Proceedings of the Second International Conference on Dependency Linguistics (DepLing 2013), pages 98–107,Prague, August 27–30, 2013. c© 2013 Charles University in Prague, Matfyzpress, Prague, Czech Republic

Pragmatic Structures in Aymara

Petr HomolaCodesign, s.r.o.Czech Republic

phomola@codesign.cz

Matt ColerINCAS3

The NetherlandsMattColer@incas3.eu

Abstract

The paper describes overt marking of in-formation structure in the indigenous An-dean language Aymara. In this paper weshow that although the word order is free,Aymara is not a discourse-configurationallanguage (Kiss, 1994); instead, informa-tion structure is expressed only morpho-logically by pragmatic suffixes. The mark-ing of ‘topic’ is more flexible that themarking of ‘focus’ (be it at the clause levelor NP-internal). Since overt marking ofinformation structure is partial, this pa-per also devotes considerable attention tothe resolution of underspecification in Ay-mara.

1 Introduction

Chomsky’s original approach to formal syntax as-sumes that sentences consist of constituents andthat the type and order of these constituents defineconfigurations that specify grammatical relations.As has been shown by Hale (1983), there are lan-guages in which word order has no or limited rele-vance for grammatical relations and the respectiveconstituent trees are flat. In most such languages,word order is said to specify information struc-ture (hence the name discourse-configurationallanguages coined by Kiss (1994), as opposedto syntax-configurational languages). We showthat Aymara is neither syntax-configurational nordiscourse-configurational. In this language, in-formation structure is expressed solely morpho-logically. This is very rare and therefore Ay-mara is very important for the research of informa-tion structure in general. As Bossong (2009) putsit (referring to Aymara and Quechua): “Cross-linguistically, grammemes explicitly expressingthe function of theme are not very frequent; gram-memes expressing the function of the rheme are a

highly marked typological rarity [. . . ]. Still moreidiosyncratic is the combination of the thematicmarker with a grammeme combining the two func-tions of question and negation.”

We are not aware of any other language (exceptfor, to some extent, Quechua) in which word or-der is irrelevant to information structure. Hard-man (2000) says about Aymaran languages that“[t]he structural elements of a sentence may occurin any order and are at the disposal of the speakerfor stylistic play.” Hardman et al. (2001) say thatword order in Aymara “affects only style, not thegrammar nor basic semantics.” Bossong (1989)says about Quechua (which is typologically verysimilar) that “word order is not only free but itis not primarily used as a means for expressingpragmatic functions as such.” A detailed analy-sis of morphological information structure mark-ing in Aymara can thus shed light on this moduleof grammar which is not expressed morphologi-cally in most languages.

Section 2 presents basic facts about Aymara.Section 3 describes the overtly marked discoursecategories in Aymara. Section 4 describes themeans of morphosyntactic, semantic and prag-matic referent identification. Section 5 treats prag-matic marking in complex predicates. Finally, weconclude in Section 6.

2 Basic Facts about Aymara

Aymara is spoken by communities in a region en-compassing Bolivia, Chile and Peru, extendingnorth of Lake Titicaca to south of Lake Poopó, be-tween the western limit of the Pacific coast val-leys and eastward to the Yungas valleys. The lan-guage is spoken by roughly two million speakers,over half of whom are Bolivian. The rest residein Peru and Chile. The Aymaran family (com-prised of Aymara, Jaqaru, and Kawki) is a lin-guistic isolate with no close relative. Aymara is

98

an affixal polysynthetic1 language (according toMattissen’s (2006) classification) with a rich mor-phology. It is SOV with modifier-head word order.Aside from the morphologically unmarked sub-ject, all syntactic relations are case-marked, typi-cally on the NP head. Roots can be divided intonouns (including qualitative words), verbs, andparticles. Suffixes, which may have a morpholog-ical or syntactic effect, can be classified as nomi-nal, verbal, transpositional, independent, or prag-matic suffixes. The category of independent suf-fixes is comprised of three suffixes which are notclassifiable as members of either nominal or verbalmorphology which likewise cannot be said to bepragmatic suffixes. These suffixes generally occurprior to inflectional morphology and/or the prag-matic suffixes. Pragmatic suffixes, by comparison,typically suffix to the last word of the entire sen-tence and/or the NP or VP. As their name suggests,their function is overall pragmatic in nature. Nom-inal and verbal suffixes can be subdivided intoinflectional and derivational categories, but giventhe ease with which category-changing transposi-tional suffixes attach to words of any category, of-ten multiple times, it is common to find words withseveral nominal, verbal, transpositional, and inde-pendent suffixes.

Detailed information can be found in (Hardmanet al., 2001; Briggs, 1976; Adelaar, 2007; Cerrón-Palomino and Carvajal, 2009; Yapita and Van derNoordaa, 2008).

3 Overt Marking of InformationStructure

As mentioned in the introduction, word order inAymara is not used to express information struc-ture. As an example, consider the following sen-tences in English (syntax-configurational), Rus-sian (discourse-configurational) and Aymara (sub-script N marks contextually new, i.e. nonpre-dictable information):

(1)(a) Peter cameN

It’s PeterN who came (*Came Peter)(b) Pëtr prišëlN

Prišël PëtrN(c) Pedrox jutiwaN or JutiwN Pedroxa

PedrowN jutixa or Jutix PedrowaN

1See (Baker, 1996).

Due to overt morphological marking of infor-mation structure, there are discontinuous phrasesin which the “gap” is not motivated by pragmat-ics, such as (2)2. The corresponding tree is givenin Figure 1.

(2) Juma-n-xyou-gen-ref

jiw-i-wdie-smpl3→3-npred

kimsathree

ch’iyarblack

phisi-ma-xacat-poss2-ref

“Your three black cats died.”

jiwiwNPRED

phisimaxaREF

jumanxREF kimsa ch’iyar

Figure 1: The surface syntax tree of (2) with a dis-continuous noun phrase

Of course, the possessive pronoun jumanx couldalso form a continuous constituent with the rest ofthe NP. Or, it might be omitted since the posses-siveness is already expressed with the suffix -ma(phisimaxa).

Overt marking of information structure is oblig-atory in Aymara. A sentence without proper prag-matic suffixes (or with marking that would destroyintersentential cohesion) is considered ill-formed.It should also be noted that the marking of infor-mation structure is orthogonal to the morphosyn-tactic expression of evidentiality.3

3.1 ReferentialityAymara has overt marking of referentiality. Thenotion of referentiality roughly corresponds to

2We use the following abbreviations in the glosses:ACC=accusative, AG=agentive, AGGR=aggregator,ALL=allative, BEN=benefactive, CAUS=causative,COM=comitative, CONJ=conjectural, DIR=directive,FUT=future tense, GEN=genitive, HON=honorific,HUM=human, IMPER=imperative, INCMPL=incompletive,INF=infinitive, LIM=limitative, LOC=locative, MIR=mirative,NEG=negative, NPRED=nonpredictable, PAST=past tense,PART=past participle, POSS=possessive, POSS1=1st personpossessive, POSS2=2nd person possessive, POSS3=3rdperson possessive, POT=potential, PROGR=progressive,QM=question marker, REF=referential, REFL=reflexive,SMPL=simple tense, TRGRDS=transgressive (differ-ent subject), TRGRSS=transgressive (same subject),VRBL=verbalizer

3Aymara has a three-way system of evidentiality: wit-nessed, knowledge through language, and inferred.

99

what Sgall et al. (1986) call ‘contextual bound-edness’. We follow Sgall’s (2009) terminologyhere: “[Contextually bound] items are presentedby the speaker as referring to entities assumed tobe easily accessible by the hearer(s), in the pro-totypical case ‘given’. They refer to ‘established’items, i.e. to those that were mentioned in the pre-ceding co-text and thus still are sufficiently salient,or to the permanently established ones (indexicals,those given by the relevant culture or technical do-main, etc.).”4

For the majority of bilingual speakers, referen-tiality is optionally marked with the suffix -xa,5

however for monolinguals it is done so more con-sistently. There are also slight dialectal differ-ences.

For example, in (3) the verb is marked as refer-ential because the event is known from the context(the speaker supposes that the hearer knows thatsomeone came, otherwise he would not ask).

(3) Khiti-swho-qm

jut-i-xa?come-SMPL3→3-REF

“Who did come?”

On the other hand, in (5) the verb is not markedas referential because it expresses an event whichis apparently unknown to the hearer (since he hasexpressed surprise in (4)).

(4) Aka-n-ka-sk-ta-säthis-loc-vrbl-progr-smpl2→3-mir“Oh, you are here!”

(5) Jïsa,yes

wasüru-wyesterday-npred

kutin-x-tareturn-COMPL-SMPL1→3

“Yes, I came back yesterday.”

The adverbial expression wasüruw “yesterday”is marked as nonpredictable6 (even though it isinherently referential) because it is more salientthan the verb7 (B is surprised to see A because

4We do not use the terminology ‘topic-focus’ or ‘theme-rheme’ because the paradigms of suffixes that express infor-mation structure do not correlate with it. For example, thetwo paradigms are not always mutually exclusive (two suf-fixes from different paradigms can mark the same word).

5In the literature this suffix is usually called “topicmarker” or “attenuator”.

6This grammatical feature is described in the next subsec-tion.

7“More salient” means that it is more important in the cur-rent context. Informally, we could also say “less predictable”.The scale of salience is important for intersentential referentidentification.

he thought that A had left or was about to leave).Nonetheless, the verb is nonpredictable, too, be-cause B does not know what exactly happened (Amay have come back earlier or may not have leftat all).

Note that -xa marks referentiality, notknown/given information, as illustrated in (6).

(6) Qharüru-xtomorrow-ref

sara-:-wago-fut1→3-npred

“I will go tomorrow.”

The adverbial expression qharürux “tomorrow”is marked as referential because it is inherently ref-erential but there is no marking which would spec-ify if it is predictable or not (the verb is alreadymarked as nonpredictable and there can be at mostone NPRED-marker per clause, as explained in thenext subsection).8

It should be noted the the REF-marking is priva-tive, i.e. the absence of -xa does not mean that theentity is not referential.

3.2 NonpredictabilityEntities are nonpredictable if they are ‘new’ tothe hearer (cf. the usual dichotomy of informa-tion structure known/given vs. unknown/new).9

Overt marking of nonpredictability in Aymara ismuch more consequent than that of referentiality.In affirmative sentences, the most salient nonpre-dictable entity is marked with the suffix -wa.10

The placement of the NPRED-marker can be bestexplained by question/answer pairs, such as (7)and (8).

(7) Khiti-taki-swho-ben-qm

ut-xhouse.acc-ref

uta-ch-ta-xahouse-caus-smpl2→3-ref“For whom did you build the house?”

8A reviewer suggested to compare pragmatic markingin Aymara with ‘topic’-markers in Japanese and/or Korean.First of all, whereas in the mentioned languages, topic mark-ing interferes with case marking, in Aymara they are orthog-onal. Also, in Aymara the use of pragmatic markers does notaffect word order. Finally, there are no free topics as in thefollowing sentence from (Iwasaki, 2013):

Zoo-waTOP hana-gaNOM nagai“The elephant; its nose is long.”In Aymara, even if the word order of a NP is relaxed, all

parts of the NP agree with each other.9Again, we follow Sgall’s (2009) terminology: “[Contex-

tually non-bound] items are presented as not directly pre-dictable in the given context, as ‘new’ information (at leastas chosen from a set of available alternatives).”

10In the literature this suffix is often called “affirmative”,“emphasizing”, or “declarative”.

100

(8) Jilata-ja-taki-wbrother-poss1-ben-npred(uta-ch-t-xa)house-caus-smpl1→3-ref“(I built it) for my brother.”

In negative sentences, the NPRED-marker isusually attached to the negative particle jani, asin (9).

(9) Jani-wnot-npred

kullaka-ma-rsister-poss2-all

uñj-k-t-tisee-incmpl-smpl1→3-neg“I did not see your sister.”

However, if an argument of the verb is moresalient than the verb (i.e., it is contrastive), it at-tracts the NPRED-marker, as in (10).

(10) Kullaka-ma-ru-wsister-poss2-all-npred

jannot

uñj-k-t-tisee-incmpl-smpl1→3-neg“It is your sister whom I did not see.”

There are constructions in which NPRED-marksare disallowed, for example in imperative con-structions or in those marked with the conjecturalevidential suffix:

(11) Chur-ita-yagive-imper2→1-hon“Give it to me, please.”

(12) Jut-chi-nicome-conj-fut3→3

“Maybe he will come.”

NPRED-unmarked affirmative sentences in thefuture tense tend to have imperative meaning:

(13) Sara-ñänigo-fut4→3

“Let us go.”

An affirmative sentence without NPRED-marking in the future tense does not alter thecontext of the present discourse and may havemodal meaning (Hardman et al., 2001):

(14) NayI

sar-ä-xago-fut1→3-ref

“I will go, right? / Should I go?”

Noun phrases introduced by mä “one” are indef-inite and therefore always nonpredictable. Nounphrases with the determiners aka “this”, uka“that”, khaya (khä) “yonder” and khuri “awayyonder” are definite and therefore usually pre-dictable unless they are explicitly NPRED-marked(in which case they are contrastive). The followingexample illustrates that referentiality and nonpre-dictability do not exclude each other.

(15) Akathis

warmi-mpi-wwoman-com-npred

mäone

jisk’asmall

marka-nvillage-loc

jiki-s-t-xameet-refl-smpl1→3-ref

“It is this woman whom I met in a smallvillage.”

In verbal complexes, the main (nonfinite) verbmay be NPRED-marked, as in (16) and (17):

(16) T’ant’bread.ACC

ala-ñ-wbuy-inf-npred

mun-tawant-smpl1→3

“I want to buy bread.”

(17) T’ant’bread.ACC

al-iri-wbuy-ag-npred

sara-sk-tago-progr-smpl1→3

“I am going to buy bread.”

3.3 Deeply Embedded NonpredictableElements

Nonpredictability is usually marked at clause levelbut it can occur inside a noun phrase11 if itis required by the discourse context, as in (18)(from (Hardman et al., 2001)).12

(18) Naya-n-xI-gen-ref

pusi-wfour-NPRED

uta-ja-xhouse-poss1-ref

utj-ituexist-smpl3→1

“As for houses, I have four.”

This example shows that REF and NPRED mark-ers need not appear at the end of a (semantic)phrase. Similarly, a nominal modifier may follow

11Hajicová et al. (1998) call this kind of information-structural marking proxy focus.

12In this example, pusiw is not contrastive, it is simply ananswer to the question How many houses do you have? Ofcourse, one could say just Pusiwa.

101

its governor if its information structure status dif-fers from that of its head in which case both wordsreceive pragmatic marking, as in (19) (from (Hard-man et al., 2001)).

(19) Uka-xthat-ref

apilla-xoca-REF

luxu-cha-ta-wafreeze-caus-part-npred“That is FROZEN oca.”

Aymara is a radically nonconfigurational lan-guage.13 However, even languages from this classmay pose word order constraints on “deeply” em-bedded phrases. Indeed, in Aymara noun phraseshave more rigid word order than clause con-stituents.14 For example, while (20) is well-formed, (21) is ill-formed because the modifiersuma “good” does not precede its head.

(20) Sumagood

chuq’i-wpotato-npred

akathis

marka-n-xvillage-loc-ref

ach-ugrow-smpl3→3

“In this village grow good potatoes.”

(21) *Chuq’i-wpotato-npred

sumagood

akathis

marka-n-xvillage-loc-ref

ach-ugrow-smpl3→3

Intended meaning: “In this village growgood potatoes.”

Nevertheless the order of a nominal modifierand its head is not restricted if both have a prag-matic marker, as in (22)–(24).

(22) Suma-wgood-npred

akathis

marka-n-xvillage-loc-ref

chuq’i-xpotato-ref

ach-ugrow-smpl3→3

“GOOD potatoes grow in this village.”

(23) Uta-cha-ña-taki-xhouse-caus-inf-ben-ref

qala-xstone-ref

alluxa-wa-lot-npred

apthapi-ñagather-inf

“One must gather MANY stones to build ahouse.”

13In terms of the generative grammar, languages with therule S → C+ at clause level.

14Unlike, for example, in Latin, where the word order ofNPs is less rigid although not completely free.

(24) Qullq-xmoney.acc-ref

allux-wa-lot.acc-npred

ap-kat-taput-dir-smpl1→3

“I collected A LOT of money.”

Thus Aymara, like many nonconfigurationallanguages, allows discontinuous constituents. Forexample, in (22), the noun phrase sumaw chuq’ix“good potatoes” is discontinuous because akamarkanx “in this village” is not part of its surfacesyntax subtree.

3.4 Focalizers“Focalizers” (i.e. focusing operators, in Englishadverbs such as only, even, also, always, at least,etc., see (Hajicová et al., 1998)) are mostly ex-pressed by suffixes in Aymara. In many cases, theword with a focalizing suffix attracts the NPRED-marking, as in (25)–(28).

(25) Kimsa-ni-ki-wthree-hum-lim-npred

sar-igo-smpl3→3

“Only three went.”

(26) Wawa-pa-xchild-poss3-ref

may-ni-ki-waone-hum-lim-npred

“He has only one child.”

(27) Iki-ña-k-wsleep-inf-lim-npredmun-t-xawant-smpl1→3-ref“I only want to sleep.”

(28) NayI

kuna-wwhat-npred

sar-t-xago-smpl1→3-ref

“Even I went.”

Another pattern is attaching the independent ag-gregator -sa to the focalized word together withNPRED-marking of the verb, as in (29) and (30).

(29) Naya-sI-aggr

sara-rak-t-wago-‘also’-smpl1→3-npred

“I also went.”

(30) Juma-ki-syou-lim-aggr

yati-sma-waknow-pot2→3-npred

“At least you should know it.”

The focalizing negative particle jani “not” at-tracts the NPRED-marking in unmarked cases (but

102

see (10)). Two focalizers can be combined, as inthe case of jani and puni “always” which givesthe form janipuni “never” or jani and raki “also”which gives the form janiraki “neither”. Bothforms attract the NPRED-marking.

3.5 Resolving UnderspecificationAs was demonstrated in the above description,overt marking of information structure in Aymarais partial, as it leaves some elements underspec-ified. Since Aymara is a radical pro-drop lan-guage15 (any argument of the verb may be unex-pressed, not only the subject), predictable argu-ments are often omitted. As a consequence, directobjects are usually omitted if they can be inferredfrom the context, as in (31).16

(31) Tata-ja-taki-wfather-poss1-ben-npredala-sk-t-xabuy-progr-smpl1→3-ref“I am buying it for my father.”

It follows that if a verb is NPRED-marked andhas an overt object, the latter is nonpredictable,too, unless it is REF-marked, as in (32).

(32) Naya-wI-npred

um-xwater.acc-ref

wayu-ni-:-xacarry-dir-fut1→3-ref“It is me who will bring the water.”

The same holds for other verbal arguments.The only ambiguity arises from inherently referen-tial expressions which appear REF-marked or un-marked (e.g., qharürux/qharüru “tomorrow”) andthis marking does not correlate with their nonpre-dictability.17

3.6 Intersentential CohesionOvert marking of information structure, describedin the previous section, has crucial importancefor intersentential cohesion and can be helpful forcoreference resolution. Consider, for example, thefollowing two sentences:

15See (Cole, 1987).16The sentence is an answer to the question Who are you

buying it for?17It is possible that the category of referentiality is under-

going a reanalysis as a result of language contact. A moredetailed analysis of the speech of monolingual speakers isneeded.

(33) Mäone

marka-nvillage-loc

mäone

jisk’asmall

imilla-wgirl-NPRED

utj-iexist-smpl3→3

“There was a little girl in a village.”

(34) Jupa-xshe-ref

MariyaMaría

suti-ni-waname-poss-npred

“Her name was María.”

In (33), both noun phrases as well as the verbare nonpredictable. The noun imilla “girl” isNPRED-marked (i.e., marked as the most salientpart of the utterance) because in (34), it is referredto by the pronoun jupa “he/she”.

The analysis of utterances in texts such as sto-ries and narratives reveals that Aymara speakersconsequently take intersentential cohesion into ac-count when they decide where to place pragmaticmarkers.

4 Referent Identification

Surface and deep syntax as well as semantics oper-ate on isolated sentences. Now we will discuss theformalization of pragmatics, the level of discoursecontext.

For the purposes of this subsection we assumethat we have a discourse that consists of sentencess1, . . . , sn and that we have the corresponding fea-ture structures f1, . . . , fn. An entity we call a fea-ture structure that represents a person, an object oran event (an event may be dynamic if described bya verb or statal if described by a nominal predi-cate). Every entity has a special attribute, INDEX,to represent coreferences.

The discourse context is formally a list ofindices (values of the INDEX attribute) C =〈i1, . . . , im〉. The sentences are processed one byone. At the beginning, C = ∅. For every fi, we dothe following:

1. For every entity in fi, we try to find its refer-ent in C (we describe below how referents areidentified). If a referent was found for an en-tity, its index in C is moved to the beginningof the list. Otherwise, a new index is assignedto the entity and prepended to the list.

2. The index of the NPRED-marked entity ismoved (or prepended) to the beginning of C.

There are various strategies of identification ofreferents in the preceding discourse that can com-bine to resolve ambiguities.

103

4.1 Morphosyntactic Referent IdentificationAymara has a rich system of switch-reference suf-fixes that help to identify referents in the discoursecontext. For example, the sentence

(35) Tumasi-xiThomas-refklasi-n-ka-ska-:n-waclass-loc-vrbl-progr-past3→3-npref“Thomasi was in the class room.”

may be followed by the following sentence:

(36) Yatichiri-xjteacher-ref

manta-n-isin-xenter-dir-trgrss-ref

nuw-i-wahit-smpl3→3-npred“When the teacherj entered, hej hit himi.”

The first sentence adds the index of the en-tity Tumasixi “Thomas” to C. The second sen-tence adds the index of the entity Yatichirixj “theteacher” to C. Furthermore, there are two unre-solved (covert) pronouns that represent the actorand the patient of nuwiwa “to hit”. In this case theswitch-reference transgressive suffix -sina speci-fies that the actor of nuwiwa is the actor of man-tasinx “entered”, thus the actor of nuwiwa has theindex i. The patient is coindexed with the nextentity in C with which it agrees morphologically(e.g., in terms of animacy) or semantically. On theother hand, the sentence

(37) Yatichiri-xjteacher-ref

manta-n-ipan-xenter-dir-trgrds-ref

nuw-i-wahit-smpl3→3-npred“When the teacherj entered, hei hit himj .”

changes the indexes of the pronouns in the ma-trix sentence because the switch-reference trans-gressive suffix -ipana specifies that the subject ofmantanipanx is different from that of nuwiwa.

4.2 Semantic Referent IdentificationIf there is the sentence

(38) Tumasi-mpiThomas-com

Marya-mpjMary-com

uñj-t-wasee-smpl1→3-npred“I saw Thomasi and Maryj .”

followed by

(39) Jani-wnot-npred

usuri-:-ta-ppregnant-vrbl-part-poss3

yat-k-t-tiknow-incmpl-smpl1→3-neg“I did not now that shej was pregnant.”

the referent identification of the covert pronounwhich is the actor of usurïtap “that she was preg-nant” is not morphosyntactically restricted (Ay-mara has one pronoun, jupa, for both “he” and“she”) but it is semantically restricted. It is obvi-ous that the semantic information of this kind hasto come from the lexicon. Likewise, there will belexically encoded semantic gender for words suchas tayka “mother”, jilata “brother”, imilla “girl”,etc.

4.3 Pragmatic Referent IdentificationAs described in Subsection 3.6, explicitly NPRED-marked entities are more salient than other non-predictable entities in the same sentence. If thereis the sentence

(40) Tayka-ma-wi

mother-poss2-npredyatichiri-ma-mpjteacher-poss2-com

jik-i-simeet-smpl3→3-refl

“Your motheri met your teacherj .”

followed by

(41) Usuta-:-ta-m-xsick-vrbl-part-poss2-refyat-x-i-waknow-compl-smpl3→3-npred“Shei already knew that you were sick.”

the actor of yatxiwa “s/he already knew” is coin-dexed with taykamaw “your mother” because thisentity is NPRED-marked in the first sentence andtherefore more salient (it precedes other entities inC). If we move the NPRED-marker in the first sen-tence to the “teacher”, the meaning of the secondsentence will change:

(42) Tayka-ma-ximother-poss2-refyatichiri-ma-mpi-wj

teacher-poss2-com-npredjik-i-simeet-smpl3→3-refl“Your motheri met your teacherj .”

104

(43) Usuta-:-ta-m-xsick-vrbl-part-poss2-refyat-x-i-waknow-compl-smpl3→3-npred“Hej already knew that you were sick.”

5 Complex Predicates

A special case of pragmatic marking representthe so-called complex predicates, i.e. monoclausalpredicates composed of (at least) two predicativeelements. In such constructions, a (fully) semanticverb combines with a modal or auxiliary elementto express complex predication such as causative,volitive, supine, etc. The concept of complexpredicates, elaborated by Alsina (1996), has beenapplied to a number of phenomena and languagesincluding, for example, Hindi light verbs (Mo-hanan, 1994) or Turkish causatives (Çetinoglu etal., 2008). We will leave aside the rather compli-cated formal treatment of these constructions inunification-based grammars (Homola and Coler,2013) and focus on the linguistic description ofthese form in Aymara from the pragmatic point ofview here.

Sentences (16) and (17) are examples of com-plex predicates. The constructions contain onlyone NPRED-marker, i.e. they are monoclausal andshould be represented by a single feature structurewith a complex functor. For (16), we get:

(44)

FUNC ‘muna-’

ARGS

⟨subjact,

FUNC ‘ala-’ARGS 〈subjact,

dobjpat〉

pat

⟩

For (17), we get:

(45)

FUNC ‘sara-’

ARGS

⟨subjact,

FUNC ‘ala-’ARGS 〈subjact,

dobjpat〉

pat

⟩

In a more concise notation, (44) can be writtenas:

(46) muna(subjact, ala(subjact, dobjpat)pat)

Likewise, (45) can be written as:

(47) sara(subjact, ala(subjact, dobjpat)pat)

Unlike in languages with morphologicallyformed volitive verbal complexes, such as

Guaraní, in Aymara such constructions are formedsyntactically (i.e. the complex predicate value isnot created in the lexicon). As for motion verbs,such as (45), there is a morphological alternative:

(48) Jichha-xnow-REF

t’antbread.ACC

ala-ni-rapi-:ma-wabuy-DIR-BEN-FUT1→2-NPRED

“I will go to buy bread for you now.”

Further evidence that such predicates are mon-oclausal is the fact that polypersonal agreement isexpressed on the auxiliary or modal verb, not thefull one, as in (49) and (50).

(49) Ch’uqpotato.ACC

alja-ñ-wsell-INF-NPRED

mun-smawant-SMPL1→2

“I want to sell potatoes to you.”

(50) Tump-iri-wvisit-AG-NPRED

jut-smacome-SMPL1→2

“I came to visit you.”

It is noteworthy that a complex predicate with amotion verb can have two complements, namely alocative phrase and a verbal complement:

(51) Al-iri-wbuy-AG-NPRED

ChukiawLa.Paz.ACC

sara-:nago-SMPL3→3

“He went to La Paz to buy it.”

Even more evidence for monoclausality can befound in sentences like (52):

(52) ChukiawLa.Paz.ACC

sara-ñ-wgo-INF-NPRED

mun-t-xwant-SMPL1→3-REF

irnaqa-ña-taki-xawork-INF-BEN-REF

“I want to go to La Paz for work.”

In (52), the nominalized verb irnaqañatakixa“to work” depends on sarañw “to go” creating along-distance dependency. Without consideringthe verbal complex sarañw muntx a monoclausalpredicate, it would be linguistically counterintu-itive and computationally hard to parse the sen-tence.

It is obvious from the mentioned examplesthat the NPRED-marker tends to attach to the full

105

verb. This fact supports the hypothesis that modal(deictic) syntactic elements should be consideredsynsemantic, i.e. at the level of deep syntax theyshould be represented as attributes of the nodes oftheir heads rather then autonomous nodes or fea-ture structures. This approach, adopted by us, isconsistent with (Sgall et al., 1986).

We omit from the discussion morphologi-cally built complex predicates, such as causatives(e.g. (53)). In other languages where they are ex-pressed syntactically, they would be treated in thesame way as the constructions described above.

(53) Jacha-y-t-wacry-CAUS-SMPL1→3-NPRED

“I made him/her cry.”

(54)

FUNC ‘caus’

ARGS

⟨subjact,

[FUNC ‘jacha-’ARGS 〈dobjact〉

]pat

⟩

6 Conclusions

We have described the overt marking of informa-tion structure and its pivotal role both in intersen-tential cohesion in Aymara as well as coreferenceresolution. Through an analysis of morphologicalinformation structure marking with pragmatic suf-fixes in this language, we illustrate the irrelevanceof word order for information structure. Whileovert marking of referentiality is not always conse-quent, nonpredictability is marked with strict regu-larity. Having demonstrated that nonpredictabilitycan be marked both at clause level and inside a NPand that focalizing suffixes tend to attract NPRED

morphology, we explained how underspecificationis resolved, noting, for example, that the overt ob-ject of a NPRED-marked verb is also unpredictableunless it is REF-marked. We also treated the iden-tification of morphosyntactic, semantic and prag-matic referents through a formalization of prag-matics. The specifics of pragmatic marking incomplex predicates show how the NPRED-markertypically attaches to the full verb thus supportingthe hypothesis that modal syntactic constructionsshould be considered monoclausal.

ReferencesWillem Adelaar. 2007. The Languages of the Andes.

Cambridge University Press.

Alex Alsina. 1996. The Role of Argument Structure inGrammar: Evidence from Romance. CSLI Publica-tions.

Mark C. Baker. 1996. The Polysynthesis Parameter.Oxford University Press.

Georg Bossong. 1989. Morphemic marking of topicand focus. In Michel Kefer and Johan van der Auw-era, editors, Universals of Language, pages 27–51.

Georg Bossong. 2009. Divergence, convergence, con-tact. In Kurt Braunmüller and Juliane House, edi-tors, Convergence and divergence in language con-tact situations, pages 13–40.

Lucy Therina Briggs. 1976. Dialectal Variation in theAymara Language of Bolivia and Peru. Ph.D. thesis,University of Florida.

R. Cerrón-Palomino and J. Carvajal Carvajal. 2009.Aimara. In M. Crevels and P. Muysken, editors,Lenguas de Bolivia. Plural Editores, La Paz, Bolivia.

Özlem Çetinoglu, Miriam Butt, and Kemal Oflazer.2008. Mono/bi-clausality of Turkish Causatives. InProceedings of the 14th International Conference onTurkish Linguistics.

Peter Cole. 1987. Null Objects in Universal Grammar.Linguistic Inquiry, 18(4):597–612.

Eva Hajicová, Barbara H. Partee, and Petr Sgall. 1998.Topic-focus articulation, tripartite structures, andsemantic content. Kluwer Academic Publishers,Dordrecht.

Kenneth L. Hale. 1983. Warlpiri and the grammar ofnon-configurational languages. Natural Language& Linguistic Theory, 1:5–47.

Martha Hardman, Juana Vásquez, and Juan de DiosYapita. 2001. Aymara. Compendio de estructurafonológica y gramatical. Instituto de Lengua y Cul-tura Aymara.

Martha Hardman. 2000. Jaqaru. LINCOM EUROPA.

Petr Homola and Matt Coler. 2013. Causatives asComplex Predicates without the Restriction Opera-tor. In Miriam Butt and Tracy Holloway King, edi-tors, Proceedings of the LFG Conference.

Shoichi Iwasaki. 2013. Japanese. John BenjaminsPublishing.

Katalin É. Kiss. 1994. Discourse Configurational Lan-guages. Oxford University Press.

Joanna Mattissen. 2006. On the Ontology and Di-achrony of Polysynthesis. In Dieter Wunderlich,editor, Advances in the theory of the lexicon, pages287–354. Walter de Gruyter, Berlin.

Tara Mohanan. 1994. Argument Structure in Hindi.CSLI Publications.

Petr Sgall, Eva Hajicová, and Jarmila Panevová. 1986.The Meaning of the Sentence in Its Semantic andPragmatic Aspects. D. Reider Publishing Company.

106

Petr Sgall. 2009. Where to look for the fundamentalsof language. Linguistica Pragensia, 19(1):1–35.

Juan de Dios Yapita and J.T. Van der Noordaa. 2008.La dinámica aymara. Conjugación de verbos. Insti-tuto de Lengua y Cultura Aymara.

107