1 Introduction
In clausal nominalization, the internal properties of the clause point to a verbal core (e.g., finiteness morphology), while its external properties are nominal (e.g., nominal distribution, case marking). A frequently discussed issue about this topic is how the clause can behave as a nominal externally. Currently, there is a general consensus that a nominal layer lies on top of the clausal content. However, there is less agreement regarding the specific structure of the nominal layer. Two main positions have emerged, as schematized in (1).
- (1)
- a.
- [dp d [… clausal structure …]] (The D-assumption of clausal nominalization)
- b.
- [nP n [… clausal structure … ]] (The n-assumption of clausal nominalization)
In particular, some argue that in clausal nominalization, a determiner head D merges directly with the clausal content, as posited in (1a); I refer to this as the D-assumption of clausal nominalization, or simply the D-assumption. This position is taken in several works (Davies & Dubinsky 1998; Borsley & Kornfilt 2000; Takahashi 2010; Miyagawa 2011; Pietraszko 2019; Alexiadou 2020; Iordăchioaia 2020; Hankamer & Mikkelsen 2021; Bochnak & Hanink 2022); cf. Grimshaw (2000) for a theoretical issue with this approach. In contrast, other scholars argue that the clausal content can be merged with n (Nasu et al. 2024; Lim 2025; 2026), as in (1b); throughout the paper, I call it the n-assumption of clausal nominalization, or just the n-assumption.
Studies that support the D-assumption in (1a) typically consider data indicating the presence of a determiner in clausal nominalizations, and rule out the n-assumption in (1b) based on expectations that are not met if an nP is assumed to exist within the same structure. On the other hand, proponents of the n-assumption argue that a determiner element is not possible, suggesting that no DP is projected in these structures, even though the clausal constituent is nominal in distribution and morphosyntax. This leads to the proposal that the nominal layer nP must be present, although its head is phonologically null.
Against this backdrop, I examine clausal nominalization – specifically, complement clauses – in the Tupí-Guaraní language Ka(g)wahíva from South America. Like several other languages from South America (Comrie & Estrada-Fernández 2012; Zariquiey et al. 2019), Ka(g)wahíva requires nominalization of dependent clauses. Furthermore, Ka(g)wahíva presents a convenient set of properties in which to examine the nominal structure of clausal nominalization and test them against the predictions of the D-assumption and n-assumption. Notably, it has overt realization of the nP head through an enclitic, =a; overt exponence of elements within the determiner category; and it permits bare nPs (or small nominals, to borrow this terminology from Pereltsvaig (2006)).
Therefore, if Ka(g)wahíva clausal nominalizations follow the D-assumption structure, then we expect these structures to exhibit no enclitic =a, and exhibit the Ka(g)wahíva determiners. In contrast, if the language follows the n-assumption structure in clausal nominalizations, then these structures should exhibit =a, but not a determiner (overt or covert). As we will see, concerning the nominal structure within nominalized complement clauses, I argue that these structures are bare nPs and do not project a DP (or any higher nominal projection, such as PossP). As a result, Ka(g)wahíva clausal nominalizations support the n-assumption structure in clausal nominalizations, as in (1b).
This paper also addresses another important issue regarding clausal nominalization: the relationship between the clause size and the type of nominalizer used. Recent proposals indicate that the type of nominalizer determines how much structure it can embed (Alexiadou 2020; Iordăchioaia 2020; Lim 2025; 2026). This paper considers the claim that TP nominalizations are possible when the head D is involved, but not when using n (at least not when a DP layer is missing). I argue that this position is too restrictive for Ka(g)wahíva. I demonstrate that complement clauses are truncated structures that project up to TP. Thus, combined with the finding that the nominal structure in clausal nominalization only consists of nP, it leads to the conclusion that Ka(g)wahíva complement clauses are TP-embedding nPs.
If correct, these results add to recent proposals that nominalizations do not always require a DP layer, as also found in Japanese and the Mongolic language Khalkha (Nasu et al. 2024; Lim 2025), besides contributing to the only existing proposal that nPs may embed TPs without further nominal projections, as in the Bantu language Oshiwambo (Lee & Ndapo 2025).
This paper is organized as follows. After presenting more information about Ka(g)wahíva ecology and data methodology in §2 and briefly introducing the complement clause in §3, I present the properties of Ka(g)wahíva clause that will serve as a background to assess the clause-like nature of nominalized clausal complements, including clause structure 4.1 and verbal morphology 4.3. After that, I show that most, though not all, clausal properties are present in the complement clause in §5. This result will allow me to posit that these structures are truncated TP embeddings. Following that, I provide background on the properties of Ka(g)wahíva nominals. This section will be a more extensive discussion, since it will also motivate key claims about the structure of nominals in the language, chief among them the claim that there is no covert DP in §6. The next section discusses the nominal properties of complement clauses §7 to support the claim that nominalization in complement clauses only involves an nP projection, and no DP. Finally, in Section 8, I explore the implications of this work and conclude in §9.
Before proceeding to the next section, I would like to emphasize that the amount of data and detail presented here is warranted. Ka(g)wahíva (and Tupían languages, more generally) has rarely been the subject of systematic studies on the nominalization of embedded clauses within a formal framework, despite the pervasiveness of nominalization in these languages. The notable exceptions include previous research on Ka(g)wahíva itself (dos Santos 2024b: in particular, Chapter 5 on nominalized relative clauses) and its sister language Tenetehára (Camargos & Castro 2013). More generally, some languages from distantly related branches within the broader Tupían family have been examined to explore the finiteness properties in dependent clauses as compared to root clauses, both in the Tuparí branch (Singerman 2021; Nogueira 2019: for Tuparí and Wayoro), and in the Arikém branch (Rocha da Silva 2016; Nogueira 2019; Vivanco 2021: for the single family member Karitiana).
2 Language ecology and data
Kagwahíva, or Kawahíva (in lects lacking the phoneme /gw/), is a critically endangered Amazonian language belonging to the Tupí-Guaraní language family. It is spoken by approximately 560 individuals out of a population of some 1060 individuals in total (dos Santos 2024a).
There are eight ethnic Ka(g)wahíva lects, often referred to by the names of the communities where they are spoken. In the state of Amazonas, the languages include Jiahui [iso: pah], Júma [iso: jua], Parintintin [iso: pah], and Tenharin [iso: pah]. In the state of Mato Grosso, there is Piripkura, and in the state of Rondônia, the languages spoken are Amondawa [iso: adw], Karipuna [iso: kuq], and Jupaú [iso: urz] (also known by the exonym Uru Eu Wau Wau).
Unless otherwise indicated, the core set of examples provided in this paper is based on the author’s fieldwork with native speakers of the Júma and Jupaú lects since 2019. These examples draw on annotated texts from spontaneous speech, primarily personal stories and traditional narratives. Additionally, targeted elicitation with speakers was conducted to confirm preliminary assumptions based on positive evidence from the texts.1
This data is part of the online Ka(g)wahíva Language Documentation Archive, hosted at the California Language Archive (CLA); this is an accessible collection of audiovisual recordings of several Ka(g)wahíva lects.2
Work on Ka(g)wahíva grammar is still limited, but there have been recent advances on Ka(g)wahíva clause structure and verbal agreement (dos Santos 2023; 2025a; b). A recent grammar sketch is also available (dos Santos 2024b). The current paper draws on the current understanding of these two properties of Ka(g)wahíva.
In the following section, I briefly introduce the complement clause in Ka(g)wahíva.
3 The structure of interest
In almost all Tupí-Guaraní languages, including Ka(g)wahíva, nominalization is the strategy for embedding. Consider the notional complement clause within brackets in (2) below, from Ka(g)wahíva.
- (2)
- Okwaham
- o-kwaham
- 3.a-know
- ki
- ki
- pst
- Boreá-hẽa
- Boreá=hẽa
- B.=3sg.f
- amãna
- [amãna
- rain
- ikyre’yma
- i-kyr-e’ym=a
- obj-rain-nom.neg=nmlz
- ko’emamẽ]
- ko’emamẽ]
- tomorrow
- ko.
- ko
- realis
- ‘Boreá thinks that it will not rain tomorrow.’ (Júma: Elicit)
We will see that complement clauses like (2) exhibit the profile of nominals while also having a clause-like internal structure. For instance, they exhibit nominal negation (via -e’ym) and nominal tense (not shown in this example), in addition to appearing in typical nominal positions, like verbal objects; See Singerman (2018: on negation as an exclusively nominal category in the distantly related language Tuparí) and Tonhauser (2006; 2007) and Thomas (2014: on nominal temporal morphology in the sister languages Paraguayan Guaraní and Mbyá). Furthermore, these structures possess all the properties that matrix clauses exhibit up to TP. For instance, the clause-like syntax is evinced via adverbial modification in (2).
All in all, Ka(g)wahíva complement clauses can be best described as truncated TPs embedded under a nominal projection.
A note is in order regarding the English translations of Ka(g)wahíva complement clauses. Given the claim in section 5.1.4 that these structures are TP-embedded nominalizations in Ka(g)wahíva, I have tried to provide gerund translations for these examples whenever that was possible (e.g., ‘You saw the jaguar killing a tapir’, from (20) below), on a par with the idea that English gerunds are TPs. When this choice was not available, I gave translations with a finite CP, as in the that-clause in (2). These choices also reflect speakers’ translations of Ka(g)wahíva into the target language, Portuguese, during elicitation tasks.
4 Clausal-like properties in Ka(g)wahíva
4.1 Clause structure
The clause-initial position in Ka(g)wahíva hosts a single element at a time. This initial element can be a verb (3a), an interrogative phrase (3b), a focused phrase (3c), or a topic phrase (3d).3
- (3)
- a.
- Vepiang
- v-epiang
- 3.a-see
- po
- po
- irr
- txi
- txi
- fut
- jirembiriko-ga
- ji=rembiriko=ga
- 1sg=partner=3sg.m
- Kanindé-gã
- Kanindé=gã
- Kanindé=3.pl
- nãhẽ.
- nãhẽ
- prosp
- ‘My husband might see the Kanindé NGO group.’ (Júma: Elicit)
- b.
- Mãngã
- mãngã
- who
- po
- po
- irr
- jiruvyra
- ji=ruvyra
- 1sg.poss=uncle
- vepiang
- v-epiang
- 3.a-see
- rimba’e?
- rimba’e
- distant.pst
- ‘Who did my uncle see back in the olden times?’ (Júma: Text)
- c.
- Pirapetxinguhua
- pirapetxinguhua
- pirapetxinguhua.fish
- te’i
- te’i
- only
- ki
- ki
- pst
- hẽa
- hẽa
- 3sg.f
- opyhy-pyhy
- o-pyhy∼pyhy
- 3.a-catch∼red
- ko.
- ko
- realis
- ‘It was only pirapetxinguhua fish that she caught.’ (Júma: Text)
- d.
- Takãi Kajuvi
- Takãi.Kajuvi
- Takãi.Kajuvi
- ki,
- ki,
- pst
- tapy’ynha
- tapy’ynha
- non-indigenous
- opurun
- o-purun
- 3.a-hit
- garehe.
- ga=rehe
- 3sg.m=at
- ‘(As for) Takãi Kajuvi, the non-Indigenous person ran over him.’ (Júma: Text)
Moreover, the first initial constituent always precedes the mood and tense particles; when the two are present, the particles follow the strict order Mood-Tense.4 In fact, no more than one initial element is allowed before the particles, no matter the linear order among the initial constituents, as in (4a).
- (4)
- a.
- *Vepiang
- v-epiang
- 3.a-see
- Kanindé-gã
- Kanindé=gã
- Kanindé=pl
- po
- po
- irr
- txi
- txi
- fut
- jirembiriko-ga.
- ji=rembiriko=ga
- 1sg.poss=partner=m
- Intended: ‘It is Kanindé that may see my husband.’ (Júma: Elicit)
- b.
- *Kanindé-gã
- Kanindé=gã
- Kanindé=pl
- vepiang
- v-epiang
- 3.a-see
- po
- po
- irr
- txi
- txi
- fut
- jirembiriko-ga.
- ji=rembiriko=ga
- 1sg.poss=partner=m
- Intended: ‘It is Kanindé that may see my husband.’ (Júma: Elicit)
Assuming a layered approach to clause structure, the representation of the matrix clause structure preliminarily suggests itself as in (5). The CP, MoodP, and TP projections are supported by the properties of matrix clauses mentioned above, namely a single slot generally occupied by information-structurally prominent phrases in clause-initial position, as well as the mood and tense-related particles rigidly ordered after that initial position. The other projections, VoiceP, vP, and VP, are accounted for under a layered approach to clausal structure, as developed in various schools of syntactic analysis (Kratzer 1996; Pylkkänen 2002; Harley 2013).
- (5)
- CP– MoodP – TP – VoiceP – vP – VP
Excepting for the verb-initial orders, I assume that phrase-initial elements are derived via [Ā] movement. As for the verb-initial order VSO of pragmatically neutral contexts (see example (3a), I follow the account in dos Santos (2025a), in which the Ka(g)wahíva VSO results from long head movement to Spec,CP triggered by a categorial feature [V] with the EPP property on C. Diagram (6) represents this idea.
- (6)
It is not crucial for the main claim of this paper, however, that information-structurally prominent phrases surface in the initial position by means of [Ā] movement, and verbs do so via long head movement. What matters for our purposes is that matrix clauses exhibit a dedicated position before the mood and tense particles (or TP and MoodP in clausal-structural terms), which hosts pragmatically prominent phrases or the verb, identified as the CP.
4.2 Tense and negation
In Ka(g)wahíva, the expression of tense and negation notoriously differs across clause types (matrix clause versus dependent clauses). I first present the matrix tense particles, since, to my knowledge, no cognate particles with this semantics are found in sister languages.
The matrix tense particles that will matter for the discussion are the past tense particle ki and the future tense particle txi. The general past tense particle ki indicates that the action occurs anytime before the utterance time. In that way, it is compatible with past temporal adverbials, such as oi’i (∼ oji’i) ‘some time ago’.
- (7)
- a.
- Gara
- gara
- what
- ki
- ki
- pst
- nde
- nde
- 2sg
- ereapo
- ere-apo
- 2sg.a-do
- oi’i?
- oi’i
- some.time.ago
- ‘What did you do yesterday?’ (Júma: Elicit)
- b.
- Humaitá
- Humaitá
- Humaitá
- uvi
- uvi
- from
- ore
- ore
- 1.excl
- juarinẽ,
- ju=a=rinẽ,
- come=nmlz=after
- ki
- ki
- pst
- ore
- ore
- 1.excl
- hoi
- ho-i
- go-ais
- kavyripe
- kavyr=ipe
- jungle=in
- ko.
- ko.
- realis
- ‘After we returned from Humaitá city, we went into the jungle.’ (Júma: Elicit)
The future tense particle txi indicates that the action will occur sometime after the utterance time. It is compatible with future temporal adverbials, including koi’iramẽ ‘later’, and adverbial clauses with a future reference time.
- (8)
- a.
- Gara
- gara
- what
- po
- po
- irr
- txi
- txi
- fut
- nde
- nde
- 2sg
- ereapo
- ere-apo
- 2sg.a-do
- koi’iramẽ?
- koi’iramẽ
- later
- ‘What will you do later?’ (Júma: Elicit)
- b.
- Wesley-ga
- Wesley=ga
- W.=3sg.m
- horamẽ
- ho=ramẽ
- go=when
- ko’emamẽ,
- ko’emamẽ,
- tomorrow
- txi
- txi
- fut
- jie
- jie
- 1sg
- mitakwanha
- mitakwanha
- bread
- poi
- po-i
- make-ais
- nãhẽ.
- nãhẽ
- pot
- ‘When Wesley leaves tomorrow, I will make bread.’ (Júma: Elicit)
As for matrix clause negation, pragmatically neutral negation is done through a circumfix, n(a)-…-i (or the phonologically conditioned allomorph, nd(a)-…-i).
- (9)
- Ndajirepiagi
- nd(a)=ji=repiag-i
- neg-1sg.b=see-neg
- gãmõ.
- gã=mõ
- 3.pl=indf.det
- ‘Nobody saw me.’ (Júma: Elicit)
In contrast, in dependent clauses, the exponents of tense and negation are bound morphemes that are also shared with simple nominals. Starting with tense, dependent clause tense surfaces through the suffixes -ver ‘past tense’ and -ram in Júma (and -ham in Jupaú) ‘future tense’. Focusing on the past tense, the example in (10) illustrates this suffix within the complement clause. This morphology also occurs in simple nominals (11)–(12), where it exhibits lexically conditioned allomorphy (-ver ∼-kwer).
- (10)
- Kwarupi
- kwarupi
- today
- emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- ga
- [ga
- 3sg.m
- vuva-ga
- v-uva=ga
- 3sg.cor=father=3sg.m
- repiagavera
- repiag-(a)ver=a
- see-nom.pst=nmlz
- oi’i!
- oi’i]
- some.time.ago
- ‘Tell (them) today that he saw his father yesterday!’ (Júma: Elicit)
- (11)
- Ayvera
- ay-ver=a
- wood-nom.pst=nmlz
- ‘Stick’ (Also applicable to any piece of wood that is no longer part of a tree) (Júma: Text and Elicit)
- (12)
- Jirembirekokwera-hẽa
- ji=rembireko-kwer=a=hẽa
- 1sg.poss=partner-nom.pst=nmlz=3sg.f
- ‘My ex-wife.’ (Common in speech)
The negation exponent in dependent clauses is the suffix -e’ym (in Jupaú, -i’im). The same morpheme appears in simple nominals, where it realizes the idiosyncratic negative meaning of privative.
- (13)
- a.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- gã
- [gã
- 3.pl
- mãjioka
- mãjioka
- manioc
- rerure’ymavera!
- rerur-e’ym-(a)ver=a]
- bring-nom.neg-nom.pst-nmlz
- ‘Tell (them) that they did not bring manioc!’ (Júma: Elicit)
- b.
- U’ie’yma
- u’i-e’ym=a
- flour-nom.neg=nmlz
- ‘(I ate) flourless (rice).’ (Júma: Text)
The distribution of nominal tense and negation morphology across simple nominals and dependent clauses is shared across the Tupí-Guaraní family (Seki 2000; Magalhães 2007; Praça 2007; Thomas 2014; Estigarribia 2020: to cite the best-known languages in this regard, Kamaiurá, Guajá, Tapirapé, Mbyá, and Paraguayan Guaraní, respectively). In fact, in the distantly related Tuparí language, the default marking of negation is nominal (Singerman 2018).
Next, I expand this background section on clausal properties by presenting those associated with verbs in matrix clauses.
4.3 Verbal properties
In this section, we will see that the typical verb in Ka(g)wahíva is subjected to reduplication, can be modified by adverbial PPs and suffixes, and exhibits valency morphology. An additional piece of morphosyntax discussed is verbal agreement.
4.3.1 Reduplication and adverbial modification
Reduplication is considered a morphological process restricted to verbs across the family (Rose 2023). In Ka(g)wahíva, reduplication encodes pluractionality. We can see verbal reduplication in (14a), which indicates multiple participants in an action, and in (14b), which indicates an iterated action.
- (14)
- a.
- Arerety-areretyk
- are=rety∼areretyk
- 1.excl.b=pull.down∼red
- ki
- ki
- pst
- gã
- gã
- 3.pl
- ko.
- ko
- realis
- ‘They pulled us all down.’
- b.
- Anhãnga
- anhãnga
- anhãnga.spirit
- okutu-kutupam
- o-kutu∼kutu-pam
- 3.a-pierce∼red-completely
- jurua.
- j-urua
- obj-mouth
- ‘It was the anhãnga spirit that pierced (his) mouth (many times).’ (Jupaú: Text)
Adverbial modification can be indicated by affixation and postpositional phrases. Among the former are: -pav ‘completely, -jipe ‘already’, -jav ‘again’, -tehe ‘unpurposefully’, and -ahy ‘forcefully’, and in the form of temporal postpositional phrases, common phrases are ko’emamẽ ‘tomorrow’ and ypyjivete ‘early morning’.5
4.3.2 Valency morphology: causatives, applicative, reflexive, and reciprocal
The morphological expression of valency-augmenting and valency-decreasing semantics will also be considered, as it is traditionally treated as verbal within the family. I start with valency-increasing morphemes.
Ka(g)wahíva has the causative prefix mõ- (∼ mbo- in oral roots). It forms transitive verbs from intransitive verbs by introducing an agent/causer subject; compare (15a) to (15b).
- (15)
- a.
- Ohin
- o-hin
- 3.a-fall
- jimẽna-ga.
- ji=mẽn=a=ga.
- 1.sg=partner=nmlz=3.sg.m
- ‘My husband fell.’ (Júma: Elicit)
- b.
- Ombohin
- o-mbo-hin
- 3.a-caus-fall
- kurumĩ-gã
- kurumĩ=gã
- kid=3.pl
- nhãpepoa.
- nh-ãpepo=a.
- obj-pan=nmlz
- ‘The kids dropped someone’s pan.’ (Júma: Elicit)
Ka(g)wahíva also has the applicative morpheme -okan (∼-ukan). The suffix -okan adds an argument with a goal or benefactive role to a transitive verb, lexical or derived. This can be seen by contrasting the examples in (16a) and (16b), the latter of which has the applicative suffix.
- (16)
- a.
- Ombohin
- o-mbo-hin
- 3.a-caus-fall
- kurumĩ-gã
- kurumĩ=gã
- kid=pl
- nhãpepoa.
- nh-ãpepoa.
- obj-pan
- ‘The kids dropped (someone’s) pan.’ (Júma: Elicit)
- b.
- Ombohirukan
- o-mbo-hir-ukan
- 3.a-caus-fall-appl
- hẽa
- hẽa
- 3.sg.f
- mbiara
- mbiara
- meat
- javatxinga
- javatxinga
- dog
- pe.
- pe.
- for
- ‘She dropped the meat for the dog.’ (Júma: Elicit)
Reflexivity and reciprocality are also expressed morphologically on the verb. Reflexivity is indicated by the prefix ji-, as in (17a), to mark the identity between subject and object. Reciprocality is expressed by jo-, which marks mutual action between the subject individuals, as in (17b).
- (17)
- a.
- Ojipe’a
- o-ji-pe’a
- 3.a-refl-distance
- javytxinga
- javytxinga
- dog
- jawara
- jawara
- jaguar
- uvi.
- uvi.
- from
- ‘The dog distanced itself from the jaguar.‘ (Jupaú: Elicit)
- b.
- Ojovepiang
- o-jo-vepiang
- 3.a-recp-see
- ki
- ki
- pst
- gã
- gã
- 3.pl
- ko.
- ko.
- realis
- ‘They met.’ (Lit.: They saw each other) (Júma: Elicit)
In sum, the typical morphosyntactic properties of Ka(g)wahíva verbs are reduplication, adverbial modification, and valency-changing morphology.
Finally, I introduce verbal agreement.
4.3.3 Verbal agreement
Ka(g)wahíva distinguishes grammatical relations primarily through verbal agreement with core arguments. In matrix clauses, verbs agree with either the subject or object; the choice is based on the highest-ranking person feature in the Person Hierarchy 1>2>3 (dos Santos 2023). The examples in (18a)–(18b) illustrate that in sentences involving 1st and 2nd person, it is 1st person arguments that are privileged for verbal agreement over 2nd person arguments, regardless of their grammatical roles.
- (18)
- a.
- Ahepia
- a-hepia
- 1sg.a-see
- ki
- ki
- pst
- ji
- ji
- 1sg
- nde
- nde
- 2sg
- ko.
- ko
- realis
- ‘I saw you.’ (Júma: Elicit)
- b.
- Jirepia
- ji=repia
- 1sg.b=see
- ki
- ki
- pst
- nde
- nde
- 2sg
- ko.
- ko
- realis
- ‘You saw me.’ (Júma: Elicit)
Table 1 shows the full set of agreement markers. As in several other sister languages, subject agreement markers are truly inflectional, whereas object agreement markers are pronominal clitics (Jensen 1990; 1998; Meira & Drude 2013; Rose 2015; dos Santos 2023). Note that there is no agreement marker for 3rd person objects.
Table 1: Ka(g)wahíva agreement markers.
| set a ‘subject agreement’ | set b ‘object agreement’ | |
| 1st person singular | a- | ji= |
| 2nd person singular | ere- | nde= |
| 3rd person | v- ∼ o- | – |
| 1st person inclusive | txi- | nhãnde= |
| 1st person exclusive | oro- | ore= |
| 2nd person plural | pe- | pe= |
In contrast, in dependent clauses, the lexical verb may only agree with an object, using one of the object markers. In local scenarios of 1→2, as in (19a), agreement yields to 2nd person; agreement with the subject is not allowed, as (19b) shows.
- (19)
- a.
- Okwaha
- o-kwaha
- 3.a-know
- ki
- ki
- pst
- ga
- ga
- 3sg.m
- ji
- [ji
- 1sg
- nderepiaka.
- nde=repiak=a]
- 2sg.b=see=nmlz
- ‘He knew I saw you.’ (Júma: Elicit)
- b.
- *Okwaha
- o-kwaha
- 3.a-know
- ki
- ki
- pst
- ga
- ga
- 3sg.m
- ji
- [ji
- 1sg
- nde
- nde
- 2sg
- ahepiaka.
- a-hepiak=a]
- 1sg.a-see=nmlz
- Intended: ‘He knew I saw you.’ (Júma: Elicit)
This pattern of object agreement in dependent clauses is further evidenced in non-local scenarios (i.e., 3>3). Recall from Table 1 that Ka(g)wahíva does not exhibit agreement with 3rd person objects, while agreement with 3rd person subjects renders the prefix o- on the lexical verb. In dependent clauses with third person (3→3), the presence of the subject marker prefix o- creates an ungrammatical structure, as shown by (20).
- (20)
- a.
- *Erepia
- ere-pia
- 2sg.a-see
- ki
- ki
- pst
- nde
- nde
- 2sg
- jawara
- [jawara
- jaguar
- tapi’ira
- tapi’ira
- tapir
- ojuka
- o-juka=a]
- 3.a-kill=nmlz
- ko.
- ko
- realis
- Intended: ‘You saw the jaguar killing a tapir.’ (Júma: Elicit)
- b.
- Erepia
- ere-pia
- 2sg.a-see
- ki
- ki
- pst
- nde
- nde
- 2sg
- jawara
- [jawara
- jaguar
- tapi’ira
- tapi’ira
- tapir
- juka
- juka=a]
- kill=nmlz
- ko.
- ko
- realis
- ‘You saw the jaguar killing a tapir.’ (Júma: Elicit)
The Ka(g)wahíva agreement system is analyzed in dos Santos (2025a; b). Assuming the standard view that agreement is the result of a probe-goal relationship mediated through Agree Chomsky (2001), dos Santos argues that these differences in agreement require the view where subject and object agreement markers reside in different loci, and each probe has a different featural specification in Ka(g)wahíva. In particular, subject markers are due to a probe relativized to [spkr-part-Φ] on C, while object markers result from a flat [Φ]-probe on v. The combination of this idea with the clause structure proposed in the previous sections yields the representation of matrix clauses with two probes in (21).
- (21)
- [CP C [MoodP [TP [VoiceP EA [Voice [vP [v[Φ] [VP IA V ] ] ] ] ] ] ] ]
In this account, because only the C head bears a probe that c-commands (and, therefore, may copy features from) both core arguments, person hierarchy effects/subject agreement may only arise in the combination of two processes: the presence of the C head and the movement of the verb to CP to serve as a host for this morphology. Let us call these the Ka(g)wahíva Syntactic Conditions on Agreement:
- (22)
- Ka(g)wahíva Syntactic Conditions on Person Hierarchy Effects/Subject Agreement:
- a) Merging of a C-head with the relativized probe [spkr-part-Φ]
- b) Movement of the verb to CP
These two conditions straightforwardly capture the agreement facts in VSO clauses. In VSO clauses, as in example (23), the verb appears in clause-initial position – and by hypothesis undergoes movement to the CP level – and therefore bears agreement with the highest-ranking person feature in the Person Hierarchy 1>2>3; if the subject is the 1st person, then subject agreement yields:
- (23)
- Ahepia
- a-hepia
- 1sg.a–see
- ki
- ki
- pst
- ji
- ji
- 1sg
- nde
- nde
- 2sg
- pevo
- pevo
- there
- ko.
- ko
- realis
- ‘I saw you there.’ (Júma: Elicit)
In contrast, in dependent clauses, as in (19a)–(20b), verbs are strictly in final position relative to core arguments, which is taken to indicate the absence of movement to CP. We also see that agreement is restricted to objects in these instances. Under the idea that features from the external argument are only available on C, verb movement to CP in VSO clauses, but not in SOV clauses, correctly predicts that VSO clauses may yield agreement to the external argument, but SOV clauses should not.6
In sum, Ka(g)wahíva subject agreement results from a probe that sits high in the clause structure, the CP level; a similar agreement system in the sister language Mbyá-Guaraní has recently received an analogous analysis (Thomas et al. 2025).
4.4 Interim summary
The characterization of Ka(g)wahíva matrix clauses and the verbal morphosyntax provides several properties for assessing how clause-like other structures are, such as complement clauses. That is, if we find all or only a subset of these properties, this strongly suggests that the other structures under examination are either structurally akin to matrix clauses or truncated at a certain height.
For Ka(g)wahíva, we will see that complement clauses exhibit most of the properties reviewed here for matrix clauses, but lack some: thus, they have the morphosyntax that matrix clauses also show up to the TP, but entirely lack those associated with the CP, such as reality mood and subject agreement.
In turn, determining the size of complement clauses will later serve to investigate the different expectations that approaches to clausal nominalization, the D-assumption and n-assumption, make for complement clauses if they are TP-embeddings. In particular, under the D-assumption structure, TP is expected. In contrast, if these nominalizations involve the n-assumption, a TP structure is not expected.
5 Clause-like properties of complement clauses
In this section, I show that complement clauses may exhibit clausal material up to TP, but no further. This will lead me to posit that these clauses are truncated, projecting up to TP.
5.1 Amount of clause structure
5.1.1 Presence of TP in complement clauses
Complement clauses involve at least a TP. That tense is projected in embedded clauses is shown by the fact that they allow temporal postpositional phrases like ko’emamẽ in (24), which unambiguously modify the embedded verb.
- (24)
- Okwaham
- o-kwaham
- 3.a-know
- ki
- ki
- pst
- Boreá-hẽa
- Boreá=hẽa
- B.=3sg.f
- amãna
- [amãna
- rain
- ikyre’yma
- i-kyr-e’ym=a
- obj-rain-nom.neg=nmlz
- ko’emamẽ
- ko’emamẽ]
- tomorrow
- ko.
- ko
- realis
- ‘Boreá thinks that it will not rain tomorrow.’ (Júma: Elicit)
In fact, each clause may appear with a temporal adjunct that mismatches in tense interpretation, as in (25).
- (25)
- Kwarupi
- kwarupi
- today
- emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- ga
- [ga
- 3sg.m
- vuva-ga
- v-uva=ga
- 3sg.cor=father=3sg.m
- repiagavera
- repiag-(a)ver=a
- see-nom.pst=nmlz
- oi’i!
- oi’i]!
- some.time.ago
- ‘Tell (them) today that he saw his father yesterday!’ (Júma: Elicit)
A further piece of evidence that a TP is projected in complement clauses is evinced by the presence of the nominal past tense marker -ver in examples like (26), showcasing long-distance extraction of the subject of the most embedded complement clause. In both dependent clauses, the most embedded verb and the intermediate verb appear with the past tense marker -ver.7
- (26)
- Mãngã
- mãngãi
- who
- ga
- ga
- 3sg.m
- omõmbe’u
- o-mõmbe’u
- 3.a-tell
- nde
- [nde
- 2sg
- epiakavera
- ipira
- epiak-(a)ver=a
- see-nom.pst=nmlz
- imĩmavera?
- [_i
- i-pira
- obj-clothes
- i-mĩm-(a)ver=a]]?
- obj-hide-nom.pst=nmlz
- ‘Who did he tell that you saw stealing clothes?’ (Júma: Elicit)
Taken together, these facts suggest that Ka(g)wahíva complement clauses project TPs.8 Next, I present evidence that projections below TP in matrix clauses are also present in complement clauses.
5.1.2 Finiteness properties of complement clauses
Complement clauses exhibit several verbal properties (akin to English -ing gerunds) (Alexiadou 2020; Iordăchioaia 2020: among others). Beginning with their verbal properties, these nominalizations show adverbial modification in the form of verbal suffixes (27a)–(27b) and postpositional phrases (27c) (see also (24)–(25) from above).
- (27)
- a.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- tapy’ynha-gã
- [tapy’ynha=gã
- non-indigenous=3.pl
- rurahyavera!
- rur-ahy-(a)ver=a]
- come-forcefully-nom.pst=nmlz
- ‘Tell (them) that the non-indigenous people came (on a motorcycle) crazily!’ (Júma: Elicit)
- b.
- Emõmbe’u
- e-mõmbe’u
- 2sg.a-tell
- gã
- [gã
- 3.pl
- mãjioka
- mãjioka
- manioc
- reruratehevera!
- rerur-(a)tehe-ver=a]
- bring-in.vain-nom.pst=nmlz
- ‘Tell (them) they brought manioc for no reason!’ (Júma: Elicit)
- c.
- Emõmbe’u
- e-mõmbe’u
- 1sg.a-tell
- tapy’ynha-gã
- [tapy’ynha=gã
- non-ind.=3.pl
- pira
- pira
- fish
- ipyhykavera
- i-pyhyk-(a)ver=a
- obj-catch-nom.pst=nmlz
- orekavyra
- ore=kavyra
- 1.excl=land
- pyri!
- pyri]
- near
- ‘Tell (them) that the non-indigenous people caught fish in our land!’ (Júma: Elicit)
Other projections include voice morphology in the form of causative (28a), reflexive (28b), and reciprocal (28c), pluractional aspect in the form of reduplication (28c), applicative morphology (28d), negation (28e), and tense morphology in the form of the past suffix -(a)ver (all examples).
- (28)
- a.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- ga
- [ga
- 3sg.m
- mbahira
- mbahira
- light
- imbovevavera!
- i-mbo-vev-(a)ver=a]
- obj-caus-turn.off-nom.pst=nmlz
- ‘Tell (them) that he turned off the light!’ (Júma: Elicit)
- b.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- gã
- [gã
- 3.pl
- japoavera!
- j-apo-(a)ver=a]
- refl-do-nom.pst=nmlz
- ‘Tell (them) that they hurt themselves!’ (Júma: Elicit)
- c.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- gã
- [gã
- 3.pl
- jomõngita-mõngitavera!
- jo-mõngita-mõngita-ver=a]
- recp-narrate.story-red-nom.pst=nmlz
- ‘Tell (them) that they repeatedly narrated stories to each other!’ (Júma: Elicit)
- d.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- Agui-ga
- [Agui=ga
- A.=3sg.m
- mbiara
- mbiara
- meat
- mbohirukaravera
- mbohir-ukar-(a)ver=a
- drop-appl-nom.pst=nmlz
- javatxinga
- javatxinga
- dog
- pe!
- pe]
- to
- ‘Tell (them) that Agui dropped meat for the dog!’ (Júma: Elicit)
- e.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- gã
- [gã
- 3.pl
- mãjioka
- mãjioka
- manioc
- rerure’ymavera!
- rerur-e’ym-(a)ver=a]
- bring-nom.neg-nom.pst-nmlz
- ‘Tell (them) that they did not bring manioc!’ (Júma: Elicit)
All these examples also illustrate that clausal complements have event participants (subjects and objects) and event readings.
In the following section, I show that clausal complements still lack several other properties of matrix clauses.
5.1.3 Absent finiteness properties and main clause phenomena in complement clauses
Complement clauses are structurally reduced from full clauses – they do not allow mood marking, evidentiality, some types of aspect, subject agreement, and clause-internal fronting, all associated with projections higher than TP. The initial piece of evidence for the reduced size of embedded clauses is their lack of proposition-marking elements, such as reality mood (29a) and propositional morphology, such as -aha ‘unfortunately’ (29b).
- (29)
- No mood marking in clausal nominalizations
- a.
- *Arovia
- a-rovia
- 1sg.a-believe
- jie
- jie
- 1sg
- Puré-ga
- [Puré=ga
- Puré=3sg.m
- po
- po
- irr
- tapi’ira
- tapi’ira
- tapir
- juka.
- juka=a]
- kill=nmlz
- Intended: ‘I believe that Puré may have killed a tapir.’ (Júma: Elicit)
- b.
- *Arovia
- a-rovia
- 1sg.a-believe
- jie
- jie
- 1sg
- po
- [po
- irr
- Puré-ga
- Puré=ga
- Puré=3sg.m
- tapi’ira
- tapi’ira
- tapir
- juka.
- juka=a]
- kill=nmlz
- Intended: ‘I believe that Puré may have killed a tapir.’ (Júma: Elicit)
- (30)
- *Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- gã
- [gã
- 3pl
- mãjioka
- mãjioka
- manioc
- eruraha!
- erur-aha]
- bring-unfortunately
- Intended: ‘Tell (them) they unfortunately brought manioc!’ (Júma: Elicit)
Subject agreement, too, is ruled out from dependent clauses, as shown by the ungrammaticality of 3rd person subject agreement in the bracketed complement clause in (31).
- (31)
- *Erepia
- ere-pia
- 2sg.a-see
- ki
- ki
- pst
- nde
- nde
- 2sg
- jawara
- [jawara
- jaguar
- tapi’ira
- tapi’ira
- tapir
- ojuka
- o-juka=a]
- 3.a-kill=nmlz
- ko.
- ko
- realis
- Intended: ‘You saw the jaguar killing a tapir.’ (Júma: Elicit)
Additionally, these structures do not allow two other properties of matrix clauses: evidentiality and habitual and progressive aspect. The former category, in the form of suffixes like auhu ‘seemingly’ and particles, creates ill-formed structures in complement clauses (32).
- (32)
- a.
- *Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- gã
- [gã
- 3pl
- auhu
- auhu
- seemingly
- mãjioka
- mãjioka
- manioc
- reruravera!
- rerur-(a)ver=a]
- bring-nom.pst=nmlz
- Intended: ‘Tell that they seemingly brought manioc!’ (Júma: Elicit)
- b.
- *Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- avoipava
- [avoipava
- ancestor
- ka’ia
- ka’ia
- monkey
- i’ue’yma
- i-’u-’ym=a
- obj-eat-nom.neg=nmlz
- mãipo.
- mãipo]
- n.wit.pst
- Intended: ‘Tell that the ancestors did not eat monkey in the olden times!’ (Júma: Elicit)
A last piece of finiteness property from matrix clauses that is absent in complement clauses is the expression of two subcategories of aspect – habitual aspect, through the subject inflected auxiliary ko ‘be’, and progressive aspect, via inflected posture auxiliaries like ten ‘sitting’. Recall from Section 5.1.2 that the pluractional aspect is nonetheless allowed in such structures.9
- (33)
- *Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- jie
- [jie
- 1sg
- matera
- matera
- food
- ‘ua
- ‘u=a
- eat=nmlz
- itena!
- i-ten=a]
- 1sg.cor-sit=nmlz
- ‘Tell (them) that I am eating food, sitting!’ (Júma: Elicit)
- (34)
- *Emõmbe’u
- E-mõmbe’u
- 2sg.imp-tell
- gã
- gã
- 3pl
- mãjioka
- [mãjioka
- manioc
- erua
- eru=a
- bring=nmlz
- okovo!
- o-ko-vo]
- 3.cor-be-depend
- Intended: Tell that they habitually bring manioc!’ (Júma: Elicit)
Some syntactic phenomena from main clauses are also lacking in complement clauses. Examples (35) show that complement clauses do not allow embedded questions. In elicitation, the strategy of questioning a phrase within the embedded clause is always rejected, as the examples below demonstrate.
- (35)
- a.
- *Akwaham
- a-kwaham
- 1sg.a-know
- jie
- jie
- 1sg
- mõrõmõ
- [mõrõmõ
- which
- pe
- pe
- 2pl
- jimitakwanha
- ji=mitakwanha
- 1sg=bread
- i-‘u=a.
- i-‘u=a]
- obj-eat=nmlz
- Intended: ‘I know which one of you ate my bread.’ (Júma: Elicit)
- b.
- *Ndakwahavi
- nd-a-kwahav-i
- neg-1sg.a-know-neg
- jie
- jie
- 1sg
- mãngã
- [mãngã
- who
- pe
- =pe
- =to
- W.-ga
- W.=ga
- W.=3sg
- u’ia
- u’ia
- flour
- rerohoa.
- reroho=a]
- take=nmlz
- Intended: ‘I don’t know who Wesley took manioc flour to.’ (Júma: Elicit)
Instead, speakers use two independent clauses to convey the functional correlate of an embedded interrogative question, as shown in (36).
- (36)
- Mãngã
- mãngã
- who
- pe
- =pe
- =to
- Wesley-ga
- Wesley=ga
- W.=3sg.m
- u’ia
- u’ia
- manioc.flour
- erohoi?
- eroho-i?
- take-ais
- Ndakwahavi
- nd-a-kwahav-i
- neg-1sg.a-know-neg
- jie.
- jie
- 1sg
- ‘Who did Wesley take the manioc flour to? I don’t know.’ (Júma: Elicit)
Finally, dependent clauses do not exhibit verb fronting to the clause edge, unlike matrix clauses. Examples in (37) show that a non-final position for the verb within complement clauses is unacceptable.
- (37)
- No verb movement within dependent clauses.
- a.
- Ahepiang
- a-hepiang
- 1sg.a-see
- ki
- ki
- pst
- jie
- jie
- 1sg
- jiruva-ga
- [ji=ruva=ga
- 1sg.poss=father=3sg.m
- pira
- pira
- fish
- ‘ua.
- ‘u=a]
- eat=nmlz
- ‘I saw [my father eating fish].’ (Júma: Elicit)
- b.
- *Ahepiang
- a-hepiang
- 1sg.a-see
- ki
- ki
- pst
- jie
- jie
- 1sg
- ‘ua
- [‘u=a
- eat=nmlz
- jiruva-ga
- ji=ruva=ga
- 1sg.poss=father=3sg.m
- pira.
- pira]
- fish
- Intended: ‘I saw [my father eating fish].’ (Júma: Elicit)
- c.
- *Ahepiang
- a-hepiang
- 1sg.a-see
- ki
- ki
- pst
- jie
- jie
- 1sg
- jiruva-ga
- [ji=ruva=ga
- 1sg.poss=father=3sg.m
- ‘ua
- ‘u=a
- eat=nmlz
- pira.
- pira]
- fish
- Intended: ‘I saw [my father eating fish].’ (Júma: Elicit)
To summarize, in this section, I have shown that complement clauses exhibit fewer finite properties than matrix clauses in Ka(g)wahíva.
5.1.4 Proposal to the size of complement clauses
Recall the structure of matrix clauses assumed in Section 4.1, following dos Santos (2025a):
- (38)
- CP– MoodP – TP – VoiceP – vP – VP
The results achieved in the previous subsections suggest that complement clauses exhibit VP, vP, and VoiceP – for they project argument structure with subjects and objects and allow valency-changing morphology. These structures also project TP – for they display tense morphology and temporal adverbials.
However, the nominalized complement clauses do not exhibit evidence for the CP and MoodP, as matrix clauses do. For CPs, evidence would have come from embedded interrogative phrases forming embedded interrogative questions, verb fronting, and subject agreement. Bearing on the projection of MoodP, we would have expected the occurrence of the reality mood particles. None of these are allowed in complement clauses, however.
I take these facts to mean that complement clauses project a TP in the language, but no further than that. This idea is schematized as follows:
- (39)
- TP– VoiceP – vP – VP
In brief, the Ka(g)wahíva matrix and complement clauses exhibit asymmetries in the amount of structure they project. Matrix clauses project a CP and MoodP, which are missing in complement clauses.
In this way, Ka(g)wahíva complement clauses, as TP embeddings, are similar to TP embeddings in languages like Turkish (Kornfilt & Whitman 2011). The parallel becomes even more evident when nominalization is considered, as Turkish TP embeddings involve nominalization, which we will see later that Ka(g)wahíva also does. We will see, however, that a complete assimilation of the Ka(g)wahíva TP embedded nominalization into a pattern like Turkish is not possible: while Ka(g)wahíva subjects in TP embeddings do not pattern like possessors, subjects in Turkish do (the Turkish genitive subject).
Having motivated this claim, I will, in Section 7, argue that clausal nominalizations contain nominal structure. First, I provide a detailed characterization of nominals in the language.
6 Nominal-like properties in Ka(g)wahíva
In this section, I present the two key properties of Ka(g)wahíva nominals relevant for the discussion of nominalization in complement clauses. The first one is that Ka(g)wahíva nominals are the result of an acategorial root that undergoes syntactic derivation with the functional head n, spelled out as =a. Secondly, Ka(g)wahíva determiners only combine with nPs whose referents are living humans (and God). Alongside the discussion of the first property, I also provide a brief comparison to the cognate =a across the family and show that bare nouns are just bare nPs, not covert DPs.
These properties will serve as a framework against which clause-sized nominals can be compared and probed for their nominal structure and whether they are amenable to the D-assumption or n-assumption of clausal nominalizations. A summary of what each approach to these structures predicts is given in Table 2.
Table 2: Properties of Ka(g)wahíva nominals under approaches to clausal nominalization.
| n=a | determiner | |
| D-assumption | ✗ | ✓ |
| n-assumption | ✓ | ✗ |
6.1 The morpheme =a in Ka(g)wahíva
A hallmark of Ka(g)wahíva nominals is the presence of the enclitic =a, as shown for a common noun in (40). Proper names, too, require the enclitic =a (and a definite marker enclitic) (41a), unless they are used as a vocative, as in (41b).10,11
- (40)
- tajahua
- tajahu=a
- peccary=nmlz
- ‘Peccary’
- (41)
- a.
- Aviva-ga
- Aviv=a=ga
- Avip=nmlz=3sg.m
- pamẽ
- pamẽ
- together
- ga
- ga
- 3sg.m
- hoi,
- ho-i,
- go-ais
- ‘iko.
- ‘i:ko
- dir.evid
- ‘He went together with Avip.’ (Júma: Text)
- b.
- Avip,
- Avip,
- Avip,
- erun
- e-run
- 2sg.imp
- yhya
- yhya
- water
- oreve.
- ore=ve
- 1.excl=for
- ‘Avip, bring water for us.’ (Jupaú: Overheard)
This distribution is further demonstrated in (42): the absence of =a on words in the syntactic position of a noun is ungrammatical, such as heads of possessive phrases and objects of a transitive clause.
- (42)
- a.
- javatxinga
- javatxing=a
- dog=nmlz
- ra’yra
- ra’yr=*(a)
- offspring=nmlz
- ‘Dog’s puppy.’ (Júma: Elicit)
- b.
- Ahepiang
- a-hepiang
- 1sg.a-see
- ki
- ki
- pst
- jie
- jie
- 1sg
- javatxinga
- javatxing=a
- dog=nmlz
- ra’yra
- ra’yr=*(a)
- offspring=nmlz
- ko.
- ko
- realis
- ‘I saw a dog’s puppy.’ (Júma: Elicit)
The formative =a also appears with members from other lexical classes, such as postpositions, number words (which are adverbial), and verbs, in which case the resulting combination distributes as a noun. This is demonstrated in examples in (43) for a postposition-based nominal (43a) and an adverbial number word (43b), which must bear the nominalizer as a complement of a postposition and as the subject of an intransitive verb, respectively.12
- (43)
- a.
- Pea
- pe=*(a)
- to=nmlz
- uvi
- uvi
- from
- gã
- gã
- 3.pl
- hoi.
- ho-i
- go-ais
- ‘From there they went.’ (Text and Elicit)
- b.
- A’ipea
- a’ipe=*(a)
- one=nmlz
- inhãni
- i-nhãn-i
- obj-run-ais
- ovovo.
- o-vo-vo
- 3.cor-go-depend
- ‘(Then), one ran away and went.’ (Text and Elicit)
As indicated, a pervasive use of =a is in combination with verbs, creating deverbal nouns, as in (44), or clause-sized nominals, as in (45); in the latter, the resulting nominal can serve as a complement to clause-embedding verbs, such as hepiang ‘see’ and kwaham ‘know’.
- (44)
- a.
- Jiu’uavera
- ji=u’u-(a)ver-a
- 1sg.poss=bite-nom.pst-nmlz
- ‘My bite.’ (Júma: Text)
- b.
- ’aravera
- ’ar-(a)ver=a
- be.born-nom.pst=nmlz
- ‘Birthday (lit. someone’s falling).’ (Common in speech)
- (45)
- a.
- Ahepia
- a-hepia
- 1sg.a-see
- ji
- ji
- 1sg
- Puré-ga
- [Puré=ga
- Puré=3sg.m
- piauhua,
- piauhua
- piauhua.fish
- ipyhyga.
- i-pyhyg=a]
- obj-catch=nmlz
- ‘I saw Puré catching a piauhua fish.’ (Júma: Elicit)
- b.
- Akwaham
- a-kwaham
- 1sg.a-know
- ji
- ji
- 1sg
- Puré-ga
- [Puré=ga
- Puré=3sg.m
- piauhua
- piauhua,
- piauhua.fish
- ipyhyga.
- i-pyhyg=a]
- obj-catch=nmlz
- ‘I know Puré caught a piauhua fish.’ (Júma: Elicit)
Next, I show that the distribution of cognate =a across the Tupí-Guaraní language family is consistent with that presented here.
6.2 =a across the Tupí-Guaraní family
Cabral (2001) provides a broad survey of =a within the Tupí-Guaraní family, according to which a set of languages has =a as a grammatically productive formative, while the other set comprises languages in which =a only shows as crystallized in some nominal forms.
In languages with productive =a, this formative generally patterns as shown for Ka(g)wahíva above. This fact can be demonstrated for Kamaiurá and Tapirapé, for which there is a systematic description of cognates of =a that allows a comparison to Ka(g)wahíva. I follow the original sources in notating this morpheme as a suffix, and thus set it off with a dash, rather than the equals sign as with the Ka(g)wahíva cognate.13
In Kamaiurá, according to Seki (2000), =a appears with simple nouns occurring as verbal arguments (46a), complements of postpositions (46b), and possessed nouns (46c). Furthermore, =a combines with the verb of clausal nominalizations, as in the notional complement clause in (46d); the source draws on this distribution, and others, to diagnose the nominal status of clausal embeddings.14
- (46)
- a.
- Moĩ-a
- snake-n
- wararuwijaw-a
- dog-n
- o-u’u.
- 3-bite
- ‘The snake bit the dog.’ (Seki 2000: 166, ex. 530)
- b.
- Kunu’um-a
- boy-n
- o-jae’o
- 3-cry
- motawa-a
- food-n
- rehe
- for
- ‘The boy is crying because of food.’ (Seki 2000: 58, ex. 11)
- c.
- Ãaaa
- Ah!
- je=rajyr-a
- 1sg=daughter-n
- te=he=pa.
- foc=vis=ms
- ‘Ah! It’s my daughter indeed.’ (Seki 2000: 346, ex. 1315e)
- d.
- A-kwahaw=ete=rak
- 1sg-know=actual=att
- [mokaw-a
- gun-n
- mepy-taw-er-a].
- buy-nmlz-pst-n
- ‘I know that (he) bought a gun.’ (Seki 2000: 171, ex. 556)
In contrast, =a does not appear in vocatives (contrast (47a)–(47b)) and forms that would otherwise be simple nouns with this formative distribute as verbal predicates, being marked, e.g., for negation identically to how verbs are (47c)–(47d).
- (47)
- a.
- A-ha
- 1sg-go
- kor=a’e=wa
- fs=n.int=ms
- api
- voc
- ‘I’ll go, Uncle.’ (Seki 2000: 60, ex. 20)
- b.
- ’Aŋ
- now
- a’ikĩ
- aten
- api-a
- uncle-n
- rur-i
- come-ais
- a’e
- n.int
- ‘Now, poor Uncle is coming.’ (Seki 2000: 61, ex. 21)
- c.
- Na=je=pyr-ite.
- neg=1sg=house-neg
- ‘I do not have a house.’ (Seki 2000: 161, ex. 499)
- d.
- (Na)= je='amor-ite=ne=ko=py.
- (neg)=1sg=miss.someone-neg=ass=fs=ms
- ‘I do not miss (someone).’ (Seki 2000: 330, ex. 1228)
A similar distribution of =a is found in Tapirapé (Praça 2007), where it is referred to as the ‘referrer’ (hence, glossed as ref). Thus, it appears with verbal arguments, complements of postpositions, and possessed nouns. The formative is also required in clausal nominalizations. Note that =a does not arise in Tapirapé nouns ending in a vowel, which the source represents with a zero allomorph.
- (48)
- a.
- Miãr-a
- deer-ref
- mĩ
- hab
- a-yj
- 3.a-run
- kã’ã-pe.
- jungle-loc
- ‘The deer always run around the jungle.’ (Praça 2007: 40, ex. 88)
- b.
- Takãr-a
- takãra-ref
- ypype
- near
- xãytãtã’i-ø
- start-ref
- i-exãk-i
- 3.b-see-ais
- xyre-ø=agỹ-ø.
- boy-ref=pl-ref
- ‘Near the (men’s house) takãra, the boys saw the stars.’ (Praça 2007: 149, ex. 416)
- c.
- Xe=memyr-a
- 1sg=son-ref
- i-xinyk.
- 3-be.sad
- ‘My son is sad.’ (Praça 2007: 40, ex. 89)
- d.
- Ãpi-ø
- mother-ref
- a-kwããw
- 3.a-know
- [tãtã-ø
- banana-ref
- ’o-pã-ãw-a].
- eat-compl-nmlz-ref
- ‘Mother knows (they) ate all the bananas.’ (Praça 2007: 119, ex. 329)
Again, as with Kamaiurá and Ka(g)wahíva, =a is absence from vocatives in Tapirapé (49a). What is more, in the absence of this morpheme on forms that otherwise are simple nouns with it, the result distributes akin to a verbal predicate; this is demonstrated in (49b), where negation marking is by means of the standard negator of prototypical verbs (49c).
- (49)
- a.
- Kãtowyg
- Kãtowyg
- e-xar
- 2sg.imp-come
- ãpy.
- before
- ‘Kãtowyg, come here, please.’ (Praça 2007: 43, ex. 97)
- b.
- Nã-xe=memyr-i.
- neg-1sg=offspring-neg
- ‘I don’t have children.’ (Praça 2007: 192, ex. 568)
- c.
- Kã’ã-pe
- jungle-loc
- mĩ
- hab
- n=a-kãxym-i.
- neg=3.a-lose.way-neg
- ‘In the jungle, he never gets lost.’ (Praça 2007: 211, ex. 630)
As shown, the distribution of =a is consistent across TG languages. It can be generalized as follows: =a satisfies the syntactic requirement that a word must bear it to function as a noun.
Next, I turn to the analysis of this morpheme in Ka(g)wahíva. This will be followed by a review of non-nominalizing approaches to the cognate morpheme in the family in Section 6.4.
6.3 =a, a nominal categorizing head
Assuming, with Halle & Marantz (1993: and following works), that members of lexical categories, like nouns (verbs, and so on), are not atomic elements, but rather uncategorized roots that distribute as nouns by means of syntactic derivation with a functional categorizer (‘little’ n, ‘little’ v), I characterize =a as a nominal categorizing functional head (‘little’ n) in Ka(g)wahíva. That is, this enclitic derives nouns from acategorial roots, or from members of other lexical classes. To show this schematically, in (50), =a categorizes the root as a noun and heads an NP (50), while in (51), =a categorizes an already categorized root (here, as a verb, by v) to form a noun.
- (50)
- [nP [ ] [n=a ] ]
- (51)
- [nP [vP [v [ ] ] ] ] n=a ]
I note that, with members of other lexical classes, =a’s role is more similar to that of a ‘nominalizer’ that converts a non-nominal element into a nominal one. Therefore, the traditional sense of ‘nominalization’ is appropriate to characterize the syntactic derivation involved in the combination of =a with postpositions, numerals, and verbs.
Further evidence for =a’s nominal categorizing of acategorial roots comes from a non-nominal distribution in the absence of =a. Speaker translates the same word without =a as an existential predicate. For example, in (52a), the form kapyr ‘leaf, money’ functions as a predicate ‘to have money’. Distributionally, in these uses, these forms are verbs, as they bear sentence negation, similar to how more typical verbs like ’u ‘eat’ can be negated, as in (52b). In contrast, the same form kapyr cannot appear with sentential negation in combination with =a, as seen in (52c). I take it, by hypothesis, that its verbal distribution means that in these instances, the a-less forms have been categorized by a null v.
- (52)
- a.
- Ndakapyri
- nd-a-kapyr-i
- neg-1sg.a-leaf-neg
- jie.
- jie
- 1sg
- ‘I do not have money.’ (Elicit)
- b.
- Nda’ui
- nd-a-‘u-i
- neg-1sg.a-eat-neg
- ki
- ki
- pst
- jie
- jie
- 1sg
- pira.
- pira
- fish
- ‘I did not eat fish.’ (Elicit)
- c.
- *Ndakapyrai
- nda-kapyr=a-i
- neg-leaf=nmlz-neg
- jie.
- jie
- 1sg
- ‘Intended: I do not have money.’ (Elicit)
A reviewer asks if the behavior of a-less words like kapyr in (52a) is not more indicative of an underlying verb, rather than an acategorial root, in which case =a attaches not to an acategorial root. We will now see that the proposed analysis of acategorial roots, but not the alternative where they are verbal, also makes the right predictions for other phenomena in Ka(g)wahíva. The argument will come from allomorphic and allosemic patterns involving the nominal past tense and negation morphology with simple nouns, but not derived nominals.
We have seen that Ka(g)wahíva nominals, simple (50) and derived (51), exhibit nominal tense morphology. Importantly, nominal tense, in particular the past tense marker, is lexically conditioned in simple nouns (-ver ∼ -kwer) (53a)–(53b), but only -ver surfaces in deverbal nominals, as in (53c) and (53d).
- (53)
- a.
- Ayvera
- ay-ver=a
- wood-nom.pst=nmlz
- ‘Stick.’ (Also applicable to any piece of wood that is no longer part of a tree) (Júma: Text and Elicit)
- b.
- Jirembirekokwera-hẽa
- ji=rembireko-kwer=a=hẽa
- 1sg.poss=partner-nom.pst=nmlz=3sg.f
- ‘My ex-wife.’ (Common in speech)
- c.
- Jiu’uavera
- ji=u’u-(a)ver-a
- 1sg.poss=bite-nom.pst-nmlz
- ‘My bite.’ (Júma: Text)
- d.
- Kwarupi
- kwarupi
- today
- emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- ga
- [ga
- 3sg.m
- vuva-ga
- v-uva=ga
- 3sg.cor=father=3sg.m
- repiagavera
- repiag-(a)ver=a
- see-nom.pst=nmlz
- oi’i!
- oi’i]!
- some.time.ago
- ‘Tell (them) today that he saw his father yesterday.’ (Júma: Elicit)
I take it that T being inside =a in all nominals, by Mirror Principle reasoning, means that T is closer to the root than n. I propose that the T head is acategorial and can combine with a root before it is categorized; (54) updates the structure in (50) for simple nouns when T is present. The structure of deverbal nominals, with T present, is given in (55).
- (54)
- [nP n=a [TP T [ ] ] ]
- (55)
- [nP n=a [TP T [vP [v-ø [ ] ] ] ] ] ]
Following the intuitions that i) categorizing heads are cyclic/spell-out-inducing heads in the morphology (Moskal 2015; Sande et al. 2020: among many others) and that ii) contextual allomorphy and allosemy are constrained by spell-out domains (Marantz 2013: among others), this proposal predicts that we should see freer allomorphy of T with respect to the root in nominal structures than in verbal ones. The lexically conditioned allomorph of the past tense marker in simple nominals is expected, then, if, in nominalized clauses, there is a v head that instantiates a domain boundary for allomorphy not present in simple nominals, where T attaches to an uncategorized root.
The negator -e’ym also exhibits an idiosyncratic pattern in simple nominals, such as (56), but not derived nominals (57). First, observe that in both instances, Neg comes inside =a (and T, for that matter). By the Mirror Principle again, Neg is closer to the root than n.
- (56)
- U’ie’yma
- u’i-e’ym=a
- -nom.neg=nmlz
- ‘(I ate) flourless (rice).’
- (57)
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- gã
- [gã
- 3pl
- mãjioka
- mãjioka
- manioc
- erure’ymavera!
- erur-e’ym=aver=a]!
- bring-nom.neg=nom.pst=nmlz
- ‘Tell (them) that they did not bring manioc!’
Negation, however, is constrained by the choice of root in simple nominals (58); no such restriction is found in nominalized clauses (59):
- (58)
- * -e’ym=a
- -nom.neg=nmlz
- kavarapea
- cockroach
- ‘Intended: Cockroaches do not have blood.’
- (59)
- E-mõmbe’u
- 2sg.imp-tell
- [kavarapea
- cockroach
- -ø-e’ym=a]!
- -v-nom.neg=nmlz
- ‘Tell (them) that cockroaches do not have blood!’
The item vyhy ‘blood’ is used in (58) as an element of a nominal predicate; in this simple nominal structure, -e’ym is unacceptable (cf. u’i-e’ym=a ‘flourless’ above). In (59), the very same root is used as the predicate of an embedded clause, where (by hypothesis), it has first been verbalized before being negated. As predicted, -e’ym cannot select/not select for a particular root in this context because the verbalizer v creates an additional domain boundary blocking such selection.15
On the alternative that roots are underlying verbs without =a, these systematic idiosyncrasies involving tense and negation in simple, but not derived nominals, are unexpected. There would be no spellout boundary between roots and tense and negation heads in either type of nominal, and thus, we would expect them to arise across nominals, contrary to truth.
I believe the data presented in this section can also speak favorably to the status of =a as realizing n in Ka(g)wahíva, rather than realizing some other bit of morphology, as a reviewer suggests. Let us assume that under the suggested alternative, the noun-categorizing head n is null. In this case, the null formative would create a spellout boundary between the root and tense and negation across all nominals. If so, then the pattern of allomorphy/allosemy just described would still be surprising, this time because it exists at all, as T and Neg would never have access to the root across n.
6.4 Other approaches to =a across TG
Although cognates of =a are regarded as a nominal formative in all descriptive studies within the family (Rodrigues 1953; Lemos Barbosa 1956; Rodrigues 1996; 2001; Vieira 1995; Queixalós 2006), there have been two main opposing views on its function: a nominalizing vs. a non-nominalizing formative.
The earlier works that considered this morpheme, while working within no particular framework, treat -a in the dormant language Tupinambá and in Tapirapé (or label the morpheme) with a nominalizing function (Rodrigues 1953; Almeida et al. 1983; Lemos Barbosa 1956).16 To my knowledge, the generative-based work by Vieira (1993; 1995), which includes a chapter on NP structure and NP quantification in Asuriní do Trocará, is the only other proposal with this view, and also the most systematic defense of the nominalizing view of -a among all.17
Two other proposals, which are framed in the functionalist tradition, have in common that -a is not a nominalizing marker: the case marker and the ‘referential’ marker. The analysis of -a as a case marker comes from Rodrigues (1996; 2001). Based on data from Tupinambá, Rodrigues suggests that Tupinambá -a is the ‘argumentative’ case; he also refers to -a as ‘nominative’ and ‘nominal’ case; this analysis has been extended to other languages in the family (Seki 2000; Borges 2006; Duarte 2007; Magalhães 2007; Cardoso 2008: to cite just a few). Finally, -a is regarded as a ‘referential’ marker in Queixalós (2006). As a referential marker, -a is described as being capable of instituting a referential expression, but does not necessarily have to.
In dos Santos (Under review), I review and compare these approaches as applied to the Ka(g)wahíva dataset. Here, I summarize the main reasons these approaches cannot account for the distribution of =a, at least in Ka(g)wahíva.
As far as the case marking analysis is considered, this analysis solely rests on Rodrigues (2001)’s observation that Tupinambá -a is in complementary distribution with other morphemes that he also considers realizing Case. In Ka(g)wahíva, however, cognate members of the putative paradigm in Tupinambá may co-occur. Furthermore, Rodrigues (2001)’s alternative labeling of -a as ‘nominative’ leads to the expectation that -a should also behave as a nominative case marker. This expectation is not confirmed in Ka(g)wahíva either: while =a appears in all nominal elements in a clause, including non-subjects, cross-linguistically, nominative case is tied to subjects (i.e., external argument in unergative and transitive, and internal argument in unaccusatives) and is unique per clause.
As for the ‘referential’ marking approach, in dos Santos (Under review), I clarify the terminology ‘referential’, which likely denotes some form of definiteness in this approach. Based on this understanding, I demonstrate that under no sense of a definite marker, based on a typology of definites (Dawson & Jenks 2023: 108–109), is =a a definite marker in Ka(g)wahíva.
I have indicated that the non-nominalizing accounts of -a (i.e., the case marker and ‘referential’ marker) are aligned with the functionalist tradition. I do not take that as a coincidence. In fact, it is the very assumption from this framework that lexical categories are inherent that stands between the two opposing views for -a. An explicit rejection of the nominalizing status for this morpheme is found in Cruz & Praça (2019: 631), and serves to illustrate the theoretical nature of this dissent: “It is important to observe that the suffix -a does not promote lexical class change, and therefore is not a nominalizer, since it occurs with both nouns and verbs.”
Since the present work is not committed to the same view that lexical categorization is inherent, but assumes that it arises by means of syntactic derivation (indeed, it is warranted in Ka(g)wahíva given its straightforward account of several patterns of allomorphy and allosemy involving tense and negation morphology), the distribution of =a across simple and other nominals (e.g., deverbal nouns) is straightforward: =a is a root-attaching formative only in the former instances. In the first use, as the root is category-neutral, no lexical change is promoted, just like Cruz & Praça (2019: 631) observed for simple nouns. However, in the second use with e.g., verbs, a change in lexical category is evident; this claim is not disputed in the literature on Tupí-Guaraní, where dependent clauses are viewed as nominalizations that contain an internal verbal core.
So far, we have concluded that Ka(g)wahíva =a is the exponent of the nominalizer n, and that nouns with =a can be the sole elements of a noun phrase that appear in argument position. This distribution makes them bare nouns in argument position. Next, I provide arguments that argumental bare nouns are indeed bare.
6.5 Bare nouns are not covert DPs
The freer distribution of bare nouns in argument position in Ka(g)wahíva is not matched by the behavior of bare nouns in other well-known languages, like English, where only a smaller set of nouns (bare plurals and mass nouns) can occupy an argument position without a determiner.
This raises the oft-explored question of the structure of bare nouns. Some argue that, as nominal arguments are necessarily DPs, bare nouns are always DPs (Longobardi 1994); by hypothesis, languages without overt articles involve a covert DP. Others argue that bare nouns are NPs (Chierchia 1998; 2009), or are ambiguous between NP and DP (Franks & Pereltsvaig 2004), an example of which is the Algonquian language Inuu-aimun (Gillon 2011).
Against this backdrop, probing the structure of Ka(g)wahíva bare nouns becomes a relevant issue given the claim that complement clauses are nominalized TP-embeddings under a nP, lacking a DP. However, if a covert DP is available in some languages, it could, in principle, be available for Ka(g)wahíva as well.
To address this issue, in this section I draw on semantic diagnostics to defend that simple nouns with =a are bare nPs, and not a covert DP. The same diagnostics also show that Ka(g)wahíva bare nouns do not vacillate between nP and DP structures, as Gillon (2011) has shown for bare nouns in Inuu-aimun. Table 3 summarizes the diagnostics and the results expected under each approach to bare nouns laid out above, including the results for a language with an ambiguous nP and DP. The rightmost column, then, includes the results for Ka(g)wahíva under each diagnostic. As one can see, a covert DP structure is not available for simple nominals. Consequently, it must then not be available for derived nominals, like nominalized complement clauses, either.
Table 3: Ka(g)wahíva bare nPs are not covert DPs.
| nP | DP | Innu-aimun | Ka(g)wahíva | |
| wide scope/escape scope | ✗ | ✓ | ✓ | ✗ |
| narrow scope | ✓ | ✗ | ✓ | ✓ |
| uniqueness effects | ✗ | ✓ | ✗ | ✗ |
| can be used anaphorically | ✗ | ✓ | ✓ | ✗ |
| must be used anaphorically (in stretches of discourse) |
✗ | ✓ | ✗ | ✗ |
| obeys law of contradiction | ✗ | ✓ | ✓/✗ | ✗ |
| generic and kind readings | ✗ | ✓ | ✗ | ✗ |
Bare nouns overwhelmingly take narrow scope. In pragmatically neutral contexts, which are associated with VSO clauses, bare nouns have a narrow scope reading. We see this in (60).
- (60)
- Ojuka
- o-juka
- 3.a-go
- Tampinha-ga
- Tampinha=ga
- T.=3sg.m
- karuvoruhua
- karuvoruhu=a
- paca=nmlz
- ra’e.
- ra’e
- today.pst
- ‘Tampinha killed (a) paca (we said).’
- Context: The speaker directly quotes what she and others said as they saw Tampinha arrive with a paca. (Júma: Text)
Example (61) provides additional textual confirmation for the narrow scope interpretation associated with bare nouns. The second sentence states the outcome of an agouti-hunting outing, and it also reveals that the bare noun iramburava ‘agouti’ in the first sentence has a narrow scope reading.
- (61)
- Ore
- ore
- 1.excl
- hoi
- ho-i
- go-ais
- iramburava
- iramburav=a
- agouti=nmlz
- reka
- rek=a
- look.for=nmlz
- orokovo,
- oro-ko-vo
- 1.excl.cor-be-depend
- ‘itekiko
- ‘i:te:ki:ko
- dir.evid
- rõmĩa.
- rõmĩa
- frust
- Ndorepiaki
- nd-or-epiak-i
- neg-1.excl-see-neg
- ki
- ki
- pst
- ore
- ore
- 1.excl
- ko.
- ko
- realis
- ‘We had looked for an agouti. We did not see (any).’ (Júma: Text)
A narrow scope reading is the only alternative available with propositional attitude verbs like pota ‘want, desire’ (62a), in conditional clauses (62b), and in questions (62c).18
- (62)
- a.
- Opotaete
- o-pota-ete
- 3.a-want-really
- ga
- ga
- 3sg.m
- tapy’ynha
- tapy’ynh=a
- non.indigenous=nmlz
- kwarupi.
- kwarupi
- today
- ✓ Context 1: ‘He wants (to date) any non-indigenous person today.
- ✗ Context 2: ‘He wants (to date) a specific non-indigenous person today.’ (Júma: Elicit)
- b.
- Agui-ga
- Agui=ga
- Agui=3sg.m
- oje’o
- o-je’o
- 3.a-cry
- javatxinga
- [javatxing=a
- dog=nmlz
- ekiramẽ]
- ekir=a=mẽ]
- enter=nmlz=if
- nãhẽ.
- nãhẽ
- pot
- ✓ Context 1: ‘Agui will cry if any dog enters (the house).’
- ✗ Context 2: ‘Agui will cry if a specific dog enters (the house).’ (Júma: Elicit)
- c.
- O’u
- o-‘u
- 3.a-eat
- po
- po
- irr
- ndera’yra-ga
- nde=ra’yra=ga
- 2sg.poss=offspring=3sg.m
- pira?
- pir=a
- fish=nmlz
- ✓ Context 1: ‘Does your son eat any fish?’
- ✗ Context 2: ‘Does your son eat a particular fish?’ (Júma: Elicit)
Another diagnostic for the presence of a covert D is uniqueness. If Ka(g)wahíva bare nouns had a covert definite DP, then we would expect them to be associated with uniqueness effects, that is, they must refer to a unique referent within a context. (Note that, as uniqueness is associated with a definite determiner, the lack of a uniqueness effect is not evidence for a lack of DP structure. Therefore, this diagnostic only tests whether bare nouns are built as definite DPs.)
Ka(g)wahíva texts show that bare nouns lack uniqueness effects. Consider the following excerpts from the same text, in which the speaker narrates an eventful weekend in a different village. Among the activities was going hunting karuvoruhua ‘paca’. The text has five occurrences of karuvoruhu=a, (63a)–(63e), which cannot be a unique referent.
- (63)
- a.
- Aramẽ
- aramẽ
- after.that
- txiro
- txiro
- progr.
- ore,
- ore,
- 1.excl
- txaho
- txa-ho
- opt-go
- kavyripe
- kavyr=ipe
- jungle=to
- rehe,
- at
- at
- karuvoruhua
- karuvoruhu=a
- paca=nmlz
- rehe,
- rehe
- at
- ore
- ore
- 1.excl
- ’ei,
- ’e-i
- say-ais
- ’itekiko
- ’i:te:ki:ko
- dir.evid
- nõmĩa.
- nõmĩa
- frustr
- ‘After that, we go to the jungle after pacas, we said.’ (Line 72)
- b.
- Karuvoruhua
- karuvoruhu=a
- paca=nmlz
- txaho
- txa-ho
- opt-go
- epiaga
- epiag=a
- see=nmlz
- ore.
- ore
- 1.excl
- ‘We go to see pacas (he said).’ (Line 80)
- c.
- Karuvoruhua!,
- karuvoruhu=a
- paca=nmlz
- ’i
- ’i
- say
- ki
- ki
- pst
- ka
- ka
- 3sg.m
- kora.
- ko=ra
- realis=today.pst
- ‘A paca!, he said.’ (Line 88)
- Context: Said as someone in the group spots something under the water which, to him, appeared to be a paca.’ (Line 88)
- d.
- Karuvoruhua
- karuvoruhu=a
- paca=nmlz
- txia
- txi=a
- that=nmlz
- jive
- ji=ve
- 1.sg=to
- ko
- ko
- realis
- nõmĩa.
- nõmĩa
- frust
- ‘To me, it was a paca (he said).’ (Line 96)
- e.
- Ojuka
- o-juka
- 3.a-go
- Tampinha-ga
- Tampinha=ga
- T.=3sg.m
- karuvoruhua
- karuvoruhu=a
- paca=nmlz
- ra’e.
- ra’e
- today.pst
- ‘Tampinha killed a paca (we said).’ (Line 114)
Anaphoricity can also be used to probe the structure of bare nouns. If bare nouns occur with a covert determiner, we would expect all uses of a bare noun in a stretch of discourse to refer to the same entity. We have just seen that this is not true: the several occurrences of karuvoruhu=a in (63) do not pick out the same referent.
In fact, the strategy for subsequent reference to an entity previously introduced by a bare noun is simply not pronouncing that noun again. In the same text from which (63e) from above draws upon, the referent of the last occurrence of karuvoruhu=a in (63e) is referred back to in (64). As seen, no overt noun is used.19
- (64)
- Kwari-ga
- Kwari=ga
- Kwari=3sg.m
- rekoi,
- reko-i
- be.with-ais
- ipiroga,
- i-pirog=a
- obj-peel=nmlz
- ’itekiko.
- ’i:te:ki:ko
- dir.evid
- ‘(Then), Kwari stayed peeling (the paca).’ (Line 115)
In the rare instances where a bare noun matches the description of a previously introduced NP, the reference is disjoint. The three sentences in (65) are uttered one right after the other. The first sentence has pira ‘fish’, as well as the second one, but neither refers to the same referent. In the third sentence, the referent of pira from (65b) is referred back; as expected, no overt NP is used.
- (65)
- a.
- Karukamẽ
- karuka=mẽ
- afternoon=to
- ki
- ki
- pst
- ji
- ji
- 1.sg
- hoi
- ho-i
- go-ais
- pira
- pir=a
- fish=nmlz
- rerua
- reru=a
- bring=nmlz
- oi’i.
- oi’i
- some.time.ago
- ‘In the afternoon, I went to bring fish (by fishing).’
- b.
- Pira
- pir=a
- fish=nmlz
- reruri.
- rerur-i
- bring-ais
- ‘(I) brought (a) fish.’ (Júma: Text)
- c.
- Ko’emamẽ
- ko’ema=mẽ
- morning=to
- jie
- jie
- 1.sg
- imõtyryryi.
- i-mõtyryry-i
- obj-fry-ais
- ‘On the next morning, I fried (the fish).’ (Júma: Text)
Additionally, bare nouns with =a do not pass the ‘test of consistency’ or ‘law of contradiction‘ (Löbner 2002). According to this test, a noun referring to an individual cannot both hold and not hold a given property.
- (66)
- a.
- Tapi’ira
- tapi’ir=a
- tapir=nmlz
- ehe.
- ehe
- be.tasty
- Tapi’ira
- tapi’ir=a
- tapir=nmlz
- ndahei.
- nda-he-i
- neg-be.tasty-neg
- ‘There is tapir that is delicious, there is tapir that is not delicious.’ (Júma: Elicit)
- b.
- Kunhã
- kunh=ã
- woman=nmlz
- ikatu.
- i-katu
- obj-be.pretty
- Kunhã
- kunh=ã
- woman=nmlz
- ndikatui.
- nd-i-katu-i
- neg-obj-be.pretty-neg
- ‘There is a woman who is pretty, and there is a woman who is not pretty.’ (Júma: Elicit)
Finally, bare nouns are also used for generic and kind readings.20
- (67)
- a.
- Ndavyi
- nda-vy-i
- neg-blood-neg
- kavarapea.
- kavarape=a
- cockroach=nmlz
- ‘Cockroaches do not have blood.’ (Júma: Elicit)
- b.
- Mbahira
- Mbahira
- Mbahira
- vapo
- v-apo
- 3.a-make
- kunhã.
- kunh=ã
- woman=nmlz
- ‘Mbahira created women.’ (Jupaú: Text)
In sum, bare nouns in Ka(g)wahíva exhibit the semantic profile of bare nPs, and not the profile of a noun with a covert DP structure.
Next, I present data showing that Ka(g)wahíva simple nouns may still project the category D through a semantically restrictive set of determiners. This discussion complements the background on nominals, which I will draw upon to assess the nominal status of complement clauses.
6.6 Determiners and demonstratives
In Ka(g)wahíva, the elements that are commonly analyzed as members of the determiner category are (in)definiteness markers and demonstratives.
Starting with the definite enclitics, the full set of enclitics is presented in Table 4. In the ethnolects under discussion here (Júma and Jupaú), these enclitics impose a strict semantic restriction on their complement nouns – they combine only with nouns whose referents are living humans or God.21
Table 4: (In)definiteness determiners.
| determiner | gloss |
| =ga | ‘sg.m definite determiner’ |
| =hẽa | ‘sg.f definite determiner’ |
| =gã | ‘pl definite determiner’ |
| =mõ | ‘indefinite determiner’ |
These enclitics exhibit the profile of definite determiners. For instance, nominals marked with them cannot appear in existential contexts, as the pairs with and without the definite enclitics illustrate below.22
- (68)
- a.
- Ndokoi
- nd-o-ko-i
- neg-3.a-be-neg
- akwemba’ea
- akwemba’e=a
- man=nmlz
- jahya
- jahya
- moon
- pyri.
- pyri
- near
- ‘There is no man on the moon.’ (Júma: Elicit)
- b.
- *Ndokoi
- nd-o-ko-i
- neg-3.a-be-neg
- akwemba’ea-ga/gã
- akwemba’e=a=ga/gã
- man=nmlz=3sg.m/3.pl
- jahya
- jahya
- moon
- pyri.
- pyri
- near
- Intended: ‘There is no man on the moon.’ (Júma: Elicit)
- (69)
- a.
- Ndokoi
- ndo-ko-i
- neg-3.a-be-neg
- 21mẽ
- 21=mẽ
- 21.village=to
- tapy’ynha.
- tapy’ynh=a
- non.indigenous=nmlz
- ‘There are no non-indigenous people in the 21 (village).’ (Júma: Elicit)
- b.
- *Ndokoi
- ndo-ko-i
- neg-3.a-be-neg
- 21mẽ
- 21=mẽ
- 21.village=to
- tapy’ynha-gã.
- tapy’ynh=a=gã
- non.indigenous=nmlz=3.pl
- Intended: ‘There are no non-indigenous people in the 21 (village).’ (Júma: Elicit)
The enclitic determiner =ga ‘3rd singular masculine’ is used to refer to a unique referent like Tupan ‘God’.
- (70)
- Mãrãnũhũ
- mãrãn-ũhũ
- why-aug
- po
- po
- irr
- txia
- txia
- dem
- txirove?
- txirove
- now
- Tupana-garõi
- Tupan=a=ga=rõi
- God=nmlz=3sg.m=const.neg
- auhu.
- auhu
- seem
- ‘Why (is) this now? It seems it is not God.’ (Lamenting about the COVID situation) (Júma: Text)
Nouns, in combination with the enclitic determiners, pass the ‘test of consistency’ or ‘law of contradiction‘ (Löbner 2002). This is expected, as definite determiners create arguments that refer to individuals, and definite nouns obey the law of contradiction. For example, in (71), kunhã ‘woman’ must refer to the same individual in each instance. Therefore, the enclitic must be a definite determiner.
- (71)
- #Kunhã-hẽa
- kunh=ã=hẽa
- woman=nmlz=3sg.f
- ikatu.
- i-katu
- obj-be.pretty
- Kunhã-hẽa
- kunh=ã=hẽa
- woman=nmlz=3sg.f
- ndikatui.
- nd-i-katu-i
- neg-obj-be.pretty-neg
- Intended: ‘The woman is pretty, and the woman is not pretty.’ (Júma: Elicit)
Ka(g)wahíva also exhibits a nonspecific indefinite determiner =mõ in contexts where the speaker does not have a particular referent in mind, as in (72).
- (72)
- Ndahepiagi
- nd-a-hepiag-i
- neg-1sg.a-see-neg
- ki
- ki
- pst
- ji
- ji
- 1sg
- piramõ.
- pir=a=mõ
- fish=nmlz=indef.det
- ‘I did not see any fish.’ (Said upon returning from a fishing trip with no fish) (Júma: Text)
Nouns in combination with =mõ exhibit the typical behavior of non-specific indefinites: they must take narrow scope relative to negation and verbs like pota ‘want, desire’ and within conditional clauses; see dos Santos (2024b: 70–71) for examples.
Finally, I present the demonstratives on account that they may instantiate the category D in Ka(g)wahíva. Thus, demonstratives show a three-way deictic distinction, which distinguishes between proximal, medial, and distal. Each of these, in turn, shows a two- or three-way categorization of their referent, distinguishing between elongated, round, and flat. This system yields the rich paradigm presented in Table 5.
Table 5: Demonstratives.
| elongated | round | flat | |
| proximal | ‘ãngã | koro | kohoa |
| medial | mĩhĩa | txiro | pehea |
| distal | mĩhĩa | txijia | txijia |
| examples | mortar | ball, puppy, tapir | hose |
Demonstratives follow the noun (73a), but they can also appear to its left and discontinuous from the NP when focused (73b).
- (73)
- a.
- Javatxinga
- javatxing=a
- puppy/dog=nmlz
- txijia
- txijia
- dem:dist:round/flat
- ‘That (distant) puppy’ (Júma: Elicit)
- b.
- Txijia
- txijia
- dem:dist:round/flat
- ki
- ki
- pst
- javatxinga
- javatxing=a
- dog=nmlz
- omõnhãn
- o-mõnhãn
- 3.a-corral
- tapi’ira
- tapi’ira
- tapir
- ko.
- ko
- realis
- ‘That dog corralled the tapir.’ (Júma: Elicit)
Taken altogether, the data presented in this subsection show that some Ka(g)wahíva nouns may project a DP. However, excluding the indefinite enclitic =mõ, the other determiners are very selective of the semantics of their noun complements. Therefore, if there is any chance a clausal nominalization can exhibit a D, it is likely to be with =mõ.
To summarize this background section, Ka(g)wahíva nominals may be bare nouns, the combination of an acategorial root and the formative =a, or a noun with one of the best candidates for a D-head in the language, the (in)definiteness markers, or the demonstratives.
Given these results and the hypothesis that clausal complements are nominalizations, it should be straightforward to see whether they also project a DP: they should be able to appear with one of the Ka(g)wahíva D-elements.
I turn now to the nominal properties of complement clauses to examine this possibility as well.
7 Nominal-like properties of complement clauses
Clausal complements exhibit several nominal properties. Syntactically, they occur in object position, as several of the examples illustrated thus far show; example (74a) is given for convenience. They also occur in (derived) subject positions (74b–c).
- (74)
- a.
- Ahepiang
- a-hepiang
- 1sg.a-see
- ki
- ki
- pst
- jie
- jie
- 1sg
- jiruva-ga
- [ji=ruva=ga
- 1sg.poss=father=3sg.m
- pira
- pira
- fish
- ‘ua.
- ‘u=a]
- eat=nmlz
- ‘I saw [my father eating fish]’ (Júma: Elicit)
- b.
- Ahy
- ahy
- be.hard
- ki
- ki
- pst
- ndererekwavera
- [nde=rereko-(a)ver=a]
- 2sg.b=be.with-nom.pst=nmlz
- oi’i.
- oi’i
- some.time.ago
- ‘Caring for you is hard.’ (Júma: Elicit)
- c.
- Ahy
- ahy
- be.hard
- moja
- [moja
- snake
- jiu’uavera.
- ji=u’u-(a)ver=a]
- 1sg.b=bite-nom.pst=nmlz
- ‘That the snake bit me hurts.’ (Júma: Elicit)
Additional evidence for their nominal status draws on their morphosyntax. First, observe that these structures require the nominal-exclusive enclitic =a, which I have proposed to be the overt exponence of the nP head.
- (75)
- Ahepiang
- a-hepiang
- 1sg.a-see
- ki
- ki
- pst
- jie
- jie
- 1sg
- jiruva-ga
- [ji=ruva=ga
- 1sg.poss=father=3sg.m
- pira
- pira
- fish
- ‘ua.
- ‘u=*(a)]
- eat=nmlz
- ‘I saw my father eating fish.’ (Júma: Elicit)
Perhaps unsurprisingly, the nominal tense and negation morphology is also common to complement clauses.
- (76)
- Tense is nominal in complement clauses
- a.
- Garembiriko-kwer=a=hẽa
- ga=rembiriko-kwer=a=hẽa
- 3sg.m=partner-nom.pst=nmlz=3sg.f
- jie
- jie
- 1sg
- arekoajam.
- a-reko-ajam
- 1sg.a-marry-again
- ‘His ex-wife, I married (her).’ (Júma: Elicit)
- b.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- Aguia-ga
- Aguia=ga
- A.=3sg.m
- mbiara
- mbiara
- meat
- mbohirukaravera
- mbohirukar-(a)ver=a
- drop-nom.pst=nmlz
- javatxinga
- javatxinga
- dog
- pe!
- pe
- to
- ‘Tell (them) that Agui dropped the meat for the dog!’ (Júma: Elicit)
- (77)
- Negation is nominal in complement clauses
- a.
- Ere’u
- ere-‘u
- 2sg.a-eat
- nde
- nde
- 2sg
- mbotavoruhua.jauhuva’ea
- mbotavoruhua.jauhuva’ea
- rice
- u’ie’yma
- u’i-e’ym=a
- flour-nom.neg=nmlz
- rehe?
- rehe
- at
- ‘Do you eat rice without flour?’ (Júma: Elicit)
- b.
- Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- gã
- [gã
- 3.pl
- mãjioka
- mãjioka
- manioc
- rerure’ymavera!
- rerur-e’ym-(a)ver=a]
- bring-nom.neg-nom.pst-nmlz
- ‘Tell (them) that they did not bring manioc!’ (Júma: Elicit)
These results strongly suggest that Ka(g)wahíva does project a nominal structure in clausal nominalizations, and that this structure contains at least nP, whose head is spelled out by =a.
Does Ka(g)wahíva also project the crosslinguistically more common nominal layers in clausal nominalizations, that is, DP and PossP? In the remainder of this section, I argue that neither projection is afforded in these structures.
Starting with PossP, one piece of evidence that other works have relied on to argue for its presence in clausal nominalizations is the case expression of subjects. For instance, in Khalkha Mongolian, subjects in nominalized complement clauses may appear in the genitive form, similar to possessive nominals, as in examples (78a–b).
- (78)
- a.
- Pronoun possessor in genitive form in Khalkha
- Chi
- 2sg.nom
- minii
- 1sg.gen
- nom-iig
- book-acc
- unsh-san
- read-pst
- ‘You read my book’ (Lim 2026: 6)
- b.
- Nonfinite nominalized complement clause with genitive subject in Khalkha
- Naraa
- N.nom
- [minii
- 1sg.gen
- unta-j
- sleep-prog
- bai-sn]-iig(=min’)
- cop-pfv-acc=1sg.poss
- med-ne.
- know-n.pst
- ‘Naraa knows that I was sleeping’ (Lim 2026: 34)
Likewise, subjects in Turkish are marked genitive in TP nominalizations, but not in other types of embeddings; contrast (79)–(80).
- (79)
- Nominative subject in finite embeddings in Turkish
- Hasan
- Hasan
- [uşak
- servant.nom
- oda-yı
- room-acc
- temizle-sin]
- clean-opt/sbjnct.(3.sg)
- isti-yor.
- want-pr.prg.(3.sg)
- ‘Hasan wants that the servant should clean the room.’ (Kornfilt & Whitman 2011: 1301)
- (80)
- Genitive subject in nominalized embedding in Turkish
- Hasan
- Hasan
- [uşağ-ın
- servant-gen
- oda-yı
- room-acc
- temizle-me-sin]-i
- clean-nf.nmlz-(3.sg)-acc
- isti-yor.
- want-pst.(3.sg)
- ‘Hasan wants that the servant should clean the room.’ (Kornfilt & Whitman 2011: 1301)
For Kalkha (Lim 2026), on the assumption that the genitive case is assigned by the nominal structure PossP, the author argues that the expression of genitive subjects in nominalized clauses suggests the projection of PossP.
The diagnostic from genitive subjects is not available in Ka(g)wahíva – pronominal possessors are not formally distinct from regular pronouns that occur in argument positions, whether they are in a matrix or in an embedded clause. Examples (81) readily illustrate this isomorphism. The same data, however, reveals a prosodic difference: the first person singular pronoun is a proclitic in (81a), but a free-standing word in (81c).23,24,25
- (81)
- a.
- Epiang
- e-piang
- 2sg.imp-see
- jikyva!
- ji=kyva
- 1sg=lice
- ‘See my lice!’ (Júma: Elicit)
- b.
- Amõka’ẽ
- a-mõka’ẽ
- 1sg.a-roast
- ji
- ji
- 1sg
- e’yi
- e’yi
- many.ex
- pira.
- pira
- fish
- ‘I roasted many fish.’ (Júma: Elicit)
- c.
- Okwaha
- o-kwaha
- 3.a-know
- ki
- ki
- pst
- ga
- ga
- 3sg.m
- ji
- [ji
- 1sg
- nderepiaka.
- nde=repiak=a]
- 2sg.b=see=nmlz
- ‘He knew I saw you.’ (Júma: Elicit)
These differences would follow naturally under a view in which Ka(g)wahíva embedded subjects are syntactically distinct from pronominal possessors. To see how, let us assume that prosodic constituents must match syntactic constituents, e.g., a phrase in a syntactic constituent must be matched by a phonological phrase, which is its corresponding prosodic type (Selkirk 2011). That being said, I suggest that pronominal possessors in possessive phrases, but not pronominal subjects in embedded clauses, undergo procliticization into the head of the possessive phrase because it forms a syntactic unit with that head, which is then matched by a correspondent prosodic constituent in the phonology. In other words, pronouns do not occur in the same syntactic position in these two structural contexts; if they did, they would be mapped into identical prosodic constituents and be equally susceptible to procliticization. One can visualize the position of subjects within embedded clauses as in (82), which represents a clausal complement to an embedding verb with the embedded subject in its original merge site.
- (82)
- [vP v [nP n=a [TP T [VoiceP EA [Voice [vP [v [ IA ] ] ] ] ] ] ] ] ]
Further evidence for a distinct structural position between embedded subjects in clausal nominalizations and possessors in possessive nominals comes from an asymmetry under Ā extraction. First, observe that embedded subjects, as in (83a), can be extracted from the complement clause. In contrast, possessors in possessive phrases, once targeted for extraction, only extract if they pie-piped the entire possessive phrase to clause-initial position, as in (83b). Extraction of the possessor only, leaving behind the possessed noun, is not an option (83c).
- (83)
- a.
- Mãngã
- mãngã
- who
- nde
- nde
- 2sg
- erepiang
- ere-piang
- 2sg.a-see
- piauhua
- [__
- ipyhyga?
- piauhua
- piauhua.fish
- i-pyhyg=a]
- obj-catch=nmlz
- ‘Who did you see catching the piauhua fish?’ (Júma: Elicit)
- b.
- Mãngã
- mãngã
- who
- pira
- pira
- clothes
- ka
- ka
- 3sg.m
- txiro
- txiro
- progr
- omõngi
- o-mõngi
- 3.a-dress
- (a’i
- __
- ki
- (a-‘i
- 1sg.a-say
- ji
- ki
- pst
- kora).
- ji
- 1sg
- ko=ra)
- dir.evid
- ‘Whose clothes is he dressing? (I asked).’ (Júma: Text)
- c.
- *Mãngã
- mãngã
- who
- ka
- ka
- 3sg.m
- txiro
- txiro
- progr
- omõngi
- o-mõngi
- 3.a-dress
- pira?
- [__
- pira]
- clothes
- Intended: ‘Whose clothes is he dressing?’ (Júma: Elicit)
Under the idea that subjects of nominalized embedded clauses and possessors in possessive phrases do not occupy the same position in these nominal structures, this asymmetry makes sense: pronouns can be targeted for extraction as subjects of nominalized complement clauses, but not as possessors.
Note that this distribution directly contrasts with the pattern in English gerundive nominalizations, where both possessors and genitive subjects behave alike under extraction: neither can be extracted.
- (84)
- *Whose did you regret playing the piece?
- (85)
- *Whose did you see book? (Pires & Milsark 2017: 9-10)
With that, I take this Ka(g)wahíva pattern, in which embedded subjects, but not possessors, can be extracted, to be highly suggestive that they occur in different structural positions. Otherwise, we would not be able to explain the asymmetry in extraction. This fact also leads me to conclude that Ka(g)wahíva nominalized clauses do not project a PossP, which houses subjects of nominalized embedded clauses in other languages.
I turn now to the question of whether a DP projection is also present in clausal nominalizations in Ka(g)wahíva in addition to nP. In some languages, a DP layer is diagnosed by the presence of a definite determiner or a demonstrative, which may sometimes directly merge with a clause (Takahashi 2010; Miyagawa 2011; Hanink 2016; Pietraszko 2019; Hankamer & Mikkelsen 2021; Bochnak & Hanink 2022: and many others). The bracketed structures from Modern Greek and Persian illustrate this alternative below:
- (86)
- a.
- Greek to-clause
- [To
- the-nom
- [oti
- that
- ehis
- have-2sg
- filus]]
- friends-acc
- simeni
- mean-3sg
- pola.
- much
- ‘That you have friends means a lot.’ (Roussou 1991: 78 apud Takahashi 2010: 353)
- b.
- Persian in-clause
- [In
- this
- [ke
- that
- to
- you
- u-rā
- he-obj
- da’vat
- invitation
- na-kard-i]]
- neg-did-2sg
- mādar-at-ro
- mother-2sg-obj
- nārāhat
- upset
- kard.
- did.3sg
- ‘That you did not invite him made your mother upset.’ (Farudi 2007: 10 apud Takahashi 2010: 353)
As for Ka(g)wahíva, recall that the elements that can be best described as realizing D surface through a set of definite enclitics, an indefinite enclitic =mõ, and demonstratives in the language. They, however, do not appear with nominalized clausal complements. We can see this in examples (87)–(88) for a 3pl definite determiner and for a demonstrative, respectively.
- (87)
- *Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- (gã)
- [(gã)
- 3pl
- ejuva-ga
- e-juva=ga
- 2sg.cor=father=3sg.m
- gã
- gã
- 3pl
- repiaga
- repiag=a
- see=nmlz
- (gã)!
- (gã)]
- 3pl
- ‘Tell [your father saw them]!’ (Júma: Elicit)
- (88)
- *Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- txiro
- [txiro
- dem:med:round
- ejuva-ga
- e-juva=ga
- 2sg.cor=father=3sg.m
- gã
- gã
- 3pl
- repiaga!
- repiag=a]
- see=nmlz
- ‘Tell [your father saw them]!’ (Júma: Elicit)
The incompatibility between definite determiners and demonstratives and clausal complements is not surprising, as these elements are highly selective of the complements they take – the definite determiners combine with nouns that refer to living humans or God. In contrast, demonstratives are used with referents that are categorized as either elongated, round, or flat (see Section 6.6). Put simply, nominalized complement clauses do not exhibit the suitable semantics that can combine with these determiners, as they do not denote a property of individuals that are living humans, God, nor do they have a shape.
However, the indefinite enclitic =mõ in principle presents no similar semantic incompatibility with a nominalized clausal complement. Nevertheless, complement clauses with =mõ are rejected, as shown by (89).
- (89)
- *Emõmbe’u
- e-mõmbe’u
- 2sg.imp-tell
- ejuva-ga
- [e-juva=ga
- 2sg.cor=father=3sg.m
- gã
- gã
- 3pl
- repiagamõ!
- repiag=a=mõ]
- see=nmlz=indf.det
- ‘Tell [your father saw them]!’ (Júma: Elicit)
Therefore, while the highly restrictive animacy/humaneness-combining enclitics and shape-combining demonstratives are likely ruled out from combining with nominalized complement clauses on semantic grounds, no such semantic restriction can be used as the culprit for the ill-formedness of =mõ with clausal complements. Ka(g)wahíva complement clauses simply do not project a DP.
In sum, Ka(g)wahíva complement clauses exhibit nominal properties that are contributed by the nominal layer nP, and no other nominal layers, such as PossP or DP, is projected in them.
8 Discussion and implications
Having shown that complement clauses in Ka(g)wahíva present both nominal and verbal properties, I propose that these facts can be straightforwardly accounted for by assuming the Mixed Extended Projection or Functional Nominalization Thesis analysis to nominalizations (Borsley & Kornfilt 2000; Kornfilt & Whitman 2011). Under this view, nominal properties of clausal nominalizations are contributed by a nominal structure that is projected immediately above the verbal core. Thus, assuming this approach and given the results from Section 5 and the previous section – that clausal nominalizations project at most a TP, and the nominal layer includes only nP – I propose to represent clausal nominalizations in Ka(g)wahíva as in (90):
- (90)
- [vP v [nP n=a [TP T [VoiceP Voice [vP [v [ ] ] ] ] ] ] ] ]
The results of the study of clausal nominalizations in Ka(g)wahíva can inform two questions that have figured prominently in the literature on clausal nominalization. First, what nominalizes the clause? A second question is, does the nominalizer type restrict the amount of clause structure that can be embedded?
Regarding the identity of the nominal layer in clausal nominalizations, several works postulate a DP that directly merges with the clausal content (Davies & Dubinsky 1998; Borsley & Kornfilt 2000; Takahashi 2010; Miyagawa 2011; Pietraszko 2019; Alexiadou 2020; Iordăchioaia 2020; Hankamer & Mikkelsen 2021); cf. Grimshaw (2000) for arguments that a clause-embedding DP structure should not exist as DPs are the extended projection of NPs, and therefore, the latter should figure between the clausal content and the DP. I have dubbed this idea the D-assumption of clausal nominalization, which contrasts with the competing proposal that nP may also nominalize a clause, under the n-assumption of clausal nominalization; both assumptions are schematized below:
- (91)
- a.
- [dp d [… clausal structure … ] ] (The D-assumption of clausal nominalization)
- [nP n [… clausal structure … ] ] (The n-assumption of clausal nominalization)
In Ka(g)wahíva, however, nominalized clauses lack a DP layer. Instead, they are bare nPs, with nominalization proceeding through the functional nominalizer =a. This is not in itself a surprising result for the language, as it allows nonclausal bare nPs to appear in argument positions (see Section 6); borrowing the terms from Pereltsvaig (2006) and Lim (2026), Ka(g)wahíva has (clausal and nonclausal) small nominals.
Additionally, this result furnishes a novel example to the growing body of literature showing that the nominal layer in clausal nominalizations may not include a DP layer, as argued for some Japanese nominalized embedded questions (Nasu et al. 2024), and for nominalized converbial adjuncts in the Mongolic language Khalkha (Lim 2025; 2026). Crucially, however, they are still reduced – they are not CPs.
A second related question regarding clausal nominalizations concerns the relationship between nominalizer type and the amount of structure they can nominalize. Recent work has attempted to identify crosslinguistic restrictions on the amount of content that a given nominalizer may embed (Alexiadou 2020; Iordăchioaia 2020; Lim 2025; 2026). An upshot of these discussions for the present discussion is that high layers of clausal structure, such as CP or TP, may be nominalized only by a high layer of nominal structure, such as DP (or PossP for TP-embedding nominalizations in Lim (2025; 2026)). On the other hand, low levels of clause structure (below TP) are in principle compatible with both levels of nominal structure (cf. Alexiadou (2020: 102), who suggests that DPs cannot merge directly with vPs, and it needs nP before taking a vP). Particularly relevant to us is their claim that n-nominalizations should not be able to embed TPs, but should allow a structure smaller than a TP.
Given the results arrived at for Ka(g)wahíva in this paper, these theories are too restrictive. I have shown that Ka(g)wahíva complement clauses are TP-embedding nominalizations by nP, not DP. In fact, no DP is projected in them. This pattern is simply unexpected under those proposals.26 Providing an alternative explanation as to why Ka(g)wahíva allows TP-sized nominalizations by nP will take us too far in this paper. This, I leave for future research.
For now, I should note that the proposals in Alexiadou (2020) and Lim (2026: footnotes 35, 36) allow some flexibility in their account, which may be explored for a potential account for the Ka(g)wahíva pattern. In particular, the authors consider the cut-off points for nominalization – how much clausal structure content can be nominalized by what – not as driven by an absolute cut-off, but as based on a partitioning of the clausal structure in domains that are semantically-grounded (into discourse linking, anchoring, point of view, classification) (Ramchand & Svenonius 2014; Wiltschko 2014). In this view, nominalizations target these semantic domains: for instance, because discourse linking is commonly attributed to CP in the clausal structure and to DP in the nominal domain, D-nominalization, but not n-nominalization, can target CP structures. In contrast, the domain classification is part of vP in the clausal structure, and to nP in the nominal structure; therefore, n-nominalizations may target vP for nominalization.
Under this idea, we could say that in Ka(g)wahíva, discourse linking is realized by CPs and DPs, while anchoring is realized by TPs and nPs. Initial support for the idea that nPs pertain to the anchoring domain comes from the fact that bare nouns (bare nPs) are allowed in argument position. Being on the same domain allows n to target a TP for nominalization.27
Finally, by the same idea that nominalization targets semantically-grounded domains, why then does Ka(g)wahíva not allow CP nominalizations by D, since the language exhibits elements that may project a DP layer with the (in)definiteness enclitics and demonstratives? The only firm answer is the one I suggested in section 7. That is, the absence of D in clausal nominalizations is likely due to the restrictor nP, at least in the instances with definite enclitics and demonstratives. That is, clause-sized nPs cannot serve as a restrictor nP because they do not refer to living humans (or God), nor can they be categorized by shape, as demonstratives require from their nPs.28 However, a similar semantic restriction is apparently not in place for the indefinite enclitic =mõ. Nevertheless, it is simply the case that a clausal nominalization with the indefinite is not possible (see Example (89)).29
9 Conclusion
In this paper, I have set out to investigate the structure of clausal nominalization through Ka(g)wahíva, focusing in particular on nominalized complement clauses. A common assumption in works about clausal nominalization is that they consist of a nominal layer on top of the clausal content, which exhibits verbal properties (the Mixed Extended Projection / Nominalization Functional Thesis) (Borsley & Kornfilt 2000; Kornfilt & Whitman 2011). A question that emerges in this literature is what exactly the nominal layer in clausal nominalizations looks like. Another related question concerns the amount of clausal content a given nominalizer may embed. In exploring Ka(g)wahíva nominalized complement clauses in detail, I have argued that their nominal layer simply consists of nP, with no further nominal projections (e.g., a DP). Furthermore, complement clauses are TPs nominalized by nP. The first result stands in contrast to proposals for other languages, which exhibit clausal nominalization by D, while the second result indicates that TP nominalization by n is possible, a pattern not predicted in the recent literature that seeks to relate nominalizer type to the size of the embedded structure. Each result has only very recently been achieved for Japanese and Khalkha (Mongolic) (nominalizations by nP) (Nasu et al. 2024; Lim 2025), and Oshiwambo (Bantu) (TP-embedded nominalizations by nP) (Lee & Ndapo 2025).
Abbreviations
a = subject agreement, acc = accusative, ais = adjunct-initial suffix, ass = assertive, aten = attenuative, att = attested, b = object agreement, circumst = circumstantial, compl = completive aspect, const.neg = constituent negation, cop = copula, cor = coreferential marking, dem:med:round = demonstrative for referents at a medium distance from the speaker and having a round shape, dem:dist:round = demonstrative for referents at a distance from the speaker and having a round shape, depend = dependency marker, dir.evid = direct evidential marker, distant.pst = distant past, f = feminine grammatical gender, foc = focus, frustr = frustrative, fs = final sentence particle, gen = genitive, hab = habitual aspect, indf.det = indefinite determiner, irr = irrealis mood, loc = locative, many.ex = exact quantifier, obj = moved (or dropped) object complement, m = masculine grammatical gender, ms = male speaker, n = nominal, nf.nmlz = non-factive nominalization, n.int = non-interrogative, nmlz = nominalizer, nom = nominative, nom.neg = nominal negation, nom.pst = nominal past, n.pst = non-past, n.wit.pst = non-witnessed past, obj = object, opt = optative, pfv = perfective, pot = potential, progr = progressive, prosp = prospective, q = interrogative, realis = realis mood, red = reduplicant, ref = referrer, sbjnct = subjunctive, vis = visual, voc = vocative
Ethics and consent
Uncited data was obtained through field research that has been approved by the Institutional Review Board at the University of California-Berkeley (number 2019-04-12071) and the Museu dos Povos Indígenas/FUNAI.
Funding information
Field research that led to the uncited data reported in the paper has been made possible due to the following funding opportunities: several iterations of the Oswalt Grant for Field Research (UC Berkeley), a grant from the Programa de Documentação de Línguas Indígenas (Prodoclin) (Museu dos Povos Indígenas/FUNAI), a Language Legacies Grant (Endangered Language Fund), and a Glottobank small grant (Endangered Language Documentation Programme).
Acknowledgements / Agradecimentos
Minha incomensurável gratidão aos povos Ka(g)wahiva cujas línguas estão representadas aqui, em especial àqueles que contribuíram diretamente para este trabalho. Em especial, agradeço a minha mestra de Júma, Mandeí Júma, e aos meus mestre de Jupaú (Uru Eu Wau Wau), Awip (Davi) e Puré Uru Eu Wau Wau, que participaram das sessões de elicitação. Também agradeço a Mandá Uru Eu Wau Wau, da aldeia Jaikara/621, que se disponibilizou a narrar estórias e relatos pessoais cujos excertos também fazem parte do conjunto de dados Jupaú apresentado neste trabalho.
I also thank Peter Jenks for feedback on this work. Additionally, I would like to thank the Editor David Erschler and the anonymous reviewers for their instigating comments that made this paper better. Errors are my own.
Competing interests
The author has no competing interests to declare.
Notes
- The particular methods employed are as follows: i) Recording, transcription, and translation of Ka(g)wahíva spontaneous speech; ii) Translation from Portuguese to Ka(g)wahíva; iii) Well-formedness judgments of Ka(g)wahíva utterances; iv) Comparison of the well-formedness of two or more Ka(g)wahíva utterances; v) Acceptability judgments of Ka(g)wahíva utterances in a constructed context; and vi) Description of visual stimuli in Ka(g)wahíva. The recordings made using the method in (i) form a corpus of high-quality, carefully annotated audiovisual recordings of Ka(g)wahíva verbal art and conversational texts, including personal histories, traditional oral narratives, procedural descriptions, songs, and staged conversations. The annotation tasks in (i) followed a workflow that involves text segmentation, text respeaking (adapted from Basic Oral Language Documentation (BOLD) transcription method outlined in Reiman (2010), and text annotation. The elicitation interview tasks described in (iii–vi) were primarily conducted with bilingual speakers in Ka(g)wahíva and Portuguese. For Júma, interviews were conducted with Mandeí Júma. For Jupaú, interviews were conducted with the Jupaú speakers Awip (Davi) and Puré. All interviews were conducted with one speaker at a time. Translation tasks (ii) were used to obtain lexical, syntactic, and semantic information about Ka(g)wahíva. The tasks in (iii–iv) aimed to capture aspects of Ka(g)wahíva syntax. To indicate combinations that are not possible in the language, I marked sentences that were judged ill-formed in any context with an asterisk (*). The tasks in (v–vi) aimed to capture aspects of Ka(g)wahíva meaning that place constraints on the acceptability of well-formed utterances in certain situations. Utterances that are well-formed but not felicitous in a particular context are presented with a pound sign (#). During elicitation interviews, utterance contexts were constructed and supplied in Portuguese or supplied visually using picture stimuli, followed by a Ka(g)wahíva utterance intended to be spoken in that context. [^]
- The collection is available through the following link: 10.7297/X2P26W9H. [^]
- The full list of abbreviations is provided at the end of the paper (see Section Abbreviations). Furthermore, Ka(g)wahíva examples are presented using the practical Ka(g)wahíva writing system, which is similar to the IPA with the following exceptions: <y> = [ɨ], <’> = [ʔ], <kw> [kw], <tx> =[tʃ], <ngw> = [ŋgw], <gã> = [ŋã]. Additionally, a nasal(ized) vowel is indicated with a tilde. [^]
- A parallel pattern, where a single initial element occurs in the clause-initial position and is followed by two sets of particles with a strict ordering (CP and TP particles), is found in Tuparí matrix and embedded clauses (Singerman 2020). Unlike Tuparí, however, Ka(g)wahíva lacks the mood and tense particles in embedded clauses, as shown in this paper for complement clauses (see Section 5.1.2). [^]
- Other adverbial suffixes include -katu ‘well’, -uhu ‘a lot’, -ete ‘really’, -’i ‘a bit‘. However, these elements, but not the ones from above, are also shared with simple nominals; with simple nouns, they mean -katu ‘good’, -uhu ‘big’, -ete ‘genuine’, -‘i ‘small’. Therefore, I consider these adverbial suffixes unreliable as diagnostics of verbal modification. [^]
- A reviewer expressed curiosity about subject agreement distribution in non-verb-initial clauses with DP extraction, as in examples (3b), (3c), and (3d). In these, we see that verbs are not initial, yet they appear with the agreement associated with the CP. Since the details of how subject agreement in these instances arises are beyond the scope of this paper, I will only refer the interested reader to dos Santos (2025a), which deals with the distribution of agreement across other clause types, including those in the non-verb-initial sentences. [^]
- A reviewer wonders if (26) characterizes a violation of the Complex NP Constraint, a parallel of which is the English *Who did Mary tell [the rumor [that _ was stealing clothes]]? It does not, as Ka(g)wahíva structures like (26) cannot be assimilated into the profile of NP complement constructions. Unlike the English example, the Ka(g)wahíva one does not involve an overt noun (claim, rumor, fact), nor are the nominalized complements CPs, as the present section argues. [^]
- Other diagnostics to probe the presence of a TP are unavailable in Ka(g)wahíva. For instance, Lim (2025; 2026), probing the existence of TP in complement clauses in Khalkha, relies on scope evidence relating the interpretation of universally quantified subjects to negation, which he assumes is under TP; if universal quantifiers as subjects scope above negation, then it is because the subject has A-moved to TP. This diagnostic is not available in Ka(g)wahíva because the language shows no evidence that subjects move to TP. [^]
- A reviewer wonders why the choice for -vo, in okovo from example (34), and not =a. While confirmation that this combination with =a will have to wait, I suspect it should not be possible as habitual semantics is encoded via eko ‘be’ marked with -vo. [^]
- Speakers may also drop the enclitic from Portuguese proper names. [^]
- I briefly return to this distribution in vocatives in Section §6.5, footnote 21. [^]
- Words that express nominal quantification in other languages, such as numerals and quantifiers, pattern as adjuncts in the language. [^]
- The other languages that Cabral (2001)’s survey indicates as having a grammatical -a include (several etnovarieties of) Ka(g)wahíva, Asuriní do Tocantins, Parakanã, Suruí (Aikewara), Tembé, Tupinambá, Guaraní Antigo, Avá-Canoeiro, Asuriní do Xingu, Turiwára, Anambé, Wajampí do Amaparí, Emérillon, and Jo’é. Only for some of these languages has there been later grammatical work that provides a complete, or almost complete, characterization of -a with illustrative examples. In all of them, the cognate -a shows the same distribution as in Ka(g)wahíva, Kamaiurá, and Tapirapé. For instance, this is true for Asuriní do Xingu (Pereira 2009), Asuriní do Trocará (Vieira 1993), Avá-Canoeiro (Borges 2006), and Emérillon (Rose 2003), with some caveats for the last three: the distribution of -a cannot be shown for (complement) clauses in Asuriní do Trocará, as they only involve direct quotation; no data is presented for embedded clauses in Avá-Canoeiro. As for Emérillon, notional complement clauses do not show -a. This absence of -a can be explained in Emérillon as it has lost the nominalization strategy for dependent clauses; for instance, in Emérillon, but not in sister languages with that strategy, these structures can exhibit subject agreement, and there is no distinction in negation marking between matrix and dependent clauses. Of the remaining languages in the list, most lack a comparable dataset for -a. These are: Guaraní Antigo, Jo’é, Parakanã, Tembé, Tupinambá, and Turiwára. Finally, for Anambé, the only grammatical work on the language explicitly states that it has no -a (Julião 2005). This scenario also extends to one variety of the closely related language Wayampí: one variety does not have -a (Copin 2012), while the other variety apparently does, at least given its segmentation in simple nouns in Copin (2026); however, no grammatical work exists on the function of this morpheme in this variety. The presence of the formative -a is also observed in the only accessible work for Suruí (Aikewara), Lopes (2014), as it contains examples with an -a segmented off simple nouns; however, discussion about its distribution is lacking in that source. [^]
- Some data in Seki (2000) may, at first sight, contradict the claim that =a occurs with possessed nouns in this language. Thus, several possessive phrases are shown where the possessed noun does not appear with =a:
- (i)
- a.
- Kunu’um-a
- boy-n
- rup
- father
- ‘Boy’s father’ (Seki 2000: 56, ex. 2b)
One also notes from the source that, in the instances where possessed nouns are presented without -a and translated with adnominal possession, they are presented in isolation, as opposed to being within a sentence. I suspect the possessive relationship encoded by these structures is not the one reflected in the translation of these examples, namely, adnominal possession. That this seems to be the case is supported by instances in which the same structure is translated with predicative possession when presented within a sentence in the source (iia), but receives a translation of adnominal possession when shown in isolation (iib). By contrast, when used within a sentence, the same form pyt always appears with =a and is translated with adnominal possession (iic).- b.
- Kunu’um-a
- boy-n
- akaŋ
- head
- ‘Boy’s head’ (Seki 2000: 56, ex. 3b)
- (ii)
- a.
- Po
- q
- ne=pyt
- 2sg=house
- ‘Do you have a house?’ (Seki 2000: 160, ex. 498)
- b.
- Je=pyt
- 1sg=house
- ‘My house’ (Seki 2000: 57, ex. 5b)
[^]- c.
- Je=pyr-a
- 1sg=house-n
- amoete.
- far
- ‘My house is far.’ (Seki 2000: 163, ex. 515)
- In simple nominals, negation has the idiosyncratic privative meaning. In the present account, this can similarly be made to follow from negation being closer to the root in simple nominals but not in deverbal nominals, thus allowing for allosemy of negation in this context. [^]
- According to Almeida et al. (1983: 31–32), “Nominalizing the verbal root through the nominalization process of nominal roots, that is, adding the suffix -a to roots ending in a consonant, the result is ‘an action noun’, which can be used with verbs expressing perception”. [^]
- As Vieira (1993: 25) writes: “It is through the suffix -a that nouns are derived in Asurini”, as well as, “[-a] is a syntactic operator whose function is to turn a stem into a noun” (Vieira 1995: 704). [^]
- With negation, bare nouns can scope under negation, as in (i), a pattern that is commonly found for bare nouns crosslinguistically (Deal & Nee 2018). Additionally, dos Santos (2024b: 69–70) had previously shown that Ka(g)wahíva bare nouns also outscope negation. Since this behavior is exceptionally found only with negation, I set it aside here.
[^]
- (i)
- Ndoropyhygite
- nd-oro-pyhyg-i-te
- neg-1.excl.a-catch-neg-really
- ore
- ore
- 1.excl
- pira,
- pira=a
- fish=nmlz
- ’i
- ’i
- say
- ki
- ki
- pst
- gã
- gã
- 3.pl
- ko.
- ko
- realis
- ✓ Context 1: ‘They caught no fish.’
- ✗ Context 2: ‘There is a particular fish that they did not catch.’
- Additionally, the verb is obligatorily marked with a prefix i-. In Ka(g)wahíva, the pattern of i-prefixation is triggered just in case the complement of a head is moved or dropped; thus, it is found with verbs, postpositions, and inalienable possessed nouns. Despite the initial impression that examples like (64) may give, this morpheme is not a resumptive pronoun, if we consider that resumptive pronouns draw from the inventory of a language’s pronouns (McCloskey 2017: 4), while i- is not part of the pronominal inventory of Ka(g)wahíva. Furthermore, the pattern of i-prefixation is widespread across the Tupí-Guaraní family. The literature on these languages regards i- as a marking of syntactic adjacency between a complement and its selecting head (Rodrigues & Cabral 2012). [^]
- Note that the word for blood is pronounced as vy in (67a), in contrast to vyhy in the earlier example in (58). The surface form vy is the result of two regular phonological processes in Júma: /h/ deletion between identical vowels in an unstressed syllable (here, the syllable /hy/ becomes unstressed due to stress shift driven by the negative suffix -i), followed by vowel hiatus resolution, after the former process creates a vowel sequence. [^]
- Furthermore, nouns in combination with any of the definite enclitics always exhibit =a. Relatedly, as examples from Section §6.1 already showed, neither =a nor the definite enclitics appear in the vocative. A reviewer suggests this distribution of proper names in the vocative may lend substance to the idea that =a is an overt D-related element. The idea draws on a potential parallel to Greek vocatives, in which vocatives do not exhibit a determiner (Stavrou 2014). As a result, Stavrou treats vocatives as structurally smaller than a DP. In turn, the reviewer suggests a paralell between Greek and Ka(g)wahíva vocatives: the latter’s vocatives are smaller than a DP as well, and since =a is absent in the vocative, the absent material is a D-head. I believe this parallel would be less than a coincidence had Ka(g)wahíva vocatives also lack overt material that is otherwise required. In particular, as I already pointed in this note, not only =a is absent in the vocative, but the definite enclitics discussed in the present section cannot appear in the vocative either. Following closely the suggestion above, pairing the assumption that Ka(g)wahíva vocatives are structurally smaller to the absence of =a does not ultimately require that =a be a D-head. Indeed, in the view that =a is a categorizing n in simple nouns and the definite enclitics are determiners, as I argue, the structural parallel with Greek vocatives could still be partially maintained. Specifically, Ka(g)wahíva vocatives are structurally smaller, lacking the enclitics, but that does not automatically mean that =a is the D-head. Furthermore, the idea defended here that =a is a n-head and the definite enclitics realize a D-head is harmonic with the semantic behavior of nominals, as described by the several pieces of evidence put together in Section §6.5 and the present section. In contrast, if =a were a D-head, the striking nP-behavior of bare nouns would be mystifying. [^]
- Other diagnostics for definiteness are not available in Ka(g)wahíva, given the restriction that these determiners combine with nouns whose referents are living humans. Therefore, it is not possible to use them to describe parts of previously-introduced wholes (e.g., I bought a bicycle for my son, but the tires were worn out). [^]
- I thank a reviewer for drawing my attention to this prosodic distinction between the pronominal forms. Incidentally, a variation in the pronunciation of the first person singular ji as subjects in clauses, but not in possessor phrases, might also be suggestive that the first person singular is treated differently in these two contexts. As subjects, the first person is sometimes pronounced as [ʤjɛ], [ʤɛ], or [ʤi]. In contrast, such variation is never found in possessive phrases with the first person singular, where only [ʤi] is heard. Like procliticization, we can see this difference as stemming from the distinct syntactic structures in which the pronoun is used. [^]
- Pronominal forms notoriously surface as proclitics when occurring as the object of postpositions, the object of verbs, and inalienable possessors across Tupí-Guaraní languages (Jensen 1990; 1998; Meira & Drude 2013; Rose 2015; dos Santos 2023). [^]
- That ji in (81b) is a standalone word, rather than the possessor of fish, receives confirmation from the intervening e’yi, which is an adjunct. [^]
- A parallel pattern seems to exist in Oshiwambo, as discussed in Lee & Ndapo (2025). They argue that TPs are the topmost layer in subject nominalizations in the Bantu language Oshiwambo, which they assume is a nP-nominalization. However, they also include a representation for these nominalizations on p. 14 that contains a richer nominal structure than that proposed for Ka(g)wahíva, where a DP is projected on top of the nP projection that directly embeds the TP. They point out that this structure poses a challenge to Alexiadou (2020)’s proposal that rules out TP-nominalization by n. As far as I can tell, it does not challenge Lim (2025; 2026)’s, which predicts TP-embedding nominalizations by nominal structure that also includes a DP. [^]
- A referee is concerned about the idea that Ka(g)wahíva bare nouns are nPs and that they can also be related to an anchoring domain – much like TPs are considered as part of an anchoring domain (Ramchand & Svenonius 2014; Wiltschko 2014) – unless “bare nouns do indeed signal some reference”. For anchoring, the important property is having a time parameter, which clausal tense do. Do bare nouns in Ka(g)wahíva exhibit a time parameter? While going into the details to answer this empirical question is out of the scope of the paper, I can initially offer the observation that the presence of nominal tense is suggestive of a time parameter in these nouns. In particular, since nominal tense is understood as semantically similar to clausal tense in languages like the sister language Mbyá (Thomas 2014), and Ka(g)wahíva bare nouns exhibit nominal tense cognates with Mbyá (dos Santos 2024b: 57–58), then Ka(g)wahíva nominal tense is the best indication for a time parameter in these nouns. [^]
- Interestingly, the semantic restriction imposed by the determiner onto its restrictor, as observed in Ka(g)wahíva, is not observed with the D elements that nominalize clauses in languages that realize the D-assumption of clausal nominalization, where the determiner simply turns predicates into individuals, without asking whether the NP referent is animate or has a particular shape. [^]
- If it turns out that this combination is also ruled out on semantics grounds, then a defender of the D-assumption of clausal nominalization could say that the Ka(g)wahíva facts do not actually affect the soundness of their theory because the presence of all D elements is ruled out on semantic grounds, not syntactic ones. I thank a reviewer for raising this point. [^]
References
Alexiadou, Artemis. 2020. D vs. n nominalizations within and across languages. In Alexiadou, Artemis & Borer, Hagit (eds.), Nominalization: 50 years on from Chomsky’s Remarks, 87–110. Oxford University Press. DOI: http://doi.org/10.1093/oso/9780198865544.003.0005
Almeida, Antônio & de Jesus, Irmãzinhas & Paula, Luiz Gouvêa de. 1983. A língua Tapirapé [The Tapirapé language]. Rio de Janeiro: Biblioteca Reprográfica Xerox. http://biblio.etnolinguistica.org/almeida_1983_tapirape.
Bochnak, M. Ryan & Hanink, Emily A. 2022. Clausal embedding in Washo: Complementation vs. modification. Natural Language & Linguistic Theory 40. 979–1022. DOI: http://doi.org/10.1007/s11049-021-09532-z
Borges, Mônica Veloso. 2006. Aspectos Fonológicos e Morfossintáticos da Língua Avá-Canoeiro (Tupi-Guarani) [Phonological and morphosyntactic aspects of Avá-Canoeiro (Tupi-Guarani)]. Campinas: Universidade Estadual de Campinas dissertation.
Borsley, Robert D. & Kornfilt, Jaklin. 2000. Mixed extended projections. In Borsley, Robert D. (ed.), The nature and function of cyntactic categories (Syntax and Semantics 32), 101–131. Academic Press. DOI: http://doi.org/10.1163/9781849500098$_$006
Cabral, Ana Suelly Arruda Câmara. 2001. Observações sobre a história do morfema -a da família Tupí-Guaraní [Observations about the history of the morpheme -a in the Tupí-Guaraní family]. In Queixalós, Francesc (ed.), Des noms et des verbes en tupi-guarani: état de la question, 133–162. Berlin: Lincoln Europa.
Camargos, Quesler Fagundes & Castro, Ricardo Campos. 2013. Paralelismo entre DP e CP a partir de nominalizações na língua Tenetehára [Parallelisms between DP and CP in the nominalizations of the Tenetehára language]. Revista da Anpoll 1(34). 285–303. DOI: http://doi.org/10.18309/anp.v1i34.675
Cardoso, Valéria Faria. 2008. Aspectos morfossintáticos da língua Kaiowá (Guarani) [Morphosyntactic aspects of the Kaiowá language]. Campinas (São Paulo): Universidade Estadual de Campinas dissertation.
Chierchia, Gennaro. 1998. Reference to kinds across languages. Natural Language Semantics 6. 339–405. DOI: http://doi.org/10.1023/A:1008324218506
Chierchia, Gennaro. 2009. What will you have, DP or NP? Proceedings of NELS 37, 101–114.
Chomsky, Noam. 2001. Derivation by Phase. In Kenstowicz, Michael (ed.), Ken Hale. A life in language, 1–52. Cambridge: MIT Press. DOI: http://doi.org/10.7551/mitpress/4056.003.0004
Comrie, Bernard & Estrada-Fernández, Zarina. 2012. Relative clauses in languages of the Americas: A typological overview. Amsterdam/Philadelphia: John Benjamins. DOI: http://doi.org/10.1075/tsl.102
Copin, François. 2012. Grammaire Wayampi [Wayampi grammar]. Paris: Département de Sciences du Langage: Université Paris 7 – Denis Diderot dissertation. DOI: doi: https://hal.science/tel-01547977
Copin, François. 2026. The Wayampi verb: Structure, form, and function. Cadernos de Etnolinguística 13(1). 1–64. DOI: doi: https://hal.science/hal-05664241
Cruz, Aline & Praça, Walkíria. 2019. Innovation in nominalization in Tupí-Guaraní languages: a comparative analysis of Tupinambá, Apyãwa and Nheengatú. In Zariquiey, Roberto & Shibatani, Masayoshi & Fleck, David W. (eds.), Nominalization in languages of the Americas, 627–657. Amsterdam: John Benjamins Publishing Company. DOI: http://doi.org/10.1075/tsl.124.18cru
Davies, William D. & Dubinsky, Stanley. 1998. Sentential subjects as complex NPs: New reasons for an old account of subjacency. In Proceedings of the Chicago Linguistic Society (CLS), 83–94. Chicago Linguistic Society.
Dawson, Virginia & Jenks, Peter. 2023. Describing definites and indefinites. In Jenks, Peter & Michael, Lev (eds.), Key topics in language documentation and description, 99–174. Language Documentation and Conservation, Special Publication n.26. https://hdl.handle.net/10125/75034.
Deal, Amy Rose & Nee, Julia. 2018. Bare nouns, numbers, and definiteness in Teotitlán del Valle Zapotec. In Truswell, Robert & Cummins, Chris & Heycock, Caroline & Rabern, Brian & Rohde, Hannah (eds.), Proceedings of Sinn und Bedeutung 21, 317–334. Edinburgh, Scotland: University of Edinburgh. https://semanticsarchive.net/Archive/DRjNjViN/DealNee.pdf.
dos Santos, Wesley. 2023. Asymmetries among person indexes in Kawahíva. International Journal of American Linguistics 89(4). 427–455. DOI: http://doi.org/10.1086/726149
dos Santos, Wesley. 2024a. Kawahíva (Brazil) – Language snapshot. Language Documentation and Description 24(1). DOI: http://doi.org/10.25894/ldd.2549
dos Santos, Wesley. 2024b. Topics on the syntax of Kawahíva: A Tupí-Guaraní language from the Brazilian Amazon. Berkeley, CA: University of California, Berkeley dissertation. Available: https://escholarship.org/uc/item/1h26s7sq.
dos Santos, Wesley. 2025a. Verb-initial clauses in Kawahíva. Natural Language and Linguistic Theory 43. 563–628. DOI: http://doi.org/10.1007/s11049-024-09646-0
dos Santos, Wesley. 2025b. Person hierarchy effects from ϕ-agreement at the left periphery in Kawahíva. Proceedings of the 55 North East Linguistic Society (NELS). Available: https://ling.auf.net/lingbuzz/009238.
dos Santos, Wesley. Under review. On the nature of =a in Tupí-Guaraní: The view from Ka(g)wahíva.
Duarte, Fábio Bonfim. 2007. Estudos de Morfossintaxe Tenetehára [Morphosyntactic Studies on Tenetehára]. Belo Horizonte: Faculdade de Letras da UFMG.
Estigarribia, Bruno. 2020. A grammar of Paraguayan-Guarani. London: UCL Press.
Farudi, Annahita. 2007. An antisymmetric approach to Persian clausal complements. Manuscript.
Franks, Steven & Pereltsvaig, Asya. 2004. Functional categories in the nominal domain. Proceedings of FASL, 339–405.
Gillon, Carrie. 2011. Bare nouns in Innu-aimun: what can semantics tell us about syntax? Proceedings of WSCLA 16, 29–56. Available: https://lingpapers.sites.olt.ubc.ca/wscla-volumes/.
Grimshaw, Jane. 2000. Mixed extended projections. In Coopmans, Peter & Everaert, Martin B. H. & Grimshaw, Jane (eds.), Lexical specification and insertion, 115–134. John Benjamins Publishing Company. DOI: http://doi.org/10.1075/cilt.197.07gri
Halle, Morris & Marantz, Alec. 1993. Distributed morphology and the pieces of inflection. In Hale, Ken & Keyser, Samuel Jay (eds.), The view from building 20: Essays in linguistics in honor of Sylvain Bromberger, 111–176. Cambridge, MA: MIT Press.
Hanink, Emily. 2016. Internally headed relatives and event nominalizations in Washo. Proceedings of Berkeley Linguistics Society 42. Available: https://emhanink.pages.iu.edu/blspaper.pdf.
Hankamer, Jorge & Mikkelsen, Line. 2021. CP complements to D. Linguistic Inquiry 52(3). 473–517. DOI: http://doi.org/10.1162/ling_a_00387
Harley, Heidi. 2013. External arguments and the Mirror Principle: On the distinctness of Voice and v. Lingua 125. 34–57. DOI: http://doi.org/10.1016/j.lingua.2012.09.010
Iordăchioaia, Gianina. 2020. D and N are different nominalizers. Glossa: A Journal of General Linguistics 5(1). 53. DOI: http://doi.org/10.5334/gjgl.1111
Jensen, Cheryl. 1990. Cross-referencing changes in some Tupí-Guaraní languages. In Payne, Doris L. (ed.), Amazonian linguistics: Studies in lowland South American languages, 117–158. Austin: University of Texas Press. DOI: http://doi.org/10.7560/704145-007
Jensen, Cheryl. 1998. Comparative tupí-guaraní morphosyntax. In Derbyshire, Desmond C. & Pullum, Geoffrey K. (eds.), Handbook of Amazonian languages, vol. 4, 489–618. Berlin: Mouton.
Julião, Maria Risolêta Silva. 2005. Aspects Morphosyntaxiques de L’Anambe [Morphosyntactic aspect of Anambe]. Toulouse: L’Universite de Toulouse (Le Mirail) dissertation.
Kornfilt, Jaklin & Whitman, John. 2011. Afterword: Nominalizations in syntactic theory. Lingua 121(7). 1297–1313. DOI: http://doi.org/10.1016/j.lingua.2011.01.008
Kratzer, Angelika. 1996. Severing the external argument from its verb. In Rooryck, Johan & Zaring, Laurie (eds.), Phrase structure and the lexicon, vol. 33, 109–137. Springer. DOI: http://doi.org/10.1007/978-94-015-8617-7
Lee, Soo-Hwan & Ndapo, Olivia. 2025. TP in Oshiwambo subject nominalizations. Glossa: a journal of general linguistics 10(1). 1–19. DOI: http://doi.org/10.16995/glossa.16344
Lemos Barbosa, Antonio. 1956. Curso de Tupi Antigo [A course on Old Tupi]. Rio de Janeiro: Livraria Sao Jose.
Lim, Jun Jie. 2025. (A)symmetries in the morphosyntax of embedded clauses in Khalkha Mongolian. San Diego, CA: University of California, San Diego dissertation. Available: https://escholarship.org/uc/item/3x7738sf.
Lim, Jun Jie. 2026. Small nominalized clauses in Khalkha Mongolian: Asymmetries in mixed extended projections. Natural Language & Linguistic Theory 44(1). 8. DOI: http://doi.org/10.1007/s11049-025-09692-2
Löbner, Sebastian. 2002. Understanding semantics. London: Arnold Publishers.
Longobardi, Giuseppe. 1994. Reference and proper names: A theory of N-movement in syntax and logical form. Linguistic Inquiry 25. 609–665. https://www.jstor.org/stable/4178880.
Lopes, Jorge Domingues. 2014. Uma interface da documentação linguística e modelos lexicográficos para línguas indígenas brasileiras: uma proposta para o Suruí-Aikewára [An interface of language documentation and lexicography models for Brazilian Indigenous languages: A proposal for Suruí-Aikewára]. Brasília: Universidade de Brasília dissertation.
Magalhães, Marina Maria Silva. 2007. Sobre a morfologia e a sintaxe da língua Guajá (família Tupí-Guaraní [On the morphology and syntax of the Guajá language (Tupí-Guaraní)]. Universidade Federal de Brasília dissertation.
Marantz, Alec. 2013. Locality domains for contextual allomorphy across the interfaces. In Matushansky, Ora & Marantz, Alec P. (eds.), Distributed morphology today: Morphemes for Morris Halle, 95–115. MIT Press. DOI: http://doi.org/10.7551/mitpress/9780262019675.003.0006
McCloskey, James. 2017. Resumption. In Everaert, Martin & van Riemsdijk, Henk (eds.), The Wiley Blackwell companion to syntax, Second Edition, 1–30. Hoboken, NJ: John Wiley & Sons, Ltd. DOI: http://doi.org/10.1002/9781118358733.wbsyncom105
Meira, Sérgio & Drude, Sebastian. 2013. Sobre a origem histórica dos “prefixos relacionais” das línguas tupí-guaraní [On the historical origin of the “relational prefixes” in Tupí-Guaraní languages]. Cadernos de Etnolinguística 5(3). 1–31. Available at http://www.etnolinguistica.org/issue:vol5n1.
Miyagawa, Shigeru. 2011. Genitive subjects in Altaic and specification of phase. Lingua 121(7). 1265–1282. DOI: http://doi.org/10.1016/j.lingua.2011.01.009
Moskal, Beata A. 2015. Domains on the border: between morphology and phonology. Storrs: University of Connecticut dissertation.
Nasu, Norio & Takayuki, Akimoto & Koji, Shimamura & Yusuke, Yoda. 2024. Clausal nominalization and embedded questions in Japanese. Glossa: a journal of general linguistics 9(1). 1–38. DOI: http://doi.org/10.16995/glossa.10051
Nogueira, Antônia Fernanda de Souza. 2019. Predicação na língua Wayoro (Tupi): propriedades de finitude [Predication in the Wayoro language (Tupian): finiteness properties]. São Paulo, São Paulo: Universidade Estadual de São Paulo dissertation. DOI: http://doi.org/10.11606/T.8.2019.tde-14082019-101006
Pereira, Antônia Alves. 2009. Estudo morfossintático do Asurini do Xingu [Morphosyntax study of Asurini do Xingu]. Campinas: Universidade Estadual de Campinas dissertation.
Pereltsvaig, Asya. 2006. Small nominals. Natural Language & Linguistic Theory 24(2). 433–500. DOI: http://doi.org/10.1007/s11049-005-3820-z
Pietraszko, Asia. 2019. Obligatory CP nominalization in Ndebele. Syntax 22(1). 66–111. DOI: http://doi.org/10.1111/synt.12167
Pires, Acrisio & Milsark, Gary. 2017. Gerundive nominalizations. In Everaert, Martin & van Riemsdijk, Henk (eds.), The Wiley Blackwell companion to syntax, Second Edition, 1–25. Hoboken, NJ: John Wiley & Sons, Ltd. DOI: http://doi.org/10.1002/9781118358733.wbsyncom121
Praça, Walkíria Neiva. 2007. Morfossintaxe da língua Tapirapé [The Morphosyntax of Tapirapé]. Brasília: Universidade de Brasília (UnB) dissertation.
Pylkkänen, Liina. 2002. Introducing arguments. Cambridge, MA: MIT Press. DOI: http://doi.org/10.7551/mitpress/9780262162548.001.0001
Queixalós, Francesc. 2006. The primacy and fate of predicativity in Tupi-Guarani. In Vapnarsky, Valentina & Louis, Ximena (eds.), Lexical categories and root classes in Ameridian languages, 249–288. Bern: Peter Lang.
Ramchand, Gillian & Svenonius, Peter. 2014. Deriving the functional hierarchy. Language Sciences 46. 152–174. DOI: http://doi.org/10.1016/j.langsci.2014.06.013
Reiman, D. Will. 2010. Basic oral language documentation. Language Documentation and Conservation 4. 254–268. Available at http://hdl.handle.net/10125/4479.
Rocha da Silva, Ivan. 2016. Não-finitude em Karitiana: subordinação versus nominalização [Non-finiteness in Karitiana: Subordination versus nominalization]. São Paulo, São Paulo: Universidade Estadual de São Paulo dissertation. DOI: http://doi.org/10.11606/T.8.2016.tde-25102016-122036
Rodrigues, Aryon D. & Cabral, Ana Suelly A. C. 2012. Tupían. In Campbell, Lyle & Grondona, Verónica (eds.), The indigenous languages of South America: A comprehensive guide, vol. 2, 495–574. Berlin/Boston: De Gruyter Mouton.
Rodrigues, Aryon Dall’Igna. 1953. Morfologia do verbo Tupi [the morphology of the Tupian verb]. Letras 1. 121–152. DOI: http://doi.org/10.5380/rel.v1i0.20083
Rodrigues, Aryon Dall’Igna. 1996. Argumento e predicado em Tupinambá. Boletim da ABRALIN 19. 57–66. http://www.etnolinguistica.org/biblio:rodrigues-1996-argumento.
Rodrigues, Aryon Dall’Igna. 2001. Sobre a natureza do caso argumentativo [On the nature of the argumentative case]. In Queixalós, Francesc (ed.), Des noms et des verbes en tupi-guarani, 103–114. Lincom Europa.
Rose, Françoise. 2003. Morphosyntaxe de l’emerillon: Langue tupi-guarani de guyane française [The morphosyntax of Emerillon: A Tupi-Guarani language from French Guiana]. Lyon: Sciences du Langage et Arts: Université Lumière Lyon 2, Facult’e des Lettres dissertation.
Rose, Françoise. 2015. When “you” and “i” mess around with the hierarchy: a comparative study of Tupi- Guarani hierarchical indexing systems. Boletim do Museu Paraense Emílio Goeldi (Ciências Humanas) 10(2). 347–369. DOI: http://doi.org/10.1590/1981-81222015000200008
Rose, Françoise. 2023. Word classes in Maweti–Guarani languages. In van Lier, E. (ed.), The Oxford Handbook of word classes. Oxford Handbooks. DOI: http://doi.org/10.1093/oxfordhb/9780198852889.013.27
Roussou, Anna. 1991. Nominalized clauses in the syntax of Modern Greek. In van de Koot, Hans (ed.), UCL working papers in linguistics vol. 3, 77–100. London: University College London, Department of Phonetics and Linguistics.
Sande, Hannah & Jenks, Peter & Inkelas, Sharon. 2020. Cophonologies by Ph(r)ase. Natural Language Linguistic Theory 38. 1211–1261. DOI: http://doi.org/10.1007/s11049-020-09467-x
Seki, Lucy. 2000. Gramática do Kamaiurá: Língua Tupi-Guarani do Alto Xingu [Kamaiurá grammar: A Tupí-Guaraní language from Alto Xingu]. Campinas, São Paulo: Editora da Unicamp.
Selkirk, Elisabeth. 2011. The syntax-phonology interface. In Goldsmith, John & Riggle, Jason & Yu, Alan C. L. (eds.), The Handbook of phonological theory, 435–484. Blackwell Publishing Ltd. DOI: http://doi.org/10.1002/9781444343069.ch14
Singerman, Adam. 2021. On finite embedded clauses in Tuparí: their synchrony, diachrony, and typology. Glossa: a journal of general linguistics 6(1). 1–46. DOI: http://doi.org/10.5334/gjgl.1394
Singerman, Adam Roth. 2018. Negation as an exclusively nominal category. Language 94(2). 432–467. DOI: http://doi.org/10.1353/lan.2018.0022
Singerman, Adam Roth. 2020. The clausal organization of Tuparí, an indigenous Brazilian language. Acta Linguistica Academica 67(4). 429–479. DOI: http://doi.org/10.1556/2062.2020.00025
Stavrou, Melita. 2014. About the vocative. In Schürcks, Lilia & Giannakidou, Anastasia & Etxeberria, Urtzi (eds.), The nominal structure in Slavic and beyond, 299–342. Cambridge: De Gruyter Mouton. DOI: http://doi.org/10.1515/9781614512790.299
Takahashi, Shoichi. 2010. The hidden side of clausal complements. Natural Language & Linguistic Theory 28(2). 343–380. DOI: http://doi.org/10.1007/s11049-010-9091-3
Thomas, Guillaume. 2014. Nominal tense and temporal implicatures: Evidence from Mbyá. Natural Language Semantics 22. 357–412. DOI: http://doi.org/10.1007/s11050-014-9108-2
Thomas, Guillaume & Duarte, Germino & Ismael, Akil. 2025. Case-dependent agreement in an active-stative language. Languages 10(9). DOI: http://doi.org/10.3390/languages10090221
Tonhauser, Judith. 2006. The temporal semantics of noun phrases: Evidence from Guaraní. Standford, CA: Standford University dissertation.
Tonhauser, Judith. 2007. Nominal tense? The meaning of Guaraní nominal temporal markers. Language 83(4). 831–869. DOI: http://doi.org/10.1353/lan.2008.0037
Vieira, Marcia Damaso. 1995. The expression of quantificational notions in Asurini do Trocará: Evidence against the universality of determiner quantification. In Bach, Emmon & Jelinek, Eloise & Kratzer, Angelika & Partee, Barbara H. (eds.), Quantification in Natural Languages, 701–720. Kluwer Academic Publishers. DOI: http://doi.org/10.1007/978-94-017-2817-1
Vieira, Marcia Maria Damaso. 1993. O fenômeno da não-configuracionalidade na língua Asuriní do Trocará: um problema derivado da projeção dos argumentos verbais [The phenomenon of non-configurationality in the Asuriní do Trocará language: A problem from the verbal argument projection]. Campinas (São Paulo): Universidade Estadual de Campinas dissertation.
Vivanco, Karin Camolese. 2021. Nominalization and adverbial clauses in Karitiana. Amerindia 45. 37–69. Available: https://amerindia.cnrs.fr/nominalization-and-adverbial-clauses-in-karitiana/.
Wiltschko, Martina. 2014. The universal structure of categories: Towards a formal typology. Cambridge: Cambridge University Press. DOI: http://doi.org/10.1017/CBO9781139833899
Zariquiey, Roberto & Shibatani, Masayoshi & Fleck, David W. 2019. Nominalization in the languages of the Americas. Amsterdam/Philadelphia: John Benjamins. DOI: http://doi.org/10.1075/tsl.124
