Reference constraints and information-structure management in Kokama purpose clauses: A typological novelty? International Journal of American Linguistics, 80.1: 39-67. 2014.

[IJAL, vol. 80, no. 1, January 2014, pp. 39–67]© 2014 by The University of Chicago. All rights reserved.0020–7071/2014/8001–0003$10.00

39

REFERENCE CONSTRAINTS AND INFORMATION STRUCTURE MANAGEMENT IN KOKAMA PURPOSE CLAUSES: A

TYPOLOGICAL NOVELTY? 1

ROSA VALLEJOS

UNIVERSITY OF NEW MEXICO

Cross-linguistically, implicit purposive subjects are controlled to an overwhelmingly degree by the matrix subjects (Schmidtke-Bode 2009:56). In fact, little evidence has been found of constructions in which main and dependent events are not performed by the same entity, and no evidence about constructions in which the performer of the main event does not control the realization of the dependent one (Cristofaro 2003:157). Kokama (Amazon) provides a challenge to these generalizations and so constitutes a typological novelty. The language has three positive purpose constructions formed by attaching tara, mira, or tsen to the subordinated verb. In mira constructions, the matrix absolutive controls coreference with the omitted accusative in the purpose clause. In tara constructions, the matrix absolutive controls coreference with the omitted nominative in the purpose clause. In tsen constructions, coreference is free. Overall, tara/mira constructions are syntactically more integrated than tsen constructions. Discourse data shows that syntactic integration correlates with semantic-pragmatic parameters: (i) temporal integration between events, (ii) successful outcome of the purposive event, and (iii) whether the omitted argument belongs to the discourse context.

[KEYWORDS: Kokama, Amazonian languages, purpose clauses, information structure, subordination]

1. Introduction. Typological works have found that purpose clauses outrank all other adverbial clauses in terms of the lack of overt arguments (Cristofaro 2003:173). Further, implicit purposive subjects are controlled to an overwhelmingly degree by the matrix subjects (Schmidtke-Bode 2009: 56). There is little evidence about purpose relations in which main and dependent events are not performed by the same entity, and no evidence of constructions in which the performer of the event expressed in the matrix clause does not control the realization of the event expressed in the pur-pose clause (Cristofaro 2003:157). Kokama (ISO code: cod), a Tupian-based mixed language spoken in the Peruvian Amazon, provides a challenge to the latter generalizations and so constitutes a typological novelty.

1 I gratefully acknowledge the contribution of each of the Kokama-Kokamillas, also known as Kukama-Kukamiria, who have shared their knowledge with me since 1997. I am especially grateful for the dedication and friendship of two elder speakers, Victor Yuyarima Chota and Rosa Amías Murayari. Fieldwork for this study was supported by the National Science Foundation (BCS-0965604). Comments by Spike Gildea, Francesc Queixalos, and two anonymous reviewers have improved this paper tremendously. All omissions and misinterpretations are mine alone.

This content downloaded from 64.106.42.43 on Thu, 9 Jan 2014 19:47:54 PMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/page/info/about/policies/terms.jsp

!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/40

Kokama has three positive purpose constructions, formed by attaching -tara, -mira, or -tsen to the subordinated verb. These constructions exhibit di!erent degrees of syntactic integration with the matrix clause and entail di!erent coreferentiality conditions between arguments. In mira constructions, the matrix absolutive (S/O) controls the obligatorily omitted accusative (O) in the purpose clause (1). In tara constructions, the matrix absolutive argument (S/O) controls coreference with the obligatorily omitted nominative (S/A) in the purpose clause (2). In tsen constructions, there are no reference constraints and all the purposive arguments must be expressed by either NPs or pronouns (3). In addition, the language has a negative purpose construction, formed by attaching -maka to the subordinated verb. This construction is similar to tsen constructions in terms of argument realization and lack of coreference con-straints (4). This paper, however, focuses on positive purpose constructions. 2

(1) A V Oi [Oi A Mijiri erura Mararina-uy [" Kutsi(=pura) Miguel bring Magdalena-PAS1 Jose=FOC

V-mira] mutsanaka-mira] cure-!"#1

‘Miguel brought Magdalena [in order for Jose to cure (her)]’(2) A V Oi [O Ai

Mijiri erura Mararina-uy [Kutsi(=pura) " Miguel bring Magdalena-PAS1 Jose=FOC

V-tara] mutsanaka-tara] cure-!"#2

‘Miguel brought Magdalena [in order (for her) to cure Jose]’(3) A V O [O A

Mijiri erura Mararina-uy [Kutsi(*=pura) ra Miguel bring Magdalena-PAS1 Jose=FOC 3SG.M

V-tsen] mutsanaka-tsen] cure-!"#3

‘Miguel brought Magdalena [in order for her/another to cure José]’

2 Abbreviations used in this paper are: AUG = augmentative, CAU = causative, com = comitative, DAT = dative, FOC = focus, LOC = locative, NMLZ = nominalizer, N.PUR = negative purpose, PAS = past, PL.F = plural female speaker, PL.M = plural male speaker, PROG = progressive, PUR = purpose, REI = reiterative, REL = relativizer, RSN = reason, 1SG.F = #rst-person singular female speaker, 1SG.M= #rst-person singular male speaker, 1PL.EX.F = #rst-person plural exclusive female speaker, 1PL.EX.M = #rst-person plural exclusive male speaker, 3SG.F = third-person singular female speaker, 3SG.M = third-person singular male speaker, 3SG.OBJ = third-person singular object.



!"!#$# %&'%"() *+#&()( 41

(4) A V O [O A Mijiri erura Mararina-uy [Kutsi ra Miguel bring Magdalena-PAS1 Jose=FOC 3SG.M V-maka]

mutsanaka-maka] cure-N.!"#

‘Miguel brought Magdalena [in order for her/another not to cure José]’

Since coreference is not restricted in tsen constructions, they are syntactically less integrated than -tara/-mira constructions. The distribution of -tara and -mira is conditioned by syntax, while -tsen is not; speakers have two choices to express purpose: -tara/-mira versus -tsen. What, then, drives the speaker’s choice of one construction over another? This paper demonstrates that in Ko-kama purpose clauses, syntactic integration between clauses correlates not with the semantic integration between events alone (Givón 2001 and Cristofaro 2003) but with discourse-pragmatic factors. I argue here that purpose clauses in Kokama are grammaticalized information structure management strategies: -tsen implies that the event indicated in the subordinated clause is to be con-strued as more independent from the main event 3 than with -tara/-mira. First, in (1) and (2), the interpretation is that the event in the subordinated clause occurs right after the main event, while in (3) this temporal integration does not exist. Second, in (1) and (2), the purpose event is successfully realized and the main event is a condition for the purposive event to take place, while in (3) it is not. Third, the elided arguments in (1) and (2) belong to the discourse context, whereas in (3) even entities new to the addressee can be introduced in the purpose clause. Fourth, in (1) and (2), the (non-coreferential) argument in the subordinate clause can be marked with a focus marker (=pura); this is not possible in (3). This study shows that pragmatic tightness between events is re$ected by syntactic integration between clauses and suggests that IS can operate within some types of subordinate clauses.

Section 2 o!ers the theoretical notions that serve as a frame for the dis-cussion throughout the paper; 3 provides the typological features of Kokama relevant to this study; 4 gives the formal properties of the purpose construc-tions; 5 explores the frequency of each construction in a text database; 6 in-vestigates the discourse-pragmatic parameters triggering the distribution of purpose constructions; and 7 o!ers conclusions.

2. Preliminaries.2.1. Clause combining: syntactic integration and event integration.

One assumption made in this paper is that dependency, both functionally

3 In this paper, the term EVENT covers event, action, and state.



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/42

and syntactically, is not a discrete property but a matter of degree (Givón 2001:327). In one view, “coordinate” and “subordinate” clauses lie at the op-posite ends of a continuum (see, for instance, Haiman and Thompson 1984, Lehman 1988, and Van Valin 1993). Clauses can be combined in various ways in this continuum. Subordination then results from a composite of factors, including the identity of subject and TAM between the two clauses; reduction of one of the clauses (e.g., argument ellipsis); formal incorporation of one of the clauses; intonation linking; one clause being within the scope of the other (e.g., negation, interrogation); absence of tense iconicity; and identity of speech-act perspective (e.g., direct/indirect speech) (Haiman and Thomp-son 1984:511–20). What discourse factors underlie these formal parameters is still an open question. Clause integration correlates with the quantity of elements in common between two clauses. When consequent clauses share subject, TAM and polarity are placed under the same intonation contour, etc.; they are syntactically more integrated than those that have less in common (Givón 2001).

On the functional side of complex constructions, semantic integration is the degree of interconnectedness of two events. This conceptual integration results from the number of shared participants, the role(s) of the shared par-ticipants, control over realization of the subordinate event by a participant from the main clause, and temporal integration between the events, among others (Givón 1980; 2001, Croft 2001; 2003, and Cristofaro 2005). The idea is that syntactic integration correlates with event integration. 4 Givón (2001: 238) notes that propositional-semantic features of event integration shade into the discourse-pragmatic features of cross-event coherence. The grammar of Kokama purpose clauses illustrates this transition.

A related but di!erent approach to complex construction advocates a con-tinuum in the form of a conceptual space of complex sentences with the four traditional types: coordination constructions, complement clauses, relative clauses, and adverbial clauses at the corner ends (Croft 2001:322). Although constructions spread and converge in various ways, “complex sentences can be organized into a conceptual space on the bases of a Gestalt distinction between complex #gure and #gure–ground constructions, and by the property of e-site elaboration” (Croft 2001:361). 5 It needs to be highlighted that in

4 This idea has been also developed into the so-called binding hierarchy (Givón 1980:369), which states that the degree of force exerted by the agent of a matrix clause over the agent of a complement clause correlates with the degree of morphosyntactic restrictions imposed on the complement clause.

5 Building on Langacker, Croft (2001) explains that the salient substructure of a dependent concept is called elaboration site or e-site. Dependent concepts (e.g., predicate) can be made more speci#c by another concept (e.g., argument), the latter being the autonomous concept in the relation. In Croft’s view, both complement and relative clauses share the e-site elaboration property: the subordinated clause event elaborates on a salient substructure of the main clause



!"!#$# %&'%"() *+#&()( 43

a number of languages purpose clauses do not #t well within the adverbial clause category. Kokama is one of those languages.

Another key notion in clause combining is pivot constraints (Dixon 1994). The pivot controls the reference of the target element. The syntactic constraints can apply to only some kinds of clause combination within a single language; that is, languages can have S/A pivot, S/O pivot, or mixed-pivot, as illustrated by languages such as Yidiny (Dixon 1977) and Jakaltek (Craig 1977). 6 In Kokama, the notion of pivot constraints is relevant to account for a few com-plex constructions that involve obligatory ellipsis, including purpose clauses.

2.2. Purpose clauses cross-linguistically. Purpose clauses encode intentional action to achieve an intended outcome. Intention-in-action is among the most basic functional concepts (Jackson 1995:57 and Searle 2006:2). Purpose clauses provide reasons, accounts, or explanations for the occurrence of the event described in the matrix clause. Typological studies have found that deranked forms (which lack morphology associated with #nite verbs) are the most common encoding strategy for purpose construc-tions (Cristofaro 2005 and Schmidtke-Bode 2009). This fact supports the idea that the -tara and -mira constructions are of this sort.

Multiple purpose constructions are well documented cross-linguistically. Some languages have di!erent syntax for purpose clauses to encode distinc-tions such as same subject versus di!erent subject and realized versus unre-alized events (see Thompson, Longacre, and Hwang 2007); negative versus positive purpose (Lichtenberk 1995); simultaneous, consecutive, or temporally distant events (e.g., visual versus non-visual, as in Tariana; see Aikhenvald 2003:393!.), among others.

Purpose clauses are semantically unlike other adverbial clauses. Data from unrelated languages show that purpose clauses span complementation and adverbial subordination or coordination and adverbial subordination (Croft 2001:326). Unlike typical adverbial relations, the purposive event does not hold a #gure–ground relationship with the main event. Purpose clauses are generally under the scope of the main clause’s TAM and, as shown in (5), they tend to share arguments. These facts indicate that purpose clauses are more semantically integrated to the matrix clause than prototypical adverbial clauses. In Kokama, purpose clauses do not align toward complementation; to

event (e.g., a complement clause is an argument of the main clause). Conversely, adverbial and coordinate clauses do not have this property. Adverbial clauses hold a #gure–ground relation-ship with the main event, whereas coordinate clauses encode two events conceived as a single whole, i.e., a complex #gure.

6 In Yidiny, mixed-pivot occurs in coordinate constructions: when the coreferential NPs are pronouns, the pattern of control and omission is accusative, whereas when they are full NPs, the pattern is ergative (Dixon 1977).



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/44

assume complement functions, they must be nominalized. However, two of them (mira and tara clauses) obligatorily share arguments with the main event.

Perhaps the most salient variables in the typology of purpose constructions are the instantiation of arguments and the reference constraints between main clause participants and the dependent clause participants. Cristofaro (2003) claims that languages tend not to express arguments in purpose relations and suggests the implicational hierarchy in (5). In this hierarchy, purpose clauses outrank every other adverbial clause type in terms of lack of overt argument realization. Two of the three Kokama purposive constructions conform to this hierarchy: two purpose constructions (tara-/mira-) involve obligatory zero anaphora, but the third one (tsen-) does not.

(5) Adverbial Argument Hierarchy (Cristofaro 2003:173) Purpose > Before, After, When, Reality condition, Reason

The grammar of purpose clauses across languages shows that the controller of coreference plays a greater role than the target. It has been found that (i) the prototypical purpose relation is one in which the main and dependent events are performed by the same entity, and that (ii) the main clauses are typically agentive (Givón 2001:337 and Cristofaro 2003:157). According to Cristofaro, there is “no evidence about purpose relations in which the performer of the main state of a!airs cannot control the realization of the dependent one” (2003:157). Building on this claim, Schmidtke-Bode (2009) #nds that in an overwhelming majority of languages in his sample, the implicit purposive subject is controlled by the matrix subject. The agent-binding phenomenon can be functionally explained on the grounds of the content of purpose clauses. Given that purpose clauses encode intention-in-action, the likelihood of the intended outcome being achieved is higher if the one having a desire is in-volved in the realization of that purpose. Thus, there is a communicative preference that binds purposive subjects to matrix agents (Cristofaro 2003 and Schmidtke-Bode 2009).

Yet there are a few languages that have purpose clauses with implicit non-subjects. Object-gap purpose clauses constitute the bulk of this group, in which the matrix O controls the implicit purposive O (Schmidtke-Bode 2009:64); however, given the paucity of available data, no typological studies on this phenomenon have been advanced.

Kokama purpose clauses constitute a system not yet documented in typo-logical studies. The novelty does not concern the implicit purposive argument but rather the controller of coreference with that implicit argument. The two purpose constructions that involve obligatory ellipsis are consistent with the typological patterns described above in that the implicit argument is either the subject (S/A, in tara constructions) or the object (in mira constructions). However, coreference with this implicit argument is obligatorily controlled not by the matrix subject (S/A) but by the matrix absolutive (S/O). In that



!"!#$# %&'%"() *+#&()( 45

sense, tara/mira constructions do not conform to the communicative prefer-ence of agent-binding.

The characterization of purpose constructions in 4 below employs the theoretical ideas presented here. Section 5 o!ers an examination of language use to demonstrate that the distribution of Kokama purpose constructions is licensed by several semantic-pragmatic parameters.

3. Typological profile of Kokama. This section brie$y highlights some aspects of Kokama relevant for the discussion in subsequent sec-tions. These include clausehood (3.1), adverbial clauses (3.2), and clause nominalization (3.3).

3.1. Clausehood. Kokama does not employ case marking or verb in-dexation to express grammatical relations. Instead, subject and object are coded via their position relative to the verb. In order to have well-formed sentences, subject and object need to be overtly instantiated by a noun phrase or pronominal form substituting an NP. In the intransitive construc-tion, the subject precedes the verb: [SUBJ V] (6). In a pragmatically un-marked transitive construction, the subject precedes the verb and the object follows it: [SUBJ V OBJ] (7). In natural discourse, several ordering patterns are possible; the alternations between them are determined by grammatical relations, information structure, and verbal aspect. In main clauses, SVO is the pragmatically unmarked constituent order (7). While OSV is employed with salient objects (8), SOV is only possible if the verb is marked by progressive aspect (9).

(6) yawara yapana SV dog run‘The dog runs’

(7) mixiri chikuarata yawara(=uy) SVO Miguel follows dog(=PAS)‘Maybe Miguel follows (followed) the dog’

(8) yawara mixiri chikuarata(=uy) OSV dog Miguel follow(=PAS1)‘The dog, Mixiri follows(ed)’

(9) yawara mixiri chikuarata-ri SOV dog Miguel follow-PROG

‘The dog is following Mixiri’

Kokama is a relatively isolating, analytical language. Major grammatical categories like number, tense, and modality are conveyed by positionally #xed clitics. Note in (7) that tense appears attached to the NP object. Although



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/46

tense, aspect, and modality are not obligatory in a basic clause, the verb can take several aspectual su%xes and modal clitics (10). 7

(10) uruputini upaka-ta-ka-pa=ura=mía condor wake.up-CAU-REI-CPL=3M.O=HYP

‘The condor may wake him up completely’

Finally, oblique phrases can optionally occur in a clause. In contrast to sub-jects and objects that occur in relatively #xed positions in the clause, obliques appear either at the beginning or toward the end of the clause. Obliques are marked by a set of postpositions, such as =ka ‘locative’ and =muki ‘comita-tive’ (11) (Vallejos 2010a).

(11) pai uka=ka ta kurata kaitsum=uy ra uncle house=LOC 1SG.M drink masato=PAS1 3SG.M

mirikua=muki wife=COM

‘At the uncle’s house I drank masato (yucca beer) with my wife’

3.2. Adverbial clauses. Because they do not ful#ll a core semantic or grammatical role in a matrix clause, adverbials are not required. To form adverbial clauses, Kokama employs a set of adverbial subordinators that indicate a speci#c semantic relationship between the main and adverbial clause. Verbs within adverbial clauses do not take any aspectual or modal morphology. In adverbial clauses, subjects and objects are neither obligato-rily nor optionally ellipsed. 8 Generally, adverbial clauses exhibit OSV order (12), and they may precede the main clause (12) or follow it (13). Impor-tantly, oblique phrases and adverbial elements can appear within adverbial clauses to specify, for instance, the temporal frame (12) or the location of the subordinate event (13).

(12) [mukuika kuashi ta!ra ra=chikari-npu] r=ichima !aku two day son 3SG.M=lost-after 3SG.M=exit creek

ts!ma-ka shore=LOC

‘After being looking for his son for two days, he gets out to the shore of a creek’

(13) tuntu-k!ra ra ipu-ta [ima=muki iwirati=kuara drum-DIM 3SG.M sound-CAU brother=COM jungle=INE

7 In language use, as many as four su%xes can occur together on a single verb, followed by up to two clitics. For a detailed description of each morpheme, see Vallejos (2010b).

8 In elicitation, same subjects can be optionally ellipsed if the adverbial clause follows the main clause.



!"!#$# %&'%"() *+#&()( 47

r=uwata-puka] 3SG.M=walk-when

‘She plays the little drum when she walks with his brother into the jungle’

Kokama purpose clauses show all the properties of adverbial clauses, except for obligatory ellipsis in -tara/-mira constructions. This is not surprising since, as indicated in 2 above, purpose clauses cross-linguistically do not always align with typical adverbial clauses.

3.3. On clause nominalization. In Kokama, there is the A-relativizer morpheme -tara which resembles one of the purpose markers. This, together with the fact that clause nominalization is employed to functionally relativ-ize S and O (but not A), 9 suggests a historical connection between purpose clauses and clause nominalization. Thus, before I proceed with the discus-sion, an explanation of why the purpose clauses under examination are better analyzed as adverbial clauses rather than clause nominalizations is in order. 10

First, strictly speaking, the relativization of A is not achieved through nominalization. To relativized S and O, the absolutive is gapped with the absolutive nominalizer -n, as in (14) and (15). Note in (14) that ‘that fell down’ together with the head noun ‘house’ constitutes a NP and so it can take morphology associated with NPs, such as focus and plural. The plural marker =kana occurs at the end of the complex NP, but it is clear that its scope includes ‘house’. In contrast, the ergative is gapped with the A-relativizer tara (16). Unlike the true nominalizations in (14) and (15), V-tara in (16) cannot take any morphology associated with NPs.

(14) ukai ["i uwari]-n=pura=kana tseniapupe-ra-pa house fall-NMLZ=FOC=PL.M knee-VBZR-CPL

‘The houses that fell down are kind of kneeling’(15) ra=yatukupe arawatai ["i ra aya]-n

3SG.M=carry.on.back monkey.sp 3SG.M shoot-NMLZ

‘He carries on his back the choro monkey [that he shot]’(16) inu umi napitsarai ["i tukini muna]-tara (*=kana)

3PL.F see man hammock steal-REL.A‘They see the man who stole the hammock’

9 In this paper, I take the position that, in some languages, clause nominalization can be a rela-tivization strategy; that is, it restricts or identi#es a head noun (see, for instance, Andrews 2007).

10 Two anonymous reviewers have suggested that clause nominalization should be enter-tained as a hypothesis for -tara and -mira constructions. Also, the semantic and formal a%nity between purpose clauses and relative clauses has been reported for several languages including Mapudungun, Ma’di, Shoshone, Somali, etc. (Schmidtke-Bode 2009:167).



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/48

The tara purpose construction exhibits some features similar to the embed-ded constructions in (14)–(16). First, it is verb-#nal; and, second, the verb takes none of the morphology associated with #nite clauses (i.e., aspectuals and modals). In fact, in the database for this study, there are a few instances like (17) where the sequence NP V-tara could have either a relative clause or a purpose clause interpretation.

(17) r=erura ikian yawara [mari chikari-tara] 3SG.M=bring this dog thing look.for-#$%.&/!"#1

(i) ‘He brings this dog [to look for things]’(ii) ‘He brings this dog [that looks for things]’

Yet the tara purpose construction di!ers from the tara relative construction in various respects, which can serve as tests to disambiguate the syntactic structure of utterances like (17). The #rst piece of evidence comes from the distribution of tense and modal clitics, which in Kokama show up at the right edge of the verb phrase (see 3.1). In (17), if tense shows up attached to ‘dog,’ then the object of the clause is ‘this dog’ and the [NP V-tara] is a purpose clause outside of the verb phrase (as in 18). Consequently, in (18), [mari chikari-tara] can be fronted. The speaker used this utterance when narrating that at the airport the police brought a dog to inspect his luggage. In contrast, if tense attaches to V-tara, it can only be interpreted as a relative clause (as in 19). Here, ‘this dog’ together with the RC constitutes the object noun phrase, hence [mari chikari-tara] cannot appear sentence-initially. This example could be used to describe a rescue/hunting dog that was brought in. 11

(18) r=erura ikian yawara=uy [mari chikari-tara] 3SG.M=bring this dog=PAS1 thing look.for-PUR1‘He brought this dog [to look for things]’*‘He brought this dog [that looks for things]’

(19) r=erura [ikian yawara [mari chikari-tara]]=uy 3SG.M=bring this dog=PAS1 thing look.for-REL.A=PAS1‘He brought this dog [that looks for things]’*‘He brought this dog [to look for things]’

11 Additional examples that tara relative clauses and tara purpose clauses appear in rather di!erent syntactic slots within the clause are provided below. In (a), the relative clause follows a head noun, whereas in (b) the tara purpose clause simply appears sentence-#nally, outside of the verb phrase.

(a) apu [arutsu yumi-tara] uri=uy chief rice give-REL.A come=PAS1

‘The chief [that gives rice] came’ (ED)(b) apu uri=uy [arutsu yumi-tara]

chief come=PAS1 rice give-PUR

‘The chief came [to give rice]’ (ED)



!"!#$# %&'%"() *+#&()( 49

The second piece of evidence is the distribution of oblique phrases. Oblique phrases can occur within tara purpose constructions but not within tara relative clauses. Adding an inesive phrase to the purpose construction in (18) would give (20). An oblique phrase within the tara relative clause is not possible (21).

(20) r=erura ikian yawara=uy [tsaku=kuara mari 3SG.M=bring this dog=PAS1 bag=INE thing

chikari-tara] look.for-PUR1

‘He brought this dog [to look for things in my bag]’(21) *r=erura [ikian yawara [tsaku=kuara mari

3SG.M=bring this dog=PAS1 bag=INE thing chikari-tara]]=uy look.for-REL.A=PAS1

In addition, the tara purpose construction does not #ll a NP slot in the clause. To do so, the purpose constructions must be nominalized via the morpheme -n (22). Note in (22) that, once the purpose clause is nominal-ized, it can take a possessor, ra, and function as the object of the purposive verb ‘look for.’

(22) r=utsu=uy [[ra kamata-tara-n] chikari-tara] 3SG.M=go=PAS1 3SG.M work-PUR1-NMLZ look.for-PUR1‘He went to look for his job (something to work on)’

Since Kokama has a purposive postposition -ra, another hypothesis to consider would be that perhaps -tara and -mira consist of the nominalizers -ta/-mi plus the purposive postposition -ra. However, nominalizers -ta and -mi do not exist in Kokama. The language has three argument nominalizers: the absolutive nominalizer -n which derives a subject nominalization from intransitive verbs (23a) and an object nominalization from transitive verbs (23b); the agent nominalizer -wara (23c); and the pro#cient agent nominal-izer -tsuri (23d) (Vallejos 2010b:230–34). Note in (23c) that -tara does not derive nouns. (24) illustrates once again that V-tara can only function as noun if further derived by -n.

(23a) tsaipura-n ‘drunk person’(23b) yupi-n ‘woven product’(23c) yupi-wara ‘weaver’ *yupi-tara(23d) yupi-tsuri ‘skilled weaver’(24) emete ikua-tara-n=kana ini=kumitsa

exist know-REL.A-NMLZ=PL.M 1PL.IN=language‘There are those who know our language’



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/50

A look at the potential sources of the -tara/-mira subordinators could shed light on today’s Kokama purpose constructions. In Vallejos (2010b) I hy-pothesized that the sources of -tara and -mira seem to be the Tupi-Guaraní nominalizer of agent *-tsar/-tar and the patient nominalizer *-pyr, respec-tively (Jensen 1998:540–41). Further, these forms could have fused with the -ra purposive postposition over time. The source of -tsen is less obvious. However, further examination of these hypotheses lies outside the scope of this work. To conclude this section, it is clear that in today’s Kokama, -tara, -mira, and -tsen are not nominalizers.

4. Kokama purpose clauses: formal properties. As indicated earlier, Kokama has three positive purpose constructions, formed by attaching -tara, -mira, or -tsen to bare verbs. There is also one negative purpose construc-tion created by attaching -maka to the subordinated verb. In 4.1–4.4, the discussion focuses on positive purpose constructions which are the ones that exhibit functional overlap.

The minimal composition of purpose constructions is a verb marked by the respective subordinator. If the purposive verb is transitive, tara/mira con-structions must include an argument, and tsen constructions two arguments. Like other adverbial clauses, purpose clauses: (i) are not marked by modality or tense marker; (ii) the subordinate verb does not receive any other verbal in$ectional or derivational morphology; (iii) may include adverbial words and oblique phrases; and (iv) are generally postposed to the main clause, but they can also appear preposed (see table 5). Purpose clauses cannot be ne-gated by the negative particles t!ma and ni that operate within matrix clauses; there is a construction dedicated to encode negative purpose. 12 In tara and mira constructions, the verb occurs in #nal position (OV), whereas in tsen constructions and maka constructions, the object follows the verb (VO). 13

Kokama positive purpose clauses carve up the coreference space in very #ne-grained ways. As already mentioned, the choice of purpose clause marker is triggered by several parameters, two of them being (i) argument

12 For expressing negative purpose, the subordinated verb is su%xed with maka. As indicated earlier, maka constructions are similar to tsen constructions: they do not exhibit coreference re-strictions and all their arguments must be overtly expressed. Note in (a) that both the matrix clause and the purpose clause are under their own negation scope but under the same intonation contour.

(a) ina yanuki !k! churank!ra eyu-maka=ura PROH put chili child eat-N.PUR=3SG.OBJ

‘Do not put chili for the child not to eat it’ [so that he doesn’t #nd it in the food]13 Although the most frequent ordering pattern attested in typical adverbial clauses is OSV,

when discussing tara/mira constructions, zeros are not employed to make a claim about their linear position within the subordinate clause. They are meant to help the reader keep track of the coreference among arguments.



!"!#$# %&'%"() *+#&()( 51

realization and (ii) coreference control. In terms of argument realization, in tara and mira constructions, the target argument in the subordinate clause must be elided, whereas in tsen constructions, all the arguments of the pur-pose clause have to be expressed. As for the second parameter, in purpose clauses marked by -tara, the absolutive (S/O) argument controls coref-erence with the nominative (S/A) argument in the subordinate clause. In purpose clauses marked by -mira, the absolutive (S/O) argument controls coreference with the absolutive (O) argument in the matrix clause. In tsen constructions, no coreference is required with any main clause arguments. This is summarized in #gure 1.

As #gure 1 shows, the choice between -tara and -mira is automatically conditioned by the semantics of reference; however, there is a clear overlap between these two and -tsen, which can occur in any coreference conditions. I come back to this point in 6 below to demonstrate that the choice between -mira/-tara and -tsen is driven by information structure factors. The following subsections o!er a brief characterization of each purpose construction. From now on, notations like [S = S] indicate that the main clause controller is an S and the coreferential NP in the subordinate clause is an S.

4.1. Tara constructions. This construction contains minimally a verb (25). Beyond conveying purpose, the subordinator -tara indicates that the A/S argument is missing in the adverbial clause, with its referent being either the S or the O argument of the main clause. In other words, the Ab-solutive argument of the main clause controls the ellipsis of the Nominative argument in the adverbial clause. Schematically: [S/Oi [O " (S/A)i V-tara]. This construction is illustrated in (25)–(28). Examples (25) and (26) have an intransitive main clause and, in each, the S of the main clause, ‘my husband’, controls referentiality of the missing argument of the adverbial

FIG. 1.—Coreference constraints in positive purpose clauses.

T A R G E T (Subordinate)

CO

NT

RO

LL

ER

(Mat

rix) S A O

S

O

A

-tara -mira

-tsen



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/52

clause. In (25), the purpose clause is intransitive, hence coreference is with the missing S argument, [S = S]; in (26), the purpose clause is transitive, and so coreference is with the missing A argument, [S = A].

(25) Si V Si V-tara tsa mena uts=uy !w!rati [" aya-tara] 1SG.F husband go=PAS1 jungle hunt-PUR1‘My husband went to the forest to hunt’

(26) Si V O Ai V-tara tsa mena uts=uy !w!rati [!rara " yauki-tara] 1SG.F husband go=PAS1 jungle canoe make-PUR1‘My husband went to the forest to make a canoe’

The examples in (27) and (28) have transitive matrix clauses, and here it is the O argument in the main clause that controls coreference with the missing argument in the adverbial clauses: the intransitive S, [O = S], in (27) and the transitive A, [O = A], in (28), correspond to:

(27) A V Oi Si V-tara rana erutsu ta=ta!ra [" ikara]-tara 3PL.M bring 1SG.M=son sing-PUR

‘They bring my son [in order (for him) to sing]’ (ED) *‘They bring my son so that he helps them/they all sing’.(28) A V Oi [O Ai V-tara]

rana erura ta=ta!ra [nai " mutsanaka-tara] 3PL.M bring 1SG.M=son grandma cure-PUR1‘They bring my son [in order (for him) to cure grandmother]’

*‘They bring my son [in order for them to cure grandma/grandma to cure him]’

As indicated by the asterisks, no other interpretation of (27) and (28) is possible.

4.2. Mira constructions. Purpose clauses formed with the morpheme -mira are also obligatorily missing an argument, in this case the transitive O, which is understood as being coreferential with the main clause abso-lutive. Schematically, [S/Oi [" (O)i A V-mira]. In (29) and (30), the main clauses are intransitive. However, as shown in (29), an intransitive verb cannot bear the morpheme -mira. Consultants #xed example (29) by deriv-ing a transitive verb from ‘take a bath’ (30). In this example, the causative -ta is added to the verb and a new participant is introduced in the purpose clause. As a result, the new participant automatically gets interpreted as the



!"!#$# %&'%"() *+#&()( 53

A argument and the main clause S controls coreferentiality with the missing O of the purpose clause, [S = O].

(29) *Si V Si V-mira ami uri=uy [" yatsuka-mira] grandpa come=PAS1 take.bath-PUR2

(30) Si V Oi A V-mira ami uri=uy [" rana yatsuka-ta-mira] grandpa come=PAS1 3PL.M take.bath-CAU-PUR2‘Grandpa comes so that they make (him) take/give (him) a bath’

For completeness, the examples in (31) and (32) show transitive clauses in both main and subordinate clauses, so the O argument in the main clause is coreferential with the O argument in the purpose clause, [O = O].

(31) A V Oi [Oi A V-mira] yama purepeta p!t!ma " tuaminu ts!ki-mira other buy tobacco elders pull-PUR2‘Another buys tobacco for the elders to smoke’

(32) A V Oi [Oi A V-mira] !p!sa ta tsuwa n=uri [" ta erutsu-mira] night 1SG.M catch 2SG=AUX 1SG.M bring-PUR2‘At night I’ll pick youi up to bring (you)’

In (31), the object of main clause ‘buy’ is ‘tobacco’, which in turn is the unexpressed object of the subordinate verb ‘pull/smoke’; that is, [O = O]. A similar pattern can be seen in (32), where the matrix clause A is ‘I’ and the O is ‘you’, and the same two participants have the same two roles in the purpose clause. Note, however, that only the object is omitted in the purpose clause, whereas the A argument is explicitly stated. 14

A #nal point to make is that when the logical object of the purpose verb is not coreferential with the matrix O, it must be expressed as a purposive oblique (33).

(33) A V Oi [Oi A V-mira OBL] tsa mena ka!tsa !w!ra [" ya yauki-mira !rara-ra] 1SG.F husband cut tree 3SG.F make-PUR2 canoe-PUR

‘My husband cuts wood to make it into a canoe’*‘My husband cuts wood to make a canoe’

14 I have con$icting data about this point. In elicitation, if the A in the mira construction is coreferential with the matrix clause A (as in 32), the pronoun in the subordinate clause can be optionally ellided. In texts, however, the subject in mira constructions is not omitted.



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/54

4.3. Tsen constructions. Tsen constructions di!er from tara/mira con-structions in three ways. First, the reference of the purposive participants is unconstrained. There is no requirement for the tsen clause to share any arguments with the main clause. Second, no argument is missing in the tsen clause; they are always explicitly stated, whether or not coreference occurs. The arguments in the main clause are expressed by either pronominals or full NPs; if an argument is shared, in the purpose clause it is generally indicated by an anaphoric pronoun, but not by zero. Third, unlike tara/mira clauses, which are verb-#nal, tsen clauses generally exhibit SVO order, similar to the default order for main clauses. Schematically: [clause [S/A V-tsen O].

Examples provided in (34) and (35) involve intransitive main clauses. In (34), the purpose clause is intransitive and the main clause S is coreferential with the purpose clause S; in (35), the purpose clause is transitive and the main clause S is coreferential with the purpose clause A. Note that in both examples, the purpose clause nominative argument is indicated with a proclitic pronoun attached to the subordinate verb.

(34) Si V ay kuashi !wati-n ra uyepe tuyuka=ri already sun get.up-NMLZ 3SG.M go.down ground=DIF

Si V tupapenan [r=utsu-ka-tsen] again 3SG.M=go-REI-PUR3

‘When the sun is already up, he again goes down to the ground [to keep going]’

(35) Si V [Ai V kanata=ay ya=chasu uchima y=umi-tsen light=already 3SG.F=AFF go.out 3SG.F=see-PUR3

O] !r!kari=pura mosquito.net=FOC

‘At dawn, poor him gets out [to see the mosquito net]’The examples in (36)–(38) demonstrate the lack of coreference restrictions

in tsen constructions. In (36), the main clause A and O are corereferential with the purpose clause A and O, respectively. In (37), none of the arguments are coreferential with each other (note that the “outcome” of the making, ‘my house’, is encoded as a postpositional purpose phrase which must be nominal-ized by means of -n to function as the syntactic object of the tsen clause). 15

15 Postpositional phrases marked by the purpose adposition -ra function as any other oblique in the language.

(a) tsa papa chikari kawapuri ya !rara=ra 1SG.F father look.for tree.sp 3SG.F canoe=PUR

‘My husband looks for cahuapuri tree for his canoe’



!"!#$# %&'%"() *+#&()( 55

In (38), the coreference is between the main clause S and the comitative oblique in the purpose clause.

(36) Ai V Oj Ai V ra t!k!ta ikian arawata=tsuriay [ra= ichima-ta-tsen 3SG.M tie this monkey=PAS3 3SG.M= get.out-PUR3

Oj ura] 3SG.O

‘He tied this monkey (to his back) [to take him out (from the deep jungle)]’

(37) A V O A V ta yumunu kuriki=chasu [ta=mirikua yauki-tsen 1SG.M send money=AFF 1SG.M=wife make-PUR3

O t=uka-ra-n] 1SG.M=house-PUR-NMLZ

‘I send money [for my wife to make my house]’(38) V Sj

uri-ari ikian rama mai=kana ikia-ka come-PROG this other mestizo=PL.M here=LOC

S V COMj [epe kumitsa-ka-tsen rana=muki] 2PL say-REC-PUR3 3PL.M=COM

‘These other mestizos are coming here [for you to talk with them]’

4.4. Combining purpose clauses. Purpose clauses often appear in se-quence in language use. (39) and (40) illustrate how these constructions interact. In (39), because the matrix dative is the antecedent for coreference, the subsequent purpose clauses can only be tsen clauses. However, in (40), the matrix O is the antecedent and now tara is accepted. (41) shows that tsen is also accepted here, as long as the subject is overtly expressed.

(39) S V OBLi [Si V-tsen OBL] tana kumitsa ikuatawara-tsui (r=)iriwa-tsen/*tara yurimaka 1PL.M talk teacher-DAT 3SG.M=return-PUR3 Yurimaguas

[Si V OBL] (ra=) kumitsa-tsen/*tara mai=kana=muki 3SG.M= talk-PUR3 mestizo=PL.M=COM

‘We talk to the teacher [for him to go back to Yurimaguas city] [to talk to the mestizos]’

(40) A V Oi [Si V-tara OBL] tana yumunu ikuatawara " iriwa-tara yurimaka 1PL.M send teacher return-PUR3 Yurimaguas



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/56

[Si V-tara OBL] " kumitsa-tara mai=kana=muki talk-PUR3 mestizo=PL.M=COM

‘We send the teacher [to go back to Yurimaguas city] [to talk to the mestizos]’

(41) A V Oi [Si V-tsen OBL] tana yumunu ikuatawara r=iriwa-tsen yurimaka 1PL.M send teacher 3SG.M=return-PUR3 Yurimaguas

[Si V-tsen OBL] ra=kumitsa-tsen mai=kana=muki 3SG.M=talk-PUR3 mestizo=PL.M=COM

‘We send the teacher [to go back to Yurimaguas city] [to talk to the mestizos]’

It is worth pointing out the presence of oblique phrases within the purpose clauses in (39)–(41). Having discussed in detail the three positive purpose constructions available in the language, I next present a frequency study that examines the functional load of each construction.

5. Frequency study. Given that -tara and -mira present themselves when the coreference conditions are right, my initial hypothesis was that the overwhelmingly most frequent purpose clause in discourse would be the simple -tsen with coreference between matrix A/S and purpose A/S. In other words, the tsen construction could perhaps be thought of as the default purpose clause. However, this hypothesis was soon to be proved incorrect.

This section reports a corpus study of Kokama purpose clauses in a data-base that consists of 36 texts containing 4,851 intonation units (about 5,500 clauses, 700 pages). The texts comprise a variety of genders and topics, in-cluding traditional stories, personal narratives, and spontaneous conversations collected from a total of 35 speakers. There are 273 purpose constructions in the database. The number of tokens per construction is presented in table 1. 16

The #rst question I explore is: what is the nature/amount/signi#cance of the overlap among constructions in language use? Table 2 gives the distribu-tion of purpose constructions in the text database, merging the arguments according to the grammatical patterns presented in #gure 1. Table 2 focuses only on the overlap between tara/mira constructions and tsen constructions; that is, it ignores tsen constructions with patterns A = S/A/O and others (for the functional load of -tsen, see table 4).

From table 2, it is possible to rule out potential correlations between par-ticular constructions and coreference conditions. In fact, the use of tara versus

16 The numbers in table 1 include nominalized purpose clauses (n=12, 4%) and unclear cases (self-corrections, repetitions, etc.; n=9, 3%). These tokens are ignored in tables 2–4.



!"!#$# %&'%"() *+#&()( 57

mira constructions is not statistically di!erent; whereas the use of mira versus tsen construction is signi#cantly di!erent. 17 Notice that, contrary to expecta-tions, the number of appearances of mira is twice that of tsen. If tsen were the default construction, it should be expected to be more frequent that mira.

But could more #nely grained coreference conditions show predictive re-lationships? This question was pursued by looking more closely at the core-ference relationships for each construction and breaking down the argument clusters. The results are given in table 3, which is in fact a close-up of table 2.

Table 3 shows that, under certain coreference conditions, the overlap seems more extensive than in others. For instance, while in the S = S pattern both -tara and -tsen are equally frequent, in S = A -tara is twice as frequent as -tsen. Another noteworthy point is the very low frequency of constructions

17 tara vs. tsen (&2 (1) = 0.7463, p = .3877). mira vs. tsen (&2 (1) = 8.3208, p = .003919).

TABLE 1 APPEARANCES OF PURPOSE CONSTRUCTIONS

IN THE DATABASE

n %-tara 82 30-mira 47 17-tsen 144 53Total 273 100

TABLE 2 OVERLAP BETWEEN -tara/-mira AND -tsen CONSTRUCTIONS

Purpose Nominative (S/A) OMatrix n % n %

Absolutive (S/O) tara 72 54 mira 37 70tsen 62 46 tsen 16 30

Total 134 100 53 100

TABLE 3 PURPOSE CLAUSES AND THEIR COREFERENCE CONDITIONS

Purpose S A OMatrix n % n % n %

S -tara 22 29 -tara 44 49 -mira 17 32-tsen 22 29 -tsen 17 19 -tsen 0 0

O-tara 2 3 -tara 4 4 -mira 20 38-tsen 17 22 -tsen 6 7 -tsen 16 30



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/58

with implicit subjects where shifts such as O = S and O = A occur. However, this may be due to the costs of online processing. Experimental research on processing of relative clauses has found that certain argument con#gurations are harder to process than others. Taking the matrix subject as the concep-tual ground, a shift in perspective such as S = O is harder to process than O = S, O = O, or S = S (MacWhinney and Pléh 1988:106, as reported by Schmidtke-Bode 2009:64). Given the limited amount of data, I do not pursue this hypothesis here.

In 4.3 above, I showed that tsen constructions do not have coreference constraints. In fact, compared to tara/mira clauses, tsen clauses show up in a much wider range of conditions. Table 4 o!ers a breakdown of the functions of tsen clauses. Note that clauses without coreference represent 16% of its functional load. This #nding, discussed in 6.3, is an important clue for my hypothesis that the information status of participants plays a role in speakers’ preference of one construction over the other.

The last parameter I looked at is the position of the purpose clause with respect to the matrix clause. Thompson (1985:61) found that depending on their position in the sentence, purpose clauses can assume di!erent functions. Initial purpose clauses generally guide the reader on how to associate the information following the purpose clause with the material preceding it. In contrast, #nal purpose clauses modify the matrix clause locally. Initial purpose clauses are then less tightly integrated to the matrix clause than #nal purpose clauses (Schmidtke-Bode 2009:125). In Kokama, the initial–#nal distinction is not a strong parameter, as table 5 indicates. First, the overwelming major-ity of purpose clauses are postposed. Second, there are no real di!erences in patterns of distribution among the three types of clauses. 18

18 Another parameter I explored is collocational restrictions. A quick examination of the database reveals that the three purpose constructions apparently do not exhibit collocational restrictions, at least with reference to the verbs that occur in them. As is shown throughout the paper, a number of verbs occur in all three purpose clauses, including aya ‘shoot’, erutsu ‘bring’, yauki ‘make’, among others.

TABLE 4 FUNCTIONAL LOAD OF -tsen CONSTRUCTIONS

Coreference Conditions n %S = S/A 39 28O = S/A/O 39 28A = S/A/O 32 23DAT = S 2 1No coreference 22 16Questions, imperatives 6 4Total 140 100



!"!#$# %&'%"() *+#&()( 59

Thus, it is posible to say that, in language use, there is a signi#cant over-lap between tara and ts en constructions, on the one hand, and mira and tsen constructions, on the other (table 3). The question, then, is what triggers the speaker’s choice in the areas of overlap? In 6, I propose that this choice is driven by a set of interrelated discourse-pragmatic organization parameters. I show that at least one parameter is needed to motivate selection of one construction over a competing construction.

6. Kokama purpose clauses: functional load. As discussed above, while the distribution of -tara and -mira can be accounted for on gram-matical grounds, tsen clauses cannot. More speci#cally, speakers have two choices when they want to indicate purpose by means of adverbial clauses: -tara/mira clauses (whose distribution depends on the missing argument in the adverbial clause) or -tsen clauses (see #gure 1). However, what trig-gers the speaker’s choice? In this section, I examine the conditions for the distribution of purpose constructions in language use. The results suggest that speaker’s choice is driven by interrelated information structure factors, such as temporal integration (6.1), succesful outcome of the desired event (6.2), and the discourse status of referents (6.3).

6.1. Temporal integration. This parameter refers to how both events are depicted in the timeline. The main event and the purposive event could be construed as occurring almost simultaneously, immediately contiguous, distant from one another, or in other ways. First, in (42) and (43), the in-terpretation is that the event in the subordinated clause occurs immediately after the main event, while in (44) this temporal integration does not ex-ist. This claim is based on speakers’ observations about these examples. In (42), the speaker is talking about a young woman who comes to the speaker’s house only at night and proceeds directly to sleep. The situation is depicted as if ‘sleeping’ follows ‘showing up’ without delay. Likewise, in (43), the speaker reports that as soon as the community members were given some donations, they took them to the village. He portrays the two

TABLE 5 INITIAL VERSUS FINAL PURPOSE CLAUSES

tara mira tsen

Initial 7 9% 1 2% 7 5%Final 75 91% 46 98% 137 95%Total 82 100% 47 100% 144 100%



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/60

events, ‘getting out’ and ‘being heard,’ as immediately contiguous. In (44), the implication of ‘going back’ right after ‘being sent’ is not there.

(42) rai katupe !p!tsa ikia-ka ["i uk!r!-tara] 3SG.M show.up night this-LOC sleep-PUR1‘She shows up at night here [to sleep]’

(43) tsawa=nui yayti uchima ["i ya=irua=nu tsenu-mira] spirit=PL.F also go.out 3SG.F=brother=PL.F hear-PUR2‘The spiritsi also get out [for their relatives to hear themi] (to be

heard by their relatives)’(44) yantsui doctor yumunu penu=tsuri [penu

there doctor send 1PL.EX.F=PAS3 1PL.EX.Firiwa-tsen penu uka-ka] come.back-PUR3 1PL.EX.F house=LOC

‘Then the doctor sent us [to go back to our house]’

Given that in (44) the coreferential conditions for tara- are met, I asked two speakers if the tsen clause could be replaced by a tara clause. That is, I o!ered the apparently parallel sentence in (45). The two speakers agreed that, while (45) is good, in the context of the narrative -tsen is the appropriate construction. One speaker explained it this way: “With -tsen it means that the doctor told us we were free to go, by ourselves, by our own means. With -tara it would mean that he not only told us what to do, but also that perhaps they provided us with a boat to travel immediately.”

(45) yantsui doctor yumunu penu=tsuri [iriwa-tara . . .] there doctor send 1PL.EX.F=PAS3 come.back-PUR1‘Then the doctor sent us to go back. . . ’

The integration between the main event and the purpose event becomes even more obvious in (46) and (47), provided by one consultant. In a context where a baby is cryng hard, (46) would mean that if the mother nurses the baby he will eventualy stop crying. That is, both ‘nursing’ and ‘shutting up’ are presented as two separate events. The consultants explained that in this scenario, tara would not work. (47), however, describes a scene of a mother nursing a baby who is already silent. In this case, tsen would be odd.

(46) chuchu-t-ay ya=chinta-tsen (*tara) breast-CAU=3SG.F 3SG.F=silence=PUR3‘Nurse him so that he shuts up’ [stops crying]

(47) ya yumi chuchu ya=chinta-tara (#tsen) 3SG give breast 3SG.F=silence=PUR3‘She feeds (him) for him to shut up’



!"!#$# %&'%"() *+#&()( 61

6.2. Successful outcome. Typologically, the likelihood of a purpose being achieved is higher if the person having a purpose is directly involved in the realization of that purpose (see 2.2). However, in Kokama, achieve-ment is implied in constructions that share arguments, even when the shared arguments are not agents. Both mira and tara constructions imply that the purposive event takes place successfully. In (48), repeated from (11), the sit-uation is presented as if the hunting happened. One consultant explains that if the husband went to the forest with the intention to hunt but something prevented him from doing it (e.g., he was bitten by a snake), the speaker would have used a tsen clause. Likewise, in (49), the speaker reports that when the community members were given some donations, they took all the gifts to the village. In this scenario, ‘carrying things away’ de#nitely took place. The sequence of purpose clauses in (50) illustrates this further. The speaker says that during curative events, elders smoke tobacco to achieve visions. The implication in (50) is that ‘smoking’ takes place, but such an implication is not present for the tsen event.

(48) tsa mena uts=uy !w!rati [aya-tara] 1SG.F husband go=PAS1 jungle hunt-PUR1‘My husbandi went to the forest to "i hunt’

(49) aytsemeka rana yumi upi mari=pura tana=tsui truth 3PL.M give all thing=FOC 1PL.M-DAT

[" tana erutsu-mira] 1PL.M bring-PUR2

‘For real they give us all kinds of thingsi [for us to take "i]’(50) yama yankata p!t!ma [" inu tuaminu ts!ki-mira],

other put tobacco 3PL elders suck-PUR2[inu warika-tsen], [p!t!ma ts!ki-wa inu warika-tsen] 3PL go.up-PUR3 tobacco suck-GER 3PL go.up-PUR3

‘Others put tobacco [for their elders to smoke], [for them to go up], [for them to go up smoking tobacco]’

In some purpose clauses, the main event is construed as a necessary con-dition for the event expressed in the subordinate clause to take place. In mira/tara clauses, the desired event is totally dependent on the main event. In contrast, tsen clauses convey a desire, a wish, but the main event is not a necessary condition. In (51), the context is a group of kids looking for something to eat, and the only way to get to the fruit is to climb the tree. A similar situation is given in (52). This sense is absent in (53); there is nothing about the money being indispensable to make the house nor the implication of the house-making being a successful outcome—i.e., the wife could spend the money on something else.



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/62

(51) ra=tu=kana warika-pa raepe [k!ma " eyu-tara] 3SG.M=AUG=PL.M go.up-CPL there guava eat-PUR1‘They climb up there [to eat guava]’ (they do eat it)

(52) aykua-n ikaku [" t!ma awa timi-n=nu be.sick-NMLZ diet NEG person disconnect-NMLZ=PL.F

umi-mira] see-PUR2

‘The sick onei goes on diet-seclusion so that the strangers don’t see (himi)’

(53) ta yumunu kuriki=chasu [ta=mirikua yauki-tsen 1SG.M send money=AFF 1SG.M=wife make-PUR3

t=uka-ra-n] 1SG.M=house-PUR-NMLZ

‘I send money for my wife to make my house’

Perhaps this idea is more obvious in the examples below, provided by one consultant. The speaker explains that in (54) taking a bath is not necessary to be able to go somewhere. The context o!ered for (55) is this: we were looking for transportation to go somewhere, and we found out a canoe was available.

(54) ini yatsuka utsu-tsen/*tara 1PL bathe go-PUR3‘We bathe (in order) to go’

(55) emete !rara ini utsu-tara exist canoe 1PL go-PUR1‘There is a canoe for us to go’

6.3. Discourse status of referents. Another parameter conditioning the dis-tribution of purpose clauses is the discourse status of the referents in the desired situation. While mira/tara clauses are banned from introducing new participants into the discourse, tsen clauses are allowed to do this. The omitted arguments in both mira/tara clauses belong not only to the local context but also to the discourse context. In the sentences that precede (56), a hunter is introduced. In that sense, the elided subject of the tara clause is the topic of the preceding discourse. Also, the tara clause introduces ‘animal’, and so in the subsequent mira clause, it can be omitted. In contrast to mira/tara clauses, the tsen clause in (57) introduces a participant new to the addressee: ‘old lady.’

(56) !w!rati y=utsu=uy [animaru chikari-tara] forest 3SG.F=go=PAS1 animal look.for -PUR1

[y=aya-mira] 3SG.F=shoot-PUR2

‘He went to the forest [to look for animals] [to shoot]’



!"!#$# %&'%"() *+#&()( 63

(57) marawe tseta-tara=nu piyata ya=ukua tsa=tsui, fan want-NMLZ.A=PL.F ask 3SG.F=habitual 1SG.F=DAT

[ajan wija=chasu yauki-tsen inu marawe=ra-n] this elder=AFF make-PUR3 3PL.F fan=PUR-NMLZ

‘The ones who want a fan usually ask for it to me, [so that this old lady (will) make them something to fan themselves with]’

It has been found that in some purpose clauses, both the main and the em-bedded event might be asserted (Cristofaro 2003 and Croft 2001). According to Cristofaro, if purpose clauses can have both adverbial (e.g., he went to give them rice) and coordinate (e.g., he went and gave them rice) interpretations, then both clauses are asserted. From this observation, it follows that pieces of information within purposes clauses can be focalized. This seems to be the case in tara/mira clauses but less likely in tsen clauses. Vallejos (2009) found that the morpheme =pura, together with constituent order and prosody, can convey a range of focus subtypes. It was also found that =pura is not restricted to marking constituents within the matrix clause; it can also mark speci#c arguments within subordinate clauses, like in (58). 19

(58) A V Oi raepetsui rana yumunu tana [ikian !m!nan after 3PL.M send 1PL.EX.M this long.ago

O Ai V-tara rana ritama=pura " umi-tara] 3PL.M community=FOC see-PUR1

‘After that they sent us [to see their old town]’

The context for (58) is the following: the speaker is talking about one of his visits to a neighboring country. He mentions that his host took him on a tour to visit several places. In this example, the subordinate clause is ‘to see their old town’. Looking at the prosody, we see that (58) includes a clear stress on =pura, attached to ‘community’. Within the discourse context, it would be responding not to a question like “what did they take you for?” but to a question like “what did they take you to see?” Thus, the focused portion, ‘their old town’, is the O argument of the subordinate verb ‘see’. This means two

19 Similar examples are found in adverbial clauses expressing the reason the event in the matrix clause is realized. In (a), the speaker is talking about an earthquake that destroyed the city. The subordinate clause is ‘because the earthquake destroyed all their houses’, but the part under the scope of =pura is ‘all their houses’, the O argument of ‘destroy’. Notice that the stress is on upi ‘all’. In the communicative context, this construction is not about ‘why they left’ but about ‘what the earthquake destroyed’.

(a) [karura uwari-ta [upi rana uka=pura]-ikua] rana ichari ikian earthquake fall-CAU all 3PL.M house=FOC-RSN 3PL.M leave this

‘Because the earthquake destroyed all their houses, they left this (place)’



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/64

things: (i) the interpretation of the whole subordinate clause as being in focus is not likely; and (ii) the presupposition for this utterance contains an open variable, i.e., does not evoke a set of variables. That is, this does not seem to be a case of contrastive focus but of information focus (Dick et al. 1981).

An example of a tsen clause in which one of its arguments is the host of =pura is given in (59). Given the discourse context, I would argue though that the focus domain in (59) is not ‘mosquito’, the O argument of ‘see’, but the entire purpose clause, ‘to see the mosquito net’. However, given the limited data, this must remain as a hypothesis to be examined with a larger database.

(59) kanata=ay ya=chasu uchima [y=umi-tsen light=already 3SG.F=AFF go.out 3SG.F=see-PUR3 !r!kari=pura] mosquito.net=FOC

‘At dawn, poor him goes out [to see the mosquito net]’

Finally, purpose clauses in sequence operate as a reference-tracking mecha-nism. As illustrated throughout this paper, tara/mira not only convey purpose, they also indicate which argument is omitted in the purposive event and what the reference of the omitted argument is. In (60), tara indicates that A is missing in the purposive event and that its antecedent is the matrix S, ‘I’. The following mira clause indicates that O is missing and that its antecedent is ‘bulb of palm’, the O of the preceding tara clause.

(60) Si ya=puka ts=utsu !w!rati aja-mia 3SG.F=when 1SG.F=FUT1 forest this-MOD

[Oj Ai V-tara] [Oj A V-mira tsuwa " chikari-tara " ts=yauki-mira palm.bulb look.for-PUR1 1SG.F=make-PUR2 OBL] marawe=ra fan=PUR

‘When (they want it), I go to the forest [to look for bulb of palm] [to make it into a fan]’

From this section, we can conclude that tara/mira clauses convey not only successful outcome but also that the main event needs to be construed as indis-pensable for the desired outcome to come about. In tsen clauses, however, the purposive event can not only be conceived as unrealized but also as dispens-able. When no coreference exists, as in (53), the arguments of the dependent event have more control/volition about whether to execute the desired event or not. In addition, mira/tara clauses generally do not introduce new partici-pants into the scene, that is, they belong to the discourse context. However,



!"!#$# %&'%"() *+#&()( 65

tsen clauses can introduce new participants. These #ndings nicely correlate with the fact that tsen clauses are syntactically less integrated than tara/mira clauses. Kokama purpose clauses reveal themselves as a reference-tracking mechanism, especially when they interact within a single complex sentence.

7. Summary and conclusions. The main conclusion of this study is that purpose clauses in Kokama can be fully accounted for by taking into account considerations of syntax, semantics, and pragmatics. Two crucial parameters are coreference constraints and argument realization. In tara clauses, the purposive subject must be omitted and its reference is recov-ered from the matrix absolutive; in mira clauses, the purposive object is omitted and is coreferent with the matrix absolutive; and in tsen clauses, the reference of the arguments is unconstrained. Given these facts, tsen constructions are syntactically less integrated than tara/mira constructions. A similar coreferentiality system has not yet been documented, and so this study contributes to the typology of adverbial clauses in general and of purpose clauses in particular.

However, syntax alone cannot account for the distribution of Kokama pur-pose clauses. As #gure 1 above shows, there is a considerable overlap be-tween tara clauses and tsen clauses, on the one hand, and mira clauses and tsen clauses, on the other. Semantic and discourse-pragmatic variables play a role in determining the form of the utterance. While intention is a constant feature across the three types, di!erent constructions are associated with dif-ferent degrees of temporal integration and di!erent probabilities of successful realization, as well as the speaker’s assessments of the discourse status of the referents. In tara/mira constructions: (i) the event in the subordinated clause needs to be construed as occurring immediately after the main event; (ii) there is an implication that the purposive event is successfully realized, and to some extent the main event is portrayed as a condition for the desired event to come about; and (iii) the omitted argument in the purpose clause belongs to the discourse context; it is already in the addressee’s knowledge set, i.e., it is generally the topic of the preceding sentences. In tsen clauses, none of these conditions apply. These #ndings correlate nicely with the fact that tsen clauses are syntactically less integrated than tara/mira clauses. Also, both the main and purpose events can be asserted—hence pieces of information within the purpose event can be focused.

From typology, we know that the absence of arguments in dependent clauses is commonly due to the fact that the information pertaining to the missing argument is easily recoverable or irrelevant in the discourse context. This is coherent with our #ndings in tara/mira clauses. It is also known that con-structions with explicit arguments tend to non-coreferentiality. Tsen clauses require overt arguments, and although this is the only construction where coreferentiality is not required, many tsen clauses display coreferentiality.



!"#$%"&#!'"&( )'*%"&( '+ &,$%!-&" (!".*!/#!-/66

Based on these facts, I argue that purpose clauses in Kokama are grammati-calized information structure management strategies. Correlations between semantic integration of events and syntactic integration of clauses have long been reported (Givón 1980). Kokama adds a new parameter: pragmatic tight-ness between events is re$ected by syntactic integration between clauses.

A noteworthy #nding is the typological novelty of tara/mira constructions with respect to the controller of coreference. Cross-linguistically, implicit purposive subjects are not only generally controlled by the matrix subjects, but also little evidence has been found of constructions in which the main event and the purposive event are not performed by the same entity. As demonstrated here, in tara/mira constructions, the implicit argument in the purposive event is obligatorily controlled not by the matrix subject (S/A) but by the matrix absolutive (S/O). In this sense, Kokama purpose clauses do not conform to the communicative preference of agent-binding. What, then, is the functional motivation for such constructions? Attempts to answer this ques-tion will only be possible when more descriptions of such systems become available (Schmidtke-Bode 2009:64).

Tracing the diachrony of these constructions could shed light on the syn-tactic coreference requirements of today’s Kokama purpose constructions. As indicated earlier, the most obvious hypothesis to explore would be that the purposive morphemes -tara and -mira grammaticalized from the Tupi-Guaraní nominalizer of agent *-tar and the patient nominalizer *-pyr, respectively (Jensen 1998). Although other Tupi-Guaraní languages do not exhibit a system comparable to that of Kokama, an examination of the pathways of the above nominalizers in those languages could help us understand the intermediate steps in the grammaticalization process.

Overall, this study shows that syntactic choices are not determined purely syntactically. To understand why speakers employ speci#c kinds of construc-tions over others, we need to take into account semantic and information structure parameters. This paper supports the position that a thorough account of complex sentences needs to include both formal properties and pragmatic structure. As shown here, information structure has repercussions in the form of the sentences and is a crucial part of communicative competence in a language.

REFERENCES

AIKHENVALD, A. Y. 2003. A Grammar of Tariana, from Northwest Amazonia. Cambridge: Cam-bridge University Press.

ANDREWS, A. D. 2007. Relative clauses. Language Typology and Syntactic Description: Complex Constructions, 2nd ed., ed. T. Shopen, vol. 2, pp. 206–36. Cambridge: Cambridge University Press.

CRAIG, C. 1977. The Structure of Jacaltec. Austin: University of Texas Press.CRISTOFARO, S. 2003. Subordination. Oxford: Oxford University Press.



!"!#$# %&'%"() *+#&()( 67

. 2005. Purpose clauses. The World Atlas of Language Structures, ed. M. Haspelmath et al., pp. 506–7. Oxford: Oxford University Press.

CROFT, W. 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press.

. 2003. Typology and Universals. Cambridge: Cambridge University Press.DICK, S.; M. HOFFMANN; J. R. DE JONG; S. I. DJIANG; H. STROOMER; AND L. DE VRIES. 1981. On the

typology of focus phenomena. Perspectives on Functional Grammar, ed. T. Hoekstra, H. van der Hulst, and M. Moortgat, pp. 41–74. Dordrecht: Foris.

DIXON, R. M. W. 1977. A Grammar of Yidiny. Cambridge: Cambridge University Press. . 1994. Ergativity. Cambridge: Cambridge University Press.

GIVÓN, T. 1980. The binding hierarchy and the typology of complements. Studies in Language 4:333–77.

. 2001. Syntax: An Introduction. Vol. 2. Amsterdam: John Benjamins.HAIMAN, J., AND S. THOMPSON. 1984. “Subordination” in universal grammar. Proceedings of the

Berkeley Linguistics Society 10:510–23.JACKSON, H. 1995. Grammar and Meaning: A Semantic Approach to English Grammar. London:

Longman.JENSEN, C. 1998. Comparative Tupí-Guaraní morphosyntax. Handbook of Amazonian Languages,

vol. 4, ed. D. C. Derbyshire and G. K. Pullum , pp. 487–618. Berlin: Mouton de Gruyter.LEHMAN, C. 1988. Towards a typology of clause linkage. Clause Combining in Grammar and

Discourse, ed. J. Haiman and S. A. Thompson, pp. 181–226. Amsterdam: John Benjamins.LICHTENBERK, F. 1995. Apprehensional epistemics. Modality in Grammar and Discourse, ed. J. L.

Bybee and S. Fleischman, pp. 293–327. Amsterdam: John Benjamins.MACWHINNEY, B., AND C. PLÉH. 1988. The processing of restrictive relative clauses in Hungarian.

Cognition 29:95–141.SCHMIDTKE-BODE, K. 2009. A Typology of Purpose Clauses. Amsterdam: John Benjamins.SEARLE, J. R. 2006. What is language: Some preliminary remarks. <socrates.berkeley.edu/~jsearle/

whatislanguage.pdf>.THOMPSON, S. A. 1985. Grammar and written discourse: Initial vs. #nal purpose clauses in En-

glish. Text 5:55–84.THOMPSON, S. A.; R. E. LONGACRE; AND S. J. HWANG. 2007. Adverbial clauses. Language Typol-

ogy and Syntactic Description: Complex Constructions, 2nd ed., ed. T. Shopen, vol. 2, pp. 237–300. Cambridge: Cambridge University Press.

VALLEJOS, R. 2009. The focus function(s) of =pura in Kokama-Kokamilla discourse. IJAL 75:399–432.

. 2010a. Is there a ditransitive construction in Kokama-Kokamilla? Studies in Language 34:75–107.

. 2010b. A grammar of Kokama-Kokamilla. Ph.D. dissertation, University of Oregon.VAN VALIN, R. D., JR. 1993. Advances in Role and Reference Grammar. Amsterdam: John

Benjamins.



Reference constraints and information-structure management in Kokama purpose clauses: A typological novelty? International Journal of American Linguistics, 80.1: 39-67. 2014.

Documents