LANGUAGE CONTEXT: ON FRANCOPROVENÇAL

Francoprovençal is the label given by linguists to a fragmented Romance language spoken transnationally across French, Italian, and Swiss borders. While the language was spoken in the region long before the formation of modern nation-state boundaries, there has traditionally been significant disagreement over such fundamental issues as the linguistic criteria that have been adopted in demarcating the Francoprovençal-speaking region, the limits of these borders, and the glottonym that has been adopted to refer to the language (for an overview, see Kasstan & Nagy Reference Kasstan, Müller and Nagy2018).

Although a Francoprovençal literary heritage emerged in the thirteenth century, the folk-linguistic perception among speakers has always been one of fragmented oral vernacular tradition. Accordingly, there is no pan-regional linguistic identity comparable to, for example, Breton and bretonnant identity in Brittany: speakers have never embraced membership in a common linguistic system (Matthey & Meune Reference Matthey and Meune2012:108), and will instead refer to their language affectionately as ‘patois’ rather than ‘Francoprovençal’. The status accorded to the language varies widely: Francoprovençal in the Aosta Valley (northern Italy) is afforded provision in schools under Law 482, having long been recognized as a regional language. Conversely, in France, Francoprovençal was not officially recognized until 1999 (Cerquiglini Reference Cerquiglini1999:6); it is accorded no official status or suitable provision, and speakers have been encouraged (explicitly and implicitly) to abandon Francoprovençal over generations. The concomitant effects of Francoprovençal's history and variable status have contributed to its current state as a severely endangered language. Measures of ethnolinguistic vitality vary significantly across national borders (Zulato, Kasstan, & Nagy Reference Zulato, Kasstan and Nagy2018:13), and speakers have long been dwindling as language shift progresses. Francoprovençal is no longer transmitted inter-generationally in France, where French has long been entrenched in the critical domain (the home), and the decline is well advanced in Switzerland, too, in spite of local efforts to maintain traditional practices.

For the most part, then, Francoprovençal has retreated almost exclusively to the private sphere: it is mostly used by the eldest speakers inside the family (if at all), and exclusively in informal contexts. The overall picture is thus one of gradual language death (Campbell & Muntzel Reference Campbell and Muntzel1989:182): a context of long-term language contact, typically characterized by unstable diglossia, where there is a contraction in the domains of usage available to the threatened variety, leading to (a) periods of bilingualism, (b) a decrease in linguistic competence, and therefore (c) an identifiable continuum of speakers with variable proficiency. In Dorian's seminal work, this continuum ranges from fluent speakers to ‘semi-speakers’ (Reference Dorian1981:114). Semi-speakers have been referred to as ‘harbingers of language death’ (Jaffe Reference Jaffe2015:23) as they represent both a downward trajectory in language vitality at the level of the speech community, and a general pattern of so-called linguistic ‘decay’ at the level of the individual. This general pattern is so commonly reported that Dorian (Reference Dorian1977:29) has proposed a number of semi-speaker universals, chief among them their apparent reduction of stylistic options.

STYLE VARIATION IN LANGUAGE OBSOLESCENCE

As Mougeon & Beniak (Reference Mougeon, Beniak and Dorian1989) note, most sociolinguistic research investigating the linguistic effects of language obsolescence has focused on simplification in the grammar and style reduction.

Prototypically, the model predicts that, as the threatened variety progressively loses ground to the superordinate contact variety, there is a subsequent loss of more or less formal language styles associated with those domains, leaving speakers to control only casual styles in the most intimate domains.Footnote ² This pattern of stylistic shrinkage has been demonstrated cross-linguistically and at most levels of linguistic description. For instance, Dressler & Wodak-Leodolter (Reference Dressler and Wodak-Leodolter1977) report stylistic shrinkage in Breton lexicon. Breton has long been subordinated to French in most formal domains, and so style differentiation does not, they argue, ‘seem to be important anymore… of all Breton styles, only casual speech remains’ (Reference Dressler and Wodak-Leodolter1977:37). Holloway (Reference Holloway1997) adopts the label stylistic shrinkage to describe the contraction in morphosyntactic variation among speakers of Brule Spanish (Louisiana). In his sample, some speakers do not distinguish between variable pro-forms usted and tú to mark formal and informal distinctions: he cites examples of speakers employing informal second-person singular inflectional morphology regardless of speech style. Holloway argues that ‘the formal registers are so impoverished that Brule speakers for the most part have little underlying awareness of the distinctions which speakers of many other Spanish dialects use to distinguish formal and informal speech’ (Reference Holloway1997:149). Similarly, Gal (Reference Gal and Schiffrin1984) illustrates stylistic shrinkage in the phonological systems of her Oberwart Hungarian-German bilinguals. While older speakers in her sample consistently style-shift in traditional sociolinguistic interviews for three Hungarian variables, her younger speakers acquire the variants for these linguistic variables, but not the associated interactional constraints, and so use the variants interchangeably irrespective of speech style (Gal Reference Gal and Schiffrin1984:297). Gal reasons that the younger speakers maintain different social-network structures to the older speakers in that they are predominantly German speaking, and so there are far fewer occasions to employ Hungarian. The absence of normative pressure is also highlighted.

For the purposes of the present article, the absence of overt normative infrastructure and its implications for style shifting in language obsolescence is noteworthy. Stanford & Preston (Reference Stanford, Preston, Stanford and Preston2009:8) have also highlighted that traditional variationist approaches to lesser-studied languages face the challenge posed by a lack of established normative pressures. In the same volume, Clarke's (Reference Clarke, Stanford and Preston2009:120) data, taken from research on Sheshatshiu-Innu-aimun (Canada), are characterized by a void of stylistic variation for fourteen of the eighteen variables studied, which she suggests may result from ‘the absence of… a prestige standard [or] orthographical norm’. However, the absence of a standard does not entail the absence of style variation, as work on variation in underdocumented languages has tended to assume (Skilton Reference Skilton2017:108). That Clarke identifies four variables operating like sociolinguistic markers (per Labov Reference Labov1972:238) is not inconsiderable, but no explanation is offered to account for this variation: ‘why some features… should be more subject to style shifting is not fully clear’ (2009:120). Similarly, Rau, Chang, & Dong (Reference Rau, Chang, Dong, Stanford and Preston2009:274) identify style shifting between wordlist and narrative styles for word-final (ay) and (aw) among four speakers of Yami (Orchid Island, Taiwan), another unstandardized endangered language. Their data show that three speakers produce the conservative (unraised) variant in wordlists to categorical levels, and variable raising in the more informal narrative style (cf. Lai & Gooden Reference Lai and Gooden2018:134). While the sample is small, in this case it appears that speakers orient to the most conservative variant when most attention is paid to speech. Skilton's (Reference Skilton2017) qualitative work on obsolescent Máíhɨ̃ki (Peru) also demonstrates that style shifting is still present in the absence of a standard norm, and Dorian has already proposed that the absence of such norms may encourage rather than inhibit ‘the persistence of conspicuous interspeaker and intraspeaker variation’ (Reference Dorian1994a:694). Nevertheless, these examples notwithstanding, it is clear that, in the vast majority of gradual-death case studies, stylistic shrinkage has been commonly documented. The contraction of style repertoires has been linked to external factors such as domain loss and to the absence of normative pressures.

As Dorian (Reference Dorian, Biber and Finegan1994c:331) suggests, such characterizations of monostylistic behavior challenge the principle that ‘there are no single-style speakers’ (Labov Reference Labov1971:112).Footnote ³ Labov's principles of linguistic methodology are predicated on this postulation, from which he derives three further principles (Table 1) that now constitute standard assumptions in variationist fieldwork, be it in sites closer to or far from New York City.Footnote ⁴

Table 1.

Principles of linguistic methodology (Labov Reference Labov1971).

Under his attention-to-speech model of style shifting, increased focus on speech is taken to indicate a more formal context where speakers shift away from the vernacular (or ‘casual’ style). However, Labov's model is not without its critics, many of whom have instead prioritized the role of stances, activities, and interactions in style shifting, seeing the process as much more agentive and less unidimensional than his model implies (Meyerhoff Reference Meyerhoff2016). Among the early agentive approaches, Bell (Reference Bell1984) was careful to note the implications of the loss of style repertoires. Under audience design, Bell posits the style axiom, which holds that intraspeaker (stylistic) variation ‘derives from and echoes’ interspeaker (social) variation (Reference Bell1984:151), and, therefore, ‘If a variable has no interspeaker variation, it will have no intraspeaker variation (Reference Bell1984:158). In gradual language death, Bell (Reference Bell1984:158) adds that loss of style variation (what has been termed stylistic shrinkage) is a response to and result of the contracted range of social variation resulting from domain loss. His interpretation of domain loss thus also implies a reduction in occasions, contexts, and types of speakers with which to interact. The below discussion departs from this view that loss of domains entails monostylism.

It is also noteworthy that the cline of language obsolescence is not necessarily terminal. In language-revitalization contexts, there can be a ‘reshift’ (Fitzgerald Reference Fitzgerald2017:e288) in that communities commit to reclaiming and reacquiring the indigenous variety. Very often, revitalization groups acknowledge the pitfalls of absent standardized or normative practices for their own broader goals associated with, for example, more favorable language policies, or reversing the decline in speaker numbers. Therefore, revitalization strategies might, in the process of reversing language shift (a macro-level change), also reverse processes such as stylistic shrinkage (a micro-level change), as they promote the use of the threatened variety in a greater number of domains. While the expansion of the sociostylistic range through language revitalization remains understudied, recent work on new speakers has considered the implications for broader questions related to language variation and identity construction in severe language endangerment.

LANGUAGE REVITALIZATION AND NEW SPEAKERS

The ‘new speaker’ label is used in work on language endangerment and revitalization to characterize the lived experiences and sociolinguistic realities of individuals who acquire a minority language not through traditional transmission contexts (e.g. home, family), but as adults through revitalization initiatives (O'Rourke et al. Reference O'Rourke, Pujolar and Ramallo2015). The adoption of this label breaks with a tradition in sociolinguistics that prioritizes the study of ‘native’ speakers—seen as the most traditional or authentic (and thus legitimate) representatives of a speech community—at the expense of more recent arrivals, whose language production has been viewed in terms of deficit (these speakers are typically labelled ‘L2’ or ‘learner’, etc.). Such essentialist perspectives on language and identity follow from an ideology of language (and fluency) as inherited by birth (Rampton Reference Rampton1990). Essentialist beliefs are particularly common in endangered-language contexts, which are very often characterized by (a) dwindling numbers of older more fluent speakers, and (b) a preponderance of younger speakers with more limited proficiency.

New speakers have become the focus of a number of (predominantly qualitative) sociolinguistic studies in language endangerment. Kasstan (Reference Kasstan2017) highlights some general characteristics of new speakers based on the available evidence from threatened languages such as Breton. Broadly, they can be characterized as well-educated middle-class urbanites, who are highly politicized in the sense that they are often engaged in language revitalization initiatives. They are therefore in most cases far removed from more ‘traditional’ endangered-language speakers, who tend to be described as older, nonmobile, rural dwelling, and working class (e.g. in France, Pooley Reference Pooley2000:141–43). While the bulk of this work has focused on interaction-level analyses of speaker legitimacy and power relations in the community, far less attention has been paid to linguistic variation. This is surprising, given that new speakers are potentially important and influential arbiters in communities where no standard or prestige norm is established, or where ‘personal pattern variation’ (Dorian Reference Dorian1994a:634) abounds.

In language-revitalization initiatives, new speakers have significant input on and influence over emergent or fluctuating normative practices. For example, although Breton has been described as a clear-cut case of gradual death for some time, new speakers have been cultivated in Breton-medium (Diwan) schools in Britany now for several decades. This has given rise to a generation of néo-bretonnants who are said to employ a standardized variety of Breton far removed from the traditional norms associated with older speakers (Jones Reference Jones1998:323). Much of the qualitative sociolinguistic work on Breton has emphasized the social and linguistic disparity between ‘traditional’ and ‘new’ speakers. Adkins (Reference Adkins2013:66), for instance, describes how the former can view néo-breton as inferior or inauthentic, and Jones (Reference Jones1998:321) has suggested that néo-breton speakers will typically simplify complex Breton structures. However, this has been questioned in recent production studies that evidence little difference in quantitative terms distinguishing néo-bretonnants from older speakers in structured elicitation tasks (e.g. Kennard & Lahiri Reference Kennard and Lahiri2017). Nonetheless, the Breton case does illustrate that there are perceptible production differences between traditional speakers of endangered languages and emergent new speakers, and this linguistic differentiation has led to more recent quantitative work that has sought to investigate new-speaker variation in more naturalistic settings.

Nance, McLeod, O'Rourke, & Dunmore (Reference Nance, Wilson and Dunmore2016) present a quantitative analysis of word-final rhotic variation among Scottish Gaelic new speakers in Glasgow and Edinburgh. While the older ‘traditional’ Gaelic speakers approximated to palatalized rhotics in the study, some new speakers displayed substantial variation in their production of rhotics, in some cases preferring weakly rhotic or nonrhotic variants in spite of high fluency rates. These observations were supported by sociolinguistic interviews that illustrated that these speakers ‘preferred an ideal self that was more oriented towards a new-speaker model and considered a native-speaker target as inauthentic’ (Nance et al. Reference Nance, Wilson and Dunmore2016:185). Nance and colleagues show then that new speakers can produce linguistic variants that (a) differ from traditional norms, and (b) are socially constructive as markers of newspeakerness rather than traditional practice. Nance has further argued that Scottish Gaelic new-speaker variation in Glasgow is becoming enregistered, with speakers recognizing an emergent and discrete new-speaker variety, described as ‘distinctive and innovative within Gaelic-speaking circles’ (Reference Nance, Smith-Christmas, Murchadha, Hornsby and Moriarty2018:223). The Scottish Gaelic study shows that adult learners not only acquire sociolinguistic variation, but that they can also innovate features that can come to convey social functions without precedence in their respective speech communities. However, as little if any attention has so far been paid to style shifting among new speakers, it is not clear to what extent they are able to control for variation across language styles that are constrained by external factors, and how such variants (and associated constraints) are diffused. More evidence is therefore needed to establish the extent to which variation observed in language obsolescence is analogous to the systematic variability described in variationist studies on dominant languages. As language styles play a pivotal role in socially informed (Labovian) theories of linguistic variation and change, it is suggested here that style variation will provide a window on these processes in severe language endangerment, too.

VARIABLE PALATALIZATION OF LATERAL APPROXIMANTS

An investigation into style variation in language obsolescence formed part of the research design for a project on variation and change in Francoprovençal (Kasstan Reference Kasstan2015a). This article considers one dependent variable: the palatalization of /l/ in obstruent + lateral onset clusters /kl, ɡl, pl, bl, fl/.

Palatalization of laterals is an old and well-attested feature in Romance (e.g. Jänicke Reference Jänicke1997), and, in Francoprovençal, there is substantial variation both in terms of the clusters that palatalize and the surface variants that emerge (see Table 2).Footnote ⁵

Table 2.

Attested variants in C + /l/ onset clusters (adapted from Stich Reference Stich1998:47–50).

In Les monts du Lyonnais (France), early twentieth-century atlases (Gilliéron & Edmont Reference Gilliéron and Edmont1902–1910; Gardette Reference Gardette1950–1956) illustrate /l/ as palatalizing only in the velar + lateral clusters, where the median approximant [j] is attested, while [l] is produced in the labial + lateral clusters (see Table 3). This pattern holds across communes.

Table 3.

Palatalization in les monts du Lyonnais (adapted from Gardette Reference Gardette1950–1956).

In the Canton of Valais (Switzerland), the picture is comparatively more complex. Gilliéron's (Reference Gilliéron1880) atlas indicates that palatalization in all five obstruent + lateral clusters varies from commune to commune. Gilliéron grouped his data into two sets: variants in the /kl, ɡl, fl/ set (map 24),Footnote ⁶ and those in the /pl, bl/ set (map 25). Table 4 summarizes Gilliéron's observations from map 24, where four variants are identified.

Table 4.

(l) variants for /kl, ɡl, fl/ clusters in Valais (adapted from Gilliéron Reference Gilliéron1880).

This article concerns itself primarily with those sites identified as ‘eastern’ in Table 4, where two forms are attested: [ʎ] in velar clusters, and [l] in labial clusters. Conversely Jeanjacquet (Reference Jeanjacquet1931:39–40) describes the feature as variable (see Table 5).

Table 5.

Regional variation in Valais (after Jeanjacquet Reference Jeanjacquet1931:40).

Table 5 shows that, in eastern Valaisan communes, palatalization varies between [ʎ] ~ [l] in all clusters, contra Gilliéron, who reports palatalization in the velar clusters only. This indicates a possible change in progress in the 1930s, some fifty years after Gilliéron (Reference Gilliéron1880), whereby a categorical rule (palatalization before velar obstruents only) has become variable; optionality emerging out of formerly categorical rules is commonly attested in language obsolescence (e.g. Campbell & Muntzel Reference Campbell and Muntzel1989). Moreover, at the time of writing, Jeanjacquet concedes that palatalization was ‘tending to disappear’ (i.e. likely recessive) in the eastern varieties (Reference Jeanjacquet1931:40). The discussion below focuses on one eastern variety spoken in the commune of Savièse, which is motivated by two concerns. First, if Jeanjacquet's assessment concerning the progressive loss of palatalization in the East is accurate, then this may have implications for the broader discussion on stylistic shrinkage under investigation here. Second, evidence from local orthographic standardization in Savièse suggests that, since at least the 1960s, it has become convention to transcribe the lateral in obstruent + lateral clusters as <l>, for example, <cl, gl> etc. (Favre & Balet Reference Favre and Balet1960). This is significant, as the orthographic norm in Savièse is a phonetic-based spelling system, and so the use of <l> would indicate that palatalization is no longer present in the speech community, particularly when compared with other regions, such as Évolène, where, for example, <kly, gly> is found to mark /l/-palatalization (Barmaz & Pannatier Reference Barmaz and Pannatier2013:40).Footnote ⁷ In the intervening years, then, a change that was described as incipient in the 1930s may have gone to completion in Savièse by the 1960s, which would account for local spelling conventions. Some recent empirical evidence, however, does call this into question. Data collected from two speakers between 1994–2001 for the Atlas linguistique audiovisuel du francoprovençal valaisan (ALAVAL) (Diémoz & Kristol Reference Diémoz and Kristol2018) show that, while the male participant (born 1935) produces a small number of palatalized tokens (in variation with [l]), the female participant (~1937) produces no palatalized tokens at all. This may indicate then that any shift to [l] is akin to a change-from-above (Labov Reference Labov2001:274).

To summarize, in Savièse, historical evidence indicates that a categorical rule of palatalization may have become variable, whereby [ʎ] has come to co-vary with [l] across the obstruent + lateral clusters. However, none of the authors circumscribe this variation in any way, nor do they provide details of the conditions under which this variation is constrained, which is typical of the dialectological tradition. There is no comparable evidence for Les monts du Lyonnais, where very little sociolinguistic spadework has hitherto taken place, and where atlas data only illustrate the presence of palatalization to [j] before velars.

Two further considerations concerning palatalization need to be highlighted. First, the palatalization of laterals in both regions has long been subject to metalinguistic commentary, where it is evaluated as a typical Francoprovençal feature (e.g. Duraffour Reference Duraffour1932:238). Moreover, during fieldwork, speakers offered commentary on lateral palatalization, describing it as the ‘famous L’ or ‘impossible to explain to nonspeakers’. In a few cases, speakers evoked a ‘performative style’ (Schilling-Estes Reference Schilling-Estes1998:56) in which the feature was offered in a rote phrase as a shibboleth. Lateral palatalization can therefore be said to carry icon status among speakers as a ‘Francoprovençal-like’ target. Secondly, this has been recognized by language revitalizers. Recent pan-regional standardization efforts in the region have given rise to a unified multidialectal orthography termed Orthographe de référence B (ORB; Stich, Gouvert, Favre, & Walter Reference Stich, Favre and Walter2003). ORB has been heavily criticized by speakers and linguists alike, but it has been adopted by the Arpitan revitalization movement, and remains popular among new speakers in particular (Matthey & Meune Reference Matthey and Meune2012:108). Under ORB, /l/-palatalization is transcribed with the characteristic diagraph <ll>, that is, <cll, gll, pll, bll, fll>, indicating that palatalization occurs in all obstruent + lateral sets. In an earlier iteration of the orthography, the ‘recommended’ pronunciation [ʎ] is offered for <ll>, which is described (inaccurately) as the ‘majority’ form for the Francoprovençal-speaking region, and the form that speakers seeking a ‘standard’ accent should emulate (Stich Reference Stich1998:78).

A review of the literature shows that /l/-palatalization is both variable and subject to metalinguistic evaluation. The feature is therefore suitable for variationist analysis, and is referred to as (l) in the Labovian tradition. In the following sections, data are presented from the empirical investigation. The first dataset comes from participants in Les monts du Lyonnais (France).Footnote ⁸ These data are then compared with a second dataset collected in Savièse (Switzerland). First, the research design is outlined.

METHODOLOGY IN TWO FRANCOPROVENÇAL-SPEAKING COMMUNITIES

Fieldwork was conducted in France and Switzerland in 2012. In both sites, Francoprovençal has long been subordinated to French. The data come from a judgement sample of participants, via the friend-of-a-friend technique, who fall into three categories: ‘older’, ‘younger’, and ‘new’ speakers (participant demographics are provided in Appendix 1).Footnote ⁹ ‘Older’ speakers are defined here as an inter-war generation who acquired Francoprovençal as a mother-tongue: they are disproportionately retired rural dwellers. The ‘younger’ speakers are a postwar generation who acquired Francoprovençal in late adolescence. While French was prioritized in the home by immediate caregivers, they were nonetheless exposed to Francoprovençal through its presence in the wider sociocultural environment (e.g. grandparents, older siblings, the community). Like their reference group (the older speakers), they are also typically rural dwelling. Younger speakers exhibit significant variation in their production, often evidencing changes and simplifications in the grammar. In this respect they are analogous to the ‘semi-speakers’ described above. The new speakers, by contrast, are much younger tertiary-sector workers who began acquiring Francoprovençal as adults in an educational context, through evening classes provided by local language associations, or through the use of available pedagogy on the language. These speakers are much more likely to write and read than speak Francoprovençal, and there can be substantial variation in proficiency levels. For the purposes of this study, it is the extent to which they are able to acquire internal and external constraints on variation that is of most interest, and whether they can style shift. The distribution of the final sample is presented in Table 6.Footnote ¹⁰

Table 6.

Study final sample.

In order to compare the linguistic practices of these three groups, the research design borrows from Schmidt (Reference Schmidt1985), who adopted an index of language use to account for production differences between younger and older speakers of Dyirbal, another context of gradual language death. Schmidt's motivation for this approach lay not in quantifying language and social networks in the traditional sense (Milroy Reference Milroy1980), but in uncovering the boundaries of a focused peer-group variety that she describes as different to the traditional Dyirbal of older speakers (Schmidt Reference Schmidt1985:147–48). There are therefore clear and fruitful parallels with the context of the present study (i.e. older speakers of Francoprovençal and new speakers of Arpitan). Schmidt's index is based on scores assigned to speakers, which she then employs as a proxy for group structure. For the present study, an Arpitan Engagement Index (AEI) was devised to establish the extent to which traditional production patterns are oriented to by all three groups of participants. In other words, if speakers do not approximate community norms, can this behaviour be accounted for in general terms using the AEI. To arrive at the AEI, a score was given on the basis of responses to questions in interviews. For each participant in the final sample, an index ranging from 0–6 was calculated posthoc according to the following indicators (participant scores are presented in Appendix 1), given in (1) below.

1. (1)

   1. a. labels their variety ‘Francoprovençal’ or ‘Arpitan’ rather than ‘patois’

   2. b. has made use of some pedagogy to advance proficiency

   3. c. reads Francoprovençal literature from other regions

   4. d. uses Francoprovençal on the internet

   5. e. engages in language activism

   6. f. participates in the Arpitan revitalization movement

On the basis of these scores, participants were then categorized by the totals given in (2).

1. (2)

   1. a. an index score of 0–2 (low), reflecting very traditional practice

   2. b. a score of 3–4 (mid-way), reflecting some traditional practice

   3. c. a score of 5–6, constituting an independent category (high), defined here as a self-conscious category of speakers, whose sociolinguistic practices are predicated on most (if not all) components of the indexFootnote ¹¹

Data collection involved two recorded sessions. In the first meeting, a wordlist translation task was administered to participants in order to elicit a more monitored speech style. Such tasks have been fruitful as a substitution for reading passages among largely illiterate endangered-language speakers as a means of maintaining the formality of Labov's reading style (Klingler & LaFleur Reference Klingler, LaFleur, Detey and Nouveau2007:336). Owing to high illiteracy rates in Francoprovençal, it was necessary to prompt participants in French.Footnote ¹² The items included in the wordlist (Appendix 2) contain instances of the dependent variable, and fillers were also interspersed to conceal the features of interest. Participants were then invited to attend sociolinguistic interviews in groups. Group interviews were conducted in the target language between older, younger, and new speakers so as to elicit a more casual, less monitored speech style.

While conversational modules (Labov Reference Labov, Baugh and Sherzer1984) were adopted, the aim of the group discussions was to elicit the most vernacular style (Table 1), and so rather than the author coordinating the group discussions (as a nonnative speaker and a noncommunity member), they were directed by community volunteers. This enabled the author to fade into the background so as to allow the participants to guide the discussion themselves. This approach has been demonstrably successful in endangered-language research (Haddican Reference Haddican2007:689). As far as possible, both styles are analyzed below. It was not possible, however, to elicit tokens across styles for every participant in the study, owing to the old age, frailty, and isolation of many of the speakers (and thus the fact that not all participants could attend group interviews), and so the data are fragmentary. This is, however, to be expected of research undertaken on severely endangered languages, as compared with the bulk of variationist research on dominant languages in urban settings (Blainey Reference Blainey2017; Chirkova, Stanford, & Wang Reference Chirkova, Stanford and Wang2018). The author acknowledges here the constraints that the nature of the data places on the discussion and interpretation of findings.

As the dependent variable is binary in both fieldwork sites, with sufficient auditory salience to distinguish competing variants, coding was undertaken impressionistically. Where doubt arose between two possible forms, however, coding was backed up by spectrogram readings in Praat (version 5.3.17, Boersma & Weenink Reference Boersma and Weenink2012). Spectrogram settings were kept to a frequency view range of 6000 Hz, with a dynamic range of 50.0 dB, suitable for measuring formant values at F1, F2, and F3. Owing to the fragmentary nature of the dataset, there are limitations on what inferences can be drawn in the analysis. The primary statistical measure adopted below is Fisher's exact test of significance (two tailed), which has been demonstrated to perform well on small datasets of the type described above (e.g. Nance et al. Reference Nance, Wilson and Dunmore2016:177). Where inferences cannot be drawn, generalizations are provided based on descriptive statistics.

RESULTS: (1) AND STYLE VARIATION

A total of 833 tokens were extracted from both interview conditions across fieldwork sites. The only significant internal constraint that emerged in the dataset was the type of syllable initial obstruent preceding the lateral. No other segmental environment effects were found to be palatalizing triggers (see Kasstan Reference Kasstan2015a).

Palatalization in Les monts du Lyonnais

Among the older speakers, 99% of palatal tokens were observed in velar + lateral sets (n = 332); one token was recorded in the /pl/ cluster (see Table 8).

Table 7.

Group-interview structures.

Table 8.

Distribution of variants for Lyonnais older speakers.

Observations from fieldwork undertaken six decades after Gardette (Reference Gardette1950–1956) confirm then that internal-linguistic constraints operating on (l) have been maintained: palatalization remains confined to the velar sets, and this pattern was replicated in the ‘younger’ speaker category, where no palatalization occurred in the labial sets (n = 16 tokens were recorded in the velar sets; see Table 9).

Table 9.

Distribution of variants for Lyonnais younger speakers.

However, both older and younger speakers show variable palatalization in the velar sets, confirming that this rule is no longer categorical in Les monts du Lyonnais. Individual older-speaker variation is illustrated in Figure 2 (given that so few younger speaker tokens are present in the dataset, only the older speaker data are discussed further, below).

Figure 1.

Francoprovençal-speaking region in Europe (after Kasstan & Nagy Reference Kasstan, Müller and Nagy2018:2).

Figure 2.

Distribution of variants by Lyonnais older speaker (careful speech style).

The granular evidence shows that there is substantial variation within the most careful speech style (cf. Figure 4). While some speakers are categorical, others show variable (and in some cases very low) rates of palatalization. The distribution is presented for sex in Figure 3, which is not statistically significant (Fisher's exact test, two tailed p = .69). However, it is noteworthy that the females sampled do show a slight preference for the lateral variant over males in the aggregate, in spite of low token counts.

Figure 3.

Distribution of variants for Lyonnais older speakers and sex.

Figure 4.

Distribution of variants for Lyonnais older speakers and speech style.

A comparison of Figures 3 and 4 shows that, of the available conventional external-linguistic factors under analysis in the older speaker category, only speech style emerged as a statistically significant factor in the Lyonnais dataset (Fisher's exact test, two tailed p < .001).Footnote ¹³ As Figure 4 illustrates, among a relatively homogeneous sample of sixteen older speakers, there is a significant difference observed for the distribution of [j] (evaluated as the typical Francoprovençal variant) in the most formal style (wordlist), and [l] in the most casual style (group interviews). Therefore, much like the findings reported by Clarke (Reference Clarke, Stanford and Preston2009) and Rau et al. (Reference Rau, Chang, Dong, Stanford and Preston2009), here the variable appears to behave in an analogous way to a sociolinguistic marker as described by Labov (Reference Labov1972:238): there is clearly sensitivity to speech style. The nature of the style shift, however, requires further discussion (below), as speakers are not orienting to a standard variant.

Among the three new speakers sampled in Les monts du Lyonnais, broadly the same pattern is observed in that palatalization takes place most often in the velar sets.

As Table 10 and Figure 5 illustrate, the new speakers have to some extent acquired the internal constraints for (l), and the distributions are comparable to the older and younger speaker groups (although it is only possible to compare in very general terms, as just four (l) tokens were observed in spontaneous speech for this group, all of which were [l]). Thus, new speakers are also predominantly palatalizing in the velar sets (Fisher's exact test, two tailed p < .001). However, innovative palatalizing tendencies are also observed in the Lyonnais new-speaker dataset. As Figure 5 illustrates, tokens are observed for palatalization in the labial sets, and a third possible [lʲ] variant is also present, in addition to [j] and [l], which is unattested in the literature for this part of the Francoprovençal-speaking region (Kasstan Reference Kasstan2015b:351).

Figure 5.

Distribution of variants for Lyonnais new speakers.

Table 10.

Distribution of variants for Lyonnais new speakers.

Table 11 presents a more nuanced view of the distribution, where one new speaker (A18-23) is responsible for the production of palatalized lateral [lʲ] variant in the velar sets, which occurs in greater frequency than the traditional median approximant. Moreover, a small number of palatalized tokens are produced by participants A18-23 and S07-24 in the labial sets, too. These observations are discussed in detail below.

Table 11.

Distribution of variants by individual Lyonnais new speaker.

Palatalization in Savièse

Figure 6 shows that speakers sampled in the study who had acquired Francoprovenççal through traditional transmission produced zero palatalized variants.

Figure 6.

Distribution of variants across Savièse sample.

This observation lends empirical support to Jeanjacquet's (Reference Jeanjacquet1931) claim concerning recessive palatalization, and supports the above argument regarding local orthographic practice, too. Conversely, five palatalized tokens were recorded among the new speakers in the sample. While this is a very small number (~4% as a proportion of the total token count for these speakers), the principle of accountability (Labov Reference Labov1972:22) mandates that an explanation be sought for the greatest possible range of the variation. Further, Trudgill (Reference Trudgill1999:322) has warned of the dangers inherent in ignoring small token counts in sociolinguistic research, as this may mask patterns reflecting incipient change. Therefore, it is necessary to consider why new speakers might produce palatalized variants of (l), when they are not only more likely to be influenced by contact with French (a nonpalatalizing variety) than any other group, but when they are also exposed to local norms in a community that has been shifting from [ʎ] to [l] for nearly a century. Before discussing the data further, the distribution is further broken down by individual speaker and speech style (Tables 12, 13).

Table 12.

Distribution of variants among Savièse new speakers.

Table 13.

Distribution of variants for J13-26 by style (Savièse).

As both tables show, one speaker (J13-26) is responsible for the production of palatalized variants in the Savièse dataset: four tokens of [lʲ] in the /k/ + /l/ cluster, and just one instance of [j] in an /f/ + /l/ cluster, none of which are attested variants for Savièse (Diémoz & Kristol Reference Diémoz and Kristol2018). Conversely, C08-63 produces [l] in all contexts. These observations are further interrogated below.

Palatalization and index scores

The results presented above illustrate that all three groups in the Lyonnais sample are (to a greater or lesser extent) toeing traditionally attested community norms, in that, where [j] is anticipated, speakers are palatalizing (albeit at variable rates). Conversely, in the Savièse sample, it has been demonstrated that palatalization no longer takes place. However, some new speakers across fieldwork sites have been shown to be more innovative in their production of palatal forms, whereas others are following attested patterns more closely. To augment the analysis of this variation, the distribution of (l) tokens is compared against the AEI scores outlined in the research design (see Appendix 1 for speaker scores).

Figure 7 illustrates a cross-tabulation of AEI scores for all three categories of speakers (labelled ‘low’, ‘mid-way’, and ‘high’), and the total count for (l) observed in the corpus across Lyonnais and Savièse samples. As the figure shows, tokens of [l] and [j] are observed for speakers irrespective of their AEI score (which reflects the data presented above). Only those participants who accrued a ‘high’ score, however, were found to produce the innovative palatal variant [lʲ]. In a very general sense then, there is a relationship between speakers’ orientation to the Arpitan movement and speech production. However, it is again pertinent to highlight that token counts remain very low. Therefore, this limited quantitative evidence is supplemented with interactional data in the discussion.

Figure 7.

AEI categories and total count for (l).