2.2 Database of the study

We selected and manually tagged a comprehensive selection of historical legal French texts (with one exception discussed below) to reveal the full diachrony of CC: our corpus contains 19 texts that cover a period of more than 700 years, from 1150 to 1856 (Table 1). This time span aims at casting a wide net to collect data from the earliest texts available until shortly after the definite loss of CC in order to reveal the full evolution of the phenomenon.

Most of the texts in the corpus are from Normandy, except for Corpus Philippicum and Actes Royaux du Poitou. We have chosen this region in particular because of the availability and amount of textual material.

As noted above, the corpus is very homogeneous but there are a few exceptions that we elaborate on here. The corpus includes the verse text Le Roman de Brut. We have added this text to counter the otherwise low quantity of clitics found for the 12^th century with the expectation that clitic placement in Brut would help draw more robust empirical generalisations for that century. One reviewer questions the inclusion of this non-legal text in an otherwise homogeneous corpus, however we did not identify any configuration that is specific to this text – in fact, even if Brut and Lois de Guillaume are different text types, trends of clitic placement are similar. Despite our methodological choice to focus on legal material, the comparison between Brut and Lois de Guillaume (which were written five years apart) may suggest that clitic placement does not change much across text types, if at all.

Most of the corpus is constituted of texts issued from the Corpus ConDÉ (Larrivée & Goux 2021),⁶ the online corpus SCRIPTA (Bauduin 2016),⁷ the Corpus Philippicum (which is not Normandy specific, as it contains texts from the North of France, Lalou & Hélary 2017), and the online corpus Actes Royaux du Poitou (which is not from Normandy either yet was necessary to augment the number of clitics of the early 14^th century, École des Chartes 2011).

2.3 Distribution of clitic climbing per century

As we can see below, CC in infinitival clauses is present in all texts of our corpus with the exception of Pannier, our most recent text. Table 2 reports the percentage of CC with restructuring predicates per century. From a descriptive standpoint, we note that the ultimate loss of CC took place recently in the history of French.

The construction is frequent in Old French (12^th to 13^th centuries) and it is on the increase until it reaches 100% during the Middle French period (14^th to 15^th centuries). These findings echo Foulet (1919: 112) and Roberts’ (1997: 448) claims that CC was obligatory in earlier French, which we return to in sections 3 and 4. By the end of the 16^th century, the construction shows the first sign of a decrease that will accelerate during the following century. The frequency of CC is relatively stable until the 18^th century, where it is found in similar quantities as proclisis, until it abruptly disappears at the turn of the 19^th century.⁸

Most occurrences of CC are found where the infinitive is introduced by a modal verb, which for the Old French period includes: deveir ‘must’, poeir ‘can’, voleir ‘want’, saveir ‘know’, soloir ‘be in the habit of’, estovoir ‘be necessary’. These modals are typical restructuring verbs across Romance, and the latter two do not exist in Modern French anymore. Although CC is by far more common with modals, we counted sporadic examples of CC with other predicates: convenir ‘be suited’ (7 occurrences, which we may arguably consider to be a modal), valoir ‘be worth’ (4 occurrences), aler ‘go’ (2 occurrences), oser ‘dare’ (2 occurrences), cuidier ‘believe’ (1 occurrence). Still for the Old French period, we found cases of CC where the lower infinitive is introduced by what has been called a complementizer, a preposition, or a subordinator in the literature, namely à: tenir a ‘have to’ (3 occurrences), enconvenancier a ‘promise’ (2 occurrences), trover a ‘find’ (1 occurrence), obliger a ‘force’ (1 occurrence).

CC is found with more predicates during the Middle French period. Similarly to the previous period, modal verbs are the main environment in which the clitic climbs: devoir ‘must’, pouvoir ‘can’, vouloir ‘want’, savoir ‘know’, souloir ‘be in the habit of’, falloir ‘be necessary’. Other predicates selecting a bare infinitive includes: cuider ‘believe’ (17 occurrences), aller ‘go’ (13 occurrences), convenir ‘be suited’ (10 occurrences), oser ‘dare’ (9 occurrences), venir ‘come’ (6 occurrences), décerner ‘declare’ (1 occurrence), esconvenir ‘be suited’ (1 occurrence), tenir ‘hold’ (1 occurrence), retourner ‘go back’ (1 occurrence), penser ‘think’ (1 occurrence). In a similar fashion to the Old French period, instances of CC ‘over’ a complementizer are not common, however in addition to climbing over à, we attest cases of climbing over de as well: menacer de ‘threaten’ (2 occurrences), tenir de ‘have to’ (1 occurrence), promettre à ‘promise’ (1 occurrence), convenir à ‘be suitable to’ (1 occurrence). Interestingly, we find tenir Ø, tenir à and tenir de, a variation that has been discussed by Martineau & Motapanyane (2000). According to these authors, French initially went through a stage of alternation between Ø and à, before a rise of de during the Middle French period and a decrease of Ø. We formally analyse this construction and its diachrony in section 6.1.

In terms of the types of clitics we find, the language of the law is largely restricted to what he, she and they must and mustn’t do. Consequently, our corpus is mostly composed of third person clitics and locative clitics, with only a handful of occurrences of first and second person clitics. Crucially however, our overall findings are perfectly in line with Martineau’s (1990) study, whose corpus contains more first and second person clitics for the Middle French period. The fact that historical data are constrained by the material and indeed the genre they use is well-known and unavoidable (see for instance Stolk 2015, whose work on Greek is limited to the first person and the second person, exactly because she focuses on data from papyri, which in turn are predominantly letters).

Traditionally, the loss of CC in French is situated shortly after 1650 (Iglesias 2015; Amatuzzi et al. 2020), with a longer persistence in the south where the language was in contact with Occitan, where CC remains today (Bekowies & McLaughlin 2020). To further pin down the loss of CC in our corpus, Table 3 presents its frequency for each text from the 16^th century onwards.

In line with the aforementioned studies, we observe a dramatic decrease in the frequency of CC during the mid-17^th century, even though the construction remains present for more than a hundred years after. Moreover, its frequency does not decrease steadily, but instead it has some fluctuation within Basnage, Pesnelle, and Merville until its eventual decline in Pannier.

3.1 Syntactic versus phonological cliticisation

Based on empirical evidence from Italo-Romance, Ledgeway & Lombardi (2005) show that cliticisation is necessarily verb-adjacent in some grammars (e.g. Italian), whilst it allows interpolation in others (e.g. Cosentino). Interpolation is a word order characterised by the presence of an intervening element between the clitic and its verbal host, as in (5), where the adverb ggià can appear between the clitic and the verb.

Whilst in Italian the clitic is always adjacent to the verb (4), and interpolation is ungrammatical, Cosentino allows adverbs to intervene (5), thus splitting the clitic from the verb. Interpolation is reported in a wide range of Romance varieties, both in modern varieties and their diachrony (Triestino, Calabrian, Portuguese, Rumanian, early Spanish, Middle French and a handful of Italian dialects show or have shown this phenomenon at some point).

Ledgeway & Lombardi (2005) use (4) and (5) as evidence for the existence of two different cliticisation mechanisms: syntactic (cf. Italian) and phonological (cf. Cosentino). With this proposal, they argue that cliticisation is an operation found early in the derivation in varieties that have syntactic cliticisation, whereas it is only a PF-phenomenon in varieties that allow interpolation. This view addresses the long-standing observation that clitics must be analysed on two levels, namely syntax and phonology (Kayne 1984).

Assuming that the clitic is a φ-head (Dechaine & Wiltschko 2002) that Agrees with v (Roberts 2010) we can formalise Ledgeway & Lombardi’s (2005) hypothesis in (6) and (7) below. Cliticisation is phonological when the clitic does not need to ‘fuse’ with its selecting verb, and hence interpolation can take place. We interpret this empirical observation as indicating that the clitic and the verb target different heads in the clausal spine, allowing for the insertion of elements such as adverbs. Contrarily, in the case of syntactic cliticisation, the clitic and its verb must be adjacent, and interpolation results in ungrammaticality. This suggests that the clitic and the verb form a complex head in narrow syntax. In section 7 we show that this distinction is crucial to account for changes in clitic placement in French.

We now turn to enclisis with infinitives in Italian, Catalan, and Spanish. According to the traditional view, the infinitive targets a position above the clitic (Kayne 1991). This hypothesis has been supported by Schifano’s (2018) study on verb placement in Romance, as she shows that infinitives move higher in the abovementioned varieties. Since, in this context and in these languages, the verb and the clitic occupy different heads, we conclude that cliticisation is phonological (8), that is the clitic leans on the verb at Spell-Out (but not in the syntax). We follow the assumption that the HMC does not exist (see Roberts 2010: Chapter 5).

One anonymous reviewer points out that interpolation, which we take as a useful theoretical tool for diagnosing phonological cliticisation, is typically not found in the context of enclisis. It is nevertheless possible in other varieties, as observed with finite verbs in Borgomanerese (Tortora 2010):

Why does Borgomanerese allow interpolation with enclisis, whereas Italian infinitives do not, is challenging to account for if we assume that both contexts display phonological cliticisation (unless perhaps if we assume that clitics target a different projection in Borgomanerese, as proposed by Tortora 2010). The ungrammaticality of interpolation with enclisis in some languages has been raised by Mavrogiorgos (2013), who distinguishes several ways for grammars to derive proclisis and enclisis. Regarding the case under investigation here, Mavrogiorgos’ (2013) hypothesis is that a head c-commanding the cliticisation site contains (or may contain, as we understand it) syntactic features that require to be phonologically realised by a verbal host. Building on his view, we interpret the difference between Borgomanerese finite clauses (enclisis and interpolation) and Italian non-finite clauses (enclisis and compulsory verb-clitic adjacency) as residing in the absence of syntactic features in the verb in the former language. Further empirical investigation is needed to evaluate this claim, however it allows us to maintain the claim that enclisis is essentially a PF-phenomenon.

Both interpolation and enclisis share the characteristic that the clitic does not adjoin to its verb in the syntax, or in other words, the clitic maintains a certain syntactic autonomy by not forming a complex head with its verb. As such, enclisis involves structural proclisis on a functional head, but linear enclisis at Spell-Out. Put differently, since we take cliticisation to be left-adjunction to v, cliticisation (at the syntactic level) is always proclitic. What we see as proclisis and enclisis is a manifestation of linearisation at Spell-Out. The ‘autonomy’ between the clitic and its selecting verb is also found in the case of CC, as the clitic does not adjoin to the infinitive (although it does so with the higher verb, a matter we address in section 7). Consequently, we conclude that the ability of the clitic to ‘climb away’ from the infinitive into the main clause is facilitated by phonological cliticisation. We further explore this concept and apply it to our data in the remainder of the paper. Before doing so, we discuss our theoretical assumptions with regards to restructuring to frame our analysis.

3.2 Restructuring clauses

As we mentioned in the introduction, CC occurs when the domain of the upper verb (e.g. a modal or aspectual verb) and the domain of the infinitival verb are united, or form a unique domain (Rizzi 1982 et seq.). The architecture we assume for restructuring clauses is given in (10) and adapted from Roberts (2010: 84). It is monoclausal and clitics (φ-heads, cf. Dechaine & Wiltschko 2002) are generated in the complement of the infinitive before cliticising on the appropriate v-head at Spell-Out (Solà 2002; Mavrogiorgos 2010; Gallego 2016).⁹ The restructuring verb is the projection of a functional head, more precisely a v-head that does not select for internal and external arguments (Cinque 2001; 2004; Wurmbrand 2004; 2006; 2016; Roberts 2010).¹⁰ As such the main verb really looks like a light verb/auxiliary.

Further, we adopt the view that cliticisation is an operation that values interpretation via Agree (Mavrogiorgos 2010; Roberts 2010). In restructuring clauses, the higher v-head bears a set of unvalued φ-features and acts as a probe for the clitic. Upon Agree, the clitic appears on the restructuring verb (11): this is CC.

This approach accounts for the apparent ‘long movement’ of the clitic (i.e. from the lower to the upper domain) without the requirement that the clitic move ‘through’ an intermediate head (pace Kayne 1989; Martineau 1990). In systems where CC is obligatory, the higher v-head necessarily bears a set of unvalued φ-features that probes the clitic. Once the features of the higher v-head are interpreted, the clitic is realised as left-adjoined on the restructuring verb.

A system that does not allow CC is the mirror image of (11), as the higher v-head cannot probe the clitic. Upon Agree, the clitic is realised in the lower domain and remains there, as is normally found in Modern French (12).¹¹

Between the late 13^th and the early 16^th century, CC is found in 100% of all restructuring clauses in our corpus. This indicates that cliticisation was systematically as in (11). This obligatoriness is attested in other Romance languages during the same period (Pescarini 2021),¹² and a handful of modern languages,¹³ therefore this proposal can easily be extended to the analysis of CC in other varieties.¹⁴

4.1 Stage 1: 12^th–16^th century

Our data show that clitics necessarily target the higher v-head in early French restructuring clauses, until the 16^th century (Table 4). Based on our theoretical proposal, the higher v-head is associated with φ-features valued by the clitic.

This first stage is characterised by obligatory climbing: the clitic forms a complex head with the higher v, namely the restructuring predicate. This latter point is further supported by the absence of interpolation in this context (which we revisit in section 5.1) and the ability of the complex [clitic+verb] to move further upward to T.

4.2 Stage 2: late 16^th to early 19^th century

Shortly before the the 17^th century, the higher v-head became optionally associated with φ-features, thus giving rise to the non-climbing option (as proposed by Roberts 2010 for Italian). In this new grammar, where the clitic can optionally be probed by the upper v-head, CC is optional – when the clitic does not climb, we find proclisis (Table 5). This is evidenced below with data from Bérault, a text which alternates between the two orderings: whilst the higher v-head can probe the clitic in (14), it does not have φ-features in (15).

Stage 2 is the result of a shift in the grammar that occurred across the Romance family. Most contemporary Romance varieties remained at stage 2, namely Italian, Catalan, Spanish, European Portuguese and Occitan – some of these varieties have enclisis instead of proclisis, yet we believe that this connects to infinitive placement rather than clitic placement, as argued by Kayne (1991), Roberts (2010) and Olivier (2022a; 2023).

4.3 Stage 3: early 19^th century to present day

By the early 19^th century, we do not find any instance of CC anymore (Table 6). We take this as evidence for a second shift in French, where the higher v-head entirely loses its ability to probe the clitic. From then on, proclisis on the infinitive is the only option.

This stage concludes the diachrony of clitic placement in restructuring clauses. In summary, throughout the last millennium, we observe that clitics have got closer to the verb that selects them. At one end of the evolution, clitics are systematically realised high on the structure (this is CC), whereas at the other end they are necessarily proclitic on the infinitive, in a low position.

4.4 Interim summary

Our investigation reframes the development of CC in French as a series of two shifts (Table 7). The shift from Stage 1 to Stage 2 is traditionally identified in the literature as marking the loss of CC (Iglesias 2015; Amatuzzi et al. 2020), but based on synchronous cross-linguistic evidence we argue that this shift has taken place uniformly across most Romance varieties. Based on data available for Stage 2, we have shown that both CC and proclisis are available in restructuring clauses, and that the percentages presented do not foresee the ultimate loss of CC. Unlike other Romance languages however, with the exception of Brazilian Portuguese (Davies 1996), French underwent a second shift in ca. 1800 which resulted in the loss of CC.

If our interpretation of the data is on the right track, CC was optional in French similarly to what is found in Spanish, European Portuguese, Catalan and Italian, albeit for a shorter period.

5.1 Evidence for the functional nature of restructuring verbs

We now turn to the nature of restructuring verbs, which we have taken to be functional following Cinque (2004) and Roberts (2010). In particular, we propose that clitic placement in this context can be used to diagnose the functional nature of restructuring verbs. Our rationale goes as follows: if restructuring verbs are v-heads, and that this same v-head can probe clitics, then interpolation (that is, the intervention of a phrase between the clitic and the verb) should not be possible.

Our dataset does not contain any occurrence of interpolation where CC has taken place, that is we did not find the order [φ+XP+v+Vinf]. Martineau (1990) cites one Old French example from Moignet (1976) where interpolation appears to occur despite CC (17). The intervening element is an adverb of intensification, which she claims is not a genuine case of interpolation but rather a clitic cluster, a hypothesis based on an analogous phenomenon in Romanian where the adverb mai ‘anymore’ is a clitic (see also Hinzelin 2010).

In finite non-restructuring clauses, however, a few instances of interpolation are reported in the literature (see Moignet 1976 in particular), mainly for the Old French period.

Whether these examples show interpolation is debated for several reasons: first, this construction is extremely rare and de Kok (1985) does not provide any example in her comprehensive documentation. Secondly, it is not clear whether the intervening element is a phrase since they have been analysed as clitics by Martineau (1990) and prefixes by Skårup (1975). Lastly, Hinzelin (2010) argues that apparent cases of interpolation in this context are scribal errors. Martineau (1990) attests rare instances in Middle French with the negation ne with finite lexical verbs, which she also attributes to scribal errors.

To conclude our observation, Old French is a language that never allows interpolation with restructuring verbs. If our analysis is on the right track, this is because interpolation is impossible with functional verbs as they are the locus of cliticisation – whereas clitic clusters with adverbial clitics should be possible. We interpret the absence of interpolation with finite lexical verbs as evidence for syntactic cliticisation, that is the clitic and the finite lexical verb necessarily form a complex head in Old French. In section 7, we will show that this was not the case with infinitives. Before doing so, we discuss clitic placement in V2 clauses in light of our claim that CC to restructuring predicates yields syntactic cliticisation.

5.2 Clitic climbing in V2 clauses

V2 clauses support the claim that the clitic incorporates with the restructuring verb early in the derivation. In Old French, the finite verb moves to the CP-layer once its features are valued (Adams 1987; 1989; Roberts 1993; Vance 1997; Labelle & Hirschbühler 2005; Labelle 2007; Ledgeway 2007; 2021; Wolfe 2016a; b; 2018; 2019; 2021a; b; Mathieu 2013; Holmberg 2015; Klævik-Pettersen 2018; 2019). Consider examples (20) to (22) where CC has taken place and the restructuring verb is in second position. Crucially, the inflected verb precedes the subject – the initial constituent is in brackets.

In these three examples, an XP is fronted in Spec,CP and both the clitic and the modal verb have moved to C. We argue that the clitic climbs to the modal first and they then move together, primarily because the clitic syntactically cliticises onto the modal verb before the two elements move as one head to C. The structure of (22) is given in (23).

In light of the evidence introduced above, CC is necessarily a phenomenon of left-adjunction to a functional head that Spells Out as a modal, an aspectual or a motion verb. Crucially, once the clitic is in the domain of the higher functional verb, cliticisation is necessarily syntactic.

5.3 Evidence for a lower vP

Upon the assumption that restructuring verbs are functional heads merged in the vp-domain,¹⁵ Cinque (2004) argues that they lack argument structure and therefore do not select for an external argument (nor an internal one, but this is not relevant to us). Thus, the subject of the restructuring clause must be generated in the lower domain. On the cross-linguistic level, there is empirical evidence that (at least some) languages project an embedded vP in such clauses (see Wurmbrand 2016 on German and Austronesian). Our dataset allows us to support Cinque’s (2004) view, specifically the existence of a lower v-head in restructuring clauses.

Consider enclisis in examples (24) in Old French and the equivalent Italian example in (25). These are restructuring clauses without CC.

Infinitives leave the vP domain in French until ca. 1300 (Olivier 2022a; 2023), a date which marks the shift from enclisis to proclisis in the grammar. Recall that we formally analyse cliticisation as φ-Agree with v. Following Kayne’s (1991) hypothesis, the infinitives in (24) and (25) target a Functional Projection higher than the clitic, but structurally lower than the restructuring predicate. In languages where the clitic can target the lower v and is enclitic (Old French, Italian, Spanish, Catalan), we propose that a light Functional Projection (FP)¹⁶ directly above the lower vP triggers V-movement (26).

Consequently, although there are no clausal projections in restructuring environments, there is an embedded v-head which (i) assigns thematic role to the subject, and (ii) probes clitics.

6.1 Ad/de-complementizers

Romance infinitives can be introduced by a series of complementizers (sometimes called subordinators in the literature) which derive from Latin prepositions ad/de (Ledgeway 2012). These complementizers have been formally analysed as Fin-heads by Rizzi (1997). Under a monoclausal analysis, they should not be expected in restructuring clauses where CC is found, since we have seen that the structure lacks (at least) a CP-layer.¹⁷

We found 9 occurrences of CC over ad (7 in Old French, 2 in Middle French), and 3 occurrences with de (only in Middle French), spread from the 13^th to the 16^th century. Similar examples are found in Italian and Catalan.¹⁸ This construction is admittedly rare in our corpus, yet its presence requires further attention as it bears important consequences on any formal model of restructuring.

In sentences (27) to (30), the clitic is the syntactic object of the infinitive yet it cliticises on the finite verb, therefore these are clear cases of CC ‘over’ a complementizer. de Kok (1985: 224), Martineau (1990: 143) and Pearce (1990: 229) have reported similar examples in their corpora, and they also attest of the rarity of the construction. Martineau (1990) investigates two texts, from 1462 and 1505–1515, and finds 57 cases of CC with ad (for 107 infinitives introduced by ad) and 9 instances with de (for 509 infinitives introduced by de), which is higher than what we report. Although the construction is more frequent in her corpus, she also notices the scarcity of CC over de compared to ad. This difference in frequency can be explained through the diachrony of ad/de-complementizers: ad is found in high frequencies until the 16^th century, before a shift took place and verbs started selecting de over ad (Martineau & Motapanyane 2000). Thus the rarity of de within the grammar naturally reduces chances to attest CC with it for the older periods.

Crucially the complementizers under scrutiny are always adjacent to the infinitive they introduce, and often in our corpus they are written as one word (e.g. arendre instead of a rendre). This adjacency is not found with genuine C-complementizers, which allow adverbs, shifted objects and auxiliaries to intervene (Martineau & Motapanyane 2000). Although there lacks a systematic empirical analysis of the distribution and behaviour of ad/de-complementizers in the diachrony of French,¹⁹ independent accounts have noted their affixal properties (Martineau & Motapanyane 2000), which may indicate that they trigger V-movement with infinitives (Kayne 1999; Olivier 2022a). Martineau (1990) also claims that such complementizers have an inflectional flavour, and she therefore analyses them as I-heads, similar to English to. We follow Kayne (1999) and Olivier (2022a) in assuming that ad/de-complementizers are located in a projection directly above vP.

Supporting evidence for the presence of a functional head hosting ad/de directly above vP is found in cases where the clitic does not climb but instead spells out enclisis: in (31) the ad-complementizer is in a Functional Projection above vP, where the clitic is realised. Given that these complementizers attract the infinitive outside v/VP, they necessarily pattern with enclisis. Note that enclisis was the standard ordering in non-restructuring environments until the early 14^th century: according to Olivier (2022a; 2023), infinitives target the IP-layer (similarly to infinitives in modern Italian, Spanish and Catalan, see Schifano 2018). We expand on this claim in arguing that movement of the infinitive outside vP is also found in restructuring clauses where a Functional head hosting a complementizer is present.

Our proposal therefore is that ad/de-complementizers obligatorily trigger V-movement and are affixes (or perhaps clitics) on the infinitive, which explains why we never find proclisis in this context. This is supported by Martineau & Motapanyane (2000), who find evidence of two series of homophonous ad/de-complementizers until the 16^th century, namely F-complementizers and C-complementizers. The former are found in restructuring clauses and are eventually reanalysed as C-complementizers, with which proclisis is found.²⁰

To summarise these observations, we attest 4 different configurations in the diachrony of French (Table 8): in pre-1300 French, the clitic can Agree with either v-heads in restructuring clauses,²¹ and between the 14^th and the 16^th century the higher v-head always had unvalued φ-features in this environment.²² After the 16^th century, clitics cannot Agree with the higher v-head in a presence of ad/de, which have been reanalysed as genuine C-complementizers: from this moment on, these clauses do not have a restructuring make-up anymore. Independent evidence for this comes from the fact that clitics keep climbing with modal verbs for two further centuries after being lost from clauses with ad/de.

In (33) the clitic is probed by the higher v-head and the subordinator ad/de attracts the infinitive, resulting in CC:

In (34) the clitic is probed by the lower v-head and the subordinator ad/de attracts the infinitive, resulting in enclisis:

Where ad/de is merged as a C-head, the environment is biclausal (35). As said earlier, the infinitive targets a high position in Old French and Spells Out enclisis. Unless an element intervenes between ad/de and the infinitive, it is virtually impossible to differentiate (34b) from (35) at Spell Out. Our corpus does not contain any example where an adverb or a shifted object appears, which should nonetheless have been possible (Martineau & Motapanyane 2000).

From ca. 1300, infinitive movement out of v/VP is lost and the grammar has proclisis (see Olivier 2023 for a discussion). The only difference between (35) and (36b) is the lack of high V-movement in the latter.

In conclusion we argue that in clauses where CC is found over ad/de-complementizers, the latter occupy a light functional head directly above vP and trigger V-movement, which allows us to maintain a monoclausal analysis of restructuring.

6.2 Intervening Wh-phrases

Cases of CC over a moved Wh-phrase pose a second potential challenge to the monoclausal analysis of restructuring. Because Wh-elements are standardly assumed to move to the specifier of CP in Romance, Kayne (1989) argues that the construction in (37) is a strong indication of biclausality. His hypothesis takes CC to be head-movement through C, which Martineau (1990) applied to Middle French. The following two examples are drawn from our corpus.²³

This construction has been discussed for Italian (39), and the literature contains an additional example for Catalan (40).

We found only two examples in our corpus of CC over a Wh-phrase, but Martineau (1990: 110–111) reports more instances for Middle French. This construction is subject to several restrictions that have been reported for Italian (Cinque 2004; Cardinaletti 2014), and that Medieval French also shows: (a) it is only available with savoir/sapere ‘know’, similar Italian verbs like chiedersi and domandarsi ‘wonder’ yield ungrammatical results, (b) the main verb must be negated for the Wh-phrase to be preverbal, (c) the Wh-phrase must be adjacent to the infinitive, and (d) restrictions apply as to which Wh-phrase can be found in this construction: ‘when’ and ‘why’ are not accepted.

Based on this series of restrictions, Cardinaletti (2014) claims that this construction is not a case of Wh-movement analogous to what is found in questions, where such restrictions do not normally apply. Applying a cartographic method, she introduces evidence that the Wh-item follows all low adverbs, indicating clause-internal movement to a low position in the IP, directly above the VP. Furthermore, she claims that this position can only be filled by weak elements (in the sense of Cardinaletti & Starke 1999), which explains why only a subset of Wh-phrases are found in this construction: ‘when’ and ‘why’ are bisyllabic and strong in Italian (quando and perché). Lastly, she connects the necessary presence of the negation by the fact that weak Wh-phrases are indefinite, and as such she analyses them as polarity items that must be licenced by the sentential negation non – see Paradís (2019: 299–302) for an extensive discussion. Our data are similar to the ones she discusses, lending independent support for Cardinaletti’s (2014) perspective. Given that, as we previously saw, restructuring clauses do not project an embedded IP, the Wh-element in French must target a different position. Chomsky (2001) proposes that vP, being a weak phase, can serve as a landing site for XP-movement. Building on this claim, Nunes (2021) demonstrates that this applies to Wh-movement with ‘wonder’ in English.²⁴ Thus, given that Wh-elements can move to the edge of v, we propose that examples (37) and (38) offer additional evidence for the existence of an embedded vP introducing the infinitive as discussed in section 5.3 above.²⁵

7.1 Pre-19^th century: phonological cliticisation

We attest the ultimate loss of CC in our corpus at the turn of the 19^th century, which we argued is evidence that the higher v-head cannot probe the clitic anymore. Let us consider the diachrony of clitic placement in non-restructuring infinitival clauses in French. In clauses where the infinitive is not the complement of a functional verb, proclisis is the standard order since the 14^th century (de Kok 1985; Olivier 2022b) – and before then enclisis is found. In this environment, the embedded v probes the clitic (41).

Let’s consider verb placement. French infinitives can optionally move above ‘low’ adverbs, for instance with souvent ‘often’ (Pollock 1989). We recast this observation as optional movement to v where the infinitive precedes the adverb (42).

In addition, (42) shows that adverbs can potentially intervene between the clitic (in v) and the infinitive (in V). Unlike what we have seen earlier in finite clauses, proclisis was not necessarily adjacent on the infinitive in pre-19^th century French, and adverbs are found to intervene (44). In addition, negation (45) and quantifiers (46) can also occur in this position (de Kok 1985: 337; Martineau 1990: 79; Hirschbühler & Labelle 1994).

As we saw earlier, interpolation of this kind has been documented across several Romance varieties (Pollock 2002; Miller & Monachesi 2003; Poletto & Pollock 2005; Ledgeway & Lombardi 2005), and we argued that it can serve as a diagnostic for phonological cliticisation: once the clitic is realised on v, it cliticises at Spell-Out on the nearest prosodic word. In other words, the clitic is ‘blind’ with regards to the element it cliticises on as long as it is one that can act as a prosodic host. Under this assumption, languages that have interpolation necessarily have phonological cliticisation (i.e. the clitic does not needs to ‘fuse’ with its prosodic host in the syntax). Contrarily, the mechanism is necessarily syntactic in languages where the clitic is obligatorily adjacent and proclitic on the verb – specifically because they form a complex head in narrow syntax. In such varieties (e.g. Modern French), the clitic attaches to the verb before Spell-Out and is akin to head-movement that is found in narrow syntax.

The interaction of optional Vinf-to-v movement and the possible presence of low adverbs generates four combinations that were synchronically available in Middle and Early Modern French (47). This optional movement further connects to the absence of a requirement of the clitic and the infinitive to form a complex head.

In (47a), the infinitive remains within the VP whereas it targets v in (47b). A difference is visible at Spell-Out once an adverb is introduced: interpolation is found when the infinitive remains low (47c). Out of the four combinations that were synchronically available in Middle French, three had the clitic adjacent to the infinitive (47a), (47b) and (47d). Furthermore, it is reported in the literature that interpolation is available with a subset of adverbs only (Hirschbühler & Labelle 1994; Pollock 2002; Miller & Monachesi 2003; Poletto & Pollock 2005; Ledgeway & Lombardi 2005), namely those that sit low in a cartographic hierarchy, and that it is not a frequent construction. We make the same observation, in our corpus we found only 13 examples scattered between the 15^th and the 19^th century.²⁶

Interpolation with infinitives is arguably not common in the diachrony of French, but it is not a rare phenomenon either. We argue that it was available in the grammar exactly because cliticisation was phonological. Importantly, we demonstrated earlier that cliticisation was syntactic in finite contexts, therefore the mechanism of cliticisation must be domain-dependent: until the 19^th century, finite and non-finite domains did not show the same type of cliticisation.

Turning to CC, we hinted towards the fact that the absence of interpolation with restructuring predicates indicated that cliticisation is syntactic in this context: we are now in a position to revisit this claim. We propose that CC was available as long as cliticisation was phonological in non-finite domains, which allowed the clitic not to form a complex head with its verb – put differently, clitics and infinitives were independent from each other, therefore the clitic could reach the upper domain as long as the higher v-head could probe it. Once the clitic climbed to the finite domain, and for the reasons outlined above, syntactic cliticisation yields. Precisely, our data show that interpolation and CC were lost at the same time, giving rise to systematically Vinf-adjacent proclisis.

7.2 Post-19^th century: syntactic cliticisation

From the 19^th century on, French shows systematically verb-adjacent proclisis, which we interpret as a new grammar showing syntactic cliticisation in non-finite domains.

We conclude that the loss of CC stems from the reanalysis of cliticisation from phonological to syntactic. We can hypothesise that what prompted this change to verb-adjacent proclisis is the fact that this order was overall more prominent in the grammar since it is always found with finite verbs, and it is the non-climbing option with infinitives (which as we have seen, is roughly 50% of the time in late Middle French). As such, (Middle) French differs from most Romance varieties, in that proclisis is frequent in all domains. Under the assumption that change spreads from main/finite to embedded/non-finite clauses (if we follow Lightfoot’s 1993 proposal that children are Degree-0 learners and utilise input mostly from main clauses, and therefore constructions that are found exclusively in embedded clauses are more susceptible to change and/or loss) syntactic cliticisation generalised from finite clauses to infinitives, where verb-adjacent proclisis was already frequent enough (in non-restructuring clauses, and as the non-climbing option).²⁷ The diachrony of Brazilian Portuguese would be an ideal testing ground for this hypothesis, since it has lost CC and has proclisis on infinitives in some dialects (Davies 1996). It would be interesting to pursue this more systematic comparison between the two languages but this is something we have to leave open for future research.