1 INTRODUCTION
Most historical accounts of French are ‘unidimensional [. . .,] treating language change in a social and demographic vacuum’ (Lodge, Reference Lodge2004: 250). This paper brings a historical sociolinguistic perspective to aspects of a well known diachronic change in the history of French, the loss of verb-second (V2) in favour of subject-verb-object (SV) ordering in main declaratives (Adams, Reference Adams1987). The analysis focuses on a subset of main declaratives, those preceded immediately by a tensed subordinate clause (see Combettes, Reference Combettes, Béguelin, Avanzi and Corminbœuf2010; Grad, Reference Grad1956, Reference Grad1961). According to Vance, Donaldson, and Steiner (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 318), this context is perhaps the first to reveal V2 loss in Old French (OF). The study exploits intra-textual variation between represented speech and narrative in four major OF prose texts. The data reveal stylistic variation between these two registers and also gender variation, in which the represented speech of women appears more conservative than that of men. These findings refine the findings of Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010), document significant syntactic variation across different registers within individual texts, and corroborate hypotheses by Foulet (Reference Foulet1928) and Herman (Reference Herman1954) on the origins of SV in French. Specifically, the data support the view that matrix SV originated in OF informal speech before diffusing upward to higher registers and eventually the written language; couched in modern terms, this pattern represents change from below (Labov, Reference Labov1994: 78). The paper is organised as follows: Section 2 presents both a descriptive overview and generative analyses of V2 in OF. Section 3 discusses relevant aspects of work in historical sociolinguistics. Section 4 details the methods and gives an overview of the data. Section 5 presents the results, followed by discussion in section 6 and concluding remarks in section 7.
2 V2 IN OLD FRENCH
Since Thurneysen (Reference Thurneysen1892), OF has been considered a V2 language (e.g., Adams, Reference Adams1987; Buridant, Reference Buridant2000; Foulet, Reference Foulet1928; Marchello-Nizia, Reference Marchello-Nizia1995; Roberts, Reference Roberts1993; Vance, Reference Vance1997).Footnote 1 In descriptive terms, the finite verb is often the second constituent in main declaratives (see Holmberg, in press), as in (1).Footnote 2
- (1)
This tendency was already strong in twelfth-century OF: in data from the prose Quatre livre des reis (ca. 1170), the verb is second in 81.4% (577/709) of main declaratives (Herman, Reference Herman1954: 362). By the early thirteenth century, the V2 schema appears nearly categorical: in Vance's (Reference Vance1997: 38) sampling of the prose Queste del saint graal (ca. 1225), the verb appears second in 99.5% (1738/1746) of main declaratives. Old French presents another characteristic feature of a V2 grammar, in that the preverbal position accepts nearly any type of maximal projection (XP), including noun phrases (subject or object), prepositional phrases, adverbial phrases, and so forth, resulting in a range of possible word orders. Marchello-Nizia (Reference Marchello-Nizia1995: 52), who limits her analysis to declaratives that contain an overt nominal object, catalogs some 17 distinct word-order permutations in twelfth- and thirteenth-century OF. The inclusion of intransitive declaratives or pronominal objects further swells this inventory. Furthermore, unlike in modern French, subjects in OF main declaratives are frequently null, and overt subjects can occur both preverbally and postverbally. In the panoply of OF main declarative word orders, SV – which in later stages of French becomes the principal declarative order – occurs as a possible but not dominant option.
Den Besten's (1983) analysis of modern German V2 provides the basis for generative accounts of OF V2 (e.g., Adams, Reference Adams1987; Roberts, Reference Roberts1993; Vance, Reference Vance1997; Vanelli, Renzi, and Benincà, Reference Vanelli, Renzi and Benincà1985), in which main declaratives are analysed as CP projections.Footnote 3 Verb-second effects result from the combination of (a) obligatory verb raising to C0 to satisfy uninterpretable features such as case or agreement (e.g., Roberts, Reference Roberts2010: 168; Vance, Reference Vance1997: 27) and (b) saturation of the preverbal position (SpecCP) by a single raised XP (Holmberg, in press; Roberts, Reference Roberts and Gallego2012: 396; Vance, Reference Vance1997: 11). The clausal architecture of OF main declaratives can be represented as in (2), adapted from Adams (Reference Adams1987: 8).
(2) [CP XP [C’ Vi [+fin] [TP (subject) [T’ [VP . . . ti . . .]]]]]
This structure explains another generalisation of OF syntax, namely that subordinates typically present SV ordering (Posner, Reference Posner1997: 377; Vanelli, Renzi and Benincà, Reference Vanelli, Renzi and Benincà1985: 169), especially during the time period of interest in this study (Vance, Reference Vance1997: 133).Footnote 4 Following den Besten (Reference Besten and Abraham1983), verb movement to C0 is blocked in subordinates by an overt complementiser, which occupies the same position (C0) that attracts finite verbs in main declaratives. As a result, in a prototypical subordinate, the verb raises only as high as TP/IP and thus remains under the subject.
Returning to main declarative orders, OF sometimes allows the verb to occur in third (V3) or higher position. One such context, central to the present paper, involves a main declarative preceded immediately by a subordinate, as in (3). The matrix finite verb in (3) is the third XP, preceded by both a tensed subordinate and the adverb si ‘thus.’Footnote 5
- (3)
Following Rizzi's (Reference Rizzi and Haegeman1997) proposal of an extended CP-domain, recent work (e.g., Benincà, Reference Benincà, Zanuttini, Campos, Herburger and Portner2006; Labelle, Reference Labelle2007; Mathieu, Reference Mathieu and Arteaga2013; Meklenborg Salvesen, Reference Meklenborg Salvesen and Lohndal2013) has posited a richly articulated clausal left-periphery to account for V3 declaratives within the V2 system of OF and medieval Romance. The consensus that emerges is that left-peripheral material does not count for V2 purposes.Footnote 6 That is, left-peripheral elements are syntactically optional and occur outside – to the left – of the core of the main declarative in which V2 operates. In Donaldson (Reference Donaldson2012), I argue that OF initial subordinates usually occupy a left-peripheral Frame (i.e., topic) position. Thus, in (3), the quant subordinate is followed by a typical V2 declarative (beginning with the adverb si) whose grammaticality and completeness are independent of the left-peripheral material.
2.1 Loss of V2 in French
Verb-second was lost in French in favour of SV as a basic word order, and modern French is said to be a SVO language (Harris and Vincent, Reference Harris and Vincent1988: 227).Footnote 7 Most accounts place the beginning of this change in the early fourteenth century (e.g., Adams, Reference Adams1987: 26; Roberts, Reference Roberts1993: 142–143), corresponding approximately to modern divisions between Old and Middle French (Buridant, Reference Buridant2000: 23; Hall, Reference Hall and Valdman1972: 215). By sometime in the fifteenth (Posner, Reference Posner1997: 359) or sixteenth century (Roberts, Reference Roberts1993: 197) – roughly the end of Middle French – the V2 grammar, no longer productive, was completely lost. Although restricted contexts in modern French allow matrix subject-verb inversion – one characteristic feature (among others) of V2 grammars – modern French is not a V2 language.Footnote 8
Of the many word order possibilities in OF main declaratives, SV gained in importance to emerge as the dominant order in the new non-V2 grammar. From the sizeable literature on this change (e.g., Adams, Reference Adams1987, Reference Adams1989; Foulet, Reference Foulet1928; Herman, Reference Herman1954; Lemieux and Dupuis, Reference Lemieux, Dupuis, Battye and Roberts1995; Marchello-Nizia, Reference Marchello-Nizia1995; Roberts, Reference Roberts1993; Vance, Reference Vance, Battye and Roberts1995, Reference Vance1997; Vanelli et al., Reference Vanelli, Renzi and Benincà1985) emerges a common view that increasing frequencies of matrix SV in OF prompted grammatical reanalysis, such that speakers eventually acquired SV as the basic main declarative structure. As the V2 grammar lost ground and proportions of SV increased, progressively fewer phrases displayed unambiguous V2 behaviour. That is, the input available to each new generation of speakers contained fewer and fewer declaratives that could be accounted for uniquely under the V2 grammar.
In generative terms, evidence for verb raising to C0 was progressively lost, and in the innovative SV grammar, the uninterpretable features previously associated with C0 were now realised in TP/IP (Adams, Reference Adams1987: 25; Roberts, Reference Roberts1993: 153; Vance, Reference Vance1997: 319, 351; Vanelli et al., Reference Vanelli, Renzi and Benincà1985: 171–172; but see Lemieux and Dupuis, Reference Lemieux, Dupuis, Battye and Roberts1995: 103). Dwindling evidence for verb raising to C0 culminated in a parameter resetting, such that speakers of French came to analyse matrix SV in TP/IP, without verb raising to C0 (see Vance, Reference Vance1997). This reinterpretation, as schematised by Adams (Reference Adams1987: 25), appears in (4).
- (4)
Beyond V2, the loss of verb raising to C0 also entailed the loss of simple inversion and postverbal matrix null subjects (Roberts, Reference Roberts1993: 187).
Although the V2 grammar is still active in thirteenth-century OF – the period of interest in this study – the steady rise of matrix SV over this period is also empirically attested. Setting aside for the moment questions as to why SV increased, the important observation concerns grammatical instability. As Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 317) argued, if the grammar of OF was stable throughout the thirteenth century, steady increases in SV would be unanticipated. Assuming a model of change like Lightfoot (Reference Lightfoot, Joseph and Janda2003), the quantity of cues that unequivocally compel a speaker to posit a V2 grammar decreases as frequencies of matrix SV increase, given that SV is ambiguous as a cue, compatible with either a V2 or a SV grammar (Adams, Reference Adams1987: 25). Computational models demonstrate that relatively small frequency fluctuations can trigger grammatical change. For example, Niyogi and Berwick (Reference Niyogi and Berwick1997: 197–198) show, specifically for OF V2 loss, that even a 10% increase in matrix SV is sufficient for V2 to be lost within several generations of speakers.
The rise of SV and concomitant decline of V2 was gradual, affecting generations of speakers over the course of Old and Middle French. Kroch's (Reference Kroch1989, Reference Kroch, Baltin and Collins2001) notion of grammars in competition captures the progressive nature of such change. According to Kroch, as a change transpires, speakers become diglossic, possessing both the conservative, outgoing V2 grammar as well as the innovative, incoming SV grammar (1989: 212). During a change, diglossic speakers are capable of producing either variant, with preferences for the incoming and outgoing variant determined by, among many other influences, social factors (discussed subsequently in section 3). In Kroch's model, during the transition from V2 to SV in French, matrix SV could reflect either an underlying conservative V2 structure (within CP) or an innovative SV structure (within TP/IP; see Lemieux and Dupuis, Reference Lemieux, Dupuis, Battye and Roberts1995: 82; Roberts, Reference Roberts1993: 156). A problematic consequence facing the modern analyst is the inability to discern, in most cases, whether the underlying structure of matrix SV is conservative or innovative. As V2 was progressively lost, however, the proportion of innovative (non-V2) structures must inevitably have increased, given that SV without verb raising to C0 eventually supplanted V2 entirely. In keeping with Kroch's reasoning, Clark and Roberts (Reference Roberts1993: 331) argue that ‘it is likely that the non-V2 parse for SVO . . . was a close rival for the V2 parse, even in (later) OF.’ One way to interpret this situation is to assume, as I do here, that increases in surface SV over time in OF mirror the decline of V2.
2.2 Initial subordinate + main declarative
Traditional accounts place the beginning of V2 loss around 1300. Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010), however, identified a context in which SV supplants orders unique to V2 at an earlier date. Specifically, in main declaratives preceded immediately by a subordinate clause, proportions of SV eclipse proportions of SV in declaratives more generally. Examples of initial subordinate + main declarative sequences appear in (5) and (6); the initial subordinate is italicised.
(5)
(6)
In (5), the matrix sequence si + verb + subject pronoun is characteristically V2. By contrast, (6) presents matrix SV, possible under either a V2 or a SV grammar. In their classification of matrix word orders, Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 307) contrasted SV with any order unique to the V2 grammar (and not a SV grammar; e.g., postverbal subjects in Germanic inversion, null subjects, si + verb).
Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010) examined thirteenth- and early fourteenth-century OF prose and found that proportions of matrix SV varied according to the type of initial subordinate. Table 1, adapted from Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 308–310), contrasts proportions of matrix SV with non-SV orders in four major texts. Data are given for three frequent types of initial subordinate: se ‘if,’ quant ‘when,’ and por ce que ‘because.’ Following se, proportions of matrix SV are high, even in the early thirteenth century. Following quant, a clear evolution is visible: proportions of matrix SV climb, steadily and then rapidly, throughout the century. Finally, following por ce que, matrix SV is not attested until a later date, and in the latest text, por ce que occasions the lowest proportions of matrix SV.Footnote 9 The differences in matrix word order following each type of initial subordinate led Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 310) to consider se an innovative context, quant a less innovative context, and por ce que a conservative context.Footnote 10
Table 1. Rise of matrix SV following initial subordinates (adapted from Vance et al. Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 308–310)
Note. ‘V2’ refers to any word order compatible with a V2 but not a SV grammar; ‘SV’ refers uniquely to surface subject-verb sequences.
Table 1 shows that SV has nearly ousted non-SV orders in main declaratives preceded by certain initial subordinate types, even during the ‘classical’ OF period (see Buridant, Reference Buridant2000: 23). Especially noteworthy are the high proportions of SV after se, even in Villehardouin. By comparison, proportions of SV in main declaratives not preceded by a subordinate are markedly lower. In Villehardouin, the first 200 main declaratives not preceded by a subordinate evince SV at a rate of only 32%; after se, Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 309) found a rate of 88% (see Table 1). Similar results obtain for the Queste. Whereas initial se triggers a rate of 76% SV in immediately following declaratives, other main declaratives evince only between 36% (Marchello-Nizia, Reference Marchello-Nizia1995: 82) and 46% SV (Vance, Reference Vance1997: 38), depending on the extract and coding method.Footnote 11 These findings led Vance et al. to argue that sequences of subordinate + main declarative provided ‘perhaps the earliest available evidence’ of V2 loss in OF (2010: 318).
Over the course of the Old French period, sequences of initial subordinate + main declarative increased in frequency. Whereas in the Chanson de Roland (ca. 1100), only about 3.5% of main declaratives are preceded by a subordinate (Donaldson, Reference Donaldson2012: 1044), the figure climbs to 12% in Villehardouin (ca. 1208) and 15% in Joinville (ca. 1306). By the thirteenth century, it is tenable that initial subordinate + main declarative sequences were sufficiently frequent to influence V2 loss: as Roberts (Reference Roberts2007: 321) observed about algorithms that model V2 loss in French (including Niyogi and Berwick, Reference Niyogi and Berwick1997), ‘V2 becomes unstable if only a very small proportion of non-V2 examples are introduced into’ the input for the algorithm.
3 HISTORICAL SOCIOLINGUISTICS
Living languages are by nature heterogeneous, as Weinreich, Herzog, and Labov (Reference Weinreich, Labov, Herzog, Lehman and Malkiel1968: 99–100) claim in their discussion of ‘orderly heterogeneity’, and part of the orderliness in linguistic variation is social in nature. Sociolinguistic patterns in past states of languages cannot be observed directly, but neither can their existence be doubted (J. Milroy, Reference Milroy1992: 52). Commentary from the classical Latin period already alludes to socially patterned variation, including regionalisms and differences between urban and non-urban speech (Adams, Reference Adams2007: 117, 126). Similarly, Dees (Reference Dees1980, Reference Dees1987) details regional variation among the OF dialects, and Lodge (Reference Lodge1993: 35) shows that this type of variation was salient to OF speakers. Distinctions between urban (Parisian) and non-urban varieties emerge in thirteenth- and fourteenth-century OF texts (Lodge, Reference Lodge2004: 86–97). Moving beyond diatopic variation more squarely into the purview of modern sociolinguistics, linguistic forms index socio-economic status to some degree in French texts by the fifteenth century (Lodge, Reference Lodge2004: 136–137).
Despite the inherent limitations of historical data (Labov, Reference Labov1994: 11), work in historical sociolinguistics has successfully adapted modern approaches to reveal previously unnoticed sociolinguistic patterns in historical data. Romaine's (Reference Romaine1982) seminal study revealed stylistic variation across text registers in Middle Scots: formal genres like legal prose evinced greater syntactic complexity than informal genres like letters and narrative prose (1982: 152–157). With respect to diachronic change in early modern English, Nevalainen and Raumolin-Brunberg (Reference Nevalainen and Raumolin-Brunberg2003) examined the social stratification of six changes in progress to ascertain whether they represented changes from above or below (Labov, Reference Labov1994: 78), and in further work, Nevalainen, Raumolin-Brunberg, and Mannila (Reference Nevalainen, Raumolin-Brunberg and Mannila2011) identified linguistically innovative and conservative speakers, giving some idea of who led these changes. Other work has focused on regional origins of changes: Lodge (Reference Lodge2004: 59–68) examines dialectal isoglosses to identify the regional provenance of nine phonological and morphological variants that ousted local variants in medieval Parisian French.
3.1 Textual register variation
Romaine (Reference Romaine1982: 15) notes the ‘primacy of speech’ in modern sociolinguistics. For most historical investigations, however, only written data are available. However, as Koch and Oesterreicher (Reference Koch and Oesterreicher1985) and Koch (Reference Koch, Selig, Frank and Hartmann1993) show, different types of texts resemble the spoken language to different degrees. Printed interviews and diaries, for instance, constitute a more oral genre than administrative regulations. During changes in progress, innovative variants appear more robustly, and at an earlier date, in oral-like texts than in more formal texts (Nevalainen and Raumolin-Brunberg, Reference Nevalainen and Raumolin-Brunberg2003: 29). Studies by Kytö (Reference Kytö1993) on changes in the English third-person singular present tense (sayeth vs. says) and Ayres-Bennett (Reference Ayres-Bennett2004) on the decline of inversion as the canonical interrogative form in French confirm that ‘the level of formality of a text should influence the choice of the form’ in the context of changes in progress (Kytö, Reference Kytö1993: 127). The distribution of conservative and innovative variants is thus modulated by register, and informal and formal texts represent distinct registers or situations of use (Halliday, Reference Halliday1978; Finegan and Biber, Reference Finegan, Biber, Eckert and Rickford2002).
Beyond variation across textual genres (inter-textual variation), individual texts sometimes contain multiple registers or styles (intra-textual variation), as Romaine (Reference Romaine1982: 157) found. One type of intra-textual variation contrasts narrative passages with passages intended to report speech. The extract from the OF Queste del saint graal in (7) contains both narrative and ‘speech,’ the latter shown here in italics.
(7) Et ele descent et vient devant le roi; si le salue, et il dit que Diex la beneie. ‘Sire, fet ele, por Dieu, dites moi se Lancelot est ceenz.’ – ‘Oïl voir, fet li rois, veez le la.’ Si li mostre. Et ele va maintenant la ou il est et li dit: ‘Lancelot, je vos di de par le roi Pellés que vos avec moi venez jusqu'en cele forest.’
She alighted and went straight to greet the king, who wished God's blessing on her. ‘Sire,’ she said, ‘in God's name, tell me if Lancelot be here.’ ‘Yes, indeed, there he stands,’ he answered, and pointed him out. And at once she went up to him and said: ‘Lancelot, in the name of King Pellés, I bid you accompany me into the forest.’ (English translation from Matarasso, Reference Matarasso1969: 31)
Even without modern punctuation, certain utterances obviously represent speech. Typically introduced by a verbum dicendi (e.g., dire ‘say,’ faire ‘do, say’), such passages contain ‘textual dialogue’ (Fleischman, Reference Fleischman1990: 65), ‘reported speech’ (Romaine, Reference Romaine1982: 158), or ‘quoted’ or ‘indirect’ speech (Ayres-Bennett, Reference Ayres-Bennett2001: 161, Reference Ayres-Bennett2004: 30). Price (Reference Price1971: 147) distinguished between ‘conversational’ and ‘narrative’ passages, and Marchello-Nizia (Reference Marchello-Nizia, Guillot, Combettes, Lavrentiev, Oppermann-Marsaux and Prévost2012) spoke of ‘l'oral représenté’ (represented oral [language]), with an insistence on represented. I adopt ‘represented speech’ to capture, on the one hand, the opposition with narrative, but also to acknowledge that represented speech is not a direct and faithful rendition of actual spoken language. Although represented speech might ‘approximate speech to some extent’ (Romaine, Reference Romaine1982: 125), it remains artificial, filtered through the grammar of the author, who, in the medieval period, possessed a degree of education well above that of the general population (Posner, Reference Posner1994: 92).
Regardless of how closely represented speech mirrors actual speech, Romaine (Reference Romaine1982: 157–158) justified analysing it as a separate register from narrative. Logically, at the intra-textual level, represented speech is closer to the spoken language than narrative, just as, across text genres, informal texts are closer to speech than formal texts. It follows that, for changes in progress, represented speech should be more grammatically innovative than the narrative in the same text. Previous work on changes in French substantiates this claim. For example, subjects came to be obligatorily expressed after the OF period. Intra-textual variation in OF is prescient, in that represented speech evinces significantly higher rates of subject pronoun expression than narrative (Price, Reference Price1971: 148; Vance, Reference Vance1981, cited in Vance, Reference Vance1997: 245). Furthermore, a similar change in early Italian yields analogous findings (Palermo, Reference Palermo1998). For a more recent development in French, loss of negative ne, Martineau and Mougeon (Reference Martineau and Mougeon2003: 131, 135) uncovered fine-grained intra-textual variation in nineteenth-century theatrical works. In their data, the represented speech of low-class characters, with high rates of ne-deletion, was more innovative than that of the linguistically conservative high-class characters, who deleted ne less frequently. Variation of this type in a single text suggests that the author has consciously or unconsciously varied his or her register for socio-stylistic reasons. Such a claim is consonant both with Weinreich et al. (Reference Weinreich, Labov, Herzog, Lehman and Malkiel1968: 101), who argue that mastery of heterogeneous forms (i.e., multiple options or variants) is an integral part of monolingual competence, and with Kroch (Reference Kroch1989, Reference Kroch, Baltin and Collins2001), whose model of language change posits competing conservative and innovative grammars within individual speakers.
3.2 Gender variation
Beyond intra-textual register variation, the present data permit an investigation of variation between the represented speech of men and women. Linguistic gender variation is ‘an apparently universal phenomenon’ (L. Milroy, Reference Milroy1992: 178) in modern society and represents ‘the single most consistent finding to emerge from sociolinguistic work around the world in the past thirty years’ (Trudgill, Reference Trudgill2000: 73). Whereas commentaries from antiquity identify women as linguistically conservative (Key, Reference Key1996: 131; Yaguello, Reference Yaguello1978: 51), women are frequently identified as leaders of linguistic change in modern societies (Labov, Reference Labov2001: 283; Romaine, Reference Romaine, Ammon, Dittmar and Mattheier1988: 1456) and have also been found to lead some changes in historical English data (Nevalainen et al., Reference Nevalainen, Raumolin-Brunberg and Mannila2011). Importantly, as Labov (Reference Labov2001: 277; also Bucholtz, Reference Bucholtz2002) noted, linguistic differences between men and women arise for social rather than biological reasons.
Fundamental differences in the status of men and women in medieval French society entail that linguistic gender variation must have existed, whether or not such variation is reflected in extant texts. The most salient contrast was one of public versus private sphere. Whereas men were ‘overburdened with the demands of government and war’ (Ennen, Reference Ennen1989: 83) and business outside the domestic arena, women remained primarily in the domestic sphere (Power, Reference Power1975: 47, 50). Thus, in medieval times, the ‘center of respectable female activity’ was the home (Rees Jones, Reference Rees Jones, Erler and Kowaleski2003: 190). Contemporary evidence comes from a French nobleman, Guibert de Nogent, who lived in the eleventh or twelfth century and whose autobiography recounts aspects of his mother's domestic duties, including child-rearing (Amt, Reference Amt2010: 122–123). One consequence is that, whereas men were mobile in medieval society, women as a rule were not (Riddy, Reference Riddy, Erler and Kowaleski2003: 220). The few cases in which women did have a visible public presence had little effect on the overarching social roles of the sexes. Women involved in trade, for example, were generally assistants to men, and there were no gilds specifically for tradeswomen (Power, Reference Power1975: 61). Among the nobility, women who held powerful positions (e.g., the queen) were often foreigners – and hence frequently non-native speakers – imported through marriages designed for political alliances (Ennen, Reference Ennen1989: 75).
3.3 Origins of SV in French
The present paper examines the rise of SV and concomitant decline of non-SV orders in sequences of initial subordinate + main declarative. Scholars of OF have hypothesised that SV emerged in speech much earlier than it appeared robustly in writing (see Posner, Reference Posner1997: 349). Foulet (Reference Foulet1928: 351) speculated that SV became characteristic of the spoken language in the twelfth century, driven in part by the need to unambiguously mark grammatical relations as nominal declensions were lost. Herman (Reference Herman1954: 85–86) concurred that SV originated in informal spoken language, as a pragmatically driven word order, a position compatible with Price's (Reference Price1971: 147–148) view that SV emerged in preliterary periods.
If SV originated in OF (informal) speech, one could hypothesise on the basis of previous historical sociolinguistic findings that it should appear more robustly, and probably at an earlier date, in texts or textual registers that lie close to the oral end of Koch and Oesterreicher's (Reference Koch and Oesterreicher1985) continuum (also Koch, Reference Koch, Selig, Frank and Hartmann1993). Conversely, more formal (less oral) texts or registers should contain higher proportions of conservative non-SV variants. If present, such a skewing of innovative and conservative variants by register would mirror findings for other historical changes in progress (Ayres-Bennett, Reference Ayres-Bennett2004; Kytö, Reference Kytö1993; Martineau and Mougeon, Reference Martineau and Mougeon2003; Price, Reference Price1971). Taking the hypothesis a step further, the represented speech of men may differ quantifiably from that of women, in light of men and women's fundamentally different social roles in medieval France.
4 STUDY
The present study examines intra-textual register variation in OF and reveals socio-stylistic patterning in the distribution of conservative and innovative variants during a syntactic change in progress, the loss of V2. The analysis focuses on matrix word order in sequences of initial subordinate + main declarative. First, I examine register differences between represented speech and narrative to ascertain whether they are stylistically distinct. If proportions of innovative and conservative variants are consistently modulated by register, the results will allow further empirical evaluation of the hypothesis that SV originated in the spoken language. I also seek traces of sociolinguistic patterning in the distribution of conservative and innovative variants by contrasting the represented speech of women and men.
The data come from the four thirteenth- and early fourteenth-century OF prose texts studied by Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010): Villehardouin's Conquête de Constantinople (ca. 1208), the Queste del saint graal (anonymous, ca. 1225), the Roman de Cassidorus (anonymous, ca. 1267), and Joinville's Vie de Saint Louis (ca. 1306). Each text was analysed in its entirety.Footnote 12 In Cassidorus, the occasional short verse sections were excluded. These texts were chosen by Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010) based on regional provenance, genre, and dating, without regard to an eventual historical sociolinguistic analysis (see Ayres-Bennett, Reference Ayres-Bennett2001: 162). As a result, the corpus does not present a fully elaborated stylistic continuum akin to what Romaine (Reference Romaine1982: 125) advocated. Nonetheless, as the results will demonstrate, the texts contain sufficient intra-textual variation to merit investigation. An additional difficulty is that these texts were likely all penned by men. This is clearly true for Villehardouin and Joinville and probably also for the anonymous Queste and Cassidorus, as literary works by women – Marie de France and Christine de Pisan excepted – are scarce in medieval French. As a consequence, the data contain representations (by male authors), rather than direct evidence of women's speech or language produced directly by women. As a reviewer points out, it is audacious from a methodological point of view to suppose that representations of women's speech by male authors might be sensitive enough to reflect potential differences in men's and women's speech. Here again, however, the patterns that emerge from the data will demonstrate the interest of examining the effects of gender in represented speech.
The analysis included all instances of a main declarative immediately preceded by a tensed initial subordinate clause headed by a ce que ‘so that,’ ainsi que/comme ‘so that,’ apres ce que ‘after,’ devant ce que ‘before,’ en ce que ‘in that,’ endementiers que ‘while,’ por ce que ‘because, so that,’ po(u)r que ‘because,’ puis que ‘since,’ quant ‘when,’ se ‘if,’ selon ce que ‘according to,’ si comme ‘just as,’ si tost que/comme ‘as soon as,’ and tandiz que ‘as long as.’ Orthographic variants of each type of clause were included. To be considered initial, the subordinate could be preceded by no more than a coordinating conjunction (e.g., et ‘and,’ car ‘because’). If another constituent type preceded the subordinate, the subordinate was by definition not initial, and the sequence was not included. For example, the subordinate in (8) is not initial: it intervenes between the subject and verb, acting as an interpolation (Grad, Reference Grad1961: 6–7; Skårup, Reference Skårup1975: 420–424) or a parenthetical (Vance, Reference Vance1997: 194n). Furthermore, such sequences, which by definition present SV order (albeit interrupted), do not present the type of variation attested after initial subordinates that is of interest here.
- (8)
Similarly, I excluded a handful of examples in which an (apparently initial) subordinate was separated from the main declarative by a complementiser or a coordinating conjunction.
A total of 2,911 tokens of initial subordinate + main declarative were analysed. The different types of subordinate did not occur with equal frequencies. Whereas the token counts for initial se (n = 608), quant (n = 1936), and por ce que (n = 108) were deemed sufficient for separate analyses, other types yielded low token counts. Apres ce que ‘after’ occurs only nine times in the corpus, in a single text; an extreme example is a ce que ‘because,’ a hapax in this corpus. To allow for a meaningful quantitative analysis, I followed Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 308) in conflating these low frequency types, all formed with an adverbial plus the complementiser que, into an umbrella category, designated as __que clauses (n = 259).Footnote 13
Each token was coded for (a) type of subordinate; (b) the exact word order of the declarative; (c) narrative or represented speech; and for represented speech, (d) whether the speaker was male or female. Like Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 307; see also Donaldson Reference Donaldson2012: 1034), I opposed SV and all orders that could be generated by a V2 – but not SV – grammar. Thus, all instances of an immediately preverbal subject (with optional preverbal object clitics or negative ne) in the main declarative were counted as SV.Footnote 14 In contrast, orders characteristic of the OF V2 grammar but not of the emerging French SV grammar were counted as V2; these include declaratives with postverbal subjects, null subjects, or the sentence adverb si or any other non-subject constituent immediately before the verb. Represented speech was identified as the discourse directly following a verbum dicendi. Passages in which the author directly addressed the reader in narrative first-person, frequent in Cassidorus, were coded as narrative. Tokens from talking animals and mysterious voices (in the Queste and Cassidorus) were analysed as represented speech but were not assigned gender. A total of 647 tokens could clearly be attributed to a human speaker; of these, 494 were from men.
5 RESULTS
Table 2 presents the overall results, with no division between narrative and represented speech; the texts are treated as internally homogeneous. The results are similar to those reported by Vance et al. (Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010: 308–310; see also Table 1) and confirm these authors’ observation that the initial subordinate type conditions the behavior of the following main declarative, with initial se triggering the highest rates of SV, followed by quant, other __que subordinates, and finally por ce que. Following initial se, matrix SV is already dominant at the beginning of the thirteenth century. After quant and the __que class, a sharp rise in matrix SV occurs in the second half of the century.Footnote 15 After por ce que, matrix SV is attested at a later date, and non-SV orders persist more strongly in this context than elsewhere.
Table 2. Overview of matrix word order after initial subordinates
Note. Texts: V = Villehardouin, Q = Queste, C = Cassidorus, J = Joinville. V2 refers to any non-SV order; SV refers uniquely to surface subject-verb sequences.
5.1 Register variation
The extent to which represented speech represents a stylistically distinct register from narrative is shown in Tables 3 through to 6. For initial se, the data in Table 3 reveal nuances without drastically altering the overall trends reported in Table 2. The direction of the variation is of interest, however: in each text, the represented speech shows (marginally) higher rates of SV than the narrative.
Table 3. Matrix word order after initial se: Represented speech and narrative
Table 4. Matrix word order after initial quant: Represented speech and narrative
Table 5. Matrix word order after initial __que types: Represented speech and narrative
Table 6. Matrix word order after initial por ce que: Represented speech and narrative
The data for quant clauses in Table 4 make a stronger case for register differences between represented speech and narrative. The figures for Villehardouin change dramatically, in that represented speech evinces only SV, although on the basis of only three tokens. In the Queste and Cassidorus, where the token counts are robust, represented speech varies considerably from narrative, revealing intra-textual stylistic heterogeneity. In the Queste, the global rate of matrix SV after initial quant (10%; see Table 2) masks important register differences. The represented speech, with a matrix SV rate of 36.3%, is clearly more innovative than the narrative, in which SV is barely represented (5.3%). By the time of Cassidorus, roughly 40 years later, proportions of SV have advanced in both represented speech and narrative but are still modulated by register. In represented speech, the rate of SV is 58.4%, compared to only 27.9% in narrative. In Joinville, unexpectedly, the narrative shows a slightly higher proportion of SV than the represented speech.
Examples of __ que subordinates do not appear in represented speech in Villehardouin or Joinville. In the Queste and Cassidorus, however, represented speech is again visibly more innovative than narrative, as Table 5 shows.
The data for initial por ce que, the most conservative type, also confirm that represented speech is more innovative than narrative (Table 6). Although the differences are less pronounced than after quant and __que, the direction of the variation between the two registers is consistent.
For each type of initial subordinate, distinguishing between represented speech and narrative yielded a more nuanced view of syntactic change than an analysis that treated each text as stylistically homogeneous. The direction of the variation shows that represented speech is regularly more innovative than narrative. Table 7 presents a summary of rates of matrix SV in represented speech and narrative for each initial subordinate type. When the represented speech is more innovative than the narrative, the cells are shaded; this is the case in 12 of the 16 pairs of cells. In two of the four remaining pairs of cells, no comparison is possible due to the non-occurrence of the initial subordinate type in represented speech, and in one pairing, SV is attested in neither narrative nor represented speech. In other words, of the 13 contexts in which a meaningful comparison is possible, 12 (92%) confirm that represented speech is in advance of narrative in favouring SV. The validity of the pattern is bolstered by several cases that reach statistical significance, indicated by asterisks and boldface type. These contexts are as follows: quant clauses in Villehardouin, Fisher's exact text, p < .05; quant clauses in the Queste, χ2 (1, N = 622) = 55.91, p < .0001; quant clauses in Cassidorus, χ2 (1, N = 884) = 13.02, p < .001; and __ que clauses in Cassidorus, Fisher's exact text, p < .05. For __ que clauses in the Queste, differences between represented speech and narrative approached significance, Fisher's exact test, p = .089.
Table 7. Percentage matrix SV in represented speech and narrative: Summary
*p < .05.
5.2 Gender variation
Both the Queste and Cassidorus contain represented speech of women in sufficient quantities to allow meaningful comparisons with that of men. On the other hand, Villehardouin and Joinville, both military chronicles, contain nearly no women characters, precluding a meaningful analysis of gender variation in these two texts.Footnote 16 The analysis in the Queste and Cassidorus is limited to a basic division by gender, as the size of the corpus rules out a valid analysis of other factors with potential social significance, such as age or social rank.
In both the Queste and Cassidorus, the represented speech of women is, as a rule, grammatically more conservative than that of men. Men are represented as more innovative, with higher rates of matrix SV after initial subordinate clauses than women. Table 8 presents the percentages of matrix SV (with raw token counts) following each type of initial subordinate. The represented speech of men is more innovative than that of women in six of the eight (75%) contexts; these pairs of cells are shaded. In the case of quant clauses in the Queste, the difference reaches statistical significance, χ2 (1, N = 91) = 3.98, p <. 05.
Table 8. Percentage matrix SV: Represented speech of men versus women
*p < .05.
6 DISCUSSION
Intra-textual variation surfaced in each text, a finding that justifies an analysis of represented speech and narrative as distinct registers rather than treating the texts as stylistically homogeneous. In some cases, such as after initial se, separating narrative and represented speech produced only slight divergences, but elsewhere, particularly after quant, the two registers evinced markedly different syntactic behaviour. The distribution of SV and V2 orders across represented speech and narrative reveals a consistent pattern in each text. Consistent with findings from other studies of changes in progress, the less formal, more oral-like register contains greater proportions of the innovative variant, and the narrative appears linguistically conservative in comparison. In other words, represented speech is consistently in advance of narrative in adopting SV.
In classifying SV as innovative (see also Vance et al., Reference Vance, Donaldson, Steiner, Colina, Olarrea and Carvalho2010), I make no direct claims about underlying syntactic structure, as surface SV sequences in later OF were generated both in CP, by the conservative V2 grammar, and in TP/IP, by the incoming SV grammar (Lemieux and Dupuis, Reference Lemieux, Dupuis, Battye and Roberts1995; Roberts, Reference Roberts1993), which were in competition (Kroch, Reference Kroch1989, Reference Kroch, Baltin and Collins2001). Instead, following Adams (Reference Adams1987), Clark and Roberts (Reference Clark and Roberts1993), and Roberts (Reference Roberts1993), I focus on frequencies of SV as they relate to mechanisms of change. Every increase in the proportion of matrix SV at the expense of unequivocally V2 orders reduces the unambiguous cues necessary to posit verb raising to C0 and hence to maintain V2. The presence of syntactic variation modulated by register in individual texts supports Kroch's (Reference Kroch1989, Reference Kroch, Baltin and Collins2001) model of diachronic change, in which individual speakers possess multiple competing grammars. The data show that represented speech facilitates the innovative grammar, whereas the conservative grammar persists more strongly in narrative.
The mastery of variant forms also represents a central tenet of sociolinguistics (Weinreich et al., Reference Weinreich, Labov, Herzog, Lehman and Malkiel1968). Register variation of the type documented here has a social dimension, as it is a type of socio-situational variation (or stylistic variation or intra-speaker variation; see Schilling-Estes, Reference Schilling-Estes, Chambers, Trudgill and Schilling-Estes2002: 375–376). To produce this particular register variation, the medieval French authors must have been sensitive to the syntactic change in progress, without necessarily being consciously aware of it. The variation between represented speech and narrative never approaches the deliberate caricature that Fleischman (Reference Fleischman, Bloch and Nichols1996: 403) described in Molière's use of the archaic adverb si to mock the speech of peasants. If the rise of SV is a change from below (Labov, Reference Labov1994), as I will suggest subsequently, then it is likely to operate ‘below the level of social awareness’ (Labov, Reference Labov2001: 279).
This is not to deny the likelihood that rates of SV varied across social groups in medieval France. Bell (Reference Bell1984: 151) argued that ‘variation on the style dimension within the speech of a single speaker derives from and echoes the variation which exists between speakers on the ‘social’ dimension’ (also Bell, Reference Bell, Eckert and Rickford2002; Finegan and Biber, Reference Finegan, Biber, Biber and Finegan1994, Reference Finegan, Biber, Eckert and Rickford2002). Indeed, for Romaine (Reference Romaine1980: 228), the link between register variation and social variation is the ‘classic sociolinguistic finding.’ At a given moment during a change in progress, conservative variants appear most robustly in formal registers, and innovative forms are perceived as less formal, less correct, or less careful (Kroch, Reference Kroch, Baltin and Collins2001: 723; Posner, Reference Posner1997: 71; Roberts, Reference Roberts2007: 324; Trudgill, Reference Trudgill2000: 69).
Although class-based stratification of the modern sociolinguistic type is not relevant in the feudal society of medieval France (see Lodge, Reference Lodge2004), the represented speech data reveal social patterning according to gender. The trend in Table 8 shows the represented speech of women to be more conservative than that of men. In light of the social differences between men and women in medieval France (section 3.2), the direction of this linguistic gender variation is not surprising (although, as discussed subsequently, the fact that gender variation appears in these texts at all is noteworthy). The linguistic conservatism of the women in the corpus can be explained partially by the social networks (Milroy and Milroy, Reference Milroy and Milroy1985) in which they participated. Men's participation in the public sphere, which took them away from the domestic sphere frequently and often for long periods of time, entailed geographic mobility and contact with speakers from a broad range of social groups. Women, in contrast, lacked such mobility (Riddy, Reference Riddy, Erler and Kowaleski2003: 220) and were largely confined to the domestic sphere, which in medieval times involved ‘intertwined living of people of different generations who ate, slept, and worked alongside one another’ in tight quarters (Riddy, Reference Riddy, Erler and Kowaleski2003: 216). Linguistically, men and women in medieval France would typically have participated in distinct social networks.
As J. Milroy (Reference Milroy1992: 177) argues, linguistic innovations diffuse most easily in loose-knit networks in which speakers ‘form (relatively weak) ties with very large numbers of others.’ As a generalisation, the role of men in medieval French society would have entailed participation in this type of loose-knit network. Women, on the other hand, would have interacted primarily with a closed set of close acquaintances, thereby participating in strong close-knit networks, which ‘function as a conservative force, resisting pressures to change from outside the network’ (J. Milroy, Reference Milroy, Bolton and Kwok1992: 121). As V2 declined in OF and rates of matrix SV increased in kind, women may have been more sheltered than men from the change in progress and, as a consequence, slower to adopt matrix SV after initial subordinates, as the present data show. Furthermore, noble women – who account for most of the represented speech data in this corpus – were well educated, perhaps more so than men (Ennen, Reference Ennen1989: 83), and could often read and write (Power, Reference Power1975: 86); high levels of education often favor the use of prestige or conservative variants (see discussion in Trudgill, Reference Trudgill2000: 200–203; also Labov, Reference Labov1972: 115).
Perceptions of women in medieval France could also have contributed to the conservatism of women's represented speech. Recall that these data contain representations, rather than direct evidence of women's speech, and that the two anonymous texts (Queste and Cassidorus), which provide all the data for gender variation in the present corpus, were likely penned by men. As a first observation, it is of interest that the male authors appear to have been sensitive to gender variation in the speech around them. Two arguments lend credibility to their representations: first, the direction of the gender variation in these data is consistent, in that women's represented speech is regularly more conservative than that of men; second, this situation aligns well with predictions from work on social networks. One conclusion is that this gender variation was sufficiently salient in the speech of the time to be perceived (consciously or unconsciously) by the authors of Queste and Cassidorus. Another (complementary) possibility arises from the fact that the represented speech of women in the corpus has been filtered through male authors. The authors would not have been immune to the dominant attitudes toward women, which were disseminated top-down, notably by the church and the aristocracy (Power, Reference Power1975: 9). Whereas such views considered women to be subordinate to men, medieval ethics also posited women as models of virtue (Bertini, Reference Bertini, Bertini, Cardini, Leonardi and Fumagalli Beonio Brocchieri1991: 18). It is reasonable to extend this ethic to include linguistic virtue as well, especially as linguistic conservatism is generally valued over innovation, whether in the Appendix Probi (third century AD; Herman, Reference Herman2000: 26) or scores of more recent prescriptive works. Trudgill (Reference Trudgill2000: 69) notes that women's language is often judged – or perceived – as better or more pure than that of men; furthermore, women are frequently subjected to a double standard, with respect to both behaviour and language (p. 74). That is, women are expected not only to behave better than men but also to speak better, especially, as a reviewer points out, in light of the paramount role of the mother in transmitting language to children (see also Labov Reference Labov2010: 199). Thus, women's speech was likely conservative by virtue of the types of social networks in which they participated and on account of high levels of education (at least among elite women), but medieval ethics and perceptions of women likely also contributed to how their speech was represented in the present data. If this is so, a confluence of factors underlies the conservative tendencies in the represented speech of women.
Foulet (Reference Foulet1928) and Herman (Reference Herman1954) hypothesised that matrix SV originated in the spoken language during the OF period; writing would have lagged behind in adopting the innovation (Posner, Reference Posner1997: 349). Although direct access to OF speech is impossible, represented speech, as a less formal and more oral register than narrative, nonetheless provides indirect evidence. The results reveal higher rates of matrix SV after an initial subordinate in represented speech than in the neighboring narrative, consistent with Herman's hypothesis that SV developed in informal registers and diffused upward, an example of Labov's (Reference Labov1994: 11) change from below. Sociolinguistic patterning would plausibly have accompanied the upward diffusion of SV. The indications of conservatism in the represented speech of women, whose affiliation with close-knit social networks probably impeded linguistic change, provide a glimpse of such patterning.
A final comment is reserved for the role of initial se clauses, which occur far more frequently in represented speech than in narrative (Table 3). This imbalance does not appear anomalous, as all the texts in the corpus demonstrate it to some degree, as do two earlier verse texts by Chrétien de Troyes that I consulted. In Erec et Enide (ca. 1165), 11 tokens of initial se occur in narrative versus 54 in represented speech; similarly, in Perceval (ca. 1185), 6 tokens of se occur in narrative versus 77 in represented speech. In each text in the corpus, quant and se are the most frequent types of initial subordinates (Table 2). Grad (Reference Grad1961: 5–6) observed that, whereas main declaratives after initial quant frequently present the order si ‘thus’ + verb, si is rare after initial se clauses. Indeed, given the ubiquity of si as the first constituent in OF matrix declaratives more generally (e.g., Marchello-Nizia, Reference Marchello-Nizia1985), its quasi-absence after se is surprising. In slightly different terms, why do initial se clauses strongly favour the appearance of SV in the following main declarative?
One possible answer involves subject continuity versus discontinuity between the initial subordinate and the main declarative. The two most frequent types of initial subordinate – quant and se – evince markedly different behaviour in this regard. After initial quant, there is a strong tendency toward subject continuity with the main declarative (Table 9); the Queste offers an extreme case, in that 100% of the data present subject continuity.
Table 9. Subject continuity between initial quant subordinate and main declarative
In stark contrast, after initial se, there is a strong tendency for subject discontinuity between the initial subordinate and the main declarative (Table 10).
Table 10. Subject continuity between initial se subordinate and main declarative
In each text, the difference in behaviour between initial quant and se clauses is statistically significant at the p < .0001 level (χ2 tests). Initial quant and se clauses establish different discourse structures with respect to subject continuity: the high rates of the adverb si, a marker par excellence of subject continuity (Fleischman, Reference Fleischman1991), are unsurprising following quant clauses. On the other hand, the subject discontinuity that frequently characterises sequences of se + main declarative necessitates switch-reference, for which preverbal subjects (and hence matrix SV) are a common marker.
The differential behaviour of subject continuity in the context of initial quant and se is already present in the Chanson de Roland (ca. 1100), where 60% of initial se clauses involve switch-reference, versus only 24% for initial quant. On the other hand, initial subordinates of any type are infrequent in Roland compared to later texts, as discussed in section 2.2. If the present data reflect trends in OF, then initial se clauses, more frequent than initial quant in the spoken language, provided one toehold for increased rates of matrix SV. Another possible factor enabling SV after se, as Nissen (Reference Nissen1943: 59) suggested, could have been a desire to avoid the sometimes homophonous se and adverb si within a single utterance.
7 CONCLUSION
This study exploited intra-textual variation in OF texts to add a socio-stylistic dimension to a syntactic change in progress, in which matrix SV superseded V2 orders, specifically in sequences of initial subordinate + main declarative. The results, although confined to what Lodge (Reference Lodge2004: 4) called ‘macro-level tendencies,’ reveal stylistic differences between represented speech and narrative. As in studies by Martineau and Mougeon (Reference Martineau and Mougeon2003), Price (Reference Price1971), and Vance (Reference Vance1981), represented speech was demonstrably in advance of narrative in adopting the innovation, and the higher rates of matrix SV in represented speech lend support to Foulet (Reference Foulet1928), Herman (Reference Herman1954), and others who have hypothesised that SV originated in the spoken language. The represented speech of women appeared more linguistically conservative – that is, slower to adopt SV – than that of men. Appealing to work on social networks (Milroy and Milroy, Reference Milroy and Milroy1985), I hypothesise in broad terms that women's social status in medieval France led them to participate primarily in strong close-knit networks. Although a credible investigation of socio-stylistic differences between individuals is not feasible given the small size of the present corpus, Nevalainen et al.'s (Reference Nevalainen, Raumolin-Brunberg and Mannila2011) work on early modern English demonstrates the interest of such an approach. It also remains to be seen if other diachronic changes in OF give rise to similar socio-stylistic variation and, if so, whether the patterns of variation mirror those reported here. For example, the rise of left-dislocated objects with pronominal reprise (Priestley, Reference Priestley1955) is cited as a contributing factor to the loss of V2 in French (Kroch, Reference Kroch1989: 214–215; Vance, Reference Vance, Battye and Roberts1995: 187), but the emergence and proliferation of this construction remain incompletely understood.Footnote 17
Finally, the especially innovative behaviour triggered by initial se subordinates merits investigation beyond the speculations put forth here, as do explanations of the conservative behaviour of initial por ce que subordinates.
Acknowledgements
I wish to thank the anonymous reviewers for their thoughtful comments and critiques. For their questions, discussion, and suggestions, I would also like to thank Carl Blyth, Barbara Bullock, Michael Johnson, Devan Steiner, Barbara Vance, and the audience at the 39th New Ways of Analyzing Variation (University of Texas at San Antonio). All remaining shortcomings are of course my own.
