2.1 Data and distribution

The fullest published accounts of syncretism between preterite and imperfect subjunctive forms in the first and second person plural are provided in historical-comparative surveys of Occitan verb inflection by Ronjat (Reference Ronjat1937: 262; data reproduced in Table 2)Footnote ⁸ and Camproux (Reference Camproux1962: 428–432, 440–441); the pattern is also attested in individual dialect descriptions (Brunel & Camproux Reference Brunel and Camproux1931: 23–29; Teissier Reference Teissier1964: 27–35;Footnote ⁹ Rambier & Tichit Reference Rambier and Tichit2006: 12–16) and as a sub-standard dialectal variant in Alibèrt’s normative grammar (Reference Alibèrt1976: 118, 121). Sources concur that this pattern occurs systematically across all lexemes and conjugational classes, and that it is almost exclusively confined to Occitan varieties spoken in the historical province of Gévaudan (coextensive with the modern département Lozère, in the région Occitanie).

Table 2 Preterite and imperfect subjunctive forms for the three traditional conjugations in the Gévaudan (Ronjat Reference Ronjat1937: 262). Above, orthography as in source; below, conversion into IPA.

The Gévaudan (Figure 1) falls within the geographical area covered by the regional atlases of the Massif Central (ALMC, Nauton Reference Nauton1957–Reference Nauton1963) and eastern Languedoc (ALLOr, Boisgontier 1981–Reference Boisgontier1986). Complete inflectional paradigms for a range of conjugational types were elicited for both atlases: the ALMC data were published in tabular form, while the ALLOr data remain as unpublished fieldwork transcriptions archived by the University of Toulouse. Of the 55 ALMC survey points and 86 ALLOr survey points (141 total), only 12 survey points attest the presence of the Gévaudan syncretism pattern: 11 points within Lozère and a further point within Gard (Figure 2).

Figure 1 Map of southern France, showing location of the Gévaudan area.

Figure 2 ALMC and ALLOr survey points at which the Gévaudan syncretism pattern is attested.

The data for all 12 survey points confirm those given by Ronjat (Reference Ronjat1937: 262), Camproux (Reference Camproux1962: 428–432, 440–441), and Alibèrt (Reference Alibèrt1976: 118, 121). In the relevant varieties, only in the first person plural and the second person plural, there is complete syncretism between the preterite and the imperfect subjunctive. This syncretism consistently involves introduction of forms with thematic /es/ into the preterite, contrasting with thematic /εt/ in the third person singular and thematic /εr/ in the other persons (note /r/ may be realised [ɾ], [r] or [ʀ] according to speech variety). Further illustrative examples, from ALLOr survey point 48.03 Saint Germain de Calberte (southern Lozère), are given in Tables 3 and 4.

Table 3 Preterite and imperfect subjunctive forms for the three traditional conjugations in Saint Germain de Calberte (ALLOr 48.03). Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

Table 4 Preterite and imperfect subjunctive forms for èstre ‘be’, aver ‘have’, and far ‘do’ in Saint Germain de Calberte (ALLOr 48.03). Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

2.2 Historical context

The modern Occitan preterite and imperfect subjunctive comprise the surviving reflexes of Latin perfectum forms (for discussion of Latin perfectum and infectum inflection, see e.g. Stump Reference Stump2016: 90–92; Maiden Reference Maiden2018: 29–33, 37–38). The modern imperfect subjunctive is a continuant of the Latin pluperfect subjunctive, while the modern preterite results from late-mediaeval conflation of inflectional series associated with two separate TAM categories: the continuant of the Latin perfect (mediaeval preterite) and the continuant of the Latin pluperfect indicative (mediaeval old conditional or second conditional). Examples of the mediaeval series are shown in Table 5, based on the descriptions of Anglade (Reference Anglade1921), Skårup (Reference Skårup1997), and Wheeler (Reference Wheeler2012).

Table 5 Imperfect subjunctive, preterite, and ‘second conditional’ forms of vendre ‘sell’ in mediaeval Occitan (Anglade Reference Anglade1921: 294–295; Skårup Reference Skårup1997: 107–121; Wheeler Reference Wheeler2012: 20). Left, attested orthography; right, IPA conversion.

The history of all three series is characterised by pervasive analogical remodelling, including levelling of conjugational class distinctions. By the mediaeval period, only three broad conjugational types are found for the preterite, imperfect subjunctive, and second conditional. The majority of lexemes, including the large and productive first conjugation, present forms of the type shown in Table 5, with unstressed roots and thematic /ε/, reflecting widespread generalisation of Latin perfects in -dedī, -dedistī, -dedit, -dedimus, -dedistis, -dedērunt (see e.g. Wheeler Reference Wheeler, Maiden, Smith, Goldbach and Hinzelin2011, Reference Wheeler2012). The exceptions to this pattern are reflexes of the Latin i-conjugation, which present forms with unstressed roots and thematic /i/ (e.g. partí [parˈti] ‘leave.pret.3sg’, partís [parˈtis] ‘leave.ipf.sbjv.3sg’, partíra [parˈtira] ‘leave.cond2.3sg’); and a small group of high-frequency lexemes which retain extensive root allomorphy, including stressed roots in a subset of preterite and second conditional forms (e.g. dec [dek] ‘have_to.pret.3sg’, degués [deˈɡes] ‘have_to.ipf.sbjv.3sg’, dégra [ˈdeɡra] ‘have_to.cond2.3sg’).

Subsequently, both these patterns are replaced by the majority /ε/ type, but perfectum reflexes in non-first-conjugation verbs typically retain or develop a characteristic inflectional formative, often with a thematic velar (e.g. as in Table 3, a thematic augment /iɡ/ for verbs of the i-conjugation and a thematic augment /eɡ/ for verbs outside the a-conjugation and i-conjugation). The preterite and second conditional are conflated into a single TAM category, a development favoured by the reduced functional load of the second conditional, and the formal similarity between the third person plural forms of each series (Ronjat Reference Ronjat1937: 181; Camproux Reference Camproux1962: 430; Allières Reference Allières1971: 255; Allières Reference Allières and Abry1988; Esher Reference Esher2021a,Reference Esherb,Reference Esherc, Reference Esher2022). In modern Occitan, the third person singular preterite form consistently continues the mediaeval preterite; in the majority of varieties, all other person forms display a thematic /r/ originating in the second conditional, although localised variants are found in which the first and second person plural forms retain reflexes of the mediaeval preterite and, more rarely, in which the first person singular form retains a reflex of the mediaeval preterite (for a typology, see Bybee & Brewer Reference Bybee and Brewer1980). The imperfect subjunctive retains its historically regular forms with thematic /s/. Illustrative examples from varieties in areas adjoining the Gévaudan are shown in Table 6.

Table 6 Preterite and imperfect subjunctive forms of vendre ‘sell’ in the varieties of Sorbs (Hérault, ALLOr 34.10), Saint Martin d’Ardèche (Ardèche, ALLOr 07.05), and Monteils (Gard, ALLOr 30.07). Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

2.3 Syncretism, analogy, and take-over

The Gévaudan first and second person plural preterite forms necessarily result from analogical innovation: in Occitan, there is no precedent for forms with thematic /s/ in the preterite, whereas such forms are etymological and systematic throughout the imperfect subjunctive, offering a potential template for analogical remodelling. Initial comparison of modern and mediaeval forms suggests that the Gévaudan development may be a case of take-over (Carstairs Reference Carstairs1987; Baerman et al. Reference Baerman, Brown and Corbett2005: 151–160), i.e. syncretism resulting from replacement of certain full inflected wordforms in a lexeme by full inflected wordforms originating elsewhere in the inflectional paradigm of the lexeme. However, for the Gévaudan, take-over is at most only part of a more complex mechanism of change.

The interest of the Gévaudan syncretism pattern resides in the fact that, although obviously implicating the preterite and imperfect subjunctive, reference to these TAM categories alone proves inadequate to account for the specific features of the pattern: namely, its directionality (from imperfect subjunctive to preterite), the restricted paradigmatic incidence of syncretism (confined to first and second person plural forms), and the restricted geographical incidence of the syncretism pattern (confined to the Gévaudan). Section 3 discusses the insights and limitations of an account based on established patterns of formal relationship between imperfect subjunctive and preterite forms, prior to a more detailed examination of comparative dialect data (Section 4) which will identify the further-reaching formal relationships crucial to the emergence of the Gévaudan syncretism pattern.

3.1 The metamorphome PYTA

The Gévaudan syncretism pattern is discussed by O’Neill (Reference O’Neill2011, Reference O’Neill2014) in two studies which explore the definition and formalisation of morphomes (i.e. irreducibly morphological structures; Aronoff Reference Aronoff1994; Maiden Reference Maiden2009a, Reference Maiden2016, Reference Maiden2018), particularly the Romance metamorphome ‘PYTA’ identified by Maiden (Reference Maiden2001, Reference Maiden, Booij and van Marle2005). Metamorphomes are conventionally given abstract labels in recognition of their intrinsic independence from functional motivations; the label ‘PYTA’ is derived from the conventional expression perfecto y tiempos afines ‘perfect and related tenses’ used in the Spanish grammatical tradition to refer to the surviving reflexes of Latin perfectum forms. In modern Romance varieties (Maiden Reference Maiden2018: 44–83), PYTA typically comprises continuants of the Latin perfect (Romance preterite), pluperfect subjunctive (Romance imperfect subjunctive in -se) and pluperfect indicative (Romance imperfect subjunctive or conditional in -ra). These categories are functionally diverse, but consistently share inflectional exponents; they are ‘related’ in terms of morphological form, independently of function.

The diachronic behaviour of PYTA across the Romance languages has been extensively documented (see e.g. Maiden Reference Maiden2001, Reference Maiden, Booij and van Marle2005, Reference Maiden2009b, Reference Maiden2011a,Reference Maidenb, Reference Maiden2016, Reference Maiden2018: 44–83; O’Neill Reference O’Neill2011, Reference O’Neill2014; Wheeler Reference Wheeler, Maiden, Smith, Goldbach and Hinzelin2011; Esher Reference Esher2015, Reference Esher2016, Reference Esher2017, Reference Esher2021a,Reference Esherb,Reference Esherc, Reference Esher2022). These studies consistently evidence the diachronic productivity of PYTA and thereby the psychological reality of this metamorphome for historical speakers (Maiden Reference Maiden2018: 12–17, 49–50). For any given lexeme, the Romance reflexes of Latin perfectum forms inherit a shared, often distinctive, lexical root or stem; while the inflectional exponents of these forms may undergo considerable historical changes, the inherited distributional pattern of identity between perfectum reflexes is robustly maintained. Thus, the constituent cells of PYTA in any given lexeme undergo the same analogical changes as each other (Maiden Reference Maiden2018: 50–54); and where lexemes have a distinctive PYTA root, this may acquire a characteristic phonological shape across lexemes, such as the presence of a high root vowel in Castilian and Portuguese (Maiden Reference Maiden2018: 63–64), a root-final long consonant in Italo-Romance (Maiden Reference Maiden2018: 66–67), a root-final sibilant in French and Romanian (Maiden Reference Maiden2018: 66–67), or a root-final velar in Catalan and Occitan (Wheeler Reference Wheeler, Maiden, Smith, Goldbach and Hinzelin2011; Esher Reference Esher2016, Reference Esher2022).

Although metamorphomes typically act as templates for the distribution of inflectional exponents associated with the lexical root or stem, there are also indications that they can act as domains for the distribution of inflectional exponents correlated with morphosyntactic feature values such as person and number (Maiden Reference Maiden2009b, 2018: 60–62). For example, in Daco-Romance, a formative -şi originating in the second person singular pluperfect form is spread across the second person singular forms of all and only TAM categories within PYTA (Maiden Reference Maiden2018: 61): in this example, the domain for analogical extension is jointly defined by the specific morphosyntactic feature value combination and the wider metamorphome.

3.2 Abstractive morphology and analogical change within metamorphomes

O’Neill (Reference O’Neill2011: 148–149, 2014: 63) treats the Gévaudan syncretism pattern as a further example of inflectional material undergoing extension within the bounds of an existing metamorphome:

The key insight of O’Neill’s analysis (Reference O’Neill2011: 149–150, Reference O’Neill2014: 63) is that the emergence of the modern Gévaudan system is incompatible with a constructive, morphemic account. Historically, -ess- occurred exclusively in imperfect subjunctive forms (see Tables 6, 9–10).Footnote ¹⁰ Within a morphemic account, this distributional fact would be captured by treating -ess- as a morpheme embodying the inflectional features of the TAM category ‘imperfect subjunctive’ – in this case morphosemantic features (Corbett Reference Corbett2012: 49) – the imperfect subjunctive might be characterised as tense-neutral, aspect-neutral and subjunctive, see Maiden (Reference Maiden2011a: 179). However, if speakers had made a morphemic analysis of this type, incompatibility between the morphosemantic features of the preterite (past, perfective, indicative) and the morphosemantic content of -ess- would preclude extension of -ess- to the preterite. Within a lexical–realisational approach (in the typology of Stump Reference Stump2001) a speaker producing a preterite form would avoid selecting the morpheme -ess- because there is nothing in the feature specification of the preterite to license it; within a lexical–incremental approach, a speaker producing an inflectional form would introduce the morpheme -ess- bearing imperfect subjunctive meaning and thus create an imperfect subjunctive form rather than a preterite. The fact that, historically, speakers in the Gévaudan were able to extend the substring -ess- to the preterite, modifying the shape of the inflectional wordforms without modifying their function, indicates that these speakers did not make a morphemic segmentation associating -ess- with the meaning ‘imperfect subjunctive’ (O’Neill Reference O’Neill2011: 150), or indeed with any specific TAM value. O’Neill argues instead for an abstractive approach, in which

a view which this study also supports.

In order to account for the Gévaudan syncretism pattern, O’Neill appeals to a principle of analogical levelling within metamorphomes: the imperfect subjunctive and the preterite are members of the same metamorphome (PYTA) and as such are liable to increase their formal similarity via analogical change (Reference O’Neill2014: 64; compare Section 3.1), with the ultimate similarity being full identity. Yet the concept of metamorphome-internal levelling falls short as an explanation for the Gévaudan syncretism pattern. It does not predict the distinctive properties of this pattern, namely the directionality of the syncretism, the paradigmatic restriction of the syncretism to the first and second person plural, and the geographical restriction of the pattern to a single small area. Take-over of the imperfect subjunctive by the preterite, or syncretism between preterite and imperfect subjunctive for any other person/number combination(s), would constitute equally good examples of metamorphome-internal levelling, yet neither is attested in southern Gallo-Romance varieties (to the author’s knowledge at the time of writing).Footnote ¹¹ Furthermore, it is not clear that the Gévaudan syncretism pattern can legitimately be classified as an instance of levelling at all: the outcome of the change is a redistribution of allomorphy, not reduction or elimination (compare Fertig’s Reference Fertig2016 definition, cited in Section 1.1). In this respect, the Gévaudan development differs significantly from known examples of metamorphomes acting as productive domains for inflectional analogy: analogy affecting the stem or theme vowel typically generalises a single form to all constituent cells of a metamorphome, while analogy affecting personal desinences typically generalises a single form to all cells with the relevant person and number features within the metamorphome (Section 3.1).

3.3 Maintaining an abstractive account

O’Neill’s discussion of the Gévaudan syncretism pattern frames it as offering support for abstractive theories of inflectional systems and attempts to motivate the emergence of the pattern via a trend for analogical levelling within metamorphomes. The attempt is ultimately unsuccessful, since the paradigmatic scope of the syncretism contrasts with existing examples of levelling within metamorphomes, and a general principle of levelling is not equipped to predict the specific directionality, paradigmatic distribution, and geographical distribution of the syncretism pattern. A more satisfactory account is developed here by reconstructing the detail and context of the historical change, based on comparative dialect data (Section 4) and implicating morphomic relationships well beyond PYTA (Sections 5 – 7). While the substance of this account differs considerably from O’Neill’s analysis, it will ultimately offer much stronger empirical support for O’Neill’s theoretical conclusions concerning the explanatory power of abstractive theories of inflection, and the crucial importance of relationships of form independently of relationships of function.

4.1 Preterite and imperfect subjunctive inflection in neighbouring varieties

Common and widespread patterns of preterite and imperfect subjunctive inflection found in Occitan varieties of the areas surrounding the Gévaudan are illustrated for the first-conjugation verb cantar [kanˈta] ‘sing’ in Table 7 (for corresponding forms of vendre [ˈbendɾe]/[ˈvendʀe] ‘sell’ in the same varieties, see Table 6).

Table 7 Preterite and imperfect subjunctive forms of cantar ‘sing’ in the varieties of Sorbs (Hérault, ALLOr 34.10), Saint Martin d’Ardèche (Ardèche, ALLOr 07.05) and Monteils (Gard, ALLOr 30.07). Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

The variety of Sorbs (northern Hérault, ALLOr 34.10) exemplifies a pattern typical of the Languedocien dialect area (Alibèrt Reference Alibèrt1976: 118), south-west of the Gévaudan. In this system, all preterite forms outside the third person singular are paroxytonic (i.e. stressed on the penult) with thematic /εr/, and all imperfect subjunctive forms are paroxytonic with thematic /εs/. The exponents following the thematic formative are consistent across the two TAM categories (for Sorbs, [e] in the first person singular, [os] in the second person singular, [en] in the first person plural, [es] in the second person plural, [u] in the third person plural), excepting only the third person singular which has a unique preterite form in -[ˈεt]. In this variety, thematic /r/ from the second conditional occurs in all preterite forms other than the third person singular. Primary stress in the first and second person plural has moved from the final syllable (the historically regular outcome; see Section 2.2) to the penult (historically expected in all other forms with thematic /r/, i.e. the first person singular, second person singular, and third person plural). Final [en] and [es] in the first and second person plural forms originate in the imperfect subjunctive, e.g. mediaeval Occitan vendessem [venˈdesen] ‘sell.ipf.sbjv.1pl’ < uēndedissēmus, vendessetz [venˈdesets] ‘sell.ipf.sbjv.2pl’ < uēndedissētis (see Section 2.2 and Table 5).

The variety of Saint Martin d’Ardèche (south-eastern Ardèche, ALLOr 07.05) exemplifies a pattern typical of the Provençal dialect area (Martin & Moulin Reference Martin and Moulin2007: 89), south-east of the Gévaudan. This system displays many distributional similarities with that exemplified by Sorbs: /r/ is present throughout preterite forms other than the third person singular; /s/ is present throughout imperfect subjunctive forms; inflectional formatives following thematic /r/ or /s/ are parallel across the two categories, as are thematic vowels preceding thematic /r/ or /s/. The Provençal type differs in that the first and second person plural preterite and imperfect subjunctive forms retain historically expected final stress, and maintain the historically regular alternation between stressed [ε] and unstressed [e].

Also of note in Saint Martin d’Ardèche is the phonological form of the first and second person plural desinences: -eriam -[eˈʀjaŋ], -eriatz -[eˈʀjas] in the preterite, and -essiam -[eˈsjaŋ], -essiatz -[eˈsjas] in the imperfect subjunctive. The forms in this area were traditionally of the type -eram -[eˈʀaŋ], -eratz -[eˈʀas], -essam -[eˈsaŋ], -essatz -[eˈsas] (Ronjat Reference Ronjat1937: 186), with stressed [a] as is historically expected in the second conditional: e.g. uendedeˈramus > vendeˈram [vendeˈʀan] ‘sell.cond2.1pl’, uendedeˈratis > vendeˈratz [vendeˈʀas] ‘sell.cond.2pl’ (compare Table 5). The modern forms with yod are due to analogical extension of formatives etymological in the synthetic conditional and non-first-conjugation imperfect indicative: e.g. uēndēˈbāmus > vendiam [venˈdjaŋ] ‘sell.ipf.ind.1pl’, uēndēˈbātis > vendiatz [venˈdjas] ‘sell.ipf.ind.2pl’; for the preterite, this extension is dated to the mid-to-late sixteenth century (Koschwitz [1894] Reference Koschwitz1973: 115–117; Ronjat Reference Ronjat1937: 186).

Finally, the variety of Monteils (central Gard, ALLOr 30.07) exemplifies a system occurring immediately south of the Gévaudan and geographically intermediate between the Languedocien and Provençal types described above. The Gard system is characterised by conservatism of the first and second person plural preterite, which, like the third person singular preterite, retain reflexes of the mediaeval preterite without influence from the second conditional: oxytonic (i.e. final-stressed) forms without thematic /r/: vendèm [venˈdεn] ‘sell.pret.1pl’ < venˈdem [venˈdem] < uendedimus; vendètz [venˈdεs] ‘sell.pret.2pl’ < venˈdetz [venˈdεts] < uendedistis. In these varieties, strong formal parallelism between preterite and imperfect subjunctive forms for a given person/number combination is limited to the first person singular, second person singular, and third person plural. In varieties such as these, with first and second person plural preterite forms of the -èm -[ˈεn], -ètz -[ˈεs] type, first and second person plural imperfect subjunctive forms are always of the -essiam -[eˈsjan], -essiatz -[eˈsjas] type.

4.2 Geographical distribution of inflectional types

Figure 3 shows the geographical distribution of first and second person plural preterite and imperfect subjunctive endings in the ALMC and ALLOr datasets;Footnote ¹² the forms are given in ‘classical’ orthography (Alibèrt Reference Alibèrt1976: 7–36), abstracting over localised phonetic variation, particularly in the realisation of final consonants.

Figure 3 First and second person plural endings in the preterite and imperfect subjunctive according to the ALMC and ALLOr data. The exemplar varieties Saint Germain de Calberte (Gévaudan type), Sorbs (Languedocien type), Monteils (Gard type), and Saint Martin d’Ardèche (Provençal type) are marked with a triangle.

The inflectional patterns illustrated in Table 6 occur across the southern part of the survey area. The Provençal type -eriam, -essiam (e.g. Saint Martin d’Ardèche: -[eˈʀjaŋ], -[eˈsjaŋ]), widespread further east, is visible along the south-eastern edge of the survey area, while the Languedocien type -èrem, -èssem (e.g. Sorbs: -[ˈεren], -[ˈεsen]) is robustly attested in the south-western part of the survey area. The Gard type -èm, -essiam (e.g. Monteils, -[ˈεn], -[eˈsjan]) with conservative preterite forms occupies a middle ground between these two. The -èm, -essiam and -èrem, -èssem types directly adjoin the area in which the Gévaudan syncretism pattern is attested. Each area displays a high degree of internal uniformity overall.

To the north of these areas, the attested forms display much greater local variation and the groupings are more approximate. In some of these varieties, the first and second person plural preterite and/or imperfect subjunctive forms are differentiated from the other person/number combinations by a distinctive theme vowel /a/ (locally realised [ɔ] as is common in northern Occitan, see e.g. Moulin Reference Moulin2006: 20) originating in the first conjugation: this phenomenon is most systematic in the north-west of the survey area (southern Cantal), where forms are of the type -ariam, -ariatz, as described by Ronjat (Reference Ronjat1937: 259) and Alibèrt (Reference Alibèrt1976: 118). In the northern part of Cantal, a system similar to the Gard type is found: first and second person plural preterite forms are of the conservative -èm, -ètz type without a thematic vowel or thematic consonant, while the corresponding imperfect subjunctive forms are of the -essam or -essiam type.

To the north-east of the Gévaudan syncretism area, the theme vowel /a/ is ordinarily confined to the first and second person plural imperfect subjunctive, while the preterite retains a mid vowel. The north-eastern area is also characterised by syncretism between the first and third person plural forms, historically a take-over, which introduces the vowel [u] (etymological in the third person plural form) into the first person plural form, e.g. -eriom -[eˈʀjuŋ] as described by Moulin (Reference Moulin2006) for the variety of Aubenàs (southern Ardèche), or -erom [εˈrun] as attested in Quint’s (Reference Quint1999) description of the variety of Albon (central Ardèche). Finally, in the north-eastern corner of the survey area, preterite forms are usually of the -eram type (occasionally -erem) and imperfect subjunctive forms of the -essam type (occasionally -erem).

4.3 Inferring historical change from geographical distribution

For first and second person plural forms, varieties along a corridor reaching from Cantal to Gard (see Figure 3; also Casagrande Reference Casagrande2011: 174) display distinctive inflectional formatives, lacking the rhotic which these forms exhibit in other Occitan varieties (compare the illustrative examples in Table 6). At the northernmost and southernmost extremes of the corridor, conservative forms in -èm, -ètz are retained, while in the north-west a distinctive theme vowel is introduced (Sections 4.2 , 5.3), and in the central area the original preterite forms are evicted by imperfect subjunctive forms. In all these areas, first and second person plural imperfect subjunctive forms are near-exceptionlessly of the oxytonic (final-stressed) -essiam, -essiatz type.

It is significant that the Gévaudan, with its unique innovative forms lacking /r/ in the preterite, is located in the midst of two conservative areas also lacking /r/ in the preterite. This geographical distribution argues against historical presence of rhotic preterite forms of the -eriam, -eriatz type in the Gévaudan. The most plausible reconstruction is instead that the historical preterite forms in the Gévaudan were initially of the conservative, oxytonic -èm, -ètz type (e.g. cantèm [kanˈtɛn] ‘sing.pret.1pl’, cantètz [kanˈtɛt] ‘sing.pret.2pl’, aguèm [aˈɡɛn] ‘have.pret.1pl’, aguètz [aˈɡɛt] ‘have.pret.2pl’) and were directly replaced by the corresponding oxytonic imperfect subjunctive forms (reflexes of the Latin pluperfect subjunctive, e.g. cantessiam [kanteˈsjɔn] ‘sing.pret/ipf.sbjv.1pl’, cantessiatz [kanteˈsjat] ‘sing.pret/ipf.sbjv.2pl’, aguessiam [aɡeˈsjɔn] ‘have.pret/ipf.sbjv.1pl’, aguessiatz [aɡeˈsjat] ‘have.pret/ipf.sbjv.2pl’). This reconstruction is implicit in Casagrande’s analysis:

While Casagrande’s view of the change as a morphological repair for speaker confusion is implausible, the substance of the change which he assumes (replacement of -èm, -ètz by -essiam, -essiatz) is entirely consistent with the comparative geographical evidence presented in Figure 3.

4.4 Chronology of developments

For most Occitan varieties in the Languedoc, Provence, and northern Occitan regions, forms with thematic /r/ emerged throughout the preterite (excepting the third person singular; see Bybee & Brewer Reference Bybee and Brewer1980 for the resistance of this form to change), due to merger with the second conditional. Forms with thematic /r/ are robustly conserved, despite the close parallelism between preterite and imperfect subjunctive forms for any given person/number combination. Indeed, the parallelism of stress patterns and word-final exponents may result in part from metamorphome-internal levelling; but such levelling never eliminates the contrast between thematic /r/ in the preterite and thematic /s/ in the imperfect subjunctive. A variant development occurs within the corridor identified, where the spread of thematic /r/ through the preterite never extends to affecting the first and second person plural preterite. Subsequently, in a subpart of the corridor, an innovation occurs, in which the etymological first and second person plural preterite forms are replaced by imperfect subjunctive forms. The innovation spreads throughout Lozère, but does not attain the full geographical extent of the original corridor.

The potential window for the emergence of the Gévaudan syncretism pattern extends from the sixteenth to the nineteenth century. Textual evidence for other Occitan varieties indicates that the spread of -[r]- through the preterite took place between the mid-fifteenth and late sixteenth centuries (Esher Reference Esher2021b,Reference Esherc), a development which the Gévaudan innovation must post-date. The data collated in the early twentieth-century Atlas linguistique de la France (Gilliéron & Edmont 1902–Reference Gilliéron and Edmont1910), although providing only scant information on verb inflection, clearly indicate use of reflexes of the Latin pluperfect subjunctive with the value of first person plural preterite in Lozère (see map 1154B nous ne le revîmes pas ‘we did not see him again’) and thus suggest that the introduction of these forms was at least well advanced, perhaps fully complete, by the end of the nineteenth century.

5.1 Formal parallelism between the imperfect indicative of èstre ‘be’ and the second conditional

In most Occitan varieties, including those of the survey area, there exist four patterns for imperfect indicative inflection. First-conjugation verbs exhibit forms with thematic /av/, the regular reflex of Latin first-conjugation imperfect indicative forms in -ābam, and so on (e.g. cantābat > cantava [kanˈtaβɔ] ‘sing.ipf.ind.3sg’). Non-first-conjugation verbs exhibit final-stressed forms with thematic yod, which continue Latin imperfect indicative forms in -(i)ēbam, and so on (e.g. uendēbat > vendiá [benˈdjɔ] ‘sell.ipf.ind.3sg’; see Esher Reference Esher2018); among this group, the i-conjugation is distinguished by the analogical introduction of a thematic augment -iss- (e.g. bastissiá [bastiˈsjɔ] ‘build.ipf.ind.3sg’; see Maiden Reference Maiden2004; Esher Reference Esher2016). Finally, èstre ‘be’ retains reflexes of the idiosyncratic Latin forms eram, erās, erat, erāmus, erātis, erant, which were unique to the single lexeme esse ‘be’.

The significance of the forms eram, etc. is that this series of full inflectional wordforms is near-identical with the rightmost inflectional formatives within Latin pluperfect indicative forms, as shown in Table 8. The two series are initially distinguished by stress placement for the singular and third person plural – on the penult in the imperfect indicative of ‘be’; on the antepenult in the pluperfect indicative, but this distinction is lost between late Latin and mediaeval Romance due to haplology: [venˈdεderam] > [venˈdεram] via deletion of unstressed [de] (see Wheeler Reference Wheeler2012: 13–15, on whose reconstructions the present proposal is based). By mediaeval Occitan, systematic formal identity between the imperfect indicative forms of èstre and the endings of the second conditional is robustly established (for verbs of the majority cantar ‘sing’ and vendre ‘sell’ classes). The historical mechanisms by which this pattern of inflectional identity arises (conservation of inherited identity between forms; loss of inherited distinction between forms, due to regular sound change) are familiar from established cases of subsequently persistent and productive inflectional identity patterns (e.g. Maiden Reference Maiden2009a, Reference Maiden2011a,Reference Maidenb, Reference Maiden2016, Reference Maiden2018; Esher Reference Esher2017, Reference Esher2018, Reference Esher2020).

Table 8 Historical development of imperfect indicative forms of èstre < *essere ‘be’, and pluperfect indicative/second conditional forms of vendre < uendere ‘sell’, based on data from Anglade (Reference Anglade1921: 294–295, 314), Skårup (Reference Skårup1997: 99, 117), and Wheeler (Reference Wheeler2012: 13–15). Left, traditional orthographic forms; right, IPA conversion.

In the subsequent merger of the preterite and the second conditional, the majority of varieties retain the sequence /ˈεr/ in the first person singular, second person singular and third person plural preterite forms (note that for the third person plural, this sequence was etymological in the preterite as well as in the second conditional, e.g. late Latin [venˈdεderont] > mediaeval Occitan vendero(n) [venˈdεro(n)] ‘sell.pret.3pl’, see also Table 5), whereas the third person singular form is always a direct continuant of the preterite (e.g. vendet [venˈdεt] ‘sell.pret.3sg’) and undergoes no influence from the second conditional (e.g. vendera [venˈdεra] ‘sell.cond2.3sg’). The first and second person plural forms show influence from the second conditional in many varieties, but not all; notably, varieties of the Cantal–Gard corridor discussed in Sections 4.2 and 4.3 retain first and second person plural forms uninfluenced by the second conditional.

A consequence of this merger is that in modern Occitan preterite inflection, a subset of person/number combinations display partial formal identity with the imperfect indicative forms of èstre ‘be’, i.e. they retain the inherited pattern of identity between reflexes of eram, and so on, and reflexes of the pluperfect indicative, despite functional reorganisation of the paradigm and some formal redistribution of inflectional material. Modern systems of this type are illustrated in Tables 9 and 10: the data from Sorbs include an example of shared analogical changes undergone by first and second person plural forms in the imperfect indicative of èstre ‘be’ and in the preterite, namely, the shift from final stress to penultimate stress, and the change in thematic vowel quality from [e] to [ε].

Table 9 Finite synthetic forms of èstre [ˈεstɾe] ‘be’ in the variety of Sorbs (ALLOr 34.10), Languedocien type. Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

Table 10 Finite synthetic forms of èstre [ˈεstʀe] ‘be’ in the variety of Saint Martin d’Ardèche (ALLOr 07.05), Provençal type. Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

5.2 A novel alternation pattern in èstre as a model for the preterite

In a significant minority of Occitan varieties, the first and second person plural imperfect indicative forms of èstre ‘be’ are subject to analogical influence from other imperfect indicative inflection patterns. The source of influence is sometimes the majority first-conjugation type, as in the variety of Nice where innovative eravam [eraˈvaŋ] ‘be.ipf.ind.1pl‘, eravatz [eraˈvas] ‘be.ipf.ind.2pl’ correspond to etymological cantavam [kantaˈvaŋ] ‘sing.ipf.ind.1pl‘, cantavatz [kantaˈvaŋ] ‘sing.ipf.ind.2pl’ (Toscano Reference Toscano1998: 97, 100; Esher Reference Esher2018; see also Anglade Reference Anglade1921: 314 for the marginal presence of eravam, eravatz in mediaeval Occitan). More commonly, however, the model for change is the non-first-conjugation type in -iam, -iatz, hence forms such as seiam [seˈjan] ‘be.ipf.ind.1pl’, seiatz [seˈjas] ‘be.ipf.ind.2pl’ in Cantal, and siam [ˈsjan] ‘be.ipf.ind.1pl’, siatz [ˈsjas] ‘be.ipf.ind.2pl’ in Gard and Lozère (Camproux Reference Camproux1962: 442). These last examples also illustrate the eviction of etymological [er]-forms in favour of innovative [s]-forms, generalising an initial consonant common throughout the paradigm of èstre. Footnote ¹⁴

An incidental consequence of the innovation which introduces siam, siatz into the imperfect indicative of èstre ‘be’ is the creation of a novel pattern of formal identity between first and second person plural forms, this time involving full imperfect indicative wordforms for èstre ‘be’ and the rightmost inflectional material in imperfect subjunctive wordforms, as exemplified in Table 11 (variety of Monteils, Gard). From the point of view of the paradigmatic distribution of inflectional formatives, systems such as that of Monteils are near-identical to those of the Gévaudan (e.g. variety of Saint Germain de Calberte, Lozère, Table 12). The inflectional similarity of these paradigms indicates that they are indeed closely related, lending weight to the hypothesis that, prior to the emergence of the Gévaudan syncretism pattern, verb inflection in Occitan varieties of the Gévaudan was of the type found in modern varieties of Gard (an argument developed in Section 4.3 with reference only to the preterite and imperfect subjunctive, but which the data in Tables 11 and 12 show to be of wider paradigmatic validity).

Table 11 Finite synthetic forms of èstre [ˈεstʀe] ‘be’ in the variety of Monteils (ALLOr 30.07), Gard type. Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

Table 12 Finite synthetic forms of èstre [ˈεstʀe] ‘be’ in the variety of Saint Germain de Calberte (48.03), Gévaudan type. Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

The difference between the systems in Tables 11 and 12 amounts to the respective absence or presence of syncretism between the preterite and the imperfect subjunctive for first and second person plural forms. For varieties with syncretism, the analogical extension of (originally) imperfect subjunctive forms into the preterite replicates a pattern of alternation which emerged in the imperfect indicative forms of èstre due to an independent innovation affecting the first and second person plural. This alternation involves a contrast within a given TAM category, between forms ending in [ˈεʀV(C)] in the first person singular, second person singular, and third person plural, and forms ending in [ˈsjVC] in the first and second person plural.

The data presented in Tables 11 and 12 indicate a plausible motivation for the extension of imperfect subjunctive forms (e.g. for Saint Germain de Calberte, seguessiam, seguessiatz) as opposed to the creation of novel forms (e.g. *segueriam, segueriatz or *seguiam, *seguiatz): elsewhere in the inflectional system, there is a precedent for alternation between -[ˈεʀV(C)] and -[ˈsjVC] forms within a given TAM category, with exactly the distribution that would result from extension of imperfect subjunctive forms; and the series of wordforms in which this alternation pattern exists (imperfect indicative of èstre) already exhibits partial formal identity with the inflectional formatives occurring in the preterite.

If this hypothesis is correct, one would expect to find that the Gévaudan syncretism pattern only occurs in varieties in which forms of the siam, siatz type emerged in the imperfect indicative of èstre. Section 5 .3 presents evidence that this is indeed the case.

5.3 Geographical and comparative evidence for the role of èstre

Figure 4 illustrates the variety and geographical distribution of first and second person plural imperfect indicative forms of èstre ‘be’ in the ALMC and ALLOr survey area, compared to first and second person plural preterite and imperfect subjunctive forms. The geographical range of the siam, siatz and seiam, seiatz types corresponds almost exactly to the corridor along which first and second person plural preterite forms are sharply distinguished from other preterite forms (Figure 3; Sections 4.2 and 4.3). Forms of the type siam, siatz and seiam, seiatz are thus attested over a much wider zone than the Gévaudan syncretism pattern, a finding consistent with the view that the emergence of siam, siatz and seiam, seiatz in the imperfect indicative of èstre occurs independently of, and prior to, the emergence of -essiam, -essiatz in the preterite.

Figure 4 Comparison of geographical distribution: above, first and second person plural imperfect indicative forms of èstre ‘be’; below, first and second person plural preterite and imperfect subjunctive formatives for all verbs (reproduction of Figure 3). Classical orthography, abstracting over localised phonetic variation. The exemplar varieties Saint Germain de Calberte (Gévaudan type), Sorbs (Languedocien type), Monteils (Gard type), and Saint Martin d’Ardèche (Provençal type) are marked with a triangle.

Further evidence potentially corroborating the role of siam, siatz as a necessary though not automatically sufficient precondition for the emergence of the Gévaudan syncretism pattern is provided by Ronjat’s (Reference Ronjat1937: 259) data for the locality of Aurillac in south-western Cantal (Table 13).Footnote ¹⁵ Ronjat notes two patterns of preterite inflection: one, characteristic of Aurillac itself, in which first and second person plural preterite forms are of the -ariam -[ɔˈrjɔn], -ariatz -[ɔˈrja(s)] type, as in the ALMC data (see Figures 3 and 4); and a variant attested 10–15 km east of Aurillac, in which first and second person plural preterite forms exhibit syncretism with the corresponding imperfect subjunctive, as in the Gévaudan. Note that the theme vowel [ɔ] in first and second person plural forms originates in historical [a] (see Moulin Reference Moulin2006: 20), although mediaeval and comparative data indicate that [e] would be expected; [ɔ] is imported from the first and second person plural forms of the first-conjugation synthetic conditional (e.g. tirorion [tirɔˈrjɔn] ‘drag.cond.1pl’) and/or first-conjugation imperfect indicative (e.g. tirobion [tirɔˈβjɔn] ‘drag.ipf.ind.1pl’).

Table 13 Preterite and imperfect subjunctive forms for Aurillac (Ronjat Reference Ronjat1937: 259); source orthography with IPA conversion.

If the data given by Ronjat are indeed authentic, they indicate a sporadic innovation in this single locality: the replacement of established preterite forms in -ariam, -ariatz (found consistently across the southern part of Cantal) by forms syncretic with the corresponding imperfect subjunctive forms. It is significant that, although distant and isolated from varieties exhibiting the Gévaudan syncretism pattern, Aurillac falls within the zone where the imperfect indicative of èstre systematically instantiates alternation between [s]-forms in the first and second person plural, and [r]-forms in the other persons. These data are consistent with two separate actuations of the same innovation (i.e. syncretism between preterite and imperfect subjunctive in first and second person plural, mediated by the inflectional precedent of èstre). The data for Aurillac further indicate that replacement of -ariam [-ɔˈrjɔn], -ariatz -[ɔˈrja(s)] forms by -assiam -[ɔˈsjɔn], -assiatz -[ɔˈsja(s)] forms is empirically possible. However, the sporadic nature of this change, which does not spread to adjacent varieties and is not attested in the ALMC data, suggests that -ariam, -ariatz preterite forms, which display close formal similarity to inflectional formatives of other TAM categories, are more resistant to analogical remodelling than the -èm -[ˈɛn], -ètz -[ˈɛs] preterite forms, which are more isolated within the inflectional paradigm.

6.1 First and second person plural forms as a domain for analogical change

Isolated instances of analogy affecting first and second person plural forms respectively across distinct TAM categories are noted by several authors for Gallo- and Ibero-Romance varieties (see e.g. Pato & O’Neill Reference Pato and O’Neill2013; Esher Reference Esher2018, Reference Esher2022; Maiden Reference Maiden2018: 111, 291; also Note 11), where such analogies are ordinarily explicable in terms of propagating or aligning existing morphomic distributions of inflectional exponents (Enger Reference Enger2014; Esher Reference Esher2015). A more systematic study is conducted by Milizia (Reference Milizia2016) for Italo-Romance verb inflection, whose findings offer an instructive comparison with the Gévaudan data.

Milizia’s study documents cases in which desinential exponents originating in one TAM category are extended by analogy to another TAM category, resulting in syncretism either between desinential material or between entire wordforms. Consistently, both and only first and second person plural forms are affected, but diverse pairings of TAM categories are involved (preterite and imperfect subjunctive; preterite and imperfect indicative; imperfect subjunctive and conditional; imperfect indicative and conditional), with diverse directionalities (in some varieties, desinences originating in the preterite are extended into the imperfect subjunctive; in other varieties, desinences originating in the imperfect subjunctive are extended into the preterite). The data resist explanation in terms of inflectional feature values since no overall tendency can be discerned; in any case, there is no compelling functional reason why morphosyntactic distinctions of tense, aspect and mood should be of lesser importance in the first and second person plural than for other person/number combinations (Milizia Reference Milizia2016: 111–112).

Milizia argues instead for an account in terms of relative frequency, based notably on the low token frequency of first and second person plural cells compared to other personal forms for a given TAM category, and a principle of dispreferment of cumulative exponents unique to low token frequency cells (Reference Milizia2016: 90).Footnote ¹⁷ Analogical extension of desinences from one TAM category to another typically evicts exponents uniquely associated with low token frequency forms, introducing exponents of higher paradigmatic type frequency in their place. Such extension may take the form either of semi-separate exponence in which desinences can be segmented into a formative correlated with TAM category and a formative correlated with person/number combination, maintaining existing contrasts while reducing cumulative exponence (2016: 102–108); or of compensatory syncretism between TAM categories for the relevant person/number combinations, which maintains cumulative exponence but neutralises existing contrasts (2016: 92–102).

6.2 Applying Milizia’s approach to Occitan

For most Occitan varieties in the survey area, first and second person plural forms outside the present indicative (which is of comparatively high token frequency, Milizia Reference Milizia2016: 89) can be described in terms of semi-separate exponence. Of the four major systems discussed in the present study, three consistently combine a distinctive stem with one of two series of desinences: either -em, -etz (-[ˈen], -[ˈes]) or -iam, -iatz (-[ˈjan], -[ˈjats]). These are the Languedocien type (exemplified by the variety of Sorbs; Table 14), the Provençal type (e.g. Saint Martin d’Ardèche; Table 15), and the Gévaudan type with syncretism between preterite and imperfect subjunctive (e.g. Saint Germain de Calberte; Table 16).

Table 14 Finite synthetic forms of vendre ‘sell’ in the variety of Sorbs (Languedocien type; ALLOr 34.10). Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

Table 15 Finite synthetic forms of vendre ‘sell’ in the variety of Saint Martin d’Ardèche (Provençal type; ALLOr 07.05). Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

Table 16 Finite synthetic forms of vendre ‘sell’ in the variety of Saint Germain de Calberte (Gévaudan type; ALLOr 48.03). Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

The exception is constituted by the Gard type with conservative preterite forms (e.g. Monteils; Table 17), historically present in the Gévaudan. This type displays a third, additional series of desinences, unique to the preterite: -èm, -ètz (-[ˈɛn], -[ˈɛs]). The low token frequency and low paradigmatic type frequency of the unique first and second person plural preterite forms identify them as preferential targets for analogical remodelling.

Table 17 Finite synthetic forms of vendre ‘sell’ in the variety of Monteils (Gard type; ALLOr 30.07). Above, orthographic forms according to ‘classical’ system; below, IPA conversion from original Gilliéron–Rousselot transcription.

6.3 The metamorphome PYTA

Also of significance in these data is the consistency of inflectional formatives within the metamorphome PYTA, illustrated in Section 4.2 . Imperfect subjunctive forms in -èssem, -èssetz (-[ˈɛsen], -[ˈɛses]), are consistently associated with preterite forms in -èrem, -èretz (-[ˈɛɾen], -[ˈɛɾes]) while imperfect subjunctive forms in -essiam, -essiatz (-[eˈsjan], -[eˈsjas]) are commonly associated with preterite forms in -eriam, -eriatz (-[eˈʀjan], -[eˈʀjas]), as shown in Tables 14 and 15; see also Figure 3.

Thus, for a variety of the type exemplified in Table 15, if the original preterite forms in -èm, -ètz (-[ˈɛn], -[ˈɛs]) undergo replacement, the novel forms are predicted to be of the -iam, -iatz type (-[ˈjan], -[ˈjas]). Such replacement would more closely resemble the metamorphome-internal levelling of personal desinences described for Daco-Romance in Section 3.1, in which a given rightmost exponent is spread to all cells with a given person/number combination within the metamorphome. The introduction of originally imperfect subjunctive forms into the preterite produces consistency between the desinences of the preterite and the imperfect subjunctive (i.e. within PYTA), in addition to replacing desinences of low paradigmatic type frequency with desinences of higher paradigmatic type frequency.

Among the notable properties of the Gévaudan development is that, at the historical period during which speakers sought to extend the identity between the preterite desinences and the imperfect indicative forms of èstre ‘be’, forms closely resembling the template ‘PYTA stem + siam/siatz’ were already available within the paradigm of each lexeme. Where imperfect indicative forms such as seiam, seiatz and potentially siam, siatz (see Section 5.2) did not previously occur within the paradigm of èstre ‘be’, imperfect subjunctive forms ending in -siam, -siatz were already familiar to speakers, occurring with equal lexical type frequency to the preterite forms they would ultimately replace, and could simply be co-opted into the preterite.Footnote ¹⁸