Clause chaining and switch-reference in Aikanã and Kwaza Encadeamento de cláusulas e referência alternada em Aikanã e Kwazá

: In Aikanã and Kwaza, neighbouring endangered isolate languages of Rondônia, Brazil, sentences can include chains of medial clauses and end with a predicate in a matrix sentence mood, such as declarative, interrogative etc. In Kwaza, traditional narratives may even consist of a single long string of medial clauses, terminated by a fixed formula in the declarative mood. In both languages, subject (dis)continuity is expressed by a switch-reference system that indicates on the predicate of the current clause whether the subject of the next clause will be different or not. In this descriptive article I present similarities and differences between the systems of switch-reference in Aikanã and Kwaza. Among the differences, I discuss cases of formally marked switch-reference that appears to express topic discontinuity rather than subject discontinuity, in Kwaza. Also, I analyse the unusual anticipatory nature of the Kwaza system. As a special feature of this article I include two versions of a traditional mythological narrative, originally told in Aikanã and later retold in Kwaza, to illustrate clause chaining and switch-reference through a coherent and culturally relevant text.


INTRODUCTION
Aikanã and Kwaza are two neighbouring isolate languages of southeastern Rondônia, Brazil, spoken by the Aikanã and Kwaza ethnic groups.Estimated in approximate figures, the Aikanã language has 250 speakers on a population of 400 and the Kwaza language has 25 speakers on a population of 45.Originally, the Aikanã and Kwaza inhabited adjacent territories and were neighbours to other indigenous groups, such as the Kanoé (isolate), Kepkiriwat (extinct Tupian, Tupari), Salamãy (Tupian, Mondé), and different "Mekens" (Tupian, Tupari) subgroups.They shared important aspects of culture, and traditional multilingualism must have led to diffusion of linguistic traits.First contacts with western culture probably occurred around the start of the 20 th century, and brought diseases, violence and diaspora upon the indigenous peoples of the region.In the course of the second half of the 20 th century, a big part of Rondônia was deforested and indigenous reserves were demarcated for what had remained of the indigenous groups.Nowadays, the Aikanã and Kwaza share two indigenous reserves, in both of which the Kwaza speakers form a minority.One of the reserves is also shared with speakers of the Latundê (Nambikwaran) language and with descendants of the Sabanê (Nambikwaran), Kanoé, and Salamãy ethnic groups.
The Aikanã language has been studied by linguists since the 1980s, and has been partially described by Hinton (1993), Vasconcelos (2002), Silva (2012), van der Voort & Birchall (in press) and several articles by van der Voort.The Kwaza language has been studied since the 1990s, and has been described by van der Voort (2004) in an extensive monograph and several articles.Both languages are genealogical isolates, as is evidenced by their lexical and grammatical properties.Nevertheless, when painted with a broad brush, they are typologically rather similar, both being polysynthetic, morphologically very complex suffixing languages.They furthermore share some rather specific grammatical traits with each other and with certain neighbouring languages, which may be due to many centuries of language contact.One of those traits concerns switch-reference systems.
Clause chaining refers, roughly speaking, to the possibility in many languages to express simultaneity or consecutivity of events, and explanations of purpose, causation, reason, etc., in the form of chains of subordinate dependent clauses (instead of allowing coordination of full sentences, as in English).Some languages have initial-consecutive chaining structures, in which the initial clause represents the main clause, whereas other languages have medial-final chaining, in which the final clause represents the main clause.Often, dependent clauses are limited with regard to certain inflectional or derivational possibilities.A rich body of literature exists on clause chaining.A useful overview is Longacre (2007Longacre ( [1985]]) and an important recent collection of articles on different kinds of clause linkage is Bril (2010).
The term switch-reference, which was coined by Jacobsen (1967), refers broadly speaking to the phenomenon that languages may indicate by grammatical marking whether the subject or agent of a subordinate clause is identical with or different from the subject or agent of the main clause.In languages that allow clause chaining, 'same subject' and 'different subject' marking may occur repeatedly throughout the chain.Switch-reference is sometimes also put to pragmatic uses.In that case, the 'different subject' marker does not necessarily indicate a different subject, but a conspicuously different topic or different event.Switch-reference systems have been described for many languages in different parts of the world, and are especially well-known for languages of New Guinea, Australia and the Americas.Languages with medial clause chaining and switch-reference systems tend to be suffixing, head-marking, pro-drop languages and tend to display Subject-Object-Verb clause constituent order (Stirling, 2005).Important collections of articles on switch-reference systems in diverse languages are Haiman & Munro (1983) and Gijn & Hammond (2016).
In the following subsections I will describe and illustrate the switch-reference systems and clause chaining properties of Aikanã and Kwaza.In the concluding section, I will discuss the similarities and differences between these systems.The linguistic data are from my personal fieldwork with native speakers from 1995 to 2021.

COMPLEX SENTENCES AND SWITCH-REFERENCE IN AIKANÃ
A complete sentence in Aikanã consists minimally of a verb root, inflectional person marking and inflectional mood marking.Other morphology is derivational and can be optionally inserted between the verb root and inflectional morphemes.It includes classifiers, directionals, valency morphemes, aspect and modality morphemes.Non-future tense is unmarked, whereas future tense has both derivational and inflectional properties, involving recursion of person marking (Voort, 2013(Voort, , 2016)).Person markers are usually suffixes, but there is a limited set of verbs that may occur with an alternative prefix paradigm.In addition to paradigms for canonical subject marking there are paradigms for causative, reflexive, patientive and beneficiary arguments (e.g., Voort, 2013;Voort & Birchall, in press).Mood markers are verb final, and include declarative, interrogative, imperative and several other matrix sentence moods.In addition, Aikanã has a small set of mood markers used for adverbial subordinate clauses: conditional, concessive, and reason.
Adverbial clauses tend to precede the main clause, and together they form a complex sentence, as the following conditional (1) and concessive ( 2) examples illustrate: (1) sing-nmlz-3sg.refl-cond elder angry-nmlz go-dir:close-3pl.pat-decl' Always if one would organise a party, the old-timers would go there.' (2) hisa hideka-a-dupa detya hane pau~pau-pa-ẽ I strong-1sg.pat-concwoman water run~run-clf:fruit-decl 'Even though I am strong, my wife fetches the water.' In addition to subordinate clauses with adverbial semantic content, Aikanã has medial clauses: subordinate clauses the semantic content of which is coordinate with that of the subsequent clause.The mood value of Aikanã medial clauses is identical to that of the final main clause, whereas person reference may be different.There are two medial clausefinal markers: -pü 'same subject', which indicates that the subject of the next clause will be the same, and -na 'different subject', which indicates that the subject of the next clause will be different.The contrasts between the following basic examples demonstrate how the system works: (3) João ware-dua-pü dara-ne düni-ẽ João arrive-dir:in-ss hammock-loc lie-decl 'João entered (the house) and lay down in the hammock.' (4) João ware-du-na   Maria düni-ẽ João arrive-dir:in-ds Maria lie-decl 'João entered (the house) and Maria lay down.' (5) hisa Maurício-xa-pü tipa-xa-isuwã- Aikanã medial clauses can be coordinated with one another, and thus form a chain that ends with a main clause in one of the sentential moods.Traditional narratives consist of complex sentences that contain long chains of medial clauses, which indicate continuity or discontinuity of the subject by the -pü and -na suffixes.The following short traditional narrative is an example of the high productivity of the system, as well as its complexities.All lines demonstrate the basic characteristics this system, i.e., same subject and different subject morphemes in medial clause chains.Certain particular traits are evaluated at the end of the text.(a1) kureza-deri detya-deri-ye bu~bu-ke-pe-na man-3.posswoman-3.poss-accbeat~beat-3.ag-clf:round-ds(a2) wãwã 'ĩ-deri-'ene-'ete xa-na-xüne-re-'ẽ   küine-'ete-ne he-pü child-3.poss-coll-all1pl-dir:hither-dir:return-fut-imp 1 old.garden-all-locsay-ss 'she said to her children "Let's leave and return to our old farmstead!",' (a3) xü-pa-ne-'ee-pü kya-pü küine keza dir:return-tr-pfv-3pl.pat-ssthen-ss old.garden house 1 The occurrence of bound morphology without a lexical root is highly context dependent, and is discussed in Voort & Birchall (in press).

(a7)
dupa-na iza-yiye penene-ne dupa-na cntrx-ds far-again afternoon-loc cntrx-ds 'however, some time later, it was getting really late in the afternoon, then,' 'the child saw it and was about to cry,' 2 That is a highly unusual thing to happen, and therefore considered an omen of something serious. 3Recursive person marking in Aikanã future tense constructions and Kwaza quotation constructions is discussed in, among other places, Voort (2013Voort ( , 2016)).

paw=hika-paipa-'i-ye katene xa-ne-ta-xa-'i-'ete-ne
run=leave-dir:ahead-nmlz-acc there 1pl-pfv-rem.fut-1pl-nmlz-all-lockaza-i i'iwe he-pü be-int jaguar say-ss '"Now look, where the jaguar is going to run ahead, is it in the same direction where we are going home?", she wondered,' In some sentences, (a10-a11), (a15-a16) and (a21), the same subject marker -pü occurs in spite of a change of subject in the next clause.On the other hand, whenever the marker -na occurs, it does systematically indicate a different grammatical subject in the next clause.This can perhaps be accounted for by a difference in semantic markedness.Whereas different subject marking in some languages can also signal a different or unexpected event instead of a different subject (e.g.Gijn, 2012), same subject marking in Aikanã might also signal that the next episode is part of an expected course of events, whether the subject is the same or not.This possible explanation of non-default use of same subject marking in Aikanã cannot be confirmed, however, on the basis of the few examples mentioned here.

COMPLEX SENTENCES AND SWITCH-REFERENCE IN KWAZA
A complete sentence in Kwaza consists minimally of a verb root, and inflectional person and mood suffixes, in that order.Any derivational morphology should occur between the root and the person marker (third person is unmarked).It includes classifiers, directionals, valency morphemes, tense, aspect and modality morphemes.Some modality morphemes may originate from degrammaticalised mood inflections and require recursive person marking.Some expressions of remote past tense and of habitual aspect involve morpheme-based reduplication of person markers (Voort, 2003(Voort, , 2009(Voort, , 2016)).Mood markers are verb final, and include declarative, interrogative, imperative, volitive and several other main clause moods.Furthermore, Kwaza has a set of mood markers used for adverbial subordinate clauses: conditional, concessive, and manner.
Adverbial clauses tend to precede the main clause, and together they form a complex sentence, as examples ( 7) and (8) illustrate: Kwaza also has medial clauses.These are subordinate clauses ending in a dependent clause marker that represents the semantic content of the main clause mood marker 4 , as in examples ( 9) and ( 10). 4 And can therefore perhaps be regarded as a dummy mood marker.
There are two medial clause markers: -ta 'dependent', which is neutral with regard to person, and -si 'switch reference', which indicates the exclusion of a specific person value of the main clause subject.The suffix -si anticipates that the subject of the next clause will either be a non-third person or a third person, depending on whether the reverse is the case in the medial clause, as in examples ( 11) and ( 12), unless both subjects are different non-third persons, as illustrated by ( 13).Multiple medial clauses of Kwaza can be juxtaposed, forming a chain that ends with a main clause in one of the sentential moods.A narrative can consist of a single long chain of medial clauses, which contain suffixes that indicate continuity or discontinuity of the subject.The following short traditional story was retold in Kwaza from the Aikanã story of the previous section.It does not contain cases of non-third person switch-reference, but same-reference and third person different subject are well represented, as well as the impressively persistent clause-chain.The above story consists in fact of one single long chain of medial clauses, which ends in a main clause mood only at the formulaic expression in (k26), and which represents the main clause of the entire preceding subordinated chain that started with sentence (k1).One may wonder about the main clause moods that are encountered inside this narrative.However, these represent direct speech quotations in different moods, such as the exhortative in (k2) and (k22), interrogative and declarative in (k6), (k20) and (k25), negative imperative in (k11), and others.They are embedded in the clause chain, and may contain subordinate and medial clause chains themselves.
Throughout the chain, the subject of a clause may differ from the subject of the next clause, which is indicated by the different subject morpheme -dɨ-.Also clause chains that are embedded inside quotations may contain internal reference switches, as in (k20-k21).In some some cases, however, -dɨ-should be interpreted not as signalling a different subject, but as announcing a different event or topic, as in sentences (k17) and (k22).
Note that, in addition to the dedicated different subject morpheme -dɨ-, the concessive clitic wara 'but' also has the effect of announcing a different subject, as in (k5), (k10) and embedded in (k25).
Note furthermore that the functional difference between dependent clause marker -ta and adverbial clause marker -tya 'manner, background' can be unclear.At a cursory glance, in same subject third person clauses they both seem to represent medial clause markers.However, unlike -ta, the morpheme -tya is never preceded by person marking or the different subject morpheme -dɨ-, is never instrumental in clause chaining and is independent of reference switches.6CONCLUSION Both Aikanã and Kwaza have very productive medial clause chaining constructions involving switch-reference systems.Certain similarities can be perceived in their chaining constructions and in their switch-reference systems.Table 1 lists typological similarities and differences between Aikanã and Kwaza switch-reference systems attested in this article.

mε͂ -tyaryɨ ̃-ta-nãi-ko-tya
ta-ki beat-much-1sg.obj-nmlz-inst-mantalk-decl 'I went to live together (there in the old house), but the jaguar nearly grabbed us, because of (your) beating me always", she said to him.'Table 1.Similarities and differences between Aikanã and Kwaza switch-reference systems.-forms: Aikanã (-pü 'ss'; -na 'ds') vs. Kwaza (-ta 'ss'; -dɨ-/-si 'ds') Some similarities are rather common for medial clause chaining languages, such as the fact that the final clause tends to occur at the very end of the chain, that clause chains can be embedded at a separate structural level (as seen in quoted utterances in Aikanã and Kwaza), and that the reference switch is announced by the preceding clause for the next clause (Longacre, 2007(Longacre, [1985]]).

Similarities
In both Aikanã and Kwaza the distinction between same subject and different subject is made by dedicated morphological suffixes.Also, the medial clause is inflected for person, and does not resemble the "stripped-down structure" that Longacre (2007Longacre ( [1985]], p. 412) observed in other clause chaining languages.Even in the case of nonthird persons, where person marking would leave no doubt as to the distinctness of the subjects, subject discontinuity is marked overtly and in some situations quite redundantly, as in Aikanã example (16) and Kwaza example (17): Some of the characteristics of the switch-reference systems shared by Aikanã and Kwaza could be due to long term contact between the languages.Their speakers have been direct neighbours since time immemorial, they share the same cultural area, and the languages also show other traits of mutual interference (e.g.Crevels & Voort, 2008;Voort, 2005Voort, , 2016)).
There are also some notable differences between Aikanã and Kwaza clause chaining and switch-reference.As seen in the above Aikanã text, as well as in other Aikanã texts (e.g.Birchall et al., 2017), the clause chain is sometimes 'interrupted' by sentential mood clauses that are not embedded.Such interruptions are conspicuously absent in the above Kwaza text, as well as in other Kwaza texts (e.g.Voort, 2004 andVoort et al., 2017, which also contain several cases of non-third person switch-reference).
Furthermore, in this and other texts the Aikanã different subject marker seems to function exclusively to signal a reference switch.In Kwaza, however, subject discontinuity marking is not the only function of the switch-reference system.This is confirmed by examples where the subject is identical across a clause chain, but nonetheless bears different subject marking: Apparently, as seen in examples ( 18), (k17-k18) and (k22-k23), pragmatic functions such as a new topic are signalled by switch-reference too.In his overview of similar phenomena widespread in South American languages, Gijn (2012) refers to non-'default' use of switch-reference systems as 'switch-attention', and proposes that it could be subject to diffusion through contact.Between the neighbouring languages Aikanã and Kwaza, however, switch-attention does not seem to have spread, as the data presented here suggest.It is nevertheless interesting that some Aikanã examples, (a10-a11), (a15-a16) and (a21), showed an opposite phenomenon, in which same subject marking has been attested in spite of the next subject being different.Note that Aikanã-Kwaza bilingualism is exclusively found with native speakers of Kwaza today.Even though there are some notable structural similarities between both isolate languages, we do not know how common such bilingualism was in traditional times.
Another significant difference with Aikanã is that the switch-reference system of Kwaza is anticipatory.Apart from announcing that the subject of the next clause will be different from the current clause or not, it also signals whether the person value of the subject of the next clause will be different or not.Thus, if the subjects of two chained clauses are different, but their person value is the same-which is obviously only possible if both are third persons-the preceding clause is marked by the derivational suffix -dɨ-.This suffix indicates that the different subject of the next clause is a third person.However, when one or both subjects in a biclausal chain are not third persons, the first clause is marked by the inflectional suffix -si.In effect, the different subject marking system in Kwaza anticipates (by blocking the possibility of) a specific person value.Table 2 demonstrates the difference between the marking systems in Aikanã and Kwaza.Anticipatory subject agreement, where the preceding clause actually identifies the subject of the next clause, is known from some Papuan languages, such as Kanite and Hua (Roberts, 1997, p. 129).Aikhenvald (2012, p. 342) suggests that Kwaza is possibly unique among South American languages in this respect.7However, she also points out that some Macro-Jê languages, such as Maxakali, display anticipatory switch-reference.Because this is marked on conjunctional particles, she considers this different from switch-reference in clause chaining.Nevertheless, the Kĩsêdjê (Jê) language of Brazil also shows anticipatory switch-reference on its conjunctional particles or clitics (meaning 'and'), and in doing so, it can form extended clause chains, as Nonato (2014) and Nonato et al. (2017) demonstrate.
Finally, as Table 2 shows, the forms used by Aikanã and Kwaza for distinguishing same subject and different subject are different and do not seem to be etymologically related.In this respect, a similarity between the Kwaza and Mamaindê (Nambikwaran) forms is noteworthy, as Gijn (2016) points out.Although in Mamaindê (Eberhard, 2009) switch-reference is marked by suffixes on conjunctional particles, and although other important differences exist in the system, it involves the same reference marker -ta̰ ʔ and the different reference marker -sḭ̃ ʔ.Because other Nambikwaran languages do not seem to have a switch-reference system, van Gijn suggests that Mamaindê could have developed it under influence from Kwaza.The formal and functional similarities are striking, to say the least.

A NOTE ON THE ORTHOGRAPHIES USED
The orthographies used for the Aikanã and Kwaza language data are partially phonemic and partially in accordance with standard practical orthographies that were established for the languages.With regard to the Aikanã data, <'> represents IPA

ABBREVIATIONS
The abbreviations used in this article sometimes differ from those used in previous work on Aikanã and Kwaza.
huge debt to my friends and patient teachers Raimunda Aikanã and Mario Aikanã.I furthermore want to thank the communities of the Tubarão-Latundê and Kwazá do Rio São Pedro indigenous reserves for their hospitality, as well as the Fundação Nacional do Índio (FUNAI) for the necessary authorisations.I thank Rik van Gijn and two anonymous peer reviewers for their comments, which considerably improved the article, and I thank the guest editors of this special Boletim issue, Suzi Lima and Tonjes Veenstra, for inviting me.The research involved was generously financed by the Netherlands Organisation of Scientific Research (NWO: TSL 300-72-021, VIDI 276-70-005) and the VolkswagenStiftung (DoBeS 85611, 92740).All errors are mine only.