Sak-relatives in Reunion Creole: examining the distinction between light-headed and free relatives

4.1 Sa-headed light-headed relative

To ascertain whether sak-relatives derive from a light-headed structure, it is first useful to introduce RC’s headed relative clauses. In RC, relative clauses are externally-headed, postnominal and optionally marked with an invariant relative marker, k(e).^[12]

A distinguishing feature of headed relatives in RC is that they favour zero-marking, illustrated in (21). This preference for zero-marking is strongest in subject relative clauses (McLellan 2019, 2023a, 2023b).

Returning to sak-relatives, Chaudenson (1974: 365) and Corne (1995: 61), in their own terms, suggested that sak-relatives are (or were) light-headed, segmenting sak into two words and referring to sa as the antecedent of a relative clause:

The relativiser k(e) is the same marker found in headed relatives in RC (cf. [20]), and sa is a demonstrative pronoun, likely deriving from French demonstrative ça. Outside the relative clause context, sa occurs as an independent pronoun, exemplified in (23).

Unlike French ça, RC sa can have human antecedents, illustrated by (24a-b), which also show that sa can refer to singular or plural entities.

Assuming that sak-relatives do derive from the LHR structure described, the first question to ask is whether or not this structure still exists in modern-day RC. If sak-relatives are light-headed, they do not behave like RC’s headed relatives with respect to relative marking. Chaudenson (1974) and Corne (1995) noted that the antecedent sa must be followed by the relativiser k(e), and my data, collected several decades later, suggest that this remains the case.^[13] Tables 2 and 3 illustrate, respectively, the patterns of relative marking found in headed relative clauses (with a lexical antecedent) in my corpus data and that found in sak-relatives (under the assumption that sak-relatives are light-headed with sa as an antecedent). In Table 3, the number of examples where sa is found alone in a relative clause (not followed by k/t/d) is compared with that where sa is followed by k. I did not count examples where sa is followed by t or d because these are not relative markers in RC.

The figures in Table 3 indicate that relative k is virtually obligatory with sa, occurring in 96.4 % of examples compared with only 24 % in regular headed relatives. Given the high frequency, or virtual obligatoriness, of k following sa, in Section 5.1 I will consider the proposal that sa + k has been reanalysed as one word, which, in relative clause contexts, functions as a new free relative pronoun. The first clue in support of the proposal is the very occurrence of the phonological variants sat, sét and sad. Note that these non-k forms are by no means rare: there were 167 examples in my corpus, making them more frequent than k-forms (of which there were 80). The existence of these t- and d-forms suggests that the combination of sa and k was reanalysed as one word, sak, which subsequently underwent phonological change. If we consider the number of examples where sa is found in a zero-marked relative as compared with any of the five variants of sak, including the t- and d-forms, sa is even rarer in the headless relative clause context, occurring in 3/250 (1.2 %) examples.

Despite observations that sak may have been reanalysed as a single unit by most speakers, there do remain some remnants of the light-headed structure, as indicated by the figures in Table 3: there were three examples where sa was found alone in a relative clause, offering some evidence that an LHR exists with sa as light head. One such is example is given in (25).

This evidence indicates that sa can behave as the light head of a zero-marked relative clause, following the pattern found in RC’s headed relative clauses.^[14] In the next section, I reveal a second LHR structure for sak-relatives.

4.2 Sak-headed light-headed relative

Relevant to our investigation of the structure of sak-relatives is the observation that the same five forms are found outside of a relative clause context as demonstrative pronouns in RC. These demonstrative pronouns often occur as the head of a possessed phrase, illustrated by (26a-c).

These demonstrative pronouns cannot occur independently: they must either be followed by their possessor NP as in the examples above in (26) or by adverbial la as in the examples below in (27).

The presence of sak, sat, sét, sék and sad as demonstrative pronouns with no following relative clause indicates that these forms exist in RC as units where k, t and d cannot be analysed as relative markers. It is likely that sad was the original form in this context, deriving from French ça de ‘that of’. However, in present-day RC the other four variants are interchangeable with sad in this context: they are all attested in corpora and/or the literature (e.g., Albers 2019; Quartier and Gauvin 2022; Watbled 2021a) and received equally positive acceptability judgements in interviews.

Bearing in mind the distribution of sak and variants, the next question to ask is whether the five forms that occur in this non-relative context are the same as those that occur in a headless relative context. In other words, given that demonstratives are good candidates for light heads (cf. Section 3), are the forms sak, sat, sét, sék and sad that occur in headless relatives actually demonstrative heads of an LHR structure, which has developed as a distinct structure from the original LHR headed by sa?

Evidence that sak and variants can indeed be demonstrative heads of an LHR was found in interviews. Speakers were asked for acceptability judgements of examples like (28), in which sat/sak is followed by a relativiser ke.^[15] Note that no such examples were found in the corpus.

The judgements of (28) and similar examples were varied but a non-negligible number (9/14 participants asked) accepted an example with sat ke.^[16] Those speakers judged the same sentences with sak ke as considerably worse, noting that k is redundant because it already occurs in sak, which itself lends support for the argument that sa k has been reanalysed as one word (see Section 5.1). Although there was a discrepancy between the judgements of sak and sat in examples like (28), example (29), produced by an interview participant, offers further evidence that sak can also be a light head. The participant found sak and sat to be interchangeable in this example.

In (29), sak/sat is the head of an LHR in which the relative pronoun ki, whose antecedent is sak/sat, functions as the complement of a preposition. The presence of ki indicates that sat/sak does not function as a free relative pronoun in this sentence and that the k/t of sak/sat is not a complementiser, but rather, that sat/sak is a demonstrative pronoun, which is co-referent with the preceding NP lo fiy.

To return to the question of whether the sak found in sak-relatives is the same functional element - a demonstrative pronoun - as that found outside of relative clauses, an implication of the evidence presented in this section is that we could, in theory, propose that all sak-relatives are LHRs with sak and variants as their head, but that usually they occur with zero relative marking, as represented in (30).

For the speakers who did not accept such sentences with an overt relativiser (ke) (cf. (28)), their judgements may be related to the language’s preference for zero-marking in headed relatives (cf. Section 4.1). In other words, even if sak-relatives are light-headed with sat and variants as the light head, they may be consistently zero-marked.

To summarise, I have presented evidence to indicate that there are two LHR structures for sak-relatives: one headed by demonstrative sa, and one headed by a newer demonstrative, sak and variants. In the next section, I consider the evidence that these two LHR structures may be giving rise to a third structure for sak-relatives, which is not light-headed.

5.1 Reanalysis in sak-relatives

Reanalysis is understood as “change in the structure of an expression or class of expressions that does not involve any immediate or intrinsic modifications of its surface manifestation” (Langacker 1977: 58). As pointed out by Detges et al. (2021: 3), Langacker’s definition implies that the form of the expression is open to two alternative structural interpretations. I argue here that sak-relatives are open to not only two but three structural interpretations (cf. Table 1).

Although orthography is not a reliable indication of syntactic structure, particularly in a language with no standard orthography as is the case for RC, it does give an indication of how native speakers analyse a sequence and is thus useful for understanding reanalysis. Of the 52 written instantiations of sak or sék in my corpus, all but one was written as one word. The orthographic choices of native speakers thus support the argument that the sequence sa + k was reanalysed as one word. This reanalysis implies a syntactic re-bracketing from the structure in (31a) to that in (31b), involving a ‘boundary shift’ in Langacker’s (1977) terms. Crucially, there is a structural ambiguity in the surface form allowing the reanalysis to take place.^[18]

The change in syntactic structure can be explained using the Role and Reference Grammar (RRG; Bentley et al. 2023; Van Valin 2005, 2008a; Van Valin and LaPolla 1997 and others) notion of the Reference Phrase (RP). RPs are referring expressions which serve as an argument of a verb or ad-position and are typically described as NPs or DPs in other frameworks.^[19] RRG does not assume a universal set of lexical categories, nor that a syntactic category is a projection of its head. A predicate can be of any lexical category, and a similar approach is taken for RPs. RPs have a parallel structure to a clause, which themselves have three layers: the nucleus, which contains the predicate; the core, containing the predicate and its arguments, and the clause, which contains the predicate, arguments and non-arguments. Non-arguments are found in what is termed the periphery, which can modify any of the three layers of the clause. Like the clause, RPs have a layered structure, with a nucleus that can be of any category, though it is often a noun. In RRG, a headed restrictive relative clause together with its antecedent is analysed as a complex RP, where the antecedent occupies the nucleus of the RP and the relative clause is a modifier found in the periphery of the nucleus of the RP (McLellan 2023a; París 2023; Pavey 2004; Van Valin 2005, 2012; Van Valin and LaPolla 1997). In McLellan (2023a), a parallel structure is proposed for RC’s LHRs, with the light antecedent occupying the nucleus of the complex RP. This structure is illustrated in Figure 1 using example (31a).^[20] Relative k in Figure 1 is a relativiser, which is represented as a Clause Linkage Marker (CLM) in RRG - an element that links units in complex constructions.

Figure 1:

Syntactic representation of LHRs in RC.

Reanalysis of the LHR structure (cf. Figure 1) into an FR results in a structure that differs in a few key respects. Rather than being a complex RP containing nuclear modification, an FR is an RP which itself contains a clause, illustrated for (31b) in Figure 2.

Figure 2:

Syntactic representation of true FRs in RC.

In Figure 2, sak is analysed as a free relative pronoun, found in the pre-core slot, a position in the structure of the clause that is motivated by word order and information structure considerations. The pre-core slot is reserved for fronted, focal elements and wh-words in languages like English, in which they occur at the beginning of the clause, not in-situ. Although sak is not a wh-pronoun, I place it in the pre-core slot because it is found at the front of the clause irrespective of its relation to the predicate. If instead sak was analysed as a complementiser, under a super-free relative analysis, then sak would be represented as a CLM and the structure would lack a pre-core slot. This would pose a challenge for the linking between syntax and semantics (for details of this linking in RRG, see Van Valin 2023) because, while see links to sak in the pre-core slot in Figure 2, if sak was represented as a CLM, see would have a missing argument. I discuss the issue further in Section 5.2.6.

In Section 4.2, I offered evidence that an LHR exists with the demonstrative pronoun sak as its head, the key evidence being that sak-relatives can be marked with an overt relative marker for some speakers, even if they are usually zero-marked. This evidence points to there being a second LHR structure for sak-relatives, illustrated in (32a), which presents another opportunity for reanalysis of an LHR structure to an FR structure (32b) to occur, with the same result as that described above:

Also an LHR, example (32a) has the same syntactic template as example (31a) (cf. Figure 1), but instead of sa, the demonstrative head sak occupies the nucleus position and the structure often lacks a CLM position as the complementiser is usually absent. The three structures - two LHRs and one FR - are illustrated in Figure 3.

Figure 3:

Three syntactic structures for sak-relatives.

The consistency of the relative marking patterns of both LHR structures means there is consistently an ambiguity between the LHR structures and the FR structure, making reanalysis into the true FR structure possible. LHRs headed by sa are virtually obligatorily marked with k (cf. Table 3), and LHRs headed by sak virtually always occur with zero-marking - recall that, although some speakers accepted the structure in (31a) with an overt complementiser (cf. [(28)]) other speakers rejected it, and such a structure did not occur at all in corpora nor has it, to my knowledge, been reported in the literature.

Another factor that may be relevant in the reanalysis is that the original LHR structures and the resulting FR structure are all RPs. The fact that phrases are not necessarily endocentric in RRG means that the RP can be of any lexical category. The syntactic reanalysis from LHR to FR results in a less complex, more cohesive RP because there is no nuclear modification in the resulting structure, and reanalysis often results in simplification of syntactic structures, an example of which would be boundary reduction (Langacker 1977: 103), which is what I argue has happened in the case of sak.

There has been much debate about the relationship between reanalysis and grammaticalisation (see Detges et al. 2021: 36–37), but many authors conceive of reanalysis as a mechanism that is required for grammaticalisation to occur, though the reverse is not true: reanalysis can occur without grammaticalisation (Hopper and Traugott 2003: 39–63). Given that reanalysis is typically conceived of as a change that is not overtly manifested in the form of what is being reanalysed (Detges et al. 2021), the development of the variants sat, sét and sad presupposes the reanalysis of sak as a single unit, and, as will be explored in the following sections, supports the suggestion that these variant forms have subsequently grammaticalised as free relative pronouns. Note that all instances of a t- or d-form were written as one word in the corpus, and interview participants commented that sak was one word, interchangeable with sat, sét, sék and sad, corroborating the view that sa + k was reanalysed as a single unit that subsequently underwent phonological change. This phonological change is seen as one of the several signs of grammaticalisation of sak-relatives, discussed in the next section.

5.2 Arguments for the grammaticalisation of sak as a free relative pronoun

Grammaticalisation is understood as “the development from lexical to grammatical forms and once the grammatical form has evolved, the development of further grammatical forms” (Narrog and Heine 2021: 1). As observed in this definition, grammaticalisation can involve the development of grammatical items from lexical items but also from already grammatical ones. Here I consider the proposal that sak and variants are grammaticalising into free relative pronouns from a demonstrative, which is typically considered to be an already grammatical item. Demonstratives are a well-known source for grammaticalisation into various other functional items, including definite determiners, copulas, complementisers and relative pronouns (Bollée 2004; De Mulder and Carlier 2011; Diessel 1999a, 1999b, 2006, 2013; Diessel and Breunesse 2020; Diessel and Coventry 2020; Himmelmann 1997; Lehmann 2015[1982] among others); however, one that has received less attention is the grammaticalisation of free relative pronouns.

It is often noted that free relative pronouns are identical to interrogative pronouns, but it is less clear whether this is because free relative pronouns typically derive from interrogative pronouns or rather whether there is something inherently similar about the functions of these two types of pronouns that makes many languages not distinguish between them; I return to this point in Section 5.3. In RC, interrogative pronouns encoding the ontological categories of person or thing – kisa ‘who’ and kosa ‘what’ – are only marginally accepted in a headless relative context. Instead, sak and variants are preferred, indicating that RC’s headless relative system does not reflect crosslinguistic trends and, as I will argue, demonstratives may instead be the source of RC’s FR pronouns. The origins of the demonstrative itself have not been clearly traced back to a lexical source (Diessel 1999a), which Diessel (1999a, 2006, 2013) takes as evidence for the argument that deictic demonstratives are a basic element of human communication and constitute a class of their own, distinct from lexical and grammatical classes. Diessel argues that the development of new grammatical items comes from either lexical items or deictic demonstratives. Whatever its own origins, the RC demonstrative sa is the source of both the demonstrative sak and variants (cf. Section 4.2) and, as I propose here, a new free relative pronoun, so we are not dealing with the emergence of a new grammatical item from a lexical item.

Diessel (1999a: 118) proposes the following criteria to assess the grammaticalisation of demonstratives.

In what follows, I explore RC’s sak-relatives with respect to Diessel’s criteria above, arguing that they meet the majority of these criteria as well as exhibiting more general principles of grammaticalisation.

5.3 Why are demonstratives a source for free relative pronouns?

In Section 5, I suggested that two LHR structures appear to be grammaticalising into an FR structure, implying that demonstratives could be a source of FR pronouns in RC. If demonstratives are indeed another source for the function of FR pronoun in languages, then we should seek explanation for why that is. Diessel (2003), who considers the relationship between demonstratives and interrogatives and their place in language, offers compelling insights on the matter. Diessel distinguishes demonstratives and interrogatives from grammatical markers, arguing that they have a special status in language. He argues that the core function of demonstratives (cf. Section 5.2.1) and interrogatives is not the same as that of grammatical markers, which serve to relate lexical expressions to one another or qualify them. Demonstratives and interrogatives, on the other hand, are concerned with the interaction between speaker and hearer, rather than being language-internal (although they can be). Diessel (2003) argues that demonstratives and interrogatives are both directives: they both initiate a search for the identity of a referent. In the case of demonstratives, the speaker focuses the hearer’s attention on a particular referent, which is identified in the surrounding situation with the help of perceptual and contextual information from the physical and discourse contexts. In the case of interrogatives, the speaker also initiates a search for a particular referent, which, with the help of the discourse context, is identified by the hearer’s knowledge. Although the speaker knows the identity of the referent in the case of demonstratives, but not in the case of interrogatives, the commonality is that the speaker initiates a search for information, which in both cases is also aided by the ontological features encoded by the demonstrative or interrogative, e.g. person/thing, place, manner. Furthermore, along with sharing a similar pragmatic function and encoding the same ontological features, Diessel (2003) points out that interrogatives and demonstratives also often include the same derivational morphemes, which is indeed the case in RC.^[34] RC’s interrogatives share sa: kisa ‘who’, kosa ‘what’, kansa ‘when’, ousa ‘where’ (cf. Table 4).

Concerning the commonality between demonstratives and interrogatives, I would argue that initiating a search for the identity of a referent is also what free relative pronouns do. While the search for the referent is aided by the surrounding situation and discourse context for demonstratives and by the hearer’s knowledge for interrogatives, the identity of the referent of a free relative pronoun is aided by the information given in the relative clause. The finding that demonstratives are a source of free relative pronouns in RC thus offers support for Diessel’s argument concerning the similarity between demonstratives and interrogatives, and simultaneously helps to explain why demonstratives have developed into free relative pronouns in this language.