Word order in Early Old Catalan

1 Introduction

The clausal syntax of Catalan before the 13th century has never been explored in the literature concerned with the history of this language since literary texts have traditionally been the pool of data explored in studies concerned with diachronic Catalan syntax (Badia i Margarit 1951; Batllori 1993; Batllori et al. 2005; Batllori and Hernanz 2009, 2015; Fischer and Xavier in press; Par 1923; Pujol i Campeny 2025a; Pujol i Campeny, Afra 2018), and literary texts in Catalan language are not attested until the 13th century. In this article we examine the sentential syntax and information structure of four texts dating from the 10th and 11th centuries (to which we refer as Early Old Catalan or EOC) to establish that: (i) Early Old Catalan patterns with contemporary varieties in presenting low V2 syntax; (ii) it shares with other Old Romance varieties the use of expletive elements to satisfy the V2 parameter; (iii) Early Old Catalan differs notably from 13th century Catalan in the use of the left periphery and the distribution of V2, and the use of expletive elements. Three of the texts examined are written in Latin, and either display non-canonical Latin word order or syntactic glosses that inform us about the syntax of the Catalan of the period under study. The remaining text is written entirely in Catalan and offers a glimpse of the language of the turn of the 11th century. In incorporating texts in Latin, this article echoes Troberg and Whitman (2022) in defending the usefulness of Latin texts of early Romance provenance as a source of data for historical syntax research. These authors specifically engage with texts with syntactic glosses, but here we also engage with texts displaying non-canonical Latin word order, widening the pool of sources used to inform us about the syntax of Early Romance.

By maximising the evidence available to us, this article fills a gap in the literature on Old Catalan syntax and provides the broader field of Romance linguistics with a nuanced account of the clausal syntax of Early Old Catalan in order to consolidate our understanding of the evolution of the Romance languages in their early days. By comparing the robust low V2 syntax of Early Old Catalan to that of 13th century Catalan, where V2 only surfaces in archaising pockets such as embedded clauses (Pujol i Campeny 2025a, in press), contrasting with main declarative clauses' unmarked SVO, we are able to establish a syntactic periodisation of the language. This allows us to position Catalan diachronically in the Romance linguistic continuum and to reassess its syntactic fragmentation after the joint development of V2 by exploring how, when, and where V2 started to weaken within the continuum, which, by the 13th century, might be more appropriately referred to as linkage (François 2014; Ross 1988).

For the analysis of Old Catalan word order within the wider Romance context, we assume a generativist definition of V2 and a cartographic conception of the left periphery (Benincà 2006; Rizzi 1997). From this perspective, V2 is understood not as a linear phenomenon but as a syntactic requirement that involves two separate operations:

A language can be considered to be V2 when it exhibits the features highlighted in (1) in unmarked declarative main clauses (Cruschina and Sailor 2022; Roberts 1993; Sailor 2020). V2 might be present elsewhere, as is the case in most Modern Romance varieties, where verb movement to the left periphery is attested in constructions associated to marked informational readings (Cruschina and Sailor 2022; Leonetti and Escandell-Vidal 2009; Pujol i Campeny 2025a; Sailor 2020), but that does not render these varieties structurally V2.

Within the cartographic programme, the left periphery corresponds to the syntactic space located above the Inflectional Phrase. It acts as a hinge between the clause and the wider discourse, and it is further divided into the following projections and fields:

At the extremes of the left periphery, we find ForceP, a projection associated with a clause’s illocutionary force, and FinP, a projection connected to the clause’s finiteness and mood. Between ForceP and FinP, the Top(ic)P and Foc(us)P fields host constituents receiving different types of topical and focal readings. In the case of topics, we take them to be generally base-generated in the left periphery (Benincà 2004, 2006; Benincà and Poletto 2004; Cinque 1990; De Cat 2009), while constituents occurring in the Focus field undergo movement to the projection connected with their reading. Finally, above ForceP there is the Frame field, a scene-setting space that hosts constituents anchoring the proposition contained in the clause in the wider discourse in terms of speech participants and spatio-temporal deixis (Benincà and Poletto 2004).

Depending on the landing site targeted by the verb in the left periphery (Wolfe 2018b; Wolfe 2020, et seq.), two types of V2 grammars can be distinguished. On the one hand, in relaxed or low V2 languages the verb targets FinP. This allows constituents to be first-merged in the Topic and Frame fields, preceding the verb and the constituent moved to the specifier of Fin^o. Therefore, in low V2 languages, V3+ orders, where more than constituent precedes the verb, are licensed, as illustrated in (3), a 12th century Old French example:

In high or strict V2 grammars, the verb raises to ForceP, where it is only preceded by the element merged in the specifier of the projection. V3 orders are only ever licensed when adverbs are first merged in the Frame field:^[1]

The presence of an active V2 grammar has been shown to correlate with the following syntactic phenomena (Klævik-Pettersen 2019; Poletto 2014; 2019; Wolfe 2018a):

It is key to highlight that the only unequivocal cue for the acquisition of a V2 grammar are incontestable cases of inversion, as described in (5.i) (Lightfoot 1999; Westergaard 2009; Yang 2000). The rest of the phenomena listed here correlate with the presence of an active V2 grammar but can be parsed under non-V2 grammars. In the light of this, this paper focuses on inversion structures and the nature of the preverbal field to determine whether the data at hand would allow for the acquisition of a V2 grammar or not. Additionally, other aspects connected to the active presence of a V2 grammar are also considered, albeit in less detail due to space constraints: the possibility of encoding information focus in the left periphery, as a symptom of an underspecified preverbal field; clitic distribution obeying Tobler–Mussafia law whereby clitics occur enclitically whenever the verb occurs in clause-initial position (Ledgeway 2010);^[2] and the presence of expletives, which have been identified for Old Italian (Poletto 2005, 2019), Old French (Wolfe 2018b), and Old Occitan (Donaldson 2016; Pujol i Campeny 2025c). Each of these traits is explored at length where relevant in the discussion.

We cannot undertake this analysis of EOC clausal syntax without first considering the extra-linguistic variables that shape the limited pool of data available to us.

2 Syntax, register, genre, and the data at hand

The so-called ‘bad data problem’ (Weinreich et al. 1968) lays out one of the central issues that we face in diachronic linguistics: language change occurs in spoken language, and we have to rely on written data (where available) in order to reconstruct previous stages of a language. To bypass this problem, it has been suggested that researchers should focus on written documents that bring us closer to spoken language, such as familial correspondence, trial proceedings, and dialogues from prose drama (Culpeper and Kytö 2010; Elspaß et al. 2007; Kytö 2016).

However, for certain periods and varieties, as is the case for 10th-11th century Old Catalan, we only count with scant written traces of the language of the time, occurring exclusively in legal texts, which contrast heavily with oral-like texts in terms of conventionalisation and formalisation of their register and style. While in diachronic Romance linguistics there was scepticism regarding language variation connected to medium, register, and genre in early periods when vernacular Romance was not yet standardised nor ‘established’ (Posner 1997), recent studies have clearly shown that the pre-modern textual record does reflect written-medium tendencies (Ingham 2016: 383; Pujol i Campeny 2025a, in press) and is subject to register and genre constraints that intensely interact with the syntax of texts and should be taken into account.

In this article, we qualitatively examine four texts that can be described as legal given their format and content. They fall in two categories: three legal documents of Catalan provenance written in Latin, and one code of law, written in EOC. The three Latin texts of Catalan provenance were identified on the basis of philological analyses. Two of the texts, written in the 10th century, exhibit non-canonical Latin syntax and have been described as ‘oral-like’, ‘colloquial’, and narrative in nature (Alturo i Perucho and Alaix 2022, 2023). Therefore, despite their legal nature, these texts contrast with the bulk of written evidence of the period, which tends to display formulaic and relatively correct Latin syntax. The third text, the Liber Iudicorum (11th century), contains extensive syntactic glosses (Bellés i Sallent and Alturo i Perucho 2008) akin to those analysed in Troberg and Whitman (2022) in a manuscript of Occitan provenance. The fourth text, the Usatges de Barcelona, consists on a complete legal code written entirely in Catalan and compiled during the 11th century. This text lends itself to the study of clausal syntax thanks to its length and style.

In recent years, two other texts written in Old Catalan have been re-dated as having been produced in the 11th century and the turn of the 12th: the Greuges de Guitard Isarn and a fragment of a translation of the Liber Iudicum from la Seu d’Urgell. The Greuges de Guitard Isarn or Grievences of Guitard Isarn, newly dated as produced in 1105 by Alturo i Perucho and Alaix (2023), presents very particular syntax, with a prevalence of pro-drop V1 clauses headed by the verb rancurar-se ‘to complain’ and the phrase ésser rancurós ‘to be greived’, as Guitard Isarn complains about the offences committed against him by Guillem Arnau and his son Mir Arnau.^[3] The fragment of the Liber Iudicum is characterised by the use of concatenated embedded clauses and the lack of canonical and complete declarative clauses. Therefore, neither text serves the objective of this article. Finally, feudal oaths have also been excluded from the article since, despite exhibiting Catalan lexis, it is simply inserted within a formulaic Latin syntactic frame (Russel-Gebbett 1965). Finally, the last potential candidate to appear in this study would be the Cançó de Santa Fe, an 11th century poem. Two reasons have brought us to dismiss it: on the one hand, meter requirements heavily interact with word order, rendering verse an unreliable source of information. On the other, its linguistic filiation (whether the text instantiates early Catalan or early Occitan) remains contested to this day (Lafont 1998; Riquer 1964: 197–300).

The first text examined here is a 10th century diploma written by a certain Presbyter Borrell in Vic on April 5th 933 (Vic, ABEV, AC, cal. 6, núm. 152). Diplomas were legal documents emitted by someone with legal authority (the precise title and function of the person emitting them depended on the geographical location)^[4] where an event, agreement, or testimony was presented as a true fact. The text is studied from a philological perspective in Alturo i Perucho and Alaix (2021: 207), where they describe its language as follows:

We then examine two documents produced by Judge Bonsom of Barcelona, active during the last quarter of the 10th century and the first quarter of the 11th century. The first document, written in the vicinity of Barcelona in 988, consists of a diploma of narrative nature telling the story of two brothers and how they came to inherit the property they owned. Alturo i Perucho and Alaix (2022) suppose that Bonsom used notes made by someone else when composing the text, given the presence of a few errors attributed to the copying process. Bonsom’s Latin is described as grammatically correct but not spotless.

The second document produced by Bonsom analysed here is his copy of the Liber Iudicorum, a 7th century code of law containing an admixture of Visigothic and Roman law, as well as laws deriving from Conciliar activity (Alturo i Perucho 2003; Bellés i Sallent and Alturo i Perucho 2008; Collins 2004). The code was applied in the Iberian Peninsula from the fall of the Roman empire, throughout the Arab invasion, and until the High Middle Ages (Bastardas 1984). By the 11th century, it was slowly being replaced by feudal costums. Bonsom’s copy, the Liber Iudiciorum Popularis, accompanies the core legal text with a variety of additional content, including a eulogy of St Isidore of Seville, a chronicon of Frankish kings, a calendar of religious festivities celebrated in Barcelona and the surrounding area, rendering the copy a pedagogical device in addition to a legal one. We focus on two fragments of the Liber, a metaphrase, found in the opening of the first book (fol. 12v-13r, 20 lines, 155 words), and the syntactic glosses of a fragment concerned with inheritance law (fol. 103v-105r, 65 lines, 728 words).

Finally, we discuss the syntax of the Usatges de Barcelona (the Customs of Barcelona). While in the opening of the text its composition is attributed to Ramon Berenguer I and his wife Almodis of the Marche in 1068, the first compilation of the text that has reached us is dated around 1179. Costums were added to the text until the 13th century, when it was fixed and canonised under the kingship of James I (Martí i Castell 2002). Therefore, the dating of this text is to be taken with caution, but as shown below, it exhibits archaic features that set it apart from 13th century Catalan and would place it in an earlier period. We follow Alturo i Perucho and Alaix’s (2023: 191) dating of the text, grounded on philological information.

In what follows, we analyse the syntax of each of these texts to see a picture of the sentential syntax of EOC emerge clearly: it had an active low V2 grammar, patterning with the results of research on the syntax of other Early Old Romance varieties,^[6] but contrasting with 13th century Catalan, as described in Pujol i Campeny (2018, 2025a, in press). The steps followed for the analysis of each text depend on its length. Both Borrell’s diploma and Bonsom’s diploma are short texts of less than 1,000 words (300 for the former and 728 for the latter). Their syntax is analysed exclusively qualitatively, with frequency being solely taken as a potential indicator of clausal structure, and not a reflection of the underlying syntax. Texts are closely analysed to establish the presence or absence of the features listed in (5) for the identification of V2. Bonsom’s metaphrase and syntactic gloss lend themselves to a more nuanced analysis. Patterns of change from the initial Latin word order to the paraphrased or glossed word order are explored to determine whether the grammar reflected in the metaphrase exhibits features aligned with those of V2 or not. Finally, in the analysis of the Usatges, we explore the clausal syntax of 400 clauses contained in the first 10 folia of the text, belonging to the 11th century core. In this case, frequency of word order patterns is explored, but simply as an indicator of what the underlying syntax might be. The Usatges are then qualitatively analysed in order to identify the presence (or lack thereof) of structures that correlate with an active V2 grammar. Given their length and the fact that it is the first text written entirely in Catalan of its extension, it affords the possibility of a more detailed discussion clausal syntax and the conclusions reached in this paper pivot mainly on evidence from this text.

3 The 10th century: Judge Borrell’s diploma

Here, we examine the sentential word order of Judge Borrell’s diploma. Given the length of the text, evidence from this the text can only hint at the operating grammar of the time through the analysis of word order tendencies, clitic placement, and the use of expletives.

There is a clear preference for (S)VO orders in both main (8/12) and embedded clauses (12/14), contrasting with Classical Latin (Wanner 1987: 380):

Medieval Romance clitic pronouns abide by the so-called Tobler-Mussafia Law, which can be loosely formulated as the descriptive observation stating that object clitics not being able to appear in clause initial position. In Borrell’s diploma, the distribution of Latin demonstratives is akin to that of Medieval Romance clitics, consistently appearing proclitically when a suitable phonological host is available preverbally:

In (9), the demonstratives eum and illi appear proclitically, as expected given that the adverb sic ‘thus, indeed’ preceds the verb. The Romance equivalent of sic, sí, sì, or si depending on the orthography, is found in preverbal position in the earliest Langue d’Oil text, the Oaths of Strasbourg (842), and in the Greuges de Guitard Isarn, an early 12th century Catalan text:

In (10), the earliest instance of si recorded in Old French, a low V2 language at the time (Wolfe 2018b) is reproduced. Si appears between a concatenation of scene setters and the verb. In (11), si occurs twice, first, within an embedded adverbial clause introduced by quar/car ‘since’, which can be described as a peripheral adverbial clause, as it does not modify the structure of the event contained in the main clause (Haegeman 2004: 61). Peripheral adverbial clauses are characterised by not being integrated in the core structure of the main clause, and by being inserted in the left periphery. Internally, they present a fully-fledged periphery and possess their own illocutionary force, which translates into them being able to accommodate main clause phenomena (Haegeman 2012: 173; Hooper and Thompson 1973). Therefore, in the case of the quar clause, V2 syntax is expected to apply (Haegeman 2012: 179), and si satisfies the requirement of filling SpecFinP, assuming that the verb raises to Fin^o, in a low V2 configuration.

In the most recent analysis of the particle in Old French, Wolfe (2018b; 2020) proposes that si starts off as a temporal deictic adverb that undergoes movement to the left periphery to satisfy the language’s V2 requirement in the 11th century, and then, grammaticalises as a FinP expletive by the late 12th century. As such, it merges directly in SpecFinP when there would be no constituent available to move to it (Wolfe 2018b: 349).

In examples (9–11), it is not possible to establish whether sic/ sí is an adverb moved to the left periphery or an expletive. According to Wolfe (2018b), Old French si does not reach expletive status until the 12th century. Nevertheless, there are no cases of postverbal si in the Early Catalan textual record,^[7] which would be expected if it were still used as a full adverb like així/axí/aissi ‘thus’,^[8] which does appear postverbally when not fronted for specific informational purposes. This suggests that in the examples above sí being used as an expletive.

Old Occitan si exhibits the same distribution as that shown in examples (8–10). Donaldson (2016) argues that si to fulfils the second requirement of Old Occitan’s low-V2 grammar that by appearing preverbally in sentences where a dislocated subject introduces a new discourse referent, as in (12), or where the verb is preceded by an element occurring in the Frame Field, as in (13):

Old Occitan, in line with the other Medieval Romance varieties (Batllori et al. 2005; Benincà 2006), presents focus sensitive enclisis/proclisis variation (Batllori et al. 2005; Benincà 2006; Troberg and Whitman 2022; Wolfe 2021b). When si or a focalised constituent is present, it triggers proclisis, as in (13). In contrast, when the element satisfying the second step of the V2 requirement is topical in nature and si is not present, enclisis follows, as in (14):

On the basis of clitic placement, Donaldson supposes si to occur in FocP:

Donaldson’s analysis of si in the Vidas matches the distribution that we find in Borrell’s diploma, where sic occurs between the subject ‘ego Nectar presbiter’ (which acts as discourse topic), and the verb, resulting in proclisis. This suggests that the grammar of this text aligns with that of Old Occitan, a low V2 language (Donaldson 2016; 2020; Wolfe 2015a, 2015b: chap. 4).

Summing up, Borrell’s diploma (i) displays a clear preference for (S)VO word orders in both main and embedded clauses, contrasting with Latin (S)OV; (ii) shows Latin demonstrative pronoun distribution patterning with the expected distribution of Romance clitic pronouns; (iii) the distribution of sic patterns with that of contemporary texts in general, and more particularly, with that of 13th century Old Occitan si as analysed by Donaldson (2016), where it is connected with the satisfaction of a low V2 requirement in specific syntactico-pragmatic configurations.

4 The 10th century: Judge Bonsom’s diploma

Composed 55 years later than Borrell’s diploma, Bonsom’s diploma presents the expected genre features (understood in the sense of Bieber and Conrad 2009: chap. 2) of this kind of legal document, but it is set apart by its eminently narrative nature (Alturo i Perucho and Alaix 2022). We first consider word order tendencies, then turn to clitic placement and function, and finally, we examine cases of XVS or subject-verb inversion.

Like Borrell’s diploma, Bonsom’s shows a preference for (S)VO word order, especially in narrative segments, as in (16) and (17). In contrast, the opening, closing and final paragraphs, where legal matters are addressed, (S)OV is dominant due to the use of Latin formulaic expressions.

In this diploma, demonstratives standing in loco Romance clitics also inform us about the syntax of the language of the time, being used in the core of the clause to resume clitic left dislocated elements, as expected in Medieval Romance (Benincà and Poletto 2004; Benincà 2006):

In both (18) and (19), the makeshift clitic, appearing in the guise of a Latin demonstrative, occurs postverbally. Following the classic Tobler–Mussafia law, it would be expected to occur preverbally, as the verb is preceded by a constituent. Nevertheless, if we follow Batllori et al. (2005); Benincà (2006) and Donaldson (2016) analysis of clitic placement in Medieval Romance, proclisis is not connected to the presence of any constituent in preverbal position, but dependant on the presence of a focalised element (be it a constituent or a head) in preverbal position, interacting with the information structure and the prosody of the clause. (18) and (19) abide by Batllori et al. (2005)’s analysis, in that the presence of a topicalised constituent in preverbal position correlates with enclisis instead of proclisis. The clitic distribution attested in Borrell’s diploma instantiates a grammar of Catalan that is already distinctive from that of Oil dialects, where clitic distribution abides by the Tobler–Mussafia law in the classical sense. This is confirmed by cases such as (20), where proclisis is triggered in the presence of a fronted focalised constituent:

(20) also supports the presence of an active low V2 grammar in 10th century Catalan. The temporal clause acts as a scene setter, and as such, it is first-merged in the frame field and cannot check the EPP feature on FinP, assuming the low V2 analysis suggested by the presence of si and the distribution of clitics in Borrell’s diploma. The adjective onerosum ‘burdensome’, functioning as a subject complement, is fronted to the left periphery triggering linear XVS and satisfying the second requirement of low V2. The presence of onerosum ‘burdensome’ preverbally shows that the preverbal space can host information focus, which is not the case in Modern Catalan. However, since the precise position of the subject cannot be determined, as the object clause that follows it is heavy and its low position within the clause could be motivated solely by its weight, the analysis remains speculative.

In brief, Bonsom’s diploma reflects a grammar with a preference for (S)VO orders and informationally sensitive clitic placement aligned with Batllori et al. (2005)’s analysis. It further points at the presence of an active low V2 grammar with an informationally unspecified preverbal field.

5 The 11th century: Judge Bonsom’s Liber Iudiciorum Popularis

As mentioned above, Bonsom’s copy of the Liber Iudicorum was not a mere copy of the laws applying at the time, but a multidimensional pedagogical tool. Two fragments are of special interest for the study of the evolution of the syntax of Catalan (covered in more detail in Pujol i Campeny (2013; in Pujol i Campeny 2025b)): a metaphrase, found in the opening of the first book; and (fol. 12v-13r, 20 lines, 155 words), and the syntactic glosses of a fragment concerned with inheritance law (fol. 103v-105r, 65 lines, 728 words). The metaphrase consists in the reorganisation of the lexical items of the first paragraph of book 1 into EOC syntax. This mechanism for rendering the text accessible to a reader without full mastery of Latin allows for Bonsom to omit or introduce lexical items as required by EOC grammar, while striving to do his best to adhere to the Latin material as much as possible. The syntactic glosses do not afford Bonsom this flexibility. In them, Bonsom labels each individual word with a letter of the alphabet (or dots ‘·’, ‘:’, ‘⁝’, once he runs out of letters), as shown in Figure 1:

Figure 1:

Reproduction of the beginning of the syntactic gloss, (Manuscript Z-II-2, Fol. 103v).

In order to establish the presence or absence of an active V2 grammar, we assess the syntax of the metaphrase (henceforth M) and the glosses (henceforth G) in relation to (i) their sentential word order patterns; (ii) the syntax and information structure of verb-initial clauses; (iii) the distribution and syntax of XV(S) clauses and the nature informational and categorial nature of their preverbal constituent.

Table 1 shows word order changes from the original Latin text to M and G in both main and embedded clauses:

Approximately half of main and embedded clauses undergo reorganisation from Latin (S)OV to (S)VO. Another 20 % of clauses, initially presenting other word order patterns, also surface as (S)VO in M and G; as well as two initially verb initial clauses. Thus, (S)VO makes up for 75 % of the total of clauses contained in M and G. Only 7 % of clauses present XVS orders that could point at cases of inversion. Five of them derive from other word order patterns and are reorganised as XVS, while one of them preserves XVS from the original text.

Examples (21) and (22) contain two cases of XVS:

In (21), the original sentence is reorganised for the adverb certe ‘certanly’ to appear preverbally. The verb expetit ‘demands’ is immediately followed by the subject, ratio ‘reason’, which is in turn followed by the infinitival object clause headed by cognosci ‘to be known’. If we assume the verb to be in FinP, certe then needs to be a manner adverb akin to Old Catalan cert ‘surely, with certitude’, generated in the low adverbial space (Cinque 1999) and moved to SpecFinP to satisfy the second requirement of low V2. From there, it raises to TopP. This supported by the position of the reflexive clitic pronoun se, which, unlike in the original sentence, is enclitic. As we saw in Sections 3 and 4, clitic placement appears to be sensitive to the saturation of FocP, and enclisis occurs when the element that precedes the verb is topical. The fact that se occurs in enclisis can be used as evidence of the topicality of certe.

In (22), the embedded clause headed by cum is reorganised to display XVSOX. vero is maintained as the first element of the clause, following the subordinating conjunction cum. Kroon (1995) describes vero as a subjective modality marker that is used to contrast the content of a proposition against the discourse background. As such, its natural position would be FocP, if available. Cum introduces a peripheral adverbial clause that constitutes a separate illocutionary act from that of the main clause (Haegeman 2012: sec. 4.5). In terms of internal syntax, peripheral adverbial clauses present a fully-fledged left periphery (contrasting with central adverbial clauses, which have a truncated left periphery). Therefore, cum is merged in ForceP, and all projections below this head are available, including FocP. This translates into peripheral adverbial clauses exhibiting Main Clause Phenomena (Hooper and Thompson 1973), amongst which, V2. The choice of vero as the element preceding the verb in the metaphrase could be motivated to the contrastive nature of the adverb, which, after moving to SpecFinP, would have moved to FocP to receive emphasis. The verb is immediately followed by the subject, which is in turn followed by the object. The fact that the subject is followed by the object renders the position of the former ambiguous between SpecIP and SpecVP, as in both cases, the object would follow the subject.

In terms of information structure, informationally new objects are consistently located in postverbal position. The only case in which it could be argued that informationally novel elements are located left-peripherally is (20), where the adverb certe is fronted. However, the subject, informationally old, is followed by an informationally new object. The postverbal subject in (21) is also informationally old. The possibility of having informationally old subjects postverbally highlights the fact that the preverbal field is not specialised for hosting informationally old subjects, as is the case in modern SVO Romance languages, where postverbal subjects tend to be informationally new.

Summing up, the Liber Iudiciorum exhibits cases of inversion consistent with an active V2 grammar, and the distribution of new and old information is compatible with it.

6 The 11th century: Usatges of Barcelona

The extension of this text affords us the chance to obtain a more fine-grained picture of EOC syntax in the 11th century and confirms that EOC had an active low V2 grammar, a fact that was hinted at in the data examined in the previous sections. First, we examine the position of the verb in main and embedded clauses. Then, we move onto the analysis of the information structure of the preverbal field and cases of inversion, and the syntax of V1 clauses. Finally, we consider the distribution of potential expletive elements.

Table 2 shows the frequency of linear verb position in the Usatges:

In both main and embedded clauses, verb-second clauses are the dominant word order pattern. In main clauses, V1 and V3 clauses constitute roughly a quarter of the total. In embedded clauses, V1 clauses are almost as frequent as V2, constituting 42.7 % of the total, while V3 clauses are the least frequently attested word order pattern, occurring only in 13.7 % of cases. Within V1 main clauses, 35 are preceded by et (17 %), while 12 are absolute verb-initial clauses (6 %).

We now turn to the analysis of the preverbal field and cases of inversions. The nature of the preverbal element of V2 main clauses is shown in Table 3:

The most common preverbal element in V2 clauses are protases (36 %). This particularity is connected to the nature of the fragments of the Usatges examined, which present the reader with specific situations and their corresponding legal response in the form of ‘if x, then y’. Protases are followed in frequency by subjects, which constitute 33 % of preverbal constituents. They include noun phrases (28/36), pronouns (4/36), and subject que-clauses (2/36). They are followed in frequency by adjunct PPs, adverbs, and adverbial CPs. Within the total of V2 clauses, 15 present postverbal subjects, and in 11 of these cases, the subject appears adjacent to the verb.

Table 4 shows the different possible combinations of elements in the preverbal field of the Usatges:

It can be observed that adjuncts (adverbial or prepositional), scene setters, and other elements associated to the frame field (clitic left dislocated noun phrases, nominal hanging topics, protases, and dislocated subjects) tend to occur in sentence initial position and can follow and precede each other. This is so because multiple scene setters can be merged in the Frame field. Whenever objects appear preverbally, they are verb adjacent (with one exception, where an adverb intervenes between the object and the verb). Only 5/29 cases of XVS have an overt postverbal subject. Of these, four are verb adjacent, and one is not.

For the purposes of determining whether the Usatges display an active low V2 grammar, we focus on cases of XVS with overt subjects. By examining them closely, it can be established that: (i) the verb raises to the low left periphery across subjects in SpecIP and below elements in the upper left periphery; (ii) the preverbal field is informationally unspecified (it can host both rhematic and thematic constituents), and (iii) there are three subject positions available for Old Catalan subjects: the left periphery, SpecIP, and SpecvP.^[9]

Cases of XVS in the Usatges show that the preverbal field is informationally unspecified: in (23), the preverbal constituent introduces a new referent that has not been discussed before, the village of Les Roques, while in (24) it contains an aboutness topic using Frascarelli and Hinterhölzl (2007: 1) topicality scale. In both cases, the subject surfaces postverbally. Assuming an underlying V2 analysis, this would show that any constituent, regardless of its informational status, can satisfy the V2 requirement.

Supporting that the verb is located in the left periphery are several cases of Germanic-style inversion, where the subject appears between a modal verb and a lexical verb,^[10] as in (25–26):

In (25), inversion is borne with a fronted direct object that can be contrastively interpreted against the background occurring preverbally. The modal finite verb is followed by the pronominal subject hom ‘one’, which is in turn followed by the non-finite form fer ‘to do’ and the indirect object. In (26), the finite modal is preceded by a protasis and followed by a finite, informationally old subject. Like in (25), the non-finite verb is followed by verbal complements, in this case, an indirect object and an adverbial embedded clause that acts as direct object (ordering between the two probably being determined by the DO’s heaviness).

Contrasting with (25–26), in (27) the presence of a preverbal protasis does not trigger inversion. Instead, the subject appears preverbally, and the object follows the verb, in a typical (S)VO pattern.

Finally, (28) contains an example of a postverbal subject occurring after the non-finite verb form, followed by the lexical verb’s complements. Therefore, we can establish that subjects can occur preverbally, as in (27), and postverbally, in two positions: in SpecIP, as illustrated by (25) and (26) and its base-generated position, in SpecvP, as in (28).

All cases of inversion occur with transitive verbs. This is noteworthy because it shows that inversion in Usatges is not aligned with inversion in Modern pro-drop Romance languages, where unaccusative predicates favour postverbal subjects, and discards the possibility of Romance-style inversion, which is predicate and information structure conditioned, with new subjects tending to occur postverbally (Corr 2016; Hulk and Pollock 2001; Pinto 1997; Sheehan 2010; Tortora 2001).

Elements other than the subject can appear between modal auxiliaries and non-finite verbs. Consider (29) and (30):

In (29) and (30), the subject is followed by a verbal complement: a direct object in (29) and an indirect object in (30), where the direct object is dropped. In these two cases, the object has been moved to the middle field, the syntactic space between IP and vP, which according to Belletti (2006), and Poletto 2006, could be understood as a low left periphery containing projections correlating with different information categories. Leaving the analysis of Old Catalan middle field for later work, what is worth noting from these two examples is the respective order between the constituents occurring between the auxiliary and the infinitive: the subject always occurs verb-adjacent and is followed by any other element that might appear between the two verbs. This is so because the subject is not located in the middle field or the lower left periphery, but in SpecIP, and the auxiliary has moved across it to the low left periphery. Were the subject to be located in the middle field, we would expect object–subject sequences to be possible, but they are not attested in the Usatges.^[11]

As shown in (30), the verb can be preceded by more than one constituent. When that is the case, elements preceding the verb always display the same ordering, which aligns with descriptions of the Medieval Romance left periphery:

This is further confirmed by the following V3 cases:

Given that cases of inversion like (25–26) provided unequivocal evidence of V-to-C movement, and that the verb can be preceded by more than one constituent, as in (31–32), it is safe to assume that like Early Old French (until the 13th century, following Wolfe 2020: Ch. 4) and Old Occitan (Donaldson 2016; Wolfe 2018b, 2020) in EOC main clauses, the verb raises to the low left periphery.

Turning now to V1 clauses: they occur in two contexts: (i) at the beginning of a discourse unit, as in (34); (ii) after the coordinating conjunctions e, mas, sinó, o, and ne, all with subject continuity with the preceding clause, as in (35) and (36):

V1 clauses fit a low V2 analysis of Old Catalan if we assume, for verb initial clauses at the beginning of a discourse unit, the existence of a null narrative operator (Wolfe 2015a, 2015b: 157), and a Null Topic for V1 preceded by a coordinating conjunction.^[12]

As expected for a low V2 Romance language, embedded clauses exhibit a clear preference for preverbal subjects, which precede the verb in 75 % of clauses. Cases of inversion are scarce, and exclusively occur in adverbial clauses (Table 5).

This stark asymmetry between main and embedded clauses is often explained by the lack of verb movement to the left periphery in central adverbial clauses, which constitute the bulk of embedded clauses contained in the database (see the protases contained in (35) and (36) above, for instance). Since the verb remains in IP, subjects canonically located in SpecIP precede it.

Cases of inversion, however, suggest that V-to-C is possible, at least, in peripheral adverbial clauses, as in (37):

In (37), the adjunt PP en les ligs godes ‘in the Gothic laws’ is fronted to the left periphery, preceding the verb, which in turn, precedes the subject, hom ‘one’, followed by the direct object, homenatge ‘homage’. This adverbial clause contains an illocutionary act that is independent from that of the main clause (the fact that the legal act of homage is listed in Gothic laws), as expected for a peripheral adverbial clause. As discussed above, peripheral adverbial clauses can exhibit main clause phenomena, amongst which, V2.

As shown in Table 2, half of embedded clauses are verb initial. Only 4/84 clauses have overt postverbal subjects. Of the four cases with overt postverbal subjects, three present predicates that are prone to VS orders (subjects in situ with the verb in IP in Modern Romance), such as passive constructions:

In (38), the subject occurs postverbally, as expected for this type of construction. Similarly, in (39), the subject of the verb donar ‘to give, grant’, in the passive form, follows the verb. In this case, the subject is an infinitival clause. Therefore, VS in embedded clauses with V-to-T correlates with the same triggers as it does in Modern Romance, contrasting with VS in main clauses, where it correlates with V2 Germanic-inversion contexts, in which the verb raises to the left periphery.

From our overview of word order in embedded clauses, we can establish that while V-to-T is the dominant word order pattern in embedded clauses, explaining the prevalence of V1 and SVO, V-to-C is also attested (see (38) and (39) above), depending on the central/peripheral status of the clause, that postverbal subjects obey the distributional rules expected for SVO languages.

Before concluding our discussion of the Usatges, we need to consider the distribution of two elements: si and e. Si has already been introduced in Section 3, where we saw it behave like an expletive within a low V2 grammar. We have not found any examples of si being used in the same way in the Usatges. Nevertheless, we have found two sentences where e’s acts as an expletive-like element, as it did in Old Italian (Poletto 2005) and not as a coordinating conjunction. Consider (40) and (41):

In both (40) and (41), e occurs between a peripheral embedded clause, in (40) a protasis, and in (41) a reason clause, and the verb. Poletto (2005) finds similar cases of e in Old Italian:

In her analysis, Poletto proposes that in Old Italian e behaves like a topic continuity expletive occurring between the core of the clause and a scene setter. The fact that e is followed by enclisis further supports this idea. However, the two examples contained in the Usatges database do not contain clitic pronouns, and is, therefore, impossible to establish whether it triggers proclisis, like si, or enclisis (as expected for a topical element). Nevertheless, its occurrence in two pro-drop clauses that pick up on a referent active in the wider discourse and the parallelisms with Old Italian suggest that this might be a possible analysis for a particle that has not been noted in the literature.

Summing up, data from the Usatges displays an asymmetric low V2 grammar with an unspecified preverbal field. The evidence presented covered inversion patterns, clitic placement, verb-initial clauses, and main/embedded clause asymmetries. Germanic-style inversion is found when a constituent other than the subject saturates SpecFinP. Clitic placement does not abide by the Tobler–Mussafia law in the classic sense, and is not only connected to the overt saturation of the preverbal field (enclisis is systematically borne in verb initial clauses), but to the informational value of the preverbal constituent.^[13] Verb initial clauses show the distribution pattern expected for low V2, either being discourse initial or showing topic continuity. There is a sharp asymmetry between the word order pattern of main and embedded clauses, the latter only exhibiting V2 when peripheral. Otherwise, they present V-to-T, with a tendency to display preverbal subjects, postverbal subjects only occurring in contexts that favour them in SVO languages. It has further been suggested that the Usatges present a topic continuity marker, e.

7 Early Old Catalan in the early Romance context and beyond

Sections 3–6 have offered an overview of the syntax of EOC from the 10th century to the 11th. Borrell’s diploma, dating of 933, contains features that correspond to an asymmetric low V2 grammar with an unspecified preverbal field. While Bonsom’s texts do not offer the same clear picture, they provide data compatible with this analysis and with analyses provided for contemporary Romance varieties. Finally, the Usatges provide us with conclusive evidence of the passage of Old Catalan through a low V2 phase that seems operative throughout the period under study. But how does V2 emerge in Old Catalan, and how does this help us understand the evolution of the Romance languages syntactically as a continuum?

Ledgeway (2017) analyses the Peregrinatio Egeriae, a 4th century text written in Latin, as containing signs of an incipient V2 grammar: there is systematic V-to-C yielding VSO strings, the first step of V2 as defined by Holmberg (2015: 375). The second step associated to an active V2 grammar, that is, the movement of a constituent to the left of the verb, is, at this stage, optional. This is illustrated in (43):

If a V2 grammar had partially developed by the 4th century, the main change from the scenario sketched by Ledgeway to the EOC data presented here is the consolidation of the second step of V2 as a requirement of all unmarked main declarative clauses, which would have occurred during a period in which incipient Early Romance would have still exhibited syntactic continuity. Ledgeway explains the predominance of SVO in embedded clauses by suggesting that SpecIP became a subject position before V-to-C was generalised, thus rendering embedded clauses archaising with respect to main clauses.

Other authors have proposed alternative analyses for Late Latin syntax, sketching alternative hypotheses for the changes attested from Latin SOV to Medieval Romance word order. Klævik-Pettersen (2019) argues that Late Latin VSO strings are instances of V-to-I, with the subject in situ and an emergent SpecIP as a canonical subject position, accounting for SVO oscillations. According to this author, the emergence of V2 in Old French is connected to Germanic influence and syntactic transfer at acquisition level before the end of the Merovingian period. This hypothesis cannot be transferred to the case of Old Catalan, given the stark differences in the sociohistorical context and their impact on the language landscape, including the arrival of Visigoths upon the fall of the Roman Empire, who took control of the administrative government of the area, but represented only up to 10 % of the population (Collins 2004), followed by the arrival of the Arabs in the 8th century. Given that we see low V2 emerge in both Old French and Old Catalan, for the Germanic hypothesis to be applicable to both languages, both Frankish and the language and the Visigoths would have to have had a full active V2 grammar, and the proportion of Germanic population and linguistic pressure to ensure the kind of bilingualism that would grant syntactic transfer would have had to been greater than that of 10 % of the population, broadly corresponding to a ruling class that adopted Latin as their administrative language. Therefore, this hypothesis is discarded for Old Catalan. Here, V2 is assumed to have emerged within a Romance syntactic continuum pre-9th century. Further evidence for this perspective comes from the word order pattern of Old Sardinian, a syntactically archaising language within Medieval Romance. Wolfe (2015b) shows that Old Sardinian presented V-to-C, the first step associated with V2, without an EPP feature associated to the projection targeted by the verb.

By the 11th century, Old Catalan, Old Occitan, Old French, and Old Italo-Romance did not only boast low V2 systems,^[14] but they did also have an expletive use of si, as shown in examples (7) and (8). Si is not present in all Romance varieties: it surfaces, west to east, in Catalan, Occitan, French, and Italo-Romance.^[15] This suggests that the Medieval Romance languages evolve jointly until the consolidation of the V2 grammar, as it is found in across all of them, but the grammaticalisation of si in these three languages and Italo-Romance suggests a syntactic split between Ibero-Romance and Gallo- and Italo-Romance.^[16] The analysis presented here, therefore, presents evidence for the inclusion of Old Catalan in the Gallo-Romance continuum from a syntactic perspective during the period under study, and for viewing Occitan as a central variety in this dialectal linkage.

By the 13th century, Old Catalan’s low V2 system exhibits the first signs of weakening (Pujol i Campeny 2025a, in press), Old French’s low V2 gets reanalysed as a high or strict V2 system (Wolfe 2018a), while Old Occitan retains a low V2 grammar (Vance et al. 2010; Wolfe 2017, 2018a), and Old Italo-Romance varieties follow different, individual paths (Ledgeway 2008; Poletto 2014; Singh 2021; Matticchio 2024). At the same time, Old Spanish, like Old French, developed into a strict or high V2 system by the 14th century (Wolfe 2015a, 2015b: Ch. 6). Therefore, the 13th century appears as a key turning point in the evolution of this syntactic continuum, where each of the three languages takes different paths at sentential word order level. It is possible that from the geographical centre of the Romance linguistic continuum, Catalan acted as an innovative pole, being the first of the recorded Medieval Romance varieties to lose V2 after having presented a low V2 system akin to that of Old Occitan and Old French, including an expletive-like element in the earliest documents, that is, nonetheless lost by the 11th century.

The analysis presented here makes the most out of the little bad data that has reached us for the period under study: by exploring the seemingly poor syntax of legal texts written in Latin, and comparing them amongst themselves, we have been able to positon Catalan within the wider Early Romance picture, opening up new questions regarding the evolution of this linguistic continuum.

8 Conclusions and further research avenues

In this article, we have explored the syntax of four texts from the historical region of Catalonia, dating from the 10th to the 11th century: three in Latin, and one written in Early Old Catalan. It has been shown that the seemingly broken syntax of Borrell’s 10th century diploma hides several telltale signs associated with the syntax of Medieval Romance, and specifically, that of Gallo-Romance: a low V2 grammar with si, an expletive like elements used as last resort strategies in order to satisfy the second requirement of an active V2 grammar, and Romance clitic placement. Bonsom’s 10th and 11th century texts do not present si, but and display a grammar with a strong theme/rheme division between the preverbal and the postverbal space, but they contain cases of inversion that are consistent with a Romance low V2 system. The presence of an active V2 grammar is further confirmed by the Usatges de Barcelona, compiled in the 11th. The Usatges present all the features of a low V2 system, where the verb systematically raises to FinP in unmarked declarative clauses, the preverbal field is unspecified, allowing for the occurrence of informationally new constituents preverbally, with canonical Germanic-style inversion, and expletive use of the topic continuity marker e.

Thus, the evidence presented here clearly demonstrates that up to the 12th century Old Catalan patterned with its neighbouring varieties in exhibiting a low V2 grammar, suggesting that up to this point in time, central Romance varieties, understood as the Gallo-Romance continuum, including Old Catalan and at least some Italo-Romance varieties, formed a syntactic continuum, excluding Old Spanish and Old Portuguese, and potentially Old Romanian (which is not documented for this period).

This highlights the need to study the syntax of Early Romance holistically in order to get better insights on the nature of their syntactic fragmentation. It also shows the importance of adding Old Catalan into comparative Romance studies. Without this often overlooked variety, our understanding of the evolution of this language family remains incomplete.