15

Nominalizations, case domains, and

restructuring in two Amazonian languages
Andrés Pablo Salanova and Adam Tallman

One of the most influential points made by Remarks on Nominalization

concerns the parallelism between nominal and verbal projections, an idea
that opened the way to the development of X̄-theory and to much subsequent
work on the decomposition of lexical categories.
The importance of this point is lost on many linguists that work with
languages of lowland South America and elsewhere, where the parallel
between the projections headed by nouns and those headed by verbs is often
a given. Though rarely addressed explicitly in the literature, researchers
working in the Amazon would not bat an eyelash at the identity between
clausal and nominal categories of case and agreement, for instance: Systems
where nominal possessors are expressed in the same way as ergative subjects,
or where they are expressed in the same way as the absolutive arguments of at
least some verbs, are in this area of the world, if not outright the norm, at least
as widespread as cases where they are completely distinct.
We illustrate the identity in the inflection of nouns and verbs with the two
langauges that are examined in this chapter.¹
In Mẽbêngôkre (Jê), person indices indicating the possessor of inalienable
nouns are the same as those used for objects of most verbs and for subjects of
stative and nonfinite verbs. The morphological case corresponding to these
indices (as well as the morphologically unmarked case of nouns appearing
in the same function) could equally correctly be called absolutive or genitive.

¹ Following standard practice for these languages and others in the region, we treat bound person
indices as pronouns rather than as agreement. When we discuss case categories we are talking as much
about the form of these indices as about any overt case marking on independent noun phrases.

Andrés Pablo Salanova and Adam Tallman, Nominalizations, case domains, and restructuring in two Amazonian languages
In: Nominalization: 50 Years on from Chomsky’s Remarks. Edited by: Artemis Alexiadou and Hagit Borer,
Oxford University Press (2020). © Andrés Pablo Salanova and Adam Tallman.
DOI: 10.1093/oso/9780198865544.003.0015
364 ́     

In the glosses, we leave it unmarked (as opposed to the explicit marking of

, , and other cases):

(1) i-kamy
1-brother
‘My brother.’

(2) ga i-pumũ
2 1-see.
‘You see me.’

In Chácobo, subjects of transitive verbs are expressed by means of the same

case category that is used to indicate possession within a noun phrase:

(3) yoʂa= 0 tsi kiá tʃaȿo tsaya=kɨ

woman= 5  deer see=
‘The woman saw the deer.’

(4) yoȿa= 0 tʃaȿo

woman= deer
‘the woman’s deer’

The absence of a strong morphological differentiation between nouns and

verbs as far as marking of some dependents goes does not mean that the
languages do not differentiate clearly between nouns and verbs, however. In
Mẽbêngôkre, for instance, the difference between nominal and verbal lexemes
hinges on the presence, in the paradigm of the latter, of two distinct forms, one
finite and the other nonfinite (actually, as we will see later, nominal), that are
associated with two different alignment patterns. In Chácobo, different clause-
typing morphemes are used according to whether the predicate is verbal or
nonverbal.
In fact, Chácobo and Mẽbêngôkre use nominalization to a far greater extent
than what is known from the languages of Europe. Event nominalizations are
used whenever verbal clauses are subordinated. This includes many structures
that don’t involve subordination in other languages, such as manner
modification.
In Mẽbêngôkre, for instance, a verbal predicate subordinated to a predicate
of direct perception appears in its nominal form; in such a form, its first
 ,  ,   365

argument is expressed by means of a genitive or absolutive person index, while

its ergative argument is expressed by an oblique case, much in the way the rare
external argument of a nominal predicate would (see (6)):

(5) djãm nẽ ga [kubẽ kute ibê djudjê oàkĩnh] pumũ?

  2 barbarian 3 1 bow steal. see.
‘Did you see the white guy walking away with my bow?’

(6) djãm nẽ ga [kubẽ kum rop pyma] pumũ

  2 barbarian 3 dog fear() see.
‘Did you see the white guy (being) afraid of dogs?’

In this chapter, we examine two constructions that involve embedding a

nominalizations under another predicate for aspectual effect. Though the
identity in inflection between nominal and verbal predicates makes it difficult
to untangle such complex constructions, embedding of nominalized clauses
may be seen in the unusual alignment patterns that arise in them. We claim
that category-changing morphology has a crucial role in creating separate
case-assignment domains.
The structure of this chapter is as follows. In Section 15.1 we introduce the
main empirical phenomenon examined here as it has been dealt with by
previously published literature. Section 15.2 offers a description of the relevant
facts in Mẽbêngôkre, a Northern Jê language spoken in central Brazil, and
sketches an analysis of the construction and of the nominals that appear in it.
Section 15.3 offers a description of similar facts in Chácobo, a Panoan lan-
guage spoken in northern Bolivia. Section 15.4 proposes a general analysis of
constructions with two case domains, and reflects on the relevance of nomin-
alization for the delimitation of case domains.

15.1 Complex constructions and case domains

A number of theories of case rely on the idea that case is assigned in specific
structural domains. This is no doubt most obvious in dependent-case theories
where two noun phrases compete for case within a single domain, such
as Marantz (1991); Bittner & Hale (1996); Baker (2015); among others. It is
366 ́     

also implicit in more traditional theories of case involving some formalization

of Burzio’s generalization (Burzio, 1986), in the sense that the possibility of
accusative case assignment by a verb, which is assigned locally to a comple-
ment, is dependent on there being a thematic relation between the verb and
another argument, which is hence also local to the assigner (mutatis mutandis
for ergative constructions). In the functional-typological literature, the notion
of a case-assigning domain is evoked implicitly when it is argued that case
marking serves to differentiate participants within a clause (Payne, 1997: 140;
Kibrik, 2012: 212; Comrie, 2013; inter alia). When we talk about a case
domain, we mean something equivalent to a clause in the sense just sketched,
and will use the two terms interchangeably. However, in certain circumstances
a prima facie simple clause will actually contain more than one case domain.
For concreteness, we sketch of a theory of case involving competition for
case within a clause loosely based on Marantz (1991).
In the simplest case, one has a clause with a transitive verb, where the case
domain (CD), which ex hypothesi corresponds to the clause, contains the
verb’s two arguments.

(7) [NP V NP]CD

The presence of two argument chains in the same case domain triggers the
assignment of a ‘dependent case’. That is, one of the two arguments will
receive a case which is only assigned when there is another argument chain
present in the same domain. Two options, specified as a feature of the head of
the domain, exist for dependent case assignment: Dependent case can be
assigned to the ‘lower’ argument (with P grammatical role), yielding what is
normally called nominative-accusative alignment (i.e. accusative is the
dependent case, as it only appears when there is more than one argument in
the clause), or dependent case can be assigned to the ‘higher’ argument (with
A grammatical role), yielding what is called ergative-absolutive alignment
(i.e. ergative is the dependent case). Nominative and absolutive cases, to the
extent that they are overtly expressed at all, are elsewhere cases (not necessarily
the same as the default case) that are assigned after the assignment of
dependent case.
Across nominative-accusative and ergative-absolutive languages, but
primarily in the latter, dependent case theories need to contend with
 ,  ,   367

constructions where dependent case does not surface, even in the presence of a
co-argument. This situation can be seen in the following Basque examples,
taken from Laka (2006), which are an instance of a more general situation
observed by Coon & Preminger (2017) regarding aspect-based splits:

(8) a. Emakume-a-k ogi-a jaten du.

woman-- bread- eating has
‘The woman eats the bread.’
b. Emakume-a ogi-a jaten ari da.
woman- bread- eating  is
‘The woman is eating the bread.’

(9) a. Emakume-a hurbiltzen da.

woman- approach is
‘The woman gets closer.’
b. Emakume-a hurbiltzen ari da.
woman- approach  is
‘The woman is getting closer.’

Dependent case (ergative) shows up on the subject in the aspectually

unmarked construction (8a) whenever another noun phrase argument is
present in the clause. Absolutive case is unmarked. Another pattern, specific
to progressive aspect, is one where both arguments have unmarked (absolu-
tive) case. In (8b), the way each individual argument is marked does not
depend on the presence of another argument in the clause, as can be seen by
comparing that transitive sentence to the intransitive (9b).
We begin our exposition of what we consider a typical analysis of such splits
by considering Laka’s (2006) analysis of Basque progressives. Ergative align-
ment in the aspectually unmarked construction is straightforwardly dealt with
by a domain-based theory of case: the A argument is assigned dependent
ergative case in an ergative clause because it is in a domain that contains
another NP, as shown schematically in (10).

(10) One domain, dependent ergative is assigned to A argument:

[NPA-erg V NPP]CD
368 ́     

In Basque, it is possible to assume that marked case is assigned to the higher

argument in all case domains, i.e., alignment is always ergative in simple
clauses.
For progressives, Laka follows Hualde & Ortiz de Urbina (1987) in claiming
that ‘ari [ . . . ] is a main verb with its own auxiliary which may take a
nominalized clause as its complement’ (Laka, 2006: 174; and Hualde & Ortiz
de Urbina, 1987: 428). This allows her to claim that ‘[t]he contrast [between
(8a) and (8b)] results from the fact that the ari progressive involves a biclausal
syntactic structure [ . . . ]’ (Hualde & Ortiz de Urbina, 1987: 428). This gives a
structure as in (11) to the sentences above. In our terms, (8b) is partitioned
into two case domains, as in (11b):

(11) a. [[emakumeak]DP [[ogiajaten]VP du]]IP

b. [[emakumeai]DP [[[[proi ogia jate-]VP –n]PP air]VP da]]IP

The claim that the split is caused by a difference in structure would be

circular if no evidence independent of the case marking itself were brought to
bear in establishing the domains. Thus, Laka (2006) provides the following
arguments for the existence of such a biclausal structure in Basque (some
arguments ultimately come from Hualde & Ortiz de Urbina, 1987):

1. ari is a verb meaning ‘be engaged’; this can be seen in its morphology
and in that it can be nominalized with -tze and can receive aspectual
suffixes -ko , -tzen , and -tu .
2. Constructions with ari always co-occur with the intransitive auxiliary,
irrespectively of the transitivity of the lexical verb with which it is
combining. In the unmarked aspect, the auxiliary agrees in transitivity
with the main verb.
3. ari selects a PP headed by -n; this is seen not only in the morphology, but
also in the fact that the complement of -n can be a noun; furthermore, -n
may be replaced by other Ps without resulting in ungrammaticality.
4. In western dialects, the verb ari may be replaced by two other verbs ibili
and egon (both unaccusative and both taking a locative complement),
with similar meaning.

In sum, the morphology is strongly suggestive of a biclausal analysis.

However, not all of the syntactic facts point in that direction. There are
some reasons to believe that the structure of (8b) does not involve two
 ,  ,   369

syntactic domains (or, in Laka’s terms, is not biclausal); the following counter
arguments are mentioned by Laka:

1. When subject wh-movement occurs, there is inversion of verb + auxil-

iary to second position. In progressive constructions, it is possible to
move just ari + da (supporting a biclausal approach) or V + ari + da
(supporting a restructuring approach).
2. With certain verbs that have dative subjects or objects, auxiliaries will
agree with both the higher absolutive argument and the dative argument
(see Hualde & Ortiz de Urbina, 1987, (8d’,f ’)).
3. In some eastern varieties, progressive constructions with ari have erga-
tive rather than double absolutive alignment.

Laka’s answer is to propose that restructuring occurs in those dialects where

ergative-absolutive alignment arises, but argues against the other two phe-
nomena being reasons to be forced to accept a monoclausal structure. We do
not need to go into the merits of the proposal for Basque here. What we are
interested in is the question that such argumentation raises, which is central in
this chapter, namely the extent to which case domains can be defined inde-
pendently of the case phenomena that they are called upon to explain. If they
are not defined independently, then the analyses based on them are circular
and amount to a redescription of the facts, being at best a good schema to
describe the etymology or historical evolution of a construction that is syn-
chronically no more complex than other clause types (see, e.g., Gildea, 2008).
But even if there is independent evidence for the domains that are relevant for
case, one must ask the question of how much convergent evidence is needed in
order to claim that a given construction has two distinct structural domains.
We return to this question in the last section of this chapter, after considering
the evidence for biclausality in two constructions in Mẽbêngôkre and
Chácobo.

15.2 Complex constructions in Mẽbêngôkre

Mẽbêngôkre is a Jê language spoken by approximately 10,000 people in central

Brazil. Mẽbêngôkre has clauses with both ergative and accusative alignment, as
well as clauses with a nominative-absolutive alignment. The language is not
alone in the family in having these alignments (cf. Gildea & de Castro Alves,
2010, where nominative-absolutive is analyzed as one of the basic monoclausal
370 ́     

patterns, contrary to what we do here). Embedded clauses other than quoted

speech are always headed by nonfinite (or nominal) forms of the verb, and are
always ergative, while in independent clauses one may find all three types of
alignment. The following examples illustrate these various surface patterns in
independent clauses: (12) show ergative-absolutive, (13) show nominative-
accusative, while (14) show nominative-absolutive.

(12) a. A O.V
ije krẽn kêt
1 3.eat.. 
‘I haven’t eaten it.’
b. S-V
i-tẽm kêt
1-go. 
‘I don’t go.’

(13) a. A O-V
ba ku-krẽ
1 3-eat..
‘I (’m going to) eat it.’
b. S
ba tẽ
1 go.
‘I (’m going to) go.’

(14) a. A O.V
ba krẽn o=nhỹ
1 3.eat.. OBL=sit..
‘I’m eating it.’
b. S S-V
ba i-tor o=dja
1 1-dance. =stand..
‘I’m dancing.’

Nominative-absolutive alignment is associated with the progressive con-

struction, initially described by Reis Silva (1996), and reexamined in Salanova
(2008). In this construction, while the person marker on the verb is always
absolutive, there is a further nominative pronoun which either stands for the
external argument or redundantly cooccurs with an absolutive index if the
 ,  ,   371

latter indexes an intransitive subject (nominative is unmarked on nonpro-

nominal NPs). That is, while in (14a) the index on the verb and the nominative
subject index two different arguments, in (14b) the reference of the two person
indices is the same.
The morphosyntax of this construction resembles the Basque progressive
construction in many ways:

1. The lexical main verb is in a nonfinite or nominal form, glossed here as

; the nominal character of the nonfinite form is argued for in Salanova
(2007).
2. This form is subordinated to the auxiliary by means of an adposition, o,
glossed as  and used variously as an instrumental (with transitive
verbs) or as the marker for applicative themes.
3. The auxiliary is in its finite form, glossed here as .

The differences that the construction in Mẽbêngôkre has with its equivalent
in Basque are not directly relevant to the definition of case domains, but we
note them here:

1. Auxiliaries in Mẽbêngôkre are chosen from a small set, and encode

posture (sit, stand, etc.).
2. Absolutive is distinct from nominative: It is in fact the same as genitive,
as was mentioned at the beginning of the chapter; the reason for this is
that the two case domains in Mẽbêngôkre have distinct elsewhere cases,
contrary to what happens in Basque. The construction can nevertheless
be considered as parallel to the Basque progressive because neither of the
cases are dependent: Both are the case that is assigned in a domain where
no other argument chain is present.
3. Differently from Basque, intransitive subjects are not empty in the lower
clause: an absolutive person prefix, coindexed with the nominative
subject of the matrix clause, is obligatory on subordinate verbs.
4. Finally, an ergative pronoun may be redundantly present next to the
nominative subject of a transitive clause even in the progressive con-
struction. This is a general property of ergative in this language.

We sketch a formal representation of (14a), on which we will attempt to

represent the various properties that we identify in the construction.
372 ́     

(15)
VFP
thematic relation

SubjNOM VʹF

VʹF
bai …
HIGHER CASE DOMAIN
PP VF

LOWER CASE DOMAIN nhỹ

VNP P

VʹN
control
o
SubjERG

ijei ObjGEN VN

∅ ˜
kren

Arguments in favor of this structure are numerous.

The postural verbs used in the Mẽbêngôkre progressive construction are
chosen from a small but possibly extensible class that includes at least the
following: nõ/nhikwã (lie sg./pl.), dja/ ku’ê (stand sg./pl.), nhỹ/krĩ (sit sg./pl.),
wajêt/jarij (hang sg./pl.), tẽ/mõ (go sg./pl.). Though the determination of
postural verb is somewhat conventionalized for many verbs, some choice is
possible, and this choice is based exclusively on the position of the subject
while carrying out the action.² Minimal pairs for intransitive and transitive
verbs may be seen in (16) and (17), respectively.

(16) a. a-bãm nẽ õt o=nõ

2-father  3.sleep. =lie..
‘Your father is sleeping (on a bed or mat).’
b. a-bãm nẽ õt o=wajêt
2-father  3.sleep. =hang..
‘Your father is sleeping (on a hammock).’

² One could insist that the determination of the postural verb according to the lexical requires a
selectional relation. This is not a problem: a thematic relation may be said to exist between the postural
verb and the lexical verb. The adposition o would be transparent to thematic relations in this case.
 ,  ,   373

(17) a. ba pi’ôk jarẽnh o=nhỹ

1 book say. =sit..
‘I’m reading (sitting down; i.e. studying).’
b. ba pi’ôk jarẽnh o=dja
1 book say. =stand..
‘I’m reading (standing up; i.e. lecturing).’

A further index of a selectional relation between postural verbs and subjects

may be seen in number alternations. A few verbs in Mẽbêngôkre come in
mostly suppletive pairs that are chosen based on the number of one of the
arguments or on iterativity or durativity of the action. Postural verbs are
among the verbs that reflect the number contrast. In progressive construc-
tions, the subject of the action is the only relevant argument for determining
the choice of postural verb based on number (cf. (18)). In cases where the verb
is chosen based on iterativity or durativity rather than on number of one of the
participants, number reflects the iterativity of the overarching situation, in
which the subject is repeatedly engaged. This may be seen in (19):

(18) a. mẽ’õnire nẽ mry bôr o=dja

woman.one  meat roast.. =stand..
‘A woman is/was roasting meat.’
b. mẽnire nẽ mẽ mry bôr o=ku’ê
woman   meat roast.. =3.stand.
‘Women are/were roasting meat.’

(19) a. i-bãm nẽ ujarẽnh o=nhỹ

1-father  3..say. =sit..
‘My father is / was telling a story / stories.’
b. i-bãm nẽ ujarẽnh o=krĩ
1-father  3..say. =3.sit.
‘My father was sitting for a long time / sits frequently to tell stories / a
story.’

In short, the evidence strongly suggests that the subject of the clause is also
the postural verb’s logical subject. One further morphosyntactic fact confirms
this. Postural verbs are intransitive, and, like other intransitive verbs, do not
inflect for person in their finite forms, as the person indices in finite verbs are
for the P argument. In nonfinite forms, required in subordination, in negation
374 ́     

and a few other contexts, the postural verbs inflect for the person of the
subject, like any intransitive verb, irrespective of the transitivity of the lexical
verb:

(20) a. ba pi’ôk jarẽnh o=nhỹ

1 book say. =sit..
‘I’m reading.’
b. ba pi’ôk jarẽnh o=i-nhỹr kêt
1 book say. =1.sit.. 
‘I’m not reading.’

Contrary to the nominative subject, the absolutive noun phrase is an

argument of the embedded lexical verb, which is in a nominal form.
Diagnostics similar to those applied to the auxiliary can be applied to the
lexical verb to show that there is a selectional relation between it and the
absolutive argument. Beyond the identity between this argument and the S or
P argument of a verb in a nonprogressive construction, number on the lexical
verb scopes over it, as the following examples show:

(21) a. ba tep kur o=nhỹ

1 fish eat.. =sit..
‘I’m eating (many) fish.’
b. ba tep krẽn o=nhỹ
1 fish eat.. =sit..
‘I’m eating (one) fish.’

To conclude, there is strong evidence that points to considering that the

Mẽbêngôkre progressive (‘nominative-absolutive’) construction is one where
the S or P is an argument of the lexical verb, while the A or S is an argument of
the postural verb or ‘auxiliary’, an S argument being indeed an argument
of both. This is suggestive of a control construction.³ A morphological quirk of
Mẽbêngôkre forces a controled ergative subject to be zero, while a controled
absolutive subject appears as a genitive person prefix on the lexical verb.
Are there any pieces of evidence that point to a reanalysis, or at least to
the embedded clause being something other than a normal nominalization?
There are two places to look: (i) test the nominalized embedded clause for the

³ For arguments that the Mẽbêngôkre locative construction also involves control, see Beauchamp
(2017).
 ,  ,   375

properties that are associated to objects (i.e. compare its external properties to
those of undisputed nominal objects), (ii) verify whether the nominalized
embedded clause has the same constructional possibilities as nominalizations
elsewhere. One could also look for the traditional evidence of restructuring
(apparent cross-clausal agreement, case assignment, or movement phenom-
ena). By our discussion so far in this section, however, it should be apparent
that the latter type of evidence is not found in Mẽbêngôkre.
Salanova (2015) examines the properties of objects as opposed to adjuncts
in detail. If we apply the diagnostics from that paper, we find that there are a
number of differences between the embedded nominalized clauses in the
progressive construction and simple nominal objects. First of all, nominal
objects can be moved to the first position of the clause for contrast, stranding
the adposition; Nominalized embedded clauses cannot without a change in
meaning (see (23)):

(22) a. ba kẽn o=nhỹ

1 stone =sit..
‘I’m sitting with a stone.’
b. kẽn nẽ ba o=nhỹ
stone  1 =sit..
‘I’m sitting with a stone.’

(23) a. ba tep krẽn o=nhỹ

1 fish eat.. =sit..
‘I’m eating fish.’
b. (ije) tep krẽn nẽ ba o=nhỹ
1 fish eat..  1 =sit..
‘I’m sitting with the fish eaten (by me).’

Furthermore, while the subject of nominalized transitive clauses, if unex-

pressed, is interpreted as a generic agent, the subject of nominalized clauses in
the progressive construction is necessarily referential, even if controled by the
subject of the higher clause. Compare the complement clauses in (24) with
(25), where no possibility of having a generic subject exists:

(24) a. Ba [a-bãm kute tep djonhwỳr] pumũ.

1 2-father 3 fish pierce. see.
‘I saw your father fishing (with arrow)’, ‘I saw the fish your father
caught with an arrow.’
376 ́     

b. Ba [tep djonhwỳr] pumũ.

1 fish pierce. see.
‘I saw (people) fishing with arrows’, ‘I saw a fish caught with an
arrow.’

(25) Ba tep djonhwỳr o=dja.

1 fish pierce. =stand..
‘I’m fishing with arrows.’

In a sense, this diagnostic complements (23): If the nominalized clause is

moved to the beginning of the clause, a generic subject interpretation becomes
possible (see (23b)); however, the progressive meaning is lost.
To sum up, in Mẽbêngôkre the morphology tells us, apparently even more
clearly than in Basque, that a particular construction with unusual alignment
is synchronically complex. Some syntactic facts seem to argue against this,
however, though it is not clear whether those facts diagnose complexity of
structure. What is seen in (23) most likely has to do with the fact that the first
position of the clause is reserved for referential expressions, as it is a topic
position. What is seen in (25), on the other hand, only tells us that there is
obligatory control in the progressive construction; it is not currently clear
whether obligatory control should be used as a diagnostic for restructuring
into a single domain.
We postpone evaluating the weight of the various arguments surrounding
the interpretation of the Mẽbêngôkre progressive construction until we have
considered the somewhat more ambiguous case of Chácobo.

15.3 Complex constructions in Chácobo

Chácobo is a Panoan language spoken by approximately 1,500 people in

northern Bolivia. Simple clauses in Chácobo display ergative alignment with
full noun phrases, where ergative is marked as final high pitch, and accusative
alignment with pronouns, where accusative is marked with a suffix. This is
illustrated in the following examples:

(26) a. mi paβí=kɨ
2. dance=.
‘You were dancing / You danced.’
 ,  ,   377

b. tʃaʂo paβí=kɨ
deer. dance=.
‘The deer was dancing / danced.’
c. tʃaʂo=0 mi-a tsáya=kɨ
deer= 2- see=.
‘The deer saw you / was watching you.’

Verbal predicate constructions in Chácobo are characterized by having a

verbal root and a clause-type morpheme. In examples (26a–c) the clause-type
morpheme is kɨ ‘declarative, past’. All clause-type markers encode clause-type
(declarative, imperative, interrogative, reportative), while some also encode
tense (e.g. past, nonpast, future). Order in verbal predicate constructions is
fairly free, but the subject cannot follow the clause-type morpheme.⁴
Clauses where the predicate is nonverbal (henceforth ‘nonverbal predicate
[NVP] constructions’) are clearly distinct in Chácobo. Like verbal predicate
constructions, NVPs are marked with a clause-type morpheme. The clause-
type morphemes of NVPs are not the same as those from verbal predicate
constructions, although there is some overlap in form and function. In con-
trast to verbal predicate constructions, the canonical order of NVPs is rigidly
predicate-subject, with the clause-type morpheme occuring between the predi-
cate and the subject. Furthermore, noun phrases do not receive ergative case in
NVPs, and pronouns never appear in the unmarked nominative case, but
rather appear with case markers that are nearly homophonous to the accusa-
tive forms used in verbal predicate constructions:⁵

(27) a. tʃaʂo ʂo mi-a

deer . 2-
‘You are a deer.’
b. tʃaʂo ki mi-a
deer . 2-
‘You will be a deer.’

Certain properties of NVPs can be seen in an intermediate construction,

exemplified in (28), which displays predicate-subject order and no ergative

⁴ If the subject comes between V and the clause-type marker, an auxiliary may appear, as in (32).
⁵ We gloss the marker on pronouns in NVPs , following the argument in Tallman (2018) that
the {-a} is inserted there in order to satisfy bisyllabic minimality.
378 ́     

case marking on full noun phrases. We refer to this construction here as the
V-C-Subj construction.

(28) tʃaʂo tsaya =ki honi

deer. see =. man.
‘The man sees the deer.’

Given the rigid predicate-subject order also found in NVPs, it would be

natural to suppose that the verb in the V-C-Subj construction is a type of
nonverbal predicate, despite the absence of clearly nominalizing morphology.
The case facts in this construction are also suggestive in that direction: an
analysis of V-C-Subj constructions where they are assimilated with NVPs is
compatible with the idea that two domains exist in the construction, in a way
that we will make explicit below.
The most plausible biclausal analysis one could apply to V-C-Subj con-
structions is one where the verbal predicate appears embedded inside a
nonverbal predicate construction. The structure we propose for a sentence
such as (28) would be (29), where we consider the clause marker to be a
predicative element that heads the NVP. The label X given to the head of the
lower domain is a placeholder, and nothing hinges on it at present. Note that
we are also agnostic as to the mechanism that coindexes the overt subject of
the predicative phrase with the unexpressed subject of the XP; presumably the
element X turns precisely that argument into the predicate’s referential
argument.

(29) PredP
higher case domain
Predʹ Subj
lower case domain honii
XP Pred

=ki
PROi Xʹ

VP X

t∫aʂo tsaya

This analysis solves the problem of neutral case assignment by making the
A subject the subject of an NVP construction, and having the subordinated
 ,  ,   379

clause define a separate domain for case assignment. Since NVPs are copular
constructions that do not assign ergative case, the subject of this construction
stays in the neutral absolutive. In contrast to the analysis presented for
Mẽbêngôkre, we propose that Chácobo has no main verb in the V-C-Subj
construction, only a subordinate verb. Analyzing the dependent-absolutive
construction as an NVP entails as much.
Before moving on to the syntactic arguments in favor of such an analysis, we
consider whether, like in Mẽbêngôkre, the morphology presents prima facie
reasons for us to believe that the idea that V-C-Subj constructions are complex
is on the right path. The evidence is suggestive but inconclusive.
The clause-type markers across the verbal predicate, nonverbal predicate,
and V-C-Subj constructions overlap in form and function, but they are not the
same. An overview of the clause-type markers across the three constructions is
presented in Table 15.1.
There are between two and three pairs of clause-type markers that provide
evidence that the V-C-Subj construction should be treated as a type of
NVP. The partial identity in the reporative forms (kiá and Ɂi kiá) can be
considered evidence of identity between V-C-Subj constructions and NVPs, as
Ɂi can be shown from other constructions to be a subordinator. The pair Ɂi ní
and ní of interrogative forms is evidence for the same reason, even if Ɂi does not
occur consistently across all clause-type markers in the V-C-Subj construction.
On the other hand there are also three clause-type markers that are identical
in form and nearly identical in meaning across the verbal predicate construc-
tion and the V-C-Subj construction: kɨ ‘declarative, past, anterior’, ní ‘inter-
rogative, remote past’, and Ɂá ‘declarative, past, anterior’. It is nevertheless
relevant that an aspectual (‘anterior’) rather than temporal reading is associ-
ated with these markers in the V-C-Subj construction.

Table 15.1 Clause-type markers across different clause-types

Verbal predicate V-C-Subj Nonverbal predicate

Declarative kɨ ‘declarative, past’ kɨ ‘declarative, anterior’

ki ‘declarative, non past’ ki ‘declarative, future’
ʂo ‘declarative,
present’
Interrogative ní ‘interrogative, ní ‘interrogative, remote past’ ní ‘interrogative’
remote past’ Ɂá ‘interrogative, anterior’
Ɂá ‘interrogative, past’ Ɂi ní ‘interrogative, non past’ (ní ‘interrogative’)
Imperative wɨ ‘imperative’
Reportative Ɂi kiá ‘reportative’ kiá ‘reportative’
380 ́     

The fact that apparently identical clause-type markers have very different
interpretations depending on whether they occur in verbal predicate construc-
tions or V-C-Subj constructions has been discussed in Tallman (2018) and
Tallman & Stout (2016). Here we provide a brief synopsis.
In verbal predicate constructions, the marker kɨ encodes past tense in the
sense that it relates utterance time to topic time. In its default interpretation it
advances narrative time, although it cannot be considered perfective because it
does not always have this function (Tallman, 2018; Tallman & Stout, 2018).
These properties of kɨ are shown in the following examples.

(30) a. hatsi ɨ-a=rí pi=kɨ

then 1-= eat=.
‘Then, I ate as well.’
b. pi=Ɂá hɨnɨ nami-na=Ɂá=ka ɨ
eat=. chicha thick-.=.= 1
bótɨ=kɨ
descend=.
‘After I ate, I lowered the chicha that had thickened.’
c. ha-tó=bɨta=ʂó ɨ-a=rí náka náka=kɨ
3-== 1-=too chew~chew=.
tsɨmo=kana ha =kɨ
darken=go.. 3 =.
‘With them I was chewing (on the yuca) as well, while it got darker.’

However, in V-C-Subj construction kɨ can only have an anterior or relative

past interpretation. It cannot advance narrative time. The examples in (31),
taken from a narrative, illustrates this:

(31) a. bɨpana=0 panɨ =kirí i=ní=kɨ

large_house= wall = be==.
‘He (the vampire) was beside the wall of the large house.’
b. . . . nii nobá=na=ki yabo-ko=kɨ kiá ʂatʃi
stop/stand 1.== tie-=.  grass
‘He was standing, the stock of grass already tied around his arm [lit.,
our body part].’

Details on all the semantic differences between kɨ and Ɂá in verbal predicate

constructions as opposed to V-C-Subj constructions can be found in Tallman
 ,  ,   381

(2018: 715–845). A full analysis of these semantic differences and how they
relate to the constructions in which they appear is beyond the scope of this
chapter. Note, however, that if due to their semantic differences we analyze kɨ
and Ɂá as pairs of homophonous but semantically distinct morphemes, then
the morphological evidence points less ambiguously to identifying the V-C-
Subj construction with the NVP.
To summarize, rather than clearly supporting a biclausal analysis, the form
of clause-type markers suggest that the V-C-Subj construction occupies some
intermediate status between verbal clauses and NVPs. There is suggestive
syntactic and semantic evidence that points to a biclausal analysis of the
former, however, which we will go through now.
The first argument comes from fronting. Chácobo has a VP-fronting
construction where the NP object plus the verb stem front to a focused
position (see Tallman, 2018: 322–7, for discussion). Prima facie both V-C-
Subj and the verbal predicate constructions should allow the VP to front: Both
involve a VP, and there is no clear functional or pragmatic reason for V-C-
Subj constructions to behave any differently from verbal predicates. However,
only the verbal predicate construction allows VP-fronting:⁶

(32) a. yoʂa= 0 tsi kiá tʃaʂo tsaya=kɨ

woman= 5  deer see=.
‘The woman saw the deer.’
b. [tʃaʂo tsaya] tsi kiá yoʂa= 0 wa=kɨ
[deer see] 5  woman= =.
‘The woman saw the deer / ~ As for the seeing of the deer, the woman
did it.’

(33) * tʃaʂo tsaya tsi kiá =kɨ yoʂa

deer see 5  = woman
Intended: ‘it is said that the woman has been seen by the deer’

The landing position for fronted constituents is in the matrix clause, as is

signaled by the position of the tense morpheme (tsi ‘position 5’). VP-fronting,
being clause bound, cannot place the VP outside this clitic. Even if the tense
clitic seeks to be in second position, it cannot break up the subordinate clause

⁶ Note that the morphemes tsi and kiá are Wackernagel clitics which always occur following the first
constituent (NP or VP) (Tallman, 2018).
382 ́     

to appear after the subordinated VP. This explains the ungrammaticality

of (33).
An alternative analysis where V-C-Subj constructions are posited to have
the same structure as regular verbal clauses would not be able to capture this
restriction on VP-fronting other than by stipulation.
A second argument for a biclausal analysis of V-C-Subj constructions
comes from the observation that there is tighter constituency between a verb
and its complement, as attested by the lack of object-subject permutation in
such constructions.
In Chácobo verbal predicate constructions with ergative alignment, the
subject and object can be variably ordered with respect to each other, allowing
SOV, OSV, OVS orders (SVO is possible but somewhat marked and less
common). Examples of SOV and OVS are provided in (32) above. An example
of OSV can be seen in (34):

(34) tʃaʂo tsi kiá yoʂa=0 tsaya=kɨ

deer 5  woman= see=.
‘The woman saw the deer.’

However, in V-C-Subj constructions, not only is the subject obligatorily

after the clause-type marker, but in addition the object and subject cannot be
freely permuted. The construction displays a fixed OVS order. VOS is com-
pletely banned and VSO requires an utterance pause and is most felicitously
translated into two sentences. These facts are illustrated in (35).

(35) a. * tsaya=ki tʃaʂo yoʂa

see=. deer woman
Intended: ‘The woman sees the deer.’
b. #tsaya=ki yoʂa . . . tʃaʂo
see=. woman deer
Intended: ‘The woman has seen it . . . The deer.’

The impossibility of the subject intervening between the object and the verb
in V-C-Subj constructions, as opposed to the free order in standard verbal
predicate constructions, receives a straightforward explanation if verb and
object are contained in a separate domain that cannot be interrupted by the
subject.
A third argument for a biclausal structure comes from the syntax and
semantics of the reportative marker. The reportative marker can occur in
 ,  ,   383

the verbal predicates, nonverbal predicates, and V-C-Subj constructions. Its

position and scope in these three types of clause reveals that V-C-Subj
constructions align with NVPs.
In verbal predicate constructions the reportative must occur before the
obligatory clause-type rank morpheme. For instance, the sentence in (36) is
a verbal predicate construction; the reportative marker occurs before the verb
and the clause-type morpheme.

(36) hatsi kiá yobɨka=0 romɨ pí=kɨ

and  shaman= tobacco eat=.
‘[It is said that] then the shaman chewed on the tobacco.’

A minimally contrastive V-C-Subj construction would be as in (37), where

the subject noun phrase is moved to a position after the clause-type/rank
morpheme and does not have ergative case.

(37) a. * hatsi kiá romɨ pí=kɨ yóbɨka

and  tobacco eat= shaman
Intended: ‘[It is said that] the shaman has chewed the tobacco.’
b. * hatsi kiá romɨ pí=ki yóbɨka
and  tobacco eat=. shaman
Intended: ‘[It is said that] the shaman is chewing tobacco.’

However, such sentences are ungrammatical unless the reportative also

moves to a position to the right of the clause-type morpheme, right before
the subject. This is illustrated in (38).

(38) a. hatsi romɨ pí=kɨ kiá yóbɨka

and tobacco eat=  shaman
‘[It is said that] the shaman has chewed the tobacco.’
b. hatsi romɨ pí=Ɂi kiá yóbɨka
and tobacco eat=  shaman
‘[It is said that] the shaman is chewing tobacco.’

One way of understanding the distributional restriction on reportatives

would be to postulate that the candidate low domain of the V-C-Subj con-
struction cannot be modified by the reportative. This would follow from
the analysis presented in (29), as it is common for embedded clauses to
384 ́     

have restrictions on modal or evidential modification. Reportatives are thus

associated to the main predicate across clause-types in Chácobo. In the V-C-
Subj constructions, in particular, the reportatives must occur in the candidate
high domain as markers in the nonverbal predicate construction.
A distributional fact about reportative markers in NVP constructions cor-
roborates this interpretation. In nonverbal predicate constructions, reportative
marker kiá is in contrastive distribution with other clause-type markers (see
Tallman, 2018: ch. 4, for details). This is illustrated in (39).

(39) a. kaɁɨ=ʂɨni tsi ʂo yobɨka

know= 5 . shaman
‘The shaman is/was wise.’
b. kaɁɨ=ʂɨni tsi kiá yobɨka
know= 5  shaman
‘It is said that the shaman is wise.’

Notice that the typical order of the reportative in relation to the subject is
identical to its order in the V-C-Subj construction. Understanding the V-C-
Subj construction as an NVP construction naturally accounts for this fact.
A fourth argument comes from the exponence of subject plurality.
Like in Mẽbêngôkre, even though the construction in question is biclausal,
it is nevertheless impossible for two coreferential subjects to occur in the low
domain and the high domain, as illustrated in (40).

(40) a. ha/yoʂa= 0 tʃaʂo tsaya=kɨ

3/woman= deer see=.
‘She / The woman saw the deer.’
b. * haii tʃaʂo tsaya=ki yoʂaii
3 deer see=. woman
‘The woman sees the deer.’

Though an overt subject cannot precede the verb in V-C-Subj constructions

in the candidate low domain, there are exponents of subject plural marking
which can occur in the candidate low domain that suggest that it is a separate
domain from the high domain with regard to number marking. First, note that
in Chácobo the third person plural subject pronoun ha . . . =kan displays
extended exponence. One part of the pronoun occurs prior to the verb and
the other part occurs between the verb root and the clause-type morpheme.
This is illustrated in (41). The first part of the pronoun is optional.
 ,  ,   385

(41) (ha) tsaɁo=ká(n)=kɨ

3 sit==.
‘They sat down.’

This cannot be regarded as number agreement because it is ungrammatical

for kán to appear when an overt NP is in place. This is illustrated in (42).

(42) a. * hóni=bo tsaɁo=ká(n)=kɨ

man= sit==.
‘The men sat down while going.’
b. honi=bo tsaɁo=kɨ
man= sit=.
‘They men sat down while going.’

While kán cannot occur with an overt preverbal subject in the verbal
predicate construction, the kán must occur when the subject is plural in the
V-C-Subj construction. Thus, the distribution of kán in the V-C-Subj con-
struction is the mirror image of its distribution in verbal predicate construc-
tions; compare (42) with (43):

(43) a. tsaɁo=ká(n)=kɨ hóni=bo

sit==. man=
‘The men sat down while going.’
b. * tsaɁo=kɨ hóni=bo
sit=. man=
Intended: ‘The men have sat down while going.’

It should also be noted that the distribution of kán in V-C-Subj construc-

tions is identical to its distribution in nonverbal predicate constructions, as
shown by the following examples:

(44) a. tɨtɨka=bo =ka(n) ʂo/kiá/ní honi=bo

long= = // man=
‘The men are long/tall. / It is said that the men are tall. / Are the men
tall?’
b. ʂobo katʃa rɨa-mɨ=tí=kan=kiá isko ʂokɨ=bo
house outside fill-=== gorse=
‘The gorses filled the outside of the house.’
386 ́     

Another exponent of plural marking is verb root suppletion. In Chácobo,

verb roots undergo suppletion when they have a plural pronominal subject.
The pattern they follow is identical to the one that describes the contribution
of kan described above. A full NP cannot co-occur with the plural allomorph
in a regular verbal clause, as may be seen in (45c). However, in the V-C-Subj
construction, the plural form of the verb has to appear, as (46) shows.

(45) a. honi=bo ka=kɨ

man= go./=.
‘The men went.’
b. ha bo=ka(n)=kɨ
3 go./==.
‘They went.’
c. * honi=bo bo(=kan)=kɨ
man= go./==.
‘The men went.’

(46) a. * ka=kɨ honi=bo

go=. man=
‘The men went.’
b. bo=kan=kɨ honi=bo
go==. man=
‘The men went.’

In conclusion, the exponents of plurality function as if there was no overt

NP subject in the candidate low domain of the V-C-Subj construction. If the
V-C-Subj construction was monoclausal we would not expect the marking of
plurality to operate differently than it does in the normal verbal predicate
construction. However, the distribution of the plural pronominal element in
the V-C-Subj construction is the same as in the NVP, which is predicted by the
analysis sketched in this section.

15.4 When is there reanalysis?

In both Mẽbêngôkre progressives and Chácobo V-C-Subj constructions, a

number of diagnostics support the idea that two distinct domains exist, with
effects on case assignment and in a number of other operations. We have
 ,  ,   387

postulated that in both cases what is responsible for the separation of the
clause into domains is a nominalizing element, whether overt (as in
Mẽbêngôkre and Basque) or covert (as in Chácobo). We propose the following
reference structure, uniting both the Chácobo V-C-Subj construction and the
Mẽbêngôkre progressive:

(47) VP

higher case domain Vʹ Subji

nP V
lower case domain
√P n

Subji √ʹ

Obj √

The primary elements represented in this structure are the following:

1. The ‘main verb’ is a categoriless root, with its arguments, one of which is
covert. The association of arguments to roots is a simplifying assump-
tion whose motivation we cannot discuss for reasons of space.
2. The root’s projection merges with a n category head (or possibly an
underspecified category head, standing for n and a).
3. The subject is actually subject of a higher predicate, represented as V. It
is coindexed with the covert subject of the ‘main verb’, but in neither
case discussed here does it form a movement chain with it.

The basic properties of the two constructions follow almost trivially from
this representation: The absence of dependent case is a consequence of the
separation of two domains by n, and the facts surrounding constituency and
the distribution of pronouns in Chácobo are unproblematically represented in
the structure.
The differences between the constructions are of course numerous, but they
do not affect the account:

1. In the Mẽbêngôkre progressive there is a thematic relation between

V and the higher subject, while in Chácobo V is simply a copula.
388 ́     

2. The morphology of n is overt in Mẽbêngôkre, but not in Chácobo.

3. While in Mẽbêngôkre the nP is licensed by means of a P, in Chácobo it is
directly licensed by the copula.
4. Specifiers are left-branching in Mẽbêngôkre, and thus subjects are initial.

A number of traits of the constructions that seemed to contradict the basic

analysis do not force us to change this account:

1. The fact that in Mẽbêngôkre the nP cannot be fronted like other objects
while retaining its meaning is a consequence of the obligatory control
construction in which it sits: The only way in which the covert subject
can be coindexed with the overt matrix clause subject is by being c-
commanded by it; this constraint also applies to other constructions that
involve coindexing between an overt matrix subject and a covert subject
of a nominalized clause.
2. That not all the clause-type markers are the same between the Chácobo
V-C-Subj construction and nonverbal predicates most likely has to do
with semantic rather than categorial selection between the clause-type
markers and the predicate: Though nominal, V-C-Subj constructions do
not always encode stative notions, like other nonverbal predicates. It is
to be expected that temporal and aspectual markers be sensitive to that
difference.

To conclude our exposition, we need to address three interrelated questions:

(i) what is the relationship between nominalization and case domains, (ii)
when can restructuring be said to have transformed a construction with two
domains into a simple clause, and (iii) can diagnostics establish this in a
manner that avoids circularity.
Despite the many parallels between n and v in Mẽbêngôkre, there are a
number of differences which we can generalize to both of the constructions
focalized in this chapter. In addition to not tolerating merging with tense, n
has the property of heading a phrase from which phrasal movement is
impossible: n is not only a phase, but a phase without an escape hatch. This
property of n is responsible for the fact that the two arguments in the
constructions that we have examined here are in distinct case domains: The
subject, associated with the higher domain, cannot form a chain with the
thematic position inside the root phrase.
We consider that restructuring consists precisely in changing this structure
in (47) into one that allows a single chain to be formed between the two
 ,  ,   389

domains, either by emptying out V to become an Aux (and simultaneously

shifting n into a nonfinite v), or by incorporating n into the selecting V (e.g., as
Hale & Keyser, 1993, propose for unergative verbs). Both of these processes
should have as an automatic consequence the fusing of the two domains into a
single one, and reverting to the alignment found in simple clauses. The trees in
(48) represent these two restructured constructions.

(48) VP VP

V’ Subj Vʹ Subj
nP ni+V
nP V→Aux
√P ti
... √P n → υnfin
...

The types of diagnostics that would serve as good tests for the occurence of
this reanalysis follow from these structures:

1. Morphological incorporation between the lower and higher predicate,

and functioning of these as a unit (e.g. in inversion).
2. Agreement across domains.
3. Loss of any thematic interaction between the higher predicate and the
subject.
4. Acquisition of verbal features by the lower predicate.

Interestingly, if diagnostics such as these are applied to Basque as described

by Hualde & Ortiz de Urbina (1987), they suggest that reanalysis has in fact
taken place in that case. It is not surprising that in many varieties of Basque the
progressive construction behaves as a single domain for case assignment. In
the presence of clear evidence for reanalysis, taking double-nominative align-
ment as the sole diagnostic for the partition of the clause into domains is
circular, however. The Mẽbêngôkre and Chácobo constructions examined
here, on the other hand, don’t pass any of the diagnostics for restructuring.
We conclude that there is no reason to conclude that restructuring has taken
place in either of them.
390 ́     

Acknowledgments

The authors would like to thank their Mẽbêngôkre and Chácobo consultants as well as the
editors of the volume, Andrey Nikulin, and an anonymous reviewer for extremely helpful
comments. Work on this chapter was supported by Insight Grant number 435-2018-1173
from the Social Sciences and Humanities Research Council of Canada (PI Andrés
Salanova).