Aboutness Topics in Old and Middle French: A corpus-based study on the fate of V2

Abstract

This article investigates the loss of aboutness topics in preverbal position in the history of French, using a corpus-based research on preverbal accusative objects. A comparison of Old and Middle French with Modern French reveals that new-information focalization had disappeared by the 14^th century, whereas aboutness topicalization had in turn vanished by the end of the 16^th century, along with other marked constructions. Combined with the generative premise that independent pragmatic factors should not trigger syntactic change, the results of this study suggest the reanalysis of the grammar as V-to-I in Renaissance French is responsible for blocking the derivation of aboutness topicalization. An alternative proposal based on phase extension and on Relativized Minimality, in a version affecting some types of A′-movement, relates those two diachronic shifts. The article concludes with the idea that the study of marked constructions may be recast as offering diagnostics on broader syntactic changes.

Résumé

Cet article examine la perte des topiques d’à-propos (aboutness topics) en position préverbale dans l'histoire du français, à partir d'un corpus de compléments accusatifs préverbaux. Une comparaison des données de l'ancien et du moyen français avec ceux du français moderne révèle que la focalisation informative avait disparu au 14^e siècle, alors que la topicalisation d’à-propos ne disparaît qu’à la fin du 16^e siècle, en même temps que d'autres constructions marquées. Combinés à la prémisse générative selon laquelle des facteurs pragmatiques indépendants ne devraient pas déclencher de changement syntaxique, les résultats de cette étude suggèrent que la réanalyse du mouvement V-à-I, dans la grammaire du français de la Renaissance, est responsable de la perte de la topicalisation d’à-propos. Une proposition alternative basée sur l'extension de phase et une version de la contrainte de minimalité relativisée visant certains types de mouvement A′ relie ces deux changements diachroniques. Ceci permet de supposer que l'étude des constructions marquées peut acquérir un nouveau rôle, et offrir ainsi des diagnostics sur des changements syntaxiques plus larges.

1. Introduction

A long-standing tradition of research in Old French has sought to unveil the relation between information structure and syntax and the verb-second (V2) hypothesis (Thurneysen 1892, Priestley 1955, Rickard 1962, Skårup 1975, Marchello-Nizia 1995, Labelle & Hirschbühler 2005, among others).¹ This article builds on this practice by offering a corpus-based investigation of the nature of left-peripheral elements from the 11^th to the 16^th century. The data, properly referenced further below, were collected from the following corpora: MCVF (Modéliser le Changement: les Voies du Français), the Penn Supplement, and the BFM (Base de Français Médiéval). The use of automatic queries helped to build a bigger database than was found in most previous works on the same subject.

Among the left-peripheral elements identified, this study focuses on aboutness topics, that is, elements stored in the common ground, which the sentence develops. Due to the V2 character of Old French, aboutness topics trigger subject-verb inversion as illustrated in (1):²

1. (1) Aboutness topicalization

   car  cest  abit _i  me  lessa  mon  pere  et  ma  mere t _i

   for  this  clothing  me  left.3sg  my  father  and  my  mother

   ‘For my father and my mother left this clothing to me’

   (de Joinville, Histoire de Saint Louis, p.17)

Aboutness topicalization shows the connection of the left-peripheral topic to a gap inside the clause. However, Old French aboutness topics also appear in a different construction – left dislocation – where the topic is resumed by a clitic, a pronoun or even an epithet, as (2) illustrates.

1. (2) Left dislocation

   Le  fourmens  et  les  orges _i,  il  les _i  ravoient  mis  par

   the  wheat.pl  and  the  barley.pl,  he  them  again.have.3pl  put.ptcp  by

   monciaux

   pieces

   ‘He had put again the wheat and the barley by parts’

   (de Joinville, Histoire de Saint Louis, p.63)

The main issue examined in this article is the disappearance of aboutness topicalization in Modern French, which seems to follow a general Romance trend (Poletto 2006, Mensching 2012), although it does not occur in Portuguese (de Andrade 2015).

Previous literature has provided three predominant hypotheses to explain this loss. The oldest, first formulated by Foulet (1919) and von Wartburg (1946), proposes that it is caused by the ambiguity resulting from the loss of the morphological case. However, as argued by Adams (1987), this idea is not tenable because some languages lacking overt case marking, such as Dutch, still have aboutness topics and V2. Therefore, the following more recent hypotheses should instead be considered.

Hypothesis A: Loss of aboutness topicalization is an independent process, possibly related to changes in pragmatic and prosodic shifts that eventually led to the loss of V2. In other words, there has been a frequency shift in the number of OSV clauses, S being in most cases a pronoun. This type of structure was ambiguous between two structures, V-to-C and V-to-I, between which the latter finally won with the fixation of pronouns as clitics (Adams 1987).

Hypothesis B: Aboutness topicalisation in French was, and continues to be, a pure form of left dislocation, disguised by the fact that the resumptive, as a weak pronoun, could instead be realized as a null pronoun. This proposal relies on the assumption that the choice between overt and null pronouns is free (Troberg, 2004).

The remainder of the article is organized as follows. Section 2 discusses the use of marked topic constructions in Modern French so as to distinguish this grammar from the ones of Old and Middle French. Section 3 presents the methodology and the qualitative and quantitative results from the corpus-based study, which helped to assess the hypotheses presented above. Section 4 examines these findings, linking the disappearance of aboutness topicalization with the fate of V2. Section 5 provides some conclusions and directions for further research.

2. Marked topic constructions in Modern French

First of all, in order to determine the type of data to be tracked in the Old and Middle French corpora, it is necessary to observe the distribution of marked topic constructions in Modern French.

The most common constructions used to convey marked topics in Modern French are left dislocation and right dislocation, where a resumptive element, such as a clitic or a possessive pronoun, is coreferential to the peripheral topic, as shown in (3) from de Cat (2007: 98, 99).

1. (3)

   1. a. Les  malotrus _i,  on  ne  les _i  invite  pas.

      the  louts  one  neg  them  invite  not

   2. b. On  ne  les _i  invite  pas,  les  malotrus _i.

      one  neg  them  invite  not  the  louts

      ‘We don't invite louts.’

With a focus on left-hand topics, de Cat (2007) observes that the well-established distinction between clitic left dislocation and hanging topic left dislocation is not clear in Modern French. These two constructions are distinguished based on the presence of agreement in case, gender, and number between the resumptive and the topic; a phenomenon also dubbed as connectivity (Cinque 1990). For this reason and for simplicity, I use the basic term left dislocation, although I consider the construction to be closer to a “hanging topic left dislocation”. In related work, in addition to left dislocation, there is frame-setting topicalization, where the peripheral constituent is a discourse topic that bears a loose relation to the clause, as depicted in the following example from Blasco (1997:11):³

1. (4) Les  lycées  techniques…  tu  arrives  avec  un  bac de technicien.

   the  high.schools  technical.pl…  you  arrive  with  a  diploma of technician

   ‘Technical high schools, you arrive with a technician diploma.’

The constituent in italics in (4) fits the definition of frame-setter in Krifka (2007: 46), as an element that frames “the set in which the following expression should be interpreted.”

The example in (5), taken from de Cat (2007: 174), represents another kind of marked construction, which conveys a contrast between the topic and other members of a set.

1. (5) Alors,  la  soupe  de  poissons _i,  on  va  mettre t _i  là?

   then,  the  soup  of  fish.pl,  one  go.3sg  put  there

   ‘And then the fish soup, shall we put it there?’

The sentence in (5) describes a scene where a set of dishes is being arranged, and the soup must be assigned a specific place. I call this construction contrastive topicalization because it encodes a topic and a contrastive focus at the same time (see Erteschik-Shir 1997 and Krifka 2007 for analyses combining topic and focus in different layers).

The previous classification raises a question about the necessity of resumption in left dislocation: is it mandatory, or are there contexts where it is optional? Barnes (1985) has pointed out that what seems to be topicalization should, in fact, be treated as a special case of left dislocation with zero anaphora in generic contexts where ça would be an appropriate resumptive pronoun, as in (6), adapted from de Cat (2007: 85).

1. (6) Les  gâteaux _i,  j'adore  (ça _i).

   the.pl  cakes  I=adore  (this)

   ‘I love cakes.’

As a matter of fact, the same pattern has been observed by Casielles-Suárez (2004) for Spanish, where non-resumption is attributed to the clitics’ inability to refer to generic NPs. Other non-resumption contexts are found in the literature. De Cat (2007) presents those related to object omission. Likewise, Rizzi (1997) notes that indirect objects and locatives are not obligatorily resumed in Italian and French, as seen below (Rizzi 1997: 322):

1. (7) Au  Pape _i,  personne  n'oserait  (lui _i)  parler  ainsi.

   to.the  Pope,  nobody  neg=would.dare.3sg  (him)  speak  like.this

   ‘Nobody would dare to talk to the Pope like this.’

The optional nature of left dislocation is analyzed in Frascarelli (2000) as a subcase of pro-licensing (Rizzi 1986) while assuming, as many authors have since Hirschbühler (1975), that the left-dislocated topic is base-generated in the left-periphery and coindexed to a constituent inside the clause. As illustrated in (7), there is a preposition that licenses the NP topic via inherent case, which is lacking in direct objects. These provisos, however, fail to explain some cases of non-resumption, such as the one in (5).

This summary indicates that topicalization is not used in Modern French to convey an aboutness topic, that is to say, an element that only retrieves old information, without a contrastive meaning. This sort of topic, illustrated in (1), is expected to appear instead in the left dislocation construction, even though the resumptive is sometimes covert for independent reasons. This paves the way for the diachronic analysis in Section 3, where it is shown that preverbal aboutness topics represent a distinguishing factor between Modern French and its earlier stages.

3. The distribution of aboutness topics in Old and Middle French

This section presents the methodology and the results obtained from the corpus-based research.

3.1 The data

The dataset used for this research consists of 665 root clauses from Old and Middle French, as well as from the beginning of Renaissance French, where nominal objects were found in a preverbal position.

Fifteen texts, ranging from the 11^th to the 16^th century, were used. Some texts are from the MCVF corpus (Martineau et al. 2007 and Martineau et al. 2010; see also the detailed bibliography in Martineau 2008), others from the Penn Supplement corpus (Santorini and Kroch 2014), and the remainder from the BFM corpus (Guillot 2014). For more details on the sampling procedure, see Appendix A.

CorpusSearch was used to extract data from the parsed texts with the help of two queries, presented in Appendix B. For the non-annotated texts, all sentences were manually counted, and tokens were included in the dataset according to the sampling criteria. Overall, the sampling procedure involved the choice of texts, whose length ranged from 6,000 to 9,000 sentences, from which the data were retrieved. This criterion excludes texts from the 11^th century, as none were sufficiently long. For the 11^th and 12^th centuries, most of the data came from poetry texts, which may have biased the results. Nevertheless, a comparison between these data and those from the texts written in the 13^th century onwards showed that they are a worthy contribution.⁴

Preverbal objects were classified according to the informational roles listed in Table 1, based on the annotation scheme proposed in Petrova and Solf (2009).⁵

Table 1: Annotation scheme of informational role

As Petrova and Solf (2009) explain, new-information focus (nf) and contrastive focus (cf) may apply within the same focus domain. While a contrastively focussed constituent is in a contrastive relation with another constituent in the discourse, new information focus contributes information that is either requested by a preceding wh-question or necessary to develop the discourse. The authors emphasize that no direct connection exists between newness, a type of information status, and new-information focus.

Although Petrova and Solf (2009) do not clearly distinguish topics, Götze et al. (2007) propose a clear-cut contrast between aboutness topics (ab) and frame-setting topics (fs), which I have adopted. Götze et al. define aboutness topics as entities “about which the sentence under discussion makes a predication” (p. 163) and frame-setting topics as “the frame within which the main predication of the respective sentence has to be interpreted” (p. 167). Petrova and Solf (2009) provide rules to identify aboutness topics based on definiteness and information status, the second of these being discussed below. To these rules, they add that the analyst must also consider aboutness and syntactic realization in the clause. Aboutness is that the sentence is about X, where X is the topic constituent, whereas syntactic realization in the clause is the identification of a marked topic construction by the use of specific features.

Because our focus here is preverbal accusative objects, no cases of frame-setting topics have been included. In addition, I distinguish contrastive topics (cts), which I consider as constituents bearing both an aboutness topic tag and a contrastive focus tag.

1. (8) [L'armee  dudit  duc  estoit  en  deux  parties.]

   the = army  of.the.said  duke  was  in  two   sections

   L'une  menoit  le  mareschal  de  Bourgongne…

   the = one  lead.pst.3sg  the  marshal   of  Burgundy

   ‘The aforesaid duke's army was (split) into two sections; the marshall of Burgundy led one (of them)…’ (de Commynes, Mémoires, p.146)

The constituent l'une [partie] in (8) could receive an ab tag because it refers to the previous referent deux parties [de l'armée]. Nevertheless, it must be classified as a ct because it also implies a choice for one of the alternatives previously set, that is, the two parts of the duke's army; and this is a particular feature of cf. There are therefore four informational roles coded in this study: new-information focus (nf), contrastive focus (cf), aboutness topic (ab), and contrastive topic (ct).⁶

From this perspective, informational status was also used to support the classification of informational roles. The annotation scheme adopted is the tagset described in Haug et al. (2014: 33), given in Table 2.⁷

Table 2: Annotation scheme of informational status

Source: Haug, Eckhoff, and Welo (2014:33)

This annotation scheme was chosen because it was shown to be one of the most efficient for encoding givenness, according to multiple tests on the use of verb-object order in Old English carried out in Taylor and Pintzuk (2014). The scheme includes a matrix of tags, categorized according to specificity and reference contexts, in addition to kind and quant, which are considered as special.

Specific tags are used for referents available for anaphoric pickup; they occur in the main narrative storyline. Nonspecific tags occur in short-term contexts set up by the narrator for specific purposes, such as relating hypothetical, negated, desired, or possible scenarios. For this reason, they are unavailable for anaphoric uptake in the main storyline but may be inside the same short-term universe. Examples for the specific tags are provided below.

Reference contexts indicate that givenness is understood as an addressee-based notion. If the referent is found in the discourse context that precedes a certain utterance, as in (9), it receives the tag old.

1. (9) Cest  mien  seignur en  bataille  faillirent

   this  my  lord  in  battle  abandoned.3pl

   ‘They abandoned this lord of mine in (the) battle.’  (La chanson de Roland, p.195)

The nonspecific counterpart of old (nonspec-old) occurs in opaque contexts, such as in the scope of negation when it is coreferent to a previous element such as quant (see (15) for this tag). If the referent is found in a scenario context, that is, one that includes knowledge such as ‘buses have drivers’, it receives the tag acc-inf, which means accessible-inferred. It is shown in (10), where the trigger for the inference ‘items in a meal’, pain (‘bread’), is shown in boldface. Its counterpart in opaque contexts is nonspec-inf.

1. (10) si  fu  l'os  si  povre  qu'il  n'i  avoit

   then  was  the = army  so  poor  that = it  neg = loc  had.3sg

   mie  pain  a   paistre   plus  de  chent  hommes,

   at.all  bread  to   feed  more  of  a.hundred  men,

   mais  char  e    vin  avoient    il    assés.

   but  meat  and  wine  have.pst.3pl    they  enough

   ‘Then the army was so poor that there was no bread to feed more than a hundred men, although they had enough meat and wine.’

   (de Clari, La conquête de Constantinople, p.102)

Furthermore, the encyclopedic context includes knowledge shared between author and recipient at the time when the text was written, as shown in (11). There is no nonspecific counterpart for this context, nor is there for the situation context, which refers to the joint surroundings of the speaker and hearer, illustrated in (12).

1. (11) Deu servirei,  le  rei  ki  tot  guvernet

   God  serve.fut.1sg  the.acc  king  that  all  rules

   ‘I will serve God, the king that governs all.'

   (de Joinville, Histoire de Saint Louis, p.231)

2. (12) Le  blanc  osberc li  ad  descust  el  cors

   the.acc  white  hauberk  him  has  unravel.ptcp  on.the  body

   ‘He [=Marganice] has unpicked the white hauberk over his [=Olivier's] body’

   (La chanson de Roland, st.145)

If the referent cannot be found in any context, the constituent is tagged as new if encountered in the main storyline, as in (13). On the occasion that the element occurs within a short-term context, it is tagged as nonspec.

1. (13) Glaive  fortraistrent  li  pecheur

   sword  drew.3pl  the.nom  sinners

   ‘The sinners drew (the) sword.’    (Psautier d'Oxford, ch.36, l.14)

Finally, two special tags are used for generic and quantified NPs: kind and quant. The former does not refer to individuals as other specific NPs do, as shown in (14). The latter may be non-referential or include reference to a plural discourse representation, as in (15).

1. (14) Viandes envia  à  els  en  abundance

   meat.pl  sent.3sg  to  them  in  abundance

   ‘He has sent them meat in abundance.’ (Psautier d'Oxford, ch.77, l.29)

2. (15) Pluisseurs  paroles disoit  on  de  li  en  France  et  a  Paris.

   many  words  said.3sg  one  of  him  in  France  and  in  Paris

   ‘People said many words about him in France and in Paris.’

   (Froissart, Chroniques, p.483)

The classification above helps to identify non-referential constituents, which cannot be topics. The identification of nonspecific referents is also valuable because they are less likely to act as topics. The previously described methodology thus provides a safer way to identify informational roles than the correspondence rules in Marchello-Nizia (1995), which are uniquely based on the form of the NP. However, there is a special category in that analysis that consists of objects included in light-verb constructions, especially common in the older periods of French. These objects were counted as nonref because they do not have a referent, and do not fit any of the mentioned categories. Following Marchello-Nizia's observations, these objects were coded in the present study as new-information focus.

3.2 Qualitative results

Before a more extensive presentation of the results, I turn first to a systematic study of preverbal objects in the Middle French text of de Commynes (1964 [1498]), to test the predictions made by hypothesis B. This is a crucial step; if topicalization in Old and Middle French is indeed left dislocation in disguise, it would critically affect the presentation of the quantitative results, not to mention the very direction of the analysis.

Hypothesis B rests on two arguments: (i) the resumptive in the purported left dislocation is freely null, which means that overt resumptives are not favored in any structural environments; (ii) only null pronouns, not null variables, can be used in argumental positions. An examination of both arguments follows.

As for structural environments, embedded contexts have been shown to interact with only some types of marked constructions (Bianchi and Frascarelli 2010). For instance, topicalization is accepted only in root clauses and embedded clauses with root properties in English (Haegeman 2012 and references therein). In contrast, (16) and (17) suggest that both aboutness topicalization and left dislocation were possible in embedded clauses of Old French:

1. (16) je  sai  [que  vous  la  derrenier  aumosne me  donrez t _i].

   I  know  that  you  the  last  charity  to.me  give.fut.2pl

   ‘I know that you will give me the last charity’ (Roman de Cassidorus, p.215)

2. (17) Gardez  [que  ces  lettres  que  vous  receüstes

   beware  that  these  letters  that   you   received.2sg

   par  Melsius _i  que   il   les _i   vous   envoiast].

   from  Melsius  that   he  them  to.you  sent.3sg

   ‘Watch if it is indeed the case that it was Melsius who sent the letters that you have received from him.’

   (Lit.: ‘Pay attention that, these letters that you received from Melsius, that he indeed sent them to you.’) (Roman de Cassidorus, p.330)

The structure in (16) has a different distribution from Modern French. Notice that, if this construction were contrastive topicalization, it should not fit this sort of embedded context, as is the case in English. Moreover, the example in (17) shows that left dislocation is also available in embedded clauses. Therefore, syntactic contexts do not help to distinguish the two constructions in a synchronic analysis of the Middle French data, although it is diachronically different from Modern French.

Pragmatic contexts can also be taken into account to distinguish different uses of topicalization, as highlighted in (18) and (19).

1. (18) [le  secours  qui  vint  de  Bourgongne,  que  mena

   the  help  that  came  from  Burgundy,  that  led.3sg

   le  seigneur  de  Coulches,  le marquis de Rothelin,

   the  lord  of  Couches,   the marquis of Rothelin,

   le  seigneur  de  Montagu et   autres] […]

   the  lord  of  Montaigu  and  others

   Ceste  assemblee avoient      faicte    en  Bourgongne

   this    assembly   have.pst.3pl    do.ptcp  in  Burgundy

   le  conte  de   Beaujeu  et    le   cardinal  de  Dourbon…

   the  count  of   Beaujeu  and    the  cardinal  of  Dourbon

   ‘[…the help that came from Burgundy, which was led by the lord of Couches, the marquis of Rothelin, the lord of Montaigu and others] […] Le count of Beaujeu and the cardinal of Dourbon had gathered this meeting.’  (de Commynes, Mémoires, p.18)

2. (19) [Des  gens  de  cheval  qui  estoient  avec  ledict duc

   of.the  people  of  horse  that   were.3pl  with   the.said duke

   de  Bourgogne,  il   en   y  envoya  une  partie

   of  Burgundy,  he   gen  loc  sent.3sg  a   portion

   pour   donner  la   chasse;] […]

   to   give  the   hunt

   Autres  renvoya   devant    la   ville…

   others   sent.3sg   in.front  the  city

   ‘[The duke of Burgundy sent some of the horsemen that were with him to give chase…] He sent others to the front of the city…’  (de Commynes, Mémoires, p.107)

In (18), the topic constituent ceste assemblee corresponds to an ab constituent, where only its reference back to a group of personalities is relevant. On the other hand, autres, in (19), is a ct constituent. Observe that the topic antecedents or anchors, boldfaced in the contexts provided between square brackets, differ only in contrast between the members of the inferred partially-ordered set presupposed in (19), which is {groups of knights}, to be compared with {Burgundian nobles} in (18) (see Birner and Ward 1998 for more information on partially-ordered sets and their pragmatic functions).

In the examples of left dislocation found in this text, there is always a contrastive relation between the topic and other members of the partially-ordered set. This relation can be either implicit or explicit, as in (20).

1. (20) [Et  ja  estoit  commencée  une  forte  et

   and  already  was.3sg  begun.ptcp  a  important  and

   grosse  escharmouche  au   bout  du   village  de   Montlehery,

   large    skirmish   at.the  end    of.the  village  of   Montlhéry,

   toute  d'archiers  d'ung    costé  et    d'autre.]

   all  of = archers  of = a    side   and    of = other

   Ceulx de la part du Roy _i les _i conduisoit Poncet de Rivière…

   those of the side of.the king them led.3sg Poncet de Rivière

   ‘[And a large skirmish had already begun at the end of the village of Montlhéry, composed of archers on both sides.] Poncet de Rivière led those on the king's side.’

   (de Commynes, Mémoires, p.25)

Left dislocation therefore seems to be in competition with contrastive topicalization in this grammar, in the sense that both were apparently used in the same pragmatic context, unlike aboutness topicalization.

Let us now consider the second argument for hypothesis B, according to which only null pronouns can be used in argumental positions. To frame the discussion, we must first establish some technical assumptions about the argument position in relation to the derivation of topicalization and left dislocation. Most authors have assumed that the latter derivation involves base-generation of the topic in the left-periphery of the clause, with the topic and resumptive connected by coreference. In contrast, the classical derivation of topicalization implies A′ movement which leaves a null variable behind. This difference accounts for the fact that topicalization is sensitive to islands, but that dislocation is not, as demonstrated in the pair below, from Chomsky (1977: 91):

1. (21)

   1. a. * This book, I wonder [who read t _i].

   2. b. As for this book _i, I think [you should read it _i].

The question at hand is whether there is a gap left by movement, or rather a pro, in the base position of the topic in Old French. I use pro _i to indicate a null pronoun that occurs (for various reasons) instead of a resumptive pronoun, and t _i to mark a gap left by movement, which is interpreted as a variable. In this regard, debate about the nature of null objects in Old French in comparison to those of Modern French has carried on for quite some time. The prevalent view was that all null objects should be analyzed as null pronouns (Arteaga 1998). Another view, that of Donaldson (2013), is that like (22), an example far easier to analyze when the null object is treated as a variable:

1. (22) Sa  grand  valor _i,  ki  purreit  acunter t _i?

   his  great  value,  who  can.cond.3sg  measure ?

   ‘Who could measure his great value?’

   (Chanson de Roland, l.534,

   cited in Donaldson 2013: 79)

Arteaga (1998) questioned the derivation of an example similar to (22), given a minimality restriction that should block the movement of the object over the A′ element ki. Actually, topics cannot be found in such contexts in English: they are in complementary distribution with both focus constituents and wh-words, as first observed by Chomsky (1977). However, aboutness topicalization can be found in a context similar to (22) in European Portuguese, as discussed in Duarte (1987: 333) and illustrated in (23).⁸

1. (23) Esses  livros _i,  a  quem  ofereceu  o  João t _i?

   these  books,  to  whom  offer.pst.3sg  the  João

   ‘To whom did João offer these books?’

The solution to this apparent inconsistency involves adopting a finer conception of Relativized Minimality, where it is posited that different sorts of topicalization may be derived cross-linguistically, the English type being encoded in a lower projection in the left periphery (Haegeman 2012).

The interim conclusion is that hypothesis B cannot be correct. Although the two constructions have similar structural environments, Old French topicalization differs from its Modern French counterpart, where it is found only in root contexts. Furthermore, their pragmatic contexts are distinct, in that aboutness topicalization, which does not involve a contrast between the members of an inferred partially-ordered set, was available in addition to contrastive topicalization and left dislocation. Ultimately, there are reasons to believe that null variables were present in Old French, in precisely the contexts where topicalization took place.

3.3 Quantitative results

Table 3 presents the data retrieved by the automated queries, before the exclusion of specific tokens. It gives a bird-eye view of the number of preverbal objects, from which abs were then selected.

Table 3: OV versus VO in root clauses of the base texts

The data can be divided into two periods. In the texts from the 11^th and 12^th centuries, the proportion of OV was 20% or higher. From the 13^th century to the 16^th century, this percentage decreases to 10% or less. The OV percentages dwindled to less than 5% in the 15^th and 16^th centuries, after which this word order was completely abandoned.

It is questionable whether these results truly represent a case of linguistic change, given that the available texts from the 11^th and 12^th centuries are all in the form of poetry. The order may also depend on idiosyncratic features of the text, as the inconsistent percentages of OV found in the two poetry texts from the 11^th century imply. Regardless, since OV is still found in the Psautier d'Oxford (1860 [ca. 1150]), a 12^th century prose text, the presence of OV may not be genre-specific.

Another perspective, shown in Figure 1 below, illustrates the relative frequency of informational roles century by century.⁹ The most noticeable change is the decline in new-information focus (nf) occurrences, which become very rare in the 15^th century, and completely disappear by the 16^th century. The incidence of the other roles increases throughout this period, with a prevalence of abs that decreases only in the 13^th century, at the same time that cfs rose in frequency. Lastly, cts show fairly constant rates and are less common than the other roles.

Figure 1: Distribution of informational roles by century (in percentage)

The results in Figure 1 should be interpreted based on Table 3. More specifically, from the 13^th century on, the results indicate that preverbal topics and foci became marked, suggesting that nf came to be conveyed at the right of the clause, thus having an impact on the overall distribution of other marked constructions. Concretely, although the frequency of abs analogously increased, its frequency decreased in the overall corpus. This picture is consistent with the results in Labelle and Hirschbühler (2012), showing that the number of preverbal foci diminished progressively in Old French.

The next two tables shine a light on preverbal objects that are abs, which correspond to half of the collected data and the theme of this paper. Table 4 correlates these numbers with subject realization in each century.¹⁰

Table 4: Subject realization with aboutness topics in root clauses of the base texts

abs usually occur in clauses with referential null subjects. Their overall mean (69%) rises significantly (to 88%) in the 12^th century. A similar peculiarity also occurs with postverbal subjects, which are usually less frequent than null subjects, but whose frequencies exceed 50% for the 13^th, 14^th and 16^th centuries. This variable is relevant, as it may indicate whether abs are indeed related to a V2 structure and understood as entwined with V-to-C movement, since this phenomenon should trigger subject inversion.

Table 5 pursues the same rationale; it examines whether there is a potential connection between object abs and the V2 effect.¹¹ Although there were more tokens with V2 than there were with V > 2 throughout our dataset, the percentage of V > 2 still amounted to around 40% of the catalogued cases in the 11^th and 15^th centuries.

Table 5: Verb position with aboutness topics in root clauses of the base texts

Further analysis of the V > 2 tokens revealed that most of them correspond to the XOV type (33 tokens), followed by OXV (11 tokens), OSV (10) and SOV (8). Most OSV cases have a pronominal subject.

These results show that aboutness topicalization strongly correlates with postverbal subjects and V2 during the period when postverbal subjects were in use. In addition, Figure 1 demonstrates that new-information focalization disappeared by the end of the 14^th century, whereas aboutness topicalization was still in use until the 16^th century. Consequently, hypothesis A does not answer the research question. Its formulation probably did not take into account the distinction between preverbal topics and foci, which is now better understood. At any rate, abs seem to be closely associated with the very fate of V2.

4. Aboutness topics and the fate of V2

In this section, I discuss the implications of the results in the context of a previous analysis of the dissolution of the V2 system, to give an explanatory-based account for this change. From there, I develop an explanation for the demise of topicalization, based on the assumption that a syntactic change in Middle French caused the loss of aboutness topicalization. Unlike the loss of new-information focalization, this change cannot be explained by prosody specifically (i.e., the loss of sentence-initial stress; see Marchello-Nizia 1995, among others). Although topicalized constituents must be stressed by a particular intonation, prosody alone does not determine their very occurrence, as opposed to (some) focalized constituents.

4.1. A further hypothesis on topics and V2

To ensure the accuracy of the analysis of aboutness topics in Old French, the parallel-phases hypothesis, first put forward by Poletto (2006) and elaborated upon in Mensching (2012), has been incorporated here because of the finer distinctions it makes between informational roles. This particular hypothesis correlates the availability of topicalization with vP scrambling, a construction especially found in compound tenses and illustrated in (24) for Old Italian.

1. (24) ch'egli  avea  il  maleficio commesso

   that=he  had.3sg  the  crime  committed

   ‘that he had committed the crime’

   (Giamboni, Fiore di rettorica, 1994[1292]: 34, cited in Poletto 2006:265)

Mensching (2012) equates the loss of an optional movement feature in the phase head v with the loss of topicalization. Example (25a) shows the derivation in Old Spanish, and (25b) the one in Modern Spanish, where a clitic phrase (clP) replaces vP in its role:

1. (25)

   1. a. [_IP NP_DO  [_vP NP_DO … [  NP_DO]]]

   2. b. [_CP NP_DO  [_clP NP_DO cl [_IP [_vP NP_DO]]]]

According to the minimalist account of syntactic phases (Chomsky, 2001), any clause is divided into two domains that are visible to the interfaces. Similarly, this proposal also presupposes that, to be displaced to the left periphery, a topic constituent must stop at each intervening phase edge. In the basic representations above, the intermediary NP_DO stays in the position where scrambled objects are found. For some reason, the topical NP does not always complete its movement to the CP periphery; it sometimes stops in the lower vP periphery. Thus, this theory predicts that in the period when there were fronted objects, we should also find scrambled objects with similar properties.

With an emphasis on providing an explanation for the loss of topicalization, this proposal assumes, similarly to hypothesis A, that this shift is the cause of the demise of the V2 system, with an additional step: the loss of vP scrambling. However, by doing so, it formalizes a functional observation related to the use of certain marked constructions, a proposal that naturally implies that pragmatic changes trigger syntactic changes, going against basic generative premises. An additional pitfall in Mensching's (2012) specific rendering of the hypothesis for this change is the prediction that topicalization is replaced by left dislocation. This conjecture goes against empirical evidence found in the base texts studied ‒ the latter construction remains very marked until the Renaissance period.

The proposal opens up new problems – germane to the parallelism between topicalization and scrambling – whose proper discussion goes beyond this text. Not much is known about vP scrambling, but Zaring (2010) observes that OV order in infinitival and participial domains continued to be frequent in the 13^th century, when OV order was rather rare in finite contexts. This finding suggests that this type of OV is relatively independent of aboutness topicalization. To verify this, I have queried for participles in compound tenses and for “restructuring” infinitives, which project up to vP, in the base texts written in the 15^th and 16^th centuries.¹² The results show that scrambled NPs are similar either to focus or to topics, as the examples in (26) and (27) suggest.

1. (26) ung  prestre  ne  doit  ame tuer

   a  priest  neg  should  soul  kill

   ‘a priest should not kill a soul.’ (Cent nouvelles nouvelles, p.62)

2. (27) et  povez  l'absence  de  mon  bon  mary  supplier…

   and  could  the=absence  of  my  good  husband  compensate

   ‘and you could compensate the absence of my good husband…’

   (Cent nouvelles nouvelles, p.574)

Nevertheless, evidence of scrambled topics ceases to be found by the end of the 15^th century, although the presence of aboutness topicalization was detected in some of the later texts. In any case, scrambled foci of quantifiers such as plus, jamais, aucunement persisted in Renaissance French:

1. (28) J'ay  beaucoup entendu  de  son  affaire.

   I = have  much  understand.ptcp  of  his  business

   ‘I have understood much of his business.’

   (de Mailles, Histoire du Seigneur de Bayart, p.108)

This construction was retained, and is still accepted in Modern French (Abeillé and Godard 2004), although no counterpart in the CP periphery has been reported.¹³ This may suggest that vP-scrambled topics and CP-moved topics can be independently generated, and encode different sorts of elements.¹⁴ Therefore, vP scrambling needs to be further studied, but it is unlikely that it had an impact on the dynamics of change, or precisely, that a shift in the vP periphery had an impact on the CP periphery.

4.2 An alternative proposal

In order to build an alternative proposal, it is necessary to bear in mind that there are two types of topics, moved and base-generated, as discussed in Section 3.2. Salvesen (2013) conveys this, to a degree, by saying that topics are invisible to the V2 configuration. If we adopt Benincà and Poletto's (2004) cartographic proposal for the left periphery, the only elements that can count as being in the first clausal position thus occupy either SpecFocP or SpecFinP. The explanation for this invisibility is the base-generation of the topic in the peripheral position, being coreferential to a pro argument inside the clause.¹⁵ For Salvesen (2013), elements occupying SpecFinP are informationally “neutral.”

There are two ways the term “neutral” can be understood. Some use it to refer to constituents that are typically destressed and realized in a pronominal form: familiar topics (fm; see Frascarelli and Hinterhölzl 2007). This cannot be the relevant sense of “neutral”, as not only familiar topics and foci correlate with V2, but also aboutness topics, as shown in Table 5 above. In some centuries, they even showed a strong correlation with postverbal subjects, a sign of Germanic inversion occurring when the subject is adjacent to the verb:

1. (29) [et  puis  se  le  revesti  on  le  palle]

   and  then  adv  him.dat  put.on.3sg  one  the.acc  coat

   et  puis  le   mantel   li  frema   on   seur   l'espaulle

   and  then  the.acc  surcoat  him.dat  fasten   one   over   the = shoulder

   ‘And then they put the coat on him, then they fastened the surcoat over his shoulders.’

   (de Clari, La conquête de Constantinople, p.95)

In (29), the constituent le mantel (‘the surcoat’) illustrates a typical case of ab, used in a descriptive context where the king wears different pieces of clothing. Therefore, a more reasonable proposal is to interpret “neutral” as nonspecific. Hence, I adopt the idea that [Spec,FinP] is a multifunctional position, as other traditional accounts for V2 have done (e.g., Roberts 2004). A product of this idea is shown in the clause structure in (30), where syntactic positions are connected to informational roles, using a simplified set of projections ‒ up arrow indicates Merge, and left arrow represents Move; ForceP is not projected because this structure applies to root contexts:

1. (30) [_FrameP  [_TopicP  [_FocusP  [_FinP  [_IP

   ↑ fs  ←ab, ct ← cf, nf ← ab, ct ←fm,

   cf, nf Scrambled XP

The constituent that moves to [Spec,FinP] is one of the informational elements that appear below FinP in the scheme in (30), from where they move further, either to [Spec,TopicP] or to [Spec,FocusP], depending on the type of information they convey.

In addition, I adapt Frey's (2004) framework, according to which some elements in the left periphery move by True A′-movement or by Formal Movement. This distinction is based on the motive for movement: in true Aʹ-movement, it is guided by a criterial feature, whereas in formal movement no interpretative effect is involved, only the need to value the EPP feature of Fin. This last case implies movement of either a fm or a scrambled XP, and occurs if none of the elements shown below FinP in (30) is available in the sentence.¹⁶ Unlike in Frey's (2004) account, there is no need to assume an intermediary KontrP projection for items with criterial features, understood as contrastive. Instead, these are assumed to have priority in their checking over formal features, implying that Formal Movement will occur if and only if there is no [topic], [focus] or [contrast] feature, nor a combination thereof, involved.

This proposal predicts that, whenever there is a preverbal object, V2 is the predominant order. Nevertheless, V3 orders such as XOV and OXV are also possible, as in the examples below:

1. (31) mes  de  tel  mestier  riens ne  sai…

   but  of  this  matter  nothing  neg know.1sg

   ‘but regarding this matter, I know nothing [at all]…’ (Roman de Cassidorus, p.356)

2. (32) Branches  d'olive  en  vos  mains  portereiz…

   branches  of = olive  in  your  hands  carry.fut.2pl

   ‘You will carry olive branches in your hands.’ (La chanson de Roland, st.13)

In (31), X corresponds to fs, and O, to nf. In (32), O embodies the ab role, whereas X plays the role of cf. This example is interesting because it indicates that locative items may function either as fs or as cf, while a contrastive interpretation is required in the configuration above. Examples (33) and (34) show SOV and OSV orders respectively:

1. (33) Li  rei  des  Arabiens  e  de  Saba  duns amerrunt

   the.nom  kings  of.the  Arabs  and  of  Saba  gifts  bring.fut.3pl

   ‘The kings of the Arabs and of Saba will bring gifts.’

   (Psautier d'Oxford, ch.71, l.10)

2. (34) Les  chars  crues  il  mettent  entre  leur  celles  et  leur

   the.pl  meat.pl  raw.pl  they  put  between  their  saddles  and  their

   paniaus

   baskets

   ‘They put the raw meat between their saddles and their baskets.’

   (de Joinville, Histoire de Saint Louis, p.239)

Unlike the cases with XOV and OXV, SOV and OSV orders were unevenly distributed through time. SOV was apparently conventional in the 11^th and 12^th centuries, whereas OSV was more frequent from the 13^th century onwards. Thus I interpret that different grammars, involving different verb positions, are involved in these cases. In the former period there was a V-final structure (V in situ) competing with the V-to-C grammar. On the other hand, in the latter period V-to-I was in competition with V-to-C. Therefore in (33), the subject is an ab, and the object is nf, but only the subject moves into the left periphery, because it is the higher element in the vP phase. In (34), the object is an ab, and the subject is a fm; only the object moves, because it is equidistant from the subject when in an IP-scrambling position, which I assume was lost at the same time as nf.

I also assume that loss of verb movement came about with the change in null subjects and is linked with the loss of new-information focalization (see, respectively, Vance 1997 and Labelle and Hirschbühler 2012). The OSV order, most common in the Middle French period, is understood as indicating a V-to-I analysis, hence triggering the reanalysis that established V-to-I as an innovative structure. Consequently, the loss of aboutness topicalization should be seen as an an effect of this change; and not its trigger, which naturally gives rise to the a third hypothesis, Hypothesis C.

Hypothesis C: The V-to-I grammar of Renaissance French blocked the derivation of aboutness topicalization.

I adhere to the framework of phase theory (Chomsky 2001, among others), further developed in den Dikken (2007) and discussed in Pesetsky (2007) regarding the notion of “phase extension.” According to den Dikken, the position into which a phase head moves determines the limits of a syntactic phase. Therefore, if verb movement stops at I⁰, [Spec,IP] becomes the phase edge, as in (35b).

1. (35)

   1. a. [_CP [_IP {_PhaseEdge [_vP V + v [_VP V]]]]

   2. b. [_CP {_PhaseEdge [_IP V + v+T [_vP V + v [_VP V]]]]

As Pesetsky (2007) points out, the movement triggered by I⁰ in (35b) counts as an A′ movement; it is a property of phasal heads. The adoption of the phase extension model combined with Rizzi's (1990) Relativized Minimality condition implies that an intervening governor, the subject, blocks a government relation of the same kind. Notice that both abs and fms have a [topic] feature, but this feature is already checked in [Spec,IP], as there is no difference in criterial features between the two. Therefore, an ab cannot move over a subject.

Two implications stem from the proposal relating V-to-C movement and aboutness topicalization. The first one is that only aboutness topics are affected, although contrastive focalization and contrastive topicalization have disappeared altogether from Renaissance French, according to the dataset used here. These elements are featurally distinct from the subject because they carry a [contrast] feature and both [contrast] and [topic] features, respectively; thus Relativized Minimality does not block their movement. Such an implication is welcome because preverbal cts are possible in Modern French, and preverbal cfs are possible in other Modern Romance languages, like Spanish.¹⁷

The second implication of the proposal is that only languages with V-to-C movement may have aboutness topicalization. This is a strong prediction, and a discussion of it would surpass the goals of the present article. Among Romance and Germanic languages, this seems to be correct. For instance, English adheres to this prediction − so much so that the very existence of topicalization is sometimes put into question (Prince 1999; see also Speyer 2010 for the correlation between contrast and topicalization in English). Portuguese, on the contrary, remains a noticeable exception. However, it also happens to have archaic feature remnants, such as enclitic pronouns in finite clauses, which are associated with the activation of the left periphery, and specifically with V-to-C movement (Raposo 2000 and others).

Since the syntax is frequently opaque to informational properties, it may or may not allow the movement of aboutness topics, as well as of other informational categories. This explains in part the existence of cross-linguistic variation on how such categories are expressed.

5. Conclusion

In this paper, I have discussed aboutness topics in the history of French to understand their disappearance in preverbal position. To do so, I have conducted a corpus-based study of preverbal accusative objects in fifteen texts from the Old and Middle French periods up to Renaissance French, from the both qualitative and quantitative points of view. The data indicate that contrastive topicalization and left dislocation used to encode the same informational role, that is to say, ct (contrastive topic). Second, contrastive focalization and aboutness topicalization were lost by the end of the 16^th century. Finally, new-information focalization was lost by the 14^th century, with only traces remaning in the 15^th century.

Based on the available data, I have concluded that V2 holds in Old and Middle French, as there is almost no competition with other data that indicate an older V-final grammar or a newer V-to-I grammar. In this scenario, [Spec,FinP] is a multifunctional position that must be realized by some constituent moved into it. True A′ Movement takes precedence over Formal Movement, which only selects the closest element available. In this sense, there is an implicit connection between the disappearance of aboutness topicalization and the fate of V2: once the V-to-C grammar was reanalyzed as V-to-I, the movement of aboutness topics was blocked by the Relativized Minimality restriction, assuming Phase Extension.

This view of the change is different from the prevalent one, according to which aboutness topics disappeared for independent reasons and caused the loss of V2. The reason to bring forward this new proposal has to do with the assumption that informational-structural notions should not change independently. Instead, the morphosyntactic profile of each language constrains the way information structures are expressed. If we take seriously the generative premise that syntax is relatively autonomous from pragmatics, the diachronic study of information structure should consider marked constructions as diagnostics or effects of more general grammatical phenomena.