1 Introduction

Why does the syntactic organisation of languages change? One view is that a variety of options exist at any period, and that what varies is their status as marked or default variables (e.g. Roussou 2016). This is illustrated by the shift from ex situ to in situ wh partial questions. A significant minority of world languages (29% of the WALS sample, map 93.A) have ex situ wh as a default option for partial questions. Some of those have in situ as a marked option for specific usage, and in particular for echoic questions, as is the case with English and Ukrainian. As with French and Brazilian Portuguese, some of the default ex situ languages are displaying a shift toward a majority of in situ uses. These in situ are initially found categorically in clauses that are explicitly discourse-old, repeating an antecedent proposition (Larrivée 2019; Rosemeyer 2019). An explicitly discourse-old value thus initially characterise marked variables. Is there a similar trajectory for marked word order that would associate to a categorical informational value at some initial or final point of their development? It might be that word order options categorically associate with an informational value when they emerge or decline, either as a cause or a consequence of that change.

The role of informational value for word-order change is what is investigated in this paper. The considered change is the loss of verb-second (V2) configurations in a well-described language, Medieval French. These configurations are particularly good candidates for the study of the relation between syntactic change and information structure, since an informational role is generally thought to attach to the preverbal complements¹ (see Labelle & Hirschbühler 2018 inter alia). Whether this actually is the case in Medieval French, and what it means for the relation between information structure and syntactic change, are questions that this paper answers.

It is organised as follows. Section 2 reviews recent studies that investigate the relation between declining Medieval French V2 configurations and information structure (IS). It suggests that such investigations should refer to a homogeneous corpus of data and an explicit informational analysis. The corpus and method are presented in section 3. Section 4 reports on the informational behaviour of preverbal complements in V2 configurations, touching upon other configurations such as clefts and dislocations. The final discussion in section 5 summarises the results, and assesses the extent to which information structure can be thought as a cause or a consequence of syntactic change. The overall conclusion is that preverbal complements have a categorical value of discourse-old information while V2 is a productive configuration, and that it is the loss of that categorical informational value that anticipates the decline of V2.

2 IS and preverbal complements

A major word order evolution in Romance is that from Latin (S)OV to SVO in contemporary Romance languages (e.g. Marchello-Nizia 2009; Danckaert 2017), reflecting a universal word-order change trend according to Gell-Mann & Ruhlen (2011). A transition stage of (some) early Romance languages is found in verb-second (V2) configurations XVS. In such configurations, the V2 preverbal complements are very generally believed to involve an informational value (such as e.g. Focus). The purpose of this section is to review recent work on the relationship between information structure and V2 preverbal complements in early French.

The idea that an informational value relates to early French and Romance V2 complements follows from general and theoretical assumptions. On a general level, because V2 is a minority²,³ configuration in early French and other early Romance languages, and that minority configurations may take on a pragmatic value that relates to informational distinctions, such a pragmatic value could be assumed to characterise the marked structures. This is indeed the traditional view expressed by e.g. Venneman (1974) and Marchello-Nizia (2000). On a theoretical level, assuming a split-CP model à la Rizzi (1997), the V2 preverbal complements must be in the left periphery of the sentence, since they precede the verb that is itself assumed to be left-peripheral (inter alia Wolfe 2016); and since left-peripheral projections relate to IS distinctions of Focus and Topic,⁴ these values should associate to the complements.

There is however some disagreement about which informational value the preverbal complement takes on. That the V2 XP has a uniform Focus value is proposed in a recent monograph on Medieval French (Zimmermann 2014), as it is for e.g. Italian (Poletto 2014). This however contradicts the view of the XP as a Topic (Venneman 1974; Marchello-Nizia 2000), and the frequent attestations of topical deictic and anaphoric XPs, as pointed out by Mensching (2012) (and illustrated by e.g. (5) below).

The debate on the informational value of the preverbal XP in Medieval French is revisited by recent quantitative investigations. A systematic investigation is proposed by Ingham (2018a). Using a textually homogeneous corpus of 13^th century narrative prose romances, he considers XPs in V2 configurations, excluding coordinated clauses. A correlation is found between the information status of the XP and the realisation of the subject. XPs comprising the deictic ce ‘this’ occur with 93% of expressed pronominal subjects (excluding impersonal subject pronouns), and only with 7% of null subjects. Conversely, intensifier XPs are found at 3% with an expressed personal pronoun subject, and at 97% with null subjects. Thus, discourse-old Topic XPs relate to an expressed subject in a V2 configuration, whereas discourse-new Focus XPs associate with a null subject. This is interpreted as reflecting the necessity for the null subject to be licensed by an operator in a left-peripheral Topic position, that is therefore not available to host a discourse-old phrase. This provides reasons to believe that at least in homogeneous corpora, for a given period of time, predominant informational roles can be found for preverbal complements. Such a predominant informational role is supported by other studies such as Ingham & Larrivée (2015). They find in the dialogic legal material of the Anglo-Norman Yearbooks for 1282–1284 that while 66% of XPs in X V S are discourse-old anaphoric when there is no particle si inserted, 76% of the XP in X si V S configurations where the asseveration particle si is inserted between the XP and the V are non-anaphoric (see also Rahn 2016).⁵

The literature thus provides reasons to believe that information structure plays a role in the occurrence and distribution of V2 XPs, although the nature of the informational value is a matter of debate. Issues that arise are the homogeneity of the investigated texts, the criteria for informational values, and the fact that informational values may change through time. These issues inform the method and corpus proposed for the current study, which are discussed in the next section.

4 The Information Structure of V2

V2 is a rare typological feature, that is mostly found in some families of Indo-European languages, Germanic, Celtic and Indo-Aryan (see the overview in Holmberg 2015, inter alia). It is also attested in a number of Medieval Romance languages (Wolfe 2018a, 2016 and the extensive set of references therein), including French. In most languages, the configuration is asymmetric: it is a feature of main clauses, and is constrained or excluded in subordinates. That would be because subordinates have a less developed left-periphery, which is targeted by V2. However, variation has been alleged between asymmetric and symmetric stages of the same language (see Côté 1995; Hulk & Kemenade 1995 for Medieval French). Another recurrent variation is one between (states of) languages with strict V2, allowing only one item before the verb, as seems to be the case in contemporary standard German, and non-strict V2, typically (some periods of) Medieval French. According to Wolfe (2016), the variation is dependent on whether the verb moves to a relatively low FinP position in a non-strict V2, or to a relatively high ForceP projection in a strict V2, a high position that prevents other XPs from moving up. Evolution from strict to non-strict has been claimed for the same language at different periods (for early French, see the discussion between Rouveret 2004 and Labelle & Hirschbühler 2005).

A recurring allegation is that the XP complement in a V2 configuration is in a sentence-initial position because it has an informational value. The reasons why complement XPs are focused upon in the literature, as opposed to particles and coordinators, are two-fold. First, complements, as illustrated by e.g. verb arguments, adverbials and subordinates, have plausibly moved to the left periphery, rather than being base-generated there. Consider the following:

The initial complement du gage ‘about the pledge’ would be expected to be generated within the verbal phrase, as it is normally a post-verbal argument of the verb. As pointed out by Ledgeway (2008), no movement is the case for particles such as si, or for coordinators: the coordinator mès ‘but’ in (2) would be in initial position in any case. The movement out of the verb phrase is assumed, because it leads to the left-periphery, to have an informational motivation. The second reason to focus on complement XPs is that they may be assessed for informational value. It makes sense to ask whether du gage ‘about the pledge’ relates to information available in the antecedent discourse or not, since either option is available: in that case, as the gage is explicitly mentioned in the antecedent context, the XP represents discourse-old information. Such an assessment is difficult for particles and coordinators; there are no clear cases cited in the literature of variation between discourse-old and discourse-new particles or coordinators.

The results from the informational annotation of the occurrences of complement XPs in V2 configurations in the extended corpus Flan+ are provided in the Table 2 below.

Most texts from the corpus contain instances of V2, but not all do. Those that do not are generally brief. No text with 15 or fewer main clauses have V2⁹ – 1255 Moulin Anzerey has 60 clauses and 7 main clauses, 1318 Enquête Cotentin has 78 clauses and 14 main clauses – although one with 16 main clauses has one V2 (1472 Moulin des Noes), and one at 17 (1423 Ordonnances) has three. To be complete, we note that texts without V2 include 1278–97 Arrêts de l’échiquier with 100 clauses and 30 main ones, and 1191 Établissement Saint-Quentin with 300 clauses and 120 main – we interpret the latter as a reflection of the slump of V2 observed at the beginning of the 13^th century.

Regarding the general situation of V2, it is infrequent¹⁰ over the whole period. Looking at the number of all V2, an unexpected progression is visible. That is made clearer by aggregating the results by periods of roughly one hundred years, as in Table 3.

In the first period, until 1250, 2,4% of the clauses in the corpus display V2, a proportion that more than doubles at 5,1% in the intermediate 1250–1350 period, to go back down to 3,3% between 1350–1475. The pattern is the same if we remove formulaic sequences. Formulaic sequences are those where elements of the XP V S configuration are instantiated by similar strings of lexical elements. By way of illustration, in the 1386 Ordonnances, the sequence Si P, dit le conseil que Q ‘If P, says the Council that Q’, illustrated in (3), is repeated 6 times.

The repetition reflects not the currency of the construction in the ordinary usage of the speakers of the period, but the contribution of an expression to a specific professional genre. Removing formulaic sequences gives the following results.

The productive sequences represent 2,4% of all clauses in the 1150–1249 period, rising to 3,6% in the second period before going back to 2,1% between 1350 and 1475. A similar pattern, although with a less dissonant intermediate rise, is found in the proportion of productive V2 XP complement to all clauses. There is a movement from 1,7% of all clauses in the 1150–1249 period, to 1,9% in the intermediate period, to 0,7% between 1350 and 1475. A regular curve of evolution is found in the relation between productive complement XPs involved in V2 as compared to all productive cases of V2. The curve goes from 72% of complements representing all V2 in 1150–1249 to 52% in 1250–1349 to finally 35%, for an overall proportion of 50%.

Let us consider the main point of the investigation, which is the informational value of the preverbal complements in V2 configurations. Overall, discourse-old is the dominant value, at 73,6% of all 163 occurrences of V2 complements. A practically identical situation is found with productive cases provided in Table 4, where 75,9% of the 112 attestations are discourse-old. The evolution pattern goes from discourse-old being the exclusive¹¹ informational role of complements in the initial period, and it is not until 1236 that we have one XP with a discourse-new value, in Anglo-Norman Coutume d’Exeter. Discourse-old remains dominant from 1250 to 1349 at 82,6%. This dominance is lost between 1350–1475 when discourse-old stands at 42,2%. This is reflected by productive cases at 96,2%, 78,7% and finally 48% for each period respectively.¹²

The discourse-old occurrences relate to explicit previous mention, deixis and anaphor, and accommodation. In the following illustrations, the preverbal complement is mentioned previously,¹³

it deictically refers to an antecedent expression,

and it is accommodated as old by a Contrastive Topic, opposing the place where the service of the said Richard is due and where it is not.

Discourse-old status in the whole corpus is actualised by 24 explicit mention (20% of all occurrences), by 48 cases of deixis, 24 of anaphor and 1 familiar topic (60,8%), and 21 accommodation uses (of which 20 Contrastive Topics) (17,5%), and 2 inferred (1,7%) (see (10)–(11) below).

This raises the question of what type of discourse-new complements are found. In other words, is it information not previously evoked in the text, or is it Focus per se? Let us look at the three examples below (out of the five attestations before 1299). The first two cases below are the typical configuration with an initial discourse-new adverb, which has not been evoked before in the text. It might be that they are instances of Focus, in order to emphasise the contrast between the permission to buy things and that of selling them, and the obligation to pay anchorage fees depending on whether a ship has been in the port of Rouen before or not. However, such a status is difficult to apportion with certainty. Even greater uncertainty characterises the third case. It is unlikely to be Focus, which does not easily bear on whole subordinates not explicitly modified by a focus particle such as only; the conditional about ships constructed in England could be new information in a paragraph that lists the obligations of ships from Flanders; or it could be interpreted as a Contrastive Topic by way of contrast to the concessive clause regarding ships made in England and those that sail under a Flemish flag, in which case it would be classified as discourse-old; but the fact that the contrast would then be between a subordinate and the main clause led the Contrastive Topic analysis to be abandoned and the attestation to be classified as discourse-new.

These reinforce the point that discourse-new covers a range of cases that are not always easy to discern. They also illustrate the necessity to have text-based criteria for classification of informational values, to avoid the difficulties raised by relying on intuition alone.

To summarise, preverbal complement XPs in a V2 configuration relate to an informational value. But they do so categorically only for one period of the configuration’s attested development. It is the initial period from 1150 to 1236 that is concerned. A first discourse-new use is attested in 1236, and the proportion of discourse-new cases rises until these represent the majority of uses in the 1350–1475 period. Note that before 1236, variation in informational value is prefigured by two potentially ambiguous sequences, given below. They come from the very brief jurisprudence reports of the 1207–1211 Arrêt de l’échiquier de Normandie. The initial complements where the noun is introduced by a definite article are not explicitly mentioned before; but they are inferable from the rest of the sentence that refers to them. This is what Hansen & Visconti (2009) might call forward inferable (see Birner 2006).

These nonetheless have the potential to be interpreted as discourse-new contexts, allowing the future reanalysis of the V2 complement as having no characterised informational value.

The loss of categorical discourse-old value closely precedes the decline of V2 as a productive configuration. Decline is evidenced by the fact that productive V2 represents fewer than 1% of sentences after 1350. Also, the proportion of formulaic cases, first attested in the 1250s, rises significantly: comparing Tables 4 and 3, it goes from 0% in the first period to 33,6% in the second and 44,5% in the third. Productive V2 thus starts being lost after complements have ceased to have a categorical informational value.

One reason why informational complement V2 is declining as a productive configuration may be due to a push chain.¹⁵ The idea is that another emerging construction relating to informational structure is gaining currency and making the use of V2 unnecessary. One candidate is clefts. Clefts could be drawing away discourse-old¹⁶ complements from V2 and thus explain the decline of the latter as a productive configuration from 1250. However, the number of clefts is small in the Flan+ corpus. In line with the suggestion from the literature that the use of clefts is not significant before Middle French (Dufter 2008; Bouchard, Dufresne & Dupuis 2010; Stein & Trips 2012; see also Rouquier 2007), they were not initially annotated. For the purpose of this study, they are examined in the annotated part of the hearing text type from 1200 to 1400. Occurrences of c’ and ce are searched and retained when in subject position followed by the verb être ‘to be’, excluding the set phrase c’est à savoir ‘that is’. The result in Table 5 is that there are only 4 clefts in that part of the corpus, and they are all incomplete, not having the expected appended clause (as that John bought in It’s red trousers that John bought).

The paucity of clefts as compared to total number of V2 makes it implausible that they push V2 out of productive use. This is all the more so since the very few cleft structures occur in the earlier stage, when V2 is still a productive configuration.

Another candidate to a push chain is dislocations, adjoined phrases that are represented in the relevant clause by a resumptive pronoun.

A push chain scenario with dislocations is even more plausible than with clefts: dislocations are generally believed to represent discourse-old phrases (Ward & Birner 2001: 131ff), which is the informational value of productive V2 XPs. Thus, V2 loss could be brought about by XPs being realised as dislocations instead. As dislocations were originally annotated, the data could easily be assembled. The expectation, if dislocations are recruiting discourse-old complements away from V2, is that there should be a rise of the proportion of dislocations, especially from 1236 when V2 has a first discourse-new XP. An evaluation of these expectations can be provided by calculating the number, time-frame and informational value of occurrences in the corpus. The results of these calculations are provided in Table 6, which as a comparative includes per period the number of V2 with a complement XP and the total number of V2.

It shows that none of the expectations are met. Dislocations are not that common, appearing in less than one clause in 500. There is no sense of a progression of dislocation frequency with time; in fact, its proportion to all clauses does not rise, it decreases from 1,2 % for the 1150–1249 period to 0,3% in 1250–1349 to 0,5% between 1350 and 1475. Finally, while the dislocated XP tends to be discourse-old, discourse-new cases still represent 42% of occurrences. If there is a push chain accounting for the loss of categorical pragmatic discourse-old value of V2, as far as this corpus is concerned, the chain is not pushed by dislocations.

This section has provided quantitative data on the development of complement XPs in V2 configurations in order to assess their informational role. V2 XPs are categorically discourse-old between 1150 and 1236. A couple of occurrences during that period are potentially ambiguous to prefigure the XP informational value variation. The loss of categorical discourse-old value precedes that of V2 as a productive configuration. The first formulaic uses are attested in 1250, and will represent an increasing proportion of all V2 uses as V2 is slowly declining as a proportion of all clauses. The picture that emerges is that since the role of V2 is to profile informational values, when it starts losing its categorical discourse-old value, it then starts to disappear as a productive configuration. Categorical informational value loss appears to be the cause of loss of V2 in Medieval French.

5 Final discussion

This paper analyses the relation between information structure and word order change in Medieval French. An informational value is generally assumed to characterise the preverbal complement in V2 configurations to justify the existence of a minority construction with respect to the established SVO word order. The informational value of the preverbal complement in these configurations is tested in a novel homogeneous corpus of three types of prose legal texts from the Normandy region over the period 1150–1475. The corpus allows control for regional and textual variation, and legal texts have been shown to provide earlier chronology of change and possibly more regular curves of evolution than literary material. A decision-tree is proposed to assess informational value that relies on linguistic cues in the textual environment. Discourse-old status is attributed on the basis of explicit previous mention of the complement, deictic/anaphoric relation to the antecedent context, use in an accommodating construction such as Contrastive Topic or inference as information available to the hearer; discourse-new status characterises information that is not previously made available by the context. Note that discourse-new complements could be instantiations of Focus, but need not be.

V2 is a minority declining configuration. The preverbal complement is categorically discourse-old from 1150 to 1236. This converges with results from Ingham (2018a) who finds near-categorical discourse-old V2 complements with expressed subjects in 13^th century romance. Two earlier informationally ambiguous cases prefigure the loss of a categorical discourse-old value, which is realised by one discourse-new attestation in 1236. This change precedes the loss of V2 as a productive configuration. From 1250, V2 complements start to exhibit formulaic uses, their productive uses representing a gradually smaller proportion of all clauses, and of all V2 cases. In other words, after complement XPs lose their categorical discourse value, V2 is becoming a non-productive, declining configuration. The traditional explanation according to which loss of categorical informational value may be due to competition with other informational configurations that are dislocations and clefts is verified and invalidated: neither dislocation nor clefts, which occur mostly in the early part of the corpus, are represented in sufficient number to constitute a credible vector of change.

It was assumed that a characteristic informational value would relate to the preverbal complement of V2 configurations to explain the subsistence of a minority configuration that represents at most 5% of all clauses in any of the investigated periods. That assumption is confirmed for V2. Its preverbal complement has a categorical discourse-old value. That categorical value is lost, after which V2 starts being represented by formulaic sequences and thus to decline as a productive construction. This reflects a common pattern of language evolution: a phrase in a particular configuration has a distinctive interpretative value, that is lost with time. This can be illustrated by another item involved in Medieval French V2, the asseveration particle si. It undergoes an evolution according to which it goes “from a temporal deictic adverb to a Topic continuity marker and then to a FinP expletive which undergoes upwards reanalysis as a ForceP expletive” (Wolfe 2018b: 332). In other words, interpretative features become formal features that are liable to reinterpretation or decline. The proposed causality is that it is loss of categorical informational value that is responsible for the decline of V2 in French.

This perspective helps answer potential objections to the connection between IS and V2 referring to languages where V2 is an established configuration. V2 in contemporary Germanic languages does not seem to imply a categorical informational value for the preverbal XP, as pointed out by reviewers. However, on the one hand, it is not clear that the XP has no consistent informational value (see Bohnacker & Rosén 2007), pending a systematic empirical study. If it were found that in e.g. contemporary German or Swedish, V2 XPs had no determined informational role, this could be due to the bleaching of an interpretative into a formal feature. The question would then be why the historical loss of IS is not leading to the demise of V2 in Germanic, whereas it would have been in Romance. The crucial difference is that where Romance V2 is declining at the time where IS is lost, it is expanding in Germanic (outside English). The same differential pattern is found with other grammatical phenomena such as negation: whereas loss of informational value leads to the decline of French preverbal negative non in the Medieval period, it leads to the expansion of mica in Italian (Larrivée 2015 and references therein); expansion is also found with in situ interrogatives that are initially categorically explicitly discourse-old, as evoked in the introduction. In other words, the level of productivity of a phenomenon is also crucial to its future development when its pragmatic status changes. Whether thresholds of productivity can be defined is an interesting question that remains to be explored.

The results presented here raise a number of perspectives for future research. It would be desirable to verify the correlations established here on a larger set of attestations, especially for the earliest period; and provide more material to show further that the connection identified here between information structure loss and V2 loss is more than correlation.¹⁷ The association between categorical discourse-old value and V2 complement could be tested in homogeneous, non-literary corpora from other regions of northern Gallo-Romance in the 12^th and 13^th century. This could be done at the very least for Anglo-Norman and Flemish regions, where early documentation is available in edited format. The method could be applied to determine whether similar patterns of evolution can be found for other Romance languages where V2 was found and lost. Such an endeavour would allow the strong hypothesis that it is loss of a categorical informational value of the XP that leads to the demise of the V2 configuration to be verified.

Abbreviations

3S = third person singular, 3P = third person plural, ACC = accusative, CL = clitic, DAT = dative, F = feminine, FUT = future, INF = infinitive, NEG = negative, NOM = nominative, P = plural, PR = present tense, PST = past tense, PTCP = participle

Competing Interests

The author has no competing interests to declare.