are ostensibly counterbalanced (Gordon, Champion, McDonald & Whitehead 2017:66), thus native population decline likely led to decreased exposure to local variants; a key factor in the maintenance of mergers in communities undergoing change (Maguire 2007).

It is the middle generation in this study who were born into and lived through this transitional period. At the tail-end of the industry’s demise in the 1980s, participants belonging to this generation would have been between the ages of five and twenty-five. They would have grown up with the moan-mown distinction present around them—a contrast they retain themselves—but were the first generation to have potentially left Lowestoft for work or education, or to have stayed and faced significant in-person exposure to speakers with a single goat phoneme for the first time. The generation’s shift toward more centralized and fronted moan/mown variants reflects this new contact. Wealthy retirees to the coast, many of whom were likely RP speakers, will have produced goat with a central nucleus and higher F2 value in the offglide, that is, [əʊ]. It is argued that RP was more socially relevant then than it is today (Lindsey 2019:5), and Trudgill and Foxcroft (1978) reported similar trends of RP convergence in northern East Anglia, noting intermediate moan forms ([ɵʊ]). In this study, the middle generation in Lowestoft produced comparable central onsets ([əʉ ~ ɵʉ]), with the youngest generation acquiring a new target.

Demographic changes that lead to high levels of dialect contact are typically associated with abrupt mergers, as in Herold’s (1990) account of eastern Pennsylvania and Johnson’s (2007) observations on the Massachusetts-Rhode Island border. Yet, mergers by approximation—like Lowestoft’s—more commonly occur in low dialect contact situations (Herold 1990). The Lowestoft case, with its mixture of gradual progression and obvious social change, resists easy categorization into existing scenarios of either abrupt merger with social impetus (Herold 1990; Johnson 2007) or gradual merger without it (Dinkin 2011). The interplay of linguistic and social factors must therefore determine whether mergers diffuse gradually or immediately on a case-by-case basis.

7. Conclusion

This study explored the moan-mown merger in working-class Lowestoft English, focusing on how the merger is diffusing into an area of stable resistance and the extent to which the variety is assimilating/dissimilating to external varieties. Findings showed that the merger is progressing gradually via approximation, with the youngest generation nearing full merger in both perception and production, and the middle generation exhibiting phonetic changes without phonological reorganization. The data illustrated how mergers can evolve incrementally when successive generations adopt different phonetic targets (RP versus shared adoption of a southeast variant), shaped by changing sociolinguistic conditions. The application of goat fronting to the mismatched two-phoneme moan-mown system illuminates further how external phonological systems can shape phonetic outcomes in recipient dialects without altering their underlying phonological structure or perceptual categories. While these findings indicate that this particular feature may be assimilating toward other Southern British varieties, this study falls short of identifying the true extent to which Lowestoft is undergoing broader linguistic assimilation. A more comprehensive account of additional variables across the linguistic system and different socioeconomic groups during Lowestoft’s transitional phase is needed to understand the full extent of resistance.

The lagging behind of mown in the merging process shows how social salience and ongoing changes (e.g., Moan-goose [near] merger) have the potential to influence merger processes, leading to divergence in the local system amid long-term convergence, mirroring Britain’s (2005) Fenland findings. The divergence in vowel backness observed before (near) merger in the youngest generation not only underscores the methodological value of dynamic vowel analysis but also perhaps indicates that current definitions of merger by approximation may oversimplify the process, assuming a straightforward, linear convergence of phonemes between generations, marked by a gradual movement of the changing phoneme(s) into a shared space.

Finally, Lowestoft was found to diverge from other socially triggered mergers in both its mechanism and timescale (e.g., Herold 1990; Johnson 2007). However, more data from a wider range of communities and merger types are needed to clarify the motivations behind gradual or immediate merger into areas of stable resistance, which remain underexamined. Specifically, future research must investigate the moan-mown merger’s diffusion into areas of East Anglia’s northern zone that lack Lowestoft’s cultural and linguistic precarity, while this is still possible. This would elucidate whether the Lowestoft pattern is a direct product of its unique socio-cultural, historical, and linguistic situation, or a more general underreported pattern for mergers diffusing into areas of stable resistance.