Is Medieval French diglossic? New evidence on remnant V2 and register

An ongoing debate concerns the degree of diglossia of contemporary French, by which vernacular and normative registers display significant grammatical differences. Was diglossia characterizing Medieval French? This question is explored through the study of unambiguous V2 configurations. The word order has been shown to display rates of use and informational behavior correlating to register. The correlation is investigated in novel data relating to narration vs. dialogue and to correspondence by members of different social classes. Quantitative analysis shows that variation in V2 behavior remains determined by formality. The proposed methods thus help measure diglossia of previous states of languages.


Introduction
A language like French has been described as diglossic because it has a number of grammatical features that have differential realizations and behavior in the vernacular and in the normative practice (Massot 2010, Zribi-Hertz 2011, Rowlett 2013). The question is whether this is a new state of affair, or whether it was always thus. Answering that question is difficult because the available resources representing vernacular exchanges are few, and become fewer still the further one goes back in time.
Reconstructing vernacular grammar therefore relies on developing efficient methods.
Two methods are operationalized here. One is to compare narration and dialogue parts of 13 th century prose texts. The second is to consider the productions of members of different social classes in 15 th century correspondence. The sources are used to analyze the issue of one construction unambiguously instantiating verb-second word order in Medieval French. The analysis confirms that this declining word order configuration was already sensitive to register, and that in this respect Medieval French had features of a diglossic language.

Background
Diglossic languages display significant grammatical differences between vernacular and normative registers (defined as a continuum of distance along the lines of Koch & Oesterreicher 2001). Thus some syntactic phenomena are part of the immediate competence of speakers, and some are found in normative practice. An example is provided by post-verbal use of nominative clitics in French (known as 'subject inversion'). Such uses are absent from vernacular productions of contemporary speakers (Blanche-Benveniste 1997: 54, Culberston and Legendre 2008, Palasis 2013) who acquire it latter with a second-language level of command (Meisel, Elsig and Bonnesen 2011). This feature illustrates the diglossia of contemporary French, such that clitic subject inversion is exclusively a normative practice, and therefore not part of the ordinary competence of speakers.
Reconstructing the ordinary competence of speakers for earlier states of languages 1 can only be achieved from traces of written practice. Finding texts that reveal vernacular practice thus appears crucial. Such texts tend to belong to certain types, as proposed by Lodge (2009 : 212) and Ayres-Bennett (2020). Ayres-Bennett considers ego-documents (letters, diaries, travelogues) and represented speech, especially dialogue as compared to other textual modalities. The rate and behavior of grammatical features in dialogue and narration parts of the same text have been contrasted by a number of authors (Vance 1997: 245-246, Schøsler 2002, Rodríguez Somolinos 2003, Marnette 2006, Dufter 2010. Such comparisons have also been encouraged by the latter work on "represented orality" in literary texts (Marchello-Nizia 2012, Guillot et al. 2015, Guillot-Barbance et al. 2017, Lefeuvre & Parussa 2020. These studies have contributed to establish that dialogue is generally less conservative than narration (Glikman & Mazziotta 2014, Mazziotta & Glikman 2019; see also Donaldson 2018). As compared to narrative sequences, dialogue tends to contain a lower rate of declining features, and a higher rate of emerging features. The advantage of such of method is that one can compare different modalities within the same text, although it does not always yield the expected results (Steiner 2014, Pujol 2020, especially when not based on representation of real-life speech. On the other hand, Medieval records of actual dialogues presented in direct speech are exceedingly rare. Some exceptional resources are represented by the Old Bailey Corpus and the Salem Witch Trial for English, the Lio Mazor for Venetian and the Anglo-Norman Year Book Corpus for (Anglo-Norman) French, but they offer a limited temporal coverage, from 1720 to 1913 for the first, 1692-1693 for the second, 1312-1314 for the third, and 1280 to 1385 for the latter. Work based on the Year Books shows that they are less formal than e.g. petitions (Ingham 2016, Larrivée in press), and that accordingly they contain a lesser proportion of declining verb-second word order, but there has been no attempt to contrast narration and dialogue. As before, the expectation would be that dialogue should be less conservative than narration.
Another approach is to compare the speech, real or fictitious, of members of different social classes. While members of different social classes can adopt both vernacular and normative traits, there is an expectation that more vernacular registers will be used more by members of less privileged social groups. Such a comparison of class membership and register crucially relies on these dimensions being represented in a given text. This is the case in the 1422 prose allegory Quadrilogue invectif, which presents in turn speakers from different social orders, including a character meant to represent the people. Schøsler (2002) shows that the latter uses fewer null subjects than the representative of the clergy and of the knighthood. As expected, declining features are less represented in the vernacular practice, which is associated more readily to members of less privileged social groups.
Quantitative studies of dialogue and of correspondence are proposed in this paper. The method is applied to a declining configuration found in Medieval French. Verb-second word order (V2) is a configuration in which the verb follows an initial projection (XP) (Wolfe 2018 and references therein). Crucially, since Skårup (1975), Thiersch (1978) and den Besten (1982), both the initial XP and the verb are generally assumed to be in a high syntactic position, later identified as the left periphery. This is demonstrated by unambiguous configurations where the XP is a projection other than the subject, be it an argument or adjunct as in (1), a particle or connector as in (2) or a coordinator (3), that with the verb stands above the expressed subject. These are illustrated by the following examples from the beginning of 13 th century prose literary text Queste del Saint Graal.
( A number of recent studies have sought to characterize the trajectory and characteristic properties of Medieval French V2 at a period when it is being lost as a productive configuration. There is however considerable variance between the studies. Work on 13 th century literary prose cite figures of between 17% and 30% of V2 among all clauses (Radwan 2011, Salvesen and Bech 2014, Marchello-Nizia 2018, Kaiser and Zimmermann 2011, 13 th century prose legal sources yield 3.7% of unambiguous V2 (Larrivée 2019). Such quantitative variance between literary and non-literary sources is also found across the history of Portuguese, Galves 2019. As for the informational value of the XP that is thought to play an important role for the construction, whereas Larrivée (2019) and Ingham (2018) converge in finding virtually categorical discourse-old argument and adjunct XPs with unambiguous V2 in 13 th century prose, Labelle and Hirschbühler (2018) discover a near-categorical discourse-new Focus value for preverbal objects in verse texts between 1090 and 1200, and Radwan no specific value for V2 in 13th century literary prose. These issues are important because the former tells us whether V2 is a central word order of 13 th century French or merely a particular construction, as is the case in contemporary normative French (Guimier 1997); and the latter because it indicates whether the configuration is driven by formal syntactic requirements or informational dynamics. The hypothesis is that such variance relates to the register of the investigated texts, as suggested by Ingham (2012: 114, 117) and Ledgeway (2008) and documented by Larrivée (2019).
The role of register on rate of use and informational value for V2 configurations in Medieval French is verified using two complementary methods of examining narration vs. dialogue, and productions of socially-differentiated writers in correspondence. The first investigation is presented in the next section.

V2 in narration and dialogue in 13 th century legal and literary prose
The purpose of this section is to examine the behavior of Medieval French V2 in narration and dialogue. While the influence of text formality on rates of use of V2 has already been established by Larrivée (In press), the comparison of narration and dialogal has not been attempted. One reason to do such a comparison is that it might explain the variance in informational values associated to the argument and adjunct XPs of V2 across studies. Whereas Radwan finds that 49% of XP in unambiguous V2 configuration are discourse-new in a 13 th century literary text, Larrivée (2019) shows that 2,7% are in legal Norman texts from 1150 to 1250. It may be that literary texts contain more dialogue, and that this provides more opportunities to introduce discoursenew information, including of the Focus type. If this is correct, the expectation is that dialogue should contain more discourse-new and more Focus XPs than narration, and that for reasons of formality more such XPs are found in literary than legal texts.
To optimize comparability, the research is contrasting two of the texts used by Larrivée (In press) that contain both direct speech dialogue on the one hand, and on the other reported speech and narration. The Anglo-Norman legal prose Year Books represent notes from actual cases at the King's court, from which the earliest 1270s texts are used (Brand 1996). The literary prose Queste del Saint Graal is a romance dating from 1225-30 and presenting no particular regional trait. These are used to identify examples of main clause 2 unambiguous V2 in dialogue as compared to other modalities ("narration" from now on refers to both narration and indirect speech). Again, unambiguous V2 are structures such as those illustrated in (1)-(3) with an expressed postverbal subject and a preverbal XP, and although it was not possible to consider exclusively the strict cases with only one XP, these were separated in data reported below.
A continuous sequence of each text was annotated that contained 50 unambiguous main clause V2 constructions in direct speech and 50 in narration, starting from § 7 to end § 48 in Queste, and from p. 9 to p. 45 in the Year Book. A full analysis of all the clauses in the sequence necessary to obtain the target number of V2 was not accomplished, as the proportion of V2 per clause is already provided in Larrivée (In press) for the first 50 V2 across modalities (5.1% of all clauses and 14% of all main clauses in the Year Books and 9.2% of all clauses vs. 20.7% of all main clauses in the Queste), and since the distribution of direct speech is text-dependent. What was identified was first the type of XP, arguments and adjuncts (1) on the one hand, and on the other particles such as ainsi 'thus' in (2) and coordinators like et 'and' (3). While the former has an informational value, that is not the case of the latter: there is no situation in which one can identify discourse-old coordinators as opposed to discourse-new or Focus coordinators. As the objective of this study is to assess whether there is a significant relation between the language modality of narration and dialogue and the informational value of the initial XP, what was counted was the value of the initial XP. The criteria to determine that value were those used in Larrivée (In press). Discourseold XPs are those that relate to information previously made available, via e.g. a deictic/anaphoric expression as in (1) above. No relation to the antecedent context is entertained by discourse-new XPs, and among them, Focus implies a contrast to other potential values (see the illustrations provided below). The results from the analysis are as follows. We start with the data from the less formal text, the Year Books. Results 3 from dialogue is presented in Table 1, and those from narration in Table 2.  In tables 1 to 4, the number of non-strict XPs is always greater than the total number of V2, since non-strict V2 configuration have by definition at least two XPs, unlike strict V2 that have only one.
The figures from dialogue show that discourse-old XPs are very much the dominant informational value, at 84% of the 45 adjunct and arguments, and 76% of all V2. A similar situation is found in narration, with differential numbers, at 91% of the 33 arguments and adjuncts and 60% of all V2. New and Focus XPs are better represented in dialogue, with 7 occurrences (15.5% of arguments and adjuncts XPs and 14% of V2), as compared to 3 in narration (10% of the relevant XPs and 6% of V2). There is however no significant difference between Focus in narration, at 2 occurrences (4% of XPs), as compared to 1 in narration (3% of XPs). Strict and non-strict Focus cases in dialogue and narration are illustrated in turn. We now turn to the more formal, literary text. The results from dialogue are presented in Table 3, and those from narration in Table 4.  The numbers from dialogue show that discourse-old XPs account for 68% of argument and adjunct XPs and 46% of all V2. Similarly, 74% of relevant XPs and 72% of V2 are discourse-old in narration. New and Focus XPs are comparable in proportion across modalities, with dialogue represented by 32% of all relevant XPs and 22% of V2 and narration by 26% and 18%. This is strikingly higher than what is found in the legal text. Also strikingly higher is the figure for Focus alone, which is more numerous than the discourse-new in both modalities, and the only status represented in narration. Dialogue provides 21% of relevant XPs with a Focus, and narration 26%, more than 5 and 8 times the proportion in the legal text. In the morning as soon as the day rose the companions rose up A summary of proportions across texts and modalities is presented below. The differential in rates of use of Focus is following expectations regarding formalitythere are fewer Focus in the less formal texts, that is 3 in the Year Books out of 100 V2, as compared to 16 out of a 100 in the literary Queste. The expectations are not confirmed with respect to modality: there are actually fewer Focus in dialogue (with 9 occurrences across the two texts) than in narration (with 10 occurrences). It is only in the less formal text that mode has the expected impact, where although only a handful of cases are concerned, there are more Focus in dialogue than in narration. The fact that there are actually fewer Focus in dialogue than in narration in the Queste appears an unexpected result that supports the view that literary material might not best reflect the effective grammar of a period. The investigation therefore shows that V2 is sensitive to register, and that the dialogue vs. narration dimension has only an impact on rates of Focus in real-life exchanges. This correlation is examined from a different angle in the next section.

Register and social classes at the end of the 15 th century
This section reports on a second investigation regarding the correlation between register and V2 configuration. It differs from the previous investigation in that it does not rely on the differential profile of dialogue as compared to narration. Instead, it investigates the differentiation in the productions of members of different social classes. These productions are real-life exchanges, as reproduced in the Year Books and unlike in the fictional literary text, a potentially fruitful choice as suggested by the unexpected 4 A reviewer questions whether si tost come contains an element of degree, being a set phrase. Yet, as far as we can tell, the phrase indicates a high degree of rapidity (as soon as), that contrasts with an ordinary indication of temporality (when). As expressions of degree are primary candidates for Focus, I have analyzed them as such here. behaviors evidenced by the latter. The selected material is correspondence found from an early set of letters from a noble Norman family. The correspondence of the d'Estouteville family is exceptional in preserving letters both from the members of the family and from some of their employees (a horseman, a bricklayer, a servant, a butler). Although these letters are from a comparatively late period from 1460-1535, that period is certainly amongst the earliest for French when private letters include correspondence from lower social class members. Such data thus in principle allows an examination of whether members of each social class has a differentiated language practice. If so, this would lead one to expect, in view of the results of the previous section, that as a feature marked by register, V2 would be more present in the letters by noble writers than by their employees.
The method of investigating V2 with respect to the social class of the writer was defined as follows. First, we selected letters from the 1460 and 1480 period, to avoid significant impact of temporal change by going into the first quarter of the 16 th century. In that period a selection was operated in function not only of the social class of their writer (which we schematically termed High and Low), but also of the recipient. A quadrant of letters was thus defined from High to High, from High to Low, from Low to High and Low to Low. The repartition was as follows, with the number of the letters referring to edition by Le Cacheux (1935).  The numbers show clearly that V2 is a rare configuration at 2,5% of all clauses.
We provide the illustrations below. Although the number of occurrences is lower than one would wish due to the extreme rarity of Medieval correspondence produced by members of lower social-classes, there are twice as many occurrences produced by higher social-class writers as there are by lower social-class writers. But this is not reflective of a significant difference, as the V2 represent a proportion of 4,6% of all clauses in each group. What is significant is the productions not by writers but by recipients. All of the eleven occurrences of V2 are found in letters written to members of higher social classes, and none in letters addressed to lower social-class members.
The results of the analysis of this exceptional resource of socially-differentiated correspondence are revealing clear and unexpected results. A priori, if V2 is a registersensitive grammatical feature, one would have expected it to be found mostly in productions from writers of higher social classes. That is the case, but it is not the most striking behavior. What is most striking is that V2 is found only in letters addressed to members of the higher social class. This shows clearly that audience design is defining here the usage of a grammatical feature known to be sensitive to register. Let us now move to the conclusion of this and the previous investigation.

Is Medieval French a diglossic language?
This paper proposes two methods to assess the sensitivity of syntactic configurations to register. Contrasting syntactic behavior in narration vs. dialogue on the one hand and looking at the correspondence of members of different social classes reveal that unambiguous V2 is sensitive to formality. The precise impact of formality can further be assessed. The formality of literary texts defines a higher proportion of Focus XPs, which are as much as 8 times more frequent than in the legal text. Dialogue has an effect, and a marginal one, only in the legal text. The role of formality is again found in correspondence, but with a twist: it is not the social class of the writer that defines the rate of V2, but that of the recipient.
Of note are the role of audience design in the correspondence, and the unexpected results in the literary text. In the latter, there is actually a smaller proportion of Focus in dialogue than in the narration. The narration of the literary text has a massive overrepresentation of Focus XP in strict main-clause V2 configurations; if this carries to other texts, given that strict main clauses are thought to be central to V2 and that literary narration is traditionally very much the primary material of studies, that may have been instrumental in supporting the view that even in 1230, there was no dominant informational value for the XP (but see Larrivée 2019).
This study thus confirms that V2 configurations are sensitive to register in Medieval French. This can be interpreted in two different ways. One is that V2 is an option that is on the way out, and maintained longer in more conservative higher registers. Another is that it has never really been a feature of the ordinary competence of speakers, and only ever of higher registers. At this stage, it is difficult to see how these could be apportioned. Judging by the correspondence, what is clear is that V2 is not part of the ordinary competence of speakers by the end of the 15 th century, and that it is highly sensitive to register in the earliest periods of prose attestations. It is to be hoped that more data closer to the vernacular or better methods could be brought to assess the status of this and other syntactic configurations.