Throughout this article, I deal with a hand-annotated corpus of eight texts taken from the Base de Français Médiéval,10 chosen principally on diachronic grounds. Unfortunately, this required using verse texts for the earliest Old French, which is a typical issue in studies of this type and reflective of the resources available.11 Specifically in what follows I make use of the La Chason de Roland (c.1100), the Lapidaire alphabétique (c.1115), Eneas (c.1155), Robert de Clari’s Conquête de Constantinople (henceforth Clari, c.1205), the Queste del saint Graal (Graal, c.1225), the Vie de saint Eustache (Eustace, c.1225), the Grandes chroniques de France (GChron, c.1275) and the Chronique de Morée (Morée, c.1320). Crucially, these latter two texts date from a period where a number of major changes are observed in French morphosyntax at the beginning of the period referred to by some as ‘Middle French’ (see Marchello-Nizia 1980: Chapter 1 and Smith 2002 for detailed discussion). Independent of issues of nomenclature, this choice is deliberate in order to detect the possibility that the subject system too begins to undergo change.

For all the texts, a small corpus (200–247 clauses) was assembled to investigate the overall distribution of subjects, which was then supplemented where relevant as listed below in the relevant section. In recent years much progress has been made on the challenging issue of how to pragmatically annotate historical corpora, in light of the now well-established intuition that historical pragmatics is an essential component of understanding syntactic change. Although, as already seen (cf. also §3 below) a relationship between information structure and position of the subject is often suggested in the research literature, the exact diagnostics for the information structural status of the subject are often somewhat vague. In order to address this issue in the current study, the following diagnostics were used to annotate subject expressions: (i) Quantified [+Q], pronominal [+PRO] and relativised [+REL] subjects were all given an individual tag and are represented separately in the tables that follow; (ii) Subjects which were active in the discourse in the terms of Prince (1981: 243), Chafe (1987) and Lambrecht (1994: 165), here taken to correspond to being activated in the previous ten lines of text, were assigned a [+ACTIVE] tag; (iii) subjects taken to correspond to ‘accessible’ information (cf. Ariel 1988: 66), that is to say non-activated but nevertheless inferable concepts such as Dieu ‘God’ or Li rois ‘the King’ were tagged as [+ACCESSIBLE]; (iv) Subjects which were neither quantified, pronominal, relativised or tagged as ACTIVE or ACCESSIBLE, were assigned the [+NEW] label, such as in (8) where the subject expression is neither discourse-ACTIVE nor likely to form part of the common ground of ACCESSIBLE information between reader and writer:

The aim of this section is to give a broad overview of the distribution of null and overt subjects within the corpus. As already noted, an analysis of general changes in this domain is hardly a lacuna in the field of French linguistics and has already generated a large literature.12 The purpose of this section is therefore to contextualise the analysis below against the background of the general system of subjects found within the texts, with a particular focus on the distribution of preverbal, postverbal and null subjects. Consider Table 1 and Figure 1 in this regard, which summarise the main data from this aspect of the corpus analysis.

As Table 1 shows, null, preverbal and postverbal subjects are robustly attested in all the texts under examination. However, it immediately becomes clear that the texts do not pattern alike in their distribution. For the discussion that follows, the distribution of preverbal subjects such as those in (9) is not crucial to the analysis. Clear evidence for diachronic progression in this domain is also hard to establish, with both the lowest and highest proportion of preverbal subjects found within our latest three texts.

In basic terms we note therefore that the preverbal placement of the subject overwhelmingly preferred in the contemporary language (Pollock 1989: 391–407; Lambrecht 1981: 5–7; Rowlett 1998: 7; Rowlett 2007: sec. 4.3; Smith 2016: 310) has not consistently taken hold in any of the texts considered from the latest period.

It is in the domain of postverbal and null subjects that greater variation is observable. As already noted there is a long tradition in the syntactic literature of differentiating between Early Old French (pre-~1200) and Later Old French (Hirschbühler 1990; Roberts 1993; Vance 1997). Partly this distinction concerns an increase in the asymmetry between main and embedded clauses, but there is also an additional difference, namely the decline of V1 clauses from c.1200 onwards (Simonenko & Hirschbühler 2012; Wolfe 2016a). Likely due to a change in the locus of V2 (Wolfe 2016a; Wolfe 2018b), the licensing conditions for null subjects change in Later Old French, such that they are typically only licensed postverbally, whereas previously they could also be licensed preverbally, yielding a V1 order. This generalisation might lead us to expect a general reduction in the attestation of null subjects in the texts from the Later Old French period. This is in fact what we find, with null subjects accounting for 44.5%–50% of the data in the Early Old French texts (Roland, Lapidal, Eneas) but only 29–32.5% in the Later Old French texts (Clari, Graal, Eustace). The finding that null subjects again increase in frequency in the two latest texts is entirely in keeping with the qualitative literature on Middle French (Roberts 1993; Hirschbühler 1991; Hirschbühler 1995; Vance 1997: 257) where it is frequently noted that in both main and embedded clauses, Middle French texts license null subjects where they would not be licit in Later Old French.

A final point to note is that postverbal subjects, that is to say inversion structures, are most common in Clari and Graal, two of our three Later Old French narrative chronicles. Given that inversion structures have, since the very earliest work on the topic (i.e. Thurneysen 1892, von Wartburg 1958: 103, Foulet 1919: 243–245) been viewed as an integral component of the V2 syntax of Old French, the high relative attestation of postverbal subjects lends credence to the notion that 13^th-century prose texts somehow constitute a stricter instantiation of V2 than is found in earlier texts (cf. Roberts 1993: 135–136, Rouveret 2004 and Wolfe 2018:Ch4 on this notion).13 Nevertheless, the lower proportion of postverbal subjects in Eustace, which is still the third-highest overall, cautions against a blanket generalisation on texts from the first half of the 13th century.

We have therefore seen that the overall distribution of the data calls for a nuanced analysis, under which the distribution of subjects is not amenable to widespread generalisations. This said, one factor does come through strongly, the fact that null subjects are more restricted in Later Old French texts than either their Early Old French or Middle French counterparts. This finding is unsurprising if we recall the observation well established since Foulet (1919) that null subjects are especially restricted in early 13th-century prose.

The preverbal field in Old and Middle French has been extensively discussed in the research literature (Skårup 1975; Reenen & Schøsler 1992; Vance 1997; Mathieu 2006; Mathieu 2009; Mathieu 2012; Labelle 2007; Combettes 2007; Muller 2009; Steiner 2013; Hansch 2014; Wolfe 2016b; Labelle 2016; Labelle & Hirschbühler 2018), reflecting a more general theoretically-informed interest in the prefield in a range of V2 languages (cf. Cardinaletti & Roberts 2002; Frey 2004; Jouitteau 2010; Salvesen 2013; Holmberg 2015; Wolfe & Woods 2020).

Preverbal objects in particular have been the subject of considerable interest. Marchello-Nizia (1995: 95–100) for example notes that preverbal objects in Early Old French are more frequently informationally NEW than fronted objects in Later Old French prose, a finding confirmed with certain caveats in recent studies by Labelle & Hirschbühler (2018: 276–277), Steiner (2014: 205–226) and Wolfe (2018b). Indeed, in recent work Wolfe (2016a: sec. 3) has claimed that the loss of preverbal new Information Focus is a crucial emergent isogloss in the medieval period which triggers morphosyntactic changes in a range of other domains.

If we assume a close mapping between the activation of functional projections and the pragmatic status of constituents and that the Information Focus projection is only active until c.1200 in French, a clear prediction takes hold. We expect to observe a decline in the possibility for informationally NEW subjects to occupy the preverbal position. However, we note from the outset that Labelle & Hirschbühler (2018: 271), exploring the statistics for definite vs. indefinite subject placement in a diachronic corpus note ‘a small increase in the tendency to place indefinite subjects in postverbal position’ but note that indefinite (typically informationally NEW) subjects can occupy the preverbal position from the 10^th to the early 14^th centuries (pace previous work by Rinke & Meisel 2009). This suggests from the outset that preverbal objects may be more susceptible to the change in discourse-pragmatics than subjects. The data present as follows.

Looking at Table 2, there are a number of findings that become readily apparent. Firstly, there is a modest but notable decline in the proportion of subjects which are informationally NEW such as those exemplified in (10), with all three of the highest percentages occurring in the three earliest texts:

Secondly, ACCESSIBLE preverbal subjects which are inferable from context but not discourse-ACTIVE in the terms of Lambrecht (1994) show their highest proportion in two of the earliest texts, though there is not a straightforward linear decline:

Our tentative conclusion here might be that Early Old French texts show a greater tendency towards having ACCESSIBLE subjects in the prefield and that Later Old French and Middle French show less of a tendency, but that there is intertextual variation in this domain.

Finally, there is a difference between texts in terms of the proportion of subjects which are unambiguously discourse-ACTIVE, with Roland, Eneas, Lapidal, Graal and Eustace having a relatively small proportion of these and Clari, GChron and Morée a higher one:

At first glance, there is not a straightforward diachronic story to tell here, which is perhaps surprising, given the frequent claims in the literature that discourse-ACTIVE subjects increasingly occupy the prefield after 1200 (Marchello-Nizia 1995: 95–100; Steiner 2014: 205–226; Wolfe 2016a: 480). However, if we look closely at Graal and Eustace, we see that they both have ~68% of preverbal pronominal subjects. Pronominals are classed as highly topical on Givón’s (1983) Topic Accessibility Hierarchy are often cited in the literature as cases par excellence where reference to an already ACTIVE discourse referent is highly likely, though not absolute (Ariel 1988: 66; Lambrecht 1994: 172; Schwarzschild 1999: 154; Krifka 2007). A likely hypothesis is therefore that the unusually high proportion of preverbal subject pronominals is affecting the frequency of ACTIVE nominal subjects within these two texts. If we accept this, a clearer pattern emerges where the Later Old French and Middle French texts in the sample stand in contrast to earlier texts in showing a greater tendency towards hosting highly topical preverbal constituents.

We noted at the outset that the most recent corpus work on pragmatic change in the Medieval French prefield is indicative of a more gradual change in this domain that has conventionally been conceived (Steiner 2014; Labelle & Hirschbühler 2018). Our data broadly confirm this intuition, with obvious intertextual variation for all the factors considered. Nevertheless, we can tentatively conclude a tendency for a decrease in NEW information constituents occupying the prefield and a move away from more loosely topical ACCESSIBLE constituents occupying this position, to discourse-ACTIVE or pronominal preverbal constituents becoming more prominent. Abstracting away from some of the inconsistencies, this suggests that the classic intuition of a shift away from NEW/FOCAL constituents in the prefield towards OLD.TOPICAL constituents is correct, but that there is a more nuanced picture that is often suggested.

As already noted above, nominal and pronominal postverbal subjects are found in all the texts, which is unsurprising given the V2 status of Old and Middle French.14 In this section I set out to explore what the precise discourse-pragmatic status is of postverbal subjects in general and, furthermore, whether the specific discourse-pragmatic characteristics are correlated with distinct structural positions. In order to facilitate this part of the analysis the postverbal corpus was supplemented with additional data to ensure 100 tokens were analysed for each text.

Before considering the data, recall the discussion in §1.2.2 regarding the loss of New Information Focus in Medieval French and Poletto’s (2014) Uniformity of Phases hypothesis. One proposal, put forward in Wolfe (2016a), is that as French loses New Information Focus in the left-peripheral prefield around 1200, New Information Focus is then relocated ‘downstairs’ to the vP-periphery. This would lead us to expect a concomitant rise in postverbal NEW/FOCAL subjects as such subjects decrease preverbally. The second proposal, developed for Old Italian by Poletto, makes the inverse prediction, namely that the syntax-pragmatics mapping changes in tandem in each domain. This would mean that the extant evidence shows a decline in NEW/FOCAL subjects both pre- and postverbally and furthermore that both the prefield and the postverbal field become increasingly specialised in hosting OLD/TOPICAL subjects. Now consider Table 3 in this regard.

We find that postverbal subjects tagged as informationally NEW decrease diachronically, from a high of 36% in Roland to a low of 9% in Morée, the latest text within the corpus. The examples that follow come from three texts, Roland, Graal and Morée:

Furthermore, observe that in the 13^th- and 14^th-century prose texts, subjects which are unambiguously discourse-ACTIVE also increase in frequency from c. 10% in Roland and Eneas to 23–34% in the later texts. The Lapidal is an outlier for the Early Old French period, but this is likely due to its narrative structure, as it is the only text in the corpus which is not a chronicle or narrative prose. Rather, it is a scientific text recounting the properties of particular stones, where a single stone is consistently referred back to throughout a portion of text. Overall, the data are suggestive of a decline for NEW postverbal subjects within the period studied.

Is there a parallel increase in subjects which strongly encode OLD information, that is to say PRONOMINAL or ACTIVE nominal forms? Once again, we appear to observe a split between the earliest texts and the later ones, reflective of the more general split between pre- and post-1200 French texts noted in the literature. In the three earliest texts, ACCESSIBLE subjects constitute 36–37% of the data, whilst more strongly OLD ACTIVE or PRONOMINAL subjects constitute 19–32% of the data. By contrast, in all but one of the later texts (Morée), ACCESSIBLE subjects constitute 11–22% of the data and their ACTIVE (16) or PRONOMINAL (17) counterparts constitute a striking 49–67% of the data.

So here we observe that the postverbal field is becoming increasingly specialised, though not exclusively so, in hosting the subject expressions which unambiguously encode OLD information.

Taken together, the behaviour of NEW, ACCESSIBLE, ACTIVE and PRONOMINAL subjects, suggests that the ability for the postverbal field to host NEW or weakly-old ACCESSIBLE information declines. In tandem, the attestation of unambiguously old ACTIVE or PRONOMINAL subject increases overall. Although corroboration is needed at a larger scale, these data provide little support for Wolfe’s (2016a) proposal that New Information Focus is encoded postverbally after 1200. Rather, they suggest that Poletto’s (2006a; 2006b; 2014) approach to Old Italian may fruitfully be applied to French: the CP and vP phase show parallel developments regarding the syntax-pragmatics mapping.

Postverbal subjects in the specific contexts of Germanic- and Romance-inversion are ripe for reconsideration. Although both these distinct subject positions are prominent in early generative work on Old French (see in particular Roberts 1993: 117–142; Vance 1995: 174–177; 1997: 102–125 and de Bakker 1997: 39f), it is only in recent years that the prominence of information structure has come to the fore for this phenomenon (Rinke & Meisel 2009; Salvesen & Bech 2014) and Old French syntax in general (Labelle 2007; Labelle 2016; Zaring 2010; Zaring 2011; Donaldson 2012; Steiner 2014; Larrivée 2011; Larrivée 2019; Wolfe 2016a; Wolfe 2018a; Labelle & Hirschbühler 2018b; Ingham 2018).

In what follows, I set out to consider the distribution of cases of Romance- and Germanic-inversion in the corpus, breaking down the broad class of ‘postverbal’ subjects introduced above. I present cases where the subject is unambiguously of the Romance or Germanic inversion type, alongside cases where there is no unambiguous diagnostic element present. Recall that for the purposes of the analysis, in keeping with the formal literature on the topic (Vance 1997; Lombardi & Middleton 2004; Vance, Donaldson & Steiner 2009; Poletto 2014: Chapter 1; Salvesen & Bech 2014; Wolfe 2015a), a diagnostic element was taken to be a past participle, infinitival or predicative complement: items which in formal terms demarcate the left edge of the extended verbal projection (i.a. Cinque 1999).

This is in and of itself significant in two senses. Firstly, the purported minimal or non-attestation of Germanic- or Romance-inversion structures has been used as evidence against the V2 hypothesis for French and other Medieval Romance languages (cf. Kaiser 2002: 134; Lombardi & Middleton 2004: 571; Rinke & Meisel 2009: 126; Sitaridou 2011: 164). Secondly, given that Germanic-inversion is viewed by a number of scholars as the most robust piece of historical evidence for the V2 status of Medieval Romance and Old French, an understanding of the distribution of the various inversion structures and their pragmatic correlates will help develop an understanding of both change within the V2 grammar and any changes which lead to the eventual loss of the V2 property. Table 4 shows all the relevant data, whilst Figure 2 shows the distribution when the ambiguous cases are discounted.16

There are a number of observations we can make based on the data. First, witness that the proportion of ambiguous inversion structures is always substantially larger than either those of the unambiguous Romance or Germanic type. This is unsurprising on two grounds. First, we require an overt postverbal subject, which as the data in in §2 show, only ever accounts for a maximum of 35.5% of subjects. Furthermore, as an unambiguous diagnostic, the presence of a vP-edge demarcating element such as an infinitive or past-participle is required. In the latter case, recall that the compound past tense was nowhere near as ubiquitous in the medieval period as it is in the contemporary language (Harris 1978; Rickard 2003: 56; Caudal 2015). In sum, we should therefore not be surprised that the unambiguous cases are not extremely frequent in text, as we require the coalescence of two syntactic factors: an overt postverbal subject and the presence of a diagnostic element for the position of that subject.17

With this caveat, when the structurally low or high position of the subject is demarcated, Germanic-inversion is always more numerous than Romance-inversion within the sample, with the one exception of our latest text Morée. This is a particularly pertinent finding as it demonstrates that of the unambiguous data available, the subset clearly indicating matrix V-to-C movement is always more robustly attested than the nevertheless V2-compatible Romance-inversion counterpart. A second finding is that, although the highest and lowest percentage of Germanic-inversion are found in our earliest and latest texts respectively, the diachronic decline in Germanic-inversion is not straightforwardly linear. Consider for example how the second-latest text shows 21% of its postverbal subjects to be unambiguous cases of Germanic-inversion. However, this is again arguably unsurprising if inversion of this kind is integral to a V2 syntax and Old and Middle French were V2 languages (cf. Vance 1997 et seq). Finally note that the apparently high Romance-inversion figures for Morée, as will be discussed further below, should be treated with some caution, as every subject is athematic/passive (18), in contrast to all the other texts where the data are more diverse and athematic subjects are always in a minority:

As such we can conclude that Germanic-inversion outnumbers Romance-inversion in the samples of all texts and that furthermore there is some evidence for a decline in the rate of Germanic-inversion.

Considering the syntax-pragmatics mapping, recall the most relevant recent study by Salvesen & Bech (2014). As outlined above, they note that whilst postverbal subjects in a structurally high position are typically discourse-OLD, subjects in a structurally lower position can be either NEW or OLD. So far, this generalisation has not been tested against a larger corpus of texts. Tables 5 and 6 reveal data from the information structure-focussed tagging of the relevant examples in the corpus. In an attempt to normalise the relatively small number of tokens, 20 examples of Romance-inversion and 30 examples of Germanic-inversion were collected for the texts where this was possible. As Tables 5 and 6 show, this was not possible for all texts. Given the particular scarcity of such structures in the latest period, an additional text, GRChronJ2C5, was also added from the BFM. Given the small sample size we are dealing with overall, the following findings should therefore be taken as provisional results to stimulate further large-scale research. I have also removed texts with fewer than ten tokens of the relevant structure from the analysis that follows.

With the appropriate caveats in mind, it is clear that Salvesen & Bech’s (2014) generalisation for R-inversion holds to some extent to a wider sample of texts. We find that in the Early Old French verse texts and GChron, over half the subjects found in Romance-inversion contexts are indeed discourse-NEW (19–21) and that ACTIVE subjects are relatively rare.

In early 13^th-century prose onwards, however, the position appears to be non-specialised for NEW information. Note the relatively even distribution of discourse-NEW, ACTIVE and ACCESSIBLE subjects in Clari and Graal, alongside the fact that GRChronJ2C5 and Morée show a surprisingly high proportion of ACTIVE subjects for a position allegedly specialised for NEW information.18

Thus, an interim summary on the basis of this small sample is that the earliest texts provide the strongest evidence for the R-inversion structure being associated with NEW-information subjects. The numbers here are small, so caution must be exercised in reaching strong conclusions, but the data presented here are suggestive of the weakening of this pragmatic requirement after the Early Old French period, where the position appears more generalised in the pragmatic status of constituents occurring there.

Turning to G-inversion, Salvesen & Bech (2014) and Wolfe’s (2018b: 72–73, 93, 115) proposal that the structurally high postverbal position, here taken to be SpecTP, used in Germanic-inversion structures, is a dedicated position for OLD information-subjects is also borne out in the data to an extent. NEW-information subjects never constitute more than 15% of the subjects in this position and furthermore, ACTIVE and pronominal subjects which are both typically encode strongly OLD information constitute 73.3%, 80%, 84.6%, 73.3% and 66.7% of the data in Clari, Graal, Eustace, GChron, and GRChronJ2C5 respectively (23–25):

A further point of variation concerns subjects which are only weakly ACCESSIBLE, which constitute 36.7%–53.3% of the relevant subject expressions in the earliest three texts. However, this figure falls in all the later texts. This suggests that, within the confines of the corpus sample, there is an increasingly strong requirement for SpecTP subjects to be unambiguously discourse-OLD (i.e. ACTIVE or PRONOMINAL). This is thus the inverse of the trend observed for R-inversion, with the discourse-pragmatic status of G-inversion subjects becoming increasingly specialised within the corpus during the period studied,

It is noteworthy that much research on Old French in recent years has revived the tradition of noting substantive diachronic change in the period conveniently labelled ‘Old’ or ‘Medieval’ French in standard handbook treatments (Labelle & Hirschbühler 2005; Labelle & Hirschbühler 2017; Labelle & Hirschbühler 2018; Labelle 2007; Labelle 2016; Zaring 2010; Zaring 2011; Prévost 2011; Simonenko & Hirschbühler 2012; Balon & Larrivée 2016; Wolfe 2016a; Wolfe 2018a; Simonenko, Crabbé & Prévost 2019; Meklenborg 2020). We could of course dismiss the findings above on the basis of the size of the corpus, but given that many of these same texts have been used to show clear evidence for syntactic change in other domains, there is a strong motivation to explore a strong interpretation of the data.

The broad overview of the data offered in §4 hint at a discontinuity between the Early Old French texts and the early 13^th-century texts on the one hand, and the early 13^th-century texts and the late 13^th-century and early 14^th-century texts on the other.

As Table 1 and Figure 1 show, one of the major differences is in the attestation of null subjects. The Early Old French texts, Roland, Lapidal and Eneas, show widespread attestation of null subjects.19 In terms of Fleischman’s (1991: 258) adaptation of Givón’s (1983) Topic Accessibility Hierarchy for Old French, there are at least two mechanisms for marking a highly accessible/referential subject. Either a null subject can be used, with the possibility of yielding a V1 clause, or particle SI can be employed, which in these early texts marks Topic-continuity in the vast majority of cases (Fleischman 1991; Reenen & Schøsler 1992; Wolfe 2018a). Importantly, our exploration of subject position and discourse-value in these early texts, shows that neither preverbal nor postverbal position of the subject, nor, more specifically, its occurrence in an R- or G-inversion structure, is a strong predictor of its highly topical status.

Table 1 and Figure 1 show that, in line with the literature on the topic, null subjects decline in frequency in early 13^th-century French prose. This is principally due to the loss of a previous marker of highly topical antecedents, the use of a V1 clauses with a null subject, which is no longer licensed in the grammar by 1200, but which Roberts (1993: 179) shows rise again from the end of the 13th century onwards. The alternative Topic-marking strategy highlighted by Fleischman (1991) is also no longer available, as it has been shown in recent work that SI in these very same texts has grammaticalised as a V2-related expletive which does not typically mark Topic-continuity (Salvesen 2013; Meklenborg 2020; Wolfe 2018a).20 The proposal stemming from the analysis in this article is that the innovative grammar of 13th-century French needs a new way of marking subjects as unambiguously topical. This leads to the discourse-pragmatic specialisation of the SpecTP postverbal subject position in Germanic-inversion structures, as a position hosting highly referential ACTIVE or PRONOMINAL subjects See Table 7 for a schema of the relevant changes.

Taking the data as a whole, the generalisation is that the Early Old French (12^th century), Later Old French (early 13^th century) and Early Middle French (late 13^th and early 14^th century) texts have partly distinctive subject systems and partly distinctive syntax-pragmatics mappings. As we will now see, this extends to more fine-grained variation in the discourse-pragmatics of the preverbal and postverbal field.

At the outset of the article, the consensus view was noted that Early Old French licenses preverbal new Information Focus (Labelle 2007: 302–5; Mathieu 2012: 341; Wolfe 2018b: 139), whereas evidence for this is less apparent in 13^th-century prose onwards (Marchello-Nizia 1995: 99–101). This is likely correlated with a low locus of V2 within the articulated left periphery (Rouveret 2004; Labelle 2007; Wolfe 2016a; Wolfe 2018b: Chapter 4). However, how later Old French encodes new Information Focus has not been settled in the literature. We noted that there are at least two hypotheses that could be tested against the data from the subject system: new Information Focus could be licensed at the vP-periphery as is the case in Modern Italian (Belletti 2001; Cruschina & Ledgeway 2016: sec. 31.2.1; Ledgeway In Press), which might lead us to expect an increase in NEW-information postverbal subjects following the loss of CP Information Focus (25a). An alternative hypothesis, in line with argumentation in Poletto (2006a; 2006b; 2014) would lead us to expect parallel developments at the CP and vP-edge (25b). In the schema that follows, the two hypotheses are illustrated with bold text indicating an activated functional projection and grey text one non-active in the language:

The data presented above do not evidence a sudden change in either domain but do hint at a progressive loss of NEW-information subjects and an increase in subjects which are unambiguously discourse-OLD in both the C-domain and the vP-layer (i.e. a gradual version of (26)). The Old French data then lend credence to the value of Poletto’s approach to account for change away from the Old Italian data her analysis is based upon. Broadly speaking, we see similar changes in the syntax-pragmatics mapping at both phase edges. This suggests that change in one domain, may analogically condition change in the parallel domain.

Crucially, however, it has been shown that the data show yet more complexity when Germanic-inversion subjects in SpecTP are isolated from Romance-inversion structures which are clearly lower in the structure, at the vP-edge or below.

Considering first the nature of SpecTP, we find evidence from the outset that NEW-information subjects are only minimally attested in this position. However, the earliest Old French texts do pattern distinctly to the later texts. In 12^th-century verse, SpecTP hosts a large number of subjects which are more weakly referential and thus tagged as ACCESSIBLE in the methodology used here. However, there is potential evidence for diachronic reanalysis in the nature of this position in 13^th-century prose, insomuch as ACCESSIBLE subjects no longer make up a large part of the data and are instead replaced by more highly referential discourse-ACTIVE and PRONOMINAL subjects. In other words, we are dealing with a functional projection whose role is becoming increasingly specialised in pragmatic terms.

Two factors may have triggered this reanalysis. Firstly, pronouns, which are found in this position in the very earliest French texts are frequently cited in the literature as typically belonging to a highly referential/topical class and thus occupy a high position on Lambrecht’s (1994: 165) Topic Acceptability Scale.22 An increasing move towards a grammar where all types of nominal and pronominal discourse-active subjects can move to SpecTP when postverbal could therefore be conditioned by a form of syntactic analogy of the type envisaged in Roberts’ (2007: 275) Input Generalisation Principle. Putting aside irrelevant details, under this account formal features associated with a particular class of functional heads, may become associated with a larger class than in the original grammar. In this particular case, the movement trigger for pronominal subjects to evacuate their base-generated position in the vP-layer and move to SpecTP would be progressively extended to both nominal and pronominal subjects which score highly in terms of discourse-activation:23

Secondly, we must consider other changes affecting the subject system at this time. As already noted above and in §2 it is precisely in these 13^th-century texts that null subjects in general show a very restricted distribution (Marchello-Nizia 1980: 331; Adams 1987c; Dupuis 1989; Roberts 1993: 136–147; Vance 1997: 32; Rouveret 2004: 193–5). Likewise, it has been suggested recently that whilst the particle SI may be an unambiguous marker of thematic subjects in Early Old French verse (Fleischman 1991) this topic-marking strategy is no longer operative in 13^th-century prose (Wolfe 2018a: secs. 2–5). The suggestion put forward here is that the reanalysis of the SpecTP position outlined immediately above provides a novel topic-marking strategy in the grammar of later Old French. A highly topical subject in Early Old French could be rendered with a V1 clause or particle SI could be employed. In the new grammar, such subjects increasingly occur postverbally in SpecTP.

Romance-inversion within the period considered shows a distinct trajectory. Rather than gaining an increasingly specific pragmatico-semantic function, it appears to lose its specialisation in hosting NEW-information subjects during the period considered. Whilst instances of Romance-inversion show predominantly focal subjects in Early Old French, the system appears to have broken down in the 13^th- and 14^th-century texts, with an increase in subjects that are already ACTIVE in the discourse. Recall from our discussion above that this mirrors the overall diachronic generalisations about postverbal subjects in general. Given that the Romance-inversion subjects are preceded by participles and other VP-complements standardly analysed as demarcating the edge of the v-VP-complex (Cinque 2001; Cinque 2006: 12; Cardinaletti & Shlonsky 2004: 525; Ledgeway In Press), I take it that they occupy the vP left periphery postulated by Belletti (2001; 2004; 2005a; 2005b; 2006; 2008). The analysis above is that new Information Focus is being lost at the low left periphery from the beginning of the 13th century onwards. It is therefore predicted that, once isolated in the extant textual evidence, postverbal subjects unambiguously at the vP-edge will not necessarily encode NEW information. In formal terms, whilst unambiguously discourse-ACTIVE and pronominal subjects are attracted to SpecTP, the remainder remain in the extended vP-layer, yielding the non-specialised subject position noted in the literature (de Bakker 1997: 57; Vance 1997: 79; Myking 2012; Salvesen & Bech 2014: 222; Wolfe 2015b: 91). Specifically, we have argued that this non-specialised nature only holds true for 13th- and 14th-century texts.

Table 8 is a proposed schema of variation and change discussed in this and the previous sections. The terms should be taken to be used relatively.

Overall this article has aimed to show that despite being one of the best-researched areas of French historical linguistics, there are a substantial number of research questions worthy of exploration within the domain of the subject system. The core hypothesis has been that despite the inevitable challenges of assembling reliable corpora for historical varieties, there is evidence for change within the subject system throughout the medieval period. In theoretical terms, our analysis has suggested that key relevant changes in the Topic-Focus system can be captured to some extent by assuming a parallelism in the makeup of both the CP- and vP-edge in line with Poletto’s (2006a et seq.) analysis of Old Italian. However, the specific pragmatico-semantic function of particular inversion structures shows further fine-grained variation. In concrete terms, the SpecTP position targeted by subjects in Germanic-inversion specialises in function towards hosting nominal and pronominal subjects which are unambiguously discourse-OLD. This is the opposite of the central observation for Romance-inversion, where the pragmatic function of the vP-periphery appears to generalise from hosting predominantly focal, NEW-information subjects to subjects which can belong to a range of pragmatic categories.

At least two questions arise which can act as a springboard for future research. First, it has become apparent in the course of the article that both the high and low left periphery show a parallel loss in the ability to encode NEW Information Focus. Clearly, this cannot mean that speakers of later stages of French were left with no means to syntactically encode new Information Focus. Belletti (2005a: sec. 2.1) however makes a pertinent observation that in contemporary French, aside from intonation, a cleft is by far the most natural way to encode new Information Focus when answering questions. Given that cleft sentences increase in frequency only towards the end of the medieval period (Dufter 2008), future research should focus on whether the rise of this new encoding device is a result of the loss of new Information Focus being realised through movement at the CP and vP-edge. A second related topic for future research concerns how the late medieval system documented here changes into the contemporary subject system found in Modern French. Two observations are noteworthy here. Considering that in Modern French inversion structures, it is typically only pronominal subjects that can occupy SpecTP (Rizzi & Roberts 1989), this state of affairs is just a natural extension of the change observable in the medieval period, whereby the nature of this position’s use in inversion structures becomes increasingly specialised along the lines ACCESSIBLE > ACTIVE/PRONOMINAL > PRONOMINAL and is thus reminiscent of a number of changes in English syntax historically whereby the movement triggers for various types of inversion become increasingly specialised diachronically (Biberauer & Roberts 2012; Roberts 2019). In a parallel fashion, the fact that the vast majority of non-pronominal subjects in Modern French inversion structures target the structurally low position in the vP-periphery is also unsurprising if we consider the generalised nature of this inversion site at the beginning of the Middle French period. It is arguably precisely the non-specialised nature of the position which makes it a ripe target for reanalysis as the structural position of the inverted subject in a range of V2-related structures once this property is lost. However, the diachronic trajectory of both these constructions would need to be explored in a range of post-medieval texts.