Grammaticalization in Seychelles Creole: the coding of reciprocity by kanmarad

Seychelles Creole (SC) is one of the few creoles with a grammaticalized reciprocity marker. The grammaticalized use of kanmarad (< Fr. camarade ‘comrade, companion’) is mentioned in the grammars of SC (Bollée 1977; Corne 1977; Choppy 2009) but its evolution and distribution in modern SC have never been analyzed. This contribution first presents present-day data from spoken and written corpora of SC and compares them to data published in the Atlas of Pidgin and Creole Language Structures, APiCS (Michaelis & al. 2013). Appealing to several grammaticalization mechanisms discussed in the literature, it then traces back the grammaticalization process of kanmarad, a process that is not very advanced in the closely related Mauritian Creole (MC). In accordance with Michaelis & Haspelmath (2020), the evolution of kanmarad in SC can be considered to be an instance of accelerated functionalization which the authors consider to be typical of creole languages. Ultimately, the study’s findings are discussed in light of two complementary hypotheses that try to explain the acceleration of functionalization: the Extra-Transparency Hypothesis (Haspelmath & Michaelis 2017) and the Distinction during Codification Hypothesis which I suggest for SC. Both are considered to be possible factors favoring an ordinary language-internal grammaticalization process.


Introduction
SC, the first national language of the Republic of Seychelles, is one of the few creole languages possessing an innovative and grammaticalized marker of reciprocitya lexical item that has been freshly grammaticalized. Although the token frequency of the reciprocal construction with kanmarad (<Fr. camarade 'comrade, companion') 1 is not very high, I observed during recent fieldwork that it is used in all language registers. The following is an example from spontaneous oral speech of a Seychellois herbalist who explained in an informal conversation the ecology of medicinal plants he grew in his garden.
( Was this just a metaphorical expression, some kind of humanization of the plants he was talking about? I do not think so and this study aims to explain why. SC is the native language of about 95% of the Seychelles' population and is spoken by some 85,000 speakers. It is considered to be a continuation of stable varieties of Mauritian 1 Given the multiple writing conventions and the comparative perspective of this study I use the term 'companion'-based reciprocal construction (see APiCS). 2 My corpus examples are given with an interlinear version that follows the Leipzig glossing rules. The glosses are first given in French, the lexifier language of SC, followed by an English version.
Creole (MC) that were imported at the end of the 18 th century 3 . Modern MC and modern SC are mutually intelligible. In both countries we are witnessing an increased use of creole in formal contexts as a consequence of independence. This use is however much more common in Seychelles than in Mauritius where the creole language has no official status. This paper will focus on the use of kanmarad, which is slightly different between the two languages and has to be considered an example of the few differences between the two varieties on the morphosyntactic level.
Contact-induced language change has been a much-discussed topic in recent years and creole formation has been an important area of interest in this field of study. A great deal of attention has been paid to contact-induced grammaticalization in creole formation (e. g. Bruyn 1996Bruyn , 2009Hopper & Traugott 2003;Kriegel ed. 2003;Plag 2002;McWhorter 2018) and in other language contact situations (e. g. Heine & Kuteva 2003, 2005. Michaelis & Haspelmath (2020) state: "A very important phenomenon that interacts with grammaticalization in creole languages is 'semantic imitation' of substrate languages". Different cases of what Michaelis & Haspelmath (2020) summarize as "semantic imitation" have been discussed in the literature under different rubrics, e. g. ordinary contact-induced grammaticalization (Heine & Kuteva 2003, 2005, apparent grammaticalization (Bruyn 1996(Bruyn , 2009, and polysemy copying (Heine & Kuteva 2003). As Bruyn (2009) points out, distinguishing among these different mechanisms, although necessary, proves to be difficult in practice. This paper discusses the evolution of kanmarad into a grammaticalized marker of reciprocity in modern SC in the light of these approaches to grammaticalization. I argue that the evolution of kanmarad is not due to substrate influence (section 2.3) but can be explained by an ordinary language-internal grammaticalization process (section 3). This result is consistent with McWhorter's (2018) claims that "there is no 'creole' kind of grammaticalization" and that "grammaticalization has indeed occurred to an unusually vast degree in the few centuries that most creoles are known to have existed" (McWhorter 2018). Further, my explanation will appeal to the idea of accelerated functionalization put forward by Michaelis & Haspelmath (2020) (section 4.1). Of course, my argument in no way negates the fact that contact-induced change has occurred in other areas of SC morphosyntax.
In Sections 2.1 and 2.2, I present and compare data from spoken and written modern SC with data from APiCS (Atlas of Pidgin and Creole Language Structures) and with data from other languages. Section 2.3 addresses the question of possible contact-induced change. Section 3 deals with the grammaticalization of kanmarad. Using data from both earlier and contemporary SC and MC, I discuss several grammaticalization mechanisms that played a role in the grammaticalization of kanmarad. Lastly (section 4), after a brief discussion of the notion of accelerated functionalization suggested by Michaelis & Haspelmath (2020), I consider possible reasons for the emergence of the 'companion'-based reciprocal construction. I discuss the Extra-Transparency Hypothesis and the Distinction during Codification Hypothesis as possible factors favoring an ordinary language-internal grammaticalization process.

Kanmarad in SC and the data from APiCS
I am interested in the morphosyntactic marking of reciprocal constructions without focusing on their highly complex semantics (see e. g. Nedjalkov 2007). I am perfectly aware of the fact that, as in many languages, reciprocity can stay unmarked, above all with symmetrical verbs (Haspelmath 2013) as illustrated in (2). 3 Chaudenson (e. g. 2013) also considers the input from Bourbonnais, less important than the one from MC.
(2) SC Pyer ek Lise pe anbrase. Pierre avec Lise PROG embrasser Peter with Lise PROG kiss 'Pyer and Lise are kissing/hugging.' (data elicitation) However, I concentrate on morphosyntacticly marked examples with a special focus on the 'companion'-based reciprocal construction.

The data in modern SC
The first linguists to observe the evolution of the lexeme kanmarad as a reciprocal pronoun in SC were Bollée and Corne, in their grammars, both published in 1977 The pharmacy works in collaboration with the Logan-Hospital, in a partnership in which they complement one another.' (Seychelles Nation 8-11-2016) Among the alternative techniques mentioned by Choppy (2009) (see above), only enn a lot is restricted to the expression of reciprocity as it is the case for kanmarad. I will briefly discuss this expression: enn a lot derives from the French pattern un à l'autre ('one another') (Guentchéva & Rivière 2007) and also exists in several other French-based creoles (see table 1). Besides the example given in Choppy (2009) it can be found in corpus example (3), in which the speaker uses kanmarad and, additionaly, enn a lot to reinforce his statement (also see Bollée 1977: 50-51). As to example (3), my Seychellois colleagues gave me the following information: "Kanmarad can be left out here with no change in meaning. In fact, enn a lot is a synonym of kanmarad in this context." (Penda Choppy, p.c.) The use of enn a lot is not mentioned by Michaelis & Rosalie (2013) (APiCS). The only example in our spoken corpora is example (3) where enn a lot is used in addition to kanmarad. A variant with the agglutinated French definite article l' is attested in an example drawn from a written register 4 : (9) SC a. Siport lenn a lot pou adapte soutenir un à autre pour adapter support one to other to adjust 'to support each other so as to adjust' (Verbatim Report, National Assembly of Seychelles, https://www.nationalassembly.sc) Data elicitation revealed that (l)enn a lot can be used as an alternative to kanmarad in some contexts. Its use can be considered to be marginal. Its French model, the compound reciprocal pronoun l'un l'autre is marked for gender and number and takes prepositions which are inserted between the two components (Guentchéva & Rivière 2007: 564). The prepositions change following to the valency pattern of the verb. In addition to this complex morphosyntactic behavior, the use of l'un l'autre is semantically restricted (for details Guentchéva & Rivière 2007: 564). In addition to the frozen form (l)enn a lot going back to a verb used with the preposition à attested in our oral corpus of SC (example 3) and in example (9a) with agglutination of the French article, attestations with verbs introducing their complement by other prepositions are sparse. Example (9b) is drawn from a written register: are felt to be grammatically correct but "not really used by the Seychellois community" (Penda Choppy, p.c.).

Comparison with other creole and non-creole languages
In WALS (World Atlas of Language Structures, Maslova & Nedjalkov 2013) as well as in APiCS, the chapter about reciprocal constructions (Haspelmath 2013) deals almost exclusively with the question of whether reciprocal and reflexive construction have the same markingthis is the case in a lot of languages (German: 'Alfred und Bernhard schlagen sich', French: 'Alfred et Bernard se frappent.'). This question is not central to my study because the French-based creoles have lost the clitic personal pronoun 'se' (and its equivalents for other persons) and have developed partially innovative techniques to mark reflexivity 5 , on the one hand, and reciprocity, on the other. This is the reason why this study considers only the values linked to reciprocal constructions not identical to the reflexive, so-called special reciprocal constructions (APiCS), with a special focus on the French creoles. Most French creoles belong to the group of creoles with a special reciprocal construction based on 'other' which maintains the French pattern l'un l'autre (used in 5 The French creoles use body part reflexives and, alternatively, the object form of the personal pronoun, often followed by an intensifier (e. g. Kriegel & Ludwig & Pfänder 2019). addition to the clitic personal pronoun). As has been shown in section 2.1, in SC the use of (l)enn a lot based on 'other' is also a marginal possibility.  As MC will be studied in more detail in Section 3.3, I will briefly consider the situation in Guianese French Creole: The first novel in a French Creole is Atipa, written in 1885 in Guianese French Creole by Alfred Parépou, a pseudonym for an unknown writer (but see  I would like to thank the anonymous reviewers who mentions the existence of an earlier form without the agglutinated French definite article l', quoted by Baissac (1880: 22): "L'un l'autre. Une construction récente place la préposition à entre les deux pronoms pour marquer la réciprocité. Ex. : Mariez-les vite, ils s'aiment, Marié zautes vitement, zautes content éne à laute.
A comparable corpus example from modern Guianese French Creole is cited by Jennings & Pfänder (2018: 150). It should be noted that unlike in modern SC the possessive determiner is used in both examples (10) and (11).
(11) Wonm-yań tchoué yé kompannyen Men-DEF kill 3PL RECP 'They killed each other.' (Jennings & Pfänder 2018: 150) According to APiCS, the other creole languages to possess a special reciprocal construction based on 'companion' are the three varieties of Portuguese-based Cape Verdean Creole as well as Berbice Dutch and Creolese, the English-lexifier Guyanese Creole spoken in vicinity of Berbice Dutch.
Beyond creole languages and in a wider perspective of language typology, reciprocal constructions based on 'companion' exist in several unrelated languages throughout the world. Evans (2008: 52) reports cases from Zapotec spoken in Mexico. Heine & Kuteva (2002: 92) quote examples from Russian, from several West African languages of the Niger-Congo family (Gabu (Adamawa-Ubangi), Gola (West Atlantic), Fulfulde (West Atlantic), Koromfe (Voltaic)) as well as from SC. They state: "More research is required on the exact nature and the genetic and areal distribution of this process. This is an instance of a process whereby concrete nouns are grammaticalized to pronouns expressing relations among clause participants." (Heine & Kuteva 2002: 93) 7 2.3. The possibility of contact-induced change as an explanation I will now examine the possibility that the 'companion'-based reciprocal construction of SC could be the result of contact-induced change. As stated in section 1, SC can be considered to be a form of MC that was imported at the end of the 18 th century. Though there was a short period (1727-1735) at the beginning of the colonization of Mauritius when West African slaves were imported, the main substrate languages present in the Mascarenes (today Reunion and Mauritius) after 1760 were Malagasy and several Eastern Bantu languages. After the abolition of slavery in 1835 another wave of Eastern Bantu speakers arrived in the Seychelles 8 . The 'companion'-strategy for forming reciprocals is not present in the main substrate languages of SC. Malagasy uses the infix -if-in combination with the prefix m-. 9 As to Eastern Bantu 10 , the dominant pattern marked by -ana is clearly not related to 'companion' 7 For a discussion of the strategy concerning nouns meaning 'comrade', 'mate', 'companion', 'friend', etc., as well as examples from a wider range of languages see Heine & Miyashita (2008: 178-181). 8 As a lot of speakers of Tamil arrived in the Seychelles after the abolition of slavery it is also interesting to check data from Tamil: According to Annamalai (2000), the reciprocal pronoun in Tamil is not related to a lexical item of the type kanmarad. "The reciprocal form is the reduplication of the nominal form oru 'one' with gender and number appropriate to the subject." (Annamalai 2000: 175) 9 'c)Verbes réciproques: Pour exprimer l'idée de réciprocité il faut rajouter l'infixe -if ou ses variantes -ifamp et -ifank au préfixe m-: Exemples: -mifanojo/ se rencontrer -mifampijery/se regarder -mifankatia/s'aimer.' (Jaozandry 2014) (c) Reciprocal verbs: To express the idea of reciprocity, you must add the infix -if or its variantsifamp and -ifank to the prefix m-: Examples: -mifanojo/to meetmifampijery/to look at one another mifankatia/to love one another (my translation)). 10 Woodward (1926: 306) states in his grammar of Makwa, one of the most probable substrate languages for SC: "Reciprocal form: This is made by changing the final -a into -ana: wiwa, to hear; wiwina, to hear one another; wata, to beat; watana, to fight." but goes back to a NP conjunction marker. 11 In addition, Maslova (2007: 350) mentions other strategies to express reciprocity using body-part reflexives in several Bantu languages, but does not mention the existence of 'companion'-based reciprocals.
The situation is different in the case of the other creole languages possessing a 'companion'-based reciprocal construction (see section 2.2), it is highly probable that its existence can be explained by substrate influence from West African Niger-Congo languages. So, for Guianese French Creole Jennings & Pfänder (2018: 150) explicitly mention an example from Ewe and observe: "This syntactic strategy [the 'companion'-strategy, S.K.] may be a transfer from the very similar Ewe strategy." Even if theoretically West African influence in SC via MC cannot be excluded, most of the enslaved people spoke languages that are not good candidates as models for 'companion'-based reciprocals. It should also be pointed out that the grammaticalized use of kanmarad as a reciprocal is limited to SC and that in MC the grammaticalization process is less advanced (see section 3.3). In conclusion, it is improbable that there is a West African model for the kanmarad reciprocal in SC.

Non-grammaticalized uses in modern SCthe process of divergence
In section 2.1, I presented corpus data containing a 'companion'-based reciprocal construction that is clearly grammaticalized. I now want to trace back this grammaticalization process. But firstly, it is important to state that kanmarad continues to exist as a full lexical noun, completely independent from a reciprocal interpretation in SC:  (Bollée & Rosalie 1994: 186) The status of kanmarad in example (12) is clearly lexical: It is used in subject position and could easily be exchanged with other lexemes like zanmi, dalon 'friend' (data elicitation). Moreover, the presence of the possessive determiner son shows that we are dealing with a 11 Maslova (2007: 343) quotes Schladt (1996), who argues in favor of the existence of a grammaticalization path. 12 For a critical discussion of the term "bleaching" see Hopper & Traugott (2003: 94-98). I systematically use "semantic generalization or bleaching". noun. The use of kanmarad as a full lexeme is mentioned in the relevant dictionaries (DECOI, Bollée 2000: 203) which contradicts König & Haas' statement: "La grammaticalisation de kamarad en marqueur de réciprocité est confirmée par le fait que kamarad ne s'emploie plus comme lexème indépendant". (König & Haas 2007: 144) (The grammaticalization of kamarad as a marker of reciprocity is confirmed by the fact that kamarad is no longer employed as an independent lexeme. (my translation)) The existence of kanmarad as an independent lexeme in no way contradicts an interpretation as an instance of grammaticalization. It simply illustrates its synchronic dimension. Raible (1992: 263) speaks of the "Gleichzeitigkeit des Ungleichzeitigen", the simultaneity of the non-simultaneous, Hagège calls it "le principe de la Preuve par Anachronie", the principle of the proof by anachrony (Hagège 2001: 1617), Roberts & Roussou label it "lexical split" whereas Hopper (1991: 22) and Hopper & Traugott (2003: 118-122) speak of divergence.
Grammaticalized uses of camarade, even incipient grammaticalization, do not exist in the lexifier language French. This observation is important because work on grammaticalization in creole languages should not neglect the possibility that grammaticalization may have started in the varieties of the lexifier language present in the creolization process. This is for instance the case of the reflexive marker son lekor (see Kriegel & Ludwig & Pfänder 2019). In the case of camarade, the first attestation found in the FEW is in Middle French from 1560 where the lexeme is used in the sense of 'chambrée de soldats', a soldier's barrack (FEW), https://apps.atilf.fr/lecteurFEW/index.php/page/lire/e/47251. The FEW does not mention uses from Regional Frenches that could be interpreted as incipient grammaticalization.

The widening of contextsmetonymic change in early SC
The first text of a certain length (14,056 words) in SC was written around 1900 (see below). As SC must be considered an offspring of MC (see section 1), it is appropriate to examine early Mauritian texts which have the advantage of going back to the 18th century. As a base of my study, I will work with a corpus of 99,209 words gathered by Philip Baker and Guillaume Fon Sing (Baker & Fon Sing & Hookoomsing 2007 The speaker obviously does not want to fight with the previously mentioned referent cet homme, 'this man'. It is highly improbable that he refers to this manwhom he does not seem to knowas being a companion. 15  comrade like yourself 'D: Don't take your neighbour's wife. (…) Don't be envious of goods that belong to your neighbor. (…) You must love your God with your heart and your neighbor as you love yourself.' (Fon Sing, corpus de textes anciens en créole mauricien) In French, a semantic feature of camarade is clearly 'a familiar person' but its use in the context of the 10 commandments is not attested.
To summarize the use of camarade in early Mauritian texts, we can say that although in most cases it refers, like in French, to a familiar person of the same social condition, we can observe a semantic generalization in some of our examples. 15 Note that this occurrence is one of the very few where camarade is used without any determiner. I do not think that this is sufficient for negating a nominal status as I do for the modern examples.
Let us now discuss the first occurrences of the word camarade in the only 'early' text in SC, an adaptation of La Fontaine's fables by Rodolphine Young, a Seychellois school teacher (Young 1983). This text was probably written around 1900 but we do not know the exact date (see footnote 13). My first impression was that the token frequency of camarade in Young's text is very high. A comparison with the French original confirms this impression: Out of 14,056 words we find 27 occurrences of camarade in the creole version whereas the corresponding fables of the French version contain 11,475 words with only three occurrences of camarade. Camarade is definitely used much more frequently in the creole adaptation. To examine whether the number of occurrences of the word camarade found in Young's adaptation significantly differs from that found in the corresponding French fables, a χ 2 test was performed (see the contingency table in Table 3) 16 . The results revealed that the proportion of occurrences of the word camarade is indeed significantly greater in the creole text than in the corresponding French fables (χ 2 = 13.44; p < 0.001). If we consider the context and the French original of the fable, a translation of son camarade by 'the others' is much more appropriate than a literal translation. In some lines below the author uses les autres, 'the others' in a perfectly similar context. A second example from the same text collection given in (17a) is even more convincing because it establishes an opposition between son camarade and son lé kô ('its body'), the other and the self. We can conclude that in the first text in SC, camarade functions as an independent lexeme, a noun. We are not dealing with grammaticalized uses. No decategorialization in the sense of Bybee (2015: 129-131), Hopper (1991: 30) or Hopper & Traugott (2003: 106-115) has taken place. Every time Young uses camarade in the meaning 'the others' in her fables, the lexeme appears with the corresponding possessive determiner, in most cases with the 3 rd person singular son, but the 3 rd person plural zott 17 is also attested (example 17b). Camarade clearly has nominal status.
In the examples discussed above a semantic generalization has occurred, we observe metonymic change. 'Comrades' are part of the whole constituted by 'the others'. Even if the grammaticalization process has not started yet, the conditions for its triggering are fulfilled.
Following Haspelmath (1999), I consider semantic generalization or bleaching to be a prerequisite for the increase in the frequency of a word: "Semantic generalization or bleaching usually is a prerequisite for use in a basic discourse function, that is for the increase in frequency that triggers the other changes". Bybee (2007: 354) also insists on the fact that repetition or increase in frequency is essential to grammaticalization: "Repetition is universal to the grammaticalization process. Repetition and its consequences for cognitive representation are major factors in the creation of grammar". Hopper & Traugott (2003: 127) state that "the repetition of forms may lead to their 'liberation' or 'emancipation' (Haiman 1994), from their earlier discourse functions and to increased freedom to associate with a wider variety of forms".
In the light of these observations, I argue that the semantic generalization we observe in our examples leads to an increase in frequency (see the high token frequency in Young's fables). This increase by repetition is the prerequisite to the start of grammaticalization proper.
The fact that we do not find grammaticalized uses of camarade in Young's text raises the question of how reciprocity is expressed in the fables: While searching for reciprocal 17 I adopt the original writing. Young alternates between the writing zott and zot but this alternation does not seem to be linked to the desire to distinguish morphosyntactic functions. contexts in the fables I found the following examples morphosyntacticly marked by ensembe (Fr. ensemble 'together'), an adverb used to express comitative and related semantic relations. This example corresponds to a use still encountered in modern SC: in the examples given by Choppy (2009), and in corpus example (4) the verb kontan is used in the same way. The absence of the possessive determiner points in the direction of a possible decategoralization. Example (19) can possibly be interpreted as the first and only instance of a grammaticalized use. However, there is a gap of more than 100 years before we can find comparable examples from the second half of the 20th century. Moreover, the grammaticalized examples without possessive determiner emerge in SC and not in MC where the use of the possessive determiner is obligatory (see Section 3.3). In an elicitation task, I asked my consultants to translate the English version of (19) into modern MC. They used the possessive determiner.  comrades 'They all love one another.'

Further grammaticalization in modern SC: semantic generalization or bleaching, decategorialization, and specialization
In most cases the subject is human but there are some exceptions: see example (1) and (3) which refer to plants and example (7)  'Il ne restera pas pierre sur pierre.' (Bollée 1977: 51) 'Not one stone here will be left on another' (my translation) • decategorialization (Hopper 1991, Hopper & Traugott 2003Bybee 2015: 129) In the corpus examples (1) and (3)- (8), kanmarad is used without determiners. Kanmarad has lost its nominal status and has become a pronoun. The unidirectional shift from the category noun to the category pronoun is labelled as being an instance of decategorialization. (see Hopper & Traugott (2003: 107) 18 .
My Seychellois consultants refused to accept the following examples in which a possessive determiner has been introduced: "When a form undergoes grammaticalization from a lexical to a grammatical form, however, it tends to lose the morphological and syntactic properties that would identify it as a full member of a major grammatical category such as noun or verb (Hopper & Traugott 2003: 107 The rejection by my consultants clearly shows that kanmarad has changed its categorial status as a noun and has lost the property of being the head for determiners. In these cases, kanmarad surely must be considered as a pronoun.

speaks of paradigmatic variability)
While different items can occur in the position of kanmarad in its not-grammaticalized uses (see the discussion of (12) in section 3.1), this is impossible with its grammaticalized uses, the only alternative often being enn a lot in some contexts (see section 2.1). This fact can be interpreted as a specialization defined as "the process of reducing the variety of formal choices available as the meanings assume greater grammatical generality" (Hopper & Traugott 2003: 116 After having studied three classical grammaticalization mechanisms at work, I conclude now that kanmarad has become a fully-fledged, unambiguous reciprocal marker in present-day data from SC.

Brief comparison with present-day data from MC
After having presented early data from MC and SC and having stated that both languages are still closely related in their modern forms, a study of present-day data from Mauritius now seems to be important. While moving lower in the animacy hierarchy (Silverstein 1976) and giving examples with non-animate referents in subject position, the use of kanmarad as a reciprocal marker becomes more and more problematic for the Mauritian consultants: While some Mauritian consultants accepted the use of kanmarad in this example, they insisted on the obligatory use of the possessive determiner, in this case zot (POSS.3PL). Others rejected the example even with the possessive determiner. On the contrary, the Seychellois consultants did not have the slightest hesitation to use kanmarad in this example, without a possessive determiner.
For my Mauritian consultants, the Seychellois example (3)  With one exception, the constructed example with an inanimate referent in (20) was never used with kanmarad by my Mauritian consultants, not even with the possessive determiner.
I conclude that we should speak of incipient grammaticalization in MC. The prerequisites are present: semantic generalization or bleaching is already seen in the early examples and is still present in modern MC. But the use of kanmarad in reciprocal contexts is in general limited to animate referents and must be interpreted as a sign of generalization or bleaching that has not been entirely accomplished. The obligatory use of the possessive determiner shows that decategorialization has not taken place.

Possible motivations for the grammaticalization of kanmarad
In the last section of this article, I would like to discuss whether the observed grammaticalization process can be explained by any external factors. As I have shown in section 2.3, the use of kanmarad as a reciprocal marker cannot be explained by substrate influence: I did not find any models in the substrate languages of SC. But other external factors could be at the origin of the grammaticalization process I examined in section 3. So before concluding, I want to discuss two hypotheses: Michaelis & Haspelmath's (2020) Extra-Transparency Hypothesis linked to second language acquisition and my own sociolinguistic hypothesis which I call the Distinction during Codification Hypothesis. (Michaelis & Haspelmath 2020) In recent creole studies, debates have largely centered on "creole exceptionalism", the question whether creoles are special languages. According to some linguists, creoles show great transparency and simplicity (e. g. McWhorter 2001;Parkvall 2008;Bakker & al. 2011;Leufkens 2013) whereas others insist on the fact that they evolve like any other natural language (e. g. Aboh & DeGraff 2015;Mufwene 2015) 20 . A slightly different perspective is taken by those who ask themselves if there may have been accelerated change processes in creolization (Bruyn 1996(Bruyn , 2009Véronique 1999;Kriegel & Neumann-Holzschuh 2000;Detges 2001;Neumann-Holzschuh & Schneider 2001;Plag 2002;Bakker 2008). These change processes have been discussed with different terminological proposals whose discussion would be beyond the scope of this paper: reanalysis, acquisitional grammaticalization, instantaneous grammaticalization. Michaelis & Haspelmath (2020) speak of accelerated grammaticalization: "(…) It appears that grammaticalization is accelerated in creolization." (Michaelis & Haspelmath 2020) According to them grammaticalization involves three main processes (i) SEMANTIC CHANGE, which often results in (ii) FUNCTIONALIZATION (content items become function items), and is then followed by (iii) COALESCENCE (cliticization, agglutination, fusion of function item). One central observation of this paper is that English-based, French-based and Ibero-Romance-based creoles show a great deal of functionalization, but little coalescence (though the former content items are often reduced). (Michaelis & Haspelmath 2020) Whereas creoles do not show a lot of coalescence, they exhibit a lot of accelerated functionalization. The new function morphemes, so to speak freshly grammaticalized materials, are typical of creole languages and did not exist in grammaticalized uses in the lexifier languages (Michaelis & Haspelmath 2020; for the example of modals see Kriegel & Michaelis & Pfänder 2003). To explain these functionalizations, Michaelis & Haspelmath (2020) discuss two hypotheses: the Loss-and-Replacement Hypothesis 21 , which they reject because "we favour a hypothesis that is compatible with the view that creoles do not (necessarily) arise from pidgins" and the Extra-Transparency Hypothesis, which they adopt.

Extra-Transparency Hypothesis
In social situations with many (or even mostly) adult second-language speakers, people need to make an extra effort to make themselves understood. This naturally leads to the overuse of content items for grammatical meanings, which may become fixed when more and more speakers adopt the innovative uses. (Michaelis & Haspelmath 2020; see also Haspelmath & Michaelis 2017: 16) 20 McWhorter (2018) insists on the fact "that in creoles, at least early in their life spans, there is indeed more grammaticalization than under ordinary processes of grammar-internal change". Nevertheless, he stresses that grammaticalization in creoles is not of a different nature than grammaticalization in other languages. 21 Loss-and-Replacement Hypothesis "In the transmission bottleneck of pidginization, inflectional and other non-salient grammatical markers are lost, because they cannot be acquired (e. g. Good 2012). This leaves a void, and when pidgins turn into full-fledged languages again, they need to fill the gaps by new material deriving from content words." (Michaelis & Haspelmath 2020) The authors suggest a list of new grammaticalized materials from several creole languages, materials which, according to them, have developed because of the need for extratransparency in societies with many or mostly second language speakers. Among the new function morphemes, they mention the 'companion'-based reciprocal markers discussed in APiCS and illustrate them with examples from Guianese French Creole, SC, and Creolese.
I find the Extra-Transparency Hypothesis plausible and I am convinced that it can be verified by careful diachronic case studies concerning e. g. TMA markers or other frequently used markers. However, with respect to reciprocity, two problems arise: the rarity of its expression, and, consequently, the paucity of data in MC of SC during the relevant time slot: we have to consider data from the second half of the 19 th century when there were a lot of second language speakers after the abolition of slavery in 1835. Unfortunately, the paucity of examples does not allow making any decisive claims regarding a grammaticalization of kanmarad in this period: As to MC, it is impossible to draw general conclusions based on one isolated example used without a possessive determiner (see example (19)) from the 19 th century without confirmation by more recent examples from MC. As to SC, I stated (section 3.2.1) that Young's fables from around 1900, the only available text from Seychelles creole diachrony, do not contain examples that allow for a grammaticalized interpretation. However, this observation is not sufficient to make a safe claim that the construction was grammaticalized only during the 20 th century. Consequently, the validity of the Extra-Transparency Hypothesis cannot be verified as far as the grammaticalization of kanmarad in SC is concerned. However, I would like to retain it as a possible factor that could have contributed to trigger the grammaticalization of kanmarad. As our data base is much more solid for data starting from the 1970s, I would like to discuss another hypothesis which I call the Distinction during Codification Hypothesis.

The Distinction during Codification Hypothesis
SC has been promoted to the status of first national language in 1981, followed by English and French, heritage of the country's colonial past. Seselwa has coexisted with English, rather than French, since 1810 and the role of French today is marginal. The officialization of Creole, which had been an almost exclusively spoken language before, led to an acceleration of its codification (e. g. Bollée & Kriegel 2016). Today, Creole is the first language of alphabetization, the language of parliament and the main language in oral media. In this context, several morphosyntactic changes have been observed where speakers (and language planners) consciously or unconsciously choose the "déviance maximale", the maximal distinction from French, the lexifier language when they are confronted with variation. Journalists, writers, and academics dealing with literacy and codification seem to systematically opt for variants which are considered to be the most authentic, the most basilectal, the most creole. This has been shown for the complementizer pourdir (Kriegel 2004) and for the reflexive marker son lekor (Kriegel 1996). 22 This state of affairs can be formulated in what I call the Distinction during Codification Hypothesis: The Distinction during Codification Hypothesis If a language has to satisfy all communicative needs in all language registers (which is the case for a first national language), we can observe accelerated functionalizations and the stabilization of constructions formerly subject to great variation. In special sociolinguistic 22 The development of a frequently used passive form without restrictions on the verb semantics using ganny (< Fr. gagner -'to get, to win') as an auxiliary is essentially due to the need for more marked forms in literacy. Additionally, it can be interpreted as the choice of a form that is perceived as being maximally distinct from French. (Kriegel 1996) contexts, constructions not found in the lexifier have more chances to become stabilized than those that continue patterns of the lexifier.
The examples (6)-(8), drawn from two recent novels and from the newspaper Seychelles' Nation (2016), can be interpreted as illustrations of this hypothesis. Writers and journalists aware of the autonomy of Creole with respect to the French lexifier and wishing to stress this fact, use techniques associated with Creoleness and veer away from the French model. 23 This is certainly not the case for the examples (1), (3)-(5) from the oral spontaneous speech of elderly people. However, the fact that the kanmarad-technique is preferred to the "more French" (l)enn a lot-technique in the media and in recent literature is an argument in favor of the validity of the Distinction during Codification Hypothesis. This desire to mark a distinction may contribute to the further expansion and grammaticalization of the kanmaradtechnique.

Conclusion
I hope I have shown how the grammaticalization of a 'companion'-based reciprocal marker took place in modern SC by discussing several grammaticalization mechanisms and by comparing my data with data from earlier SC texts and from MC. While searching for an external explanation for the studied grammaticalization or, in Michaelis & Haspelmath's (2020) term, accelerated functionalization, I discussed the Extra-Transparency Hypothesis and the Distinction during Codification Hypothesis, two not mutually exclusive but complementary hypotheses. The Extra Transparency Hypothesis may have contributed to triggering semantic changes prior to functionalization during the second half of the 19 th century, and the Distinction during Codification Hypothesis to the grammaticalization and further expansion of the grammaticalized reciprocity marker kanmarad. However, I want to stress the fact that we are dealing with an ordinary language-internal grammaticalization process. Finally, the case study reported here provides evidence that creole languages, once they have stabilized, behave just like any other language.
Yani Shamir Maury, and Georges Daniel Véronique) who helped me with data elicitation and interpretation. Thanks to Annegret Bollée and to Susanne Michaelis for commenting on a first version of this text and to Michael Pretina and Pauline Welby for reading the text carefully. Last but not least I am grateful to Martin Haspelmath who awarded me a fellowship from the ERC grant "Grammatical universals" in Leipzig during which my interest in dealing with creole reciprocals emerged.