Clause-linkage, Embeddedness, and Nominalizations in Chácobo (Pano)

Abstract

As with all Pano languages, Chácobo links clauses together through an elaborate system of switch reference clauses. This paper provides a detailed description of switch reference and clause linkage in Chácobo (Pano) from a typological perspective. While previous work on Chácobo and Pano languages in general describes such clause-linkage strategies as involving subordination, no work has provided a detailed description of the diagnostics for classifying clause-linkage types with clause-linkage strategies in Pano. If these variables are relied on, nearly all clause-linkage strategies in Chácobo fall outside of typical coordination and subordinate patterns. There is also little reason to adopt such a distinction on language-internal grounds.

1. Introduction

Pano languages have highly complex systems of same/different subject marking. Same/different subject clauses are described as subordinate in many grammars of Pano languages (Fleck 2003, p. 1001; Valenzuela 2003, p. 413; Tallman 2018a, p. 317; Camargo Souza 2020). However, a detailed investigation of such clauses in terms of the criteria typically used to distinguish coordinate and subordinate clauses has not been conducted. Neely (2019, p. 434) claims that the relative coordinate or subordinate status of such clauses requires more research. Same/different subject clauses in Pano languages appear to be, in very general terms, structurally and functionally similar across Pano languages. Such clauses are marked for whether their subject is co-referential or obligatorily not co-referential with the subject of the main clause. Same-subject clauses also display “transitivity harmony” (Valenzuela 2005, 2013). They code whether the subject of the main clause is an A (subject of a transitive) or an S (subject of an intransitive) argument.

Whether “switch reference” clauses are described as subordinate or coordinate in the linguistic literature can partially depend on theoretical considerations. Finer (1985) seems to assume that all switch reference clauses are subordinate, and Roberts (1988) argues that switch reference clauses are coordinate based on a number of diagnostics (see Keine 2013 as well). More recent literature has claimed that some switch reference clauses are subordinate and others are coordinate (Stirling 1993; McKenzie 2015), while others have advocated for a third category or some subtype of coordination (Weisser 2012). In McKenzie’s survey of switch reference in North America, he argues that the debate about whether switch reference is coordinate or subordinate is “moot” because “SR is North America occurs with all types of clause connectives” (McKenzie 2015, p. 429). In other words, whether switch reference is subordinate or coordinate is a matter of typological variation (see the work of Baker and Souza 2019, for a recent overview).

Such perspectives assume that a discrete distinction between “coordinate” and “subordinate” clauses, borrowed from traditional grammar, is necessarily theoretically valid. They assume that there is an a priori distinction between coordinate and subordinate clauses (perhaps as a matter of language design) and it is simply a matter of picking the right set of distinguishing features that home in on the ideal type. Functional–typological literature, applying wider array of diagnostics more consistently, has suggested that there is a continuum between subordination and coordination types (Haiman and Thompson 1984; Lehmann 1988; Foley and Van Valin 1984; Van Valin 1993; Croft 2001; Cristofaro 2003). From this perspective, a linked-clause construction is coordinate or subordinate to some degree. The question arises as to whether actual typological patterns organize themselves into prototypes (Bickel 2010), perhaps due to functional and diachronic “attactor points” (Hawkins 2004; Bybee and Beckner 2015; Schmidtke-Bode 2019). In order to investigate clause linkage from such a perspective, detailed descriptive works are necessary, which apply a methodology that does not reify or presuppose candidate attractor points.

As stated above, in most works on Pano linguistics, same-subject clauses are described as “subordinate”. Evidence for this in Chácobo may come from the fact that an interrogative constituent can be asymmetrically extracted from the same-subject clause as in (1) below and such asymmetric extraction is typically regarded as evidence of subordinate status (Ross 1967). We can also see that the same-subject clause yonoko=só “work before V” is center-embedded, potentially yet another piece of evidence for subordinate status (Bickel 2010).

1.	hɨnawa=ʂói	tsi	yonoko=ʂó	ti	ina	taʃi= ́	tɨpas=ʔá
	how=sa	lnk	work=prior:sa	t	dog	Tashi=erg	kill=inter:pst
	‘How, after working, did Tashi kill the dog? (‘How did Tashi work after killing the dog?’)

Not all data point to a subordinate status for this construction in Chácobo, however. First, note that main clauses can also be center-embedded. An interrogative constituent can extract from a post-posed same subject clause (producing a sentence which is difficult to translate), which is generally unavailable to adjunct subordination (Bošković 2020).

2.	hawɨi	kako	ho=ʔá	ti	kopi=ʔáʂna
	how=sa	lnk	work=inter:pst	t	buy=prior:ss
	‘What, after Caco arrived, did he buy?’

Thus, center-embedding may not apply in this case, because it also suggests that the main clause is subordinate (Weisser 2015 on problems in interpreting such diagnostics). Illocutionary scope suggests that same-subject clauses in Chácobo may be a coordinated structure (Jendraschek and Shim 2018, inter alia). Subordinate clauses will typically be presupposed information, but in Chácobo, an interrogative illocutionary marker can scope over each predicate, suggesting a more coordinate-like structure.

3.	tʃaʃo	pi=ʔi	tsi	hɨɾɨ=yá	tʃani=ka(n)=ʔá
	pig	eat=concur:ss	lnk	Gere=com	speak=3pl=inter:pst
	‘Were they eating and were they speaking with Gere?’, ‘While they were eating, did they speak with Gere?’

Furthermore, we also expect subordinate clauses to be de-ranked compared to main clauses, displaying less tense-aspect-modal contrasts, for instance. While there are some limitations in marking, overall same-subject clauses display most of the same marking as main clauses, suggesting a relatively higher coordinate status. Nor are such cases of mismatch rare (Bickel 2010; Weisser 2012, 2015; Jendraschek and Shim 2018).

One approach to this apparent ambiguity is to discard the conflicting data.1 We could choose one criterion (e.g., “extraction”) and discard the others as irrelevant to the assessment of that particular construction in Chácobo, changing which criteria are relevant or irrelevant depending on the language and classifying each construction based on whatever diagnostics give us the results that conform to our preferred theoretical position (see Hofmeister and Sag 2010 for a relevant discussion on islands). However, this approach has been criticized as methodologically biased (Croft 2001) and is foreign to the methods in all mature sciences (Mayo 2018; Tallman 2021a, among others). If the distinction between subordination and coordination is taken as a grammatical primitive or the distinction represents some sort of substantive universal, explicit conditions need to be stated for its falsification. However, positing that it is appropriate to discard conflicting evidence in order to maintain a desired hypothesis at best makes claims about the universality of the distinction confirmationally lax, and, at worst, immunizes such a claim against falsification, making it a tautology: a coordinate–subordinate distinction can be recognized because diagnostics exist and can be cherry-picked to rationalize the distinction; however, the linguist sees fit. To assume that because a distinction is used in descriptive work, it must reflect a distinction which manifests substantive universals, is to lift a heuristic methodological unit into a theoretical postulate without justification. And to insist that the distinction is a well-tested hypothesis (and not a metaphysical prejudice) while maintaining that its falsification is in principle impossible is to seriously misunderstand scientific method (see Ozerov 2018 for a discussion of similar problems with the categories topic and focus categories, and Tallman 2020, n.d. on the notion of ‘word’, ω and X0).

From a typological perspective, allowing the definition of coordination and subordination to vary leads to problems for linguistic comparison. It is not clear that one linguist’s “subordination” will correspond to the next’s, if linguists are choosing criteria inconsistently. Assuming a distinction without providing a fixed and consistent empirically operationalized definition applied rigorously from one language to another will result in non-commensurability between language descriptions and hinder our ability to make verifiable and robust cross-linguistic generalizations. One solution to this problem would be to propose a fixed definition by fiat (a “comparative concept” or “retrodefinition”) defining coordination or subordination based on a single criterion so that the concept at least as mneumonic value (Haspelmath 2010, 2018). This perspective would preserve the traditional terminology without making claims about its usefulness in accounting for constraints on cross-linguistic variation, apart for making it clearer what researchers mean by the terms.

In this paper, I take a different approach, inspired specifically by Bickel’s (2010) multivariate approach to clause linkage, but more generally by work on polythetic classification in the biological sciences and other fields (Sokal and Sneath 1963; Needham 1972; Ellen 2008; Parnas 2015). Polythetic classification refers to classification in the absence of necessary and sufficient criteria for the relevant classes. In a systematic review of the diagnostics that distinguish between coordination and subordination, Bickel (2010) deconstructs the properties that have been posited as diagnostics to distinguish between coordination, subordination and/or co-subordination into a typological variable.2 While cluster methods show that there are perhaps subordinate and coordinate prototypes, the typological variation in clause linkage swamps the simple classifications used in general linguistics. In this approach, an interesting question arises as to whether there is some “statistical order” to the patterns: there are no jointly sufficient and necessary conditions for distinguishing between coordination and subordination, but perhaps cross-linguistically and in a given language, the relevant diagnostics cluster into two groups better than would be expected if such a distinction was not relevant. The distinction between coordinate and subordinate is seen as a latent variable responsible for correlations between test results. I apply this perspective to the description and analysis of clause-linkage clauses in Chácobo.

This paper also provides the first detailed description of clause linkage in Chácobo (Pano). I show that the majority of Chácobo clause-linkage constructions (which includes all “switch-reference” clauses) are neither subordinate nor coordinate. I make this argument in the first place, by considering how Chácobo clause-linkage constructions pattern with respect to a broader typological sample, showing that they fall into neither candidate subordinate nor into candidate coordinate “prototypes”, but simply occupy a liminal middle ground (Weisser 2015; Jendraschek and Shim 2018). I also make this argument language-internally, based on a wider variety of more fine-grained features than Bickel (2010). On language-internal grounds, the clause-linkage constructions of Chácobo do not cluster into two groups much better than chance with some differences arising depending on how the variables are aggregated. They do vary substantially from one another on language-internal grounds, but characterizing this variation in terms of coordination or subordination is misleading. I make this point with hierarchical clustering models coupled with simulation methods.

Section 2 provides language background on Chácobo. Section 3 describes the dependent clauses of Chácobo. Section 4 provides a description of the clause-linkage variables in relation to the clause-linkage constructions of Chácobo. This section contains some revisions of Bickel’s criteria. Such revisions are to be expected in an autotypology approach (Bickel and Nichols 2002). Tests related to interrogative constituents need to be broken down further in Chácobo. Pano languages also display a type of “clause-skipping” in their agreement patterns that could be rallied as a diagnostic as well, since it plausibly related to Bickel’s “layer of attachment”. Section 5 and Section 6 are concerned with assessing the degree to which a coordination–subordination distinction is motivated in Chácobo. I argue that it is not, based on two types of arguments: (i) one relying on the relative closeness of Chácobo clause-linkage strategies to candidate “prototype” subordinate and coordinate constructions; (ii) another based on whether there is evidence for language-internal clustering into two types of clause-linkage strategies. Section 7 provides some concluding remarks and discusses future research and problems with the application and comparability of some of the diagnostics.

2. Chácobo Language Background

Chácobo is a southern Pano language of the northern Bolivian Amazon. The language is spoken by approximately 1500 people. It is spoken in the town of Riberalta (Beni, Vaca Diez), and villages on or close to the Geneshuaya, Ivon, Benicito, and Yata rivers. The largest Chácobo village is Alto Ivon with about 500 inhabitants (and growing). Chácobo is still learnt as a first language by children in the villages. Typically, children who grow up in Riberalta do not learn to speak the language, perhaps acquiring a passive knowledge of it.

Chácobo has a relatively simple segmental inventory with four vowels (i, o, a, ɨ) and 14 consonants (p, t, k, β, ts, tʃ, n, m, s, ʃ, ʂ, ʔ, h, j, w). Syllable structure is (C)V(C). All consonants can occur in the initial position, but only sibilants can occur in the coda position. In some dialects of Chácobo, the glottal fricative /h/ can occur in coda position, but the number of forms with the coda /h/ in the lexicon is relatively small.

Chácobo can be described as a tonal language in the sense that lexical items are distinguished by consistent indications of pitch (Hyman 2006).3 Lexical items in Chácobo have their syllables specified as either toneless or LH. The H has a relatively higher pitch on the syllable the LH is docked to. Throughout, I will mark a lexical LH with an acute accent. The timing of the L depends on morphosyntactic context. Within lexical items or highly frequent sequences of lexical items L is realized on the prior syllable. At less frequent junctures, the L is realized during on the syllable it is specified for. In other words, a form such as kamáno “jaguar” with the LH on the second syllable will be realized as [kàmáno] “jaguar”. A form such as honi “man” with the LH on the first syllable will be realized as [hǒni] with a contour tone on the first syllable. A tone reduction rule in Chácobo deletes an H if it occurs left-adjacent to a lexical LH (LHLH → LLH). The rule applies obligatorily, optionally, or not at all depending on context (Tallman 2018a; Tallman and Elías-Ulloa 2020). Chácobo also has grammatical (ergative, genitive, spatial) floating LH tones and morphemes which condition the appearance of an LH tone on an element to their left in certain circumstances. For instance, the adjectivalizer = ́ʂɨni has a floating LH to its left which docks to the final syllable of the element the morpheme combines with: tsaya “look” becomes tsayá=ʂɨni “a looker”.

While I refer to LH as a lexical tone, it should be pointed out that, in this context, I mean “lexical” as in lexical item, not an element that has lexical content. Nor should lexical here be taken to imply that tone sandhi rules that affect LH occur inside a lexical (as opposed to post-lexical) phonology. It simply means that morphemes in Chácobo are listed as having LH tones docked on certain syllables.

Chácobo is dependent-marking: it codes grammatical relations with case on noun phrases. Case marking on noun phrases display an ergative alignment. The ergative case is marked with a floating LH tone which falls on the final syllable of an A noun phrase. The grammatical relations S and P are unmarked in full nouns. Pronouns display a nominative–accusative alignment, however.

All clauses in Chácobo come with a clause-type morpheme, which codes clause-type (declarative, interrogative, imperative reportative) and other categories such as tense depending on the marker. There are two main types of clauses: verbal and nonverbal predicate. These can be distinguished according to four properties: (i) the clause-type marker; (ii) ergative marking on full noun phrases; (iii) the order of predicate and A/S role (“subject”); (iv) the part of the speech category of the predicate. Table 1 summarizes the differences between the different clause types. There is an additional mixed clause type, which combines properties of verbal and non-verbal predicates. In the work of Tallman (2018a), this is referred to as a c-subj verbal predicate construction, because the A/S subject must occur after the clause-type morpheme.

Table 1. Main clause types in Chácobo (Pano).

	Verbal	Nonverbal	“Mixed” cl-Subject
Clause-type markers	=kɨ “declarative, past” =ʔá “interrogative, past” =wɨ́ “imperative” =pá “imperative (mirative)” =páima “possibility”	ʂo “declarative” ní “interrogative” ki “declarative, future” kiá “reportative”	=kɨ “declarative, anterior” =ʔi ní “interrogative” =ki “declarative, non-past” =ʔi kiá “reportative”
Ergative	yes	no	no
Order	subject–predicate	predicate–subject	predicate–subject
predicate category	verb	adjective, noun, adverbial, stative verb, post-positional phrase	verb

Examples in (4), (5), and (6) below illustrate the verbal, non-verbal, and mixed predicate constructions in Chácobo.

4.	kamáno= ́	hóni	á(k)=kɨ
	jaguar=erg	man	kill=decl:pst
	“The jaguar killed the man.”

5.	hóni	ʂo	tóa
	man	decl	dem:dist
	“That is the/a man.”

6.	áshi=kɨ	hóni
	bathe=decl:ant/perf	man
	“The man has/had bathed.”

Salanova and Tallman (2020) suggest that the mixed construction is a non-verbal predicate construction with an embedded verbal predicate. Apart from the properties listed in Table 1, which it has in common with non-verbal predicate constructions, evidence for this comes from the fact that two of the clause-type markers of the mixed constructions contain material found in dependent clauses: =ʔi is a concurrent same subject clause marker (see Section 3 below). For the purposes of this paper, I treat mixed constructions as verbal predicate constructions. The reason for this is that, in contrast to the predictions of Salanova and Tallman (2020), transitivity agreement between the same/different subject clauses and mixed main clauses treat such constructions as verbal predicate constructions.

Chácobo verbal predicates can be modified by many temporal, aspectual, modal, and evidential categories including “lexically heavy” categories such as associated motion. Chácobo verbal predicates are coded obligatorily for temporal distance (or “graded tense”) for which there are six overtly expressed categories: =ní “remote past”, =yamɨ́t “distant past”, =ʔitá “recent past”, =yá “recent past (perfect, mirative)”, =tsi~=tsa “immediate present/past”, =ʃaɾí “tomorrow”, =ʂɨ́ ‘remote future (Tallman and Stout 2016). The language also has a highly elaborate associated motion (AM) system. The associated motion markers display suppletive allomorphy depending on the transitivity of the verb they combine with and the number of its S/A subject: (i) =kaná~=βoná “going”, (ii) =honá~=βiná “coming”, (iii) =kayá~=βayá “do and go”, (iv) =kiria~=βiria “do and come”, (v) =kó~=boʔó “do and go (distributed)”, (vi) =koná~=boʔoná “go, do and come”, among others (Tallman 2020). These facts should be kept in mind when discussing whether a given dependent clause is “finite” or not: it is unclear what exactly finiteness means in the context of Chácobo verb structure as it is unclear which of the aforementioned categories should be considered inflectional and which not.4 In this paper, I assume that the potential expression of associated motion can be considered part of the relative finiteness of a clause.

The data for this paper come from approximately 32 months of fieldwork and an annotated corpus of about 28 h, transcribed and translated in ELAN. Data from naturalistic speech are supplemented with data from elicitation. Data from elicitation come from Caco Moreno and were double-checked with Miguel Chávez. Some of the extraction data could only be verified with one speaker, however, and are thus not necessarily as reliable. Part of the corpus for these data is documented with ELAR (Tallman 2018b).

3. Dependent Clauses

All dependent clauses in Chácobo can be usefully divided into four types depending on how they constrain subject A/S coreference. Same-subject (glossed ss or sa) clauses have A/S subjects which is coreferential with the S/A subject of the main clause. Different-subject (glosses ds/a) clauses have an S/A subject which is not coreferential with that of the matrix clause. Noun-modifying clauses (nmd) and nominalized clauses (nmlz) are unspecified with respect to whether their subject is coreferential with that of main clauses. Note that noun-modifying and nominalized clauses can take on an adverbial function.

Same and different-subject clauses vary in terms of the temporal relation they have with the main clause (‘Temporal relation” in the table below). Some dependent clauses alternatively function to modify noun phrases (“Noun-modifying”) and some can function as arguments of verbs (“Referential function”). An overview of the clause-type morphemes is provided in Table 2.

Table 2. Dependent clauses in Chácobo.

Form	Relation to Main Clause	Temporal Relation	Noun-Modifying	Referential Function
ʔáʂ(na)	same S	prior	no	no
ʂó(na)	same A	prior	no	no
ʔi(na)	same S	concurrent	no	no
kí(na)	same A	concurrent	no	no
́noʂpaɾí	same A/SNVP	subsequent	no	no
pama(ʔáʂ)	same S	interrupted event	no	no
pama(ʂó)	same A	interrupted event	no	no
(asyndetic)	same S/A	planned succession	no	no
kɨ́(no)	different S/A	prior	no	no
́no	different S/A	concurrent	no	no
́noʂparíno	different S/A	subsequent	no	no
ʔá(ka(to)/na)	none	anterior	yes	no
ʔái(ka(to)/na)	none	imperfective	yes	yes (agentive)
tí	none	purpose	no	yes (instrumental)

None of the clause-linkage constructions are dedicated complementation constructions insofar as complementation is defined in terms of core arguments of the main verbs. However, the agentive nominalized clause can take on this function: it can function as a clausal argument of the verb, even though this is not very common in natural speech (Tallman, Forthcoming). This is important because Bickel (2010) claimed to only code clauses which were plausibly of an adjunct status. All clauses of Chácobo have such a status, or at least could be analyzed as such. The only caveat is that there is one clause-linkage strategy which can take on a complementation function (those clauses marked with =ʔái(na) “agentive nominalizer”).

Note that some of the markers have phonologically short and long allomorphs. The short forms appear when the dependent clause occurs before the clause-type morpheme of the main clause. The long form occurs when the dependent clause occurs after the clause-type morpheme of the main clause. For instance, the short forms of the prior same-subject markers =ʔaʂ(na)~=ʂó(na) occur when the dependent clause occurs before the clause-type markers as in 7 and 8. Examples of the long forms are found in 35 and 36 (Section 4.1). These examples also illustrate that same subject-clauses code the transitivity of the main clause. This is called inter-clausal participant agreement in Pano linguistics (Valenzuela 2005).5

7.	hawɨ́	poko	pi=ʂó	tsi	no	ima=ní=kɨ
	3sg:gen	intestine	eat=prior:sa	lnk	1pl	roast=rempst=decl:pst
	“After eating his intestines, we roasted it.” 0027:004

8	paʔití	nima=ʔáʂ	tsi	kiá	áʃiná= ́
	jug	put=prior:ss	lnk	report	Ashina=erg
	kí-tʃa=ní=kɨ
	leg-open=rempst=decl:pst
	“Ashina put down the jug and opened her legs (over it).” 0818:0003

Same-subject clauses can also be distinguished according to the temporal relation they code. The examples in 7 and 8 above encode that the event of the dependent clause is prior to that of the main clause. The morphemes =ʔí(na) and =kí(na) encode an event which is concurrent or subsequent to the event of the main clause. Examples are provided in 9 and 10 below.

9.	hátsi	ʂokóβa	ʃita=kí	tʃoʃ-a=kɨ́
	then	children	cross=concur:ss	step.on=tr-prior:ds/a
	tsi	ratɨ=ʔi	kiá	hóni
	lnk	be.scared=c	report	man
	“Then when the children crossed (the patio), they would step on (near his penis), and the man was scared.” 0804:0038

10.	hátsi	kama	síɾi	hiá=ɾoʔá
	then	jaguar	old	good=limit
	map-a	hah	βaɾi	wɨ́sti
	close-tr	yes	sun	one
	no-kí	his-má-ʔi		kiá
	1pl:acc	see=caus-concur:ss		report
	kamáno	nokí	pi=kína
	jaguar	1pl:acc	eat=concur:sa
	“So he kept it well, yes, and after one day the jaguar visited (saw) us to eat us.” 0181:0105

Chácobo has a highly infrequent subsequent dependent clause marked with =noʂparí (there are only four examples in my corpus).

11.	hakiɾɨkɨ́	naa	ka=ʔita=ʔá=ka		βaɾi
	then	dem.prox	go=recpst=nmlz:pst=rel		day
	no	ho=noʂparí		hawɨ́	yonóko
	1pl	come=subseq:ss/a		3sg.gen	work
	mi	a=kɨ́		tsi	ní
	2sg	do=prior:ds/a		lnk	inter
	naa	no	ho=ita=ʔána
	dem.prox	1pl	come=recpst=nmlz:pst
	“After this, yesterday, before we came, “what work did you do before arriving?” (he said)” 1865:0060

Same-subject and different-subject clauses can occur in chains. In the following example, a concurrent same-subject clause marked with =ʔi “same-subject S/A concurrent” is embedded under a prior same-subject clause as in 12.

12.	βakíʃmaɾí	tsi	sani	a(k)=ʔi
	morning	lnk	fish	do=concur:ss
	tsi	kaɾo	a(k)=ʂó	hawɨniá
	lnk	lumber	do=prior:sa	what.time
	baɾí=no	kaɾá	ho=kí=a	tiá
	day=spatial	epis	come=decl:nonpst=1sg	epis
	“After getting lumber and fishing, what time/day will I come back.” 0243:0094–0095

The relation between dependent and main clause can also be aspectual. The morpheme =pama “same-subject, interrupted” encodes that the event expressed by the dependent clause is interrupted by an event of the main clause. Examples are provided in 13 and 14 below.

13.	ka=páma	tsi	kiá	ʃinó	ha
	go=intrmp:ss/a	lnk	report	monkey	3sg
	nika=ní=kɨ
	hear=rempst=decl:pst
	“As he was going (he stopped) and heard the monkey”

14.	taʂaʔa(k)=βoná=pama		tsi	kiá	mai
	sweep=going:tr/pl=intrmp:ss/a		lnk	report	earth
	ha	ɾooʔa(k)=ní=kɨ			ɨɨ
	3sg	fall.into.earth=rempst=decl:pst			ideo
	“As she started to sweep the floor, she fell through the ground and yelled ëë” 0638:0090

Chácobo also displays asyndetic clause conjunction (called asyndetic “coordination” in Tallman (2018b)). To the best of my knowledge, such an asyndetic clause linkage construction has not been described for any other Pano languages. The construction is typically used when the conjoined events display some parallelism, or even identity as in 15, respectively.

15	ható	ʃina	βɨɨ	ható	ʃina
	3pl:gen	soul	bring	3pl:gen	soul
	bɨɨ	kiá	yoʃí	táʃi	i=pao=ní=kɨ
	bring	report	spirit	Tashi	aux=hab=rempast=decl:pst
	“He used to bring the spirits and brought the spirits.” 0783:0064

Asyndetic conjunction seems to be used to highlight the fast and perhaps planned succession of events acted out by the A/S participant. For instance, the following utterance comes from a story of a man who seeks to kill his in-laws by farting in their face after mixing his farts with tar—both actions (grabbing and coming) are performed purposefully and sequentially with the intent to kill via gastrointestinal gases.

16	ʂɨto	atʃ-á	tsi	ho=ʔi	kiá
	pitch	grab-tr	lnk	come=c	report
	“He grabbed the tar and came (to fart in her face)” 0852:0076

As it will become relevant for the discussions below, I point out here that asyndetic clauses are somewhat hard to elicit. One often has to start with an instance of such clauses occurring in natural speech and then modify it to obtain elicitation judgments. This is perhaps due to the fact that I do not yet fully understand the semantics and/or pragmatics of these clauses.

There are two different subject clauses. Different subject clauses marked with =kɨ́(no) code that an event occurs prior to the main clause. Switch reference clauses marked with = ́no occur concurrently with the event of the main clause. Examples of the prior switch reference are provided in and 18 below.6

17.	hakirɨkɨ́		toa		ha	pi=kɨ́		tsi	kiá
	then		dem.dist		3sg	eat=prior:ss/a		lnk	report
	ha		toa		ɨwati= ́	yopa=ní=kɨ
	3sg		dem.dist		gra.mo=erg	look.for.not.find=rempast=decl:past
	hawɨ́		βakɨ́		kamáno	…		toa
	3sg.gen		child		jaguar	…		dem.prox
	kako= ́				pi=ʔána
	Caco=erg				eat=nmd:past
	“And after he (Caco) ate him (his father), it is said that his gran mother looked for him and didn’t find him (Caco), nor the jaguar that Caco ate.” 0032:001
18.	tíma	há		wa=kɨ́			tsi
	sound	3sg		tr=prior:ds/a			lnk
	kia	há		ráya			ho=ní=kɨ
	report	3sg		parrot			come=rempst=decl:pst
	“After he (the woodpecker) had been knocking (sounding), the parrot came.” 0780:0071

Examples of concurrent marked clauses are provided in.7

19.	háβi	tóka=ka	mai	kíni	oto
	surely	like.so=rel	earth	hole	cough
	oto	há	wa=no	tsi	kiá
	cough	3	tr=concur:ds/a	lnk	report
	hóni	wɨ́tsa	ho=ní=kɨ
	man	other	come=rempast=decl:pst
	“When they were coughing from the cave like this, another man arrived.” 0008:0110

Finally, there are dependent clauses which are not constrained with respect to whether they do or do not share an argument in common with the main clause. Clauses marked with =ʔai(na) can be coreferential with the object or the subject of the main clause as in 20 and 21, respectively.

20.	a=βona=ʔái=ka	makína	tʃipatia=βona=ʔái=ka	kará	tóa
	do=going=nmlz=rel	machine	row=going=nmlz=rel	dub	dem:dist
	a(k)=pao=ní=kɨ		yamaβo= ́	pápa
	kill=hab=rempast=decl:pst		dead=erg	father
	“While that one was rowing or going by motor (on the river), my father would kill him.” 0312:0334

21.	ʂáɾa=ka	ʂóβo	náa	paso=ní=kɨ	kiá
	inside=rel	house	dem:prox	be.slient=rempast=decl:pst	report
	nika=ʔáina
	listen=nmlz:agt
	“So the jaguar went silent listening to what was going on in the house” 0026:0019

There is a past tense =ʔá(na) which is also unspecified with respect to whether it requires coreference with the subject of the main clause. It can be coreferential with the subject as in 22, or not as in 23.

22.	ima	ima=ʃina		ha
	roast	roast=at.night		3
	wa=ʔá=ka		káʂa=kɨ		kiá
	tr =nmd:ant=rel		angry=decl:pst		report
	yóʂa
	woman
	“After roasting it all night, the woman was angry (it is said)” 0483:0945

23.	ha	ho=ʔá=ka		yoanomano
	3	come =nmd:ant=rel		for.a.long.time
	ho=tɨkɨ́(n)	tsáka	=ní=kɨ
	come=again	agouti	=rempast=decl:pst
	“After he arrived, and then after a while, the agoutis came.” 0058:0032

Note that =ʔái(na) and =ʔá(na)-marked dependent clauses can modify noun phrases. In many cases, they are ambiguous between a noun-modifying and a predicate-modifying function (see Guillaume 2011 for similar phenomena in Cavineña). This is illustrated in 24 and 25.

24.	yonoko=ʔái=ka	hɨnɨ	yoʂa= ́	á(k)=kɨ
	work=nmlz=rel	chicha	woman=erg	make=decl:pst
	“The woman who is working made chicha.”/”While the woman was working, she made chicha.”

25.	yonoko=ʔá=ka	hɨnɨ	yoʂa= ́	á(k)=kɨ
	work=nmlz=rel	chicha	woman=erg	make=decl:pst
	“The woman who had worked made chicha.”/”After the woman worked, she made chicha.”

There is a strong tendency for =ʔá(na) and =ʔái(na)-marked clauses to be predicates of non-verbal predicate constructions (Tallman 2018b). When such clauses do occur in nonverbal predicate constructions, they also strongly tend to occur after the subject, contradicting the general trend for non-verbal predicate constructions that follow a predicate–subject order (Tallman 2018b for details). Examples where the =ʔái(na) and =ʔá(na)-marked dependent clauses occur as predicates in non-verbal predicate constructions occur in 26 and 27.

26.	hati=roʔa=ka	noʔiria=bo	tsi	kiá	ho=yo=ʔáina
	all=limit=rel	people=pl	lnk	report	come=all=nmlz
	“All the people came.”/”The people were the ones who all came.” 0014:0187

27.	wɨ́tsa	tsi	kiá	naa	aka(n)=ita=ʔána
	other	lnk	report	dem.prox	be.killed=recpst=nmd:pst
	“This is the other one that was killed”/”This other one was killed.” 0056:0131

Dependent marked clauses marked with =ʔái(na) can function as arguments of a verb. One could refer to such cases as headless relative clauses or simply claim that the clauses are nominalized themselves (Shibatani 2019). Examples occur in 28 and 29. Dependent clauses marked with =ʔá(na) cannot function as arguments of a verb (independent of a head noun that they modify).

28.	hatí=ɾoʔa	tʃani=kan=(ʔ)ai=βo	hoi	ha
	all=limit	speak=pl=nmlz=pl	speech	3
	bitʃ=(ʔ)i	kiá
	take=c	report
	“It grabs the speech, all that is spoken.” 2153:0409

29.	diezaño	ha	=ʔá=ka	ɨ-a=ɾí	kai=kí
	ten.year	3	=nmd:pst=rel	1sg-epen=aug	mother=dat
	tsi	ka=kas=kí=a		i	kiá
	lnk	go=vol=decl:nonpast=1sg		say	report
	naa	rɨso=kan=(ʔ)ái=βo		ka=ʔai
	dem.prox	die=pl=nmlz=pl/assoc		go=nmlz
	kia=ʔái=ka=bo
	lie=nmlz=rel=pl/assoc
	“When they are 20 years of age “I want to go to my mother” they say, and these that are dead go and lie.” 0783:0031

Another type of clause-linkage device is marked with =tí “purpose/instrumental nomalizer”, which codes a purpose clause. An example is provided in 30.

30.	toa	toʔotí	siɾi	ɨ
	dem.dist	shot.gun	old	1sg
	bi=ní=kɨ		naa	roʔá	tsi
	grab=rempast =decl:pst		dem.prox	limit	lnk
	yona=kí=a		βikoβí	sani	a(k)=tí
	use =decl:nonpst=1sg		nail.arrow	fish	kill=nmlz:purp
	“I bought that old shot gun; I use this nail arrow to fish only.” 0903:0098

Dependent clauses marked with =tí can also function as predicates in non-verbal predicate constructions as in 27.

31.	harí	náama	ʂo	mí	βana=ka(n)=tí
	again	already	decl	2sg:gen	harvest=pl=nmlz:purp
	“It is already again time for your harvest.” 2153:0848

The marker also functions as an instrumental nominalizer. By “instrumental” I mean it creates a referent: “object is used for V”. Examples where =tí-marked forms which have a referential function are provided in 32.

32.	hawɨ́	tɨ-nɨʂ-ɨ=tí	pistia	tsi	kiá	ha
	3sg:gen	neck-tie-itr=nmlz:purp	small	lnk	report	3
	tɨ-nɨʂ=ní=kɨ
	neck-tie-itr=rempst=decl:pst
	“He (Caco) tied his little scarf around his (the Kingfisher’s) neck.” 2119:0357

All dependent clauses in Chácobo require another clause to be present in the same sentence to occur—a clause which they are dependent to. However, this other clause needs not be a main clause, as I defined it above. Dependent clauses can be “co-dependent” with another dependent clause as in 33 and 34.

33.	βotɨ	ha	=wa=kɨ́	tsi	naká
	go.down	3	=tr=prior:ds/a	lnk	chew
	naká	no	=wa=ʔána
	chew	1pl	=tr=nmd:pst
	“When she went down, we had chewed everything (the yuca).” 1156:0091

34.	βaʔi= ́	ʃita	ʃita=ʔái=ka	no
	road=spat	cross	cross=nmlz=rel	1pl
	atʃ-a=ʔána
	grab-tr=nmd:pst
	“We grabbed it when it crossed the road.” 1157:0127

Based on the Chácobo data, I add “capacity to function referentially” and ability to modify nouns as another variable in the clause-linkage typology. These variables were not considered in Bickel (2010) but they are important for fully capturing variation in clause linkage, especially in a South American context (see the papers by Zariquiey et al. 2019).

4. Parameters of Typological Variation in Chácobo-Dependent Clauses

This section applies diagnostics for the coordination–subordination distinction to the clause-linkage strategies of Chácobo. Most of these properties are described in the work of Bickel (2010). Some of these properties, or typological variables, are broken down further in order to account for the observed variation found in Chácobo. For instance, whether dependent clauses can have their own interrogative constituents depends on the part of speech of the constituent clause in question. Also, finiteness is not treated as a binary variable as it is in the work of Bickel (2010). Rather, I consider every TAAMME modification for which I have data.

As noted above, some elicitation data are used to fill gaps in my corpus or to provide negative evidence where necessary. To this end, I constructed a survey of elicitation questions designed to test all the relevant variables from Bickel. The original recordings for the data from naturalistic speech and the elicitation data can be found in the work of Tallman (2018b). The parameters are summarized in Table 3 below.

Table 3. Parameters of variation with respect to the coordination–subordination distinction.

Variable	Values
Position: The position of the dependent clause vis-à-vis the main clause with which it enters a dependency relation is	Fixed: post-main: is fixed and is always after the main clause
	Fixed: pre-main: is fixed and is always before the main clause
	Flexible-adjacent: can be before or after the main clause but must be adjacent to it
	Flexible-relational: can be before or after the main clause and can be separated from the main clause by other dependent clauses (Bickel 2010, pp. 81–82)
Wh: Question words and constituent focus inside dependent clauses are allowed	ok: allowed
	banned: not allowed (Bickel 2010, pp. 81–82)
Wh-NP-ext-main (new): An NP constituent interrogative can be extracted from a main clause.	ok: Extraction of an NP constituent interrogative can always occur out of a main clause
	local: Extraction of an NP constituent interrogative can occur out of a main clause when the main clause is local.
	banned: Extraction of an NP constituent interrogative out of a main clause is banned
Wh-NP-ext-dep (new): An NP constituent interrogative can be extracted from a dependent clause	ok: Extraction of an NP constituent interrogative can always occur out of a dependent clause
	local: Extraction of an NP constituent interrogative can occur out of a dependent clause when the dependent clause is local
	banned: Extraction of an NP constituent interrogative out of a dependent clause is banned
Wh-Adv-ext-main (new): An AdvP constituent interrogative can be extracted from a main clause.	ok: Extraction of an AdvP constituent interrogative can always occur out of a main clause
	local: Extraction of an AdvP constituent interrogative can occur out of a main clause when the main clause is local
	banned: Extraction of an AdvP constituent interrogative out of a main clause is banned
Wh-Adv-ext-dep (new): An AdvP constituent interrogative can be extracted from a dependent clause	ok: Extraction of an AdvP constituent interrogative can always occur out of a dependent clause.
	local: Extraction of an AdvP constituent interrogative can occur out of a dependent clause when the dependent clause is local
	banned: Extraction of an AdvP constituent interrogative out of a dependent clause is banned
Wh-np-atb-ext (new): An NP interrogative constituent can be across the board extracted	ok: ATB extraction is allowed
	banned: not allowed
	not applicable: There are no contexts that allow us to assess the claim
Extraction: Extraction of elements of dependent clauses is	ok: allowed
	banned: not allowed (Bickel 2010, pp. 81–82)
Foc: Focus marking on dependent clauses is	Reformulated into focus position (Bickel 2010, pp. 81–82)
Foc-position (new): Focus or topic position in the clause	ok: available, one can show that there is a focus position in both conjuncts
	banned: cannot be established that there is more than one focus/topic position
Finiteness: The dependent clause is headed by a verb that is	finite: at least as many categories must be marked as in main clauses
	Non-finite: only fewer categories are allowed
	any: either the same range or less categories can be marked (Bickel 2010, pp. 81–82)
ILL-mark: Marking of illocutionary force operators in the dependent clause is	ok: allowed
	banned: not allowed
	harmonic: allowed but subject to constraints based on the tense or status choice in the main clause (Bickel 2010, pp. 81–82)
T-mark: Marking of tense or status operators in the dependent clause is	ok: allowed
	banned: not allowed
	harmonic: allowed but subject to constraints based on the tense or status choice in the main clause (Bickel 2010, pp. 81–82)
Finiteness: The dependent clause is headed by a verb form that is	finite: at least as many categories must be marked as in main clauses
	non-finite: only fewer categories are allowed
	any: either the same range or less categories can be marked
Finiteness-multiple (new): A T(ense), A(aspect) A(ssociated) M(otion), M(odal) E(vidential) marker is	ok: allowed
	banned: not allowed
Layer: The dependent clause adjoins	ad-v: to the predicate and can be center-embedded
	ad-s: to the clause and cannot be center-embedded (Bickel 2010, pp. 81–82)
Center-embed-case (new): Center embedding via ergative case marking is	ok: allowed
	banned: not allowed
Center-embed-pa (new): Center-embedding via participant agreement is	ok: allowed
	banned: not allowed
T-scope: The scope of tense or status operators in the main clause is	conjunct: extends to the main clause and the dependent clause
	local: is limited to the main clause
	extensible: extends to either the main clause alone or to both the main clause and the dependent clause, but never to the dependent clause alone. (Bickel 2010, pp. 81–82)
ILL-scope: The scope of illocutionary operators in the main clause is	conjunct: extends to the main clause and the dependent clause
	local: is limited to the main clause
	extensible: extends to either the main clause alone or to both the main clause and the dependent clause, but never to the dependent clause alone. (Bickel 2010, pp. 81–82)
NEG-scope (new)	conjunct: extends to the main clause and the dependent clause
	local: is limited to the main clause
	extensible: extends to either the main clause alone or to both the main clause and the dependent clause, but never to the dependent clause alone
Referential-function (new): A referential function is	na: not allowed
	agentive: allowed and the clause/NP has an agentive function
	instrumental: allowed and the clause/NP has an instrumental function
Noun-modify-function (new): A noun-modifying function is	no: not allowed
	yes: allowed (e.g., the clause can function as a relative clause)

Table 3 contains the variables from the work of Bickel (2010) and additional variables that I have added to this study. The new variables are marked off with “(new)” beside the name of the variables. The justification for adding such variables is provided throughout the description. I also code variables as they are found in Bickel as well, which allows me to contextualize the patterns with respect to Bickel’s data (see Section 5). Note that, ideally, I would recode all of Bickel’s data according to the new variables I have added. This would follow autotypology methodology more faithfully (Bickel and Nichols 2002). Unfortunately, I do not have the relevant data for these variables in all the languages of Bickel’s study. My goal in adding more variables is partially to provide a richer description of Chácobo, but also to encourage researchers to consider the new variables in their own descriptive studies, an issue that I return to in Section 7.

An obvious example of a new variable I have in light of the evidence from Chácobo comes from the variable center-embed:pa. This refers to the possibility that a given dependent clause can be skipped over by a switch-reference marker. This variable may be very specific to Chácobo, or Pano languages, but its value for a given construction could be construed as evidence for subordinate or coordinate status for that construction and it good be seen as a sub-variable of Bickel’s layer. Center-embedding is plausibly more associated with subordination than with coordination. The other new variables Referential-function and Noun-modifying-function are more general. They are important to add in the context of South American languages, due to the tendency for many languages in the region to have constructions which can function as either noun modifiers or adverbial clauses. Other new variables are those that refer to the possibility of constituent interrogatives to function as arguments of or modify dependent clauses. This variable relates to both Wh and extraction.

4.1. Position in Relation to Main Clause

As Bickel (2010, p. 76) notes, the flexibility of the dependent clause in relation to the main clause is understood as an indicator of “subordinate” status. “Coordinate” or chained clauses are thought to occur in a more fixed order. In the generative literature, this criterion could be thought of following from the “Coordinate Structure Constraint” since it bans movement of conjuncts in coordinate structures, but not complex sentences with subordinate clauses (Ross 1967, p. 161; Weisser 2015, p. 11). All dependent clauses in Chácobo can occur on either side of the main clause except asyndetic conjunction, and the interrupted event =pama clauses. Thus, with respect to the position variable, only the asyndetic conjunction and =pama marked constructions are coordinate.

As noted above, some of the same/different subject markers display a different phonological form depending on whether they mark a clause that occurs after or before the main clause. The prior same subject clauses are realized as =ʔáʂna and =ʂóna rather than =ʔáʂ and =ʂó, respectively. Examples of the prior same-subject clauses occurring after the main clause, with their “long form” markers are provided in 35 and 36.

35.	haβa=ʔitá=kɨ	kiá	nika=ʔáʂna
	run=recpst=decl:pst	report	hear=prior:ss
	“He already escaped when he heard it.” 0841:0077

36.	ɨ	bi=ʔá=ka	ória=βo	ɨ́	tʃokoʔa(k)=yamɨ́t=kɨ
	1sg	grab=nmlz=rel	pot=pl	1sg	clean=distpst=decl:pst
	hɨnɨ́		páʂa		βi=yo=ʂóna
	water		raw/new		grab=cmpl=prior:sa
	“When I grabbed it, I washed the pots after gathering all the water.” 1156:0016

The prior-event different-subject clause is realized as =kɨ́no rather than =kɨ́ as in 37 below.

37.	no-ki=ti	tsi	hɨnɨ	a(k)=ki	no-a
	1pl-dat=too	lnk	water	do=decl:nonpst	1pl-epen
	ha	ka=kɨ́no
	3	go=prior:ds/a
	“When he goes, we make chicha.” 1840:0040

In asyndetic conjunction, the clauses cannot be reordered. Since there is no overt dependent marking, it is unclear which of the clauses should be regarded as dependent, but in any case, switching the order of the clauses is ungrammatical as in 39 below (the grammatical sentence on which this sentence is based is provided in 38.).

38.	ha-ʔ-ɨpa	yoa	ha-ʔ-ɨwa	tsaya	tsi
	3-epen-father	tell	3-epen-mother	see	lnk
	honi= ́	=wa=ní=kɨ
	man=erg	=tr=rempst=decl:pst
	“The man told his father and visited his mother.” elic

39.	*ha-ʔ-ɨwa	tsaya	tsi	honi= ́	=wa=ní=kɨ
	3-epen-mother	see	lnk	man=erg	=tr=rempst=decl:pst
	ha-ʔ-ɨpa	yoa
	3-epen-father	tell
	“The man told his father and visited his mother.” elic

On the other hand, reordering the conjuncts without moving the clause-type morpheme produces a difference in meaning. The conjuncts are therefore “tense iconic” in asyndetic clause linkage (see Croft 2001).

I have no examples wherein the interrupting dependent clauses marked by =pama occur after the main clause which they modify. Speakers also reject sentences where the =pama-marked dependent clause occurs after the main clause as in 41 (40 is the correct formulation).

40.	naráha	raʂo=páma	ha	kɨɨsí=kɨ
	orange	peel=intrpt:ss/a	3	cut-intr=decl:pst
	“S/he was cutting the orange when he cut himself.” elic

41.	*ha	kɨɨs-í=kɨ	naráha	raʂo=páma
	3	cut-intr=decl:pst	orange	peel= intrpt:ss/a
	“S/he was cutting the orange when he cut himself.” elic

The values for the position variable, which I have attempted to apply unmodified from the work of Bickel (2010), are provided in Table 4.

Table 4. The results for the position variable applied to clause-linkage constructions.

Form	Gloss	Value
ʔáʂ(na)	prior:ss	flexible:relational
ʂó(na)	prior:sa	flexible:relational
ʔi(na)	concur:ss	flexible:relational
kí(na)	concur:sa	flexible:relational
́noʂpaɾí	subseq:ss/a	flexible:relational
pama(ʔáʂ)	interrupt:ss/a	fixed:pre-main
(asyndetic)	quick:ss/a	fixed:pre-main
kɨ́(no)	prior:ss/a	flexible:relational
́no	concur:ss/a	flexible:relational
́noʂpaɾíno	subseq:ss/a	flexible:relational
ʔá(ka(to)/na)	nmd:ant	flexible:relational
ʔái(ka(to)/na)	nmlz:agt	flexible:relational
tí	nmlz:purp	flexible:relational

4.2. Illocutionary (Interrogative) Marking and Scope (ILL-Marked, ILL-Scope)

Illocutionary force can be used as a criterion to distinguish coordination from subordinate clause-linkage. Clauses are more subordinate if they are not scoped over by illocutionary force and if they do not have illocutionary marking. An intermediate case is where illocutionary force scopes over both conjuncts but they cannot be each be marked by their own illocutionary force independently as in 42, referred to as cosubordination in some of the literature (Foley and Van Valin 1984; Good 2003, inter alia). The fact that 43 is not grammatical suggests that the relevant construction is cosubordinate to these authors.

42.

Jeff has already left for Wittenberg and should arrive there tomorrow.

43.

*Has Jeff already left for Wittenberg and should arrive there tomorrow?

Dependent clauses in Chácobo cannot have their own illocutionary marking independent of the main clause. However, an illocutionary marker of a main clause can scope over a dependent clause. Bickel (2010) described four possibilities with respect to illocutionary scope: (i) local: the illocutionary operator scopes just over the main clause; (ii) conjunct: the illocutionary operator scopes over the main and the dependent clause; (iii) extensible: the illocutionary operator extends over the main clause or the main clause and the dependent clause, but never just the dependent clause; (iv) disjunct: the illocutionary operator extends to either the main or the dependent clause but never both.

Data from elicitation reveal that all interrogative operators are extensible across all dependent clauses in Chácobo except the “nominalized purpose/instrumental” clause marked by =tí and the interruptive same-subject clause marked by =páma. An illustration of the extensible character of interrogatives with dependent clauses is provided in 44. The interrogative marker scopes over just the main clause or the main clause and the dependent clause. The interpretation whereby the illocutionary operator scopes over both the dependent and main clause does not appear to be particularly common in naturalistic speech. Note that one knows that an extensible interpretation is possible nevertheless, because 45 and 46 are both permissible answers to the question in 44. From this point on, I will not include the permissible answers and assume that extensibility can be read off the alternative translations.

44.	tʃaʃo	pi=ʔi	tsi	hɨɾɨ=yá	tʃani=kan=ʔá
	pig	eat=concur:ss	lnk	Gere=com	speak=3pl=inter:past
	“While they were eating did they speak with Gere?”/”Were they eating pig and did they speak with Gere?”

45.	hɨɾɨ=yá	tʃani=kɑ(n)=yáma=kɨ	hama	tʃatʃo
	Gere=com	speak=3pl=neg=decl:pst	but	pig
	pi=ká(n)=kɨ
	eat=3pl=rempst=decl:pst
	“They did not speak with Gere, but they did eat pig.”

46.	tʃatʃo	pi=ká(n)=yáma=kɨ	hama	hɨɾɨ=yá
	pig	eat=3pl=rempast=decl:pst	but	Gere=com
	tʃani=ká(n)=kɨ
	speak=3pl=neg=decl:pst
	“They did not speak with Gere, but they did eat pig.”

Extensible interpretations are also found with prior same-subject clauses and different subject clauses as in 47, 48, and 49.

47.	tʃaʃo	pi=ʔáʂ	tsi	hɨɾɨ=yá	tʃani=kan=ʔá
	pig	eat=prior:ss	lnk	Gere=com	speak=3pl=inter:pst
	“Did they speak with Gere after eating pig?”/ “Did they speak with Gere, and did they eat pig?”

48.	tʃaʃo	pi=ʂó	tsi	hɨɾɨ	honi=βá
	pig	eat=prior:ss	lnk	Gere	man=pl:erg
	tsaya=ʔá
	see=inter:pst
	“Did the men see Gere after eating pig?”/ “Did they eat pig and see Gere?”

49.	honi= ́	ɾaʃa=kɨ́=ɾoʔá		tsi	ina
	man=erg	hit=prior:ds/a=limit		lnk	dog
	ɾɨso=ʔá
	die=inter:pst
	“Right after the man hit the dog, did it die?/Did the man hit the dog and did it die?”

The same pattern applies to all dependent clauses except the interruptive clause and the purposive clause. The purposive clause does not display extensibility with respect to interrogatives, as is shown in 50. Rather, the interrogative only scopes over the main clause and the information in the =tí-marked clause is presupposed. Thus, with this clause the scope property is local, rather than extensible.

50.	ʂoβo	a(k)=tí	karo	kɨɨs-a=ʔaí
	house	make=nmlz:purp	lumber	cut-tr=inter:2sg
	“Are you gathering lumber to build a house.” “*Are you building a house and are you gathering lumber?”

Thus, most dependent clauses in Chácobo pattern somewhat like cosubordination in that interrogative force scopes over them. But they are not like cosubordination in that they can also have an interpretation where the illocutionary operator does not scope over them. The instrumental nominalizer and the same-subject interruptive clause behave most like a subordinate clause in this respect as they can only display local scope. For the same-subject interruptive clause, this may be somewhat problematic because it patterns more like a coordinate clause with respect to the position variable.

The results for the ill-scope and ill-mark variables are provided in Table 5. These variables are adopted from Bickel (2010) without modification.

Table 5. The results for the ill-scope/ill-mark variable applied to clause-linkage constructions.

Form	Gloss	ill-mark	Ill-scope
ʔáʂ(na)	prior:ss	banned	extensible
ʂó(na)	prior:sa	banned	extensible
ʔi(na)	concur:ss	banned	extensible
kí(na)	concur:sa	banned	extensible
́noʂpaɾí	subseq:ss/a	banned	extensible
pama(ʔáʂ)	interrupt:ss/a	banned	local
(asyndetic)	quick:ss/a	banned	extensible
kɨ́(no)	prior:ss/a	banned	extensible
́no	concur:ss/a	banned	extensible
́noʂpaɾíno	subseq:ss/a	banned	extensible
ʔá(ka(to)/na)	nmd:ant	banned	extensible
ʔái(ka(to)/na)	nmlz:agt	banned	extensible
tí	nmlz:purp	banned	local

4.3. Negative Marking and Scope (NEG-Marked, NEG-Scope)

In contrast to illocutionary marking, in Chácobo, negation can be marked in all dependent clauses. Despite this difference, similar questions about illocutionary scope can also be asked of negation. In Chácobo, all dependent clauses can be marked with negation, although it is not common in naturalistic speech. Some illustrative examples are provided in 51 with the purposive clause and in 52 with a same-subject clause.

51.	tʃani=yáma=tí	haβá=kɨ	hɨ́ɾɨ
	speak=neg=nmlz:purp	run=decl:pst	Gere
	“Gere ran away so he wouldn’t have to speak.” ptcp obsv

52.	moʔi=yámɑ=ʔi	waaʂá=ki	honi
	move=neg=concur:ss	paddle=decl:nonpst	man
	“He is paddling without moving.” ptcp obsv

Whether negative marking is extensible or local depends on which dependent clause is involved. Asyndetic conjunction and all same-subject clauses are extensible with respect to negation. This means that, when the negative marker occurs in the main clause, the negation can have a strictly local interpretation (modifying the event of the main clause) or display scope over the main and the same-subject clause, as in 53 below.

53.	ʂoβo=kí	kaʔɨ=ʔi	tsi	honi
	house=dat	arrive=concur:ss	lnk	man
	tsaʔo=yáma=kɨ
	sit=neg=decl:pst
	“When the man arrived at this house, he didn’t sit down.” “The man did not arrive at his house, nor did he sit down.”

However, the interpretation of the negation modification must be local when the dependent clause is a different subject clause.

54.	ʂoβo=kí	yoʂa	kaʔɨ=kɨ́	tsi	honi= ́
	house=dat	woman	arrive=prior:ds/a	lnk	man=erg
	tsaya=yáma=kɨ
	see=neg=decl:pst
	“When the woman arrived at the house, the man did not see her.” *”The woman did not arrive at the house and the man did not see her.”

In contrast to same/different-subject clauses, nominalized clauses marked with =ʔái(na) require negation to be interpreted locally. That is, if a main clause is marked with the negative =yáma, the negative scopes over the main clause and not the imperfective clause, as illustrated in 55.

55.	yonoko=ʔái=ka	yoʂa= ́	hɨnɨ	a(k)=yáma=kɨ
	work=nmlz=rel	woman=erg	chicha	make=neg=decl:pst
	“As the woman worked, she made did not make chicha.” “*The woman neither worked, nor made chicha.”

This is not true of the nominal-modifying clause marked with =ʔá(na), as illustrated in 56 below. Clauses marked with this marker are extensible with respect to negation marking.

56.	yonoko=ʔá=ka	yoʂa= ́	hɨnɨ	a(k)=yáma=kɨ
	work=nmd:ant=rel	woman=erg	chicha	make=neg=decl:pst
	“After the woman worked, she didn’t make the chicha.” “The woman neither worked, nor made chicha.”

Thus, in Chácobo, all same-subject clauses display extensibility with respect to negation. This also includes asyndetic conjunction. This means that the negative marker can have a local or wide scope interpretation. However, with different subject clauses, the negation only has local scope. Finally, nominalized clauses display extensible scope. Different subject clauses would appear to be the most subordinate-like according to negative scope. Table 6 summarizes the results of applying diagnostics based on negation.

Table 6. The results for the neg-marked/neg-scope variable applied to clause-linkage constructions.

Form	Gloss	neg-marked	neg-scope
ʔáʂ(na)	prior:ss	ok	extensible
ʂó(na)	prior:sa	ok	extensible
ʔi(na)	concur:ss	ok	extensible
kí(na)	concur:sa	ok	extensible
́noʂpaɾí	subseq:ss/a	ok	extensible
pama(ʔáʂ)	interrupt:ss/a	ok	local
(asyndetic)	quick:ss/a	ok	extensible
kɨ́(no)	prior:ss/a	ok	local
́no	concur:ss/a	ok	local
́noʂpaɾíno	subseq:ss/a	ok	local
ʔá(ka(to)/na)	nmd:ant	ok	extensible
ʔái(ka(to)/na)	nmlz:agt	ok	local
tí	nmlz:purp	ok	local

4.4. Constituent Interrogatives (WH)

One of the criteria Bickel (2010) uses is whether a dependent clause can host a constituent interrogative. In Chácobo, research thus far suggests that all constituent interrogatives are fronted.8 Furthermore, one cannot have a sentence with two constituent interrogatives of the same type even when one could, in principle, be licensed by a dependent clause. This is illustrated with the ungrammatical sentences in 57 and 58. The ability for another interrogative constituent to occur when one of the dependent clauses is present has been tested with all the dependent clauses.

57.	*tsowɨ	tsowɨ	tsaya	awini= ́	=wa=kɨ́
	who	who	see	woman=erg	=tr=prior:ds/a
	tsi	taʃi= ́	raaʔak	=ʔá
	link	Tashi=erg	scold	=inter:pst
	Intended: “Who did the woman see and (after) who did Tashi scold?”

58.	*tsowɨ	awini= ́	tsaya	wa=kɨ́	tsi
	who	woman=erg	see	tr=prior:ds/a	lnk
	tsowɨ	taʃi = ́	ɾaaʔak=ʔá
	who	Tashi=erg	scold=inter:pst
	Intended: “Who did the woman see and (after) who did Tashi scold?”

4.5. Information Structure Positions and Markers (FOC)

Bickel (2010) describes having a focus position and being able to have a focus marker as potentially independent variable. However, testing for “focus” as a typologically consistent variable is made difficult by the fact that there is no cross-linguistically agreed-upon definition of focus: that is, the notions “focus” and “topic” can be similarly broken down into a number of distinct senses, uses, or “variables” (Ozerov 2018, 2021).

It is outside the scope of this paper to attempt to integrate a typology of information structure categories into clause-linkage typology. Instead, I will refer to a variable that refers to positions that have information structural definitions. The clause-initial position in Chácobo has a number of functions. It is used for contrastive focus and in answer to questions for NPs, but is also associated with givenness, especially with verbs (Tallman 2018a). In Chácobo, this position is marked off by having the Wackernagel-like morpheme tsi occur before it, referred to as “position 5 morph” in Tallman (2018a) and glossed as “linker” here (for more examples, see Tallman (2018a)).

A contrastive focus-like function of this initial, prior to tsi, position is provided in 59. The noun phrase hawɨ roʔá “his large vulture” is in a position before tsi and has a contrastive focus function in the following example.

59.	hawɨ	roʔá	tsi	kiá	kaʔɨ=ʂɨni
	3sg:gen	large.vulture	lnk	report	know=v>adjlz
	hama	kiá	hawɨ	siyaβi	…
	but	report	3sg:gen	Siyabi	…
	hawɨ́	roʔá	tsi	kiá
	3sg:gen	large.vulture	lnk	report
	βotɨ=ní=kɨ
	descend=rempst=decl:pst
	“His (Mabocorihua’s) large vulture knew, but not his siyabi… then it was his large vulture that descended.” 00063:0155–0157

The question which is relevant for clause-linkage typology is whether this focus/topic position can be projected in dependent clauses. That is, can dependent clauses have a “first position” before tsi independent from the main clause? Chácobo-dependent clauses appear to be able to contain this pre-tsi first position. Evidence for this is that tsi can occur more than when in clause-linkage constructions with same-subject clauses as in (60), (61), and (62).

60.	hama	kako= ́	tsi	kiá	toa
	but	Caco=erg	lnk	report	dem:dist
	kamano= ́	βɨ́ɾo	moto	toka=ta(n)=ʂó
	jaguar=gen	eye	chive	do.so=go&do=prior:sa
	tsi	kamano= ́	βɨ́ɾo	hana=kí
	lnk	jaguar=gen	eye	mouth=dat
	toa=ní=kɨ
	explode=rempst=decl:pst
	“But when Caco did so with the jaguar’s eye and chive, the jaguar’s eye exploded in his mouth.” 0181:0164

61.	ʂoβo	ak=(ʔ)á	tsi	tana
	house	do=prior:ss	lnk	distance
	raka-na=tan=i		ható	ɨwatí	βi=mitsa
	stay-intrc=go&do=concur:ss		3pl:gen	gra.mo	recieve=psbl
	natani=βayá		tsi	ʂoβo	a(k)=βayá
	pass.by=do&go:tr/pl		lnk	house	do=do&go:tr/pl
	tsi	ʂoβo	a(k) =βayá		tsi
	lnk	house	do=do&go:tr/pl		lnk
	kiá	ha	βo=kan=ní=kɨ
	report	3	go=pl=rempst=decl:pst
	“When they made the house, they stayed there for just one week and then right after their grandmother could have recieved them, they passed by, they made the house, and they left.” 0181:0220

62.	tana	tsi	hoi-ko		=pama
	distance	lnk	get.up-distr		=intrmp:ss/a
	tsi	kiá	yáma	tsi	ʂo
	lnk	report	neg	lnk	decl
	“After he got up “there is nothing” (he said)” 0181:0090

Different subject clauses can also contain the marker tsi in them as in 63.

63.	ha	ak=(ʔ)á=ka		wɨakɨ́	tsi
	3sg	do=nmd:ant=rel		after.day	lnk
	hɨnɨ	no	ak=(ʔ)á=ka		toʔo-ko
	chicha	1pl	do=nmd:ant=rel		stir-distr
	ha	=kɨ́	tsi	wɨakɨ́	wai=kí
	3	=prior:ds/a	lnk	after.day	garden=dat
	ká=ki		noa	toka	tsi
	go=decl:nonpst		1pl	do.so	lnk
	ha=βɨta	ká=ʔi	ɨ	i=pao=ní=kɨ
	3=com	go=concur:s	1sg	do =hab=rempst=decl:pst
	“When she did it, the day after we made the chicha, after she stirred it, the day after we go to the garden with her.” 1840:0041

The initial position is also present in nominalized clauses marked with =ʔái(na), as in 64.

64.	toa	ɨ	haβi=ʔái=ka	tsi	piʃa
	dem:dist	1sg	learn=nmlz=rel	lnk	little
	ɨ	ina=kanɑ=ʔái		tsi	ʂo
	1sg	ascend=going:itr=nmlz		lnk	decl
	toa	haʔiki …		noʔó	naama
	dem:dist	then …		1sg:gen	dream
	tsi	ʂo	toa	toa	haska
	lnk	decl	dem:dist	dem:dist	same
	=kato
	=rel
	“When I was learning, I got better and better, then something like my dream will be.” 1840:0133

The nominal-modifying clause marked with =ʔá(na) and the purpose-nominalized clause marked with =tí do not appear with tsi in them. Based on this, I assume that the first position is not available to these clauses.

The results of the informational structural variable are reported in Table 7. The results may not be comparable to the data in the work of Bickel (2010). However, this is because the concept of focus is not comparable. Thus, I have replaced the variable with something which is marked in Chácobo grammar which I consider to be relevant to the coordination–subordination distinction.

Table 7. The results for the focus/topic position variable applied to clause-linkage constructions.

Form	Gloss	Value
ʔáʂ(na)	prior:ss	ok
ʂó(na)	prior:sa	ok
ʔi(na)	concur:ss	ok
kí(na)	concur:sa	ok
́noʂpaɾí	subseq:ss/a	ok
pama(ʔáʂ)	interrupt:ss/a	ok
(asyndetic)	quick:ss/a	banned
kɨ́(no)	prior:ss/a	ok
́no	concur:ss/a	ok
́noʂpaɾíno	subseq:ss/a	ok
ʔá(ka(to)/na)	nmd:ant	ok
ʔái(ka(to)/na)	nmlz:agt	ok
tí	nmlz:purp	banned

4.6. Asymmetric Extraction of NP Constituent Interrogatives (WH-NP-EXT)

Extraction may be considered a classical test for distinguishing between coordination and subordination. In coordinate clauses, no elements from either of the conjuncts can be extracted asymmetrically (from one conjunct and not the other) (Ross 1967; Levine 2009; Weisser 2015; Bošković 2020), while “across-the-board” extraction is indicative of coordinative status.9 One of the ways this diagnostic has been used is with the extraction of interrogative constituents. To simplify matters, I will only refer to the extraction of interrogative constituents. Future research will be concerned with assessing extraction in other types of contexts (e.g., right dislocation, adverb extraction).

The issue of extraction presented in Bickel (2010) is simplified compared to the number of variables and values relevant for capturing potential cross-linguistic variation. Bickel (2010) only has a single binary-variable extraction. However, a distinction needs to be made between the (i) type of element being abstracted, as noted above; (ii) the status of the clause of the extraction site (main or dependent); (iii) whether the extraction site is local to the landing site.

In order to bring some order to these possibilities, I first make a distinction between noun phrase and adverbial phrase extraction (WH-NP vs. WH-Adv), with the latter being discussed in Section 4.7. Then, these variables are split up further according to whether we are dealing with extraction from the marked-dependent clause or the main clause (WH-NP-MAIN vs. WH-NP-DEP; WH-Adv-MAIN vs. WH-Adv-DEP). Finally, each of these variables can take three values: (i) ok: extraction of the NP/AdvP interrogative constituent is always allowed; (ii) local: extraction of the NP/AdvP interrogative constituent is only allowed when the extraction site and the landing site are not interrupted from each other by more than one clause boundary (see Section 4 above for a similar formulation); (iii) banned: Extraction of the NP/AdvP interrogative constituent is not allowed. Cases where extraction can occur only non-locally do not occur and thus this is not specified as one the potential values.

There are additional caveats and complications involved in interpreting asymmetric extraction of NP constituents in Chácobo. These are discussed in Appendix A (Appendix A.1).

Non-locally, NP constituent interrogatives cannot be extracted from a same-subject clause. If the same subject clause is on the right-side of the main clause, then the fronting cannot occur as illustrated by the ungrammaticality of the following examples.

65.	*hawɨi	kako= ́	oʂa=ʔá	___i	kopi=ʔáʂna
	whati	Caco=erg	sleep=inter:pst	___i	buy=prior:sa
	“What did Caco buy and then slept?”

66.	*hawɨi	kako	oʂa=ʔá	___i	ʃiná=ʔina
	whati	Caco	buy=inter:past	___i	think=concur:ss
	“What did Caco think about when he slept?”

Extraction of constituent NPs from main clauses is permitted whether such extraction is non-local or not in a same-subject construction. This is illustrated in 67 and 68.

67.	hawɨi	hawɨ	ʂoβo	=ki	ho =ʂó	kako= ́
	whati	3sg:gen	house	=dat	arrive=prior:sa	Caco=erg
	___i	kopi=ʔá
	___i	buy=inter:pst
	“What did Caco think about when he slept?”

68.	hawɨ	tsi	aʃi=kí	βoka= ́	___i	kopi=ʔá
	what	lnk	bathe=concur:sa	Boca=erg	___i	buy=inter:pst
	“What while/before bathing did Boca buy?”

For =páma “interruptive”-marked clauses, non-local extraction out of its main clause is not allowed, however, as illustrated by the ungrammaticality of the example below.

69.	*hawɨi	ka=pama	tsi	kiá	ha	___i	nika=ʔá
	whati	go=intrmp:s/a	lnk	report	3	___i	listen=inter:pst
	Intended: “What while going did he hear?”

Asyndetic same-subject clauses can display asymmetric extraction as in 70.

70.	tsowɨi	___i	atʃ-a	hawɨ	taʔɨ	nɨʂ-a
	whoi	___i	grab-tr	3sg:gen	foot	tie-tr
	honi= ́	=wɑ	=ʔá
	man=erg	=tr	=inter:pst
	Intended: “Who did the man grab and grab his foot?”

Note that asyndetic same-subject clauses display a fixed position (Section 4.1), which suggests that they are coordination constructions. However, they also allow for asymmetric extraction, as in the example above. A researcher who is dedicated to the coordination–subordination distinction will have to discard the position diagnostic just so and claim that the clause is fact subordination, or state that this construction violates the coordinate-structure-constraint (Hofmeister and Sag 2010 for relevant discussion).

Different-subject clause constructions marked with =kɨ́ “prior” or = ̌no “concurrent” appear to be more permissive with the extraction of noun phrase constituents. As with =ʔáʂ(na)~=ʂó(na) and =ʔi(na)~=kí(na) same-subject clauses, non-local extraction is permitted out of a main clause. In contrast to same-subject clauses, they appear to allow for non-local extraction from a dependent clause.

71.	hawɨi	kaɾá	ka	honi= ́	kamá	tʃoiʃ-á=kɨ́	tsi
	whati	epis	rel	man=erg	jaguar	shoot.at-tr=prior:ds/a	lnk
	___i	haβá=ʔá
	___i	run=inter:pst
	“What ever did Caco think about when he slept?”

72.	hawɨi	hawɨ́	βakɨ́	pistia	haβa=ʔá	honí	___i
	whati	3sg:gen	child	small	run=inter:pst	man=erg	___i
	tʃoiʃ-a=kɨ́na
	shoot.at-tr=prior:d/sa
	“What did Caco shoot and then its child escaped?”

Nominalized and noun-modifying clause constructions likewise both allow for extraction either locally or non-locall (Note, however, there is problem in interpretation of this example related to the potential presene of a null resumptive pronoun, see Appendix A.1).

73.	hawɨ	kará	kai	tsi	yoʂá	hawɨ́
	what	dub	reli	lnk	woman=erg	3sg:gen
	haʔíni	=yá	tʃani=ʔá		___i	a(k)=áina
	girl	=com	speak=inter:pst		___i	make=nmlz
	“What ever was she making while she talked to her daughter?” (“Whati was the woman who making iti talked to her daughter?”)

Nominalizing purpose/instrumental clauses, however, cannot have NP-constituent interrogatives extracted out of them. For instance, the following is not grammatical in Chácobo according to the speakers I asked:

74.	??hawɨi	ɾiberalta=kí	ha	ka=ʔá	___i	kopi=tí
	??whati	Riberalta=dat	3sg	go=inter:pst	___i	buy=nmlz:purp
	“What did he go to Riberalta to buy?”

The full results of applying asymmetric extraction to each of the clause-linkage constructions are summarized in Table 8.

Table 8. The results for the np-wh-extraction-(from)main/(from)dependent variable applied to clause-linkage constructions.

Form	Gloss	from main clause	from dependent clause
ʔáʂ(na)	prior:ss	ok	local-only
ʂó(na)	prior:sa	ok	local-only
ʔi(na)	concur:ss	ok	local-only
kí(na)	concur:sa	ok	local-only
́noʂpaɾí	subseq:ss/a	ok	local-only
pama(ʔáʂ)	interrupt:ss/a	banned	local-only
(asyndetic)	quick:ss/a	banned	banned
kɨ́(no)	prior:ss/a	ok	ok
́no	concur:ss/a	ok	ok
́noʂpaɾíno	subseq:ss/a	ok	ok
ʔá(ka(to)/na)	nmd:ant	ok	ok
ʔái(ka(to)/na)	nmlz:agt	ok	ok
tí	nmlz:purp	ok	ok

4.7. Asymmetric Extraction of AdvP Constituent Interrogatives (WH-ADV-EXT)

In same-subject clause constructions, adverbial constituent interrogatives can be asymmetrically extracted from same-subject and main clauses in local and non-local contexts. One can see that, when the main clause is final, a fronted adverbial constituent can modify either a local same-subject clause or the main clause. Interpreted in terms of extraction, the interrogative constituent can be extracted from the same-subject clause or the main clause as in 75 and 76.

75.	hawɨniai	hɨnɨ	ak=(ʔ)aʂ	___i	kako	tsaʔo=ʔá
	wherei	chicha	drink=prior:ss	___i	Caco	sit=inter:pst
	“Where, after he drank chicha, did Caco sit?”

76.	hawɨniai	___i	hɨnɨ	ak=(ʔ)aʂ	kako	tsaʔo=ʔá
	wherei	___i	chicha	drink=prior:ss	Caco	sit=inter:pst
	“Where did Caco drink chicha and then sit?”

The adverbial interrogative constituent can be extracted from a local main clause as in 77 or a non-local same-subject clause as in 78.

77.	hawɨniai	kako	tsaʔo=ʔá	___i	hɨnɨ	ak=(ʔ)aʂna
	wherei	Caco	sit=inter:pst	___i	chicha	drink=prior:ss
	“Where did Caco drink and then sit?”

78.	hawɨniai	___i	kako	tsaʔo=ʔá	hɨnɨ	ak=(ʔ)aʂna
	wherei	___i	Caco	sit=inter:pst	chicha	drink=prior:ss
	“Where did Caco drink before he sat?”

There appear to be no constraints on the extraction of adverbial constituents from prior and concurrent same-subject clauses.

Same-subject clauses marked with the interruptive =pama ban non-local extraction, even of adverbial constituents. The basic facts are illustrated in 79 and 80.

79.	hawɨʂoβai	___i	hɨnɨ	a(k)=pama	tsi	ʃinó
	with.whati	___i	chicha	do= intrmp:ss/a	lnk	monkey
	ha	nika	=ʔá
	3sg	listen	=inter:pst
	“With what was he drinking chicha, when he heard the monkey?”

80.	*hawɨʂoβai	hɨnɨ	a(k)=pama	tsi	___i	ʃinó
	with.whati	water	do=intrmp:ss/a	lnk	___i	monkey
	ha	nika	=ʔá
	3sg	listen	=inter:pst
	“With what, while he was drinking chicha, did he hear the money?”

Different-subject constructions marked with =kɨ́(na) “prior different A/S” or = “no concurrent different A/S” allow for all types of adverbial extraction. An example of local extraction of an adverbial constituent interrogative is provided in 81. An example of non-local extraction of an adverbial constituent interrogative from a different subject clause is provided in 82.

81.	hɨni=ʂói	___i	hawɨ́	βakɨ́	pistia	haβa
	how=sai	___i	3sg:gen	child	small	run
	=kɨ	honi= ́	kamá	tʃoiʃ-a	=kɨ́na
	=decl:pst	man=erg	jaguar	shoot-tr	=prior:ds/a
	“Why did hisj child escape when the man shot the jaguarj?”

82.	hɨni=ʂói	hawɨ́	βakɨ́	pistia	haβa	=kɨ
	how=sai	3sg:gen	child	small	run	=decl:pst
	___i	honi = ́	kamá	tʃoiʃ-a	=kɨ́na
	___i	man=erg	jaguar	shoot-tr	=prior:ds/a
	“Why, when hisj child escape, did the man shoot the jaguarj.?”

For clause-linkage with nominalized clauses, all extraction possibilities are available when the extracted constituent interrogative is an adverbial clause. This is illustrated in 83 through 86.

83.	hawɨniai	___i	hɨnɨ	ak	=(ʔ)ái=ka
	wherei	___i	chicha	make	=nmlz =rel
	yoʂa= ́	hawɨ́	haʔíni	=yá	tʃani=ʔá
	woman=erg	3sg:gen	girl	=comit	speak=inter:pst
	“Where was she making chicha when the woman spoke to her daughter?”

84.	hawɨniai	hɨnɨ	ak	=(ʔ)ái=ka	___i
	wherei	chicha	make	=nmlz =rel	___i
	yoʂa= ́	hawɨ́	haʔíni	=yá	tʃani=ʔá
	woman=erg	3sg:gen	girl	=comit	speak=inter:pst
	“Where, when she was making chicha, did the woman speak to her daughter?”

85.	hɨni=ʂói	___i	yoʂa= ́	hawɨ́	haʔíni	=yá
	why=pai	___i	woman=erg	3sg:gen	girl	=comit
	tʃani	=kɨ	hɨnɨ	ak	=ʔáina
	speak	=decl:pst	chicha	make	=nmlz
	“Why did the woman speak with her daughter while she made chicha.”

86.	hɨni=ʂói	yoʂa= ́	hawɨ́	haʔíni	=yá	tʃani
	why=pai	woman=erg	3sg:gen	girl	=comit	speak
	=kɨ	___i	hɨnɨ	ak	=ʔáina
	=decl:pst	___i	chicha	make	=nmlz
	“Why, while the woman spoke with her daughter, was she making chicha?”

As far as I have been able to discern, one cannot extract adverbial constituent interrogatives from purpose clauses. Evidence for this is provided in 87 and 88.

87.	hɨni=ʂó	pila	kopi	=tí	yonoko	=ʔá
	why	battery	buy	=nmlz:purp	work	=inter:pst
	“Why did s/he work to buy batteries?”??”Why after working did/she buy batteries?”

88.	hɨni=ʂó	yonoko	=ʔá	pila	kopi	=tí
	why	battery	=inter:pst	battery	buy	=nmlz:purp
	“Why did s/he work” to buy batteries?” ??”Why after working did/she buy batteries?”

The values of the adverbial constituent interrogative extraction variable are summarized in Table 9.

Table 9. The results for the adv-wh-extraction-(from)main/(from)dependent variable applied to clause-linkage constructions.

Form	Gloss	from main clause	from dependent clause
ʔáʂ(na)	prior:ss	ok	ok
ʂó(na)	prior:sa	ok	ok
ʔi(na)	concur:ss	ok	ok
kí(na)	concur:sa	ok	ok
́noʂpaɾí	subseq:ss/a	ok	ok
pama(ʔáʂ)	interrupt:ss/a	banned	local-only
(asyndetic)	quick:ss/a	ok	ok
kɨ́(no)	prior:ss/a	ok	ok
́no	concur:ss/a	ok	ok
́noʂpaɾíno	subseq:ss/a	ok	ok
ʔá(ka(to)/na)	nmd:ant	ok	ok
ʔái(ka(to)/na)	nmlz:agt	ok	ok
tí	nmlz:purp	ok	banned

As stated above, the extraction variable of Bickel only codes whether any extraction can occur out of a dependent clause, as I understand him. However, the literature reports differences in extraction from main or dependent clauses and differences in the extraction of core and adverbial arguments (Hofmeister and Sag 2010, inter alia), and in Chácobo there appears to be a difference. I, therefore, suggest that the variable may be worth expanding on in this paper.

4.8. Center-Embedding via Ergative Case Marking (Layer of Attachment)

Bickel (2010) describes a variable that he calls “layer of attachment”, making a distinction between clauses that adjoin to the predicate (ad-V), clauses that adjoin to the sentence (ad-S and clauses that adjoin to “some higher level and appear “detached” from the main clauses…” (Bickel 2010, p. 77)).

The main criterion for distinguishing between ad-V and ad-S is whether a dependent clause can appear center-embedded vis-à-vis a main verb. In Chácobo, ergative marking cannot skip over a dependent clause. An ergative case must be assigned locally. This is shown in 89 and 90.

89.	yonoko	=ʂó	tsi	ano	tɨpas
	work	=prior:sa	lnk	paca	murder
	taʃi= ́	wa	=kɨ
	Tashi=erg	tr	=decl:pst
	“After working, Tashi killed the paca.”

90.	*taʃi = ́	yonoko	=ʂó	tsi	ano
	Tashi =erg	work	=prior:sa	lnk	paca
	tɨpas	=kɨ
	murder	=decl:past
	“After working, Tashi killed the paca.”

I have no obvious cases wherein case marking from the main clause skips over a dependent clause. On this basis, all dependent clauses are classified as Ad-S. The problem with such a classification, however, is that, if we look at participant agreement, we arrive at the opposite result.

4.9. Center-Embedding via Participant Agreement

A clause which could be argued to be center-embedded could be argued to be subordinate. In Chácobo, there is a type of center-embedding that can occur when there are adjacent same-subject clauses. An example is provided in 91. In this example, tsaʔo=ʂó could be argued to be center-embedded because the participant agreement marking of the previous clause skips over it and agrees with the transitivity of the main clause. In elicitation contexts, speakers find cases with no such agreement skipping preferable, as in 92.

91.	?kaʔɨ=ʂó	tsaʔo=ʂó	tsi	honi= ́	tsáya=kɨ
	arrive=prior:sa	sit=prior:sa	lnk	man=erg	see=decl:pst
	“The man arrived and then sat down and looked at them.”

92.	kaʔɨ=ʔáʂ	tsaʔo=ʂó	tsi	honi= ́	tsáya =kɨ
	arrive=prior:ss	sit=prior:sa	lnk	man=erg	see=decl:pst
	“The man arrived and then sat down and looked at them.”

In Chácobo, the most common pattern is for a same-subject clause to agree with the clause directly to its right (or left for those clauses which occur after the clause-type morpheme of the main clause), regardless of whether this clause is dependent or not. Examples are provided in 93 and 94 below.

93.	toa	oʂa=kí		orikití	piʃa
	dem:dist	sleep=concur:a		food	a.little
	ma	βo=ʔá=ka		βɨtɨ=ʂó
	2pl	bring=nmd:ant=rel		cook=prior:sa
	pi=ʔi		ma	ʔi=ní
	eat=concur:s		2pl	aux.intr=inter:rempast
	“After sleeping, there you brought a little food, after cooking it you ate it?” 1851:0032

94.	pápi	há	a=ita=ʔáʂ			tsi
	paper	3	make=recpast=prior:ss			lnk
	kiá	ka=ʂó		há	hana=ní=kɨ
	report	go=prior:sa		3	leave=rempast=decl:pst
	“After he had made the paper, and after he went, he left it.” 2119: 368

There are a few attested cases of agreement which appear to skip over a right adjacent same-subject clause in natural speech. Examples are provided in 95 and 96. In 95, the dependent clause aʃi=ʂó “after bathing” is intransitive. However, the same-subject clause before it, nia=ʂó “after throwing it away”, takes a same subject clause which agrees with a transitive clause (kɨɨsakaʂɨkawɨ́ ‘cut it’). This example thus shows that prior same-subject clauses can be center-embedded according to participant agreement. Another example is provided in 96 below.

95.	ma-to	ʃapokotí	nia=ʂó		tsi
	2pl-acc	traditional.belt	throw =prior:sa		lnk
	aʃi=ʂó		tsi	naa	tsi
	bathe=prior:sa		lnk	dem.prox	lnk
	kɨ́ɨs-a=ká(n)=ʂɨ=ka(n)=wɨ́		…	tiaroʔa
	cut-tr=pl=fut=pl=imper		…	size=limit
	“Put this on, throw away your shapocoti, bathe yourself and cut (your skirts) to this size.” (this is what Cai Mariana said) 0146:0153

96.	bi=ʂó	tsi	naa	ho=ʂó	mia-rí
	get=prior:sa	lnk	dem:prox	come=prior:sa	2sg-too
	a=ita=ʔána
	drink=recpast=nmd:ant
	“After getting it (chicha), after arriving, you drank it (chicha) as well.” 1840:0096

Dependent clauses linked through asyndetic combination can also be skipped over with respect to participant agreement. For instance, the transitive clause a=βayá “do and go” is skipped over by the same subject marker =ʔáʂ “prior S/A>S subject” in the example below: the clause tɨ́ɾɨsa há waʔá “after cutting (someone’s) throat” has a same subject marker that agrees with an intransitive clause.

97.	tɨɾɨsa	ha	wa=ʔá(ʂ)	tsi	kiá
	cut	3	tr=prior:ss	lnk	report
	a(k)=βayá		=paɾi	tsi	kiá
	do=do&go:tr/pl		=first	lnk	report
	ha	ka =ní=kɨ
	3	go=rempast=decl:pst
	“After cutting their throat, she did this and went.” 2119:0135

Noun-modifying clauses can also be skipped over as in 98 below. Note that the verb tsaʔo “sit” is intransitive, but the clause prior to agrees with a transitive clause: paʂnariaʂó “after being hungry”.

98.	wɨakɨ́	ɨarí	paʂna-ria=ʂó		ʂoβo
	tomorrow	1sg-too	be.hungry-aug =prior:sa		house
	tsaʔo	tsaʔo=βaʔina=ʔána		ɨarí	wɨakɨ́
	sit	sit=allday=nmd:ant		1sg-aug	tomorrow
	tana=i		ka=ʃaɾi=kí=a
	fish=concur:s		go=cras=recl:nonpast =1sg
	“Tomorrow, after being hungry, and sitting around all day at home, I will go fishing.” 1154:0054

The nominalizer ʔái(na) can also be skipped over in a similar way as in the example 99 below.

99.	tʃoʃ-a=ta=ʂó	yoʂa	ho=ʔái=ka
	kill-tr=pnct=prior:sa	woman	arrive=nmlz=rel
	bɨroa	kaiti	bɨbikima=ní=kɨ
	cutuchi	door	hang.up=rempast=decl:pst
	“After she killed the snake, she hung it in the front of the house, where the woman was coming.” 0029:0028

In elicitation, I was able to collect center-embedding via participant agreement for all of the clause-linkage strategies. Bickel (2010) did not include center-embedding via participant agreement as a variable.

4.10. Finiteness Marking and TAAME Marking

Finiteness can be defined in a graded manner (Givón 2001; Cristofaro 2005; Adger 2007). Givón (2001), for example, considers finiteness to be a scale. That is, a given clause can be more or less finite depending on how many categories expressable in the main clause can be marked in it. Below, I provide a review of fourteen verbal categories (T(ense)-A(spect)-A(ssociation)-M(otion)-M(odality)-E(videntiality)) for which I have enough data. I consider whether they can be marked in dependent clauses. The relevant examples and data are provided in Appendix A. Here, I present only an overview.

The following morphemes/categories are banned from all dependent clauses: =yá “mirative perfect”; =tɨkɨ́n “again”; pistia “a little”; kaɾá “dubitative”; ní “remote past”; kiá “reportative”. The following morphemes/categories are permitted in all dependent clauses; =yó “completive”; =ʃiná “at night”; =βɨkí “interactional”; =tɨkɨ́n “again”; =wɨstí “once”; =ɾaβɨ́ “a few times”; =wɨní “before someone”. In addition, all associated motion markers appear to be permitted in dependent clauses.

Other morphemes/categories vary in terms of whether they can occur in dependent clauses. Table 10 summarizes what morphemes can occur in what dependent clauses.

Table 10. Morphemes/categories across different dependent clauses.

	prior:ss	prior:sa	concur:ss	concur:sa	subseq:ss/a	interrupt:ss/a	quick:ss/a	prior:ds/a	concur:ds/a	subseq:ds/a	nmlz:agt	nmd	nmlz:purp
habitual	✓	✓	✕	✕	✕	✓	✓	✓	✕	✕	✓	✕	✕
allday	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✕
punctual	✕	✕	✕	✕	✕	✕	✓	✕	✕	✕	✓	✓	✓
go&do	✓	✓	✓	✓	✓	✓	✕	✓	✓	✕	✓	✓	✕
tomorrow	✕	✕	✓	✓	✓	✓	✕	✕	✓	✓	✓	✓	✕
recpast	✓	✓	✕	✕	✕	✕	✕	✓	✕	✕	✕	✓	✕
distpast	✓	✓	✕	✕	✕	✕	✕	✓	✕	✕	✕	✓	✕
remfut	✕	✕	✓	✓	✕	✕	✕	✕	✕	✕	✓	✕	✕
cntrfct	✓	✓	✓	✓	✓	✓	✕	✓	✓	✓	✓	✓	✕
abil	✓	✓	✓	✓	✓	✓	✕	✓	✓	✓	✓	✓	✕
limit	✕	✕	✕	✕	✕	✕	✓	✓	✓	✓	✓	✓	✓
distal	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✕	✓	✓
anxiety	✓	✓	✓	✓	✓	✕	✓	✓	✓	✓	✓	✓	✓
volitive	✕	✕	✓	✓	✓	✓	✓	✕	✓	✓	✓	✓	✕

Note that the table above refers to what morphemes speakers accept in dependent clauses, not necessarily what appears in natural speech. If one follows Bickel’s (2010) variable of finiteness, all dependent clauses in Chácobo are finite. However, one could consider finiteness to be a matter of degree relating to what categories can be expressed in the relevant dependent clauses. If one uses this notion, dependent clauses may vary in terms of how finite they are. The nominalized purpose clause would appear to be the least finite. Most other clauses are roughly the same, however.

5. Discussion I: Cross-Linguistic Assessment

In this section, I will briefly assess whether and how clause-linkage strategies in Chácobo pattern with embeddedness (coordination vs. subordination) from a cross-linguistic perspective. In order to do this, I leverage the data in the work of Bickel (2010) adding the Chácobo clause-linkage data.

A Neighbornet (see Bryant and Moulton 2004) was run with Bickel’s (2010) data combined with the Chácobo clause-linkage data. A graph similar to the one found in Bickel’s work (2010) is presented in Figure 1 below. This was constructed with the data available in the Supplementary Materials (Tallman 2024). The data for the cross-linguistic assessment do not contain any of the new variables that I introduced throughout this study, because these have not been coded with the languages of Bickel’s study. The following variables are therefore excluded: center-embedding via participant agreement; across-the-board extraction; asymmetric extraction of AdvP constituents; asymmetric extraction of NP constituents; and all the individual variables related to marking of specific morphemes. Extraction variables are all lumped into one variable extraction and the marking of functional categories is replaced with the variable finiteness. Ideally, we would apply the new variables across all the other languages (Bickel and Nichols 2002), but this is outside the scope of this paper.

Figure 1. Neighbornet using Bickel’s (2010) data with simplified Chácobo data added.

For the cross-linguistic data, a dissimilarity matrix was developed using the statistical software R (R Core Team 2014). I use Gower’s distance to calculate the dissimilarity matrix. Gower’s distance is a metric for computing the overall dissimilarity between two elements when the variables are of different types (see Gower 1971 for the mathematical details). It can also be used when the variables are categorical. A dissimilarity matrix is a matrix that gives measurements of distance between each of the datapoints in a data frame. The values are between 0 and 1, where 0 means that the clause-linkage constructions are identical with respect to the variables and 1 means that the clause-linkage constructions are as different as possible with respect to the variables. For instance, Chácobo’s concurrent same-subject clause linkage construction has a distance of 0.36 from English’s to -ing participle construction, but a distance of 0.73 from English’s and conjunction. This translates to the concurrent same-subject clause being more similar to the participle construction than to and conjunction in English.

A Zenodo file has the full dissimilarity matrix (Tallman 2024). Table 11 provides an example of a dissimilarity matrix with only 8 of the 82 data points/clause-linkage constructions.

Table 11. A dissimilarity matrix using Gower’s coefficient and data points from the work of Bickel (2010) with data from the current study.

	Chácobo_interrupt.ssa	Chácobo_prior.ds/a	English_and	English_to.ing.detached	German_and	Nepali_chain	Hua_ss-chain	Tauya_chain
Chácobo_interrupt.ssa	0	0.27	0.73	0.18	0.73	0.40	0.44	0.70
Chácobo_prior.ds/a	0.27	0	0.73	0.36	0.73	0.40	0.56	0.70
English_and	0.73	0.73	0	0.82	0.09	0.40	0.78	0.60
English_to.ing.detached	0.18	0.36	0.82	0	0.73	0.50	0.33	0.70
German_and	0.73	0.73	0.09	0.73	0	0.50	0.89	0.70
Nepali_chain	0.40	0.40	0.40	0.50	0.50	0	0.33	0.33
Hua_ss-chain	0.44	0.56	0.78	0.33	0.89	0.33	0	0.13
Tauya_chain	0.70	0.70	0.60	0.70	0.70	0.33	0.13	0

A dissimilarity matrix can be used in various exploratory data analyses. I use a Neighbornet here because it was used in Bickel’s original study. Neighbornets are also relatively common in typological research (see Cysouw 2007; Wichmann and Saunders 2007; Nichols and Warnow 2008; Donohue et al. 2011; Grünthal and Nichols 2016, inter alia).

A nexus file built from the dissimilarity matrix was then imported into Splitstree4 (Huson 1998; Huson and Bryant 2006). The resulting Neighbornet is provided in Figure 1.

Bickel (2010) suggests that his clause-linkage data may show clusters which could be called “prototypes”. I have annotated the domains which correspond to candidate clusters of subordinate and coordinate prototypes. Note that most clause-type strategies fall in between these categories and the Neighbornet is highly reticulated.

Two properties are notable in the Neighbornet above. First, the Chácobo clause-linkage strategies do not cluster with either of the prototypes. They appear to be in some intermundia between the two types. Secondly, they tend to cluster with each other (at the top of the graph), rather than falling within some well-defined category established by easily identifiable prototypes. Without choosing some arbitrary criterion for classification as either coordinate or subordinate, it is unclear whether they should be coordinate or subordinate with respect to typological patterns. The problem with such ambiguity for syntactic theory is that it could imply that claims about coordination or subordination specific constraints cannot be meaningfully assessed in Chácobo. For example, we cannot assess whether constraints on extraction in coordinate clauses can be assessed for any given instance of clause linkage (see Section 4.6 on this point as well), because we cannot be sure we are dealing with coordination rather than subordination (Tallman 2021a for discussion on the lack of falsifiability of theories that reify and presuppose traditional categories rather than seeking to test their validity).

On the other hand, the variables (diagnostics) that have thus far been constructed are perhaps too coarse-grained, relative to what may be needed to give a reliable assessment of cross-linguistic clustering. Future research may show that there are clearly distinct clusters once the variables have been better articulated and made more precise. Furthermore, given that the criteria crosscut a number of more fine-grained distinctions in cross-linguistic categories (control constructions, nominalized clauses, etc.), interpreting the results in terms of a binary distinction between coordination and subordination is perhaps misleading. These issues are discussed in more detail in Section 7.

6. Discussion II: Language-Internal Assessment

In this section, we will assess whether some type of fuzzy distinction can be made between coordinate and subordinate clauses in Chácobo based only on language-internal evidence. We will use a wider range of more detailed variables to make the assessment. As noted in the discussion above, a number of Bickel’s clause-linkage variables can be further split up into more variables.

To summarize again the distinction between the cross-linguistic database and the one specific to Chácobo, Bickel’s extraction is expanded and/or split up into wh-np-ext-main (extraction of an NP-constituent interrogative from the main clause in a clause-linkage construction), wh-np-ext-dep (extraction of an NP-constituent interrogative from the dependent clause in a clause-linkage construction), wh-adv-ext-main (extraction of an adjunct-constituent interrogative from the main clause in a clause-linkage construction), wh-adv-ext-dep (extraction of an adjunct-constituent interrogative from the dependent clause in a clause-linkage construction), wh-np-atb-ext (across the board extraction of an NP-constituent interrogative). Bickel’s finiteness is split up into variables for every single bound marker that can elaborate or modify a verb (e.g., habitual, punctual, completive, etc.). Bickel’s layer is reinterpreted and split into center-embed:case (can a dependent clause be skipped over by case) and center-embed:participantagreement (can a dependent clause be skipped over in participant agreement). Rather than referring to the selection of a verb or a clause, I refer to the constituent identified by treating the clause-linkage-strategy as a constituency test. Constituency test results refer to the left and right edges of groupings in relation to an array of sequentially ordered syntagmatic positions identified by treating the clause-linkage strategies as coordinative-based constituency tests. The reader should consult Tallman (2021b, n.d.) and the papers of Tallman et al. (n.d.) for details. Constituency test variables add min-left-edge (the left edge of the constituent identified by the clause-linkage strategy via a planar structure and a minimal/wide-scope interpretation), max-left-edge (the right edge of the constituent identified by the clause-linkage strategy via a planar structure and a minimal/wide-scope interpretation), min-left-edge (the left edge of the constituent identified by the clause-linkage strategy via a planar structure and a maximal interpretation), max-left-edge (the right edge of the constituent identified by the clause-linkage strategy via a planar structure and a maximal interpretation). Other variables that are added are whether the clause can have a noun-modifying function, whether the clause can be referential, and the behavior of negative-scope. Definitions for these variables are provided at the beginning of Section 4 and throughout the relevant sections as well.

Based on these variables, we will attempt to assess whether clause-linkage strategies in Chácobo can be grouped into coordinative and subordinate constructions overall. The strategy that will be used to make this assessment will be bottom-up hierarchical clusters. I assume that, if some general dichotomy between coordinate and subordinate clauses is motivated in Chácobo, the data will cluster into two groups better than chance. I argue based on a hierarchical cluster model of the data in relation to a simulated null hypothesis that no such better-than-chance partition is present.10

Section 6.1 will explain the logic and basic ideas behind agglomerative hierarchical clustering (see Borcard et al. 2018 for a practical introduction among many other sources). Readers who have some basic understanding of exploratory data analysis may want to skip this section.11 Section 6.2 will apply a confirmatory analysis which leverages the simulation of a “null” distribution (e.g., Spanos 2013) modelling a hypothetical situation where there is no distinction between coordinate and subordinate constructions. It is shown that the Chácobo data on clause-linkage are not sufficiently distinct from this null distribution, even while they show a strong tendency to cluster in general better than the simulated null. I do not mean to imply that this is the only method that one could use to assess the question. I am only suggesting that it may be a useful tool when the data display a high dimensionality (many different logically distinct variables), making a qualitative assessment more difficult.

6.1. Agglomerative Hierarchical Cluster on Clause-Linkage Constructions

The objective of a clustering methodology is to find groups in data based on some measure of similarity.12 There are a large number of clustering methods based on different algorithms and different notions of what it means for datapoints to be in a “group”. Based on the logic of the problem presented in this paper, I use an agglomerative hierarchical clustering. Agglomerative hierarchical clustering starts with the assumption that every datapoint is its own cluster and successively builds higher clusters from these by taking the clusters which are closest according to a dissimilarity metric and grouping them into a new cluster. The result is a tree structure with partitions at different levels of dissimilarity. The method is not new (Sokal and Sneath 1963) and has been widely used in many fields, including linguistics (e.g., Dagmar and Fieller 2014).

As stated above, to run a hierarchical cluster model we develop a dissimilarity matrix for the clause-type data frame. This is conducted as in Section 5, but with a larger dataset which contains all the variables which have been applied to Chácobo. I use Gower’s metric as in the previous section. The more fine-grained database used for this study including relevant R code is provided in Tallman (2024).

The distance matrix for the clause-linkage constructions in Chácobo is provided in Table 12.

Table 12. Dissimilarity matrix for clause-linkage constructions in Chácobo.

	prior:ss	prior:sa	concur:ss	concur:sa	subseq:ss/a	interrupt:ss/a	quick:ss/a
prior:ss	0	0	0	0	0.008547	0.470085	0.401709
prior:sa	0	0	0	0	0.008547	0.470085	0.401709
concur:ss	0	0	0	0	0.008547	0.470085	0.401709
concur:sa	0	0	0	0	0.008547	0.470085	0.401709
subseq:ss/a	0.008547	0.008547	0.008547	0.008547	0	0.461538	0.393162
interrupt:ss/a	0.470085	0.470085	0.470085	0.470085	0.461538	0	0.547009
quick:ss/a	0.401709	0.401709	0.401709	0.401709	0.393162	0.547009	0
prior:ds/a	0.162393	0.162393	0.162393	0.162393	0.17094	0.478632	0.487179
concur:ds/a	0.162393	0.162393	0.162393	0.162393	0.17094	0.478632	0.487179
subseq:ds/a	0.162393	0.162393	0.162393	0.162393	0.153846	0.461538	0.470085
nmd	0.179487	0.179487	0.179487	0.179487	0.188034	0.649573	0.504274
nmlz:agt	0.34188	0.34188	0.34188	0.34188	0.350427	0.65812	0.666667
nmlz:purp	0.581197	0.581197	0.581197	0.581197	0.57265	0.649573	0.487179
	prior:ds/a	concur:ds/a	subseq:ds/a	nmd	nmlz:agt	nmlz:purp
prior:ss	0.162393	0.162393	0.162393	0.179487	0.34188	0.581197
prior:sa	0.162393	0.162393	0.162393	0.179487	0.34188	0.581197
concur:ss	0.162393	0.162393	0.162393	0.179487	0.34188	0.581197
concur:sa	0.162393	0.162393	0.162393	0.179487	0.34188	0.581197
subseq:ss/a	0.17094	0.17094	0.153846	0.188034	0.350427	0.57265
interrupt:ss/a	0.478632	0.478632	0.461538	0.649573	0.65812	0.649573
quick:ss/a	0.487179	0.487179	0.470085	0.504274	0.666667	0.487179
prior:ds/a	0	0	0.017094	0.17094	0.179487	0.512821
concur:ds/a	0	0	0.017094	0.17094	0.179487	0.512821
subseq:ds/a	0.017094	0.017094	0	0.188034	0.196581	0.495726
nmd	0.17094	0.17094	0.188034	0	0.162393	0.683761
nmlz:agt	0.179487	0.179487	0.196581	0.162393	0	0.615385
nmlz:purp	0.512821	0.512821	0.495726	0.683761	0.615385	0

The R package cluster() is used to build a hierarchical cluster model (Maechler et al. 2022). An agglomerative hierarchical cluster model starts from the assumption that all clause-linkage types are their own cluster. Then, it groups each clause-linkage construction into larger clusters based on how similar they are. For instance, the first agglomeration of clusters would group the prior and concurrent same-subject clauses into the same group {prior:ss, prior:sa, concur:ss, prior:sa} because they are maximally similar with respect to variables used in this study. The algorithm would then join the cluster {prior:ss, prior:sa, concur:ss, prior:sa} with the {subseq:ss/a}, because this is the shortest distance from the second cluster 0.008547. The prior and concurrent different-subject clauses would also group together since they have a relative distance of 0. Then, the {prior:ds/a, concur:ds/a} cluster would merge with [subseq:ds/a] because the distance is the next lowest at 0.017094. Agentive nominalizers would then merge with the noun-modifying construction because the distance is the next lowest 0.162393. The different-subject clauses {subseq:ds/a, prior:ds/a, concur:ds/a} would then merge with the cluster that contains noun-modifying and agentive-nominalizing constructions {nmd, nmlz:agt}. Figure 2 is a dendrogram that represents the clustering process. The nodes in the tree represent clusters. The y-axis represents their relative distance the cluster have from one another.

Figure 2. A dendrogram showing the agglomeration of different subject clauses with nominalized and noun-modifying clauses.

The clustering process continues until all clusters are merged into a single cluster. The dendrogram classifying all the dependent clauses in Chácobo is provided in Figure 3 (de Vries and Ripley 2022 for the R package).

Figure 3. Dendrogram for Chácobo-dependent clauses using all coded variables (see the Supplementary Materials).

Figure 3 shows a split with the quick-succession asyndetic clause, the same subject interruption clause, and the nominalized purpose clause in the one cluster, and the rest of the clauses on the other at the highest partition. This is because the asyndetic same subject clause and the purpose clause share a number of structural properties, not shared by other clause-linkage strategies. Neither of these clauses allow for temporal distance markers (T-mark), but all other dependent clauses do (Section 4.10). Both of these clauses are unique among Chácobo clause-linkage constructions in banning extraction of any NP arguments out of their dependent conjuncts (Section 4.6). Both of these clauses are unique in not containing their own information structurally important position independent of the main clause (Foc). Furthermore, compared to the rest of the constructions, these three strategies are relatively non-finite in the sense of having less categories overall that can be expressed in them compared to the other dependent clauses. It is the combination of these properties that they share not shared by the rest of the dependent clauses that is responsible for them being grouped together. These constructions group together despite the fact that, overall, they fall on opposite ends of the continuum from coordination to subordination. Out of the clause-linkage strategies, the asyndetic quick succession clause construction could be considered the most coordinate-like. It has a fixed position with respect to the main clause and it bans the asymmetric extraction of core arguments out of any of the conjuncts (see the coordinate structure constraint; Ross 1967, inter alia). Despite its relative similarity to the asyndetic construction, the purpose clause has more subordinate-like patterns: illuctionary operators cannot scope over it (it cannot be questioned material), and while asymmetric extraction is allowed in purpose constructions, it can only occur out of the main clause, which is what one would expect if it was an adjunct clause (Bošković 2020).13

Thus, the first partition may not reflect the best candidates for the distinction between coordination and subordination. If most Chácobo clauses were structurally intermediate between these two types (see Section 5), this may be expected. This general idea can be observed if we attempt to rank clauses in terms of the coordinate–subordinate distinction.

If we recode the data in terms of integer values, where 1 is provided to the more subordinate-like value, and 0 is the more coordinate-like value for a given property, a non-finiteness value is calculated by taking the average value of all of the TAAMME values. We can thus construct a subordination metric for each construction by summing over the variables. Higher values are more subordinate in terms of the variables of the study.

If we plot the constructions in terms of their subordination value, we see that the partition above could be seen as corresponding to a distinction between coordinate and subordinate constructions.

One important point to make in relation to the graph below, however, is that agentive nominalizations and the nominal modification clauses are nearly the same as same subject clauses in terms of the subordination metric respectively. We may wonder whether the precise dividing line between coordination and subordination is arbitrary given the intermediate status of same and different subject clauses (see Figure 4).

Figure 4. Strip plot for Chácobo-dependent clauses ranking them on a scale from coordination to subordinate like by recoding all variables that are relevant for this distinction as 0 = coordination-like and 1 = subordination-like values and summing the values for each dependent clause (see the Supplementary Materials).

6.2. Simulated Null (Testing the Coordination–Subordination Distinction)

Cluster models cannot be used as an inferential technique by themselves but are rather tools of exploratory data analysis. In order to make an argument about groupings from a cluster analysis, we need to minimally construct a type of null hypothesis against which to gauge the results. Otherwise, we may fall into the “clustering tendency” fallacy, which involves seeing groups in arbitrary partitions of the data (Jain and Dubes 1988). A clustering model will always find groups tautologically. The question is whether these groups represent some surprising result from what we would expect if the data did not, in fact, cluster.

In order to investigate the possibility that there may simply be a continuum between coordination and subordination, we see whether the data are being split into two types better than chance in the hierarchical cluster model. In order to do this, we simulate vectors of the clause-linkage variables of our study where the values are assigned randomly (see Supplementary Materials for details and code).

The notion of “chance” and the randomness requires some comment here. As stated above, the cluster model is designed to find patterns (groups) in the data regardless of whether these are natural or meaningful. A hierarchical model will group clause-linkage constructions into two groups at its highest level regardless of whether these are meaningful. A first question that can be asked to assess the meaningfulness of the groupings found in the data is how likely such patterns are to occur even if our hypothesis about the bipartite division of clause-types was false. One way of doing this is to compare our data to a hypothetical distribution (a null hypothesis) simulated to correspond to a situation where there is no meaningful bipartite pattern. Such a technique is commonly used across the sciences to make inferences about the validity of quantative hypotheses (see Spanos 2013).

There is more than one way to construct a null hypothesis. We could, for instance, assume that all values for any given variable are just as likely to occur and build a simulated null hypothesis from that. In this study, I will sample from the Chácobo data themselves to construct a null distribution. This means that, for instance, for a given variable, if a value val occurs in 2 out of 13 clause-linkage constructions, then val will be sampled with a 2/13 (=0.15) probability. To create a null distribution, we sample “random languages” from the Chácobo clause-linkage data. The difference between the actual data and the simulated data is that there is no reason to expect that variables will covary with one another in a random language except those patterns that appear due to chance relations that will appear between variables (Roberts and Winters 2013).

Note that, when we compare with our null distribution, we will not be concerned with how many clusters are found per se. Rather, we will be concerned with the distance between the two largest clusters after the first partition. This is based on the assumption that, if a coordination–subordination distinction is to be found, it will be the highest partition in the data, a position which follows logically insofar as clause-linkage variables are relevant for making the distinction between coordination and subordination. This of course assumes that the variables we chose are the right ones for assessing this question. Future research may find that one or more of the variables ought to be removed from consideration or that there are yet more relevant variables that have not been coded.14

To create a null distribution, I simulated 1000 random languages and measured their cophonetic distance and the relative height of their first partition of each of the simulated languages. The cophonetic difference is a measurement of how well a given hierarchical cluster model fits the data points (see Sokal and Rohlf 1962; Rohlf and Fisher 1968 for explanation). This measurement tells us how well the data cluster into the groups of the cluster model overall. The reason I present this is to point out that the dendrogram is capturing actual groups in Chácobo, regardless of whether those reflect a bipartite division.

Probability density distributions for random/simulated cophenetic correlations and random/simulated height differences between the first and second partition are provided in Figure 5 below. The red dot over each of these distributions is where the value is for the actual data. The cophenetic correlation of the real data is 0.9138, showing a strong tendency for clause-linkage constructions in Chácobo to cluster with respect to the clause-linkage variables. However, the tendency for clause-linkage constructions to cluster into two groups is 0.1989.

Figure 5. The probability density distributions of cophonetic correlations and height differences between the first and second partitions of simulated hierarchical cluster models based sampled randomly from the clause-linkage data of Chácobo. The red dots represent the values for the hierarchical cluster over the actual data.

The first density distribution shows that the Chácobo clause-linkage data tend to show clustering relatively well. None of the simulated values occur with a value this extreme. This is not too surprising. The clause types can be classified into different groups and the hierarchical cluster meaningfully captures that.

The second density distribution shows that Chácobo clause-linkage strategies do not tend to group into just two clusters better than chance. A total of 33% of the simulated values have stronger first (binary) partitions than the actual data or, stated another way, 33% of the time, a language created through randomly sampling from the data would cluster as well or better into two groups than the actual data. This may indicate a weak tendency to cluster into two groups, and overall, the distinction is not well supported. It is possible that further refinement of the variables may provide support for such a distinction.

7. Conclusions

This paper has provided the first detailed description of clause-linkage strategies in Chácobo. The variables described in the work of Bickel (2010) were used and expanded on and applied to each of the clause types with data from naturalistic speech where available and data from elicitation where necessary. Contextualizing the Chácobo data with respect to that in Bickel’s (2010) sample shows that most Chácobo clause-linkage strategies do not fall into candidate “prototype” subordinate or coordinate clusters, to the extent that such notions can be validly inferred from Bickel’s (2010) database to begin with. One may defend the coordinate–subordinate distinction in Chácobo on language-internal grounds. Indeed, such a distinction is made in Tallman’s (2018a) descriptive grammar of the language. However, a detailed overview of the variables and an attempt to motivate the distinction using all of these rather than simply cherry-picking a few shows that such a language-internal classification may do more to obfuscate rather than clarify the differences and similarities between Chácobo’s clause-linkage strategies. The clustering and simulation methods suggest as much. This paper thus provides an example where comparative work can help clarify issues in language description.

One of the methodological innovations of autotypology (Bickel and Nichols 2002; Witzlack-Makarevich et al. 2021) is to build variables from the bottom-up and capture linguistic variation at the finest degree of detail. In this approach, typologically informed research on individual languages should go beyond simply coding the results of a predefined questionnaire. Rather, when it is noticed that a variable may give ambiguous or uncertain results, the descriptive linguist seeks to refine and expand the variables accordingly, fracturing them into new variables, if necessary to capture the language-specific details. In this study, I have expanded the variables for center-embedding, extraction, and added a few variables such as nominal modification and referentiality. Future research should try to code these variables in more languages, where possible, and also expand the variables further.

Some of the new variables require commentary as issues of comparability arise. First, this paper experiments with quantizing the finiteness variable. This is achieved by taking all the grammatical categories that can be expressed in each dependent clause and assigning each clause-linkage strategy a relative value on a scale of 0 to 1 for how finite they are, where 0 is no categories can be expressed and 1 is all categories can be expressed. This is a very rough metric that burries some potential variation. Future research may treat every marker separately, rather than aggregating them in this fashion. The problem with this proposal, however, is that languages vary in terms of their TAME systems15 and thus issues of comparability could arise. Perhaps a quantized assessment of finiteness requires positing a typological inventory of explicitly defined TAME variables.

The extraction variables may also suffer from issues of comparability but for other reasons. Extraction or filler-gap constructions are not always an easily applicable diagnostics because their assessment needs to be mediated by what types of null elements are posited in the language. I pointed out that the availability of null third person object pronouns made some claims about NP-extraction difficult to assess. Some strict criteria perhaps need to be put in place so that filler-gap constructions can be interpreted consistently from one language to another. Another issue is that extraction constructions are typically subject to degrees of acceptability. Eventually, the typological researcher needs to find some way of systematically coding the graded nature of acceptability.

This paper has operated under a relatively simple hypothesis about the distinction between coordination and subordination. I assumed that the clause-linkage variables above were relevant to making this distinction. I assumed that a coordination–(adjunct) subordination would arise as clusters because coordination and subordination are typological prototypes or “attractor points”.16 However, the “fitness landscape” for clause-linkage constructions is likely more complex than this. A more sophisticated typology of clause-linkage constructions is likely required (e.g., Croft 2001).

A question arises as to what explains the clause-linkage patterns that do exist in Chácobo. We saw in Section 5 that same-subject and different-subject constructions tend to cluster together even while clause-linkage constructions of other languages were more dispersed across the Neighbornet. One potential reason for this is that same/different subject clauses may come from the same basic diachronic source construction. It has been posited that switch reference in some Pano languages may derive from clausal nominalizations (Valle and Zariquiey 2019), and indeed many of the same/different subject markers seem to derive from postpositions (Valenzuela 2003). Drawing the link between the diachronic development of same/different subject clauses in Pano and the properties they display as clause-linkage strategies may be the next step, but this will require the relevant clause-linkage properties to be documented in other Pano languages.

Appendix A.1. Null Pronouns and the Interpretation of Asymmetric Extraction of NP Constituent Interrogatives in Chácobo

There is an analytic issue that potentially confounds the interpretation of asymmetric extraction in Chácobo. Third-person Chácobo objects are null and Chácobo has null third person pronouns as well. This can be observed in the following sentence:

100.	∅	∅	kɨɨs-a	=kɨ
	3sg:nom	3sg:abs	cut-tr	=decl:pst
	“S/he cut it/her/him/them.”

If, however, there is an overt object, speakers will reject the sentence unless it also has an overt subject, as shown in17. Stated another way: when the object is overt, the subject must also be overt.

101.	mɨɨʃi	*ha	kɨɨs-a	=kɨ
	branch	3sg.nom	cut-tr	=decl:pst
	“S/he cut the branch.”

This generalization does not hold of same-subject clauses, however, which can never have their own subject independent of the main clause. Furthermore, in different-subject clauses, subjects must be overt.

There is no evidence for or against Chácobo allowing object pronominal resumptive in the case of constituent interrogative extraction (e.g., What did you tell him to eat it?); however, if such an analysis is permitted of null anaphora, then NP extraction in same-subject clauses cannot be meaningfully assessed. For instance, the gaps could be analyzed as null (resumptive or resumptive-like) pronouns. I have no way of discarding such an analysis, in principle, when dealing with third-person objects. This makes the assessment of some NP-extraction in same-subject and different-subject clauses difficult because the construction already places constraints on the coreferentiality of subjects and thus we cannot extract NP subjects either.

102.	hawɨi	kako= ́	___i	kopi=ʔáʂ	oʂa=ʔá
	whati	Caco=erg	___i	buy=prior:sa	sleep=inter:pst
	“What did you Caco buy and then slept?”

103.	hawɨi	___i	ʃiná=ʔi	kako	oʂa=ʔá
	whati	___i	think=concur:ss	Caco	buy=inter:past
	“What did Caco think about when he slept?”

Finally, another analytic issue arises here as well. In cases such as those above, it is unclear whether we are dealing with extraction per se because fronting internal to a same subject clause is permissible (see Section 4.6). Thus, it is not clear whether the fronting in the examples above constitute fronting over a clause boundary or not. In addition to the possibility of positing null resumptive pronouns, the issue of clause-internal movement makes the assessment of NP extraction in same-subject clauses difficult to assess. These caveats suggest that perhaps future research may require reformulation of the extraction variables into concepts whose coding is not so heavily mediated by other auxiliary hypotheses or assumptions.

All cases where extraction is allowed could be reanalyzed as cases where there is a null resumptive pronominal element. While such an analysis is difficult to refute conclusively, the problem with it is that it would be unclear why the null resumptive pronoun would not be available in cases where extraction is apparently banned, either completely or non-locally. If a null resumptive pronoun is available, why can it not be posited in examples such as 74? Another reason for rejecting a null resumptive pronoun analysis involves the possibility of extracting subject NPs non-locally from different-subject clauses. Different-subject clauses are normally rejected if they do not contain an over subject. However, if the subject is extracted to the initial position, the sentence is accepted, even if it crosses a main clause in doing so. If null resumptive pronouns are confounding, it is unclear why the subject pronoun ha results in ungrammaticality even while overt subjects are otherwise obligatory in different-subject clauses.

104.	tsowɨi= ́	kará	hawɨ	βakɨ́	pístia	háβa	=kɨ
	whoi=erg	epis	3sg:gen	child	small	run	=decl:pst
	___i/*ha	kamá		tʃoiʃ-a		=kɨ́na
	___i/3sg	jaguar		shoot-tr		=prior:ds/a
	“Who could have shot the jaguar before its little child ran away?”

The important issue is perhaps not whether these are definitely cases of extraction sensu stricto, but rather, whether in applying the test as we have done here we are capturing meaningful typological variation. Even if we claim that some resumptive null pronouns confound our interpretation, we will still have to posit that null resumptive pronouns are allowed in some cases but banned in others and still could be formulated as variable for clause-linkage constructions as well (it involves just as much stipulation).

Cases of local extraction are even more difficult to assess due to the possibility that the constituent interrogative does not actually cross a clause boundary, and the fronting applies clause-internally. I would suggest that future research should reformulate these variables in less theory-bound ways.

Appendix A.2. Across-the-Board Extraction of NP-Constituent Interrogatives

Across-the-board (ATB) extraction has been identified as one of the criteria for identifying coordinate structures (Ross 1967; Bošković 2020). Coordinate structures cannot undergo asymmetric extraction whereby an element is extracted out of just one of the conjuncts, but if the same element is extracted or appears gapped out of both, then the extraction is permitted. Illustrative examples of this ungrammaticality are provided below. But the extraction of both of these elements from the conjunct is permissible. An illustrative example is provided below.

105.

*Whom_i did Caroline wish to meet ____i and talk to Mery in La Paz?

106.

*Whom_i did Caroline wish to meet Mery and talk to ____i in La Paz?

107.

Whom_i did Caroline wish to meet _____i and talk to _____i in La Paz?

Since ATB extraction has been identified as a diagnostic for coordinate status, it is worth seeing whether it can be applied as a typological variable to the Chácobo clause-linkage constructions.

Before attempting to do this, I will briefly describe why ATB extraction is hard to apply as a diagnostic to Chácobo in practice. First, in Chácobo, third-person pronominal object arguments have null realizations, as described in Section 4.6. Due to this, we cannot use the extraction of a P argument to test ATB extraction. To illustrate this, consider the following sentence:

108.	háwɨi	____i	aʂa=ʂó	mi	____i/∅i	pi	=ʔá
	what	____i	roast=prior:sa	2sg:nom	____i/ 3:absi	eat	=inter:pst
	“Whati did you roast ___i and then eat ___i?”

We do not know whether the gap in the second conjunct is phonetically null because the interrogative constituent hawɨ “what” moved from this position or whether it is a third-person object, because the latter are null. As stated in Section 4.6, if, however, there is an overt object, the subject is now obligatorily overt as is repeated again below.

109.	mɨɨʃi	*ha	kɨɨs-a	=kɨ
	branch	3sg.nom	cut-tr	=decl:pst
	“S/he cut the branch.”

Due to this property, we could potentially test ATB-extraction by extracting the subject but with overt object arguments in both conjuncts, since these are generally disallowed. However, it is not clear whether same-subject clauses can be tested. While we can extract the subject of the main clause to the first position of the sentence, it is not clear whether this constitutes ATB extraction, because same-subject clauses cannot have an overt subject anyway.

To illustrate the problem, consider the following sentence, which constitutes an attempt to elicit ATB extraction. The sentence is considered ungrammatical. Thus, we may claim that ATB extraction does not apply to prior same-subject clauses.

110.	*tsowɨi= ́	___i	nami	aʂa=ʂó	___i	sanino	pi	=ʔá
	whoi=erg	___i	meat	barbeque=prior:sa	___i	fish	eat	=inter:pst
	“Whoi ___i roasted the meat and then ___i ate the fish?”

However, given that subjects cannot be overt in the same-subject clause anyway, we may suppose that same-subject clauses are control structures perhaps with an obligatory pro element in subject position, as depicted here. Rather than being an instance of ATB extraction, we simply have extraction over the same-subject clause which has a null pro element.18

111.

Thus, it seems that ATB extraction may be unavailable as a diagnostic for core arguments in Chácobo, at least for same-subject clauses. For completeness, I include cases wherein the subject is extracted in this fashion for same-subject clauses anyway. Concurrent same-subject clauses are marginally acceptable. Asyndetic clauses allow for the extraction of the subject in the first position. The same subject interruption clause also allows for the extraction of the subject.

112.	?tsowɨi= ́	___i	hɨnɨ	aʂa=kí	___i	sanino	pi	=ʔá
	whoi=er	___i	meat	barbeque=concur:sa	___i	fish	eat	=inter:pst
	“Whoi ___i ate fish while ___i barbaqueing meat?”

113.	tsowɨi= ́	___i	waka	kiʃi	atʃ-a	___i	hawɨ	taʔɨ
	whoi=erg	___i	cow	leg	grab-tr	___i	3sg:gen	foot
	nɨʂ-a	=ʔá
	tie-tr	=inter:pst
	“Whoi grabbed the cow’s leg and tied his trotter?”

114.	tsowɨi= ́	___i	atsa	pi=pama		___i	nami
	whoi=erg	___i	yuca	eat=intrmp:ss/a		___i	meat
	mɨʃ-o	=ʔá
	burn-intr	=inter:pst
	“Whoi while ___i roasting the yuca ___i burnt the meat?”

115.	tsowɨi= ́	___i	atsa	pi=pama		___i	nami
	whoi=erg	___i	yuca	eat=intrmp:ss/a		___i	meat
	mɨʃ-o	=ʔá
	burn-intr	=inter:pst
	“Whoi while ___i roasting the yuca ___i burnt the meat?”

ATB extraction cannot easily be assessed for different-subject constructions for the simple reason that the conjuncts require different subjects. If we attempt to extract the subject, the sentence is grammatical, but not under an interpretation where the subjects are coreferential. Rather, the sentence must be understood as asking a question about two distinct subject participants (i.e., “who … “ and “who else …”). An example is provided here. This is true of all the different-subject constructions.

116.	tsowɨ= ́	atsa	ima=kɨ́		tsi	nami
	who=erg	yuca	roast=prior:ds/a		lnk	meat
	pi	=ʔá
	eat	=inter:pst
	“Whoi ___i roast yuca and who else ate the meat?” *”Whoi ___i roast yuca and ___i ate the meat?”

Nominalized and nominal modifier clauses behave in the same way. The fronted constituent interrogative cannot function as a coreferential subject of both of the conjuncts.

117.	tsowɨ= ́	atsa	ima=ʔái=ka			nami
	who=erg	yuca	roast=nmlz=rel			meat
	pi	=ʔá
	eat	=inter:pst
	“Whoi ___i was roasting yuca and who else ate the meat?” *”Whoi ___i roast yuca and ___i ate the meat?”

Since in other contexts, nominalized and nominal-modifying clauses can have the same subject as the subject of the main clause, the example above shows that ATB extraction is not permitted in these clauses. ATB extraction appears to be permitted in purpose clause constructions.

118.	tsowɨi = ́	tapaya= ́ =ka		___i	oɾikití		inia=tí
	whoi=erg	Alto.Ivon=spat=rel		___i	food		sell=nmlz:purp
	___i	aros	kopi	=ʔá
	___i	fish	buy	=inter:pst
	“Who bought rice to sell in Alto Ivon.”

If we were to decide that the ATB extraction was the only relevant criterion for distinguishing different clause-linkage types, we may conclude that agentive nominalized and nominal-modifying clauses are subordinate in some sense. As for the nominalized purpose clause, it could be an adjunct clause which, like coordination, can display ATB extraction (Bošković 2020). However, we would not know how to characterize the rest of the clauses because the NP extraction cannot be easily applied. Future research on across-the-board extraction of adverbial elements is still required, however. This paper summarizes the results of attempting to apply ATB extraction to the clause-linkage variables in Chácobo. As stated above, this variable was not coded in the work of Bickel (2010).

Table A1. The results for the atb-extraction variable applied to clause-linkage constructions.

Form	Gloss	Value
ʔáʂ(na)	prior:ss	not applicable
ʂó(na)	prior:sa	not applicable
ʔi(na)	concur:ss	not applicable
kí(na)	concur:sa	not applicable
́noʂpaɾí	subseq:ss/a	not applicable
pama(ʔáʂ)	interrupt:ss/a	not applicable
(asyndetic)	quick:ss/a	not applicable
kɨ́(no)	prior:ss/a	not applicable (disjunct interpretation)
́no	concur:ss/a	not applicable (disjunct interpretation)
́noʂpaɾíno	subseq:ss/a	not applicable (disjunct interpretation)
ʔá(ka(to)/na)	nmd:ant	banned (disjunct interpretation)
ʔái(ka(to)/na)	nmlz:agt	banned (disjunct interpretation)
tí	nmlz:purp	ok

Table A1. The results for the atb-extraction variable applied to clause-linkage constructions.

Form	Gloss	Value
ʔáʂ(na)	prior:ss	not applicable
ʂó(na)	prior:sa	not applicable
ʔi(na)	concur:ss	not applicable
kí(na)	concur:sa	not applicable
́noʂpaɾí	subseq:ss/a	not applicable
pama(ʔáʂ)	interrupt:ss/a	not applicable
(asyndetic)	quick:ss/a	not applicable
kɨ́(no)	prior:ss/a	not applicable (disjunct interpretation)
́no	concur:ss/a	not applicable (disjunct interpretation)
́noʂpaɾíno	subseq:ss/a	not applicable (disjunct interpretation)
ʔá(ka(to)/na)	nmd:ant	banned (disjunct interpretation)
ʔái(ka(to)/na)	nmlz:agt	banned (disjunct interpretation)
tí	nmlz:purp	ok

Appendix A.3. TAME Marking Supplementary Data

Appendix A.3.1. Tense and Temporal Distance Marking and Scope

In Chácobo, tense is coded on the clause-type marker. Clause-type markers are not marked on dependent clauses (by definition). Apart from tense, Chácobo has another category of temporal distance. Temporal distance markers code the degree of remoteness of an event from the utterance time or from a temporal reference time (Tallman 2018a; Tallman and Stout 2016).

Although it is rare in naturalistic speech, same-subject clauses can have temporal distance markers. An example is provided below with a prior-event same-subject clause.

119.	ha	βo=ʔita=ʔá=ka		tʃaʔa
	3	carry.away=recpast=nmd:ant=rel		feather.head.band
	sawɨ=roʔá		kiá	raka=roʔá
	put.on=limit		report	lie.down=limit
	kɨtóma	kiá	hatsi	ʂo	naa
	patio	report	then	decl	dem:prox
	a(k)=ʔita=ʔáʂ		hatsi	ʂo	naa
	do=recpst=prior:ss		then	decl	dem:prox
	i	kiá
	say	report
	“After he (the man) brought the feather head band, she (the panther) put it on and she (the panther) was lying in the patio, this (the panther) was the one who killed him (the man).” 0056:0140

Despite being rare in natural speech, same-subject clauses with temporal distance markers are fairly easy to elicit as seen below.

120.	yonoko	=itá/=yamɨtá		=ʂó	tsi
	work	=recpast/=distpast		=prior:sa	lnk
	taʃi= ́	anó	tsayakɨ
	Tashi=erg	taitetu	see=decl:pst
	“After he had worked (the day before/a week before), Tashi saw the taitetu.”

121.	ʂoβo	tsaya	=ʃari	=kí	tsi
	house	see	=cras	=concur:sa	lnk
	anó	tsaya	taʃi= ́	=wa	=kɨ
	agouti	see	Tashi=erg
	“Tashi visited the house the day after seeing the agouti.”

122.	ʂoβo	tsaya	=ʃari	=pama	tsi
	house	visit	=cras	=intrmp:ss/a	lnk
	taʃi	pakɨ́=kɨ
	Tashi	fall=decl:pst
	“Tashi wanted to see the house the day after he fell. (he wasn’t able to visit the house as a consequence of falling).”

123.	yonoko	=ʂɨ	=ʔi	tsi	taʃi
	work	=remfut	=concur:ss	lnk	Tashi
	tsaʔó	=kɨ
	sit	=decl:pst
	“Tashi sat down in order to think about working in the future.”

Prior same-subject clauses =ʔáʂ(na) and =ʂó(na) cannot occur with future/posterior temporal distance markers such as =ʃarí “tomorrow” and =ʂɨ́ “remote future”, and concurrent same-subject clauses ʔi(na) and kí(na) cannot combine with past tense/anterior temporal distance markers =ʔitá “recent past” and =yamɨ́t “distant past”. No same-subject clause can be marked by the remote past =ní. Asyndetic (same-subject) clauses cannot occur with temporal distance markers.

I have no examples in my corpus of different-subject markers clauses being marked by temporal distance markers. I tested whether temporal distance markers could appear in different-subject clauses. Only different-subject clauses marked with =kɨ́ “prior, different S/A” can appear with a temporal distance marker. And the only temporal distance markers which are compatible with this clause are =ʔitá “recent past” ~ =yamɨtá “distant past”, as seen below.

124.	ʂoβo	tsaya	awini=́	wa
	house	see	wife=erg	tr
	=ita/=yamɨta		=kɨ́	tsi	taʃi= ́
	=recpst/=distpst		=prior:ds/a	lnk	Tashi=erg
	raaʔá	=kɨ
	scold	=decl:pst
	“After his wife visited the house (the day before/the week before), Tashi scolded her.”

Nominalized and noun-modifying clauses marked with =ʔá(na) and =ʔái(na), respectively, can be marked by the temporal distance markers. A dependent clause marked with =ʔá(na) can be marked by the past tense/anterior temporal distance markers. The nominalized clause marked with =ʔái(na) can be marked with the posterior/future tense temporal distance markers. Examples with =ʔái(na)-marked clauses are provided below.

125.	mi	a=ʃaɾi=ʔái=ka		ʃina=ʔá
	2sg	do=cras=nmlz=rel		think=inter:ant
	mí-a
	2sg-epen
	“What are you thinking of doing tomorrow?”1835:0099

126.	naa	naʔa=ʂó	ɨarí	pi=ka(s)=ʃaɾi=ʔái=ka
	dem:prox	nest=a	1sg-too	eat=want=cras=nmlz=rel
	a(k)=wɨ́
	do=imper
	“Roast this which I will eat in my nest.”0483:1109

127.	naa	no	wɨsti	yói=ka	…
	dem:prox	1pl	one	bad=rel	…
	pakɨ́=noʂparí		no	rɨso=ʂɨ=ʔái=ka
	fall=subseq:ss/a		1pl	kill=remfut=nmlz=rel
	o	no	rɨso=tí=ka		βári
	or	1pl	die=nmlz:purp=rel		day
	yói=ka	…	saipí	nokí	a=tí=ka
	bad =rel	…	machete	1pl-acc	do=nmlz:purp=rel
	maní	o	rono= ́	nokí	tɨʂa=tí=ka
	conject	or	snake=erg	1pl-acc	bite=tr=nmlz:purp=rel
	o	…	mi	naama=ní
	conject	…	2sg	dream=nmlz:rempst
	Jɨmá	naama=ʔaí		hɨma= ́
	Jema=voc	dream=inter:nonpast		Jema=voc
	tóa=kato
	dem:dist=rel
	“When one of us due to something bad, we fall, that we will die, or the day or our death, or something bad, that a machete would cut us, that a snake could bite us, Jema, have you ever dreamed of this, Jema, have you ever dreamed of this?”1851:150

128.	toa	mí	…	mí	…
	dem:dist	2sg	…	2sg	…
	naama=ʔáina		…	ɨ	…
	dream=nmlz		…	1sg	…
	mi	oʂa=ʔái=ka		hawɨ́	…
	2sg	sleep=nmlz=rel		3sg:gen
	yonoko=ʃaɾi=ʔái=ka		mi	naama=ʔái=ka
	work=cras=nmlz=rel		2sg	dream=nmlz=rel
	…	toβi		ní	mi
		dem:dist:deic		inter	2sg
	naama=ʔáina
	sleep=nmlz
	“Are you dream of this? or when you are dreaming, what work you will do tomorrow, or what are you dreaming about?” 1838: 208

Examples with ʔá(na) “noun modifying anterior”-marked dependent clauses marked with temporal distance markers are provided below.

129.	naa	ɨ	βotɨ=yamɨt =(ʔ)á=ka		inoma
	dem:prox	1sg	descend=distpst=nmd:ant=rel		contrast
	motoro = ́		ɨ	βotɨ=yamɨt=(ʔ)á=ka
	motor=spat		1sg	descend=distpst=nmd:ant=rel
	pɨ	…	hɨnɨ= ́=ka		hawɨ́
	anx	…	water=spat=rel		3sg:gen
	ʂoβá	ɨá	atʃ-a=yamɨ́(t)=kɨ		iwɨ́
	with	1sg-acc	grab-tr=distpst=decl:pst		you.know
	“When I went down, when I went down in the motor in the water, it grabbed me, you know.”1849:0258

130.	wɨakɨ́	kiá	hawɨ	ɨwatí
	tomorrow	report	3sg:gen	gra.mo
	tɨɨ-wa	=ʔita	=ʔá	=ka	a(k)
	sweet-vblz	=recpst	=nmd:ant	=rel	drink
	=yáma	=ki	mi-a
	=neg	=decl:pst	2sg-epen
	“Tomorrow, after your grandmother has made chicha, you are not going to drink to.” 635:699

Clauses marked with the purpose nominalizer =tí cannot occur with any temporal distance markers. For instance, a conceivable purpose clause such as tsaya=ʔita=tí “to have seen yesterday” is ungrammatical (does not occur in my corpus and is not accepted by speakers in elicitation). Note that the absolute tense of a clause is provided by the clause-type morpheme of the main clause =kɨ “declarative past”. The temporal distance marker =ʂɨ́ “remote past” encodes the relative past with respect to the absolute tense established by the clause-type marker of the main clause. This is illustrated in the example below from elicitation.

131.	ʂoβo	tsaya	=ʂɨ	=ʔái	=ka
	house	see	=remfut	=nmlz	=rel
	yoʂa = ́	hɨnɨ	a(k)=kɨ
	woman=erg	chicha	drink=decl:past
	“The woman that had plans to see the house in the future, made chicha.”

The example above shows that markers of absolute tense of the main clause always scope over dependent clauses. The same is true of temporal distance markers of main clauses as well. For example, the dependent clause imaʂó “after roasting” takes on the temporal distance of the main clause, coded by the marker =ní “remote past”.

132.	ʔima	=ʂó	tsi	no
	roast	=prior:a	lnk	1pl
	ho	=ní	=kɨ
	arrive	=rempst	=decl:pst
	“We roasted and then we returned.” 0027:0025

Appendix A.3.2. Aspect Marking

Chácobo has a number of aspectual markers. In this section, I will discuss =yó “completive”, =paó “habitual”, =βaʔiná “during the day”, =ʃiná “at night, while dreaming”, tɨkɨ(n) “frequentive, iterative”. Below, I assess which of these can occur with which dependent clause.

=yó “Completive”

The marker =yó encodes that the event of the verb it modifies is complete. If the verb is transitive, it encodes that the P argument is completely affected. If the verb is intransitive, it encodes a universal quantifier over the subject translated as “all”. The marker =yó can occur in all dependent clauses. Examples of the completive occurring as a prior-event same-subject clause are provided here.

133.	matoʂ	=yo	=ʂó	tsi	nó
	dice	=cmpl	=prior:sa	lnk	1pl
	hɨnɨ́	páʂa	no	βí	=kɨ
	chicha	raw	1pl	grab	=decl:pst
	“After dicing everything, we brought water up.” 1156:0089

134.	toa	a(k)	=yo	=ʔaʂ	tsi
	dem:dist	do	=cmpl	=prior:ss	lnk
	kiá	nɨáma	=ka	honi	tɨʂɨβo
	report	far	=rel	man	other=pl/assoc
	ʂoβo	wɨtsa	=ka	=βo	ho
	house	other	=rel	=pl/assoc	come
	=yo	=ní	=kɨ
	=cmpl	=rempst	=decl:pst
	“After doing this (making all the chicha), the men from far away from other clans, all arrived.” 0013:0111

The completive can also occur in nominalized clauses, as seen below.

135.	ɑʃina = ́	kiá	tʃiʔi	=yá	i
	Ashina=erg	report	fire	=comit	see
	=pao	=ní	= kɨ	rɨkɨβa= ́
	=habitual	=rempast	=decl:pst	ancestors=erg
	ʃinó	βari	hana	=ʂó	=ɾoʔá
	monkey	sun/day	leave	=prior:sa	limit
	pii	=pao	=ní	=kɨ
	eat	=habitual	=rempast	=decl:past
	ʂokóβo	kiá	ɾɨso	=yo	=ʔáina
	children	report	die	=cmpl	=nmlz
	“Only Ashina had fire, back then our ancestors only put the monkey in the sun (to dry), and after that all the children died of hunger.” 0780:0139

Dependent clauses with =yó are accepted in elicitation contexts for all types of dependent clauses.

=paó “Durative, Habitual”

The marker =paó encodes durative or habitual aspect. It can occur with nominalized or noun-modifying clauses and from spontaneous speech.

136.	yonoko	yonoko	=pao	ʔá	tsi
	work	work	=dur	=ndm:ant	lnk
	no	ho	=tɨkɨ(n)	=ní	= kɨ
	1pl	come	=again	=rempast	=decl:pst
	“We worked for a years and then came again.” 0582:0016

137.	ʃoa	hasi=yá	no	raka	=ní
	Benicito	mutun=com	1pl	stay	=rempst
	= kɨ	raka	raka	=pao	no
	=decl:pst	stay	stay	=dur	1pl
	=ʔái	=kato
	=nmlz	=rel
	“On the Benicito, in the place on the mutun, we lived and we lived there for a long time.” 0055:0031

The durative is not compatible with the purpose/instrumental nominalized clauses marked with =tí (e.g., tsaya=pao=tí see=dur=nmlz:purp “to look for a long time” is ungrammatical). Although they do not occur in my corpus, speakers accept prior same-subject and different-subject clauses with the durative marker.

138.	haβá	haβá	=pao	=ʂó	tsi
	run	run	=dur	=prior:sa	lnk
	taʃi= ́	anó	tsáya	=kɨ
	Tashi=erg	paca	see	=decl:pst
	“After running for some time, Tashi saw a paca.”

139.	haβa	haβa	=paó	awí	=kɨ́
	run	run	=dur	woman	=prior:ds/a
	tsi	taʃi= ́	raaʔá	=kɨ
	lnk	Tashi=erg	scold	=decl:pst
	“After his wife ran for some time, Tashi scolded her.”

The marker =paó “habitual, durative” cannot occur with asyndetic same-subject clauses. This is not surprising because these clauses are associated with fast succession. The durative cannot be marked on any concurrent same subject clauses either. Illustrative examples are provided here.

140.	*haβa	haβa	=paó	=ʔi
	run	run	=dur	=concur:ss
	tsi	taʃi	tsaʔó	=kɨ
	lnk	Tashi	sit	=decl:pst
	Intended: “Running for some time he sat down.”

141.	*haβa	haβa	=paó	=pama
	run	run	=dur	=intrmp:ss/a
	tsi	taʃi	pakɨ	=kɨ
	lnk	Tashi	fall	=decl:pst
	Intended: “Tashi was running for some time and then fell.”

=ʃiná “at Night, while Dreaming”

The marker =ʃiná “at night, while dreaming” combines with clauses to mark them as taking place at night or while the subject participant of the verb is dreaming. Examples of the morpheme occurring in dependent clauses are provided here.

142.	a-ʔ-á	=ʃina	ha	wa=kɨ́
	do-epen-do	=atnight	3	tr=prior:ds/a
	tsi	kiá	náa	táʂa	há
	lnk	report	dem:prox	sweep	3
	a(k)	= ní	=kɨ
	do	=rempst	=decl:pst
	“After he (the cutuchi) worked all night, she swept there.” 2123:188

143.	oʂá	=ʃiná	ha	=ʔá	=ka
	sleep	=atnight	3	=nmd:ant	=rel
	wɨakɨ́	tsi	kiá	ha
	next.day	lnk	report	3
	βoʔo	=i	βo	=kan	=ní
	carry	=concur:ss	go:pl	=pl	=rempst
	= kɨ
	=decl:pst
	“After sleeping, they went the next day, carry (it) to him.” 0028:0047

Although it is not attested in my corpus, in elicitation contexts speakers accept the marker =ʃiná “at night, while dreaming” in all dependent clauses.

=βaʔiná “all day, during the day”

The morpheme =baʔina “all day” expresses a durative aspect and that the action took place over the course of the whole day. It can occur in dependent clauses as in the examples below.

144.	wɨakɨ́	ɨa=rí	paʂna-ria=ʂó		ʂoβo
	tomorrow	1sg-too	be.hungry-aug=prior:sa		house
	tsaʔo	tsaʔo=βaʔina=ʔána		ɨarí	wɨakɨ́
	sit	sit=allday=nmd:ant		1sg-aug	tomorrow
	tana=i		ka=ʃaɾi=kí=a
	fish=concur:ss		go=cras=recl:nonpast =1sg
	“Tomorrow, after being hungry, and sitting around all day at home, I will go fishing.” 1154:0054

145.	ho~ho	=βaina	=ʔi	tsi	ʂaʔɨ
	come~come	=allday	=concur:ss	lnk	ant.eater
	ha	βɨtʃa	=ní	=kɨ
	3	meet	=rempst	=decl:pst
	“When he was coming back home all day, then he met the anteater.” 0873:0083

These are the only two obvious cases where =βaʔiná occurs in dependent clauses in the corpus of natural speech. Elicitation data show that =βaʔiná “all day” can occur in all dependent clauses.

=tapi “punctual”

The morpheme =tápi encodes that an event occurred rapidly or in a rushed fashion. This morpheme is unattested in dependent clauses. In elicitation contexts, speakers reject the appearance of =tápi in all dependent clauses except agentive nominalized marked with =ʔai(na) “agentive nominalized” and noun-modifying clauses =ʔá(na) “noun modifier, anterior”.

=tɨkɨn “Again”

The morpheme tɨkɨ(n) encodes that an event had already taken place and that the same or a similar event is taking place again. It can occur in same subject clauses.

146.	ka	=tɨkɨ(n)	=ʔáʂ	media	ora
	go	=again	=prior:ss	half	hour
	tsi	tʃani	=tɨkɨ(n)	hawɨ́wa	pasí
	lnk	speak	=again	mother	silence
	“After half an hour his mother returned and asked again, and there was silence.” 0181:0069

The marker =tɨkɨ́n “again” can occur in different subject clauses.

147.	ɨ́	pa	ɾásɨ	βi	=tɨkɨ́(n)
	1sg	mirative	crouched	grab	= again
	= no	…	áʃiná= ́	átsa	ɨ
	=concur:ds/a	…	Ashina=gen	yuca	1sg
	βi	=tɨkɨ́(n)	=no	i	= kayá
	grab	= again	=concur:ds/a	do	=do&go:intr
	βo	= kan	=(ʔ)á	=ka	haβi
	take.away	=pl	=nmd:ant	=rel	surely
	háskaɾa
	similar.to
	“Ashina grabbed me crouching, while I was grabbing her yuca, and when she did this, they took me away in the same way.” 2123:0047

The morpheme can occur in nominal-modifying clauses and nominalized clauses.

148.	ho	=tɨkɨ(n)	=ʔá	=ka	dos
	come	= again	=nmd:ant	=rel	two
	βáɾi	no	ka	=tɨkɨ(n)	= yáma
	day	1pl	go	=again	=neg
	= kɨ	noβá	patiáɾi	tsaya	=ʔána
	=decl:pst	1pl:gen	chicken	see	=nmd:ant
	“After two days, we went again to see our chicken.” 1157:0017

149.	ʂɨki	ha	βi	=tɨkɨn	=(ʔ)ái
	corn	3	grab	=again	=nmlz
	= ka	pɨ	hawɨ	ʂɨko	= ɾoʔá
	=rel	anxiety	3sg:gen	cob	=limit
	mapo	mapo	mapo
	put	put	put
	“When she grabbed corn, she put a lot of their cobs nothing more in a pile.”0638:0273

In elicitation contexts, speakers accept =tɨkɨ́n in all dependent clauses.

Appendix A.3.3. Associated Motion

As with all Pano languages, Chácobo has a complex associated motion system. Associated motion markers are bound morphemes which add a motion event (often backgrounded) to the event of the verb that they modify (Tallman 2020). I have found no restrictions on the distribution of associated motion morphemes in dependent clauses in Chácobo. Associated motion markers can occur in all same-subject clauses.

150.	ɾiɾii	ɾiɾii	i	=kana	=ʔá
	ideo	ideo	say	=going:itr/sg	=nmd:ant
	hawɨ	áni	a-ʔ-a		=βona
	3sg:gen	river	make-epen-make		=going:tr/pl
	=ʔaʂ	ha	ka	=ʔá	=ka
	=prior:ss	3	go	=nmd:ant	=rel
	naama	=ʔái	=kato
	delay	=nmlz	rel
	“When he was descending with the dounf riririri he was making the river and after some time he went.” 0181:0423

Associated motion markers can occur in all different subject clauses.

151.	ha	nia	=yáma	ik	=ita
	3	throw.away	=neg	be	=recpst
	=ʔá	haβokí	hana	=βayá	ha
	=nmd:ant	now	leave	=do&go:tr/pl	3
	= kɨ́	tsi	kiá	naa	ha
	=prior:ds/a	lnk	report	dem:prox	3
	noʔíria	wɨ́tsa	pia	βi	=ʂó
	people	other	arrow	grab	=prior:sa
	tsi	kiá	yáwa	a(k)	=ki
	lnk	report	peccary	do	=decl:nonpst
	= a
	=1sg
	“After he had thrown it away the day before, he left and went, and grabbed an arrow of another person “I am going to kill the peccary”” 0014:0352

Associated motion markers can occur in nominal-modifier clauses and nominalized clauses.

152.	pi	=βona	=ʔá	tsi	ka
	eat	=do&go:tr/pl	=nmd:ant	lnk	go
	=kí		haβi	hɨnɨ	a(k)
	=concur:sa		surely	water
	= βona		=tsi	= ká(n)	=ki
	=do&go:tr/pl		=now	=pl	=decl:nonpst
	“After eating and drinking chicha they went.” 0116:0334

153.	nawapaʂáwa	namɨ	=káss	=i	kiá
	Nahuapaxahua	kill	=vol	=c	report
	tsios	tsios	tsios	tsi	kiá
	ideo	ideo	ideo	lnk	report
	ʂaʔɨ	tsipis	=hona	=ʔáina
	ant.eater	fart	=coming:intr	=nmlz
	“They wanted to kill Nahuapaxahua “tsios tsios tsios”, the ant eater was coming farting.” 0783:0181

Appendix A.3.4. Modal and Evidential Markers

Chácobo has a number of modal markers. In this section, I will discuss =kás “volitional”; ‘kɨá “counterfactual”; kará “probably”; pi “abilitative”; kiá “reportative”.

=kas “volitive”

The volitive encodes that the subject participant wants to or tried to perform some action. It is compatible with same-subject concurrent clauses.

154.	tʃota	=kas	=kí	tsi	kiá
	have.sex	=vol	=concur:sa	lnk	report
	hinaɾá	raɑ=ta	ha	wa	=no
	penis=admon	send=pnct	3	tr	=concur:ds/a
	rɨrɨʂɨ
	head.bite
	“Everytime he tried to have sex, as he would send his precious penis in, they (the snakes living inside the woman’s vagina) would bite its head.” 0483:0317

The volitive can occur in nominalized clauses.

155.	no	a(k)	=kas	=ʔái	=ka
	1pl	do	=vol	=nmlz	=rel
	no	…	no	a(k)	=tíʂo
	1pl	…	1pl	do	=oblig
	naa
	dem:prox
	“What we want to do is what we should do.” 2153:0059

The volitional is not very common in dependent clauses. Speakers do not accept sentences where the volitional is in prior same or different-subject clauses.

=kɨá “Counterfactual” and = pi “Abilitative”

The marker =kɨá encodes counterfactual semantics and the marker =pi encodes the subject ability or possibility. I have no examples in spontaneous speech of these markers occurring in naturalistic speech. However, speakers accept the counterfactual and the ablative in elicitation.

156.	yonoko	=kɨá	=ʔi	tsi	taʃi
	work	=cntrfct	=concur:ss	lnk	Tashi
	tsaʔó	= kɨ
	sit	=decl:pst
	“When he was about to work, Tashi sat down instead”

157.	yonoko	=pi	=ʔi	tsi	taʃi
	work	=abil	=concur:ss	lnk	Tashi
	tsaʔó	= kɨ
	sit	=decl:pst
	“If its the case that Tashi worked and sat down.”

The only dependent clause which cannot be marked with the counterfactual nor the ablative is the dependent clause marked with =pama “interrupted”.

158.	*ʂoβo	tsaya	=kɨá	=pama	tsi
	house	see	=cntrfct	= intrmp:ss/a	lnk
	taʃi	pakɨ́	=kɨ
	Tashi	fall	=decl:pst
	Intended: “Tashi was watching the house and then fell.”

159.	*ʂoβo	tsaya	=pi	=pama	tsi
	house	see	=abil	= intrmp:ss/a	lnk
	taʃi	pakɨ́	=kɨ
	Tashi	fall	=decl:pst
	Intended: “If Tashi was watching the house and then fell.”

=kiá “Reportative”

The reportative can occur adjacent to a dependent clause.

160.	hatsi	kiá	yonoko=ʂó	tsi	taʃi= ́
	then	report	work=prior:sa	lnk	Tashi=erg
	anó	tsáya=kɨ
	paca	see=decl:pst
	“Then (it is said) after working, Tashi saw the paca.”

However, it is not clear whether the reportative can mark a dependent clause independent of a main clause. There is maximally one reportative marker for every main clause that is present. This suggests that dependent clauses are not marked independently for the reportative.

161.	*hatsi	kiá	yonoko=ʂó	tsi	kiá
	then	report	work=prior:sa	lnk	report
	taʃi= ́	anó	tsáya=kɨ
	Tashi=erg	paca	see=decl:pst
	“Then (it is said) after working, Tashi saw the paca.”

On the other hand, reportative markers scope over dependent clauses. The scope of the reportative markers appears to be mostly determined by context. Examples where the reportative scopes only over the main clause are provided here.

162.	gas	tsi	kiá	toa	=ʔitá
	gas	lnk	report	explode	=recpst
	=kɨ	ha-ʔá	βɨtɨ	=kan	=(ʔ)áina
	=decl:pst	3-loc	cook	=pl	=nmlz
	“It is said that the gas exploded where they typically cook.’ (context: the speaker knows where they cook).” 2153:0225

163.	ka	=ʂó	tsi	kiá	hɨma
	go	=prior:sa	lnk	report	Jema
	=kí	ha	yoa	=ʔitá	=kɨ
	=dative	3	tell	=recpast	=decl:past
	“After she (Bosi) went, (it is said) she (Bosi) told Jema.” (context: speaker knows that Bosi went somewhere, she saw her). 2153:0293