Abbreviations
We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/glossing-rules.php), with the following additions:
- 1
first singular
- 2
second singular
- 3
third (proximate) singular
- 4
fourth (obviative) singular
- 12
first person plural inclusive
- A
actor
- AN
animate
- ANA
anaphoric
- ASP
aspect
- CISL
cislocative
- CLM
clause linkage marker
- CNJ
Conjunct order
- CON
conjectural mode
- DEIC
deictic
- DIR
directional
- DITR
ditransitive
- ECS
extra-core slot
- END
endophoric
- EQ
event quantification
- EPEN
epenthetic
- EVID
evidentiality
- EXO
exophoric
- FAI
animate intransitive final stem
- FII
inanimate intransitive final stem
- FTA
animate transitive final stem
- FTI
inanimate transitive final stem
- I
inanimate singular
- IF
illocutionary force
- II
inanimate plural
- IMPER
impersonal construction
- IN
initial stem
- INAN
inanimate
- INFR
inferential or dubitative mode
- LS
logical structure
- MOD
modality
- N
nominal stem
- NMR
non-macrorole
- OBV
obviative
- PoDP
post-detached position
- PRET
preterite or narrative mode
- PrCS
pre-core slot
- PrDP
pre-detached position
- PSA
privileged syntactic argument
- REP
reportative or attributive mode
- RP
reference phrase
- TEMP
temporal
- TNS
tense
- TRNS
translocative
- U
undergoer
- VAI
animate intransitive verb
- VII
inanimate intransitive verb
- VTA
animate transitive verb
- VTI
transitive inanimate verb
- X
unspecified actor
22.1 Introduction
This chapter examines the grammar of Cheyenne or Tsėhésenėstsestȯtse (Plains Algonquian, USA) within the framework of Role and Reference Grammar (henceforth RRG) (Van Valin and LaPolla Reference Van Valin and LaPolla1997; Van Valin Reference Van Valin2005).Footnote * Although this grammatical sketch aims to cover as much ground as possible, for reasons of brevity, it only covers in some depth the interaction of syntax, semantics, and pragmatics in simple sentences. Thus, by way of illustrating the pervasive influence of pragmatics in all areas of Cheyenne grammar, the last part of the paper includes an account of the relation between word order and information structure, and the influence of pragmatics on morphosyntactic coding and macrorole assignment.
The organization of the chapter is as follows: after a brief account of the principal morphosyntactic features of Cheyenne, Section 22.2 offers a description of a number of grammatical areas which benefit from analysis in RRG terms. Section 22.3 provides an account of the relationship between information structure and two linguistic mechanisms which are particularly typical of Algonquian languages, namely the proximate/obviation distinction and the direct/inverse system. These serve to mediate between macrorole assignment and morphosyntactic coding and, consequently, compensate for the lack of a rigid word order and a proper case-marking system in Cheyenne.
22.2 Grammatical Sketch
After introducing a number of key features of Cheyenne grammar (22.2.1), in this section we discuss basic clause types (22.2.2), verb valence and transitivity (22.2.3), the marking of core arguments, argument-adjuncts and adjuncts (22.2.4, 22.2.5) and the linking algorithm (22.2.6).
22.2.1 Basic Morphosyntactic Properties of Cheyenne
Cheyenne is a polysynthetic language which, like other Algonquian languages, is believed to exhibit a pragmatically influenced word order (Leman Reference Leman1999: 37), which is concerned with the way in which the message is formulated, rather than with the identification of semantic roles or syntactic functions. In accordance with the concept of ‘newsworthiness’ (Mithun Reference Mithun and Tomlin1987), which appears to imply importance and unpredictability, the language tends to place any elements expressing relevant information preverbally and any elements expressing trivial information postverbally.
Owing to the pronominal-argument and head-marking nature of Cheyenne, verbs have an extremely complex templatic structure, which not only includes bound forms covering the grammatical information of the core arguments of the predicate, but also derivational stems and particles expressing grammatical categories like aspect, tense, modality, etc., which are treated as operators modifying the different layers of the clause, as shown in Figure 22.1 (modified from Corral Esteban Reference Corral Esteban, Nolan and Diedrichsen2017: 310).

Figure 22.1 Verb structure in Cheyenne
The relative ordering of the elements within the verbal complex is as follows. The person cross-referenced on the verbal prefix of a main clause always refers to the most pragmatically salient participant, in accordance with the Person Hierarchy 2 > 1 > 3 > 4 > I(inanimate).1
The preverbal operators between the prefix and the verb stem express the realis/irrealis distinction, past or future tense, epistemic modality, event quantification, negation, deontic modality, direction,2 and the different aspectual distinctions.
Predicates can be simple or complex, depending on whether they comprise one single stem or more than one stem. The morphemes constituting a complex predicate, which incorporate adverbial, nominal and verbal meanings into the verb, are commonly referred to as initial, medial and final, in accordance with their relative position within the verbal complex (Bloomfield Reference Bloomfield, Osgood and Hoijer1946).
The theme marker indicates voice or, more accurately in Algonquian terminology, whether the construction is direct or inverse. In broad terms, the direct construction is used when the subject of the transitive clause outranks the object in the Person Hierarchy, and the inverse is used when the object outranks the subject.
Following the theme marker – and sometimes merged with it in a portmanteau – a number of suffixes that function as pronominal affixes appear, as they cross-reference the core arguments of the predicate and cover grammatical information about them in terms of person, number, animacy, and proximate/obviative status.
The two outermost operators, the clausal operators of evidentiality and illocutionary force, follow the pronominal suffixes. To summarize, in Cheyenne operators realized as bound morphemes occur on both sides of the nucleus.
(1)
(tns - neg - mod - asp - dir - V - neg - evid - if) É - s - sáa - tónėše – éne - e’e-óo’ó’t(a)-ȯ-hé-nȯse-Ø 3-pst-neg-be.able.to-stop-up.in-look.fti-3>I.3sg.a.Isg.u-neg-evid.3sg.a.Isg.u.rep-decl ho’honáevose. mountain ‘He is said to have not been able to stop looking up the mountain.’
Example (1) and its representation in Figure 22.2 indicate that, in Cheyenne, the order of the morphemes expressing operators with respect to the nucleus reflects their relative scope and, consequently, confirms the validity of the principle governing this ordering (Van Valin and LaPolla Reference Van Valin and LaPolla1997: 49–52), since core operators such as internal negation,3 and (deontic) modality occur closer to the verb than two clausal operators like tense and evidentiality, but are positioned further from the verb than are nuclear operators, such as aspect or directionals.
Likewise, example (1) shows that the verb in Cheyenne exhibits pronominal affixes, in the form of a prefix such as é- ‘3’, and suffixes, namely ‑ȯ- and ‑nȯse, which cross-reference the different arguments in the clause. Thus, all the grammatical relations are coded in the verb, rather than in the reference phrases (RPs), which occur freely in the clause (Corral Esteban Reference Corral Esteban2014: 381).

Figure 22.2 Layered structure of the clause with constituent and operator projections
The person cross-referenced by the prefix refers to the most pragmatically salient direct core argument in terms of the Person Hierarchy 2 > 1 > 3 > 4 > I, which gives preference to local participants over animate non-local participants, and to animate non-local participants over inanimate entities. In intransitive constructions, the prefix always cross-references the only argument of the verb. However, in transitive constructions, a choice is made regarding which of the arguments of the verb is more pragmatically salient in terms of the aforementioned ranking. Furthermore, in transitive predications, the Person Hierarchy works closely with a Semantic Function Hierarchy based on animacy and topicality, whereby agent outranks recipient and benefactive, and recipient and benefactive outrank patient and theme. A harmonic alignment between the two hierarchies is said to reflect a direct construction, whereas a disharmonic alignment between them is said to exhibit an inverse construction.
(2)
Né-méhót-o. → Direct construction 2-love.vta-2>3.2sg.a.3sg.u4 ‘You love her.’ love′ (2sg, 3sg.f)
(3)
Né-méhót-a. → Inverse construction 2-love.vta-3>2.3sg.a.2sg.u ‘She loves you.’ love′ (3sg.f, 2sg)
In these two transitive constructions, the verb has two semantic arguments – a second-person singular argument and a third-person singular argument. As second person outranks third person in the Person Hierarchy, both constructions select the prefix né- ‘2’, regardless of the semantic function of the argument in each construction. However, despite having the same prefix, each construction exhibits different relations between the verb and its arguments, as indicated by the suffixes ‑o and ‑a, which reflect a direct and an inverse construction respectively. In (2), the person cross-referenced by the prefix né- indicating a second-person participant, corresponds to the higher semantic macrorole, namely the actor, so the Person and Semantic Function hierarchies are in proper alignment, and the construction is then marked as direct. In (3), by contrast, the hierarchies are not properly aligned, as the same person prefix né- corresponds to the lower semantic macrorole, that is to say, the undergoer, so the construction is marked as inverse.
22.2.2 Basic Clause Types
Cheyenne, like the other Algonquian languages, exhibits three major divisions of verb forms, commonly named ‘orders’ following Bloomfield (Reference Bloomfield, Osgood and Hoijer1946: 97–103): Independent, Conjunct and Imperative. The Independent order includes all verb forms other than imperatives that can stand alone, and the Conjunct order is used for all dependent verb forms. Hence, while the former generally occurs in main clauses (4), the latter appears in subordinate clauses (5–9).
(4)
Hetane é-mane-Ø ma’xemené-mahpe. man 3-drink.vti-3>I.3sg.a.Isg.u apple-water ‘That man drank the apple juice.’ [do′ (hetane, [drink′ (hetane, ma’xemené-mahpe)]) & ingr consumed′ (ma’xemené-mahpe)]
(5)
[…] tsé-x-heše-manė-se ma’xemené-mahpe. cnj-pst-thus-drink.vti-3>I.3sg.a.Isg.u.ind apple-water ‘[…] that he drank apple juice.’
(6)
[…] tsé-’-éše-manė-se ma’xemené-mahpe. cnj-pst-already-drink.vti-3>I.3sg.a.Isg.u.ind apple-water ‘[…] after he drank apple juice.’
(7)
[…] tsé-x-homá’xe-manė-se ma’xemené-mahpe. cnj-pst-because-drink.vti-3>I.3sg.a.Isg.u.ind apple-water ‘[…] because he drank apple juice.’
(8)
[…] vé’-manė-stse ma’xemené-mahpe. cnj-drink.vti-3>I.3sg.a.Isg.u.sbjv apple-water ‘[…] if he drank apple juice.’
(9)
[…] momȯxe-manė-stse ma’xemené-mahpe. cnj-drink.vti-3>I.3sg.a.Isg.u.sbjv apple-water ‘[…] I wish he drank apple juice.’
The main difference between the two verbal orders lies in the function of their prefix: whereas in the Independent order the prefix serves to express the most pragmatically salient person, in the Conjunct order it indicates the grammatical mood, namely realis (e.g. indicative tsé- (5–7)) or irrealis (e.g. conditional vé’- (8), optative momȯxe- (9), etc.). The prefix in a dependent clause functions as a clause linkage marker and is usually accompanied by a preverbal particle that identifies the type of subordinate clause (e.g. ‑heše- in complement clauses (5), ‑éše- in temporal adverbial clauses (6), ‑homá’xe- in adverbial clauses of reason (7), etc.), which can lead to an instance of either daughter subordination (5) or ad-subordination (6–8).
The RRG notion of clause structure is based on two fundamental contrasts: the contrast between predicating and non-predicating elements, and the one between those non-predicating elements that function as arguments and those that do not. Regarding the first contrast, Cheyenne sentences have two major types of predicate, namely verbal (10) and non-verbal, which can be nominal (11 and 12) or adjectival (13).5
(10)
Néhe hetane é-ónėstán-a he’nétoo’o. deic.end.prox.an man 3-open.vti-3>I.1sg.a.Isg.u door ‘This man opened the door.’ [do′ (néhe hetane, Ø)] cause [become opened′ (he’nétoo’o)]
(11)
Tá’tóhe ma’háéso é-ma’heónė-hetane-(é)ve-Ø. deic.exo.dist.an old.man 3-holy-man.n-be.fai-3sg.a ‘That old man is a holy man.’ be′ (tá’tóhe ma’háéso, [ma′heónėhetane′ ])
(12)
Muhammad Ali é-cassiusclay-h-eve-Ø. Muhammad Ali 3-Cassius.Clay.in-epen-fai-3sg.u ‘Muhammad Ali is Cassius Clay.’ equate′ (Muhammad Ali, Cassius Clay)
(13)
Na-né’ame é-pėhév-ȧhe-o’o 1.poss-parents.pl 3-good.in-fai-3pl.a ‘My parents are good.’ be′ (na-né’ame, [good′ ])
Unlike a verbal predicate, such as ónėstán ‘open’ in (10), the formation of a non-verbal predicate requires an initial nominal or adjectival stem and a final verb stem, which functions as a copula. This is evidenced by ma’heónėhetane ‘holy man’ and ‑eve ‘be’ in the identificational construction in (11), Cassius Clay and ‑eve in the equational construction in (12), and pėhév- ‘good’ and ‑ȧhe ‘be’ in the attributive construction (13).6
The final stem, by contrast, does not occur in specificational predications, which are expressed in Cheyenne through a different syntactic construction involving the presence of a relative clause.
(14)
Naa’háanéhe hetane tsé-ama’én-o deic.ana.dist.an man clm-drive.vti-3>I.3.a-I.u heóve-amȧho’hestȯtse. yellow-car ‘That man is the schoolbus driver.’ (lit. ‘That man is the one who drives the yellow car.’) be′(naa’háanéhe hetane, [drive′ (naa’háanéhe hetane, heóve-amȧho’hestȯtse)])
The post-copular element in a specificational construction is realized syntactically through a relative clause, such as tsé-ama’én-o heóve-amȧho’hestȯtse ‘who drives the yellow bus’ in (14). This relative clause includes a lexical verb, so it is able to express more easily the second participant of a specificational predication, as this is the more referential element of the construction, unlike in a predicational sentence where the second participant is less referential or not referential at all.
22.2.3 Verb Valence and Transitivity
Verbs in Cheyenne are divided into four types according to the criteria of syntactic transitivity, based on the number of direct core arguments, and animacy: inanimate subject + intransitive verb (II), animate subject + intransitive verb (AI), animate subject + transitive verb + inanimate object (TI), and animate subject + transitive verb + animate object (TA).7
The number of macroroles of a verb corresponds closely to its characterization according to the notion of syntactic transitivity in Cheyenne, as the number of direct core arguments is generally the same as that of macroroles, provided the arguments refer to a particular entity. Thus, single macrorole verbs are usually intransitive:
(15)
Peter he-mėšemo é-máse-am(e)-ėhn(e)-óho. Peter 3.poss-grandfather.obv 4-happily-asp.in-walk.fai-4sg.a ‘Peter’s grandfather is walking happily.’ do′ (Peter he-mėšemo, [walk′ (Peter he-mėšemo)])
The activity verb ēhne ‘walk’ occurs in an intransitive construction including an animate participant that is lexically realized by the obviative RP argument Peter he-mėšemo ‘Peter’s grandfather’. This argument appears in the logical structure (LS) of the verb and is cross-referenced on the verb through the prefix é- ‘4’ and the pronominal affix óho, which refers to a fourth person / obviative singular animate patient. As this single direct core argument of each verb is referential, the verb can be considered to have one macrorole.
An exception to such a correlation between the number of direct core arguments and macroroles in intransitive constructions is provided in Cheyenne, like in other languages, by ‘weather’ verbs.
(16)
É-ho’ééto-Ø. I-snow.vii-Isg.a ‘It is snowing. (lit. ‘It snows.’) snow′(Ø)
The direct core argument of an II verb such as ho’ééto ‘snow’ is cross-referenced on the verb by the prefix é- ‘I’ and a null pronominal suffix if their argument is singular, or the suffix ‑nėstse if it is plural. This is not actually a semantic argument of the predicate (Van Valin Reference Van Valin2008: 8). Therefore, even though these verbs are classified as intransitive in syntactic terms, semantically speaking they are atransitive.
There are other intransitive constructions showing a mismatch between the syntactic valency of a verb and its semantic transitivity, owing to the influence of referentiality on Cheyenne.
(17)
É-h-ma’xe-mané-stove taa’é-va. 3-tns-much.drink.vai-imper night-temp ‘There was a lot of drinking at night.’ do′ (Ø, [drink′ (Ø, Ø)])
This impersonal construction in (17) exhibits an intransitive verb stem: mane ‘drink’. This, along with the presence of the prefix é- ‘3’, implies the presence of an animate argument. However, the occurrence of an impersonal suffix such as ‑htove/stove/nove in this construction does not allow the addition of a pronominal affix cross-referencing the single argument of the verb, which prevents the prefix from being referential. Thus, the only direct core argument of the verb – only coded in the prefix – is not a macrorole in an impersonal construction, and the verb must be considered atransitive, which shows non-identity between S-transitivity and M-transitivity.
The fact that Cheyenne only cross-references referential arguments can also be evidenced in intransitive verbs corresponding to a two-place predicate whose second argument is only implied:
(18)
Ná-mané-me. 1-drink.vai-1pl.a ‘We drink.’ do′ (1pl, [drink′ (1pl, Ø)])
The activity verb mane ‘drink’ in (18) has one direct core argument that is coded on both the prefix ná- ‘1’ and the pronominal affix ‑me and cross-references a first-person plural animate argument, namely the first argument in the LS. However, no reference is made to the second semantic argument of the verb; hence the verb mane ‘drink’ in (18) can be said to take one macrorole, which is assigned to the only direct core argument of the verb.
Arguments with generic reference are generally cross-referenced on the verb in Cheyenne.
(19)
Vé’hó’e é-ohkė-šénetam-ovo xamaevo’ėstan(e)-oho. white.people 3-eq-loathe.vta-3>4.3pl.a.4sg.u Indian.obv ‘White people usually find Indians repulsive.’ (Fisher et al. Reference Fisher, Leman, Pine Sr and Sanchez2006: 257) loathe′ (vé’hó’e, xamaevo’ėstanoho)
In this example, both the prefix é- ‘3’ and the pronominal affix ‑ovo cross-reference two arguments – a third-person/proximate plural animate argument and a fourth-person/obviative singular argument. However, despite being cross-referenced on the verb, the two arguments are not macroroles since they are non-specific.
There is, however, a context where an argument with a generic reference is not coded on the verb in Cheyenne. This occurs when there is a close semantic relationship between the verb and the argument, which would be a typical object of the verb. In such cases, the object RP is incorporated into the verb.
(20)
a. É-o’ene-mēna-Ø. 3-pick.in-berry.fai-3sg.a ‘He is berry picking.’ (Fisher et al. Reference Fisher, Leman, Pine Sr and Sanchez2006: 215) do′ (3sgM, [pick′ (3sgM, o’ene)]) b. É-o’en-anȯtse menȯtse. (3) -pick.vti-3>I.3sg.a.Ipl.u berry.pl ‘He is picking the berries.’ do′ (3sgM, [pick′ (3sgM, menȯtse)]) & ingr picked′ (menȯtse)
The difference between activities and their telic version, that is to say active accomplishments, is observed syntactically in Cheyenne. Unlike the active accomplishment transitive verb o’en ‘pick’ in (20b), the activity intransitive verb o’enemēna ‘berry-pick’ in (20a) is formed by the initial stem of the verb ‘pick’, o’ene, and a final stem of the noun ‘berry’, mene. Also, while o’en ‘pick’ has two direct core arguments, cross-referenced by the prefix é- ‘3’ and the suffix ‑anȯtse, which indexes a third-person singular animate agent and a plural inanimate patient, the bipartite verb stem o’ene-mēna ‘berry-pick’ has only one direct core argument – a third-person singular animate agent, which is coded by the prefix and the null suffix. Additionally, as only the first argument of the activity verb in (20a) is specific, the verb only takes one macrorole. Conversely, the active accomplishment verb in (20b) has two specific arguments and, therefore, takes two macroroles.
Except for these examples showing the influence of referentiality, the number of direct core arguments and macroroles also tends to coincide in Cheyenne transitive constructions provided the arguments are referential. Consequently, two macrorole verbs are usually transitive.
(21)
Ná-ono’átam-oo’o tsé-héstoo’éšeeh-ae-se. 1-respect.vta-1>3.1sg.a.3pl.u cnj-raise.vta-3>1.3pl.a-1sg.u.ptcp ‘I respect my parents.’ (lit. ‘I respect who raised me.’) respect′ (1sg, tsé-héstoo’éšeehaese)
The two-place verb ono’átam ‘respect’ has two arguments, namely a first-person singular animate experiencer and a third-person plural animate theme, which are cross-referenced on the verb by the prefix ná- ‘1’, the theme marker ‑o- and the pronominal affix ‑o’o. The second argument is also realized by a full lexical RP, namely tsé-héstoo’ éšeehaese ‘my parents’ (lit. ‘those who raised me’). As both direct core arguments are referential, the verb can be said to have two macroroles.
Finally, unlike other Algonquian languages (e.g. Blackfoot, Plains Cree or Ojibwa), Cheyenne ditransitive verbs have a very complex system of suffixes (Corral Esteban Reference Corral Esteban2014: 419–422), which vary with respect to the animacy and number of the theme argument. Thus, verbs such as the stems mét ‘give’, véestomev ‘ask for’, nomáhtseh ‘steal’, véstomev ‘promise’, or vóo’seh ‘show’ behave like true ditransitive verbs, since they cross-reference the three core arguments morphologically.8
(22)
Né-nomáhtsėh-aenóvo mo’éhno’hāme. 2-steal.ditr-3>2>4.3pl.a.2sg.u.4.nmr horse.obv ‘They stole the horse from you.’ [do′ (3pl, Ø)] cause [become not have′ (2sg, mo’éhno’hāme) & become have′ (3pl, mo’éhno’hāme)].
The predicate nomáhtsėh ‘steal’ in (22) shows an instance of causative accomplishment where an activity is carried out by a third-person plural animate actor, cross-referenced by the suffix ‑aenóvo. This activity leads to a change of possession whereby the former participant manages to appropriate something, which is represented by the fourth-person/obviative participant mo’éhno’hāme ‘horse’, which previously belonged to the addressee. These three arguments are coded on the prefix né- ‘2’ and the pronominal suffix ‑aenóvo.
The influence of referentiality on the cross-reference of core arguments in Cheyenne (Corral Esteban Reference Corral Esteban2014: 409–410) can also be observed in ditransitive verbs not including a reference to some of their semantic arguments. In such cases, the arguments are not cross-referenced on the verb morphology, which means that both the syntactic transitivity of the verb and the number of macroroles is reduced:
(23)
Ná-mét-onóneo’o nėstámane 1-give.ditr-1>3>I.1pl.a.3pl.u.Isg.nmr 12.poss.food na-nésonėhaneo′o. our.children ‘We give our children our food.’ [do′ (1pl, Ø)] cause [become have′ (nanésonėhaneo’o, nėstámane)]
(24)
Ná-méa(’tov)-noneo′o na-nésonėhaneo′o. 1-give.vta-1>3.1pl.a.3pl.u 12.poss.children ‘We give away our children to somebody.’ [do′ (1pl, Ø)] cause [become have′ (Ø, nanésonėhaneo’o)]
(25)
Ná-mé(a’e)-ánóne nėstámane. 1-give.vti-1>I.1pl.a.Isg.u 12.poss.food ‘We give our food to somebody.’ [do′ (1pl, Ø)] cause [become have′ (Ø, nėstámane)]
(26)
Ná-méa(‘e)-ēme. 1-give.vai-1pl.a ‘We give something to somebody.’ [do′ (1pl, Ø)] cause [become have′ (Ø, Ø)]
Although the three-place predicate ‘give’ requires three semantic arguments, in Cheyenne these are only coded on the verb when they refer to specific participants. The pronominal affixes ‑o-nó-ne-o’o in (23) cross-reference the three semantic arguments of the logical structure, namely a first-person plural agent, a third-person plural recipient nanésonėhaneo’o ‘our children’, and an inanimate singular theme nȧhtámáne ‘our food’. However, the pronominal affixes ‑no-ne-o’o in (24) code only a first-person plural agent and a third-person plural theme nanésonėhaneo’o ‘our children’; the pronominal affixes ‑á-nó-ne in (25) code only a first-person plural agent and the inanimate theme nėstámane ‘our food’; and finally, the pronominal affix ‑ēme in (26) only cross-references the first-person plural agent.
The role of referentiality is not only noticed on the pronominal affixes, but also in the verb stem, which adopts different forms depending on the number of macroroles the verb takes. Thus, in Cheyenne, the predicate ‘give’ can be expressed through a ditransitive stem mét, a TA stem méa’tov, a TI stem méa’e, or an AI stem méa’e (23–26). As regards the correlation between the number of direct core arguments and macroroles, in the intransitive verb in (26) the only direct core argument of the verb is a macrorole. The same correlation holds in (24) and (25), as the two transitive verbs have two direct core arguments and both are macroroles. Since only two macroroles are allowed in RRG, in the ditransitive construction in (23) only two of the direct core arguments are macroroles – the first-person plural agent and the recipient nanésonėhaneo’o ‘our children’, the other – the theme nėstámane ‘our food’ – being the non-macrorole core argument. They are all identified through cross-referencing.9
To summarize, except for cases such as constructions including ‘weather’ verbs10 and impersonal constructions, and the problem posed by generic reference, which can be coded on the verb via either pronominal affixes or noun incorporation, M-transitivity closely corresponds to S-transitivity in Cheyenne, due to the effect of referentiality on the coding of arguments and, most importantly, on the form of the syntactic core template: the presence of a non-referential argument in a transitive construction affects not only the pronominal affixes but also the form of the verb stem, since the non-referential argument is not cross-referenced on the verb and is, therefore, not represented within the core. Finally, the Cheyenne examples also show that the number of macroroles that a verb takes is less than or equal to the number of semantic arguments in its logical structure.11
A particularly challenging case is a typical Algonquian transitive construction, commonly referred to as the ‘unspecified actor’ construction (Hockett Reference Hockett, Nichols and Ogg1996), as it includes no reference to the agent of the action denoted by the verb.
(27)
Ná-éno’eéh-án-e na-vénótse. 1-heal.vta-X>1–1sg.u 1.poss-tepee.loc ‘I was healed at home.’ (lit. ‘Someone healed me in my tepee.’) be-at′(home, [do′(Ø, Ø)] cause [become healed′ (1sg, Ø)]])
The two-place predicate éno’eéh ‘heal’ occurs in a transitive construction including an unspecified argument that corresponds to the first argument in the LS. Bearing in mind that non-referential participants are not coded on the verb in Cheyenne, it is not surprising that the unspecified agent in this construction is not cross-referenced on the verb. This assumption appears to be supported by the fact that, while the verb stem is transitive, the suffixal morphology resembles that of the AI paradigm. Furthermore, the presence of a lexicalized agent in this construction would entail changing the form of the suffixal morphology. However, we might come to a different conclusion if we take into account that this construction uses a different theme marker (e.g. ‑án or ‑Ø) depending on whether the specified argument is local or non-local. A possible explanation for this could be that the theme markers in this construction indicate direct and inverse morphology, which means that these bound markers cross-reference an unspecified participant.12 In summary, it is difficult to gauge whether this construction reveals an identity between S-transitivity and M-transitivity or not.
Finally, the transitivity of the verb shows an important difference between the unspecified actor construction and the impersonal construction illustrated by example (17), for, unlike the impersonal construction, which is built on an intransitive verb stem and eliminates any cross-reference to the single argument on the verb suffix, the unspecified actor construction involves a transitive verb but AI morphology, which implies that at least one argument is coded on the verb.
22.2.4 Argument Coding on the Verb
Only two different types of argument can be distinguished in Cheyenne: direct core arguments and argument-adjuncts, as core arguments can be direct but not oblique in this language. Direct core arguments must be referential to be cross-referenced on the verb. However, if they are third-person RPs, they can be morphologically marked or unmarked, since Cheyenne uses a reference-tracking system to establish a clausal disjoint reference between multiple third-person participants. Thus, in situations involving more than one third-person RP, the most pragmatically salient RP (i.e. the proximate) is left unmarked, and any other less pragmatically salient RP is marked with an obviative marker.
The proximate/obviative system cannot be considered to be a type of case marking, as the notions of proximate and obviative are not associated with any semantic role in particular, as is illustrated by the following constructions involving two third-person referents.
(28)
Náhe ka’ėškóne é-ohke-véstȧhém-óho deic.end.dist.an boy 3-eq-help.vta-3>4.3sg.a.4sg.u néhe ma’háhkėse-ho. deic.end.prox.an old.man-obv ‘That boy always helps this old man.’ do′ (náhe ka’ėškóne, [help′ (náhe ka’ėškóne, néhe ma’háhkėseho)])
(29)
Néhe ma’háhkése é-ohke-véstȧhém-áá’e deic.end.dist.an old.man 3-eq-help.vta-4>3.4sg.a.3sg.u náhe ka’ėškóne-ho. deic.end.prox.an boy-obv ‘That boy always helps this old man.’ / ‘This old man is always helped by that boy.’ do′ (náhe ka’ėškóneho, [help′ (náhe ka’ėškóneho, néhe ma’háhkése)])
Despite the fact that these sentences include the same arguments and are semantically equivalent, the arguments show a different proximate/obviative status. Thus, whereas in (28) the agent is the proximate náhe ka’ėškóne ‘that boy’, and the benefactive is the obviative néhe ma’háhkėseho ‘this old man’, in (29), there is a shift with respect to the proximate marking. Now, the agent náhe ka’ėškóneho ‘that boy’ is obviative and the benefactive néhe ma’háhkése is proximate.
A representation of direct core arguments (28) is provided in Figure 22.3 for a subsequent comparison with that of incorporated arguments in applicative constructions in Figure 22.4 and argument-adjuncts and adjuncts in Figure 22.5.

Figure 22.4 Representation of an applicative construction with a derived two-place verb

Figure 22.5 Representation of a clause including an argument-adjunct and an adjunct
As there is no correlation between the proximate/obviative distinction and semantic roles or syntactic functions, it is necessary to take into account the information provided by the nominal morphology and the prefix and pronominal suffixes on the verb. Thus, the difference between these two examples of transitive predication lies in that, while in the direct construction in (28) the proximate argument náhe ka’ėškóne ‘that boy’ is the actor and the obviative argument néhe ma’háhkėseho ‘this old man’ is the benefactive, the opposite occurs in the inverse construction in (29), as the obviative argument náhe ka’ėškóneho ‘that boy’ is now the actor and the proximate argument néhe ma’háhkése ‘this old man’ is the benefactive.
Cheyenne has a number of applicative constructions where the addition of a relative root in the form of an initial stem increases the syntactic valence of the verb and, consequently, allows it to take one more direct core argument.
(30)
Nȧ-htse-vést-ȯhomó’hem-óho vého he-stóna-ho 1-fut-with.in-dance.fta-1>4.1sg.a.4sg.a chief 3.poss-daughter-obv tsé-h-móheeohtsé-stove-tse. cnj-pst-meet.vai.imper-ptcp ‘I will dance with the chief’s daughter at the meeting.’ do′ (1sg, [dance′ (1sg)]) ∧ do′ (vého hestónaho, [dance′ (vého hestónaho)])
While the predicate ‘dance’ requires only one semantic argument, the complex verb véstȯhomó’hem ‘dance with’ in (30), formed by an adpositional initial vést- ‘with’ and a transitive verb stem ȯhomó’hem ‘dance’, cross-references two arguments, namely a first-person singular agent and a third-person singular comitative vého hestónaho ‘the chief’s daughter’, through the prefix ná- ‘1’ and the pronominal affix ‑o, which also includes a reference to the argument vého hestónaho ‘the chief’s daughter’. Thus, although it does not refer to a participant semantically required by the predicate ‘dance’, the RP vého hestónaho ‘the chief’s daughter’ is treated as an argument of the complex verb véstȯhomó’hem ‘dance with’, yielding a derived two-place verb.
Likewise, unlike argument-adjuncts (see Section 22.2.5), this RP is cross-referenced on the verb, and, unlike adjuncts, it is not marked as an oblique.13 The fact that the cross-reference of the RP vého hestónaho ‘the chief’s daughter’ is subject to the addition of the adpositional initial vést- ‘with’ makes this argument the closest equivalent to an oblique core argument. The only difference is the fact that the adpositional initial is not determined by the verb. Rather, it is predicative and, consequently, it contributes its own predicate to the clause. It is also of note that the possessor RP in the genitive phrase is represented like an element branching from the RP node that agrees in person with the prefix designating the possessor, namely he- ‘his/her’, which behaves as a core argument and is, consequently, represented as a daughter of the nominal core.
22.2.5 Adjuncts, Argument-Adjuncts and Adpositional Marking
Adjuncts in Cheyenne can appear in very different guises, for example: verbal premodifiers (e.g. ‑ohke- ‘usually’ in (31)), nominal phrases (e.g. méó’ne ‘on the road’ in (31)), free particles (e.g. hó’ótova ‘sometimes’ in (31)), or dependent clauses (e.g. tséhmóheeohtséstovetse ‘at the meeting’ in (30)). They function as optional modifiers because they are never cross-referenced on the verb or incorporated with an adpositional initial into the verb. Furthermore, they may receive oblique (e.g. locative (31), temporal (17) or instrumental (see example in note 13)) case marking through the nominal suffixes ‑va, ‑e, or ‑o, when they are realized through nominal phrases.
(31)
Hó’ótóva ná-ohke-ameváen-one méó(’o)-n-e. sometimes (1)-eq-pass.vta-1>3.1pl.a.3sg.u road-epen-loc ‘Sometimes we pass him on the road.’ (Fisher et al. Reference Fisher, Leman, Pine Sr and Sanchez2006: 11) [do′ (1pl, Ø)] cause [ingr overtaken′ (3sgM)]
The locative adjunct méóne ‘on the road’ is formed by the noun méó’o ‘road’ plus the oblique nominal suffix ‑e, which denotes location. This locative adjunct is neither cross-referenced on the verb nor incorporated into the verbal complex as a stem-forming element.
Despite their reference to one obligatory argument of a verb, argument-adjuncts in Cheyenne are not cross-referenced on the verb, which makes them resemble adjuncts syntactically. Cheyenne also does not distinguish adjuncts from argument-adjuncts in terms of marking, as the same oblique suffixes, ‑va, ‑e or ‑o, are added to both adjuncts and argument-adjuncts regardless of their syntactic function.
(32)
Ná-ést-o’tse-nȯtse mȯxe’ėstóonȯtse hoo’hénó-va 1-in.in-put.fti-1>I.1sg.a.Ipl.u books bags-loc éše-ē-va. day-epen-temp ‘I put the books in the bags yesterday.’ [do′ (1sg, Ø)] cause [ingr be.in′ (hoo’hénóva, mȯxe’ėstóonȯtse)]
The verb ‘put’ in (32) requires three obligatory arguments – an agent, a patient, and a locative. However, táho’tsé ‘put’ only has two direct core arguments, the agent and the patient, as can be evidenced from the TI stem and the verbal suffix ‑nȯtse, which codes only two arguments – a first-person singular animate agent and an inanimate plural patient – but does not reflect the presence of any marker representing the location. The location is expressed by an RP, hoo’heno ‘bags’, which is marked as oblique by the nominal suffix ‑va,14 but is not coded on the verb, which results in its being marked as a peripheral oblique. However, unlike adjuncts, such as éšeēva ‘yesterday’, argument-adjuncts appear to be complemented by a locative adpositional initial particle such as ést- ‘in’ in (32). This is an adpositional nucleus incorporated into the nucleus of the core functioning as an applicative marker that licenses the third core argument, the argument-adjunct hoo’hénóva ‘the bags’. The stem-forming element reflects the presence of a locative argument and it is this adpositional element, rather than the locative RP, that is obligatory in the sentence. The locative particle is an important component of the meaning of the sentence, and its form is not determined by the verb – the particle ést- ‘in’ can be replaced by ‑táh(o’k)- ‘on’, ‑áhto’- ‘under’, etc. This incorporated particle must therefore be considered to be an adpositional nucleus that functions as an applicative marker.
Finally, this construction also shows that not all three-place predicates are realized through a ditransitive construction in Cheyenne.
22.2.6 The Linking Algorithm in Cheyenne
The study of the interaction between the various components of grammatical structure generally shows language-specific variation. Bearing this in mind, for an examination of systematic relationships in the semantic content of a predicate, the syntactic behaviour it exhibits, and the pragmatic background that surrounds the act of communication, a bidirectional analysis of the linking system based on example (33) is presented in Figures 22.6 and 22.7, illustrating how the syntax–semantics–pragmatics interface works in Cheyenne.
(33)
Na-né’ame ná-met-aenóvo mȯxe’ėstoo’o. 1.poss.parents 1-give.ditr-3>1>I.3pl.a.1sg.u.Isg.nmr book ‘My parents gave me a book.’

Figure 22.6 Linking from semantics to syntax in Cheyenne
ACS = accessible; ACV = activated

Figure 22.7 Linking from syntax to semantics in Cheyenne
This linking procedure from semantics to syntax starts with the construction of the semantic representation of the sentence, which is drawn from the logical structure of the predicate.
As is the case with other languages, the selection of the actor and undergoer arguments in monotransitive constructions presents no difficulty in Cheyenne, as the highest-ranking argument in the logical structure is selected as the actor and the lowest-ranking argument is selected as the undergoer. Ditransitive constructions indicate that Cheyenne represents an instance of what Dryer (Reference Dryer1986) called a ‘primary object language’, that is to say a language where the only pattern that occurs with three-argument predicates corresponds to the marked selection for undergoer.15 Thus, when a verb has three arguments, the second-highest, rather than the lowest-ranking argument in the logical structure is selected as the undergoer macrorole and, consequently, the lowest-ranking argument is the non-macrorole direct argument. In example (33), therefore, the leftmost argument in the logical structure of na-né’ame ‘my parents’ is selected as the actor, and the second leftmost argument – the participant referring to the speaker – is chosen as the undergoer instead of the expected rightmost argument mȯxe’ėstoo’o ‘book’, which now becomes the non-macrorole argument.
The determination of the morphosyntactic coding of the arguments shows an interesting specific feature of Algonquian languages, as the choice of a specific argument on the prefix shows the existence of a privileged syntactic argument (PSA) in these languages. This PSA involves a pragmatically determined neutralization of semantic roles for syntactic purposes, as the choice of the prefix is pragmatically determined – it is only determined by the Person Hierarchy – and can correspond to different semantic roles depending on the ranking of the participants on the hierarchy (e.g. agent (4), patient (19), experiencer (1), recipient (33), etc.). In a direct transitive construction, the PSA corresponds to the actor, and in an inverse transitive construction the PSA matches the undergoer. Thus, in example (33), bearing in mind that it is an instance of inverse construction, the PSA is the undergoer cross-referenced on the prefix ná- ‘1’.
As regards the syntactic representation of the sentence, the syntactic template for a three-place predicate like mét ‘give’ must contain a core including three argument positions, which will be filled by the pronominal arguments cross-referencing the core arguments, namely the prefix ná- ‘1’ and the portmanteau verbal suffix ‑aenóvo, which cross-references the three semantic arguments of the verb.
Finally, the linking of arguments to positions in the syntactic template is also an interesting feature of Cheyenne. The fact that word order in this language is pragmatically determined allows for greater flexibility when it comes to selecting the syntactic template of the sentence because of the positions (e.g. pre-core slot (PrCS) or any extra-core slot (ECS)) to which the independent RPs are assigned.
The direction of the linking process from syntax to semantics involves the interpretation of the overt morphosyntactic form of a sentence and the deduction of the semantic functions of the elements in the sentence deriving from such an interpretation, as shown in Figure 22.7.
The information provided by the verbal affix ‑aenóvo reflects the fact that, in this inverse construction, the verb is ditransitive and has three direct core arguments – a third-person plural agent, a first-person singular recipient, and an inanimate singular theme. The prefix also enables us to identify which constituent acts as the PSA, since, according to the ranking of the Person Hierarchy, which favours first person over third and inanimate person, ná- ‘1’ acts as the PSA of the construction. As regards the identification of the semantic macroroles of the core arguments, a ditransitive verb like mét ‘give’ has three possible candidates for only two macroroles. However, taking into account the accurate grammatical information provided by the bound markers ná- ‘1’ and ‑aenóvo, and considering that Cheyenne invariably shows the marked undergoer selection in ditransitive constructions, we can then select the proximate participant na-né’ame ‘my parents’ as actor (‘x’) and the first-person participant as undergoer (‘y’), whereas the third and last core argument, namely the inanimate participant mȯxe’ėstoo’o ‘book’, becomes the non-macrorole argument (‘z’).
The linking of the core arguments to the corresponding slots in the semantic representation of the sentence may be problematic in Cheyenne due to the fact that the number of pronominal affixes in the verb does not necessarily correspond to the number of direct core arguments. This is mainly because the verbal suffix is generally a portmanteau that cross-references all the direct core arguments of a verb, and the prefix only codes one of these core arguments, which does not allow for a one-to-one correspondence between direct core arguments and semantic arguments. Thus, it is necessary to bear in mind what type of stem the verb presents and what kind of grammatical information is provided by each of the pronominal affixes in order to gauge its number of direct core arguments. It would therefore seem logical to assume that the portmanteau cross-references all the arguments and the prefix cross-references one of them again, as represented in Figure 22.7.
The linking algorithm shows the important role played by pragmatics in the interaction between semantics and syntax in Cheyenne and reveals a series of grammatical properties: the variable word order exhibited by RPs, the pragmatically determined PSA, the cross-referencing of arguments through a portmanteau, and even the lack of correlation between the proximate/obviative status of referents and their semantic function in non-local contexts.
22.3 The Role of Pragmatics in Focus Structure, PSA Selection and Macrorole Assignment
The lack of a proper case-marking system and the use of a pragmatically influenced word order, which is not concerned with the identification of grammatical relations or semantic roles and, even, not completely determined by the notion of presupposition and assertion, make the process of the transfer of information between syntax and semantics and vice versa more difficult than in other languages. Thus, the goal of this section is to discuss the role of pragmatics in clause structure, PSA selection and macrorole assignment by discussing the interaction between word order, the reference-tracking system of obviation, and the direct/inverse system.
22.3.1 Word Order and Focus Structure
As discussed in Section 22.2.1, word order in Cheyenne encodes pragmatic factors rather than argument relations. Therefore, the language exhibits a wide range of variation in word order, which appears to be sensitive to the notion of newsworthiness. Specifically, the information is distributed in descending order of importance and unpredictability in the sentence. Importantly, the fact that word order in Cheyenne is not as strict as in other languages, English for example, does not necessarily mean that it is completely free, since it may be constrained by focus structure.
Before explaining the fundamentals of Cheyenne word order from the perspective of RRG, it is necessary to examine the way in which the arrangement of discourse functions fits into the RRG model of sentence structure.
The arrangement of the pragmatic categories in a specific linear order has a functional explanation. The leftmost category (see Figure 22.8), which includes topic particles and adverbials, equates to the pre-detached position (PrDP) in syntax, whose position is set off from the clause-internal elements by a pause. The next category, a focused constituent occupying the pre-core slot (PrCS) within the clause, corresponds to elements introducing discourse topics expressing a non-contrastive change of discourse topic. Alternatively, the same position can host an element introducing a contrast, regardless of whether this element has been previously mentioned in the discourse (in which case it is a contrastive topic) or is newly asserted (contrastive focus).16 Unless it is considered newsworthy, the verbal complex follows the focused element, forming the core of the clause and serving as the point of reference for the placement of the other clausal elements within the template. Both new information and easily inferable, already evoked or predictable, information, which serves to recover background information, occupy the postverbal position. Finally, the rightmost element, called a familiar topic in Figure 22.8, can also be used for an afterthought, which is included in order to reiterate its referent by a way of summary or to help to maintain the reference.17 This is also separated from the previous clause-internal constituents by a pause, meaning that it is located in the post-detached position (PoDP).

Figure 22.8 Information structure-based sentence template
The availability of two different positions for informational focus can be explained with reference to the concept of newsworthiness. Indeed, the positioning of informational focus before or after the verbal complex appears to be related to the significance given by the speaker to this information. However, the fact that word order in Cheyenne is governed by the principle of newsworthiness and that this notion is not entirely comparable to the pragmatic notions of assertion and presupposition that underlie the discourse-pragmatic statuses of focus and topic in RRG, makes it difficult to integrate the information structure of Cheyenne into the layered structure of the clause. Nevertheless, the tendency towards verb-initial word order,18 and the identification of the preverbal position (PrCS) as the most pragmatically marked position within the clause (it is the position occupied by the elements considered more informationally prominent in terms of newsworthiness) are suggestive of a correlation between word order and the different focus types proposed in RRG.
Cheyenne appears to use a verb-initial order for predicate focus, the universally unmarked type of focus structure.
(34)
Q: É-tóneto’omenehe-Ø(-he) Richard? 3-what.happen.vai-3sg.u(-if) Richard ‘What happens to Richard?’ A: É-sáa-mé’ov-ó-he-ho neše 3-neg-find.vta-3>4.3sg.a.4sg.u-neg-4sg.u two mo’éhno’hāme (Richard). horse.obv Richard ‘Richard/He does not find two horses.’
The pragmatic presupposition in this type of focus structure includes knowledge of a certain topic, and the assertion expresses a comment about the topic. The most common sentence type in Cheyenne in this context is a comment–topic structure where the predicate mé’ov ‘find’ and the patient neše mo’éhno’hāme ‘two horses’ constitute the focus (marked with small caps), and the agent Richard corresponds to the topic. It is important to remember, however, that it is very difficult to find examples of constructions containing given or familiar topics occupying a postverbal position, so that leaving the topical referent Richard unspecified appears to be more common, unless this previously mentioned information deserves to be reiterated in order to specify or clarify the reference. Figure 22.9 represents the focus structure of (34).

Figure 22.9 Unmarked predicate-focus structure in Cheyenne
The actual focus domain (the heavy solid line in Figures 22.9 to 22.11) in the unmarked type of predicate-focus structure comprises the core verbal constituent, if the verb is intransitive, and, additionally, the postverbal ECS, when there is (are) some overt RP argument(s). Furthermore, as expected, prosodic prominence falls on the focal constituents é-sáaméevóheho and neše mo’éhno’hāme ‘two horses’ in decreasing order. If the sentence included the familiar topic Richard, it would be represented as a clause-external topic at the PoDP. In that case, this topical constituent would be out of the potential focus domain (the broken line in Figures 22.9 to 22.11) and those that follow), which is coextensive with the clause in Cheyenne.

Figure 22.10 Unmarked sentence-focus structure in Cheyenne

Figure 22.11 Unmarked narrow-focus structure in Cheyenne
While it is not so common, the positioning of a topic such as Richard in the pragmatically marked preverbal position – where it would be activated as the discourse topic – would illustrate an instance of a marked SV(O) word order for this focus type. It is of note that, due to its newsworthy nature, this constituent would be given greater prominence than the verbal complex.
The second type of focus structure is sentence focus, where no pragmatic presupposition is evoked, and all information is new and, therefore, in focus.
(35)
Q: É-ta-tónėsóotse-Ø éše-ē-va? I-trans-what.happen.vii-Isg.u day-epen-temp ‘What happened yesterday?’ A: É-onénėše-otse-Ø na-amȧho’héhame. I-broken.in-become.fii-Isg.u 1.poss-car ‘My car broke down.’
The most common word order pattern in this situation, which places the verb in clause-initial position and the RP in postverbal position, represents the unmarked structure of this focus type because there is no element occupying the PrCS slot. Figure 22.10 represents the focus structure of (35).
As there is no presupposition in this structure, the assertion extends over the entire proposition, as is indicated by the actual focus domain in Figure 22.10. Both the core and the subject RP in the ECS are associated with special intonational realization, although its intensity decreases as it approaches the clause-final position.
Placing the argument RP na-amȧho’héhame ‘my car’ in the pragmatically marked preverbal position would also be acceptable in this focus structure, but the word order would represent a case of marked sentence focus. In this case, the argument RP na-amȧho’héhame ‘my car’ would be highlighted by the speaker, possibly due to the fact that it represents the new discourse topic.19
The difference between predicate-focus and sentence-focus structures is therefore both syntactic and prosodic. Unlike sentence focus sentences, predicate focus sentences (34) have at least one topical element, which is preferably only cross-referenced by the pronominal affixes on the verb, but can also occur in the PoDP slot when overtly realized.20 In the latter case, this topical element is not stressed, unlike the postverbal RP in the sentence focus construction (35).
The third type of focus structure is narrow focus, where the focus domain is limited to a single constituent.
(36)
Q: É-onénėše-otsé-nėse ne-no’ka’éesó’hestȯtse. I-broken.in-become.fai-Isg.u.infr 2.poss-motorbike ‘I heard / It is said that your motorbike broke down.’ A: na-amȧho’héhame é-onénėše-otse-Ø. 1.poss-car I-broken.in-become.fii-Isg.u ‘My car broke down.’
The narrow-focus construction generally includes an element in the PrCS, for example na-amȧho’héhame ‘my car’ in (36). Unlike English, Cheyenne tends to mark narrow-focus constituents both syntactically and prosodically, with the narrow-focus constituent generally occurring in preverbal position – as the information it provides is normally considered by the speaker to be essential to the hearer – where it receives special prosodic prominence. Figure 22.11 represents the focus structure of (36).
The constituent under narrow focus is the only one that is prosodically stressed; hence, the actual focus domain does not generally extend into the core in Cheyenne. Thus, in Cheyenne, the unmarked narrow-focus position for single constituents with focal properties is the PrCS. However, when such focal material occurs in other positions, the marked narrow-focus structure is called for. This situation is rather infrequent, as the narrow-focus constituent provides the most important information and, consequently, the information it provides is generally considered newsworthy, but it might still occur in constructions including a transitive verb and two overt argument RPs where either the verb or either of the RPs could be considered by the speaker to convey more newsworthy information. In that case, the actual focus domain would move elsewhere within the clause. Thus, narrow focus on a newsworthy element is a case of unmarked narrow focus, while narrow focus on a non-newsworthy element is a case of marked narrow focus.
To summarize, while in Cheyenne the influence of the discourse-pragmatic concept of newsworthiness on the syntactic arrangement of constituents within a clause is paramount, the word order is not free, as it also appears to encode discourse-pragmatic functions such as focus and topic. Cheyenne selects verb-initial order as its unmarked word order in the predicate- and sentence-focus structures. The default interpretation of elements in the PrCS in these constructions is that of marked focus, for example contrastive narrow focus (but note that contrastive topics can also occur there, see Shimojo Reference Shimojo, Guerrero, Cerda and Belloro2009, Reference Shimojo2010, Reference Shimojo and Nakamura2011 and Chapter 11 on information structure). Finally, prosodic prominence is always associated with the placement of focal elements in the PrCS or, in the absence of an RP occupying such a slot, in the core-initial position.21 Thus, assuming that in predicate- and sentence-focus structures the unmarked focus position for RPs in Cheyenne is postverbal and the marked focus position for RPs is preverbal – more specifically in the PrCS – the potential focus domain extends over the entire clause in all types of focus structure, but the actual focus domain is: (1) from the core to the end of the clause in unmarked predicate- and sentence-focus structures; or (2) just the PrCS in the unmarked narrow-focus type.
Finally, while the notion of newsworthiness fails to correspond to the dichotomies topic vs. focus or even given vs. new information (Leman Reference Leman1999), a correlation between newsworthy information and focus emerges from the analysis, as evidenced by the fact that, excluding the variable positioning of informational focus, the preferred constituent order in Cheyenne is one where focal material (informationally prominent or contrastive) precedes the verbal complex, and the topic material (known or presupposed) appears after the verbal complex.22
22.3.2 The Proximate/Obviative System
Now that the link between word order and information packaging in Cheyenne has been established, this section turns to an analysis of the proximate/obviative system. According to Aissen (Reference Aissen1997: 709), this marking serves as a reference-tracking system whereby the language is able to differentiate multiple non-local participants in such a way that, in a discourse span involving two or more third-person animate arguments,23 only one of them is unmarked and placed in the foreground (the proximate) and all others are morphologically marked and relegated to the background (the obviative).
When a discourse span begins, the most pragmatically salient participant, the proximate, is usually promoted to a preverbal position, and the obviative, which refers to the other participants, tends to occupy a postverbal position. Taking into account the fact that the preverbal position is generally reserved for newsworthy information and receives special prosodic prominence in Cheyenne, it might be logical to assume a certain correlation between the notions of proximate and focus, on the one hand, and those of obviative and topic, on the other. This correspondence is not absolute, however, as it is possible to place an obviative argument in clause-initial position even when this position has inherently focal properties, as is the case with an obviative interrogative pronoun, for example. As discussed in Section 22.2.4, there is no correlation either between the distinction between proximate and obviative marking, which differentiates two non-local participants in terms of discourse salience, and semantic roles or syntactic functions (cf. 28–29).
While it is not completely clear what factor determines the choice between proximate and obviative, discourse salience must be one such factor, in the sense that the proximate is the most prominent participant of the text or the topic of discourse. The first thing a speaker must do before formulating a message is decide which of the third-person referents is the central participant in the text. Subsequently, they will leave this proximate argument, which tends to appear in preverbal position, unmarked, and mark all other third-person animate participants, which generally occupy postverbal positions, as obviative.
Indeed, it is very common to find long stretches of narrative text where the same referent, considered by the speaker as the most important participant, is left unmarked as proximate.
(37)
1. Kȧsovááhe é-h-ne’-éva-ame-vé’otsė-hoo’o. young.man 3-pst-cisl-back-continue-go.on.warpath.vai-3sg.a.pret ‘A young man returned from scouting. 2. É-š-kȧhane-otsé-hoo’o. 3-pst-tired.in.become.fai-3sg.u.pret He was tired. 3. É-x-ho’ėhahtse-Ø tósa’e nėhéóhe o’hé’e. 3-pst-make.fire.vti-3>I.3sg.a.Isg.u somewhere there river.loc He made a fire somewhere there at a river. 4. Tsé-’-ēe-hoo’ė-se nėhéóhe tóxė(ha)-ho’ėsta-va cnj-pst-around-stay.vai-3sg.u.ind there beside-river.loc As he was sitting there beside the fire, É-x-ho’ėhó’t-aehoono nevá’ėsesto. 3-pst-come.to.vta-4>3.4sg.a.3sg.u.pret someone.obv someone came up to him. 5. “Kȧsovááhe, né-ta-néhovan-ȧhtse-ma”, young.man 2-let-wrestle.vta-recp-12 é-x-het-aehoono. 3-pst-say.vta-4>3.4sg.a.3sg.u.pret “Young man, let’s wrestle!” he said to him. 6. É-h-néhovan-ahtsé-sesto. 3-pst-wrestle.vta-recp-3pl.rep They are said to have wrestled. 7. É-s-tsėheta’é(e’tov)’oe-sesto tsé-’-a’enó’ne-otse-Ø 3-pst-face.vta-4>3.4sg.a.3sg.u.rep cnj-pst-dark.in.become.fii-Isg.a.ind He is said to have been pushing him towards the dark. 8. Hápó’e, é-s-tsėheta’é(e)’(t)ov-ósesto tsé-h-vōho’ėho’ta-tse. likewise 3-pst-face.vta-3>4.3sg.a.4sg.u.rep cnj-pst-light.vii-Isg.ind Likewise, he is said to have pushed him toward the light. 9. É-’- anȧ’ham-osesto. 3-pst-throw.down.vta-3>4.3sg.a.4sg.u.rep He is said to have thrown him down. 10. “Né-hō’tah-e”, é-x-het-aesesto […]. 2-beat.vta-2>1.2sg.a.1sg.u 3-pst-say.vta-4>3.4sg.a.3sg.u.rep “You beat me”, he is said to have said to him […].’
The fact that the proximate/obviative distinction serves to track topic continuity across a discourse span can be observed in (37), an extract from an oral narrative. The main discourse topic, kȧsovááhe ‘a young man’, is established from the very outset, as it is placed in preverbal position, a pragmatically important position in Cheyenne. Once a referent has been chosen as the main character in the story, it becomes the proximate argument, and all other third-person participants receive obviative marking – which can be observed in the nominal suffix (e.g. nevá’ėsesto [obv] ‘someone’) and in the verbal suffix, (e.g. é-xhetaehoono ‘he [obv] said to him [prox]’). Thus, for example, the argument nevá’ėsesto ‘someone’ (s. 4) is introduced as new information in sentence 4, but, very possibly owing to its indefinite character, is not considered by the speaker to have pragmatic salience. Thus, it is marked as obviative and placed in postverbal position (note that the proximate counterpart would be nevá’ėsėstse [prox] ‘someone’). This implies that there is no shift in the discourse-pragmatic status of these two referents and, consequently, that the proximate status of the argument kȧsovááhe ‘a young man’ remains unaltered throughout the stretch of discourse under examination (the obviative form would be kȧsováaheho [obv] ‘a young man’), as can be observed in sentences 2 and 3, and in the dependent clause in sentence 4.
However, it is also possible to find texts where the proximate referent changes very frequently, even from clause to clause.
(38)
1. Vé’ho’e naa xaa-vo’ėstane é-’-ēe-néše-ohtsé-sesto […]. white.man and Indian 3-pst-around-continue-go.vai-3pl.a.rep ‘A white man and an Indian were going together. 2. […] Naa tsé’tóhe xaa-vo’ėstane and deic.prox.exo.an Indian é-’-eše-aahtse’-tótoéše-na-sėtse. 3-pst-lie-already-lie.with.eyes.open.vai-?-3sg.u.rep […] this Indian is said to have been already lying with his eyes open. 3. Naa vé’hoé é-’-osee-hóhta’ȧhané-tano-sėstse. and white.man 3-pst-very-tell.story.vai-want.fai-3sg.a.rep And the white man is said to have really wanted to tell his story. 4. É-h-nė-hetó-sesto tsé’tóhe xaa-vo’ėstan(e)-óho, 3-pst-ana-tell.vta-3>4.3sg.a.4sg.u deic.prox.exo.an Indian-obv He told this Indian: 5. “ná-ta-ovóe-hósésta na-ováxestȯtse!” 1-let-first-tell.vti.3>I.3sg.a.Isg.u 1.poss-dream é-x-het-ósesto. 3-pst-tell.vta-3>4.3sg.4.sg.u.rep “Let me first tell my dream!” he is said to have told him. 6. Naa tsé’tóhe xaa-vo’ėstane and deic.prox.exo.an Indian é-h-pėhév-átsėstá-nȯse. 3-pst-good.in-regard.fti-3>I.3sg.a.Isg.u and this Indian thought well of that. 7. Naa néhe vé’ho’e é-’-asėst-(h)óhta’hane-sėstse. and deic.end.prox.an white.man 3-pst-start-tell.story.vai-3sg.a.rep And that white man started to tell his story. 8. […] Naa nėhéóhe tsé-s-ta-éšė-ho’óhta’ȧhanė-se, and there cnj-pst-trns-already-arrive.in.storytelling.vai-3sg.a.ind […] And when he got to that point in the story, tse’tóhe xaa-vo’ėstane é-h-nė-het-ósesto this Indian 3-pst-ana-tell.vta-3>4.3sg.a.4sg.u.rep this Indian said to tsé’tóhe vé’hó’e, […].’ deic.end.prox.an white.man.obv this white man […].’
In this different story, shifts of proximate/obviative status occur very frequently because the speaker selects two discourse topics, rather than only one as in the previous story. The speaker considers the two referents, vé’ho’e ‘white man’ and xaa-vo’ėstane ‘Indian’, as equally important. This is supported by the first sentence of the story, where both referents appear in preverbal position forming a complex RP. From this moment onwards, whenever the speaker refers separately to either of them in the story, the referent being referred to receives the proximate status and occupies the preverbal position, for example tsé’tóhe xaa-vo’ėstane ‘this Indian’ in sentence 2 and vé’hoé ‘the white man’ in sentence 3. This leads to a rapid series of changes of discourse topic and of the discourse-status of the arguments, so that xaa-vo’ėstane ‘Indian’ is the proximate argument in sentences 2, 6 and 8, whereas vé’ho’e ‘white man’ is the proximate argument in sentences 3 and 7. Each of these two referents also receives obviative marking in one sentence, that is to say xaa-vo’ėstan(e)-óho [obv] ‘Indian’ in sentence 4 and vé’ho’e [obv] ‘white man’ in sentence 8. The fact that they are not considered pragmatically important in these two sentences is supported by their obviative marking – reflected both in the nominal suffixes and the verbal suffixes – and their placement in the postverbal position.
The analysis of the proximate/obviative distinction in these two texts shows that the main topic of discourse in Cheyenne tends to be proximate, and that the proximate/obviative status of referents remains unaltered until a new discourse topic (most commonly, a new focal element or an old topical element that is reintroduced into the discourse) is established, generally bringing about a shift in the discourse span and leading to a new proximate/obviative assignment.
However, as the proximate and obviative distinction is not invariably tied to macrorole assignment, it cannot by itself help us to determine which macrorole corresponds to each argument. This can be observed, for example, in sentences like 5, 7, 8, 9 and 10 in (37) or 5 in (38), where verbs cross-reference two third-person animate arguments with non-overt RPs, whose interpretation may confuse the hearer.
22.3.3 The Direct/Inverse System
As seen in Section 22.2.1, the Person Hierarchy and the Semantic Function Hierarchy work together as a single unified system in order to establish the link between syntactic and semantic information, as shown in Figure 22.12 (see also Wolvengrey Reference Wolvengrey2011: 57–63).

Figure 22.12 Interaction between the Person and Semantic Function hierarchies
The interaction between the two hierarchies establishes a correspondence between actors and local participants, on the one hand, and undergoers and non-local participants, on the other, in such a way that a harmonic alignment between the two hierarchies leads to a direct construction, and a disharmonic alignment between the two scales results in an inverse construction.
Both the Person Hierarchy and the Semantic Function Hierarchy rank their elements in terms of their inherent pragmatic salience or prominence. However, each hierarchy appears to be based on a different notion of salience or prominence. While the Semantic Function hierarchy is sensitive to the degree of animacy of the participants,24 the Person Hierarchy may be governed by several pragmatic factors, such as animacy or topicality, as it ranks local over non-local participants and animate arguments over inanimate arguments.25
As discussed in Section 22.2.1, the person represented by the prefix in an intransitive sentence always corresponds to the only direct core argument of the verb, so that the suffixal morphology is only concerned with the distinction between singular and plural number (see examples 11 and 13) or obviative marking (see 15). However, in transitive constructions, where more than one participant is involved, the prefix only cross-references the most pragmatically salient argument in terms of the Person Hierarchy.
(39)
Né-vóom-ȧtse. 2-see.vta-1>2.1.sg.a.2sg.u ‘I saw you.’
(40)
Né-vóom-e. 2-see.vta-2>1.2sg.a.1sg.u ‘You saw me.’
In these examples the prefix né- ‘2’ is selected as the prefix heading the verbal complex, thereby acquiring a special syntactic status, namely the PSA of the construction. Neither the prefix né- ‘2’ nor the pronominal suffix ‑ȧtse or ‑e can determine the semantic roles of the arguments of the verb by themselves. The prefix cross-references only the higher-ranking person, but is not indicative of semantic role on its own – it cross-references the undergoer in (39) and the actor in (40) – and the portmanteau formed by the theme marker plus the pronominal affixes tends to provide information about the direction of the construction, and to cross-reference the direct core arguments of the verb, but is sometimes ambiguous, especially in the Conjunct order. An examination of both elements, therefore, becomes necessary in order to link arguments and semantic roles. Thus, the prefix in both examples is a second-person participant; the portmanteaus -ȧtse and ‑e indicate that (39) and (40) are an inverse and a direct construction respectively, and that these constructions include a first-person singular actor and a first-person singular undergoer, respectively.
The direct/inverse mechanism is especially useful when the point of view changes for pragmatic reasons, and the core arguments are not lexicalized as RPs.
(41)
Ná-mó’ot-óneo’o Heévȧhetaneo’o 1-invite.vta-1>3.1pl.a.3pl.u Oklahoma.Cheyennes ‘We invited Oklahoma Cheyennes.’ naa ná-vés(e)-e’hanám-aeneo’o. and 1-with.in-eat.fta-3>1.3pl.a.1pl.u ‘… and they ate with us.’ (Leman Reference Leman1980a: 71)
In this example, the pronominal affixes ‑ae-ne-o’o, which provide unambiguous grammatical information about each of the core arguments of the verb, help us to identify that the clause ná-vés(e)e’hanámaeneo’o is an inverse construction, as it involves a third-person plural animate participant acting on a first-person plural animate participant.
To summarize, as the morphology of Cheyenne has no case distinctions, the only way to distinguish actor from undergoer in a transitive construction – see examples 5, 7, 8, 9 and 10 in (37) or 5 in (38) – is through the joint analysis of the direct/inverse system, the Person and Semantic hierarchies and the proximate/obviative status. The direct/inverse system provides a successful link between the arguments of a predicate and their semantic roles by encoding the interaction between these hierarchies. Thus, this efficient direct/inverse system, supported by the proximate/obviative distinction in non-local environments and without recourse to a fixed syntactically oriented rigid word order, plays a similar role to case marking in other languages in order to link semantic arguments with specific semantic functions in Cheyenne.
Finally, once macroroles have been assigned to semantic arguments, the speaker must select an appropriate word-order pattern following the principle of newsworthiness. In accordance with this pragmatic principle, the preverbal position in Cheyenne clauses is considered newsworthy, as it is here that discourse-prominent material is placed, and, by contrast, less discourse-salient information is placed in postverbal position.
22.4 Conclusion
The goal of this grammatical sketch has been to explore sentence structure in Cheyenne from the perspective of RRG through the analysis of a range of grammatical issues serving both to reveal the core components shared by all languages and to highlight those that are specific to Cheyenne grammar. The analysis of Cheyenne grammar in Section 22.2 showed that there is no evidence for the postulation of grammatical relations in addition to semantic predicate–argument relations, save for a pragmatically influenced PSA that is represented by the verbal prefix. It also indicates the hierarchical scope order of operators in Cheyenne, which is broadly defined in accordance with the RRG proposal. Finally, it reveals the fundamental role played by pragmatics, especially in terms of argument coding, macrorole assignment and word order.
Section 22.3 argued that, despite the word-order variability displayed by Cheyenne, it is possible to integrate information structure into clause structure and explore the intricate mechanism used by this language to accommodate semantic information into syntactic structure. Cheyenne links arguments and particular semantic roles through an intricate mechanism consisting of two components, which work in combination with the proximate/obviative distinction in contexts involving non-local participants. These components are illustrated by: (1) the correlation between the Person Hierarchy and a Semantic Function Hierarchy and (2) a binary system of verbal direction expressed by a theme marker or direction-marking morpheme on the verb, which is complemented by the grammatical information about the participants provided by the pronominal affixes. The direct/inverse system works along with the reference-tracking system of obviation, interrelates the Person and Semantic Function Hierarchies, and acts as a mediator between all grammatical components. It is thus responsible for the interplay between syntax, semantics and pragmatics. The efficiency shown by this system in coding the morphosyntactic properties of arguments from the semantic information and in linking morphosyntactic coding to macrorole assignment accounts for the existence of the ‘relatively free’ word order in Cheyenne and the absence of a true case-marking system.
All in all, these findings support the assumption of RRG that syntax is underpinned by semantic and pragmatic factors. Furthermore, the analysis of Cheyenne grammar highlights the validity of RRG as a framework for the analysis of highly polysynthetic languages.
Abbreviations
We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/glossing-rules.php) with the following additions and modifications:
- ASP
aspect
- CLM
clause linkage marker
- DEIC
deictic
- DIR
directional
- DL
dual
- ELEV
elevational
- FR.DIST
far distal
- HAB
habitual
- INACT
inactivated
- MOD
modality
- N
noun
- NINCORP
incorporated noun
- NFN
non-finite
- NR.DIST
near distal
- NUC
nucleus
- NUM
number
- PAST
past tense
- PC
paucal
- PERF
perfective
- PN
proper noun
- PRES
present tense
- POT
potential mode
- PRO
bound pronoun
- RED
?reduplicative?
- RM
remote
- RX
referential expression
- SEQ
sequence
- TNS
tense
- V
verb
- VAL
valence
- I-X
noun class
23.1 Yimas as a Polysynthetic Language
Yimas is a morphologically highly complex polysynthetic language spoken in the Sepik basin region of the northern swampy lowlands of Papua New Guinea and is one of six languages in the Lower Sepik sub-family of the Lower Sepik-Ramu family (see Foley (Reference Foley1986, Reference Foley, Pawley, Attenborough, Golson and Hide2005, Reference Foley and Palmer2017a) for comparative studies of this family). It has a typical profile for a polysynthetic language: it is heavily head marking, and, while the morphological structure of words is very extensively elaborated, the syntactic structure of phrases and clauses is only weakly so. In fact, Yimas almost entirely lacks the familiar syntactic category of phrases, and at the clause level it is highly non-configurational (Hale Reference Hale1983). The one notable difference of Yimas from most other polysynthetic languages is an extensive system of noun classes, eleven major ones and half a dozen minor ones. For a detailed exposition of Yimas grammar, see Foley (Reference Foley1991).
23.2 Nouns, Verbs and Agreement
As a polysynthetic head-marking language, the morphology of agreement is much in evidence in the language. The language has two major word classes, nouns and verbs. Words corresponding in meaning to adjectives in English or other familiar languages are divided between these two classes, with most belonging to the verb class, though there are three true adjectives. Many grammatical categories distinguish the two major classes of nouns and verbs, but a quite noticeable contrast is in their formal patterns of inflection: nouns (and the adjectives) are inflected for noun class and number by a set of suffixes, while verbs mark the noun class and number of their core argument participants for the most part with prefixes, as in this example.
(1)
yura antmaŋkl kpa-ŋkl kla-n-am-(n)tut dog.iii.sg cockatoo.vi.dl big-vi.dl vi.dl.nom-iii.sg.erg-eat-rm.past ‘The dog ate two big (sulphur crested) cockatoos.’
The marking of noun class and number on nouns is largely fusional, so it is difficult to separate the root from the inflection, and generally both noun class and number are portmanteau as well (especially in the singular), and there are morphophonemic rules at work too complex to go into here. However, the nominal agreement suffixes are isolatable when they appear on adjectives which are in concord with nouns they modify, as with ‑ŋkl vi.dl on the adjective root kpa ‘big’ in (1). Verbs have overt affixes that mark noun class and number of their core arguments, and these work under a very complex split ergative-accusative and direct-inverse alignment that will be explained below, so they are labelled in (1) simply as nominative and ergative. The distinction between the two major classes, noun and verb, is very rigid in Yimas; there is no overlap or flexibility so salient in Austronesian languages. A verb can only be used as a noun when it is overtly derived by a morphological process of nominalization (this will be the focus of the second half of this chapter), while nouns can never be used as verbs. There are no processes of denominal verbalization in the language, nothing equivalent to English I hammered the nail down into the floor, where the noun hammer denoting an instrument is used as verb denoting doing an action using that instrument; the equivalent Yimas expression would have to be something like ‘I hit the nail with a hammer piercing (the floor), going down.’ Verbs can only be used as arguments when they are derived into nouns by nominalization. Unmarked predicates are always verbs. The language does possess a copula (of very complex morphology showing agreement in noun class and number), and this is required whenever a noun functions as a predicate.
(2)
k-n akrŋ akk vi.sg-fr.dist frog.vi.sg cop.vi.sg ‘That’s a frog.’ (said on hearing the croaking)
23.3 Clause Structure, Information Structure and Head Marking
The syntactic templates to build syntactic structures in Yimas are very simple, in marked contrast to the elaborate morphological templates for verbs, which given the space limitations, I will not be able to explore in detail here, though here is a summary of the morphological possibilities for finite verbs (non-finite verbs are nominalizations); a number of these slots permit multiple fillers:
MOD-PRO-ASP-ADV-ELEV/DIR-VAL-NINCORP-V-VAL-DIR-ASP-TNS-PC-PRO
As this formula demonstrates, much of this word-level complexity has to do with the nesting of operators at various levels (Foley Reference Foley, Evans, Fortescue and Mithun2017b), but here is a typical example: ka-mpu-pay-ma-takat-ɲa-mpan-m (likely-3pl.erg-first-inside-touch-imp-3pl.dat-vii.sg.nom) ‘let them first apply it (paint) to them inside’. Essentially, below the clause level there is not very much syntactic structure; the language strongly prefers to build up clauses simply by stringing words together, and the ordering of these words within the clause is highly flexible. While Yimas, like languages of many Papuan families, has some typological markings of being right-headed, it is by no means verb-final. Any order of the verb and its arguments and adjuncts is acceptable, except verb-initial variants. Given this, there is no evidence for a pre-core slot or a post-core slot, and surprisingly, even the left-dislocated position seems weakly developed, mainly restricted to vocative uses, if these are to be analysed as such. The right-dislocated position, by contrast, is quite often used, essentially for afterthoughts, that is, when the speaker believes the pronominal affix for a core argument on the verb is insufficient to identify its referent, and this most commonly occurs when the referent is a subject, so it somewhat functions then as a switch-reference device when there are multiple animate participants in an ongoing text. As a common tendency, also, locative or temporal expressions tend to occur at the periphery of the clause, the beginning or the end (and again usually the former), but this is only a tendency, and it is not unusual to find them midway in a clause, say, following a noun functioning as a core argument.
In any case, Yimas clauses, like those of many Papuan and other purely oral languages, tend to be information-poor, containing few constituents. Clauses containing a verb and two overt core argument nouns are very rare indeed, much less than 1 per cent of all clauses. Where they might occur is at the beginning of a narrative to introduce the characters. Yimas has no specialized presentative constructions for introducing new referents like ‘there was an old woman who lived …’, so new nouns are introduced baldly in new mentions at the beginning of a story, and this usually by naming them; ‘Yapalmay and Mampalmay lived with their brother Yampwiŋkawi’, for instance, is the opening line of the legend which explains how men’s penises got shortened. There is no distinct clause type for sentence focus. About half of all Yimas clauses with transitive verbs consist of just the verb with perhaps a peripheral temporal or locative expression (often marked with the oblique suffix ‑n ~ ‑nan, the only true nominal case marker in the language), and the other half will have a single overt noun as a core argument, again with perhaps a peripheral adjunct. What determines these patterns is mainly information structure. Nouns whose referents are activated or easily accessible are normally only expressed by the corresponding bound affixes for their noun class and number. To have an overt noun and its corresponding bound affix co-occur in a context in which there is no ambiguity as to their reference makes the referent of the noun highly contrastive. It takes on this contrastive reading from the conflict in interpretation between the overt noun and the bound affix. Overt nouns are used when their referents are inactivated or inaccessible and thereby being introduced into the stream of speech; they are focal in information structure, either part of predicate focus or themselves as narrow focus, as clearly shown in this question-pair.
(3)
Q: wara ipa-n(a)-am-n? what 1pl.nom-pres-eat-pers ‘What are we going to eat?’ A: numpran ipa-n(a)-am-n pig.iii.sg 1pl.nom-pres-eat-pres *??numpran na-kay-ɲ(a)-am-n pig.iii.sg iii.sg.nom-1pl.erg-pres-eat-pres ‘We’re going to eat pork.’
The combination of the overt noun and bound verbal prefix here is very strange because it forces a contrastive reading where the question does not set a context in which this is felicitous. The overt noun signals an inactivated reading, while the bound verbal prefix, an activated or easily accessible one, two contradictory readings that could only be reconciled in a contrastive focus reading, but this makes little sense in this context. So the two most common clause types for Yimas are those with just a verb, where the referents of all core participants are activated or accessible, or a verb and one overt noun, in which the overt noun has an inactivated or inaccessible referent. This is the basic system, though there are a few wrinkles and complications, particularly with wh-questions (Foley Reference Foley1991: 430–433).
Due to the sensitivity of the bound verbal affixes to information structure, the analysis of the head-marking language Lakhota in Van Valin (Reference Van Valin, Bickel, Grenoble, Peterson and Timberlake2013) needs to be expanded a bit. Van Valin (Reference Van Valin, Bickel, Grenoble, Peterson and Timberlake2013) did not intend his analysis of Lakhota to be applicable to all head-marking languages, but rather focused on the type represented by Lakhota, and in this spirit we offer this analysis of Yimas as a further enrichment of the descriptive typology. For Lakhota, Van Valin argues that the verbal prefixes, which may be null, saturate the argument positions within the core and that any overt nouns are in core-external, but clause-internal positions, essentially sisters of the core node under the clause. Given that no basic bound affix of any person, number or noun class combination in Yimas has a null exponent, it would be extremely undesirable to appeal to null affixes in this language, but without them, to apply Van Valin’s analysis of Lakhota to Yimas would require violating the Completeness Constraint, an even more undesirable result, if we want to satisfy all the lexical requirements of the verb and its subcategorized arguments in the core, as we clearly do in this language with no voice alternations or extractions. So instead, I propose that both bound verbal affixes and overt nouns occur in the core, as long as the nouns are inactivated, that is, they do not co-occur with a bound prefix. Those that do, occur in the same core-external position that Van Valin proposes for Lakhota. All of this is illustrated in the following very typical Yimas sentence and its representation in layered structure and information structure. The only constraint on word order here is that the conjunction kanta ‘but’ occurs preferentially in second position, though this is not rigid, and the verb cannot be initial; otherwise all possibilities are acceptable, though some will be more marked pragmatically. Note particularly the separation by this conjunction of the modifier kamta- ‘empty’, which is actually a nominalized verb in a relative clause (Section 23.7.2), hence the tense marker ‑k irr, from the noun parwa ‘dock’ that it modifies, but linked to by the noun class and number concordial suffix proper to nouns of this class and number. Note too the fact that the oblique case suffix ‑n only occurs on the noun parwa ‘dock’, but applies to its linked modifier kamta- ‘empty’ as well. The bound verbal affix is a circumfix for this verb because it has negative polarity (Foley Reference Foley1991: 251–263).
(4)
parwa-n kanta kamta-k-wa ta-pu-tay-kiak-rm kay dock.ix.sg-obl but empty-irr-ix.sg neg-3-see-irr-3dl.nom canoe.viii.sg ‘But they both didn’t see a canoe at the empty dock.’
In Figure 23.1, which represents the structure of (4), RX stands for referential expression, in order to treat nouns and bound prefixes with a single label and avoid positing phrases where there are none, as such are very depauperate in Yimas. I have also simplified for purposes of exposition the representation of the relative clause, as these will be discussed in more detail in Section 23.7.2, and treated it like an adjective, as here it behaves exactly as a true adjective like kpa ‘big’ would.

Figure 23.1 Clause structure with constituent, operator and focus projections
The syntactic structure of phrases is extremely simple in Yimas, and what look like noun phrases are more akin to compounds in structure than phrases. Essentially a noun phrase can consist of no more than two constituents, a noun and its preposed modifier, and those modifiers can only be of two kinds, a possessor, which must be a pronominal or a proper name, or one of the three adjectives, and these cannot co-occur. We can say ama-na kay 1sg-poss canoe ‘my canoe’ and kpa kay big canoe ‘big canoe’ in that fixed order of constituents, but a noun phrase like *ama-na kpa kay 1sg-poss big canoe is impossible in Yimas. One would say ama-na kay kpa-y 1sg-poss canoe.viii.sg big-viii.sg ‘my canoe, the big one’, where the adjective kpa ‘big’ has been converted into a noun by the noun class and number concordial suffix ‑y and, like kamta-k-wa empty-irr-ix.sg ‘empty’ in (4), can be floated away from the noun it modifies and no longer forms a constituent with it, so that this is actually two referential phrases in apposition to each other. Ignoring nominalizations of verbs to be discussed below, referential phrases in Yimas can be made up of at most two constituents, such as ama-na kay or kpa kay, and the vast majority consist of just one, so I will continue to refer to them as referential expressions (RX) in preference to RPs, especially as they are typically at the word (X0) level. The only operator that RXs take is number, at the core level, singular, dual and plural, and adjectives and possessives are peripheral modifiers, so, as in Figure 23.2.

Figure 23.2 Phrase structure with constituent and operator projections
The only other phrase type in Yimas is the postpositional phrase headed by postpositions like nampan ‘toward, for, because of’ or kantk ‘together with’, which form the predicate and nucleus of the phrase and their core arguments are RXs. Postpositional phrases are always predicative peripheral adjuncts; there are no non-predicative peripheral adjuncts in the language, because, as we shall see next, all arguments subcategorized by a verb must be projected as arguments in the core, either as an independent RX or as a bound verbal one.
23.4 Logical Structures and Macroroles
Yimas, like some other Papuan languages (e.g. Kalam; Pawley Reference Pawley and Foley1993), has a restricted lexicon of monomorphemic verbs, only around a couple of hundred verb roots. More complex verbal expressions are constructed by these roots in serial verb constructions in nuclear junctures. Yimas has basic underived intransitive, transitive and ditransitive verbs, although there are only four roots in the last class. There are also lexical processes that derive verbs, a reciprocal prefix that removes a core argument and two causative prefixes, six applicative affixes and possessor raising that add one, but unlike some Bantu languages such as KiHaya (Duranti and Byarushengo Reference Broadwell1977), in no case can a verb be derived that exceeds the number of arguments of an underived verb; the maximum number of core arguments of any verb is three. Beyond these processes of decreased and increased transitivity, there are no derivational processes in Yimas that affect a verb and its arguments’ mapping into constituent structure. There are no voice alternations such as passive or antipassive and no alternations in realization such as we find in English between the farmer loaded hay on the truck versus the farmer loaded the truck with hay. Given a verb’s logical structure and the arguments it subcategorizes, there is one and only one mapping into constituent structure that corresponds to it. The case marking of the arguments is very complex and will be discussed in the following section; here I am just concerned with their basic realization as constituents. Consider an intransitive unaccusative verb like mal- ‘die’, as in na-mal 3sg.nom-die ‘he just died’. This is an achievement predicate and so has the following logical structure: INGR died′ (he). As he is an argument in a frame of pred′ (x), it will be assigned the undergoer macrorole by the universal Actor-Undergoer Hierarchy (AUH), and as the sole argument will appear in nominative case as a bound prefix to the verb. Now consider an intransitive unergative verb like iray ‘cry’ as in na-iray 3sg.nom-cry ‘she cried’. This is an activity predicate with the logical structure do′ (she, [cry′ (she)]). As she is an argument in a frame of do′ (x), by the universal AUH it will be assigned the actor macrorole, but again as the sole argument will appear in nominative case as a bound prefix to the verb. Now consider a canonical transitive verb such as warapa- ‘cut with a flat sharp instrument’ as in (5).
(5)
kaŋk-ɲan na-ka-warapa-ntut shell.vi.sg-obl 3sg.nom-1sg.erg-cut-rm.past ‘I cut it with a shell.’
The logical structure of this clause, including the oblique non-subcategorized RX, would be as in (6).
[do′ (I, Ø)] CAUSE [[ do′ (shell, Ø)] CAUSE [INGR cut′ (it)]]
By the universal AUH, it as the argument of pred′ (x) would be assigned the undergoer macrorole and therefore be realized by nominative case in this direct arrangement of arguments (see Section 23.5). But there are two potential actors, I and shell, as both are first arguments of do′ (x, …) (Yimas, unlike Tagalog, has no constructions which differentiate grammatically between volitional deliberate actions and those which are not, and there are no verb roots which lexicalize this contrast like see versus watch, so there is no need to postulate an operator DO in logical structures.) Yimas, unlike English, does not permit sentences like the shell cut it; the animate instigator who instigates the causal chain must be mentioned overtly and outranks the instrument. Hence it is the animate instigator, here I, which is assigned the actor macrorole and receives ergative case, realized in the core along with the undergoer, while the instrument is realized as an oblique RX in the periphery of the core. Finally, consider the canonical ditransitive verb ŋa- ‘give’ as in (7).
(7)
trawsistm tma-mpu-ŋa-ŋa-ntut trousers.v.dl v.dl.nom-3pl.erg-1sg.acc-give-rm.past ‘They gave me two pairs of trousers.’
Note that Yimas belongs to the relatively rare type of triple-agreement languages; given the proper information structure configuration, all three arguments subcategorized by the verb are realized through bound pronominal affixes on the verb, and so all three must be regarded as core arguments. Yimas seems to be a mix of a direct object language or a primary object language, but just what is it? There are two non-actor core arguments, the theme and the recipient, accessible to the undergoer macrorole, but which one or is there only one? The relevant principles in the revised AUH given in Van Valin (Reference Van Valin2005: 126) are Principle A: lowest-ranking argument in argument structure or Principle B: second-highest-ranking argument in argument structure. Consider (8), the logical structure for (7).
[do′ (they, Ø)] CAUSE [BECOME have′ (me, trousers)]
The actor macrorole is straightforward: as they is the first argument of do′ (x, …) and thereby is assigned the actor macrorole and realized in ergative case. But the undergoer is more problematic: by Principle A, trousers, the lowest-ranking argument would be the undergoer, but by Principle B, it would be me, the second-highest-ranking argument. Is there any way to determine which or must both be reckoned as undergoers? The answer to this question is not easily apparent. There are no voice alternations to check for accessibility, and the evidence is conflicting, but the bulk seems to favour the claim of two undergoers, as in Lake Bantu languages (Bresnan and Moshi Reference Bresnan and Moshi1990; Duranti and Byarushengo Reference Broadwell1977). For example, the case hierarchy to be described in Section 23.5 mandates the following ranking ACC > ERG > NOM. Note in (7) that the referent of trousers is indicated by nominative case and me by accusative. This might establish me as the real undergoer as it outranks trousers in this case hierarchy. But as we shall see, nominative is usually the only obligatory case on a verb if it has any agreement prefixes at all, so it would seem strange to claim less obligatory affixes as marking the undergoer over more obligatory ones; this favours trousers as the undergoer. Yet the nominative case affix always occurs on the periphery of the verb, either right at the beginning or right at the end, in contrast to the more highly salient internal positions closer to the verb stem occupied by the recipient affix. This, then, again may favour the claim of me as the undergoer. On the other hand, recipient arguments cannot be directly relativized like the actor or theme, but require structures that are ambiguous with coordinated clauses; the fact that the actor and theme pattern alike as against the recipient would seem to favour trousers as the undergoer. Yet in possessive-raising constructions like ‘he hit me on the arm’, for which in head-marking languages the affected person is cross-linguistically typically the undergoer, not the body part, the affected person in Yimas is always expressed exactly as the recipient of a ditransitive verb (Foley Reference Foley1991: 300–303) (for comparable data, see the Muskogean languages, Choctaw (Broadwell Reference Dixon2006), Chickasaw (Munro Reference Munro1984; Munro and Gordon Reference Munro and Gordon1982) and Creek (Martin Reference Martin2011)). The evidence is murky and inconclusive, but tentatively, I conclude that ditransitive verbs in Yimas take two undergoers, one by Principle A and one by Principle B. But paralleling the way RRG has rejected the notion of a universal grammatical category of subject, although it can serve as a useful descriptive category for some languages such as English, perhaps we should consider abandoning the concept of the macrorole undergoer as a necessary universal category present in the grammar of all languages, although again valuable for many languages. The concept of undergoer does no necessary descriptive work in Yimas, and insisting that we find one raises a host of problems. The lack of verb-based lexical alternations as with English load above also suggests its lack. The behaviour of applicative affixes in the language can be simply described in terms of adding core arguments with no reference to the concept of undergoer, which in fact just complicates their description. But in keeping with the spirit of RRG and this volume, I will use the notion in the remainder of this chapter. The question now advanced is whether Yimas is unique or do other triple-agreement languages pose the same challenge? It appears at least the Muskogean languages do.
23.5 Case Marking: Split Ergativity and Inversion
Case marking in Yimas is very complex, but there is no quirky or semantically based case. All core case marking is determined by the function and relative topicality of the arguments, and all core case is realized by verbal bound pronominal affixes. The only other case marking is the sole ‑n ~ ‑nan for non-core constituents illustrated in (4) and (5). Core case marking in Yimas is a complex blend of split ergative-accusative alignments overlaid by a direct-inverse system. The split system separates the local persons, namely the immediate speech act persons, first and second, from the non-local person, the third person. The local persons, first and second, align according to a three-way system, distinguishing the transitive actor in ergative case, from the intransitive subject, actor or undergoer, S in Dixon’s (Reference Dixon1979) terms, in nominative case, and from the transitive undergoer in accusative case. The non-local third person distinguishes the transitive actor with ergative case from the intransitive subject and the transitive undergoer, both marked with nominative case. Tables 23.1 and 23.2 present the forms.
Table 23.1 Yimas agreement affixes for local persons
| PRONOUN | ERG | NOM | ACC | ||
|---|---|---|---|---|---|
| DL | kapa | ŋkra- | kapa- | ŋkra- | |
| 1 | PL | ipa | kay- | ipa- | kra- |
| SG | ama | ka- | ama- | ŋa- | |
| DL | kapwa | ŋkran- | kapwa- | ŋkul- | |
| 2 | PL | ipwa | nan- | ipwa- | kul- |
| SG | mi | n- | ma- | nan- |
Table 23.2 Yimas agreement affixes for non-local persons
| PRONOUN | ERG | NOM | DAT | |
|---|---|---|---|---|
| SG | mn | n- | na- | -nakn |
| DL | mrm | mp- | impa- | -mpn |
| PC | mŋkt | ŋkl- | kra- | -ŋkt |
| PL | mum | mpu- | pu- | -mpun |
Note that the second-person and third-person singular ergative prefixes are homophonous. For ditransitive verbs, the theme argument, that which is given or transferred, is indicated by the relevant nominative prefix, and the recipient by the accusative prefixes for the local persons, but by a distinct set of dative suffixes for non-local persons. In the actual marking of verbs with these affixes, two principles apply, both aspects of the direct-inverse system of the language: first, local persons outrank non-local persons, specifically 1st > 2nd > 3rd, and second, surprisingly, accusative prefixes outrank ergative ones which in turn outrank nominative ones. There are also two purely structural constraints: one, the most highly ranked prefix must appear in the immediately preverbal position, and, two, every verb should have a nominative affix, and that must occur on the left edge of the verb in initial position, unless usurped or demoted by a modal or other inflection to the right edge, as in (4). Let’s see how this system all works. Consider how one would say ‘I hit them’ versus ‘they hit me’ in Yimas. The direct form, ‘I hit them’ (local person acting on non-local person) is straightforward. The local person is the transitive actor and ergative, hence ka‑ 1sg erg. The non-local person is the transitive undergoer, and the case-marking system for the non-local person is ergative-nominative, so the undergoer is realized as nominative, pu- 3pl nom, simultaneously satisfying the requirement for an overt nominative. Because local > non-local and ERG > NOM, the first-person ergative prefix will occupy the salient immediately preverbal position, and the third-person nominative the left edge.
(9)
pu-ka-tpul 3pl.nom-1sg.erg-hit ‘I hit them.’
The inverse form ‘they hit me’ is a little more complicated. The undergoer is a local person, and local persons have accusative forms (e.g. ŋa- 1sg acc). Note that the local person is higher by both local > non-local, and ACC > ERG, so ŋa- 1sg acc must occupy the salient immediately preverbal position. The non-local actor would normally take ergative case, and indeed there is an ergative form for third plural: mpu- 3pl erg. But the expected form *mpu-ŋa-tpul is ungrammatical because it runs afoul of the requirement for an overt nominative; such a verb lacks a nominative prefix on the left edge. So instead, the third plural pronominal is realized by the corresponding nominative prefix, pu- 3pl nom.
(10)
pu-ŋa-tpul 3pl.nom-1sg.acc-hit ‘They hit me.’
Ditransitive verbs are no different. Consider ‘I gave it to them’ versus ‘they gave it to me’, the first, direct, the second, inverse. The direct form is straightforward. The non-local recipient undergoer is realized by a dative suffix, so it doesn’t compete with the interaction of the prefixes where the direct-inverse system holds sway. The actor, the ergative pronominal, is first person and therefore the higher ranked by both local > non-local and ERG > NOM, so it appears in the immediately preverbal position. The theme argument undergoer is inanimate; themes with human or higher animate referents are completely prohibited for ditransitive verbs in Yimas. The affixes in Tables 23.1 and 23.2 only apply to referential expressions with human or very high animate referents, like pigs, dogs or crocodiles. Nouns with such referents belong to one of three noun classes. Nouns with all other referents and particularly all nouns with inanimate referents belong to one of the other seven major or a handful of minor noun classes. These classes also have verbal bound referential expressions, but they usually occur in only one case, nominative, again demonstrating the privilege of the nominative case. For our examples here let us take muraŋ ‘oar’, a noun of noun class VI, as the referent of ‘it’, the theme undergoer; the relevant nominative verbal affix is k- vi.sg.nom. As it is a nominative prefix, it satisfies the requirement of an overt nominative and will appear on the left edge, so the resulting form is as in (11).
(11)
k-ka-ŋa-r-mpun vi.sg.nom-1sg.erg-give-perf-3pl.dat ‘I gave it to them.’
The inverse form is more interesting, as local recipient undergoers are realized by accusative pronominals, the most highly ranked. As a first-person undergoer realized by an accusative pronominal, it ranks highest by both local > non-local and ACC > ERG > NOM, so it must occupy the immediately preverbal position. Now the third-person plural actor and the third-person singular theme undergoer slug it out. Both are third person, so local > non-local will not distinguish them, but ERG > NOM will. The actor argument is ergative, and because the theme undergoer is inanimate, it only has the option of being nominative, so by ERG > NOM, the actor argument is of the next highest rank and occupies the position next closest to the verb after the accusative prefix, while the hapless theme argument satisfies the requirement of an overt nominative on the left edge.
(12)
k-mpu-ŋa-ŋa-t vi.sg.nom-3pl.erg-1sg.acc-give-perf ‘They gave it to me.’
Let’s see how we could handle this very complex case system by extending to Yimas case assignment the rules for ergative/accusative constructions provided in Van Valin (Reference Van Valin2005: 108), reproduced here as (13). (There is a much more detailed and typologically varied description of these in Van Valin and La Polla (Reference Van Valin and LaPolla1997), but the points made below would still apply.)
(13)
Rules for accusative case constructions: a. assign nominative case to the highest-ranking macrorole b. assign accusative case to the other macrorole Rules for ergative case constructions: a. assign absolutive case to the lowest-ranking macrorole b. assign ergative case to the other macrorole
It is clear that these will need some modification for Yimas. First, we have no need for an absolutive case, and invoking it would greatly complicate the description. So-called absolutive case is simply nominative, in Yimas and, I would argue, universally. That is why, like nominative in accusative constructions, erstwhile absolutive case constituents are normally formally unmarked and why the nominative is always the privileged case in Yimas regardless of whether it shows up in the accusatively aligned local persons or the ergatively aligned non-local persons. Further, the problem of the lower-ranked macrorole in (13a) is problematic in the case of ditransitive verbs with two undergoers, but let us see how we go. For non-local persons the modification is very straightforward: simply assign nominative case to the lowest argument on the AUH (and dative case to the other one if present) and ergative case to the highest macrorole. The rules for case assignment to the local persons are more complex, because the rules for accusative and ergative case constructions apply simultaneously and partially, that is, only something like the (b) conditions apply. If we were to apply all the conditions, we will get case conflict, a transitive actor will be assigned nominative case by accusative condition (a), but ergative by ergative condition (b). But if we absorb the (a) conditions into and restate the (b) conditions, the system will work: (a) assign accusative case to the lower-ranked macrorole when there are at least two (there is a complication with ditransitive verbs with local recipients; there assign accusative case with the next to highest (or lowest, either will work) argument on the AUH); (b) assign ergative case to the highest-ranked macrorole, again where there are at least two; (c) assign nominative to any core argument that has not received case by (a) or (b). The tree diagram in Figure 23.3 represents the mapping of (11) using these rules from logical structure to constituent structure.

Figure 23.3 Linking from semantics to syntax in (11) of Yimas
23.6 Pivots or Privileged Syntactic Arguments
There is no evidence in Yimas for a privileged syntactic argument (PSA) such as subject in English or the ang phrase in Tagalog. Different constructions target different constituent types, and no one type emerges as privileged across the wide range of constructions. Most constructions in Yimas simply target the notion of core argument. Yimas, like many Papuan languages, employs clause chaining as a clause linkage device (see Roberts, this volume, Chapter 25). But unlike most, it has no restrictions for this to shared subjects or switch-reference tracking and it permits clause chains to be formed on any shared argument without any morphological difference, including none (14b).
(14)
a. tmal kray-mpi ya-kay-am-wat amtra sun.v.sg dry-seq V.pl.nom-1pl.erg-eat-hab food.v.pl ‘the sun having dried it, we eat the food’ b. tmal l-ŋka-p(u)-mpi kumpwia mnta sun.v.sg down-go-away-seq flying.fox.viii.pl then wa-kay-tay viii.pl.nom-1pl.erg-see ‘The sun having set, then I saw flying foxes.’
In (14a) the argument shared between the clauses and elided from the first clause is the undergoer, while in (14b) there are no shared arguments. Some constructions do have an intransitive subject/transitive subject (S/A) pivot, such as genitivization in nominalized non-finite complements.
(15)
a. God-na anti papak-t-wal God-poss ground.viii.sg carve-nfn-custom.v.sg ‘God’s making of the world’ b. *anti-ɲa God papk-t-wal ground.viii.sg-poss God carve-nfn-custom.v.sg ‘the world’s making by God’ c. Yakayapa(n)-na am-t-wal pn-poss eat-nfn-custom.v.sg ‘Yakayapan’s (way of) eating’
In (15a) the actor of a transitive verb papk ‘carve’ has been genitivized in the nominalized non-finite complement, but if we try this with the undergoer in (15b), the result is ungrammatical (though okay in English). However, with the single argument of an intransitive verb (15c), genitivization is again possible. Other constructions, though, work on an intransitive subject/transitive object (S/O) pivot, for example the scope of elevational/directional affixes. Consider these examples.
(16)
a. kay i-ɲa-l-ampu-n canoe.viii.sg viii.sg.nom-pres-down-float-pres ‘The canoe is down there.’ b. kay naŋ-l-arm-na-ŋkan-i canoe.viii.sg imp.pl-down-board-imp-pc-viii.sg.nom ‘You all board the canoe down below.’ ‘*You all down below board the canoe.’
Note that in (16a) the scope of the elevational prefix l- is over the sole argument of the intransitive verb ampu- ‘float’, while in (16b) it is over the undergoer of the transitive verb arm- ‘board’ and cannot be construed as modifying the actor, a clear S/O pivot. The only way to say the equivalent of the starred translation would be something like ‘you all are standing/sitting down below and now you all board the canoe’.
23.7 Clause Linkage and Nominalization
23.7.1 Non-Finite Nominalizations
Yimas contrasts with many Papuan languages and in particular with those of the Trans New Guinea family like Amele (Roberts, this volume, Chapter 25) in preferring subordination rather than the cosubordination by extensive clause chaining as its favourite clause linkage device. It accomplishes subordination through nominalizations, both non-finite and finite. Yimas completely lacks the category complementizer; all nominalizations are simply complex nouns, and as nouns, all must be marked for noun class and, if relevant, number. The formal template is easiest to see with non-finite agentive nominalizations, such as irut ampa-r-mprum mat.ix.pl weave-nfn-ii.dl ‘two female mat-weavers’; a parallel full finite clause would be: irut imp(a)-ampa-wat mat.ix.pl 3dl.nom-weave-hab ‘they both weave mats.’ Note first of all that the nominalization makes a distinction in noun class, here noun class II, the class for female humans, that the corresponding finite verbal prefix impa- 3dl.nom cannot; this is in keeping with its status as a noun, distinguishing the full range of classes that nouns do. The finite tense marker ‑wat hab is replaced by the non-finite marker ‑ru (with a number of allomorphs). All core argument marking by bound verbal affixation beyond the realization of the agent that the entire nominalization refers to is now completely prohibited; the undergoer of the finite clause can be carried over into the nominalization, but it cannot ever be realized by a bound RX: *irut w(a)-ampa-r-mprum mat.ix.pl ix.pl.nom-weave-nfn-ii.dl, *w(a)-ampa-r-mprum ix.pl.nom-weave-nfn-ii.dl. Nor can the undergoer be genitivized as a modifier of the nominalization, as it can in English: *irut-ɲa ampa-r-mprum mat.ix.pl-poss weave-nfn-ii.dl ‘the two female weavers of mats’. This indicates that the undergoer must remain inside the core of the nominalization to satisfy the Completeness Constraint for the verb. But the whole nominalization can be modified, for example by an adjective: irut ampa-r-mprum yua-mprum mat.ix.pl weave-nfn-ii.dl good-ii.dl ‘two good female mat-weavers’. The concordial pattern here tells us that the head of the nominalization is the noun class and number suffix. The non-finite verb is embedded underneath it as a core, with its subcategorized arguments. Because this is an agentive nominalization, the actor is actually bound by the head of the nominalization, the noun class and number suffix, and therefore prohibited from occurring. This all suggests the structure in Figure 23.4 for this construction.

Figure 23.4 Constituent structure of a non-finite nominalization in Yimas
All non-finite nominalizations in Yimas have the structure of Figure 23.4. As we shall see, the structure of finite ones is very similar except that the node of clause is introduced with its operators, modifiers and inflectional possibilities. The nucleus of a non-finite nominalization can be as complex as a finite verb, simply lacking tense and verbal pronominal agreement, and can consist of incorporated adverbials, applicative derivations or verbs in a serial verb construction, that is, verbs linked in a nuclear juncture, as in these examples: kpa-nti-pramuŋ-tu-mat big-adv-sleep-nfn-i.pl ‘men who sleep too much’, taŋkway-cakal-cu-mprum watch.over-feel-nfn-ii.dl ‘the two women looking after (the kids)’, namtamparawt-ɲan api-c-awt foot.ix.dl-obl put.in-nfn-sg ‘a sock (that which puts inside at the feet)’. The different kinds of non-finite nominalizations in Yimas cover the midrange of the Interclausal Semantic Relations Hierarchy (Van Valin Reference Van Valin2005: 208), psych action, jussives, etc. What distinguishes the different types of non-finite nominalizations are the different heads or noun class markers they take (or head noun, as in complements of desire), but they all have the structure of Figure 23.4. When they function as complements, that is, as core arguments of a matrix verb, they can and usually do trigger agreement with that verb, the noun class marker of the nominalization matching the affix on the verb. Beside agentive nominalizations, there are four other types of nominalization that function as complements, exemplified in (17). These are marked by specific noun class marker suffixes utilized only for this function, or in the case of complements of desire by a bound noun wampuŋ, literally ‘heart’. Furthermore, as true complements they function as core arguments of the complement-taking verb of the core of the main clause and are therefore commonly (17a, b, c), though not obligatorily (17d), indicated there by a corresponding bound verbal affix of the same noun class.
(17)
a. complement of words or thought (thinking is internal speech) patn wayk-r-mpwi pia-ka-i-c-mpun betelnut.v.sg buy-nfn-talk talk.nom-1sg.erg-tell-perf-3pl.dat ‘I told them to buy betelnut.’ b. complement of activity nam wark-t-nti tia-ka-ira-karŋkra-t house enclose-nfn-act act.nom-1SG.erg-appl-tired-perf ‘I’m tired of building houses.’ c. complement of customary or habitual action yaki am-t-wal ntak-na-k tobacco.v.pl eat-nfn-custom.v.sg leave-imp-V.sg.nom ‘Stop smoking!’ d. complement of desire tpuk am-t-wampuŋ kpa-n ama-na-t-n sago.x.sg eat-nfn-heart.v.sg big-v.sg 1sg.nom-pres-feel-pres ‘I really want to eat sago.’ (‘I have a big desire to eat sago’)
Note in (17d) that the adjective kpa ‘big’ modifies the nominalization.
Core arguments of non-finite nominalizations may be the pivot of control constructions and when they are, the pivot is strictly S/A; undergoers may never be controlled. Otherwise, genitivization applies, again strictly under an S/A pivot, illustrated above in (15) and in this example.
(18)
patn kpuc-t-wal mpu-na-kn betelnut.v.sg chew-nfn-custom.v.sg 3pl-poss-v.sg ‘their (manner of) chewing betelnut’
The controllers of the S/A pivot in control constructions can be any of the three core arguments, and what determines which is the Theory of Obligatory Control (Foley and Van Valin Reference Foley and Van Valin1984: 307–311; Van Valin Reference Van Valin2005: 243): (1) causative and jussive verbs have undergoer control and (2) all other transitive verbs have actor control. In control structures in Yimas the nominalized complement always functions as an undergoer, so with simple transitive complements there is no issue: the controller can only be the actor, the only other macrorole, as in (17b, d) above and in the examples below.
(19)
a. patn wayk-r-mpwi pia-ka-kacapal betelnut.v.sg buy-nfn-talk talk.nom-1sg.erg-forget ‘I forgot to buy betelnut.’ b. yaki am-t-wampuŋ na-na-t-n tobacco.v.pl eat-nfn-heart.v.sg 3sg.nom-pres-feel-pres ‘He wants to smoke.’
Ditransitive verbs are more complicated as there are two potential controllers, the actor or the second undergoer. In accord with the Theory of Obligatory Control, if the complement is a causative or jussive one, there is undergoer control, as in (17a) and (20).
(20)
impram pay-c-mpwi na-kra-kankantakal basket.vii.sg carry-nfn-talk 3sg.nom-1pl.acc-ask ‘He asked us to carry a basket.’
Note that crucially the jussive or causative semantics is necessary; just because a verb of saying is ditransitive does not mean it will have undergoer control.
(21)
kay yamal-c-mpwi pia-mpu-ŋa-taŋ-tmi canoe.viii.sg carve-nfn-talk talk.nom-3pl.erg-1sg.acc-appl-talk ‘They talked with me about (my, their or someone else’s) building a canoe.’
Here, although the verb is ditransitive through derivation with an applicative, there is no obligatory control, because the semantics is not causative or jussive. Ditransitive verbs that do not have causative or jussive semantics can have, as expected, actor controllers.
(22)
tpuk am-t-mpwi pia-mpu-ŋa-taŋkway-cmi sago.x.sg eat-nfn-talk talk.nom-3pl.erg-1sg.acc-appl-talk ‘He told me (while looking at me) about (his) eating sago.’
In addition to functioning as complements, non-finite nominalizations can function as peripheral adjuncts and as modifiers, essentially non-finite relative clauses. Let us look at the latter first as they are simpler. All relative clauses in Yimas, finite and non-finite, are simply juxtaposed in apposition to the noun or referring expression they modify, just as adjectives can be (17d), They are adjoined (Hale Reference Hale and Dixon1976) and not embedded, as in this example.
(23)
pu-k namat pu-ŋkl-awl-k 3pl-prox person.i.pl 3pl.nom-3pc.erg-get-irr [nampt wark-r-mat] house.pl enclose-nfn-i.pl ‘They few got these people who build houses.’
The structure of (23) is essentially that of a non-finite nominalization nampt wark-r-mat house.pl enclose-nfn-i.pl ‘the ones (masculine) building houses’ as in Figure 23.4 modifying a core argument pu-k namat 3pl-prox man.i.pl ‘these men’, in the same way as kamta-k-wa empty-irr-ix.sg ‘empty’ modifies parwa-n dock.ix.sg-obl ‘at the dock’ in Figure 23.1. Non-finite relative clauses differ from the nominalizations that we have seen thus far only in the noun class marker they take. Instead of the specialized noun class markers for complements in (17), non-finite relative clauses just use the basic adjectival concord set for the various noun classes and numbers of the nouns the modify: awruk [awt yara-t-uŋ] torch.x.sg [fire yara-nfn-x.sg] ‘a torch for picking up fire’, with just some exceptions for those modifying nouns denoting human males or females.
Non-finite nominalizations functioning as peripheral adjuncts always occur with the oblique suffix ‑n ~ ‑nan. They have a very specialized semantic function, indicating two events which are simultaneous. Such a semantic relation is quite low on the Interclausal Semantic Relations Hierarchy (Van Valin Reference Van Valin2005: 208), lower, for example, than circumstances, reasons, conditionals and concessives, but this is inaccurate for Yimas, as all of these latter are expressed by finite nominalizations, while simultaneous events must be expressed with non-finite nominalizations. While oblique non-finite nominalizations have final noun class and number markers as their heads, require ‑ru and have the basic structure of Figure 23.4, they differ in two crucial respects: first, their actors, if overt (and they are always actors, as I have no examples of states, achievements and accomplishments occurring in this construction), must occur as bare RXs like undergoers and may not be genitivized (24c). In this feature, oblique non-finite nominalizations are more like clauses than the ones previously presented, though formally they are still clearly cores attached to a nominal nucleus which is a noun class and number suffix, which is then inflected for oblique case. And second, the noun class and number suffixes do not show concord with an external RX, but rather mark those features for the actor of the nominalization, as in these examples.
(24)
a. [mpa irm-kia-r-ŋkt-ɲan] now stand-night-nfn-pc-obl paŋkra-na-ma-ŋka-pu-kia-k 1pc.nom-pres-in-go.by.land-away-night-irr ‘Standing now, we few walk inside.’ b. [wark-r-mat-ɲan nam] kumpwi mnta enclose-nfn-i.pl-obl house boy.I.pl then numa-mpu-ntak-t village.nom-3pl.erg-leave-perf ‘While building a house, the boys left the village.’ c. arm nampt ya-mpu-tawɲcak-kia-k water house.pl house.pl.nom-3pl.erg-flood-night-irr [m-um pay-kia-r-mat-ɲan num-un-mat] nr.dist-i.pl lie-night-nfn-i.pl-obl village-obl-i.pl ‘The water flooded the houses, while they, the villagers, slept.’
In (24a, b) the actor of the nominalizations is the same as that of the matrix verb, and being such, it can be elided. But such control is not obligatory, as (24c) demonstrates. There the actor of the nominalization is num-un-mat village-obl-i.pl ‘villagers’, while that of the matrix verb is arm ‘water’, which is always formally plural in Yimas, hence the bound prefix mpu- 3pl.erg. Any core argument can be the controller of an elided actor of an oblique non-finite nominalization.
(25)
pu-kra-ant-t pan-t-mat-nan 3pl.nom-1pl.acc-hear-perf pound.sago-nfn-i.pl-obl ‘They heard us when pounding sago.’ (Either they or we could be pounding sago.)
Finally, these oblique non-finite nominalizations can be used to describe locations. In that case, the noun class and number marker is invariably ‑a ix.sg, but there is no known synchronic noun of that class in Yimas that the affix could refer to. It seems to be fossilized, but the overall structure is typical of a non-finite nominalization: mawrun tu-r-a-n enemy.i.sg kill-nfn-ix.sg-obl ‘a place for killing the enemy, battlefield’, awŋkw-cpaŋ-t-a-n go.down.into.water-bathe-nfn-ix.sg-obl ‘a bathing place’.
23.7.2 Finite Nominalizations: Relative Clauses
Finite nominalizations occur with tense suffixes and bound verbal pronominal affixes for core arguments. Finite nominalizations are used in relative clauses and oblique (i.e. adverbial-type) clauses as peripheral adjuncts of the matrix clause, but they may never be used as complements (i.e. core arguments of the matrix verb). All finite nominalizations essentially have the structure of relative clauses, and the formation of relative clauses is somewhat complicated in Yimas, so I can only give a brief overview here. Their formation is complicated as they often involve yet another linear rearrangement of bound pronominal affixes for core arguments, the complexities of which, due to the split case system and the direct-inverse system, are considerable. Relative clauses are adjoined, not embedded, so the noun being modified does not form a constituent with it and, if recoverable from context, can be elided. The relative clause itself is headed by a noun class and number suffix on its verb that is in concord with the noun that it modifies or, if missing, that is recoverable from the context. This bound noun class and number suffix indexes one of the core arguments of the verb, actor or undergoer, that is the relativized noun, and in that sense, relative clauses in Yimas can be seen as internally headed, as the external modified noun, if present, is always resumed in the relative clause by a bound verbal affix according to its noun class and number. That noun class and number suffix, like those in non-finite nominalizations, occurs at the end of the verb of the relative clause and heads it, though not necessarily at the end of the relative clause, as constituents, especially peripheral adjuncts, but not restricted to them, can follow the verb. The noun class and number marker binds the missing relativized noun in the relative clause and provides its referent. The very simplest relative clauses are just of this structure; generally they do not co-occur with a noun and so function as RXs on their own: mal-k-n die-irr-i.sg ‘a corpse’, ŋa-t-ø give-perf-v.sg ‘a gift’, kalc-k-n strengthen-irr-i.sg ‘a strong person’, wa-kia-k-ra go-near-irr-v.pl ‘means/route of going’. Their structure is represented in Figure 23.5. These simple relative clauses function as kinds of patientive nominalizations in opposition to the agentive nominalizations described earlier.
(26)
tu-r-awt tu-t-Ø kill-nfn-i.sg kill-perf-i.sg ‘killer’ ‘someone killed’ amp-r-awt amp-(r)-ra kindle-nfn-i.sg kindle-perf-v.pl ‘fire lighter’ ‘firewood’ kalc-r-awt kalc-k-n strengthen-nfn-i.sg strengthen-irr-i.sg ‘hard worker’ ‘a strong person’ am-t-awt am-t-ra eat-nfn-i.sg eat-perf-v.pl ‘eater’ ‘food’

Figure 23.5 Constituent structure of a simple finite nominalization in Yimas
Most relative clauses are rather more elaborate than this. The verb of the relative clause (and many are just the verb) is typically suffixed as a whole to the near distal deictic stem m- ‘that’. Further, they can have bound pronominal affixes for their subcategorized core arguments. Consider these examples.
(27)
a. namat [m-na-taw-nt-um mnti] person.i.pl nr.dist-pres-sit-pres-i.pl there ‘the people who are sitting there’ b. namat [m-kra-tpul-c-um] person.i.pl nr.dist-1pl.acc-hit-perf-I.pl ‘the people who hit us’ c. namat [m-kay-tpul-c-um] person.i.pl nr.dist-1pl.erg-hit-perf-i.pl ‘the people who we hit’
Example (27a) illustrates a case where the head and relativized argument is the intransitive subject. This single core argument is realized solely by the head suffix, the verb bearing no other bound pronominal bound affix. In (27b) the relativized noun and head is the actor of a transitive verb, realized by the final head suffix, while the undergoer is realized by its proper prefix. Finally, in (27c) the relativized noun and referent of the final head suffix corresponds to the undergoer argument of a transitive verb, while the actor appears now as a prefix in its proper position.
All types of core arguments and even peripheral adjuncts are relativizable (though recipient undergoers with difficulty due to morphological complications (see Foley Reference Foley1991: 417–418)); here are some examples.
(28)
a. on actor Elias [m-kra-pay-pra-kia-ntuk-ŋkt-ø mota-nan] pn nr.dist-1pl.acc-carry-toward-near-rm.past-pc-i.sg motor-obl ‘(It was) Elias who brought us few by motor.’ b. on undergoer anti [God m-n-papk-ntuk-i] ground.viii.sg God nr.dist-3sg.erg-carve-rm.past-viii.sg ‘the world which God made’ panmal m-n-tpul-c-(n)ak narmaŋ man.i.sg nr.dist-3sg.erg-hit-perf-i.sg woman.ii.sg ‘the man which the woman hit’ c. on theme undergoer impram [m-(n)-nan-(n)a-ampa-ŋa-nt-m] basket.vii.sg nr.dist-3sg.erg-2pl.acc-pres-weave-appl-pres-vii.sg ‘the basket which she is weaving for you’ d. on peripheral location (again the fossilized suffix ‑a ix.sg is employed) [maramara m-mpu-t-r-a-n] goods.v.pl nr.dist-3pl.erg-lay.down-perf-ix.sg-obl m-ra ya-kay-ɲa-tacay-kulanaŋ nr.dist-v.pl v.pl.nom-1pl.erg-pres-see (red: tay-)-walk ‘Where they laid out the goods, we walked around looking at them.’ (Literally ‘at the place that they laid out the goods…’) e. as possessor [awt m-nanaŋ-pampay-caw-na-ntut- ø] na fire nr.dist-dur-appl-sit-dur-rm.past-i.sg poss kalakn na-n-tay-mpi-yara-k child.i.sg 3sg.nom-3sg.erg-see-seq-pick.up-irr ‘The son of him who was tending the fire found it.’ f. as possessed manm p-ka-tay m-ɲa cult.house.vii.sg vii.sg.nom-1sg.erg-see nr.dist-poss [m-mpu-tkam-r-m] nr.dist-3pl.erg-show-perf-vii.sg ‘I saw the cult house, yours, which they showed.’
Figure 23.6 illustrates the proposed representation for the relative clause of the second example in (28b).

Figure 23.6 Constituent structure of a relative clause (28b) in Yimas
What is unusual about Yimas in comparison to more familiar languages is that the entire relative clause, with the exception of the ‘leaked’ core argument narmaŋ ‘woman’, corresponds to a single word, and such words can be quite complex (28a, e). But that is entirely in keeping with the polysynthetic typological profile of the language; the word is its favoured grammatical unit and structures above that are very weakly syntactically developed. This fact should again caution us against building grammatical theories solely on the basis of what is familiar to us, such as constituency: they blind us to other possibilities.
The final type of relative clause to consider are those in which the relativized noun is a temporal expression. These are obliquely case-marked like relative clauses on locative adjuncts, and, also like them, the head of these temporal relative clauses is a fossilized noun class and number suffix, of the form ‑mp vii.sg. But in this case we know the noun to which it refers because it is still synchronically in the language: pucm ‘part, piece, time’. All the semantic relations on the Interclausal Semantic Relations Hierarchy that are expressed typically by adverbial subordinate clauses in more familiar languages, circumstances, reasons, conditionals and hypotheticals and concessives, are expressed by this type of temporal relative clause in Yimas. The differences between them are captured in the choice of tense and mood for both the relative clause and the matrix clause. For example, for temporal circumstances, the verb of the relative clause occurs with one of the past or present tenses or that of the definite future.
(29)
[m-mpu-ŋa-na-tay-ɲc-mp-n] nr.dist-3pl.erg-1sg.acc-pres-see-pres-vii.sg-obl pu-ka-apan-kt 3pl.nom-1sg.acc-spear-rm.fut ‘When they see me, I will spear them.’
But conditionals and hypotheticals are marked with mood affixes that mark unreal or as yet unrealized events.
(30)
[m-mpu-ya-kr-mp-n Wamur-mat] nr.dist-3pl.erg-come-rm.fut-vii.sg-obl Wambramas-i.pl mambayŋki wunt kantk banana.vi.pl sago.grub.v.pl with pay-pra-kt ama-na-ra carry-toward-rm.fut 1sg-poss-v.sg ‘If the Wambramas people will come, they can bring my bananas and sago grubs.’
Counterfactuals are slightly different. They require the potential modality prefix ant-, which expresses that an event is unlikely to happen, to co-occur with a past- or present-tense marker on the verbs of both the relative clause and the matrix clause. As the potential modality marker is a prefix, it triggers rearrangement on verbs of the matrix clause of the normally leftmost bound nominative pronominal prefix to right edge.
(31)
[tuŋkuruŋ ant-ka-tay-c-mp-n] eye.vi.sg pot-1sg.erg-see-perf-vii.sg-obl ant-ka-tu-r-(n)ak pot-1sg.erg-kill-perf-iii.sg.nom ‘If I had seen the eye (of the crocodile), I would have killed it.’
Finally, concessives are formed by having the polarity of the relative clause being positive, and that of the matrix clause, negative.
(32)
[kay i-ka-ak-r-mp-n] canoe.viii.sg viii.sg.nom-1sg.erg-push-perf-vii.sg-obl arm-n ta-ka-wul-c-i water-obl neg-1SG.erg-put.down-perf-viii.sg ‘Although I pushed the canoe, I didn’t put it down into the water.’
23.8 Conclusion
Yimas is an intriguing language that has much to offer grammatical theory in its unusual grammatical structures due to its complex and rare mix of a polysynthetic verb structure with an elaborate noun-class concordial system. Sadly, it is now rapidly approaching extinction, but fortunately it was documented when it was still in a relatively vibrant state and its full richness able to be appreciated. Without the input of data from languages like Yimas, our grammatical theory will be very impoverished indeed. But many such crucial languages like Yimas are dying before our very eyes. Grammatical theory is a scientific endeavour worthy of pursuit, but unless we document the full range of grammatical possibilities, it will never be complete. We may never know the grammatical riches the human language capacity is able to evolve. And time is rapidly running out to do that. Our discipline needs seriously to look at its priorities.
Abbreviations
We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/glossing-rules.php) with the following additions:
- C
noun class
- CLM
clause linkage marker
- CM
clause marker
- HAB
habitual
- INT
intentive
- IT
itive
- REC
recurrent
- SVM
serial verb marker
- VEN
ventive
24.1 Introduction
In this chapter we give an overview of the grammar of Avatime from a Role and Reference Grammar (RRG) perspective. Avatime belongs to the Kwa branch of the Niger-Congo language family. Within Kwa, it belongs to the Ka-branch of the group of Ghana-Togo Mountain (GTM) languages. It is spoken in the south-east of Ghana, by about 15,000 speakers.
As a background to the discussion of the grammar, two properties of Avatime phonology are important to mention. First, Avatime is a tone language. It has three level tones: low (marked à), high (unmarked) and extra-high (marked á). The extra-high tone has a limited distribution and occurs mostly on grammatical morphemes and function words. Second, there is advanced tongue root (ATR)-based vowel harmony. Vowels in most affixes and clitics harmonize in their ATR value with the nearest root vowel. +ATR vowels are i, e, o, u and −ATR vowels are ị, ɛ, a, ɔ, ụ.
The information presented in this chapter is based on data collected during several field trips to the Avatime area between 2008 and 2013.1 More elaborate descriptions of the grammar of Avatime can be found in Defina (Reference Defina2016a) and Van Putten (Reference van Putten2014a).
Part I of this chapter gives an overview of the basic grammatical structures of simple sentences. It describes lexical categories, predicate classes and logical structure, the layered structure of the clause, including the operator projection, referential phrases and adpositional phrases and the linking from semantics to syntax in simple sentences. Finally, it examines focus structure. Part II discusses the syntax and semantics of serial verb constructions and the questions they raise for the Interclausal Relations Hierarchy (Van Valin Reference Van Valin2005: 209).
PART I. Basic Grammatical Structures
24.2 Lexical Categories
Avatime has the distinct lexical categories of noun, verb, adjective, ideophone, adverb, preposition and postposition. Nouns are morphologically distinguished by their noun-class prefixes (see Section 24.5.1) and can be modified by adjectives, numerals and determiners. Nouns are used as referring expressions. They cannot function as predicates or modifiers and there is no way to derive other lexical categories from nouns.
Verbs are distinguished from other lexical categories by their ability to be marked with a subject prefix and aspect, modality, negation and direction markers (see Section 24.4.2). Verbs are used as predicates. They can be used as referring expressions only when they are nominalized. To fully nominalize a verb, the verb is reduplicated and optionally marked with the noun-class prefix ku-/kụ-. There are two constructions where the verb is marked with a noun-class prefix without being reduplicated: in non-finite complement constructions, where it is marked with the prefix ku-/kụ- and in the focus construction (see Section 24.7), where it is marked with the prefix ki-/kị-. In these cases, the verb appears to be nominalized to a certain extent, but it cannot function as a full noun in this form.
Avatime has a small class of non-derived and non-ideophonic adjectives. These modify nouns and cannot be used by themselves as predicates or referring expressions. They do not share any morphological properties to distinguish them from other lexical categories. The majority of property words are not adjectives, but verbs or ideophones. Verbs which include a property or a state as part of their meaning can be turned into adjectives by reduplication, such as kpakpa ‘dry’ from the verb kpa ‘to dry’. Adjectives can be nominalized by prefixing them with a noun-class prefix. For instance, the adjective sị̀sàmi ‘small’ can be turned into ɔ-sị̀sàmi ‘small one’ (referring to something of class 1 singular).
Avatime also has a small class of non-ideophonic adverbs. Like the adjectives, they do not share any morphological properties, but they are distinguished by their function as modifiers of predicates or clauses. These adverbs cannot be used as predicates, referring expressions or modifiers of nouns. Some examples are nyafɛ ‘maybe’, àbla ‘now’ and kóko ‘already’.
There is a large class of ideophones that overlaps in function with the classes of adjectives and adverbs. Ideophones are ‘marked words that depict sensory imagery’ (Dingemanse Reference Dingemanse2012, see Chapter 9 of this volume). They can be used to modify nouns, predicates and clauses, but they cannot be used as predicates or referring expressions. Ideophones are distinguished from other word classes by their marked phonology (e.g. long words, repeated syllables, long final vowels). Some examples are pịtịtị ‘white’, hãããã ‘intensely’ and trátrátrátrá ‘very neat’.
Finally, Avatime has small sets of prepositions and postpositions. There are only two prepositions: locative ní and comitative (a)nì. There are five postpositions, four of which have evolved from nouns. An example is ese ‘under’, which comes from the noun kese ‘ground’ (for examples of adpositions see Section 24.5.2). As we show in Section 24.5.2, the postpositions are very noun-like in their use and can be analysed as heading referential phrases (RPs). Nevertheless, they are a separate lexical category on both morphological and syntactic grounds. Morphologically, they are distinct from nouns as they have lost their noun-class prefixes. Syntactically, they cannot be modified with nominal modifiers such as definite articles and can only occur in the position following another referential phrase.
24.3 Predicate Classes and Logical Structure
We distinguish four main Aktionsart classes in Avatime: states, activities, semelfactives and accomplishments. The primary distinguishing factor is their interpretation with progressive and perfective aspects. For more detail about Avatime Aktionsart classes, see Defina (Reference Defina2009, Reference Defina2018).
States are interpreted as present in the perfective, as can be seen in the example in (1). They can be marked as progressive, where they are interpreted as persistent states (e.g. ‘the building is still in Vane’).
(1)
lịgba=lὲ lị-lị́ ní ɔ̀ʋanɔ̀ c3.sg-building=def c3.sg.sbj.pfv-be.at loc Vane ‘The building is in Vane.’ Logical Structure: be.at′ (ɔ̀ʋanɔ̀, lịgbalὲ)
Activities are dynamic events without an inherent endpoint. They are generally interpreted as occurring in the past in the perfective, as in example (2), and as occurring in the present in the progressive (e.g. ‘Afua is eating rice’).
(2)
àfua a-tá kị-mịmị Afua c1.sg.sbj.pfv-chew c4.sg-rice ‘Afua ate rice.’ Logical Structure: do′ (Afua, [eat′ (Afua, kịmịmị)])
Accomplishments are state-change events with internal duration. Like states, they are typically interpreted as present in the perfective, as can be seen in example (3), but they are interpreted like activities in progress with the progressive (e.g. ‘the door is opening’).
(3)
ò-pópo=lò ɛ-drà c2.sg-door=def c2.sg.sbj.pfv-open ‘The door is open.’ Logical Structure: BECOME open′ (òpópo)
Semelfactives are events that lack duration. Like activities, they are interpreted as past actions in the perfective (e.g. ‘she came’). However, unlike activities, they cannot be modified by duration adverbials such as ‘for 5 minutes’ and ‘quickly’. They can be used with the progressive, in which case they are interpreted as iterative or repeated actions, as in example (4).
(4)
o-ne=e ὲέ-ba c2.sg-mother=def c1.sg.sbj.prog-come ‘The mother comes regularly.’ Logical Structure: SEML come′ (one)
As yet, no clear diagnostic for the classes achievement and active accomplishment has been found for Avatime. The typical test using the modifiers ‘for X time’ or ‘in X time’ is not applicable as this specific semantic distinction is not consistently made. There is, however, a group of atomic telic predicates where the endpoint is reached in a single step rather than gradually (Caudal and Nicolas Reference Caudal, Nicolas, Maienborn and Wöllstein2005), which can be distinguished based on their incompatibility with the adverb petee ‘completely’, as in example (5).
(5)
osófo=e a-trɛ ní òholò (*petee) pastor=def c1.sg.sbj.pfv-go loc Ho (*completely) ‘The pastor went to Ho (*completely).’ Logical Structure: do′ (osófo, [go′ (osófo)]) & INGR be.at′ (Ho, osófo)
In addition to the Aktionsart predicate classes discussed above, it is worth mentioning causative predicates. Some verbs, such as trɔ ‘put on’ and bàsị ‘show’ are always causative. Other verbs have a non-causative reading when they are used intransitively and a causative reading when they are used transitively, such as dra ‘open’ in example (6) (see also (3) above).
(6)
ɔ-dɔ̀bàsị=ɛ a-dra ò-pópo=lò c1.sg-teacher=def c1.sg.sbj.pfv-open c2.sg-door=def ‘The teacher opened the door.’ Logical Structure: do′ (ɔ̀dɔ̀bàsi, Ø) CAUSE [BECOME open′(òpópo)]
24.4 Clause Structure and Operators
24.4.1 Clause Structure
Canonical constituent order in Avatime is SVO: the privileged syntactic argument (‘subject’; PSA) precedes the verb, and other arguments (‘direct/indirect object’) follow the verb. Adverbial phrases follow the postverbal arguments. There is head-marking of the PSA on the verb. A canonical transitive sentence with an adverbial phrase in Avatime can be seen in example (7) and the syntactic structure of this example is shown in Figure 24.1. Following Van Valin (Reference Van and Valin1985) and Van Valin and LaPolla (Reference Van Valin and LaPolla1997), we analyse the PSA-indexing prefix on the verb as an argument in the core, whereas the RP with which this prefix is coreferential is analysed as occurring outside the core, but inside the clause. This captures the observation that a lexical RP or independent pronoun is not necessary, and example (7) remains grammatical when leaving out the RP ɔ́dzɛɛ ‘the woman’.
(7)
ɔ́-dzɛ=ɛ á-ta á-ʋa=nà ní lị-gba=le c1.sg-woman=def c1.sg.sbj.pfv-chew c3.pl-bean=def loc c3.sg-room=def mὲ inside ‘The woman ate beans in the room.’

Figure 24.1 The layered structure of the sentence in example (7)
Avatime makes use of the pre-core slot (PrCS) for question formation and focus marking (see also Section 24.7 and Van Putten Reference van Putten2016). The question word or focused RP is placed in the pre-core slot and marked with a final extra-high tone, as shown in examples (8) and (9).
(8)
egé ɔ́-dzɛ=ɛ a-ta what c1.sg-woman=def c1.sg.sbj.pfv-chew ‘What did the woman eat?’ (elic-QUIS-foc_100714_SO)
(9)
á-ʋa=ná ɔ́dzɛ=ɛ a-ta c3.pl-bean=def:foc c1.sg-woman=def c1.sg.sbj.pfv-chew ‘The woman ate [the beans]FOC.’ (elic-QUIS-foc_100714_SO)
The pre-detached position is frequently used to host topical expressions or expressions encoding spatial or temporal background information (see Van Putten Reference van Putten, van Gijn, Hammond, Matić, van Putten and Galucio2014b). When an argument that is not the PSA is pre-detached, a resumptive pronoun occurs in its regular postverbal position, as shown in example (10). Example (11) shows a sentence with both focus-marking and a pre-detached element. The syntactic structure of this example is shown in Figure 24.2. Note that the PrCS is not the same structural position as the regular position for PSA RPs, in that it is also within the clause but outside the core. Focused elements always precede the PSA RP.
(10)
á-ʋa=nà, ɔ́-dzɛ=ɛ a-ta na c3.pl-bean=def c1.sg-woman=def c1.sg.sbj.pfv-chew c3.pl ‘The beans, the woman ate them.’
(11)
kivoe á-ʋa=ná ɔ́-dzɛ=ɛ á-ta yesterday c3.pl-bean=def c1.sg-woman=def c1.sg.sbj.pfv-chew ‘Yesterday, the woman ate [the beans]FOC.’

Figure 24.2 A sentence with pre-detached position and pre-core slot (example 11)
24.4.2 Operators
In this section, we give an overview of the operators that have scope over the different layers of the layered structure of the clause. The Avatime verb is usually marked with a prefix that indexes the PSA. This prefix (henceforth referred to as the PSA prefix) also encodes aspect or mood. It is optionally followed by other aspect, mood and directionality prefixes. There is no tense marking. An overview of the categories marked on the verb and the order in which they occur can be seen in (12) and an example with all slots occupied can be seen in (13).
PSA + Perfective/Progressive/Habitual/Potential/Subjunctive + (Negation) – (Intentive) – (Recurrent) – (Directional) – Root
(13)
mɔ́-tá-zɛ̌-zɛ-panị̀ wɔ 1.sg.sbj.pfv.neg-int-rec-it-talk 2.sg ‘I will not be going to talk with you.’ (R0811291)
There are five paradigms of PSA prefixes, as shown in Table 24.1. For the sake of brevity, the forms for the noun classes other than class 1 singular and plural have been omitted (a full table can be seen in Van Putten Reference van Putten2014a or Defina Reference Defina2016a). As the table shows, there is no one-to-one correspondence between each paradigm and an aspect/mood category. It is also clear that the prefixes are formally related to each other; it is likely that they have evolved from combinations of a prefix indexing the PSA plus an aspect/mood marker, but they cannot be synchronically analysed as such.
Table 24.1 Paradigms of PSA prefixes (most noun classes omitted)
| Person/ number/ noun class | Set 1 Perfective (most verbs) | Set 2 - Negative (with extra-high tone) - Subjunctive (some verbs) - Perfective (locative and copula verbs) | Set 3 - Subjunctive (some verbs) - Habitual (with zɛ̌- prefix) | Potential | Progressive |
|---|---|---|---|---|---|
| 1s | me-/ma- | mo-/mɔ- | mi-/mị- | máà- | mèé-/mὲέ- |
| 1p | ki-/kị | ku-/kụ- | ki-/kị- | kị́à- | kìí-/kị̀ị́- |
| 2s | wo-/wɔ- | wo-/wɔ- | wu-/wụ- | wáà- | wèé-/wὲέ- |
| 2p | mle-/mlɛ- | mla | mli-/mlị- | mláà- | mlèé-/mlὲέ- |
| C1.SG | e-/a- | o-/ɔ- | i-/ị- | áà- | èé-/ὲέ- |
| C1.PL | be-/bɛ- | ba | bi-/bị- | bị́à | bèé-/bὲέ- |
| … |
We will now briefly discuss the operators one by one. For more detail about these markers and their semantics, see Defina (Reference Defina2018).
Given that aspect marking modifies the internal temporal structure of the event itself, the aspect markers are considered to be nuclear-level operators (Van Valin and LaPolla Reference Van Valin and LaPolla1997; Van Valin Reference Van Valin2005). Avatime has four aspectual categories: perfective, progressive, habitual and recurrent. Their forms are described in (14).
(14) Aspect marking
The perfective is marked with a PSA prefix from set 1 or, in the case of locative and copula verbs, set 2. It is the most frequently occurring verb form in Avatime, some examples of which have already been presented in (7)–(11).
The progressive has its own paradigm of PSA prefixes, as shown in Table 24.1.
The habitual uses a PSA prefix from set 3, combined with the prefix zě-/zɛ̌-, see example (15).
The recurrent is one of the two optional aspect/mood prefixes and indicates repeated or recurring actions and situations. It is marked with the prefix zě-/zɛ̌-,2 which is always combined with one of the obligatory aspect/mood categories, see example (16).
(15)
mí-zě-do mawùye li-bo=lè ní sɔ̀lὲ mὲ 1sg.sbj.hab-hab-say God c3.sg-word=def loc church inside ‘I preach the word of God in church.’ (life_100614_WE)
(16)
ma-zɛ̌-wà a-xwὲ=na ní cocoa marketing board 1.sg.sbj.pfv-rec-do c3.pl-work=def loc cocoa marketing board ‘I used to work at the cocoa marketing board.’ (life_AB)
There are two directionals: the itive (‘away’), marked with the prefix ze-/zɛ- and the ventive (‘towards’), marked with the prefix bá-/bé-, shown in example (17). These forms express orientation or motion away from or towards the deictic centre. This means they are core-level operators (see Van Valin and LaPolla Reference Van Valin and LaPolla1997: 45).
(17)
a-bá-feke télefòn=ye c1.sg.sbj.pfv-ven-pick.up telephone=def ‘He came and picked up the telephone.’ (finsto_100716_DQ)
Another core operator is the intentive, which indicates the intention of the actor to carry out the action expressed by the verb. The intentive is marked with the optional prefix tá- which combines with either the perfective aspect or the subjunctive mood. For an example of the former, see (13).
Negation is expressed by a PSA prefix from set 2 with an extra-high tone. An example can be seen in (13). When the progressive is negated, this PSA prefix is followed by the prefix li-/lị-. When the habitual is negated, the prefix zě-/zɛ̌- is deleted, leaving only its tone and lengthening the vowel of the PSA prefix. Negation marking can be used for both core-level and clause-level negation.
Avatime has three clause-level operators marked on the verb: potential, subjunctive and imperative. Their properties are described in (18).
(18) Clause-level operators marked on the verb
The potential has its own PSA agreement paradigm as shown in Table 24.1. The potential PSA prefix cannot be combined with negation marking. For a negative potential interpretation, the negative perfective intentive is used.
The subjunctive is marked with a subject prefix from set 2 for some verbs and for other verbs with a subject prefix from set 3. An example is shown in (19). When the subjunctive is negated, a prefix ku-/kụ- is added while the PSA prefix remains the same. The subjunctive has irrealis and illocutionary force interpretations.
The imperative is marked by the absence of a PSA prefix, as in ba ‘Come!’. It is only used for second-person singular commands. For other commands, the subjunctive is used. To negate imperative clauses, the negative subjunctive is used.
(19)
mla-trɛ 2.pl.sbj.sbjv-go ‘You (pl) should go.’ / ‘Go! (to multiple people)’
The final operator to be discussed here is question marking, an illocutionary force operator. Polar questions are marked with a clause-final particle na or, more frequently, with intonation only. Content questions are marked with a question word in the pre-core slot (see example (8) in Section 24.4.1).
According to RRG (Van Valin and LaPolla Reference Van Valin and LaPolla1997; Van Valin Reference Van Valin2005), the order of operators in a given language always follows the same pattern whereby clausal operators occur furthest from the verb, core operators closer than clausal operators, and nuclear operators are the closest to the verb. In Avatime, this prediction is mostly borne out, but there are two exceptions: (i) the (core) directional markers occur closer to the verb than the (nuclear) aspect markers, and (ii) the (core) intentive marker also occurs closer to the verb than the (nuclear) perfective marker. The directional markers have clearly grammaticalized from motion verbs (ba ‘come’ and za ‘pass’). Before grammaticalization, they would have occurred as the first verb in a serial verb construction (SVC), directly followed by the second verb (see Sections 24.9 and 24.10 for more information about serial verb constructions). As all aspect and mood marking in SVCs is on the first verb, it now precedes these directional prefixes. A similar explanation could account for the intentive following the perfective: the intentive tá- is also likely to be the result of a grammaticalization process from a serial verb construction, probably from the verb trɛ ‘go’ or tráà ‘come’. Another possible explanation for the intentive following the perfective is the hypothesis that the perfective forms used to be the bare subject agreement markers unmarked for aspect/mood, as is the case in many related languages (Welmers Reference Welmers1973). Through language change, the default PSA prefix came to be used solely for perfective marking and thereby turned into a nuclear-level operator, which happened to occur in a position preceding the intentive marker.
Figure 24.3 shows the sentence in example (13), repeated here as (20), with both constituent and operator projections.
(20)
mɔ́-tá-zɛ̌-zɛ-panị̀ wɔ 1.sg.sbj.pfv.neg-int-rec-it-talk 2.sg ‘I will not be going to talk with you.’ (R0811291)

Figure 24.3 Sentence from example (20) with constituent and operator projections
24.5 Referential and Adpositional Phrases
24.5.1 Referential Phrases Headed by Nouns
Like clauses, referential phrases (RPs) have a layered structure. The nucleus of the Avatime RP can be either a noun or a postposition. When it is headed by a postposition it functions as an adpositional phrase, and this type of RP will be discussed in Section 24.5.2. This section will instead discuss RPs headed by nouns.
Avatime nouns, like those in many Niger-Congo languages, take part in a noun-class system. Each noun (with the exception of some loanwords) carries a noun-class prefix which indicates its gender and number. Within the referential phrase, there is agreement on numerals, articles and demonstratives (see example (24)). For more detailed information on the noun-class system and the various nominal modifiers, see Schuh (Reference Schuh1995), Van Putten (Reference van Putten2014a) and Defina (Reference Defina2016a).
The constituent order in Avatime RPs is shown in (21). Adjectives are considered to occur in a periphery of the nucleus, following Van Valin (Reference Van Valin2005), whereas the other nominal modifiers are operators on different layers of the RP.
noun – adjectives – numeral – determiner – particles
The only operator on the nuclear level is number, which is marked on the noun class prefix (e.g. ɔ̀-klị ‘leg’ vs. ị̀-klị ‘legs’). The numerals are operators on the core level, as they provide quantification. The determiners are operators on the RP level, as they ground the RP in discourse. The class of determiners contains the elements shown in (22), of which only one can occur in a given RP.
(22) Determiners
The definite article, which is a monosyllabic enclitic and agrees with the noun class of the head noun.
The indefinite article tɔ, prefixed with a noun-class agreement prefix. This article has a specific indefinite interpretation (e.g. ‘some’, ‘a certain’). For a non-specific indefinite interpretation, the determiner slot is unfilled.
The proximal demonstrative yà and distal demonstrative lɔ̀, both prefixed with a noun-class agreement prefix.
The class of particles contains words such as tsyɛ ‘also’, kɔ ‘by contrast’ and kò ‘only’. Like definiteness and deixis markers, they ground the RP in discourse and are therefore considered to be RP-level operators.
Possession in Avatime is mostly indicated by juxtaposition with a possessor–possessed word order. The regular independent pronouns are used as possessor pronouns. For possessed kinship terms, there is a slightly more complex construction in which the possessor pronoun fuses with the noun-class prefix of the possessed kinship term.3 With a nominal possessor, this fused possessor pronoun is also present, and, thus, in these cases there is head-marking of the possessor on the possessed noun. Examples of kinship and non-kinship terms with nominal and pronominal possessors can be seen in (23).
(23)
yɛ ‘c1.sg’ + ɔ̀-mà-nɔ ‘the town’ → yɛ ɔ̀-mà-nɔ ‘her town’ yɛ ‘c1.sg’ + o-ne ‘mother’ → ye-ne ‘her mother’ ɔ-nùvɔ̀-ɛ ‘the child’ + li-kuto-lɛ ‘the hat’ → ɔnùvɔ̀-ɛ li-kuto-lɛ ‘the child’s hat’ ɔ-nùvɔ̀-ɛ ‘the child’ + o-ne ‘mother’ → ɔnùvɔ̀-ɛ ye-ne ‘the child’s mother’
The possessor occurs in the RP initial position (RPIP), analogous to both the pre-detached position and pre-core slot in the layered structure of the clause (see Van Valin Reference Van Valin2005: 26).
An example of a complex RP in Avatime can be seen in (24). The constituent and operator projections of this example are shown in Figure 24.4.
(24)
bá-dzɛ kpekpe tia-bà=a ke-pe=à c1.pl.woman short c1.pl-two=def c6.sg-house=def ‘The house of the two short women.’

Figure 24.4 Complex RP with constituent and operator projections (example 24)
24.5.2 Adpositional Phrases
As mentioned in Section 24.2, there are both prepositions and postpositions in Avatime. Both of Avatime’s prepositions – the comitative preposition (a)nì ‘and/with’ and the general locative preposition ní – are predicative prepositions. This means they occur in the nucleus of a prepositional phrase (see Figure 24.5 and Chapter 10 of this volume). The five locative postpositions are indicators of what Ameka (Reference Ameka1995) calls the search domain, that is, the part of the reference object where something is located. Postpositional phrases are not locative by themselves but behave like noun phrases; they can, for instance, serve as arguments of a verb. Because of this, we analyse the postpositional phrase as an RP with the postposition as its nucleus and the noun expressing the location in an RP-initial position, analogous to possessive constructions. This makes sense from a historical perspective, as most postpositions have evolved from nouns which would have combined with the locatum in a possessive construction (e.g. ‘the upper surface of the table’ → ‘on the table’, see also Section 24.2).
In locative phrases, a postpositional phrase must be combined with the preposition ní, as shown in example (25). The constituent projection of this example is shown in Figure 24.5. Instead of a postposition, a noun can also be used to indicate the search domain, as in example (26). This follows from the analysis presented above: in this case the two nouns occur in a possessive construction, as the situation with the prepositions would have been before grammaticalization was complete. Thus, example (26) literally translates as ‘located at the back of the car’.
(25)
ní ɔ̀-kplɔ=nɔ̀ abà loc c2.sg-table=def on ‘on the table’
(26)
ní ò-hui=lò ke-de=à loc c2.sg-car=def c6.sg-back=def ‘behind the car’

Figure 24.5 Constituent projection of a PP (example 25)
24.6 Macroroles, Arguments and Linking from Semantics to Syntax
In this section, we discuss how logical structure is linked to syntactic structure, including how semantic macroroles are assigned in Avatime and how the privileged syntactic argument is selected.
Macrorole selection in Avatime simple sentences follows the Actor-Undergoer Hierarchy (AUH) shown in (27). The highest-ranking argument is selected as the actor. When it comes to the selection of the undergoer, Van Valin (Reference Van Valin2005) shows that, across languages, there are two possible principles: either the lowest-ranking or the second-lowest-ranking argument is selected as the undergoer. When a verb has only two core arguments, the two principles will always select the same undergoer. However, a difference emerges with three-place predicates. In Avatime three-place predicates, the two non-actor arguments always follow the verb with the ‘indirect object’ preceding the ‘direct object’, as shown in example (28). The logical structure of this sentence is shown in (29) and makes clear that the highest-ranking argument is Kofi, the second-highest is the first-person singular pronoun and the lowest-ranking argument is egumena ‘the cows’. Given that there is no case marking and no indexing of either of the two arguments on the verb, there doesn’t seem to be clear evidence for which of the two is the undergoer and therefore which of the two principles applies.
(27) The Actor-Undergoer Hierarchy (based on Van Valin and LaPolla Reference Van Valin and LaPolla1997 Fig. 4.2)
ACTOR UNDERGOER -----------------------------------------------------> <------------------------------------------------------------- Arg. of
DO1st arg. of
do′ (x,…)1st arg. of
pred′ (x, y)2nd arg. of
pred′ (x, y)Arg. of state
pred′ (x)[‘------------>’ = increasing markedness of realization of argument as macrorole]
(28)
kòfi a-bàsị mɛ e-gume=nà Kofi c1.sg.sbj.pfv-show 1.sg c3.pl-cow=def ‘Kofi showed me the cows.’
do′ (Kofi, Ø) CAUSE [see′ (1s, egumenà)]
After assigning macrorole status, the PSA is selected. In Avatime, the PSA is the argument that is indexed on the verb. Avatime has a nominative-accusative alignment pattern: if there is only one macrorole argument, this is selected as the PSA, and if there are two macroroles, the actor is selected as the PSA. The PSA is invariable: there is no passive construction in Avatime and non-macrorole PSAs are not allowed. In example (28), the actor, Kofi, is selected as the PSA.
After the selection of the PSA, the appropriate syntactic template is selected. The syntactic template for a canonical Avatime sentence has one core slot before the verb for a head-marked argument and up to two core slots for RP arguments after the verb (see Section 24.4.1). In the case of example (28), a syntactic template with two core slots after the verb is chosen, as there are three arguments in the logical structure. The preverbal head-marking slot is linked to the PSA, which is Kofi. This slot can only contain PSA-indexing prefixes and does not allow referential phrases. Therefore, it is filled with a PSA-indexing prefix of noun class 1 singular (class 1 is used for people) and the RP Kofi is placed in the pre-core position (see also Section 24.4.1). The undergoer and non-macrorole arguments are linked to the two postverbal core-internal slots.
24.7 Focus Structure
In addition to the constituent projection and operator projection, RRG posits a third component of grammar, representing the information structure of the sentence: the speech-act projection (see Chapter 11 of this volume). In this section, we briefly discuss focus marking in Avatime. For more detail, see Van Putten (Reference van Putten2016).
The potential focus domain in Avatime canonical clauses coincides with the entire clause, that is, any element of a clause can potentially be in focus. Focused elements in canonical clauses are not marked as such with intonation; the focus interpretation comes purely from the context.
To make clearer which part of the sentence is in focus, a syntactic focus construction can be used (see also Section 24.4.1). This construction is often used when the focus is contrastive. In this construction, a constituent is placed in the pre-core slot and marked with a final extra-high tone. This structure is most frequently used for narrow focus, that is focus on the constituent in the pre-core slot only, as was shown in example (9) in Section 24.4.1. To mark narrow focus on the verb, a copy of the verb root, prefixed with the noun class prefix ki-/kị- is placed in the pre-core slot and marked with the final extra-high tone, while the inflected verb remains in its normal position. An example is shown in (30), where the focus is on the lexical content of the verb, but the same construction can also be used to mark focus on an aspect, modality or status operator (see also Van Putten Reference van Putten2016).
(30)
ki-hɔ́ bɛ-tá-hɔ lɔ c4.sg-grind:foc c1.pl.sbj.pfv-int-grind c3.sg ‘Will they [grind]FOC it (or pound it)?’ (illness_100616_SO-DS)
The focus construction may also be used for sentence focus or predicate focus (focus on the verb + argument that is not the PSA or PP). A sentence focus interpretation is possible when the PSA is marked for focus and a predicate focus interpretation is possible when an argument that is not the PSA or a peripheral PP is marked for focus. An example of the latter can be seen in (31).
(31)
(One speaker is telling a story in which a boy riding a bicycle meets a girl. The listener asks a clarification question.) yɛ tsyɛ, gasɔ́ ὲέ-kpɛ c1.sg too bicycle:foc c1.sg.sbj.prog-put ‘Is she also [riding a bicycle]FOC ?’ (pear_100630_GoD-FB)
For the linking from semantics to syntax, the use of the focus construction means that a syntactic template with a pre-core slot will be chosen (see also Section 24.4.1). If the element to be placed in the pre-core slot is a non-PSA argument, the number of postverbal core slots will be reduced by one. PSA indexing on the verb is obligatory, so the preverbal core slot which contains the verb prefix cross-referencing the nominal PSA argument is always filled, irrespective of whether the PSA occurs in the pre-core slot or in its regular pre-core position.
24.8 Summary
In Part I of this grammatical sketch, we have discussed the grammar of Avatime simple clauses. Avatime is a language with a clear distinction between nouns and verbs, small lexical categories of adjectives, adverbs and adpositions and a large category of ideophones. It has a rather rigid constituent order, in which the PSA precedes the verb and other arguments follow the verb, followed by adverbial phrases. The PSA is indexed on the verb with head-marking. Question words and optionally focused elements occur in the pre-core slot.
Referential phrases can have either a noun or a postposition as their nucleus. If a noun heads the RP, it can be followed by a number of modifiers, most of which agree in noun class with the head noun. Possession is indicated by juxtaposition with possessor–possessed word order. When a postposition heads the RP, no modifiers are possible and the postposition is always preceded by another RP that functions like a possessor. Locative phrases are formed with the locative preposition ní.
Most operators of the layered structure of the clause are marked on the verb by the choice of PSA prefix out of several possible paradigms and/or a specialized prefix. The order of operators does not always follow the expected clausal-core-nuclear order. This may be due to the origin of some prefixes as verbs in serial verb constructions and to the change from an unmarked to a marked perfective.
PART II. The Syntax and Semantics of Serial Verb Constructions
In this part, we discuss the relation between syntax and semantics in serial verb constructions (SVCs). First, we describe the properties of SVCs in Avatime and how they are distinguished into syntactic subtypes (Section 24.9). We then discuss the semantic functions of Avatime SVCs and how they fit within the Interclausal Relations Hierarchy connecting syntactic structures and semantic functions (Section 24.10).
24.9 Serial Verb Constructions
Serial verb constructions are sequences of two or more finite verbs in one clause without overt marking for coordination or subordination. In Avatime, only the first verb is fully inflected. Subsequent verbs are either bare, as in example (32) or marked with a reduced agreement prefix, as in example (33). These reduced prefixes are specific to SVCs (for more information, see Defina Reference Defina2016b). All verbs share their PSA and may also share another argument, as is the case with example (33).
(32)
o-di ŋwὲ c1.sg.sbj.pfv-sit drink ‘S/he sits drinking.’
(33)
a-ŋwya ki-dítɔ a-kpέ c1.sg.sbj.pfv-throw c4.sg-thing:indf svm.c1.sg.pfv-put.in:loc ɔ-kà=ɛ c1.sg-father=def ‘She threw something to the father.’ (contrexp-28_120912)
Avatime SVCs can be divided into three subtypes: nuclear, core and sequential, based on a number of morphosyntactic differences. In what follows we briefly describe the properties of the different types of SVC and show how they can be analysed in RRG. For a more elaborate account, see Defina (Reference Defina2016b).
The different subtypes can be distinguished by examining their behaviour with operators at different levels. Negation and most aspect and mood markers are fused to some degree with the PSA agreement markers (see Section 24.4.2) and can only be marked on the first verb of an SVC. These markers always have broad scope over the whole SVC. Even the intentive, which is the only morphologically separate mood marker, can only occur with the first verb and scopes over the whole construction. This suggests that the verbs in all Avatime SVCs are linked below the clause level.
Aspect and aspectual adverbials are nuclear-level operators and can be used to distinguish a set of nuclear-level SVCs from the rest. Many SVCs allow aspectual adverbials between the verbs with narrow scope over a single verb, as can be seen in example (34). The recurrent aspect can also be used to mark each verb in these SVCs individually, as can be seen in example (35). Neither of these is possible with the nuclear-level SVCs. While aspectual adverbials are occasionally accepted between the verbs of these nuclear SVCs, they always have broad scope over the whole construction and the recurrent aspect can only be marked on the first verb, as can be seen in examples (36) and (37). This indicates that these SVCs are formed through nuclear co-subordination while the other Avatime SVCs are joined at a higher level.
(34)
e-bu àgbèlì=ye ì-dru=lè kóko kị́ Kwami c1.sg.pfv-remove cassava=def c2.pl-mound=def already give Kwami ‘He already dug cassava mounds for Kwami.’ (But may keep them for himself)
(35)
a. mà-zɛ̌-dzɛ Òholò a-wà à-xwὲ=na 1.sg.pfv-rec-go Ho svm.1.sg.pfv-work c3.pl-job=def ‘I was going (repeatedly) to Ho and working.’ b. mà-dzɛ Òholò a-zɛ̌-wà à-xwὲ=na 1.sg.pfv-go Ho svm.1.sg.pfv-rec-work c3.pl-job=def ‘I went to Ho and was working.’ (moved to Ho for some time)
(36)
ba-dị́ koko gu ku-nugu=yò c1.pl.pfv-sit already talk c5.sg-mouth=def ‘They already sat talking.’
(37)
a. ba-zɛ̌-dị́ ŋwὲ kù-gòda c1.pl.pfv-rec-sit drink c6.sg-palmwine ‘They were sitting drinking palmwine.’ b. *ba-dị́ zɛ̌-ŋwὲ kù-gòda c1.pl.pfv-sit rec-drink c6.sg-palmwine
Directional prefixes and locational and temporal adverbials are core-level operators and their scope properties distinguish the final two subtypes. Sequential SVCs can occur with distinct locational and temporal adverbials modifying each verb phrase, as can be seen in example (38). These SVCs also allow the directional prefix to occur on the second verb with narrow scope, as can be seen in example (39). This suggests these SVCs consist of distinct cores with their own peripheries and so we analyse them as core coordinations.
(38)
ma-tsà tomatoes=ye ní lị-vlɛ=lὲ 1.sg.sbj.pfv-cut tomatoes=def loc c3.sg-morning=def à-kpɛ ní kè-zi=a mὲ áblà svm.1.sg.pfv-put loc c6.sg-bowl=def inside now ‘I cut the tomatoes in the morning and put them in the bowl now.’
(39)
ma-kɔ̀ kà-wɛ=a zɛ-tsà ò-se=lò 1.sg.sbj.pfv-take c6.sg-axe=def it-cut c2.sg-tree=def ‘I took the axe to go cut the tree.’
The other non-nuclear SVCs also allow locational and temporal adverbials between the verbs. However, in these cases there can be only one adverbial and it has broad scope over the whole construction, as in example (40). This suggests these SVCs share a single periphery and are joined via core co-subordination. Nuclear SVCs do not allow locational or temporal adverbials to occur between the verbs, consistent with their nuclear juncture.
(40)
a. mà-dzɛ Òholò kivòe kị́ Akosua 1.sg.sbj-go Ho yesterday give Akosua ‘I went to Ho for Akosua yesterday.’ b. *mà-dzɛ Òholò kivòe kị́ Akosua òmonò 1.sg.sbj-go Ho yesterday give Akosua today
The distinguishing properties of these three subtypes of Avatime SVCs are summarized in Table 24.2.
Table 24.2 Characteristic properties of Avatime SVC subtypes
| Juncture–nexus type | Nuclear | Core | Sequential | |
|---|---|---|---|---|
| Nuclear co-subordination | Core co-subordination | Core coordination | ||
| Can aspectual adverbials occur | between verbs? | Marginally | Yes | Yes |
| with restricted scope? | No | Yes | Yes | |
| Can the recurrent occur | on subsequent verbs? | No | Yes | Yes |
| with restricted scope? | No | No | Yes | |
| Can directionals occur | on subsequent verbs? | Marginally | Yes | Yes |
| with restricted scope? | No | No | Yes | |
| Can locational or temporal adverbials occur | between verbs? | No | Yes | Yes |
| with restricted scope? | No | No | Yes | |
24.10 Semantic functions of SVCs and the Interclausal Relations Hierarchy
24.10.1 The Interclausal Relations Hierarchy
There are striking tendencies across languages in how semantic functions are expressed by syntactic constructions: closer semantic functions tend to be expressed by syntactic constructions that more closely resemble a single clause and have a stronger link between their elements, while looser semantic relationships tend to be expressed by syntactic constructions with a looser, or weaker, link between their elements. This is known as the Interclausal Relations Hierarchy, which was first proposed by Silverstein (Reference Silverstein and Dixon1976), in his study of split case marking in ergative languages, as well as by Givón (Reference Givón1980), in a more direct and detailed proposal. It has since been further extended and solidified within the theory of RRG (Foley and Van Valin Reference Foley and Van Valin1984; Van Valin Reference Van Valin2005). The hierarchy is presented in Figure 24.6.
The alignment of semantic functions and syntactic constructions is not one to one. Not all languages make use of all the listed syntactic construction types. Each construction type may express multiple semantic functions and, vice versa, a given semantic function may be expressible through several different constructions. This leads to rather complex instantiations of the interclausal relations hierarchies in individual languages (e.g. Casti Reference Casti2012; Kockelman Reference Kockelman2003) and a full consideration of the Avatime interclausal relations is beyond the current paper. However, Avatime SVCs and related semantic functions present some intriguing cases for this hierarchy, which are worth discussing here. The primary prediction of the Interclausal Relations Hierarchy is that the tightest syntactic construction used to express a semantic function should be tighter than or equal to the tightest syntactic construction used to express a looser semantic function. There are two cases involving SVCs where there is an apparent deviation from this prediction in Avatime. By examining how Avatime SVCs fit within and deviate from the predictions of this hierarchy, we can gain a better understanding of the hierarchy itself, Avatime SVCs, and their place within the wider ecology of Avatime grammar.
24.10.2 Semantic Functions of Avatime SVCs
In this section, we review the semantic functions that can be expressed using the three types of SVCs discussed in Section 24.9: nuclear co-subordination, core co-subordination and core coordination. We will then consider any alternative means of expressing these semantic functions in Avatime and review the relations between syntactic constructions and semantic functions in terms of the predictions of the Interclausal Relations Hierarchy.
The nuclear co-subordination SVCs are used to express modifying subevents such as the position, (example (32)), or manner (example (33)) in which an action is carried out. They are also used to express modification of the path of motion, as in example (41). These three semantic functions cluster neatly together in the Interclausal Relations Hierarchy and together with means form the set of ‘modifying subevents’ functions (Van Valin Reference Van Valin2005: 206). Manner can also be expressed using a core SVC, as discussed later.
(41)
be-ple e-ku ní Gbàdzɛmɛ c1.pl.sbj-descend svm-enter loc Gbadzeme ‘They descended into Gbadzeme.’ (Avatime-history_110905_BB_129)
As discussed in Section 24.5, it is also possible to express motion accompanying an action with a directional prefix in a simple clause, as in example (42).
(42)
a-zɛ-bàsị=blɔ bà-lị=à c1.sg.sbj.pfv-it-show=1.pl.obj c5.pl-palm.tree=def ‘He went to show us the palm trees.’ (Conv-ablorme_100715_SO-AS)
These directional prefix motion functions are a different type of motion semantic function from the one expressed by the nuclear SVCs. The nuclear SVCs further specify a path of motion. The directional prefixes express concomitant motion which is either towards or away from the deictic centre. This latter function is closer to the one described for the motion semantic function in the hierarchy presented by Van Valin (Reference Van Valin2005: 206).
The use of directional prefixes to express motion creates an apparent deviation from the predictions of the Interclausal Relations Hierarchy, since clauses with directional prefixes are closer to simple clauses than the nuclear SVCs used to express the tighter semantic function of manner. This deviation likely arises from the derivation of the directional prefixes from SVCs (see Section 24.4.2). If the directional prefixes were still verbs within nuclear co-subordinate SVCs, these semantic functions would be expressed by equally tight syntactic constructions and there would be no deviation from the hierarchy. It is also notable that there is a semantic split between the functions of the directional prefix and the motion nuclear SVCs where this deviation occurs. It could then be possible to split motion into two functions with accompanying motion positioned as closer than manner, and path of motion remaining where motion is currently (looser than manner). This would resolve the deviation in the hierarchy.
The core SVCs are used for adding and introducing or marking arguments. Examples (34) and (40) show the use of this type of SVC to add a beneficiary using the verb kí ‘give’. They can also be used to add instruments, manners or means, as in examples (43)–(45), respectively, all with the verb kɔ̀ ‘take’.
(43)
a-kɔ̀ kà-wɛ=a yài ò-se=lò c1.sg.sbj-take c6.sg-axe=def break c2.sg-tree=def ‘He used an axe to split the tree.’
(44)
a-kɔ̀ ku-siyeyome sὲ c1.sg.sbj.pfv-take c5.sg-anger leave ‘He left in anger.’
(45)
a-kɔ̀ ku-zò dzi ohonete c1.sg.sbj.pfv-take c5.sg-theft become rich.person ‘Through theft he became a rich man.’
These SVCs can also be used to mark or introduce theme arguments which are already called for by another verb in the SVC. For instance, in example (46) kɔ̀ ‘take’ is used to introduce the theme banana. However, as example (47) shows, kpɛ ‘put’ already licenses agent and theme arguments.
(46)
a-kɔ̀ kɔ̀ranti=ɛ kpɛ ní kà-sɔ=ya mὲ c1.sg.pfv-take banana=def put loc c6.sg-basket=def inside ‘S/he put the banana into the basket.’
(47)
a-kpɛ lì-kùto=lè c1.sg.pfv-put c3.sg-hat=def ‘S/he put the hat on.’
With the exceptions of manner and means, these functions are not included within the list of interclausal semantic functions. These argument-adding functions are relations between a predicate and its dependents and are typically intraclausal. However, the fact that they can be expressed by a combination of verbs in Avatime suggests they could be added to the hierarchy. They bear clear similarities to the causative [1] function – to directly bring about a state of affairs through an event or action: for example, paint the table red or break the bowl. This function also involves relations between a predicate and its arguments and is sometimes expressed by a single verbal lexeme. Given the overlap with manner and means functions and the general role of these functions, it would make sense for them to be added immediately below means in the hierarchy.
The core coordination SVCs are used to express sequential actions, as in examples (35), (38) and (39). These SVCs can always be paraphrased by coordinated clauses, as shown in example (48).
(48)
a. ma-tsà tomatoes=ye à-kpɛ ní 1.sg.sbj.pfv-cut tomatoes=def svm.1.sg.pfv-put loc kè-zi=a mὲ c6.sg-bowl=def inside ‘I cut the tomatoes and put them in the bowl.’ b. ma-tsà tomatoes=ye lɛ̌ mà-kpɛ ní 1.sg.sbj.pfv-cut tomatoes=def and 1.sg.sbj.pfv-put loc kè-zi=a mὲ c6.sg-bowl=def inside ‘I cut the tomatoes and put them in the bowl.’
This use of core coordination to express sequential action is another instance of a deviation from the predictions of the Interclausal Relations Hierarchy. Sequential actions are among the loosest types of semantic relations. Many tighter semantic relations, such as circumstance and reason, are expressed by the looser syntactic relations of clausal subordination and coordination. For instance, the circumstantial in example (49), where a preposed subordinate clause marked with the clause linkage marker xé indicates the point in time at which the main clause happens.
(49)
xé kụ̀ị́-sɔ pɔ́=ɛ kị-tá-halị̀ sị̀-wa=sὲ clm 1.pl.sbj.prog-sow finish=cm 1.pl.sbj.pfv-int-gather c7-weed=def petee ní ɔ̀-nyɔ=nɔ̀ mὲ all loc c2.sg-farm=def inside ‘When we finish sowing, we gather all the weeds from the farm.’ (rice_100613_EN-MM)
It is notable that, like with the previous deviation, a semantic distinction not recognized in the hierarchy plays a role. Whereas all sequential actions can be expressed by clausal coordination, the sequential actions that can be expressed by core coordination SVCs must be performed by the same actor and must constitute a single culturally relevant unit, generally with an overarching goal (see Defina Reference Defina2016b). If this type of sequential action relation were distinguished as a separate, tighter, semantic relation category in the hierarchy, the deviation would be resolved.
24.11 Summary
This part has surveyed the semantic functions and syntactic properties of SVCs in Avatime, showing how they constitute three distinct juncture–nexus types, see Table 24.3.
Table 24.3 Semantic functions of SVCs in each subtype
| Subtype | Semantic functions |
|---|---|
| Nuclear co-subordination | Modifying (manner, motion (complex path), and position) |
| Core co-subordination | Argument adding and theme marking Modifying (manner and means) |
| Core coordination | Sequential actions (same actor, coherent unit) |
Each subtype of Avatime SVC poses a question for the Interclausal Relations Hierarchy and its predicted links between syntactic constructions and semantic functions. The argument-adding and marking functions of the core co-subordination SVCs are not included within the hierarchy, but should arguably be listed below the modifying functions. The grammaticalization of SVCs indicating motion away from or towards the deictic centre has created a deviation from the hierarchy’s predictions, with the tighter directional prefix construction expressing a function that is looser than the manner function expressed by nuclear SVCs. The use of core coordinating SVCs to combine certain types of sequential actions presents another deviation. Both of these deviations relate only to more restricted semantic functions and can be resolved through separating the specific semantic functions expressed with the tighter syntactic construction from the more general motion and sequential action semantic functions.
This review of Avatime SVCs within the context of the Interclausal Relations Hierarchy provides a clearer understanding of the internal structure of Avatime SVCs and complex sentences more generally. It also suggests that the list of semantic functions expressed by syntactic linkage is far from final and may need further language-specific or general modifications or additions. This is not an entirely new finding, and the point was made by Van Valin (Reference Van Valin2005: 211) that the syntactic side of this hierarchy is better understood than the semantic side, which should not be taken as an exhaustive list of semantic functions.
Abbreviations
We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/glossing-rules.php), with the following additions:
- A
actor macrorole
- ARG
argument
- ASP
aspect
- CLM
clause linkage marker
- CONT
continuative aspect
- CORER
RP core
- DCA
direct core argument
- DEIC
deictic
- DN
DCA-non-macrorole
- DS
different subject
- DUn
DCA-undergoer
- DV
dependent verb
- EVQ
event quantification
- HORT
hortatory
- HABP
habitual past
- IF
illocutionary force operator
- IU
information unit
- INGR
ingressive
- IRIT
irregular iterative aspect
- IT
iterative aspect
- IVC
impersonal verb construction
- LS
logical structure
- MP
modifier phrase
- NEGF
negative future tense
- NMR
non-macrorole
- NUC
clause nucleus
- NUCR
RP nucleus
- NUM
numeral
- PNG
Papua New Guinea
- PP
postpositional phrase
- PoCS
post-core slot
- PoDP
post-detached position
- PrCS
pre-core slot
- PrDP
pre-detached position
- PRED
predicate
- PRO
pronoun
- PRSP
prospective tense
- PSA
privileged syntactic argument
- PSD
possessed
- PSR
possessor
- QNT
quantifier
- R(EAL)
realis status
- RG
regret particle
- REMP
remote past tense
- RP
reference phrase
- RPIP
RP initial position
- RRG
Role and Reference Grammar
- SEQ
sequential event
- SIM
simultaneous event
- SR
switch-reference
- SS
same subject
- STA
status operator
- SVC
serial verb construction
- TNS
tense operator
- TODP
today’s past tense
- U
undergoer macrorole
- YESTP
yesterday’s past tense
- -
morpheme break
- < >
infix
- =
clitic break
- ~
reduplication
- &
sequential conjunction
- {&}
seq conj in complex sentence
- ∧
overlapping conjunction
- {∧}
overlap conj in complex sentence
- *
ungrammatical
25.1 Introduction
Amele is a Papuan language spoken in Madang Province, Papua New Guinea (PNG). The Amele people inhabit an area of approximately 120 square kilometres between the Gum and Gogol rivers just south of the town of Madang in PNG. The area extends from the coast to about 14 kilometres inland. Amele is the largest of the Gum family of languages (Z’graggen Reference Z’graggen1975: 13) with a population of approximately 5,300 speakers (Lewis et al. Reference Lewis, Simons and Fennig2014).1
This chapter provides a linguistic description of the Amele language from a Role and Reference Grammar (RRG) perspective. Section 25.2 describes the basic syntax of the language and Section 25.3 examines serial verb constructions and the switch-reference system.Footnote *
25.2 Common Topics
Nominative-accusative agreement is suffixed to the verb stem and up to four core arguments can be marked on the verb. There are only two major lexical categories, nouns and verbs, with very little overlap between them. Alternative undergoer selection may be made for ditransitive verbs. There is no passive construction in the language and the only choice for privileged syntactic argument (PSA) is [S, AT]. Focus may be expressed morphologically and by incorporation of modifier elements into the verb word.
25.2.1 Basic Clause Patterns
Typologically, Amele has head-last syntax with OV order and postpositions.2 The language is also head-marking, and core arguments of the verb can be marked on the verb by cross-reference agreement. PSA agreement has one basic type of morphology and direct core argument (DCA) agreement has another basic type of morphology.3 The argument agreement morphology follows a nominative-accusative pattern. Therefore, the PSA argument agreement morphology is called nominative (nom) and the DCA agreement morphology is called accusative (acc).
The basic syntax of the clause is given in Figure 25.1. The PSA is the first element in the clause and the DCA occurs immediately preceding the verb. After the PSA is the slot for any adjunct reference phrase (RP) or postpositional phrase (PP). Following that is the slot where any argument-adjuncts occur. Typically, no more than two or three non-verbal elements are expressed in any one clause. However, up to four arguments can be encoded on the verb.

Figure 25.1 Basic syntax of the clause in Amele
There is also an order in which the different types of argument can be marked on the verb. Up to three arguments can be marked on the verb and the order is given in (1). The rightmost argument is the PSA agreement and this is obligatory on the finite verb. This is the nominative agreement morphology. The DCA-undergoer (DUn, where Un is undergoer) argument agreement attaches directly to the verb stem. This is obligatory for some verbs, optional for other verbs, and not allowed for others. Many verbs also allow optional DCA-non-macrorole (DN) argument agreement to be marked and this requires the applicative (applied object) marker. The form of the DUn and DN marking is the accusative agreement morphology in each case. It is possible to have a maximum of two DCA arguments marked on the verb, either DUn + DN or DN + DN. The linear order of argument marking on the verb in (1) is thus a mirror image of the ordering of arguments in the clause in Figure 25.1.
(1)
Order of arguments marked on the verb: verb stem ±DUn.Agr ±APPL+DN.Agr (±APPL+DN.Agr) +PSA.Agr
Verb agreement is formalized as in (2). nom agreement only applies to the finite verb form. acc agreement can apply to both the finite and infinitive verb forms, for example hel-ad-ec [throw-3pl.acc-inf] ‘to throw them’, hel-i-ad-ec [throw-appl-3pl.acc-inf] ‘to throw to them’. See Roberts (Reference Roberts1996, Reference Roberts and Newman1997c).
(2)
Verb agreement in Amele: nom agreement cross-references the highest-ranking macrorole argument in the finite verb. acc agreement can cross-reference any other macrorole or non-macrorole argument in the finite or non-finite verb.4
25.2.1.1 M-Intransitive Verbs
M-intransitive verbs have only one macrorole core argument. The main types of M-intransitive verbs are stative verbs and motion verbs. There is no copular verb like ‘be’ in Amele. Instead, the language uses several posture verbs in certain stative predications (i.e. bilec ‘to sit (down)’, nijec ‘to lie (down)’, tawec ‘to stand (up)’).5 As illustrated in Table 25.1, these posture verbs have a state function and an activity function. For the state function the single macrorole is undergoer. For the activity function the single macrorole is actor. An example of each function is given in (3).
(3)
Posture verb examples: a. Jo jobon gemo taw-ena. State: posture house village middle stand-3sg.nom.prs be-village middle′ (stand′ (3sg [jo])) ‘The house stands in the middle of the village.’ b. Dana ben taw-im-ei ma-g-en. Activity: assume posture man big stand-ss.seq-3sg.nom tell-1pl.acc-3sg.nom.remp do′ (3sg [dana], [stand′ (3sg [dana])]) … ‘The headman stood up and spoke to us.’ c. Cam qila gagadic=ca taw-ei-a. State: attributive sun today strength=add stand-3sg.nom-todp today′ (be′ (3sg [cam], [strong′])) ‘The sun is strong today.’
Table 25.1 M-intransitive posture verbs
| bilec | ‘sit’ | sit′ (x) | state: posture |
| ‘sit (down)’ | do′ (x [sit′ (x)]) | activity: assume posture | |
| ‘be’ | be′ (x) | state: attributive, identificational, specificational | |
| nijec | ‘lie’ | lie′ (x) | state: posture |
| ‘lie (down)’ | do′ (x [lie′ (x)]) | activity: assume posture | |
| ‘be’ | be′ (x) | state: attributive, identificational, specificational | |
| tawec | ‘stand’ | stand′ (x) | state: posture |
| ‘stand (up)’ | do′ (x [stand′ (x)]) | activity: assume posture | |
| ‘be’ | be′ (x) | state: attributive, identificational, specificational |
There are two basic types of M-intransitive motion verb: those that specify [move+path] and those that specify [move+manner]. [move+path] motion verbs are verb-framed and [move+manner] motion verbs are satellite-framed (Talmy Reference Talmy and Shopen2007). Each type of motion verb is M-intransitive and the single argument is actor in each case. However, there is a syntactic difference between these types of motion verb. The [move+path] motion verb allows an allative argument to be expressed either with a =ca ‘towards’ PP, as in (4a), or with an applied object marked on the verb, as in (4b). The applied object in (4b) is a non-macrorole direct core argument. The [move+manner] motion verb, on the other hand, does not allow an allative argument to be expressed, as illustrated in (5).
(4)
(5)
Allative argument disallowed with [move+manner] motion verbs: a. *Qa uqa=ca cob-oi-a. dog 3sg=towards walk-3sg.nom-todp b. *Qa cob-i-t-oi-a. dog walk-appl-3sg.acc-3sg.nom-todp
Other verbs that are M-intransitive include ededec ‘gleam’ do′ (x, [gleam′ (x)]), asalec ‘laugh’ do′ (x, [laugh′ (x)]), busuec ‘fart’ seml fart′ (x), silolec ‘ooze’ do′ (x, [ooze′ (x)]), fojec ‘vomit’ do′ (x, [vomit′ (x)]), tatiec ‘look up’ do′ (x, [look.up′ (x)]), bodoec ‘soften’ become soft′ (x), sanan mec ‘start’ ingr start′ (x).
Impersonal verbs also have to be treated as M-intransitive. A typical example of an impersonal verb construction (IVC) is given in (6). There is an optional free pronoun ija ‘1sg’ which corresponds to the subject pronoun in the English translation. However, in the Amele form this pronoun is cross-referenced on the verb with acc morphology. The nom agreement always codes 3sg in an IVC but the reference is unspecified. The agreement is dummy, neutral agreement which is part of the syntactic template for the construction and there is no actor argument with these IVCs. Such IVCs normally express a physiological or psychological experience and the 1sg.acc agreement refers to the experiencer argument.
(6)
Impersonal verb construction: (Ija) cucui-t-ei-a. (1sg) fear-1sg.acc-3sg.nom-todp feel′ (1sg, [afraid′]) ‘I am afraid.’
25.2.1.2 M-Transitive Verbs
M-transitive verbs have two macrorole core arguments: actor and undergoer. Amele has M-transitive state verbs, such as fec ‘see’ see′ (x, (y)), doc ‘know’ know′ (x, (y)), gawec ‘want’ want′ (x, y), meleec ‘believe’ believe′ (x, (y)), cucuiec ‘fear’ fear′ (x, (y)). With some M-transitive non-state verbs, expression of the undergoer argument on the verb is optional, with others this is obligatory. (7) illustrates an active verb, j-ec [eat-inf] ‘to eat’, with optional DUn marking on the verb. In (7a) this verb has the logical structure (LS) do′ (x, [eat′ (x, (y))]) with an optional second predicate argument. In (7b) this verb has the LS do′ (dana, [eat′ (dana, ha)]) with the second predicate argument position filled. This argument is realized in the syntax by the RP ha ‘sugarcane’. Since ha is neither animate nor a count noun, in this context it does not trigger acc agreement on the verb. The predicate is, therefore, an activity with a non-specific object. In (7c), on the other hand, ho ‘pigs’ is animate and a count noun and therefore triggers acc agreement ‑ad ‘them’ on the verbs qoc ‘hit’ and jec ‘eat’.6 In this case the object is also specific and ‘eat’ is an active accomplishment.
(7)
Active verb with optional DUn marking: a. Dana eu j-egi-na. man that eat-3pl.nom-prs do′ (3pl [dana], [eat′ (3pl [dana], (Ø))]) ‘Those men are eating.’ b. Dana eu ha j-ein. man that sugarcane eat-3pl.nom.remp do′ (3pl [dana], [eat′ (3pl [dana], ha)]) ‘Those men ate sugarcane.’ c. Dana eu age ho a-q-i je-ad-ein. man that 3pl pig 3pl.acc-hit-dv eat-3pl.acc-3pl.nom.remp [do′ (3pl [dana], [eat′ (3pl [dana], 3pl [ho])]) ∧ proc consume′ (3pl [dana], 3pl [ho])] & [ingr consumed′ (3pl [ho])] ‘Those men killed the pigs and ate them.’
The M-transitive verb cesul-d-oc [help-3sg.acc-inf] ‘to help him/her’ in (8) has obligatory DUn marking. In (8) the DUn marking ‑t ‘1sg.acc’ on the verb expresses the obligatory second argument. The 3pl.nom argument is actor and the 1sg.acc argument is undergoer.
(8)
Active verb with obligatory DUn marking: Age cesul-t-eig-a. 3pl help-1sg.acc-3pl.nom-todp do′ (3pl, [help′ (3pl, 1sg]) ‘They helped me.’
25.2.1.3 Three-Argument Verbs
Three-argument verbs have three core arguments in their LS. The ditransitive verb ihac-d-oc [show-3sg.acc-inf] ‘to show him/her/it’ is illustrated in (9). The LS of this verb is [do′ (x, Ø)] cause [become see′ (y, z)]. In (9a) the perceiver argument 1sg is selected as undergoer and is realized on the verb as 1sg.acc agreement attached directly to the verb stem. Because the perceiver argument has been selected as undergoer, there is no way of encoding the perceived argument, ho eu ‘that pig’, on the verb. In (9b) ho eu ‘those pigs’ is selected as undergoer and is realized on the verb as 3pl.acc agreement. In this case, the 1sg.acc perceiver is also marked on the verb as a DN applied object argument. (9c) shows that all three core arguments can be expressed as agreement marking on the verb alone. Without any clarifying RPs in the clause, (9d) is ambiguous between the first LS where 1sg is the perceiver argument and the second LS where 1sg is the perceived argument – although, the first interpretation would be deemed the more likely.
(9)
3-argument verb: a. (Uqa) ho eu ihac-t-ei-a. (3sg) pig that show-1sg.acc-3sg.nom-todp [do′ (3sg, Ø)] cause [become see′ (1sg, ho)] ‘He showed me that pig.’ b. Ho eu ihac-ad-i-t-ei-a. pig that show-3pl.acc-appl-1sg.acc-3sg.nom-todp [do′ (3sg, Ø)] cause [become see′ (1sg, 3pl [ho])] ‘He showed those pigs to me.’ c. Ihac-ad-i-t-ei-a. show-3pl.acc-appl-1sg.acc-3sg.nom-todp [do′ (3sg, Ø)] cause [become see′ (1sg, 3pl)] ‘He showed them to me.’ d. Ihac-t-ei-a. show-1sg.acc-3sg.nom-todp [do′ (3sg, Ø)] cause [become see′ (1sg, Ø)] or [do′ (3sg, Ø)] cause [become see′ (Ø, 1sg)] ‘He showed me something (unspecified).’ or ‘He showed me to someone (unspecified).’
25.2.1.4 A Four-Argument Verb
There is one verb which allows an additional fourth core argument to be marked. This is ‘give’, as illustrated in (10).7 Here the verb stem is realized as acc morphology which agrees with the recipient. The meaning of ‘give’ is constructional rather than lexical, and so there would be a constructional schema for it. Furthermore, Latrouite and Van Valin (2014) argue that constructional meaning should be represented differently from lexical meaning; the constructional meaning for ‘give’ would be [do′ (x, Ø)] CAUSE [INGR have′ (y, z)]. There is an infinitive form for each person and number (e.g. itec ‘to give me’, ihec ‘to give you (sg)’, utec ‘to give him/her’, etc.) The regular verb agreement morphology attaches to this derived stem. The first argument (recipient) of have′ (x, y) in ‘give’ logical structure is assigned the undergoer macrorole by virtue of being the argument encoded in the verb stem itself. Thus, the recipient argument is always the undergoer and there is no alternative construction where the second argument (theme) of have′ (x, y) is the undergoer. The ‘∧ feel.negatively.affected′ (z)’ represents the malefactive ‘on me’ argument. It is the LS proposed for the Japanese adversative passive (Imai Reference Imai1998; Toratani Reference Toratani2002).
(10)
The verb ‘give’: Eeta=nu ut-ad-i-t-ag-a? what=for 3sg.acc-3pl.acc-appl-1sg.acc-2sg.nom-todp be-for′ (eeta, [do′ (2sg, Ø)] cause [INGR have′ (3sg, 3pl)] ∧ feel.negatively.affected′ (1sg)) ‘Why did you give him them on me?’
25.2.1.5 Non-Verbal Predicates
Amele does not have non-verbal predicates as such. With descriptive and equational clauses a posture verb is used where it is necessary to specify a clausal operator category. However, if such specifications are not required then the posture verb is omitted. Examples are given in (11).
(11)
Stative non-verbal predicates: a. Mel eu hag=ca. Attributive boy that sickness=with ‘That boy is sick.’ a′. Mel eu hag=ca nij-en. boy that sickness=with lie-3sg.nom.remp ‘That boy was sick.’ be′ (mel, [sick′]) b. Misag uqa iwal-ad-ec. Identificational Misag 3sg teach-3pl.acc-nmlz ‘Misag is a teacher.’ b′. Misag uqa iwal-ad-ec bil-ol-oi. Misag 3sg teach-3pl.acc-nmlz sit-habp-3sg.nom ‘Misag used to be a teacher.’ be′ (Misag, [a teacher′]) c. Danben age wool-ad-ec. Specificational Danben 3pl surpass-3pl.acc-nmlz ‘Danben (village) are the winners.’ c′. Danben age wool-ad-ec nij-ein. Danben 3pl surpass-3pl.acc-nmlz lie-3pl.nom.remp ‘Danben (village) were the winners.’ be′ (Danben, [wooladec]) d. Mei ija=na cof-t-ec. Equational father.1sg.psr 1sg=of supervise-1sg.acc-nmlz ‘My father is my boss.’ d′. Mei ija=na cof-t-ec bil-en. father.1sg.psr 1sg=of supervise-1sg.acc-nmlz sit-3sg.nom.remp ‘My father was my boss.’ equate′ ([have.as.orientation.kin′ (1sg, mei-)], coftec)
25.2.2 Lexical Categories
Whereas English has four major word categories of verb, noun, adjective and adverb, which can be distinguished on morphological and syntactic grounds, Amele only has two: verbs and nouns. Words that function as nominal modifiers (adjectives in English) and verbal modifiers (adverbs in English) cannot be distinguished on morphological or syntactic grounds in Amele from words that function as the head of an RP (see Roberts Reference Roberts1987: 154–156, 158 for details). Therefore they are all categorized as ‘nouns’. In English, many words can belong to multiple lexical categories, such as break (verb and noun), red (noun and adjective), fast (adjective and adverb), round (adjective, adverb, noun, verb, preposition). By contrast, in Amele there is very little overlap between members of the verb class and those of the noun class. Only three instances of overlap have been observed: cad ‘enemy’ (noun) and cadec ‘to fight’ (verb), mele ‘truth’ (noun) and meleec ‘to believe’ (verb), and cucuiec ‘to fear’ (verb) and cucuian ‘his/her fear’ (inalienably possessed noun).
25.2.2.1 Verbs
There are six morphosyntactic forms of the verb: regular verb, impersonal verb, reciprocal verb, light verb, serial verb, and dependent switch-reference verb. Serial verbs are described in Section 25.3.1. They are not marked for tense or for nom agreement. Dependent switch-reference verbs are described in Section 25.3.2. They are not marked for tense either. Amele also has an interrogative verb adec ‘to when’, as illustrated in (12).
(12)
Interrogative verb: A~ad-eb uqa h-ugi-an? dur~when-3sg.nom.ds.sim.irr 3sg come-3sg.nom-fut ‘Whenever will he come?’
25.2.2.2 Nouns
Amele has three morphological classes of nouns: regular nouns with uninflected stems, inalienably possessed nouns with possessor agreement inflection, and deverbal nouns, derived from the infinitive form of a verb. Inalienably possessed nouns are described in Roberts (Reference Roberts1987: 171–175, 2015b). They comprise kin terms,8 body-part terms and personal attribute terms. Deverbal nouns are formed from the infinitive form of the verb. The infinitive suffix, ‑ec/‑oc, on the verb functions as a nominalizing suffix on the noun (e.g. cob-oc [walk-inf] ‘to walk’ or [walk-nmlz] ‘a walk’).
25.2.2.3 Postpositions
Postpositions are an important minor word class in Amele (Roberts Reference Roberts1987: 160–161). They are clitic words which must attach to a preceding host element and they may not be stranded (Roberts Reference Roberts1991c, Reference Roberts1992, 1996). They can be predicative or non-predicative, and we return to this in Section 25.2.4.2.
25.2.3 The Layered Structure of the Clause
In this section syntactic templates and operators are described and illustrated.
25.2.3.1 Syntactic Inventory
The basic structure of the clause in Amele is given in Figure 25.2 and Amele-specific clause-internal linear precedence rules are given in (13). Amele has pro-drop verb agreement which functions as the argument (ARG) of the predicate. The only obligatory element in the active clause is the verb.
(13)
Amele-specific clause-internal linear precedence rules: a. XP* > CORE (verb final) b. RP (PSA) > RP/PP (Adjunct) > RP/PP (Arg-Adjunct) > RP (DCA) > NUC

Figure 25.2 The layered structure of the clause
Templates for optional syntactic structures are given in Figure 25.3. The pre-detached position (PrDP) is for topical established information and the post-detached position (PoDP) is for additional information, such as clarification. The pre-core slot (PrCS) is for focal new information or ‘heavy’ constructions, such as a nominal modified by a relative clause. The post-core slot (PoCS) is for postposed clausal elements.

Figure 25.3 Optional syntactic structures
PrDP example
Example (14) illustrates a clause-external topic caja eu ‘that woman’ in the pre-detached position. There is a resumptive pronoun uqa ‘she’ in the clause for the argument in the PrDP.
(14)
Clause-external topic in PrDP: Caja eu, uqa me qee. woman that 3sg good not ‘That woman, she is no good.’
PoDP example
In (15) the RP dana eu ‘that man’ in the PoDP adds clarifying information to the identity of the PSA eu ‘that’ in the clause. This pronoun is resumptive for the argument in the PoDP.
(15)
Clarifying information in PoDP: Eu uqa jeje-g=ca m-en, dana eu. that 3sg voice-3sg.psr=with put-3sg.nom.remp man eu ‘He got his voice back, that man.’
PrCS examples
In the unmarked form, temporal adjuncts (RP or PP) occur after the PSA RP, as in (16a). Alternatively, the temporal adjunct can be placed in the PrCS as focal new information, as in (16b).
(16)
Temporal adjunct in PrCS: a. Ija cum ceta gug filfil cabi=na ceh-ig-an. 1sg yesterday yam kind different garden=in plant-1sg.nom-yestp ‘I planted different types of yam in the garden yesterday.’ b. Cum ija ceta gug filfil cabi=na ceh-ig-an. yesterday 1sg yam kind different garden=in plant-1sg.nom-yestp ‘Yesterday I planted different types of yam in the garden.’
Arguments and adjuncts with a relative clause are typically placed in the PrCS, as shown in (17).
(17)
Locative adjunct with a relative clause in the PrCS: Cudun dan ben taw-ena eu=na ija jahun-d-ug-a. place fig big stand-3sg.nom.prs that=at 1sg hide-3sg.acc-1sg.nom-todp ‘I hid it at the place where the big fig tree is.’
PoCS example
In (18) there is an instrumental argument-adjunct PP dubin=na ‘with stalk’ and a goal argument-adjunct PP camac ta=na ‘into the sago scrapings’ in the final clause. The second argument-adjunct PP is located in the PoCS.
(18)
Goal argument-adjunct PP in the PoCS: Age wa wet-i dubin=na basec-d-ogi-na camac ta=na. 3pl water scoop-dv stalk=with pour-3sg.acc-3pl.nom-prs sago scrapings=in ‘They scoop up the water and pour it in with the stalk into the sago scrapings.’
25.2.3.2 Operators
Illocutionary force
Illocutionary force (IF) is the outermost operator and the form of expression of different types of IF in Amele is illustrated in (19)–(23). Statement (19) is expressed by a declarative sentence. Command (20) is expressed by the imperative form of the verb. The imperative is identical in form to the today’s past tense form. A yes/no question (21) is expressed with the sentence-final question particle =fo. An information question (22) is expressed by an interrogative word, such as in ‘who’, eeta ‘what’, cel ‘which’. Exhortation (23) is expressed by the hortative form of the verb.
(19)
IF: statement Age aluh=na bel-ein.9 3pl mountain=to go.nsg-3pl.nom.remp ‘They went to the mountain.’
(20)
IF: command Age aluh=na bel-eig-a! 2pl mountain=to go.nsg-2pl.nom-imp ‘Go to the mountain!’
(21)
IF: yes/no question Age aluh=na bel-ein=fo? 3pl mountain=to go.nsg-3pl.nom.remp=q ‘Did they go to the mountain?’
(22)
IF: information question In aluh=na nu-i-an? who.sg mountain=to go-3sg.nom-yestp ‘Who went to the mountain (yesterday)?’
(23)
IF: exhortation Ege aluh=na bel-ec=nu! 1pl mountain=to go.nsg-inf=hort ‘Let us go to the mountain!’
Evidentials
Amele does not have evidentials.
Status
Amele expresses realis/irrealis status on the switch-reference verb (see Roberts Reference Roberts1990, Reference Roberts1994 for a wider study of this phenomenon in Papuan languages). The simultaneous ds nom agreement has different forms depending on the realis/irrealis status of the clausal operator category marked on the final verb in the linked clauses, traditionally called a clause chain in Papuan linguistics. The clausal operator categories with realis status are: present tense, today’s past tense, yesterday’s past tense, remote past tense, habitual past tense/aspect, negative past tense. The clausal operator categories with irrealis status are: future tense, prospective (about to) tense, negative future tense, and imperative, prohibitive, hortative, optative, counterfactual, apprehensive IF categories. An example of realis status is given in (24a) and an example of irrealis status is given in (24b).
(24)
a. Marking of realis status: Ho bu~busal-en age q-oig-a. pig dur~run out-3sg.nom.ds.sim.r 3pl hit-3pl.nom.todp ‘They killed the pig as it ran out.’ b. Marking of irrealis status: Ho bu~busal-eb age q-oqag-an. pig dur~run out-3sg.nom.ds.sim.irr 3pl hit-3pl.nom-fut ‘They will kill the pig as it runs out.’
Tense
The tense categories in Amele are metrical. Metrical tenses mark degrees of temporal remoteness from the deictic centre. Amele has a present tense (3a) and three degrees of past tense: today’s past tense (3c), yesterday’s past tense (22), and remote past tense (3b), as well as a negative past tense and a past habitual tense/aspect (11b′).10 The language has a regular future tense (25), a negative future tense (38), and a prospective (about to/intentional) (26) tense.11 The event referred to in the prospective tense is nearer to the deictic centre than the event referred to in the regular future tense. It is therefore metrical.
(25)
Future tense: Age cabi=na bel-oqag-an. 3pl going=to go.nsg-3pl.nom-fut ‘They will go.’
(26)
Prospective tense: Age cabi=na bel-oqag-a bil-i taw-eig-a. 3pl going=to go.nsg-3pl.nom-prsp aux-dv stand-3pl.nom-todp ‘They stood about to go.’
Modality
Categories of deontic modality, such as ability, permission and obligation are expressed lexically in Amele (Roberts Reference Roberts1987, 2001).
Event quantification
Event quantification (Roberts Reference Roberts2015a) is marked on the verb by distributive inflection that is homonymous12 with the acc agreement, ‑ad ‘plural’ and ‑al ‘dual’. Some examples are given in (27). In (27a) the motion verb belec ‘to go’ is intransitive and the ‑ad ‘plural’ marker indicates a multiple event of each woman going her own way. In (27b) and (c) calec ‘to arrive’ is intransitive. The ‑ad ‘plural’ marker in (27b) indicates multiple events of men arriving, while in (27c) the ‑al ‘dual’ marker indicates two events of men arriving.
(27)
Plural and dual event quantification: a. Caja age bud-u bel-ad-ein. woman 3pl disperse-dv go.nsg-distr.pl-3pl.nom.remp ‘The women dispersed in all directions / each to her own place.’ b. Dana age cal-ad-ein. man 3pl arrive-distr.pl -3pl.nom.remp ‘The men all arrived.’ c. Dana ale cal-al-esin. man 3du arrive-distr.du-3du.nom.remp ‘The men both arrived.’
The EVQ marker ‑ad can also indicate an exclusive action, as in (28).
(28)
Ija saen cecelac sum-i-h-ig-a qee=nu 1sg time long wait-appl-2sg.acc-1sg.nom-todp not=for ija cu~cul-h-i l-im-ig nu-ad-ig-a=da. 1sg dur~leave-2sg.acc-dv go-ss.seq-1sg.nom go-excl-1sg.nom-todp=rg ‘I waited for you for a long time but in vain. So regretfully I left and went without you.’
Aspect
There are three types of aspect that can be expressed by reduplicating some part of the verb word: durative aspect, regular and irregular iterative aspect.
Durative aspect is marked on the simultaneous ss/ds verb and conveys the idea that there is an extended temporal overlap of events. Without this marking the temporal overlap is punctiliar. Durative aspect is expressed by different kinds of CV~ or V~ reduplication of the verb stem (Roberts Reference Roberts1991a). See (24a, b), (28), for examples of CV~ reduplication and (12) for an example of V~ reduplication. With some verbs, the reduplicative (C)V~ marking of durative aspect applies to the nom agreement suffixation rather than to the verb stem, as illustrated by (29). If there is accusative marking on the verb then the reduplicative marking for durative aspect applies there, as in (30).
(29)
Durative aspect marked on nom agreement: Co-Ø a-e~en lips-3sg.psr open-dur~3sg.nom.ds.sim.r ija dunuh meci-d-ug-a. 1sg inside observe-3sg.acc-1sg.nom-todp ‘While he opened his mouth I looked inside.’
(30)
Durative aspect marked on acc agreement: Age eu cunug ihac-te~t-eig 3pl that all show-dur~1sg.acc-3pl.nom.ss.sim sa-t-ein. explain-1sg.acc-3pl.nom-remp ‘As they showed me everything they explained it to me (lit. explained me).’
The meaning of the regular iterative is a repeated, regular action. This aspect is expressed by rightward reduplication of the whole stem if the verb does not have an acc marker, otherwise the acc marker is reduplicated either in place of, or in addition to the reduplication of the verb stem (Roberts Reference Roberts1987: 252–256). For some minimal stem verbs, such as l-ec ‘to go’, the reduplicated stem is of the serial verb form (e.g. li~li-ec ‘to go repeatedly’; see Section 25.3.2). An example of regular iterative aspect is given in (31a).
(31)
a. Regular iterative: Gow-ec eu fale~fale-ei-a. light-nmlz that flash~it-3sg.nom-todp 〈IF DECL 〈TNS TODP 〈ASP IT seml flash′ (gowec) 〉〉〉 ‘That light flashed repeatedly.’ b. Irregular iterative: Gow-ec eu fale~fule-ei-a. light-nmlz that flash~irit-3sg.nom-todp 〈IF DECL 〈TNS TODP 〈ASP IRIT seml flash′ (gowec) 〉〉〉 ‘That light flashed intermittently.’
The meaning of the irregular iterative is a repeated action that is irregular in some way (i.e. haphazard, spasmodic, intermittent, etc.). This form involves reduplication of the verb stem but with a disharmonic vowel change in the reduplicated formant. There are eight types of disharmonic vowel change possibilities, which are determined by phonological factors.13 An example of irregular iterative is given in (31b).
The aspectual notions of continuative and completive can be expressed periphrastically with a serial verb construction (see examples (45) and (46) in section 25.3.1).
25.2.4 The Structure of RPs and PPs
25.2.4.1 RP Structure
The various structures of the RP are given in (32). The head noun in the RP can be a regular noun, an inalienably possessed noun or a deverbal noun. The modifier can be a noun, such as dana caub [man white] ‘white man’, dana me [man good] ‘good man’, or a modifier (a small class of words that only function as modifiers), such as dana bahic [man very] ‘real man’, dana qee [man not] ‘not a man’, or a phrase, such as dana mel iwal-ad-ec [man child teach-3pl.acc-nmlz] ‘school teacher’.
(32)
Reference phrase structures: a. noun ± modifier ± (numeric) quantifier ± deictic/indefinite article ± universal quantifier b. interrogative deictic + noun c. PP/RP + noun
A mass noun can be modified by a general quantifier, such as leih ‘some’ or geh ‘much’. A count noun can be modified by a numeric quantifier, such as lecis ‘two’ or cijed ‘three’. Amele has a pental counting system and the cardinal numbers are: osahic/osol ‘one’, lecis ‘two’, cijed ‘three’ wal oso ‘four’, ebum oso ‘one hand (≈10)’. Ordinal numbers can be formed by adding the nominalizing suffix ‑doc to the numeral (e.g. osahic-doc ‘first’, lecis-doc ‘second’, cijed-doc ‘third’). There are also a number of relationship terms that can function as ordinal numbers, for instance matu ‘firstborn/first’, milum ‘secondborn/second’, subig ‘lastborn/last’.
The quantifier can be followed by a deictic element, such as eu ‘that’ or ceheleg ‘up there’, or the indefinite article oso ‘a/one’. The article can co-occur with a deictic element, as in dana eu oso [man that art] ‘one of those men’. The universal quantifier cunug ‘all’ occurs at the end of the RP, as in jo nag cijed eu cunug [house small three that all] ‘all those three small houses’.
With respect to (32a), the head noun can be questioned with an interrogative, such as eeta ‘what’ or in ‘who’, for example Hina ija-in in? [2sg name-2sg.psr who.sg] ‘What (lit. who) is your name?’ The numeric quantifier can be questioned with an interrogative, such as ganic ‘how much/many’, for example jo ganic? [house how many] ‘how many houses?’, saab ganic? [food how much] ‘how much food?’. However, when the deictic is questioned, the interrogative form, such as cel ‘which’, is placed in the RPIP, which is the pre-core slot in the RP, as in (32b). Thus the interrogative form is cel dana [which man] ‘which man?’.
With respect to (32c), the preceding PP/RP expresses the possessor of the noun or a specification of the noun. (33) illustrates the difference in how alienable and inalienable possession is expressed.14 Alienable possession in (33a) is expressed with a possessive PP preceding the possessed noun. The possessive PP functions as an argument of the possessed noun, which is the head of the nucleus of the possessive noun phrase. The possessive PP is therefore non-predicative. Inalienable possession in (33b) is expressed by suffixal agreement in person (first, second or third) and number (singular, dual or plural) with the possessor noun phrase. The possessor need not be expressed by an overt RP or pronoun. Thus the possessor agreement functions as the argument (ARG) of the inalienably possessed head noun.15 Semantically, inalienably possessed nouns are kinship terms, body-part terms or personal attributes. In (33b) cuduni means ‘my personal place’ or ‘the place that belongs to me in some way’, whereas in (33a) ijana cudun simply means ‘my place’ without the connotations of personal ownership.
(33)
a. Alienable possession: ija=na cudun ‘my place’ 1sg=of place have′ (1sg, cudun) b. Inalienable possession: (ija) cudu-ni ‘my (personal) place’ 1sg place-1sg.psr have.as.attribute′ (1sg, cudu-)
When the modifier follows the head noun it has an attributive function. In (34a) jo us nijec describes a house where people are sleeping. When the modifier precedes the head noun it has a specificational function. In (34b) us nijec describes the type of house. Adjectival-cum-nominal modifiers are represented as predicates which take the item in the NUCR as an argument (underlined).
(34)
a. Modifier with attributive function: jo us nij-ec ‘sleeping house (i.e. a house that is asleep)’ house sleep lie-nmlz be′ (jo, [asleep′]) b. Modifier with specificational function: us nij-ec jo ‘sleeping house (i.e. a house for sleeping in)’ sleep lie-nmlz house be′ (jo, [PURP sleep′])
The basic structure of the RP is given in Figure 25.4. Most modifying elements in the RP can occur on their own as referring expressions. Some examples are given in (35). Thus, such modifiers are represented in the syntactic structure as well as in the operator projection.
(35)
a. Ben/eu/oso h-ona. big/that/one come-3sg.nom.prs ‘The big (man) is coming.’/ ‘That (man) is coming.’/ ‘Someone is coming.’ b. Cijed h-ogi-na. three come-3pl.nom-prs ‘The three (men) are coming.’

Figure 25.4 The basic structure of the RP in Amele
The structure of the alienably possessed PP and inalienably possessed RP are given in Figure 25.5. Ija cebinami ‘my brother’ is the inalienably possessed RP with cebinami as the head noun. Caja ‘woman’ is the head noun of the alienably possessed RP and ija cebinamina functions as the possessor PP. It is clear from the reference that ija ‘1sg’ is a woman.

Figure 25.5 The structure of the possessive RP in Amele
Amele uses a pronominal copy strategy where an RP is immediately followed by a personal pronoun. The pronoun indicates the person and number of the RP, as illustrated in (36). The pronoun functions as appositive to the RP and since it is not possible for other clausal elements to occur between the RP and the pronoun, it forms a constituent with the RP.
(36)
Pronominal copy strategy: Caja eu uqa/ale/age qaj-ei/-esi/-eig-a. woman that 3sg/3du/3pl cry-3sg.nom/3du.nom/3pl.nom-todp ‘That/those (du)/those (pl) woman/women cried.’
25.2.4.2 PP Structure
Amele only has postpositional phrases (PPs). They can be predicative or non-predicative. Predicative PPs occur with temporal or locative adjuncts (e.g. Mande=na [Monday=on] ‘on Monday’, be-on′ (Mande, x) or jo=na [house=in] ‘in the house’ be-in′ (jo, x)), or with goal, path, source, or instrument argument-adjuncts, as described in Section 25.2.7. The possessor-marking postposition =na ‘of’, described in Section 25.2.4.1, is non-predicative.
25.2.5 Constraints on A and U Selection
With a ditransitive verb like ihacdoc ‘to show him/her’ in (9), either the perceiver x-argument in see′ (x, y) or the perceived argument y can be selected as undergoer. The latter is the unmarked choice. When the perceived argument is selected as undergoer it is coded on the verb with acc marking attached directly to the verb stem and the perceiver argument can also be marked as a non-macrorole applied object. However, if the perceiver is selected as undergoer it is coded on the verb with acc marking attached directly to the verb stem and the perceived argument cannot be marked on the verb.
The verb ‘give’ is exceptional in that the undergoer argument functions as the stem of the verb. The first argument of have′ (x, y) (recipient) in ‘give’ LS is thus the only choice for undergoer assignment and there is no alternative argument-marking construction where the second argument of have′ (x, y) (theme) is the undergoer. An example of this was given in (10).
When the goal argument in an argument-adjunct PP is a person, an applied object construction may be used as an alternative expression, as illustrated in (4). This is also the case for the source adjunct PP argument, and the benefactive adjunct PP argument, illustrated in (37). For the malefactive argument shown in (38) there is only the applicative form and there is no corresponding argument-adjunct PP form.
(37)
Benefactive argument-adjunct applied object: a. Uqa age=nu jo eu ceh-al-ei-a. 3sg 2pl=for house that plant-3du.acc-3sg.nom-todp ‘He built both those (two) houses for you (plural).’ b. Uqa jo eu ceh-al-i-ad-ei-a. 3sg house that plant-3du.acc-appl-2pl.acc-3sg.nom-todp ‘He built both those (two) houses for you (plural).’ c. [do′ (3sg, [build′ (3sg, 3du [jo])]) ∧ proc create′ (3du [jo])] & ingr exist′ (3du [jo]) purp [become have′ (2pl, 3du [jo])]
(38)
Malefactive argument-adjunct: Ene cain salal-i-t-ag-aun. here proh slide-appl-1sg.acc-2sg.nom-negf NOT [be-here′ (2sg, [do′ (2sg [slide′ (2sg)]] ∧ feel.negatively.affected′ (1sg)] ‘Don’t slide here and annoy me (coll. don’t slide on me here).’
25.2.6 Postposition Assignment
The only non-predicative postposition assigned by rule is =na, which is assigned to the possessor in the alienable possession RP (cf. 33a).
(39)
Rule assigning =na ‘possessor’: Assign =na to the x-argument in the RP logical structure segment: have′ (x, y)
25.2.7 Coding of Adjuncts
Temporal adjuncts can be an RP, such as cum ‘yesterday’, uqadec ‘tomorrow’, qila ‘today/now’, cel saen ‘what time’, or a PP, such as Mande=na ‘on Monday’, Ogas=na ‘in August’. Locative adjuncts can be an RP, such as ene ‘here’, ono ‘there’, ceheleg ‘up there’, ana ‘where’, or a PP, such as jo=na [house=in] ‘in the house’, na=na [tree=on] ‘on the tree’, eeta=na [what=in] ‘in what’.
25.2.8 PSA Alignment(s)
In Amele the only choice for PSA with a transitive verb is [AT] as there is no passive construction in the language. However, IVCs have exceptional PSA verb agreement coding. Although the undergoer is the only DCA in an IVC and therefore the highest-ranking DCA, it is coded as acc, contrary to (2). This is because 3sg.nom is assigned by default in an IVC.
25.2.9 Information Structure
The pre-detached position is outside the potential focus domain. The information structure of (14) is illustrated in Figure 25.6. The RP caja eu ‘that woman’ is a clause-external topic in the pre-detached position and, as a topic, it is outside of the potential focus domain of the proposition. Uqa ‘she’ is the resumptive pronoun in the clause. The potential focus domain (dotted line) is the attributive clause uqa me qee ‘she (is) no good’ and the actual focus domain (solid line) is the predication me qee ‘(is) no good’.

Figure 25.6 The PrDP and the potential focus domain
Similarly, the post-detached position is outside the potential focus domain. The information structure of (15) is illustrated in Figure 25.7. The RP dana eu ‘that man’ is in the post-detached position and clarifies the identity of eu ‘that’ in the main clause. As clarifying information, it is outside of the potential focus domain of the proposition. The potential focus domain is the statement eu uqa jejegca men ‘he got his voice back’. Here eu ‘that (man)’ functions as a resumptive pronoun for dana eu ‘that man’. The actual focus domain is the predication uqa jejegca men ‘got his voice back’.

Figure 25.7 The PoDP and the potential focus domain
Amele has a range of particles that can attach to the end of the sentence which qualify the proposition, such as da ‘but, however, nevertheless’, do ‘be encouraged, let’s do it’, fo ‘yes/no question’ fa ‘dubitive question, maybe’, ijom ‘certainly’, le ‘permission granted’, and mo ‘supplication, pleading’ (Roberts Reference Roberts1990). Some of the sentence particles, such as the question particles, fo and fa, can occur either at the end of the sentence or they can be focused on a particular constituent in the sentence. In (40a) the yes/no question particle, fo, occurs at the end of the sentence. It has scope over the whole proposition and therefore expresses sentence focus. In (40b–c) the question particle is focused on a particular sentence constituent and expresses marked narrow focus on that constituent. (40d) illustrates the use of the question word in ‘who’. It is ungrammatical to have the question particle fo at the end of the sentence in this case since the question word expresses narrow focus and the particle at the end of the sentence would express sentence focus. However, as illustrated in (40e), it is possible to focus on the question word with the question particle. This expresses marked narrow focus.
(40)
Focus expressed morphologically: a. Ija aide-ni cabi=na nu-i-a=fo? [unmarked sentence focus] 1sg wife-1sg.psr garden=to go-3sg.nom-todp=q ‘Did my wife go to the garden?’ b. Ija aide-ni=fo cabi=na nu-i-a? [marked narrow focus] 1sg wife-1sg.psr=q garden=to go-3sg.nom-todp ‘Was it my wife that went to the garden?’ c. Ija aide-ni cabi=na=fo nu-i-a? [marked narrow focus] 1sg wife-1sg.psr garden=to=q go-3sg.nom-todp ‘Was it to the garden that my wife went?’ d. In cabi=na nu-i-a(*=fo)? [unmarked narrow focus] who.sg garden=to go-3sg.nom-todp(=q) ‘Who went to the garden?’ e. In=fo cabi=na nu-i-a? [marked narrow focus] who.sg=q garden=to go-3sg.nom-todp ‘Who is it that went to the garden?’
It is possible to incorporate certain modifier constituents into the verb word to express emphatic focus for the purposes of contrast or correction, or closer specification, for instance. For example, the intensifier bahic ‘very, must, really’ can occur either preceding the verb, as in (41a), or be incorporated into the verb between the verb stem and the verb suffixation, as in (41b). Other modifiers which can be focused in this way are the limiters dih ‘just’ and himec ‘only’, and the negators qee ‘not’ and cain ‘don’t’.
(41)
Emphatic focus: a. Age Anut bina-n bahic sul-eig-a! 2pl God fame-3sg.psr really lift up-2pl.nom-imp ‘Really praise God!’ b. Age Anut bina-n sul<bahic>eig-a! 2pl God fame-3sg.psr lift up<really>2pl.nom-imp ‘REALLY praise God!’
25.2.10 Linking
In this section the semantics-to-syntax linking for some two-place and three-place predicates is illustrated. Figure 25.8 shows the semantics-to-syntax linking in a transitive sentence with acc (DUn) marking. Here 3pl [dana] is the first argument in the active predicate LS do′ (3pl, [eat′ (3pl [dana], 3pl [ho])]), and is assigned the actor macrorole. 3pl [ho] ‘pigs’ is the second argument in this LS and also the only argument in the stative predicate LS ingr consumed′ (3pl [ho]). This argument is assigned the undergoer macrorole. The actor argument, 3pl [dana], is selected as PSA and coded as nom agreement on the verb. The undergoer argument, 3pl [ho], is coded as acc agreement on the verb in order to indicate the plurality of the animate referent.

Figure 25.8 Semantics-to-syntax linking in a transitive sentence with acc marking
Figure 25.9 illustrates the semantics-to-syntax linking in an IVC. The 1sg argument in feel′ (1sg, [afraid′]) is assigned the undergoer macrorole according to the Actor-Undergoer Hierarchy (AUH). The IVC is coded as 3sg.nom agreement by default. There is therefore no linking between this argument marking in the syntax and the logical structure.

Figure 25.9 Semantics-to-syntax linking in an impersonal verb construction
Figure 25.10 illustrates the semantics-to-syntax linking for a three-place predicate, ihacdoc ‘to show him’. The leftmost argument, 3sg ‘he’, is assigned the actor macrorole and the rightmost argument, 3pl [ho] ‘pigs’, is assigned the undergoer macrorole. The actor is selected as PSA and coded as nom agreement on the verb. The undergoer is coded as a DUn by acc agreement attaching directly to the verb stem. The non-macrorole core argument, 1sg ‘me’, is coded as a DN by the applicative marker and acc agreement marking on the verb.

Figure 25.10 Semantics-to-syntax linking in a ditransitive predicate
25.3 Language-Specific Topics
The language-specific topics discussed in this chapter are serial verb constructions in Section 25.3.1 and switch-reference in Section 25.3.2.
25.3.1 Serial Verb Constructions
Kroeger (Reference Kroeger2004: 229–230) shows that prototypical serial verb constructions (SVCs) have the following syntactic and semantic properties:16
(42)
Characteristic properties of SVCs: a. A prototypical SVC contains two or more morphologically independent verbs within the same clause, neither of which is an auxiliary. b. There are no conjunctions or other overt markers of subordination or coordination separating the two verbs. c. The serial verbs belong to a single intonation contour, with no pause separating them. d. The entire SVC refers to a single (possibly complex) event. e. A true SVC may contain only one specification for tense, aspect, modality, negation, etc., though these features are sometimes redundantly marked on both verbs.17 f. The two verbs in the SVC share at least one semantic argument. g. Obligatory non-coreference: a true SVC will not contain two overt NPs which refer to the same argument. h. A prototypical SVC contains only one grammatical subject.
SVCs in Amele do not have all of the prototypical properties listed in (42). In an SVC the non-final verb in the series is marked with ‑i or ‑u. Verbs with an ‑ec infinitive form take ‑i, while verbs with an ‑oc infinitive form take ‑u. The ‑i/‑u marking indicates the verb is dependent (dv).18 This contravenes principle (42b) as the dv marker indicates a nexus relationship. The PSAs of the verbs in an SVC typically have the same referent.19 There are two basic types of SVC. In one type, illustrated in (43a), the verbs are in a cosubordinate relationship. In the other type, illustrated in (44a), the verbs are in a superordinate–subordinate relationship. By comparison, the verb manimei in (43b) is fully inflected with switch-reference morphology and is therefore not a serial verb. Similarly, ehimeig in (44b) is not a serial verb.
(43)
Cosubordinate SVC: a. Caja uqa ceta man-i j-ei-a. woman 3sg yam roast-dv eat-3sg.nom-todp 〈IF DECL 〈STA REAL 〈TNS TODP [[do′ (3sg [caja], Ø)] cause [become roasted′ (ceta)] {&} [[do′ (3sg [caja], [eat′ (3sg [caja], ceta) ∧ proc consume′ (3sg [caja], ceta)] & [ingr consumed′ (ceta)]])] 〉〉〉 ‘The woman roasted and ate yam.’20 b. Caja uqa ceta man-im-ei j-ei-a. woman 3sg yam roast-ss.seq-3sg.nom eat-3sg.nom-todp 〈IF DECL 〈STA REAL 〈TNS TODP [do′ (3sg [caja], Ø)] cause [become roasted′ (ceta)] {&} [[do′ (3sg [caja], [eat′ (3sg [caja], ceta) ∧ proc consume′ (3sg [caja], ceta)] & [ingr consumed′ (ceta)]])] 〉〉〉 ‘The woman roasted yam and ate it.’
(44)
Subordinate SVC: a. Age ja eh-i n-eig-a. 3pl firewood take-dv come down-3pl.nom-todp 〈IF DECL 〈STA REAL 〈TNS TODP 〈DIR COME DOWN [[do′ (3pl, Ø)] cause [do′ (3pl, [move.away.from.ref.point′ (ja)])] 〉〉〉〉 ‘They brought firewood down.’ b. Age ja eh-im-eig n-eig-a. 3pl firewood take-ss.seq-3pl.nom come down-3pl.nom-todp 〈IF DECL 〈STA REAL 〈TNS TODP [do′ (3pl, Ø)] cause [do′ (3pl, [move.away.from.ref.point′ (ja)])] {&} [do′ (3pl, [move.down.to.ref.point′ (3pl)])] 〉〉〉 ‘They brought firewood and came down.’
In (43a) the verbs mani ‘roast’ and jeia ‘she ate’ describe a series of closely related events. Compare (43b) in which manimei is the fully inflected ss.seq form. Here the roasting and eating are interpreted as separate consecutive events and they are separate clauses in the syntactic representation. The arguments caja ‘woman’ and ceta ‘yam’ in (43a) are shared in the LS for the two predicates. In the syntax for (43a) illustrated in Figure 25.11 the argument caja is represented by an RP in the clause and by 3sg.nom agreement on the verb jeia ‘she ate’. The argument ceta is represented as a core RP argument of mani ‘roast’. It is not coded on the verb by acc agreement as it is inanimate and a mass noun. The SVC is a cosubordinate CORE [[CORE] [CORE]] juncture because it is not possible to have temporal or locative clausal adjuncts applying separately to either core. By comparison, in (43b) it would be possible to qualify jeia ‘she ate’ with the temporal adjunct hibna ‘later’ because it is a separate clause.
(44a) describes a single event of ‘take down’. The verb n- ‘come down’ is a directional modifier of the verb ehi ‘take’ and is coded in the LS as such. There is no shared argument between the verbs as n- is represented in the LS as a directional operator. (44a) therefore contravenes principle (42f). Compare (44b), where ehimeig is fully inflected for ss.seq. Here the event ehimeig ‘they took’ is linked to the event neiga ‘they came down’ by {&} in the LS. They are separate consecutive events in the LS and are separate clauses in the syntactic structure. In the syntactic structure for (44a) in Figure 25.12 the verb n- is in a subordinate relationship to ehi. This is ad-nuclear subordination. The 3pl.nom argument agreement ‑eig attaches to n- as the final verb in the series.

Figure 25.11 SVC with cosubordinate core juncture

Figure 25.12 SVC with nuclear subordination
With respect to the characteristic SVC properties detailed in (42), a particular SVC may not necessarily belong to a single intonation contour (cf. 42c). Compare the sentence in Figure 25.13. The li ‘go’ verb is a directional modifier of ehi ‘take’. The first two cores form one intonational unit and the last core is a separate intonational unit. However, the dv coding marks this as describing a series of linked events which should be interpreted as a unified complex event.

Figure 25.13 Extended coordinate SVC
The subordinate SVC has a range of modifying functions in addition to expressing directionality. Examples are given in (45)–(49). In each case the modifying verb follows the verb that is modified. (45)–(46) express aspect periphrastically. (45) illustrates how continuative aspect is expressed by a posture verb: bilec ‘to sit’, nijec ‘to lie’, tawec ‘to stand’, and (46) illustrates how completive aspect is expressed with the verb hedoc ‘to finish it’. (47)–(49) illustrate some other common forms of this type of subordinate SVC.
(45)
Continuative aspect expressed with a posture verb: Age nu-i bil-egi-na. 3pl go-dv sit-3pl.nom-prs ‘They go continuously.’
(46)
Completive aspect expressed with the verb hedoc ‘to finish it’: Age jo ceh-i he-d-oig-a. 3pl house plant-dv finish-3sg.acc-3pl.nom-todp ‘They finished building the house.
(47)
Investigative modification expressed with the verb fec ‘to see’: Uqa wehuc j-i f-ei-a. 3sg soup eat-dv see-3sg.nom-todp ‘She tasted the soup.’
(48)
Superlative modification expressed with the verb cuhadoc ‘to excel’: Caja uqa ola-Ø cus-i cuha-d-on. woman 3sg face-3sg.psr scrub-dv excel-3sg.acc-3sg.nom.remp ‘The woman cleaned her face well.’
(49)
Enumerative modification expressed with a numeral:21 Q-u lecis-d-oig-a. hit-dv two-3sg.acc-3pl.nom-todp ‘They hit it twice.’
25.3.2 Switch-Reference
Switch-reference (SR) in Amele is judged to be a local syntactic device for monitoring the referentiality of PSA arguments between adjacent clauses as to whether they have identical or non-identical reference (Roberts (Reference Roberts, Aikhenvald and Dixon2017). Dependent SR verbs occur most commonly in a clause chain.22 A clause chain comprises a string of linked clauses with only the final clause in the chain marked for clausal operator categories, such as tense or illocutionary force. The non-final clauses are dependent on the final clause for clausal operator designation. The operator-dependent clauses are also typically marked with switch-reference morphology. Amele has two basic types of ss/ds morphology. One codes sequential events and the other codes simultaneous overlap of events. The sequential verb is marked with either ‑im ‘ss.seq’ or ‑ec/‑oc ‘ds.seq’ followed by nom agreement morphology. The simultaneous verb is marked with nom agreement morphology which indicates either ss.sim or ds.sim. The ds.sim morphology is further divided into that which indicates realis status and that which indicates irrealis status of the tense or IF category marked on the final clause in the clause chain.23 The ds.sim verb agrees in status value with the status value of the final clause. ss/ds clauses are most commonly in a cosubordinate relationship with other clauses (see Chapter 13). However, ss/ds clauses can also have a subordinate function.
25.3.2.1 Cosubordinate Switch-Reference Clauses
A typical clause chain of four linked clauses in a cosubordinate relationship is illustrated in (50). The verbs neceb and tobocomin are marked for ds.seq and the verb series sumudi bibiligin is marked for ds.sim.r since the tense category of the final verb is realis status. The final verb in the chain is belowan and this verb is marked for yesterday’s past tense and declarative illocutionary force. Sequential events are indicated by ‘{&}’ in the LS and simultaneous events by ‘{∧}’. The clauses … sumudi bibiligin, neceb, tobocomin and belowan in (50) are in a cosubordinate relationship as the tense, status and IF categories marked on belowan have scope over all the clauses in the chain. Sumudi bibiligin is a continuative SVC. The verb bibiligin modifies sumudi as a continuative event and the durative marking on bi~biligin indicates the ‘waiting’ event overlapped for a period of time with the ‘came down’ (neceb) event.
(50)
Dependent switch-reference verbs: 1. Ija Malolo uqa=na ka jic anag ono=nu 1sg Malolo 3sg=of car road mother there=for sum-ud-i bi~bil-igin wait-3sg.acc-dv dur~sit-1sg.nom.ds.sim.r 2. n-ec-eb come down-ds.seq-3sg.nom 3. tob-oc-omin ascend-ds.seq-1sg.nom 4. bel-ow-an. go.nsg-1du.nom-yestp 〈IF DECL 〈STA REAL 〈TNS YESTP 〈ASP CONT 〈ASP DUR be-loc′ (ono, [be-loc′ (jic anag, do′ (1sg, [wait′ (1sg, become be-loc′ (jic anag, [have′ (Malolo, ka))])〉〉 {∧} do′ (3sg, [come.down′ (3sg)]) {&} do′ (1sg, [ascend′ (1sg)]) {&} do′ (1du, [move.away.from.ref.point′ (1du)])〉〉〉 ‘While I waited for Malolo’s car there on the main road, he came down, I climbed in and off we (du) went.’
Note that the final clause in a clause chain need not be finite. In (51) the purpose clause ho bubusaleb qoc ‘to kill the pig as it runs out’ is a clause chain with qoc ‘to kill’ as the final clause. The verb bubusaleb in the dependent clause is marked for ds simultaneous irrealis nom agreement morphology as the understood PSA of qoc is ‘those men’, and qoc, being infinitive, has irrealis status.
(51)
Infinitival final clause: Dana eu ho bu~busal-eb q-oc=nu h-oig-a. man that pig dur~run out-3sg.nom.ds.sim.irr hit-inf=for come-3pl.nom-todp ‘Those men came to kill the pig as it runs out.’
A diagram of the syntax of (50) is given in Figure 25.14. The clausal operators of tense, status and IF apply to all the dependent clauses, whereas the core and nuclear operators apply only to a particular verb. The ss/ds markers are treated as clause linkage markers. The rules for how the SR system works are set out in (52) and (53). In terms of Comrie’s (Reference Comrie, Arnold, Atkinson, Durand, Grover and Sadler1989) different types of reference-tracking systems, SR in Amele is local and interclausal.
(52) Protocol for selecting switch-reference controlling clause:
For a dependent clause in a cosubordinate relationship ({&} or {∧}), select the next cosubordinate clause as the controlling clause for switch-reference marking.
(53) Amele switch-reference coreferentiality rule:
Check the pivot of the dependent clause against the pivot of the controlling clause for coreferentiality of identity. If the referent is identical, mark ss; if not, mark ds.
Figure 25.14 Cosubordinate clause chain structure, example (50)
With respect to (53), there are complexities in determining referential identity.24 For example, Amele SR is asymmetrical where there is referential overlap between the pivot referents of the controlling clause and the pivot referents of the dependent clause. Where the pivot referent(s) of the controlling clause is (are) properly included in the set of pivot referents of the dependent clause then ss is marked. However, when the pivot referent(s) of the dependent clause is (are) properly included in the set of pivot referent(s) of the controlling clause then ds must be marked. For example, in (50) 1sg.nom in [3] is properly included in 1du.nom in [4] and in this case ds must be marked. If it was the other way around (‘we (du) climbed in and off I went’) and 1sg.nom was the controlling referent and 1du.nom was the dependent referent then ss would be marked.25
Stirling (Reference Stirling1993) suggests that SR in Amele tracks events rather than the main participant(s) in the event. As already mentioned, Amele has asymmetrical ss/ds marking when there is referential overlap between the PSA of the controlling clause and the PSA of the marked clause. When a plural number of participants controls a singular number of participants then ds must be marked. In (54) a multiple (plural) event budu beladeiga with multiple participants controls a singular dependent event age cajimeig with multiple participants. If the SR tracked events, as suggested by Stirling (Reference Stirling1993), then it would be expected that a move from a singular event to a multiple event should trigger ds marking. However, ss is marked. This shows that the Amele SR system tracks the coreferentiality of participants across clauses (both plural in this case) rather than the coreferentiality of events.
(54)
Event quantification and SR marking: Dana caja=ca age caj-im-eig bud-u bel-ad-eig-a. man woman=add 3pl arise-ss.seq-3pl.nom disperse-dv go-distr-3pl.nom-todp ‘The people got up and dispersed in all directions (multiple events).’
Further evidence that SR in Amele tracks participants rather than events is provided by the reciprocal verb form, as illustrated in (55). Here a pair of ds-marked verbs functions as the predicate in the nucleus.26 The 3pl.nom agreement at the end of the predicate refers to the whole group of women (caja). The ds-marked verbs express the notion of one womani hitting another womank and this being reciprocated by womank hitting womani. Thus the ds marking refers to individual participants in the overall event of women hitting.
(55)
Reciprocal verb and SR marking: Caja q-oc-ob q-oc-ob egi-na. woman hit-ds.seq-3sg.nom hit-ds.seq-3sg.nom 3pl.nom-prs 〈IF DECL 〈STA REAL 〈TNS PRS do′ (3pl [caja], [seml do′ (3sg [caja]i, [hit′ (3sg [caja]i, 3sg [caja]k)]) {&} seml do′ (3sg [caja]k, [hit′ (3sg [caja]k, 3sg [caja]i)])]〉〉〉 ‘The women are hitting each other.’
25.3.2.2 Subordinate Switch-Reference Clauses
Switch-reference clauses can also have a subordinate function. The following functions are described below:
core subordinate object clause of perception verbs
An SR clause may function as the object of a perception verb.27 In (56a) the second argument of the perception verb perceive′ is a proposition become arise′ (dedeman) ‘a smell had arisen’. Therefore in the syntax dedeman waseceb occupies the DUn position in the clause between the PSA caja eu ‘that woman’ and the verb don ‘she perceived’. In Roberts (Reference Roberts1988a) it is shown that overtly coordinate clauses in Amele, such as a qa ‘but’ clause, cannot be embedded within another clause. Therefore dedeman waseceb cannot be in a coordinate relationship with don. It is core subordinate, as illustrated in Figure 25.15. Since become arise′ (dedeman) is an accomplishment (ending in a result state) it is interpreted as being in a sequential relationship with the matrix predicate perceive′, that is, the smell arose before the woman perceived it. Therefore, the verb waseceb is marked for ds.seq.

Figure 25.15 Core subordinate DUn clause, example (56a)
Similarly, in (56b) the second argument of the perception predicate see′ is the proposition lie′ (3sg [ma susul] ‘the taro peelings are lying (on the ground)’. Consequently, ma susul eu ninijen occupies the DUn position in the clause between the PSA mala uqa ‘chicken he’ and the verb fen ‘he saw’. Here ma susul eu ninijen is in a core subordinate relationship with fen. Because the proposition lie′ (3sg [ma susul]) is a state it is interpreted as occurring simultaneously with see′. The verb ninijen is therefore marked as ds.sim.r.
(56)
Core subordinate DUn clause of perception verbs: a. Caja eu dedeman was-ec-eb d-on. woman that smell arise-ds.seq-3sg.nom perceive-3sg.nom.remp 〈IF DECL 〈STA REAL 〈TNS REMP perceive′ (3sg [caja], [become arise′ (3sg [dedeman])])〉〉〉 ‘That woman perceived that a smell had arisen.’ b. Mala uqa ma susul eu ni~nij-en f-en. chicken 3sg taro peelings that dur~lie-3sg.nom.ds.sim.r see-3sg.nom.remp 〈IF DECL 〈STA REAL 〈TNS REMP see′ (3sg [mala], [〈ASP DUR lie′ (3sg [ma susul])〉])〉〉〉 ‘Chicken saw those taro peelings lying (there).’
A switch-reference clause can function as a modifier of a following clause. (57) is a sentence containing five clauses. Clauses [1, 3, 5] describe the mainline events, while clauses [2, 4] provide background information. Clause [2] says ‘when the yams ripened …’ and clause [4] says ‘when the yams dried …’. The secondary nature of clause [2] is indicated by the ss marking on clause [1]. In this case, clause [3] is the controlling clause for the ss marking on clause [1]. Clause [3] is also the controlling clause for the ds marking on clause [2]. Thus clause [2] functions as a temporal modifier to clause [3]. Similarly, the ss marking on clause [3] is controlled by clause [5] and clause [4], marked for ds, functions as a temporal modifier to clause [5]. There are thus layers of cosubordination, which can be schematically represented as in (57a).
(57)
Layered cosubordination in a clause chain: a. [Clause-1]SS [[Clause-2]DS Clause-3]SS [[Clause-4]DS Clause-5] 1. Ceta bahim m-i he-d-um-ei yam store put-dv finish-3sg.acc-ss.seq-3sg.nom 2. ceta wal m-ec-eb yam ripe put-ds.seq-3sg.nom 3. ceta eu hun-im-ei yam that bore-ss.seq-3sg.nom 4. gulden h-oc-ob dried come-ds.seq-3sg.nom 5. ceta bahim=na tac-en. yam store=in fill-3sg.nom.remp b. 〈IF DECL 〈STA REAL 〈TNS REMP 〈ASP COMPL [do′ (3sgi, [make′ (3sgi, ceta bahim)]) ∧ proc create′ (ceta bahim)] & ingr exist′ (ceta bahim)〉 {&} [[ become ripened′ (ceta)] {&} [do′ (3sgi, [dig.up′ (3sgi, ceta)]]] {&}[[become dried′ (ceta)] {&} [do′ (3sgi, Ø)] cause [become be-in′ (ceta bahim, Ø)] cause [ingr full′ (ceta bahim)]])〉〉〉 ‘He finished making the yam store and, when the yams had ripened, he dug up those yams and when they dried he put them in the yam store.’
The syntactic structure of (57) is diagrammed in Figure 25.16. The modifying clauses [2] and [4] are in a cosubordinate relationship to the following matrix clause. The matrix clause is the controlling clause of the ‘sub-cosubordinate’ clause for the purposes of switch-reference marking. The PSA in the sub-cosubordinate clause is different to the PSA in the matrix clause in each case, so ds is marked.
Note that this produces an anomalous SR marking between clause [1] and the immediately following clause [2], and between clause [3] and the immediately following clause [4]. The verb in clause [1] is marked for ss when the PSA of clause [2] is different to the PSA in clause [1]. The verb in clause [3] is also marked for ss when the PSA of clause [4] is different to the PSA in clause [3]. This shows that the switch-reference marking system takes account of the layers of cosubordination. For clauses signalling the mainline events, such as clauses [1, 3, 5] in (57) and the four clauses in (50) the SR-marking system selects the next cosubordinate clause as the controlling clause for SR marking. For an SR clause indicating background information, such as [2, 4] in (57), it is interpreted being relevant to the following ‘mainline event’ expressing clause, and the SR-marking system selects that clause as the controlling clause for SR marking. A translation of (57) directly reflecting its structure would be ‘He finished making the yam store; the yams ripened, and then he dug up those yams; the yams dried, and then he put them in the yam store’.

Figure 25.16 Layered cosubordination in (57)
The protasis (condition clause) in a conditional sentence modifies the apodosis (consequence clause). The condition clause in the protasis can be a switch-reference clause. ss.seq marking for the condition clause is ‑if/‑uf followed by the regular ss.seq nom agreement morphology. For all other ss/ds marking the conjunction fi ‘if’ is cliticized to the switch-reference verb. An example of an ss condition clause is given in (58a) and of a ds condition clause in (58b). In (58a) the ss marking actually indicates the certainty of the consequence. In both cases, the ss/ds condition clause can be postposed after the consequence clause and it is ad-clausal subordination.
(58)
Ad-clausal subordination in a conditional clause: a. Qee j-i he-d-uf-eg qaga-h-ig-en. not eat-dv finish-3sg.acc-ss.cond-2sg.nom kill-2sg.acc-1sg.nom-fut ‘If you don’t eat it all (lit. finish) I will kill you.’ b. Ija ja hud-ec-emin=fi uqa saab man-igi-an. 1sg fire open-ds.seq-1sg.nom=if 3sg food cook-3sg.nom-fut ‘If I light the fire she will cook the food.’
The fact that switch-reference in Amele occurs in subordinate clauses as well as cosubordinate clauses requires a revision of the protocol for selecting the controlling clause; (52) has therefore been modified as follows.
(59)
Protocol for selecting switch-reference controlling clause: 1. For a dependent clause in a cosubordinate relationship, select the next cosubordinate clause as the controlling clause for switch-reference marking. 2. For a dependent clause in a subordinate relationship, select the superordinate matrix clause as the controlling clause for switch-reference marking.
25.3.2.3 Grammatical Functions and Switch-Reference
The coreferentiality rule as stated in (53) applies in most cases. However, it does not apply in a straightforward way when the dependent or controlling clause is an IVC. In (60a) cucuiimig ‘I feared something’ is a state verb with two arguments. It is M-transitive. The actor (emoter) argument is marked with 1sg.nom agreement and the undergoer (target) argument is unspecified. Here, cucuiimig functions as the controlling clause for ija cocobig ‘as I walked’ and the dependent clause to busali nuiga ‘I ran away’. With respect to (53), the pivot of ija cocobig is 1sg.nom and is coreferential with the 1sg.nom pivot of cucuiimig. Thus, cocobig is marked ss. Then the 1sg.nom pivot of cucuiimig is coreferential with the 1sg.nom pivot of busali nuiga and cucuiimig is marked ss. Adherence to (53) is straightforward.
(60)
a. State verb cucuiec fear′ (x, y) ‘to fear something’: Ija co~cob-ig cucui-im-ig 1sg dur~walk-1sg.nom.ss.sim fear-ss.seq-1sg.nom busal-i nu-ig-a. run away-dv go-1sg.nom-todp ‘As I walked along, I was afraid, and I fled.’ b. Impersonal verb cucuidoc feel′ (x, [afraid′]) ‘to be afraid’: Ija co~cob-ig cucui-t-ec-eb 1sg dur~walk-1sg.nom.ss.sim fear-1sg.acc-ds.seq-3sg.nom busal-i nu-ig-a. run away-dv go-1sg.nom-todp ‘As I walked along, I became afraid, and I fled.’
Now compare (60b). Here, cucuiteceb ‘I was afraid’ is an IVC.28 In this case, the verb has a 1sg.acc ‘experiencer’ argument and a 3sg.nom ‘empty’ argument. When cucuiteceb functions as the controlling clause to ija cocobig this verb is marked ss. Thus, the 1sg.acc argument in the IVC functions as pivot when it is the controlling clause. However, when cucuiteceb functions as the dependent clause to busali nuiga it is marked ds. This shows that the pivot arguments are not being compared between these clauses. Instead, the controller 3sg.nom argument in the IVC is compared with the pivot in the non-IVC controlling clause. Thus, the coreferentiality rule in (53) needs to be modified to (61). (61) applies to both regular verbs and IVCs. With a regular verb the pivot and controller are one and the same argument so monitoring either function does not select a different argument. However, in an IVC, the pivot and controller are different arguments, so it makes a difference which argument is monitored when the IVC functions as controlling clause or dependent clause.
25.3.2.4 Pragmatic Functions of the SR System
Often in Amele text, there occur what appear to be ‘anomalous’ ds markings where ds is indicated but the PSA of the marked clause and the controlling clause are the same. These ds markings are, in fact, not anomalous but are indicating a change in discourse theme. The thematic changes are primarily in the area of time, place and possible world setting (real vs. unreal). Thematic changes of time and place are often accompanied by temporal and locative modifier expressions and a ds thematic change of place is most commonly marked on a motion verb. Thematic changes in a possible-world setting are normally a switch from the real world to an intended or proposed action or vice versa, a switch from intended/proposed action to the real world. Examples of this phenomenon are given in Roberts (Reference Roberts1987: 303–305, 1988b). However, these are pragmatic extensions of the SR system. At core, the switch-reference system in Amele is syntactically motivated. It is a local device for monitoring the referentiality of PSA arguments between adjacent clauses as to whether they have identical or non-identical reference.
Abbreviations
The abbreviations and symbols used in the glosses in general follow the Leipzig glossing rules (www.eva.mpg.de/lingua/resources/glossing-rules.php), with the following additions:
- ABLT
abilitative
- ASP
aspect marker
- AV
actor voice
- CN
common noun
- FAC
factual mood
- IA
instrumental applicative
- LA
locative applicative
- LNK
linker
- NEUT
neutral voice
- NFIN
non-finite
- PPN
personal proper noun
- PrCS
pre-core slot
- PrDP
pre-detached position
- PREP
preposition
- RED
reduplication
- UV
undergoer voice
- VOL
volitative mood
26.1 Introduction
This chapter presents a sketch grammar of Amis with a focus on the discussion of the phenomena related to its case marking and voice. Amis is currently spoken by 210,839 speakers in Taiwan.1 It is classified as one of the Eastern Formosan branch languages (Blust Reference Blust, Zeitoun and Jen-kuei Li1999) and is reported to consist of five dialects (Tsuchida Reference Tsuchida, Kamei, Kono and Chino1988). The following discussion is based on the data collected from the Central dialect.
26.2 Basic Clause Patterns
Amis is a predicate-initial language. Three types of predicates can be identified: verbal, nominal and prepositional. One-place, two-place and three-place verbal predicates and their logical structures are provided in (1):2
(1)
Amis verbal predicates a. R<om>akat kako i lalan. <neut> walk 1s.nom prep road ‘I am walking on the road.’ a′. be-on′ (lalan, [do′ (kako, [walk′ (kako)]) b. Mi-palo Ø-ci Sawmah ci Mayaw-an. av-beat nom-ppn Sawmah ppn Mayaw-dat ‘Sawmah is beating Mayaw.’ Or ‘Sawmah is going to beat Mayaw.’ b′. do′ (Sawmah, [go′ (Sawmah)]) & INGR be-at′ (y, Sawmah) PURP do′ (Sawmah, [beat′ (Sawmah, Mayaw)]) b′′. do′ (Sawmah, [beat′ (Sawmah, Mayaw)]) c. Ma-palo n-i Sawmah Ø-ci Mayaw. uv-beat gen-ppn Sawmah nom-ppn Mayaw ‘Mayaw was beaten by Sawmah.’ c.′ do′ (Sawmah, [beat′ (Sawmah, Mayaw)]) & BECOME beaten′ (Mayaw) d. Pa-nanom kako t-o wawa t-o sayta. caus-water 1s.nom dat-cn child dat-cn soda ‘I gave the child soda to drink.’ (Causative, AV) d′. do′ (kako, Ø ) CAUSE BECOME have.water′ (wawa, sayta)
As shown in (1), verbal predicates are usually affixed with a voice marker (e.g. mi-), followed by various numbers of arguments that are marked by a case marker, be it a nominative case (for the privileged syntactic argument (PSA)), a genitive case (for a non-PSA actor or a possessor), or a dative case (for an oblique core argument or an adjunct). A locative argument or adjunct is marked by the preposition i.
Nominal predicates and prepositional predicates are exemplified respectively in (2a) and (2b).
(2)
Amis nominal and prepositional predicates a. O singsi cingra. cn teacher 3s.nom ‘He is a teacher.’ a′. be′ (cingra, [teacher′]) b. I loma’ Ø-ci Mayaw. prep house nom-ppn Mayaw ‘Mayaw is at home.’ b′. be-at′ (loma’, Mayaw)
Unlike verbal predicates, nominal predicates are preceded by a noun class marker3 and prepositional predicates are initiated by a preposition.
Special clause types, such as displacement constructions (e.g. 3a) and topic constructions (e.g. 3b), are given below.4
(3)
Amis displacement construction and topic construction a. O fafahian a singsi k-o ka-olah-an=ako cn woman lnk teacher nom-cn nfin-like-la=1s.gen ‘It is female teachers that I like (better).’ b. O sa-pi-senger t-o felac i, nga’ay ma-lalo cn ia-nfin-soak dat-cn rice top good.for uv-soft o sanay. cn like.this ‘Regarding the reason to soak the rice, it is for softening (it).’
Both of the constructions contain a displaced element that appears before the main predicate, but there is no pause before the element in the displacement construction and the rest of the clause while there is one in the topic construction, which means that the displaced elements in (3a) and (3b) are placed under the pre-core slot (PrCS) and pre-detached position (PrDP) respectively.
26.3 Lexical Categories
The discussion of lexical categories includes the roots and derived words in Amis. As pointed out by Wang (Reference Wang1976), all the root forms in Amis are syntactically nominal.5 In other words, to serve as a verb, root forms, even for roots with inherent verbal meaning, have to be either affixed with a verbal prefix (usually a voice marker6) or placed in the predicate position. The former strategy is more commonly found and examples can be seen in (1); the latter is only found with some state predicates such as miming ‘small’ and tosa ‘two’. Derived verbs with no verbal affixes are termed unaffixed verbs in Wu (Reference Wu2006a).
As root forms are syntactically nominal, they can readily appear after a case marker or a preposition (e.g. (4a)) to manifest arguments and adjuncts. However, the derived verbs (e.g. root form plus a voice marker) have to be nominalized7 before they can show up in a nominal position (e.g. (4b–c))8. See Wu (Reference Wu2006a: 70) for some common nominalizing or deverbal strategies in Amis.
(4)
Amis nouns and deverbal nouns a. Na’on-en k-o rakat! mind-uv nom-cn walk ‘Good-bye.’ Lit. ‘Mind your walk!’ b. *Na’on-en k-o r<om>akat! mind-uv nom-cn <neut>walk c. Na’on-en k-o r<om>akat-ay! mind-uv nom-cn <neut>walk-fac ‘Mind the one who is walking!’
The manifestation of predicates was illustrated in Section 26.2; a verbal predicate is expressed by a derived verb (e.g. (1)), a nominal predicate is composed of a noun (derived or base-generated) preceded by a noun class marker (e.g. (2a)), and a prepositional predicate is designated by a preposition plus a noun (e.g. (2b)).
26.4 Macrorole Selection and Argument-Marking
This section discusses the macrorole selection of Amis predicates with different numbers of core arguments and how arguments and non-arguments are marked.
26.4.1 Macrorole Selection of Verbs with Various Numbers of Core Arguments
Amis predicates with zero core arguments are found in meteorological or phenomenal verbs such as ma-’orad ‘rain’, and sienaw ‘cold (in terms of weather)’. These verbs can appear by themselves without any co-occurring argument, as illustrated in (5a). As there is no core argument, there is no macrorole for such verbs, and thus, they are macrorole-atransitive (M-atransitive). However, such verbs can also take one core argument, as illustrated in (5b) and become M-intransitive. Their different logical structures are given in (5a′) and (5b′) respectively.
(5)
Amis meteorological or phenomenal verbs a. Ma-’orad anini. neut-rain now ‘It is raining today.’ a′. rain′ (Ø) b. Ma-’orad k-o kakarayan. neut-rain nom-cn sky ‘The sky is raining.’ b′. rain′ (kakarayan)
Regarding the assignment of the macrorole for verbs with one core argument (i.e. syntactically intransitive or S-intransitive verbs), the presence or absence of do′ in their logical structure (LS) makes a crucial difference; the presence of do′ makes the only core argument an actor while the absence of do′ makes it an undergoer. In other words, the single argument for intransitive activity verbs such as t<om>ireng ‘stand’ and intransitive state verbs such as ma-toni’ ‘soft’ will not be assigned the same macrorole:
(6)
Amis verbs with one core argument a. T<om>ireng cingra. <neut>body 3s.nom ‘He is standing.’ b. Ma-toni’ k-o-ni a titi. neut-soft nom-cn-this lnk meat ‘This meat is soft.’
Although the only argument of the two verbs in (6) ia marked by the same case (i.e. the nominative case), the macrorole of t<om>ireng would be actor while that of ma-toni’ would be undergoer. This is where the Role and Reference Grammar (RRG) analysis differs from other studies concerning this issue in Amis (e.g. Liu Reference Liu1999, Liu Reference Liu2003, to name just a few). In such research, the single arguments of S-intransitive verbs seem to be treated as the same type and the verbs are labelled as actor-focus or actor-voice verbs. This analysis will be further discussed in the last section of this chapter.
When there are two core arguments in the LS of an Amis predicate, the macrorole assignment follows the Actor-Undergoer Hierarchy (AUH) and the default macrorole assignment principles. However, it is also possible for such verbs to have only one macrorole. As illustrated in (1b) and (1c), two-place verbs can appear with two case-marking patterns: the nominative-dative case frame in the actor voice (AV) sentences and the genitive-nominative pattern in the undergoer voice (UV) sentences. While the two-place UV verb follows the AUH and has both an actor and an undergoer, the second argument in their AV counterparts is a non-macrorole core argument, as argued in Wu (Reference Wu2006a). In other words, the two-place AV verbs are M-intransitive. The non-macrorole status of the second argument results either from the linking phase from the argument position in the LS to macrorole or from the voice operation that deprives a core argument of its macrorole status during linking from macrorole to syntactic function. The first case is found when the second argument of an activity verb is non-referential, as exemplified in (7a). The second one is illustrated in (7b).
(7)
Amis AV sentences with a two-place predicate a. Mi-nanom cingra (t-o nanom). av-water 3s.nom dat-cn water ‘He is drinking water.’ Or ‘He is going to drink water.’ b. Mi-nanom cingra t-o-ra sayta. av-water 3s.nom dat-cn-that soda ‘He is drinking that soda.’ Or ‘He is going to drink that soda.’
The principal difference between a macrorole second argument (normally a patient) and a non-macrorole one lies in the possibility for this argument to be promoted by the applicative construction; only a non-macrorole argument (or an adjunct) is eligible to appear in such constructions. See Wu (Reference Wu2006a) for the discussion in detail.
In addition to the two-place AV verbs, examples illustrating the mismatch between syntactic transitivity and macrorole transitivity is also seen in three- place predicates. To begin with, the AV construction of a three-place predicate is also deemed M-intransitive. Both of its non-actor arguments are marked by the dative case and, most important of all, both of them can be the promoted argument in the applicative construction, which is a feature for a non-macrorole argument in the AV construction. Examples are given in (8).
(8)
Amis three-place AV predicate and the applicative constructions a. Pa-nanom cingra ci Aki-an t-o-ra sayta. caus-water 3s.nom ppn Aki-dat dat-cn-that soda ‘He gave Aki that soda (to drink).’ (Causative, AV) b. Cima k-o pa-nanom-an nira t-o-ra sayta? who.nom nom-cn caus-water-la 3s.gen dat-cn-that soda ‘Who did he give that soda to drink?’ (causee/recipient as the undergoer) c. O maan k-o pa-nanom-an nira ci Aki-an? cn what nom-cn caus-water-la 3s.gen ppn Aki-dat ‘What did he give to Aki to drink?’ (causand/theme as the undergoer)
As shown in (8), both the recipient argument and the theme argument can be the PSA,9 in particular, the syntactic pivot, of the applicative three-place verb pa-nanom-an ‘give water’. This is a property for a non-macrorole (NMR) argument. In other words, neither of the two non-actor arguments in AV pa- sentences such as (8a) are macroroles.
While the UV three-place predicates are M-transitive, just like the UV two-place UV predicates, the mismatch between S-transitivity and M-transitivity is still found as there can be at most two macroroles in RRG, while there are three core arguments in a three-place predicate. This means there is competition for macrorole status between the two groups of potential undergoer participants, namely, theme/patient and recipient/beneficiary/source/goal. Hence, two principles of undergoer selection (i.e. Principle A: choosing the lowest-ranking argument in LS and Principle B: choosing the second-highest-ranking argument in LS) have been proposed in RRG under the AUH. As reported in Guerrero Valenzuela and Van Valin (2004), languages tend to exhibit a mixed type that needs both principles to account for their undergoer selection. Amis is also such a language. Wu (Reference Wu2006a, b) has shown that Amis three-place predicates vary in their selection of the undergoer, though Principle B seems to be more commonly employed. The evidence lies in the possibility of having either the theme/patient or the recipient/beneficiary/source/goal argument as the PSA in the UV constructions, and the latter group seems to be favoured by more three-place predicates.10 Some examples are given below. As one can see, verbs like pa-caliw ‘lend’ can only have the theme argument as the PSA in their UV construction, while verbs like pa-nanom ‘give water’ only allow the recipient argument as the PSA in their UV counterpart.
(9)
Some UV three-place predicates a. Ma-pa-cakay n-i Aki k-o foting ci Ofad-an. uv-caus-buy gen-ppn Aki nom-cn fish ppn Ofad-dat ‘Aki sold (other people’s) fish to Ofad.’ a′. [do′ (Aki, Ø)] CAUSE [[do′ (Ofad, Ø)] CAUSE [BECOME NOT have′ (Aki, foting) & BECOME have′ (Ofad, foting)]] b. *Ma-pa-cakay n-i Aki t-o foting Ø-ci Ofad. uv-caus-buy gen-ppn Aki dat-cn fish nom-ppn Ofad c. Ma-pa-nanom=to n-i ina t-o sayta Ø-ci uv-caus-water=asp gen-ppn mother dat-cn soda nom-ppn mama. father ‘Mother gave Father soda to drink.’ c′. [do′ (ina, Ø)] CAUSE [BECOME have.water′ (mama, sayta)]11 d. *Ma-pa-nanom=to n-i ina ci mama-an k-o uv-caus-water=asp gen-ppn mother ppn father-dat nom-cn sayta. soda
After briefly reviewing how the macroroles are selected for verbs with various numbers of core arguments, now let us see how these arguments and non-arguments are marked in Amis.
26.4.2 Argument-Marking
Arguments and non-arguments are marked in Amis through a tri-case system (nominative, genitive, dative) and the preposition i. The nominative case marks the PSA, or the so-called grammatical subject. The genitive case can mark a possessor or an actor in a non-AV sentence (e.g. (1c)), which makes it the equivalent marker of an ergative case. The dative case serves a wide range of functions. It can label a NMR core argument, as seen in the above examples (e.g. (1b)). It can also appear before an oblique core argument or an adjunct. Examples follow.12
(10)
Amis dative case marking an oblique core argument or an adjunct a. Ma-pa-nanom=to n-o wawa k-o kolong uv-caus-water=asp gen-cn child nom-cn water.buffalo t-o-ya nanom. dat-cn-that water ‘The child has already fed the water buffalo that water.’ (Theme) a′. Ya nanomi ma-pa-nanom=to n-o wawa k-o that water uv-caus-water=asp gen-cn child nom-cn kolong ____i. water.buffalo ‘That water the child has already fed the water buffalo.’ b. Ma-ota’ kako t-o sanek n-o tosiya. neut-vomit 1s.nom dat-cn smell gen-cn car ‘I feel sick from the smell of cars.’ (Reason) c. Ma-tayal kako t-o romi’ami’ad. neut-work 1s.nom dat-cn every.day ‘I work every day.’ (Time) d. Cenger-en=ako k-o kiladom t-o kohting-ay. color-uv=1s.gen nom-cn cloth dat-cn black-fac ‘I am going to colour the cloth with the black colour.’ (Instrument)
The dative case marks an oblique core argument in (10a) and various kinds of adjuncts in (10b–d). The oblique status of the theme argument in (10a) is indicated by its displaced version in (10a′), in which it is simply displaced to the sentence-initial position without affecting the rest of the sentence. This is the feature of displacing an oblique reference phrase (RP) in Amis; the displacement of a direct core RP will result in a nominal structure like an equational sentence, as will be seen in the later discussion. An Amis RP that appears clause-initially, be it a displaced one or not, does not bear any case marking. That is why the demonstrative pronoun before nanom ‘water’ has no case in (10a′).
Although both NMR core arguments and adjuncts can be marked by the dative case, there is a crucial difference between them, which lies in the mechanisms to promote their status to PSA. The core arguments can be promoted via plain undergoer voice constructions and applicative constructions:
(11)
The strategies for promoting an NMR core argument a. Ma-nanom n-i Aki k-o-ra sayta. uv-water gen-ppn Aki nom-cn-that soda ‘Aki drank that soda.’ (Plain UV) b. Mi-nanom-an n-i Aki k-o sayta. la-water-la gen-ppn Aki nom-cn soda ‘Aki drank the soda.’ (Locative applicative UV) ‘What Aki drank is the soda.’ (Locative applicative UV) c. Olah-en namo ∅-ci Panay. like-uv 2p.gen nom-ppn Panay ‘You have to love Panay.’ (Plain UV) d. Ka-olah-an=ako ∅-ci Panay. nfin-like-la=1s.gen nom-ppn Panay ‘Panay is the one I like (most).’ (Locative applicative UV)
The data in (11) illustrates the possibilities for enhancing the status of the second argument of pred′ (the one marked by the dative case) in mi-nanom and ma-olah to become a PSA (i.e. undergoer of a UV verb). Both the plain UV constructions (e.g. (11a) and (11c)) and the applicative UV constructions (e.g. (11b) and (11d)) are applicable here. Note that the number of core arguments in the two predicates has remained the same in the plain UV constructions and the applicative UV constructions.13
Now consider a different case exemplified in (12).
(12)
As shown in (12), for a reason/indirect cause adjunct RP marked by the dative case to become a PSA, only the applicative construction can be used. Note that the number of core arguments will change when the applicative constructions are employed, as can be seen from the comparison of the number of arguments in the LS of the non-applicative verb (e.g. (12c′)) and the applicative one (e.g. (12d′)).
Another way to make the adjunct in (12a) and (12c) a PSA, in addition to using the applicative UV construction, is to make the adjunct the actor of an AV construction, as illustrated in (13):
(13)
The strategies for promoting an adjunct a. Mi-patay k-o sapaiyo n-o 'edo t-o 'oner. av-dead nom-cn medicine gen-cn mouse dat-cn snake ‘The poison for killing mice may kill a snake as well.’ a′. [do′ (sapaiyo no 'edo, Ø)] CAUSE [BECOME dead′ ('oner)] b. Mi-ota’ t-o tamdaw k-o sanek n-o tosiya av-vomit dat-cn person nom-cn smell gen-cn car ‘The smell of cars makes people vomit.’ b′. [do′ (sanek no tosiya, Ø)] CAUSE [BECOME vomit′ (tamdaw)]
The reason/indirect cause adjunct RPs in (12a) and (12c) now become the actors in (13a) and (13b), respectively. As indicated in the logical structures of mi-patay in (13a) and mi-ota’ in (13b), the predicates have become causativized, and there is an effector added to the core of the predicates.14 In other words, the number of core arguments has also been changed. The addition of a core argument is not found in the examples in (11); when the to RPs of mi-nanom and ma-olah become PSAs in the plain or applicative UV constructions, there is no argument addition involved. Hence, the to RP of ma-patay in (12a) and ma-ota’ in (12c) should be analysed differently from the to RPs in mi-nanom in and ma-olah; the former are adjuncts while the latter are NMR core arguments.
The last difference between an NMR direct core argument and an adjunct (or even an oblique core argument) is that the status of the former can always be adjusted through the plain voice operation; however, for the latter, it is not always possible. In other words, some to RPs can only be promoted by means of the applicative constructions. For example, the plain voice construction is quite unlikely to be employed to promote the adjunct manifesting temporal expression in (10c), although the applicative form ka-tayal-an ‘place or time for working’ can be used. For some adjuncts that are more likely to be construed as effectors (e.g. an indirect cause like sanek no tosiya ‘smell of the car’ in (12c), they may be promoted to become an actor in AV and UV constructions, as we have seen in (13).
Based on the discussion so far, the case assignment rules in Amis can be summarized as (14):15
(14)
Case assignment rules in Amis a. Assign nominative case to the lowest macrorole argument in terms of the PSA Selection Hierarchy. b. Assign genitive case to the other macrorole argument. c. Assign dative case to other direct core argument(s).
The examples in (15) illustrate how the rules in (14) are applied in Amis.
(15)
The application of case assignment rules in Amis a. Ma-olah kako ci Panay-an av-like 1s.nom ppn Panay-dat ‘I like Panay,’ a′. like′ (kako, Panay) (Rule(s) applied: (14a) and (14c)) b. Ma-ka-olah=ako Ø-ci Panay. uv-nfin-like=1s.gen nom-ppn Panay ‘I love Panay (secretly).’ or ‘Panay was loved by me.’ b′. like′ (ako, Panay)….BECOME like′ (ako, Panay) (Rule(s) applied: (14a) and (14b)) c. Ma-stol kako t-o fekeroh. neut-stumble 1s.nom dat-cn rock ‘I stumbled over on the rock.’ c′. stumble′ (kako) (Rule(s) applied: (14a)) d. Ma-stol n-o fekeroh kako. uv-stumble gen-cn rock 1s.nom ‘The rock rolled to me and made me stumble.’ d′. [do′ (fekeroh, Ø)] CAUSE [BECOME stumble′ (kako)] (Rule(s) applied: (14a) and (14b)) e. Ma-rohem=to k-o-ra pawli. neut-ripe=asp nom-cn-that banana ‘The banana has become ripe.’ e′. (INGR/BECOME) ripe′ (pawli) (Rule(s) applied: (14a)) f. Pa-si-fana’ k-o singsi t-o wawa t-o caus-have-knowledge nom-cn teacher dat-cn child dat-cn n-o Amis. gen-cn Amis ‘The teacher is going to teach the children Amis.’ (Causative, AV) f′. [do′ (singsi, Ø)] CAUSE [BECOME have.knowledge′ (wawa, no ’Amis)] (Rule(s) applied: (14a) and (14c))
In addition to a set of case markers, there is also a preposition i in Amis. This preposition mainly marks arguments with a locative feature (i.e. x in be-loc′ (x, y) or pred-loc′ (x, y)). In addition, it also marks the first argument of the existential verbs ira or awa (i.e. (NOT) exist′ ([pred′ (x, y)])) and possibly the first argument in the embedded logical structure BECOME/INGR pred′ (y, z). Some examples are given in (16):
(16)
The functions of the preposition i a. Maro’ kako i Taypak. live 1s.nom prep Taipei ‘I live in Taipei.’ (Neutral voice) a′. live.in′ (Taypak, kako) b. Ira k-o kawas i loma’ nira. exist nom-cn ghost prep house 3s.gen ‘There is ghost in his house.’ (Neutral voice) b′. exist′ ([be-in′ (loma’ nira, kawas)]) c. Pa-nengneng kako t-o-ni~ni t-o/i wawa. caus-see 1s.nom dat-cn-this~red dat-cn /prep child ‘I showed the child this.’ Or, ‘I showed this to the child.’ (Causative, AV) c′. [do′ (kako, Ø)] CAUSE [BECOME see′ (wawa, tonini)]
However, as illustrated in (16), while the first argument of the embedded BECOME/INGR pred′ might be marked in more than one way (e.g. by a dative case or a preposition), the preposition is the only choice for the first argument of pred-loc′ (x, y). Moreover, while the first argument of the embedded BECOME/INGR pred′ can be a possible undergoer and hence a PSA in the UV construction, it is impossible for the first argument of pred-loc′ to be an undergoer, let alone a PSA. This is illustrated by the following contrast between pa-nanom ‘cause to have water’ and pa-teli ‘put’ in (17):
(17)
Comparison of the first argument of BECOME/INGR pred′ and pred-loc′ a. Pa-nanom-en k-o sayta t-o nanom! caus-water-uv nom-cn soda dat-cn water ‘Add water to the soda!’ a′. DO (x, [do′ (x, Ø)] CAUSE [BECOME have.water′ (sayta, nanom)] b. Pa-teli kako t-o konga i langa. caus-put 1s.nom dat-cn sweet.potato prep basket ‘I put the sweet potatoes in the basket.’ (Causative, AV) b′. [do′ (kako, Ø)] CAUSE [BECOME be-loc′ (langa, konga)] c. Ma-pa-teli’=ako k-o konga i langa. uv-caus-put=1s.gen nom-cn sweet.potato prep basket ‘I put the sweet potato in the basket.’ c′. *Ma-pa-tli’=ako t-o konga k-o langa. uv-caus-put=1s.gen dat-cn sweet.potato nom-cn basket
As indicated in (17), the first argument of the embedded be-loc′ (e.g. langa ‘basket’)) cannot be an undergoer in the UV construction. This follows from the claim in RRG that the first argument of be-loc′ or pred-loc′ cannot be a macrorole; in other words, two-place locative predicates are always M-intransitive. The following preposition-assignment rules are postulated for Amis:
(18)
26.5 Grammatical Relations and the Voice System
This section discusses grammatical relations and the voice system in Amis. According to Wu (Reference Wu2006a), Amis only employs a PSA in the formation of relative clauses, a nominal type of displacement constructions and wh-questions. Constructions that might need a grammatical relation in other languages, such as control constructions and reflexivization, can be taken care of by semantic roles in Amis. Hence there are no grammatical relations found in such constructions.
As for the voice system, Amis distinguishes two voices: actor and undergoer, and the latter can be further divided into the plain (or non-applicative) UV and the applicative UV. The functions of each voice will be elaborated in the following sub-sections.
26.5.1 Grammatical Relations
A relative clause (RC) in Amis is formed by gapping an RP from the modifying clause. The gapped RP is a pivot as it is omitted in the clause. This gapped RP is coreferential with the modified noun that follows the RC. To serve as a head of an RC, its coreferential gapped RP has to be the actor of an AV verb, the undergoer of a plain UV verb, or an applied argument of an applied UV verb in the RC. If the gapped RP does not belong to any of these types, the sentence will be rendered ungrammatical. This is exemplified in (19), in which the gapped RP is indicated by ‘__’ in the RC.16 Hence, there is restricted neutralization of semantic roles on the pivot of an RC in Amis.
(19)
Amis relative clauses a. Pivot: Actor of AV verb Ma-patay=to k-o-ya mi-kalat-ay ____i ci neut-dead=asp nom-cn-that av-bite-fac ppn Aki-an a wacoi. Aki-dat lnk dog ‘That dog that bit Aki is dead.’ a′. Pivot: NMR direct core argument of AV verb *Ma-patay=to k-o-ya mi-kalat-ay k-o waco ____i neut-dead=asp nom-cn-that av-bite-fac nom-cn dog a tamdawi. lnk person ‘That person that the dog bit is dead.’ b. Pivot: Undergoer of UV verb Tati’ih k-o-ya ma-kaen-ay n-i Aki ____i a bad nom-cn-that uv-eat-fac gen-ppn Aki lnk talii. taro ‘That taro that Aki ate was bad.’ b′. Pivot: (Patient) undergoer of applied UV verb Tati’ih k-o-ya mi-kaen-an n-i Aki ____i a bad nom-cn-that la-eat-la gen-ppn Aki lnk talii. taro ‘That taro that Aki ate was bad.’ c. Pivot: Actor of UV verb *Ma-soso k-o-ya ma-kaen-ay ____i k-o tali neut-fat nom-cn-that uv-eat-fac nom-cn taro a tamdaw. lnk person ‘The person that ate the taro was fat.’ d. Pivot: (Instrument) undergoer of applied UV verb Ma-pitek=ako k-o sa-pi-cikcik n-i Aki uv-break=1s.gen nom-cn ia-nfin-cut gen-ppn Aki t-o dateng ____i a po’oti. dat-cn vegetable lnk knife ‘I broke the knife with which Aki cut the vegetable.’ e. Pivot: (Locative) undergoer of applied UV verb Tayra Ø -ci Panay mi-ladom i go nom-ppn Panay neut-fetch.water prep pi-ladom-an n-i Aki ____i a tefoni. nfin-fetch.water-la gen-ppn Aki lnk well ‘Panay went to fetch water at the well where Aki fetched water.’
The same restriction is found in the formation of the displacement construction and wh-questions, both of which involve a displaced nominal element that appears in the clause-initial position. There are two types of such structure. The first type, termed the nominal type, is constructed as an equational sentence in which the displaced RP or the wh-word and the remaining elements of the clause are juxtaposed. This remaining clause is preceded by a nominative case marker, and it is structured like a headless RC. The second type, termed the verbal type, is formed simply by placing an RP or a wh-word at the beginning of the clause. The remaining clause of the verbal type stays structurally unchanged. The wh-word can even appear in situ in the verbal type though it more often appears clause-initially. It is the nominal type that involves a restricted neutralization of semantic roles; that is, its pivot has to be an actor of a gapped AV clause or an undergoer of a gapped UV clause. As for the verbal type, the restricted neutralization is not found. The nominal type is illustrated in (20) while (21), repeated from (10a′), exemplifies the verbal type.17 The displaced RP is underlined and its pivot is indicated by ‘___’ in the clause.
(20)
Amis sentences with displaced RPs (the nominal type) a. Pivot: (Patient) undergoer of an applied UV verb O fafahian a singsii k-o ka-olah-an=ako ___i. cn woman lnk teacher nom-cn nfin-like-la=1s.gen ‘It is female teachers that I like better.’ b. Pivot: Actor of AV verb Ya wawai k-o mi-pa-nanom-ay ____i t-o that child nom-cn av-caus-water-fac dat-cn kolong. water.buffalo ‘It is that child who fed water to the water buffalos.’ b′. Pivot: Actor of UV verb *Ya wawai k-o ma-pa-nanom-ay ____i k-o that child nom-cn uv-caus-water-fac nom-cn kolong. water.buffalo ‘It is that child who fed water to the water buffalos.’ c. Pivot: (Patient) undergoer of an applied UV verb Ya nanomi k-o mi-pa-nanom-an=to=ako ci that water nom-cn la-caus-water-la=asp=1s.gen ppn mama-an ____i. father-dat ‘That water is what I gave father to drink.’
(21)
Amis sentences with displaced RPs (the verbal type) Pivot: Oblique core argument of three-place UV verb Ya nanomi ma-pa-nanom=to n-u wawa k-o that water uv-caus-water=asp gen-cn child nom-cn kolong ____i. water.buffalo ‘The child has already fed the water buffalo that water.’
In (20a), the undergoer RP appears in sentence-initial position and there is a gap in the remaining clause that follows the displaced RP. As one can see, there is a case marker ko between the displaced RP and the remaining clause; that is, the clause appears in a nominal position. Furthermore, the verb inside the nominal clause is an applied UV verb. Examining the rest of the examples in (20), we can see that they demonstrate a restricted neutralization of semantic roles, as the pivot in the nominal clause following the displaced element has to be the actor of an AV verb or the undergoer of a UV verb; the latter can be either a plain UV verb or an applied UV verb.
The restricted neutralization exemplified in (20) is also observed in the nominal type wh-questions in (22):
(22)
Amis wh-questions (the nominal type) a. Pivot: Actor of AV verb Cimai k-o mi-palo-ay ____i t-o wawa? who.nom nom-cn av-beat-fac dat-cn child ‘Who is the one that beat the child?’ a′. Pivot: Actor of UV verb *Cimai k-o ma-palo-ay ____i k-o wawa? who.nom nom-cn uv-beat-fac nom-cn child ‘Who is the one that beat the child? b. Pivot: Undergoer of UV verb O maani k-o ma-kaen-ay n-i Aki ____i? cn what nom-cn uv-eat-fac gen-ppn Aki ‘What is it that Aki ate?’ b′. Pivot: NMR direct core argument of AV verb *O maani k-o k<om>aen-ay Ø-ci Aki ____i? cn what nom-cn <av>eat-fac nom-ppn Aki ‘What did Aki eat?’ c. Pivot: (Patient) undergoer of applied UV verb Cimai k-o ka-olah-an=iso ____i? who.nom nom-cn nfin-like-la=2s.gen ‘Who is the one you like?’
The sentences in (22a–a′) exemplify wh-questions concerning an actor of a predicate. As shown in the data, the clause following the interrogative pronoun is preceded by a case marker, which gives the nominal property of the clause. Furthermore, when the interrogative pronoun functions as the actor of the predicate, the verb has to be marked by the AV affix; this pronoun cannot be interpreted as functioning as the actor of a UV verb. When the interrogative pronoun refers to a non-actor in the clause, the verb has to be marked by either the plain UV markers (e.g. (22b)) or the applicative markers (e.g. (22c)). Hence, there is a restricted neutralization of semantic roles.18
26.5.2 Voice Operations and Applicative Constructions
Table 26.1 summarizes the voice system in Amis. As shown, Amis distinguishes three voices: neutral, actor and undergoer. Neutral voice is used for marking verbs that are both M-intransitive and S-intransitive as their single macrorole is not specified. The neutral voice will be further explored in Section 26.6. The other two voices will be introduced in more detail here.
Table 26.1 Voice markers and the applicative markers in Amis (adapted from Wu Reference Wu2006b: 289)
| Voice | Macrorole of the PSA | Affixes | |
| Neutral voice19 | Unspecified | mi-, ma-, <om> | |
| Actor voice | Actor | mi-, ma-, <om> | |
| Undergoer voice | Plain | Undergoer (unmarked choice) | ma-, ma-…<om>, ma-ka-, ‑en |
| Applicative | Undergoer (marked choice) | sa-, ‑an | |
Let us begin with the discussion of actor voice. AV verbs always have a nominative-dative case pattern, and they have only one macrorole (i.e. actor). The actor is assigned the nominative case and the NMR argument is marked by the dative case. The voice-marking function of the AV affixes is demonstrated in their co-occurrence with the volitative mood marker ‑aw, which shows the UV pattern when it attaches to a root form.
(23)
-aw volitative construction a. Nanom-aw=ho=ako. water-vol=asp=1s.gen ‘I will go drink water first.’ (Volitative mood, UV) a′. Mi-nanom-aw=ho kako. av-water-vol=asp 1s.nom ‘I will go drink water first.’ (The water is farther away than the one mentioned in (23a).) b. Kaen-aw=ako k-o dateng. eat-vol=1s.gen nom-cn vegetable ‘I will try that vegetable.’ (Volitative mood, UV) b′. K<om>aen-aw k-o wawa t-o sapaiyo. <av>eat-vol nom-cn child dat-cn medicine ‘(I am) afraid that the child will take the medicine.’ c. Olah-aw=ako kiso? like-vol=1s.gen 2s.nom ‘May I go to love you?’ (Volitative mood, UV) c′. Ma-olah-aw kako tisonan. av-like-vol 1s.nom 2s.dat ‘I am afraid that I will like you.’
As shown in (23), the suffix ‑aw manifests an optative reading for the derived verb. Notice that the case-marking pattern for V-aw follows the UV pattern, as the actor is marked by the genitive case. However, when the V-aw forms are affixed with mi-, <om>, and ma-, their case-marking patterns become the AV pattern, as indicated in (23a′), (23b′) and (23c′). This contrast shows the voice-marking function of the AV markers. But what kinds of functions do the AV constructions perform?
Clearly, the AV construction has a PSA-modulation function, as it makes a marked choice of PSA in terms of the PSA Selection Hierarchy. Given the fact that Amis displays ergative features in at least the case-marking system and some constructions that involve a PSA such as the relative clause and the nominal type of wh-question, one would expect the lowest-ranking argument to be the unmarked PSA choice. However, in the AV construction, it is the highest-ranking direct core argument that is chosen to be the PSA.
What about the argument modulation function? For a two-place predicate, the lowest-ranking direct core argument in the AV sentences should be assigned an undergoer based on the macrorole assignment principles, as such verbs can take at most two macroroles. However, this argument in the AV construction does not surface as a macrorole syntactically, as revealed by its case marking and the fact that its status can be promoted by the applicative construction. Instead, this argument is realized as an NMR core argument in the AV construction. In other words, the lowest-ranking argument of a two-place predicate has been stripped of its macrorole status by the AV operation. However, it is still in the core, as indicated by its behavioural property in serving as a semantic controller in the persuade-type control construction, as discussed in Wu (Reference Wu2006a). Hence, AV constructions also perform an argument-modulation function. This function is even more salient for three-place predicates, as a possible undergoer can also be marked by the preposition in the AV construction in addition to the dative case. Consider the following examples:
(24)
Amis three-place predicates a. Pa-caliw Ø-ci Kacaw t-o singsi t-o caus-borrow nom-ppn Kacaw dat-cn teacher dat-cn payso. money ‘Kacaw lent the teacher money.’ (Causative, AV) b. Pa-caliw Ø-ci Kacaw t-o payso i caus-borrow nom-ppn Kacaw dat-cn money prep singsi. teacher ‘Kacaw lent the money to the teacher.’ (Causative, AV) c. Aka pa-caliw-en k-o singsi t-o payso! neg.imp caus-borrow-uv nom-cn teacher dat-cn money ‘Don’t lend the teacher money!’
As shown in (24), the recipient RP singsi can be marked either by the dative case or the preposition in the AV construction. This RP is the second-highest-ranking argument in the LS of pa-caliw ‘lend’, and it is also a possible undergoer, as indicated in the UV sentence in (24c). The presumed undergoer RP is realized as non-macrorole in the AV construction in (24a), but it is realized as an adjunct in (24c), as the preposition i typically marks a locative RP in the periphery. From the above discussion, we can thus conclude that the actor voice not only modulates the PSA choice but also modulates the semantic status of a core argument by either stripping a macrorole argument of its macrorolehood or realizing a core argument as an oblique element.
What about the undergoer voice? As shown in Table 26.1, there are two sets of UV markers: the marked set and the unmarked set. The markers that signify a marked undergoer choice used to be treated as voice markers as well (termed instrumental voice and locative voice) in Amis, but the examples in (25) show that they serve functions other than that of a voice operation.
(25)
The applicative UV constructions in Amis a. Aka sa-pi-litek-en k-o-ra caklis neg.imp ia-nfin-chop.tree-uv nom-cn-that axe t-o-ra kilang! dat-cn-that tree ‘Don’t use that axe to chop down the tree!’ b. Ma-sa-pi-sanga n-i Aki t-o takid k-o-ya uv-ia-nfin-make gen-ppn Aki dat-cn bottle nom-cn-that aol. bamboo ‘Aki used that bamboo to make the bottle.’
The sentences in (25) show that when the UV marker ‑en or ma- and the instrumental applicative marker sa- co-occur in a sentence, only the instrument RP surfaces as the undergoer instead of the patient RP, which would be the default undergoer following the AUH. In other words, the applicative markers indicate a marked undergoer selection. The voice markers and the applicative markers show different operations in the two phases in the RRG linking algorithm. That is to say, the applicative marker affects the linking from argument positions to macroroles, while the voice marker operates at the linking from macroroles to syntactic functions. There are two functions of these applicative markers. First, they may enhance the status of a non-argument such as instrument or location to become a core argument. Second, they can also promote a non-macrorole core argument (e.g. patient in an AV sentence) to become a macrorole. The instrumental applicative construction serves the first function, while the locative applicative construction can perform both functions.20
Based on the discussion so far, the UV pattern should be deemed the default pattern in Amis, which is proven by the fact that it is the unmarked voice of the applicative constructions even when the UV markers do not show up. Although the UV pattern enjoys unmarked status in Amis, there are some predicates that seem to take the AV pattern by default, and for such predicates, UV appears to be the marked pattern. Such predicates can be illustrated by the pa- verbs. Consider the following examples of a pa- verb plus the volitative mood suffix ‑aw:
(26)
pa- verbs suffixed with the volitative marker -aw a. Pa-nanom kako t-o kolong. caus-water 1s.nom dat-cn water.buffalo ‘I feed water buffalos water.’ (Causative, AV) b. Pa-nanom-aw=ho=ako k-o kolong. caus-water-vol=asp=1s.gen nom-cn water.buffalo ‘I will feed the water buffalo water first.’ (Volitative, UV) c. Mi-pa-nanom-aw=ho kako t-o kolong av-caus-water-vol=asp 1s.nom dat-cn water.buffalo ‘I will go to feed water buffalos water first.’
Recall that in the previous discussion, I showed that when a root form is suffixed with ‑aw, it takes the UV pattern. As we can see in (26a), the pa- predicates appear with the AV case-marking pattern (i.e. nominative-dative). However, when they are suffixed with ‑aw, the case-marking pattern becomes the UV pattern. In other words, the pa- predicates behave like a bare root form in the volitative mood construction. When the volitative form pa-nanom-aw is prefixed with mi-, the case pattern becomes the AV pattern again. These examples show that, unlike mi-, pa- does not have a voice-marking function. However, pa- verbs follow the AV pattern by default. To make pa- verbs appear in the UV pattern, the plain UV markers or the applicative forms have to be used. Morphologically, the AV pattern appears to be the default pattern of pa- verbs, while the UV pattern is a marked one. However, syntactically, the UV forms actually turn a marked pattern (i.e. AV) into an unmarked one. This may explain why the UV form pa- … ‑en and the applicative form pa- … ‑an are found much more frequently than the plain pa- forms in Amis.21
As mentioned, AV constructions perform both PSA-modulation and argument-modulation functions. How about UV constructions? Though appearing to be the basic pattern of Amis based on the case marking and the default voice choice of the applicative constructions, they turn out to be the marked voice choice for some predicates that usually appear with the AV pattern by default. For such predicates, their plain UV constructions perform a PSA-modulation function but no argument-modulation function as the macrorole arguments and the NMR core arguments remain unaffected. See Wu (Reference Wu2006a) for more discussion. However, the applicative UV constructions display both functions, as now an adjunct becomes both a macrorole and the PSA. The above discussion shows that both the actor voice and the undergoer voice are deemed basic voice forms.22 Therefore, Amis exhibits a split system in verbal morphology in spite of displaying ergative features in the case-marking system and in some grammatical constructions.
26.6 Special Discussion: An RRG Account for One-Place Predicates in Amis
One-place predicates in Amis usually appear in an unaffixed manner (e.g. tayra ‘go’ and miming ‘small’), or with affixes formally identical with the AV markers such as <om> and ma- (e.g. t<om>angic ‘cry’ and ma-lalok ‘diligent’). In the RRG analysis proposed here, the voice marker that appears on one-place predicates is treated as neutral voice as the only macrorole of such verbs is not specified. This analysis is rather different from the previous studies, which seem to assume a unified semantic role for the single argument of intransitive predicates. For example, they either label intransitive predicates as AV verbs (e.g. Liu Reference Liu1999) or claim that the single argument of intransitive verbs is patient (e.g. Chen Reference Chen1987). In fact, treating one-place predicates as AV or AF (i.e. agent-focus) verbs has been a fairly common practice in the studies of other Formosan languages as well (e.g. Atayal, as seen in Huang (Reference Huang2000) and Yeh (Reference Yeh, Zeitoun, Teng and Wu2015), and Tsou in Zeitoun (Reference Zeitoun1993), to name just a few). However, this unified-semantic-role approach is implausible if we consider the rather different morphosyntactic behaviours of the single argument of the one-place predicates in Amis. In the following, I am going to show how the RRG non-unified macrorole approach for one-place predicates (Van Valin and LaPolla Reference Van Valin and LaPolla1997; Van Valin Reference Van Valin2005) can account for their peculiar morphosyntactic behaviours in two constructions: the UV ‑en verbs and the evaluative construction.
26.6.1 The -en Form of One-Place Predicates
The first piece of evidence that proves the different macrorole assignments of the sole argument of the one-place predicates comes from their corresponding UV ‑en form, in which the actor gets the genitive case while the undergoer takes the nominative case. Consider the following examples:
(27)
Amis verbs with one core argument a. T<om>ireng cingra. <neut>stand 3s.nom ‘He is standing.’ a′. Tireng-en=ako pa-kimad, ta paka-nengneng stand-uv=1s.gen caus-speech so.that ablt-see kamo. 2p.nom ‘I will stand up when making a speech so that you can see (me) clearly.’ b. Ma-toni’ k-o-ni a titi. neut-soft nom-cn-this lnk meat ‘This meat is soft.’ b′. Toni’-en=ako k-o-ni a titi. soft-uv=1s.gen nom-cn-this lnk meat ‘I will soften this meat.’
Both t<om>ireng ‘stand’ and ma-toni’ ‘soft’ are traditionally labelled as AV verbs. When they are suffixed with ‑en, the only argument in t<om>ireng (now tireng-en) is marked by the genitive case. However, the single argument in ma-toni’ (e.g. koni a titi) is still marked by the nominative case in the UV construction toni’-en. The factor affecting the case-marking pattern is the different macroroles assigned to the only arguments of t<om>ireng ‘stand’ and ma-toni’ ‘soft’. As the LS of t<om>ireng is do′ (x, [stand′ (x)], the x-argument will be an actor. But, as there is no do′ in the LS of ma-toni’ (i.e. (BECOME/INGR) soft′ (x)), the x-argument is an undergoer. When the verb is affixed by ‑en, the agentive UV marker, the actor in tireng-en is marked by the genitive case by default,23 while the undergoer in toni’-en receives the nominative case in this UV ‑en construction. The above examples indicate the inadequacy of labelling both of the two verbs as AV or AF verbs. Instead, the RRG approach can account for the distinctive nature of the semantic roles played by the single arguments of the two verbs.
However, one may run into a problem upon the application of the rules in (14), repeated as (28) below, for intransitive verbs suffixed with the UV marker ‑en.
(28)
Case assignment rules in Amis a. Assign nominative case to the lowest macrorole argument in terms of the PSA Selection Hierarchy. b. Assign genitive case to the other macrorole argument. c. Assign dative case to other direct core argument(s).
As shown in tiring-en in (27a′), the single argument of an ‑en intransitive verb is always marked by the genitive case. Applying rule (28a) to an ‑en intransitive verb will yield the wrong case assignment. Therefore, another set of case assignment rules for verbs marked by ‑en has to be postulated. These rules are stated in (29):
(29)
Case assignment rules for verbs marked by -en a. Assign genitive case to the highest-ranking macrorole in terms of the PSA Selection Hierarchy. b. Assign nominative case to the other macrorole argument. c. Assign dative case to other direct core argument(s).
For two-place or three-place ‑en verbs, all three of the rules in (29) are applicable. But for the one-place ‑en verbs, only (29a) and (29c) will apply, as there is only one macrorole in such verbs. The examples in (30) illustrate how the rules in (28) and (29) work in Amis.
(30)
The application of the revised case assignment rules in Amis a. Ma-olah kako ci Panay-an. av-like 1s.nom ppn Panay-dat ‘I like Panay,’ a′. like′ (kako, Panay) (Rule(s) applied: (28a) and (28c)) b. Ma-ka-olah=ako Ø-ci Panay. uv-nfin-like=1s.gen nom-ppn Panay ‘I love Panay (secretly).’ Or, ‘Panay was loved by me.’ b′. like′ (ako, Panay)….BECOME like′ (ako, Panay) (Rule(s) applied: (28a) and (28b)) c. Ma-stol kako t-o fekeroh. neut-stumble 1s.nom dat-cn rock ‘I stumbled over on the rock.’ c′. stumble′ (kako) (Rule(s) applied: (28a)) d. Ma-stol n-o fekeroh kako. uv-stumble gen-cn rock 1s.nom ‘The rock rolled to me and made me stumble.’ d′. [do′ (fekeroh, Ø)] CAUSE [BECOME stumble′ (kako)] (Rule(s) applied: (28a) and (28b)) e. Ma-rohem=to k-o-ra pawli. neut-ripe=asp nom-cn-that banana ‘That banana has become ripe.’ e′. (INGR/BECOME) ripe′ (pawli) (Rule(s) applied: (28a)) f. Rakat-en=ako. walk-uv=1s.gen ‘I will walk (to do something.)’ f′. DO (ako, [walk′ (ako)]) (Rule(s) applied: (29a)) g. Rakat-en=ako k-o-ni a kayakay. walk- uv=1s.gen nom-cn-this lnk bridge ‘I will walk past this bridge.’ g′. DO (ako, [walk′ (ako, kayakay)]) & [BECOME walked′ (kayakay)] (Rule(s) applied: (29a) and (29b)) h. Pa-si-fana’ k-o singsi t-o wawa t-o caus-have-knowledge nom-cn teacher dat-cn child dat-cn n-o ’Amis. gen-cn Amis ‘The teacher is going to teach the children Amis.’ (Causative, AV) h′. [do′ (singsi, Ø)] CAUSE [BECOME have.knowledge′ (wawa, no ’Amis)] (Rule(s) applied: (28a) and (28c))
26.6.2 The Evaluative Constructions
RRG’s non-unified treatment of the semantic role played by the single argument of the one-place predicates can not only account for the different case-marking patterns of the one-place ‑en verbs discussed in the previous section but can also offer a better explanation for the evaluative construction, which can render two types of English sentences such as ‘The food tastes good’ or ‘It is difficult to do this job’. Amis examples for the two types are given in (31) and (32) respectively.
(31)
Amis evaluative constructions: Type I a. Tada-fangcal a nengneng-en k-o-ra coka. very-good lnk see-uv nom-cn-that picture ‘That picture looks really beautiful.’ b. ’angerer-ay a kaen-en k-o karokot. bitter-fac lnk eat-uv nom-cn bitter.gourd ‘Bitter gourds taste bitter.’
(32)
Amis evaluative constructions: Type II a. Koesit a sowal-en ciira. difficult lnk say-uv 3s.nom ‘It is difficult to talk to him.’ b. Tati’ih a fo’enot-en k-o fanoh n-o howak. bad lnk pull.out-uv nom-cn feather gen-cn duck ‘It is difficult to pull out the feathers of ducks.’ c. Sa-koesit sa a tedal-en k-o-ni saso’ot. so-difficult like.this lnk untie-uv nom-cn-this knot ‘It is so difficult to untie this knot.’
The Amis sentences in (31) and (32) all begin with a state predicate (i.e. the evaluative predicate), which is then followed by the linker a and a UV verb. The difference between the two types of evaluative construction is that the nominative argument is a core argument shared by both predicates in (31) while the one in (32) is an argument of the second verb but not the first one. Here we will limit our discussion to the first type. A similar construction has been found in other Formosan languages such as Seediq (Tsukida Reference Tsukida, Adelaar and Himmelmann2005) and Atayal (Yeh Reference Yeh, Zeitoun, Teng and Wu2015). Atayal examples are given in (33) (Yeh Reference Yeh, Zeitoun, Teng and Wu2015: 132, original gloss).
(33)
Atayal evaluative constructions24 blaq niq-un qu’ mami’ qa’. good[av] eat-uv nom rice dem ‘This (type) of rice tastes good.’
Yeh analyses the first predicate of such constructions as an AV verb in spite of the fact that it is a state predicate, and she argues that the construction is a fusion of a serial verb construction (SVC) and a commentative complement clause construction (commentative CCC) (Yeh Reference Yeh, Zeitoun, Teng and Wu2015: 132). Her analysis is summarized in Table 26.2 (cited from Yeh Reference Yeh, Zeitoun, Teng and Wu2015: 148).
Table 26.2 The relationship between the juxtaposed verbs in the BLAQ evaluative construction, SVCs and commentative CCCs in Squlip Atayal: a comparison (adapted from Yeh Reference Yeh, Zeitoun, Teng and Wu2015: 148)
| Construction type parameter | SVCs | BLAQ evaluative construction | Commentative CCCs |
|---|---|---|---|
| (A) Manifestation of voice in verb sequences | (a) AV UV | AV UV | (a’) AV AV |
| (b) UV AV | (b’) AV UV | ||
| (B) Sharing of arguments | Obligatory | Obligatory | Optional |
| (C) Sharing of TAM (tense-aspect-modality) information | Obligatory | Obligatory | Depends |
| (D) Sharing of polarity value | Obligatory | Obligatory | Optional |
According to the table, the Atayal evaluative construction shares more features with SVC, but it is argued to also show the feature of the complement clause in its AV–UV sequence. Amis examples analogous to the Atayal SVCs discussed in Yeh (Reference Yeh, Zeitoun, Teng and Wu2015) are provided in (34). Similar to those Atayal examples, the second verb in such sentences has to be an AV form or an intransitive verb with an actor (e.g. (34a), but there is no restriction on the voice of the first verb.
(34)
Amis SVC examples a. Ma-herek=to kako a ma-lafi. av-finish=asp 1s.nom lnk neut-dinner ‘I finished eating dinner.’ b. Ma-rara k-o fafahian a mi-fihon. neut-slow nom-cn woman lnk av-put.on.makeup ‘The woman puts on make-up slowly.’ c. Tanam-en=ako a pa-rakat k-o-ra tosiya. try-uv=1s.gen lnk caus-walk nom-cn-that car. ‘I will try to drive that car.’
Wu (Reference Wu2006a) mentions that these constructions all have an actor pivot, either actor of an intransitive verb or that of an AV verb, in the linked core.
Like the Atayal examples discussed in Table 26.2, the evaluative construction in Amis also shares the following two features with SVCs: sharing of TAM information and sharing of polarity value. First, in the evaluative construction as well as in the SVC, only the first verb can be marked with TAM information.
(35)
Sharing of TAM information in SVCs and the evaluative constructions in Amis a. Mi-sawad=to kako a mi-mali av-quit=asp 1s.nom lnk av-ball ‘I have quit playing ball.’ b. *mi-sawad=to kako a mi-mali=to av-quit=asp 1s.nom lnk av-ball= asp c. Fangcal=to a nengneng-en k-o-ra coka. good=asp lnk see-uv nom-cn-that picture ‘That picture looks good now.’ d. *Fangcal=to a nengneng-en=to k-o-ra coka. good=asp lnk see-uv=asp nom-cn-that picture
Second, only the first verb in the named constructions can be conjugated with the non-finite markers when following the negator caay ‘not’.25
(36)
The negative sentences of SVCs and the evaluative constructions in Amis a. Caay ka-safon-en nira k-o tireng a mi-ngingoy neg nfin-soap-uv 3s.gen nom-cn body lnk av-shower ‘He didn’t use soap to wash (his) body.’ b. *Caay ka-safon-en nira k-o tireng a pi-ngingoy neg nfin-soap-uv 3s.gen nom-cn body lnk nfin-shower c. Caay ka-fangcal a nengneng-en kora coka. neg nfin-good lnk see-uv nom-cn-that picture ‘That picture does not look good.’ d. *Caay ka-fangcal a ka-nengneng-en kora coka. neg nfin-good lnk nfin-see-uv nom-cn-that picture
However, unlike the SVCs in (34), it is the second verb in the evaluative construction that controls the case-marking pattern of the sentence, but not the first verb. As shown in (34c), even though the second verb appears in the AV form, the arguments in the linked core are case-marked with the UV pattern, following the first verb. In other words, the second verb in the SVCs has no voice function. Nevertheless, the arguments of the linked core in the Amis evaluative constructions always follow the case-marking pattern of the UV verb, as seen in the comparison between (37a) and (37b).
(37)
The case-marking pattern of the Amis evaluative construction a. Tada-fangcal a nengneng-en=ako k-o-ra coka. very-good lnk see-uv=1s.gen nom-cn-that picture ‘I saw the picture and thought that it was very beautiful.’ b. *Tada-fangcal kako a nengneng-en k-o-ra coka. very-good 1s.nom lnk see-uv nom-cn-that picture c. Tada-fangcal k-o-ra coka a nengneng-en. very-good nom-cn-that picture lnk see-uv ‘That picture looks really beautiful.’
Yeh (Reference Yeh, Zeitoun, Teng and Wu2015) in particular explores the answer to the question of why the second verb in the evaluative construction has to be marked by the UV form, but not the AV. She proposes that it is the undergoer that is evaluated in the construction, and this pragmatic function accounts for the UV marking of the second verb. However, the RRG approach might offer an even more straightforward answer.
Following the RRG analysis, the first verb in (31) will have an undergoer instead of an actor, and the voice sequence in the Amis evaluative construction should be a neutral voice followed by a UV verb. The shared arguments in such sentences play the role of undergoer (undergoer of the intransitive verb and undergoer of a UV verb). Hence, it is quite reasonable to maintain the syntactic status of the undergoer in both cores. If the second verb takes the actor voice, the macrorole status of the undergoer will be removed, as we have seen in the function of the voice discussed earlier. The seeming paradox in Yeh’s study lies in her treatment of the only argument of the first verb as an actor or a unified thematic role, but the RRG perspective can provide an account that resolves the paradox.
As for the SVC examples, the shared argument might play different semantic roles in the two linked verbs, but this role is always an actor in the second verb. This may explain why the second verb bears AV marking though the AV does not have any voice function.








































