Factive Entailments and Clausal Complementation in Karitiana

Abstract

Attitude verbs describe mental states of their subject in relation to a proposition codified by their clausal complement. In this paper, we describe the behavior of such verbs in Karitiana, a Tupian language spoken in Brazil. Recently, it has been noted in the literature that embedded clauses with nominal features seem to lead to a factive reading of these verbs. In Karitiana, all embedded clauses are nominalized since they lack many clausal features, and conversely exhibit nominal distribution and behavior. We propose a structure with an N head for embedded clauses in Karitiana, which accounts for their morphological behavior and also explains why these constructions behave as strong islands. Judgments of contradiction with attitude verbs of Karitiana were collected and the results show that they fall into two categories: one that obligatorily entails the truth of the embedded clause and another that favors it, but that does not require it obligatorily. This shows that nominalization of embedded clauses seems to be tied to the presence of factive entailments, but this alone is not a sufficient condition to force a factive reading of the matrix attitude verb.

1. Introduction

This paper explores the interpretation of attitude verbs such as ‘believe’, ‘know’, ‘think’, ‘forget’, etc. These verbs, which select for a complement clause, denote a mental stance of the subject in relation to the proposition expressed by the embedded clause. In (1), for example, the verb ‘believes’ expresses the belief relation between the subject ‘John’ and the proposition [the Earth is flat] in the subordinate clause:

(1)	John believes that the Earth is flat.

Attitude verbs can be broken down into two categories depending on the status of the proposition denoted by the embedded clause. These verb classes are illustrated below with ‘forget’ and ‘think’. In (2), the proposition that Francis died can be false (as stated in the second clause), but the whole sentence will still be true. This is because verbs like ‘think’ denote the internal beliefs of the subject, which may be wrong. On the other hand, verbs like ‘forget’ in (3) require that the event denoted by the embedded clause be true. If one tries to negate it (by saying ‘but Francis didn’t die’ as a follow-up clause), the whole sentence becomes contradictory.

(2)	John thought [that Francis died], but Francis didn’t die.

(3)	John forgot [that Francis died], # but Francis didn’t die.

Verbs like ‘forget’, which impose this truth requirement on the complement clause, are called ‘factive’. At first sight, factivity seems to come from the matrix attitude verb, as the structure of the embedded clause in English seems to be the same in both cases. Nonetheless, cross-linguistic evidence has been brought to attention in the past years to show that the embedded clause could have some sort of impact on the interpretation of the matrix verb. The leading question of this paper is then the following: to what extent could the structure of the selected clausal complement constrain the interpretation of their matrix attitude verb? We address this question with novel data from Karitiana, an indigenous language of the Tupian family spoken in the Brazilian western Amazon. A noteworthy feature of Karitiana is that all embedded clauses exhibit a very similar structure, which is arguably nominalized. Thus, we check whether this feature affects the factivity of embedded clauses as observed cross-linguistically.

This paper is divided as follows. In Section 2, the behavior of attitude verbs is discussed with special attention to the impact that the nature of their clausal complements have on their interpretation. The syntax of embedding in Karitiana is the topic of Section 3. As will be seen in this section, subordinate clauses in Karitiana exhibit many nominal features, which lead us to analyze them as clausal nominalizations. In Section 4.1, the previous literature on factivity is briefly presented, and we discuss that the factive inference under discussion is an entailment. Section 4.2 presents the materials and the elicitation method used to collect judgments of contradiction with attitude verbs. The results of the elicitation are reported in Section 4.3, where the two patterns of behavior found in the elicitation process will be presented and exemplified. In Section 4.4, we discuss how these findings can be related to the structure of embedded clauses in Karitiana. Finally, Section 5 summarizes the main findings of the paper and points to topics for future research.

2. The Problem

As mentioned in the introduction, attitude verbs describe mental states of their subject regarding a proposition expressed by the embedded clause. Nonetheless, it was observed by Kiparsky and Kiparsky (1971) (and see also Catell 1978) that these verbs can also be broken down into two different classes, whose behavior is illustrated below:

(4)	John knows/regrets/remembers [that Peter told the secret].

(5)	John thinks/believes [that Peter told the secret].

Verbs like ‘know’/‘regret’/‘remember’ select an embedded clause depicting a true proposition. They are called factive, because the event described has to be a consolidated fact in the discursive background. On the other hand, a non-factive verb such as ‘think’/‘believe’ does not have the same requirement. In (5), the whole sentence can be true even if Peter did not tell any secret—in this case, only John’s assumption about the world is wrong.1

Since attitude verbs can be listed with respect to their meaning (see Catell 1978 for a list of verbal classes), it is reasonable to think of factivity as lexically encoded in the verb. Nonetheless, it has been noted in the literature that the interpretation of certain verbs may be influenced by the type of embedded clause they select. The following paradigm, taken from Moulton (2009), illustrates this observation with the perception verb ‘see’. One can see that the type of embedded clause in column A impacts the interpretation of the matrix verb (column B): Depending on the clausal type, the same verb can be interpreted as one act of (a/b) seeing, (c) imagining, (d) believing, or (e) acknowledging an established fact:

(6)	English (Moulton 2009)
	A. Sentence	B. Clausal type	C. Interpretation
a.	John saw [Fred leave early].	bare infinitive	direct perception
b.	John saw [Fred leaving early].	gerundive	direct perception
c.	John saw [Fred owning a house].	gerundive	imaginative
d.	John saw [Fred to be a party-pooper].	infinitive	belief
e.	John saw [that Fred left early].	finite clause	factive

Therefore, the structure of the clausal complement seems to have a significant contribution to the meaning of the matrix verb. This was also observed for factive verbs by Kiparsky and Kiparsky (1971): gerunds, a nominal form, are possible with factive verbs such as ‘regret’ in (7) but impossible with non-factive verbs (see (8)):

(7)	English (Kiparsky and Kiparsky 1971)
	I regret [having agreed to the proposal].

(8)	English (Kiparsky and Kiparsky 1971)
	∗I believe [having agreed to the proposal].

In this paper, we examine the following cross-linguistic observation, which can be roughly expressed in general terms as follows: the structure of clausal complements of factive verbs often shows nominal traits across languages (Bochnak and Hanink 2022; Bogal-Allbritten and Moulton 2018; Moulton 2009; Özyildiz 2017; Zubizarreta 1982). This is the case of English gerunds in (7) and (8) and in many other languages as well. In Washo, for example, factive verbs such as ‘forget’ in (9) select a nominalized clause with the D head -gi/-ge. On the other hand, non-factive verbs such as ‘think’ in (10) select a bare non-nominalized clause:

(9)	Washo (Bochnak and Hanink 2022)
	[∅-háʔaš-i-š]-ge	di-hámu-p’áy-i
	3-rain-int.pst-ind-ds-nm.acc	1/3-forget-ind
	“I forgot that it rained.”		(factive)

(10)	Washo (Bochnak and Hanink 2022)
	Béverli	[démlu	di-begúweʔ-é:s-aʔ]	∅-hámu-yi
	Beverly	food	1-buy-neg-dep	3-think-ind
	“Beverly thinks I didn’t buy the food.”				(non-factive)

It has also been noted that in some languages, the very same verb can be interpreted either as factive or non-factive depending on the form of the clausal complement: with clauses showing nominal traits, the interpretation is factive; with non-nominal clauses, it is non-factive. This can be seen in the examples below from Turkish (Özyildiz 2017). The presence of a nominalizer in (11a) leads to a factive interpretation of the matrix verb bil- (inflected as biliyor). Factivity can be detected by the possibility of inferring ’Hillary won’ from the sentence and also by the impossibility of uttering (11b) soon after, which would yield a contradiction. In the absence of a nominalizer as in (12a), the same bil- verb is interpreted non-factively and the same type of inference does not hold:

(11)	Turkish (Özyildiz 2017, adapted)
	a.	Tunç	[Hillary’nin	kazan-dığın-ı]	biliyor
		Tunç	Hillary	win-nmz-acc	knows
		“Tunç knows that Hillary won.” → Hillary won.							(factive)
	b.	#Ama	kazan-ma-dı.
		but	win-neg-pst.3s
		“#But she didn’t win.”

(12)	Turkish (Özyildiz 2017, adapted)
	a.	Tunç	[Hillary	kazan-dı	diye]	biliyor.
		Tunç	Hillary	win-pst	diye	knows
		“Tunç believes that Hillary won.” ↛ Hillary won.									(non-factive)
	b.	Ama	kazan-ma-dı.
		but	win-neg-pst.3s
		“But she didn’t win.”

The same behavior can be seen in Hebrew, in which the proximal demonstrative ze in the embedded clause induces a factive reading:

(13)

Hebrew (Kastner 2015)

a.

hu

hisbir

et

[ze

še-ha-binyan

karas]

(#aval

hu

lo

be’emet

he

explained

acc

this

comp-the-building

collapsed

but

he

neg

really

karas)

collapsed

“He explained the fact that the building collapsed (# but it didn’t).”

(factive)

b.

hu

hisbir

še-ha-binyan

karas]

(aval

hu

lo

be’emet

karas)

he

explained

comp-the-building

collapsed

but

he

neg

really

collapsed

“He explained that the building collapsed (but it didn’t).”

(non-factive)

According to Özyildiz (2017), these facts can be interpreted by either of the following alternatives: (1) factivity is entirely encoded in the embedded clause selected by an attitude verb; or (2) the factive meaning emerges in the semantic composition of an attitude verb in conjunction with the embedded clause. The author argues that Turkish facts support the second alternative, and we return to this question in Section 4.3.

While it is natural to think that factivity is built into the attitude verb, the configuration of subordinate clauses constrains the interpretation of these verbs in some way. More specifically, nominal structure induces factive readings. In this paper, we engage in this debate from a different starting point. In Karitiana, all embedded clauses show nominal morphosyntactic traits, and hence, we assume that they are nominalized. Since nominalized clausal complements tend to induce factivity in many languages, we investigate the impact that the structure of complement clauses may have in the attitude verbs in Karitiana. Particularly, we check whether the omnipresent nominalization of embedded clauses precludes non-factive readings of attitude verbs, especially for those such as ’believe’ or ’want’ (which are typically non-factive in other languages).

Before moving to attitude verbs, the properties of embedding in Karitiana are described in the next section.

3. Syntax of Embedded Clauses

The goal of this section is both to provide an overall picture of Karitiana and to argue that embedded clauses in Karitiana are indeed nominalized. We review the evidence put forth by Vivanco (2018, 2022) in favor of a nominal status of embedded clauses, which includes two blocks of arguments: (1) the lack of clausal structure in embedded clauses; (2) the nominal morphosyntatic behavior of embedded clauses. These arguments are presented in Section 3.2 and Section 3.3, respectively.

3.1. Karitiana

Karitiana (ISO 639: ktn) is a Tupian language spoken in Rondônia, a northwestern State of Brazil. It is still spoken by 369 people who live in an indigenous reservation (Terra Indígena Karitiana) and in the urban areas of Porto Velho, the capital of Rondônia (Storto and Rocha 2018). Even though the Karitiana language is currently being transmitted to younger generations, it faces several challenges such as the vicinity of urban areas, where Portuguese is heavily spoken.

Regarding its morphosyntactic features, the structure of noun phrases in Karitiana is noteworthy. The language only has bare nouns, as can be seen with the direct complement gooj (‘canoo’) in the example below (Müller 2011; Müller et al. 2006):

(14)	Maria	∅-naka-m-’a-t	gooj
	Maria	3-decl-caus-make-nfut	canoe
	“Maria built canoe(s).”			(Müller et al. 2006)

These elements may receive a singular or plural reading, and they can also be indefinite or definite (examples (15) and (16), respectively):

(15)	Yn	∅-naka-sadna-j	õwa	hadna	hyk
	I	3-decl-report-fut	child	story	about
	“I will tell the story of a child.”					Müller (2011)

(16)	∅-py-pyn-a	andyk-y-n	õwa
	ass-know-epen	impf.pass-epen-nfut	child
	“The child was smart.”			(Müller 2011, adapted)

Moreover, quantificational indefinites also arise without any determiner. This is the case of ’boar’ (sojxa) in the example below:

(17)	An	i-oky-t	sojxa	hỹ?
	2s	3-kill-nfut	boar	interr
	“Did you kill boars/any boar?”				(Landin 1984, adapted)

Therefore, we adopt the structure of noun phrases in Karitiana as follows, with no DP layer:

(18)

Karitiana is an ergative language, and this case alignment is mostly detected in main clauses (Storto 1999, 2005 but see also Landin 1984). As can be seen in examples (19) and (20), intransitive verbs agree with their subjects, but transitive verbs agree with their objects.

(19)	Karitiana (Storto 1999)
	A-ta-opiso-t	an
	2-decl-listen-nfut	you
	“You listened.”

(20)	Karitiana2 (Storto 1999, adapted)
	Yn	a-ta-oky-j	an
	I	2-decl-kill/hurt-fut	you
	“I will hurt you.”

Another important feature of Karitiana is a complementary distribution between matrix and embedded clauses regarding word order and verbal morphology (Storto 1999). Matrix clauses are mainly V2 and display full verbal morphology, whereas embedded clauses are obligatorily verb-final and do not show agreement, tense, and mood morphemes: in (21), the verb ‘oky’ occupies the second position of the matrix and fully inflected clause, while the same verb sits bare in the final position in the embedded clause of (22):

(21)	Karitiana (Storto 1999)
	Yn	a-ta-oky-j	an
	I	2s-decl-kill-fut	you
	“I will kill you.”

(22)	Karitiana (Vivanco 2018)
	Ivan	∅-naka-sadn-∅	[Ana	ombaky	oky]
	Ivan	3-decl-report-nfut	Ana	jaguar	kill
	“Ivan reported that Ana killed a jaguar”

These features shed light onto the structure of clauses in Karitiana, a topic that we address in the following sections.

3.2. Lack of Clausal Structure in Embedded Clauses

As first noticed by Storto (1999), embedded clauses in Karitiana do not exhibit full clausal structure. They lack CP, and agreement, mood, and tense morphemes are also obligatorily absent.

As far as complementizers are concerned, Storto (1999) claims that embedded clauses in Karitiana lack the syntactic projection CP, which is, on the other hand, present in matrix clauses. Three pieces of data support this conclusion:

(23)	Arguments for the lack of C in embedded clauses
i There are no lexical items that could be classified as complementizers; ii The complementary distribution in word order (examples (21) and (22)) can only be explained by the lack of C; iii There are no indirect questions in the language.

We begin the discussion of the lack of C in embedded clauses (but not matrix clauses) with the property (i). Embedded clauses are introduced without any complementizer-like lexical item, as examples (24)–(26) show.3 These sentences illustrate three types of subordinate clauses: a direct complement clause, an indirect complement clause (marked with the oblique marker -ty), and a relative clause:4

(24)	Karitiana (Vivanco 2018)					(=22)
	Ivan	∅-naka-sadn-∅	[Ana	ombaky	oky]
	Ivan	3-decl-report-nfut	Ana	jaguar	kill
	“Ivan reported that Ana killed a jaguar.”

(25)	Karitiana (Vivanco 2018)
	Karin	∅-na-aka-t	i-koro’op kãra-t	[Ana	médico	mĩ]-ty.
	Karin	3-decl-cop-nfut	nmz-suspect/think-cop.agr.	Ana	doctor	hit-obl
	“Karin thinks that Ana hit a doctor.”

(26)	Karitiana (Vivanco 2014)
	Yn	∅-na-aka-t	i-pyting-∅	[boet	Luciana	ti-m-’a]-ty
	1s	3-decl-cop-nfut	nmz-want-cop.agr	necklace	Luciana	inv-caus-make-obl
	“I want the necklace that Luciana made.”
	(Context: The speaker had two choose between two necklaces)

Since there is no overt counterpart of C, the simplest analysis would be to consider that the language has no C altogether. This analysis can be further supported by observations (ii) and (iii). Regarding (ii), the word order facts presented by examples (21) and (22) can easily be captured if embedded clauses lack C. If one assumes that C is absent in embedded clauses, this contrast is readily comprehensible. Storto (1999) claims that the verb raises to C and acquires agreement, tense, and mood in matrix clauses, similarly to Germanic V2. Since there would be no C in subordinate clauses, the verb stays in situ and the verb-final order emerges.5 This is also tied to the other non-clausal property of embedded clauses: the lack of agreement, tense, and mood morphology. As seen in example (22), the embedded verb oky emerges bare, without prefixes or suffixes related to agreement, tense, or mood.6 According to Storto, C would host agreement, tense, or mood morphology, and since C would be missing in embedded clauses, these morphemes are consequently absent in subordinate environments.

Finally, the last argument pointing to the lack of C in embedded clauses was presented by Vivanco (2018). This is the property (23iii), namely, the absence of "real" indirect questions in the language. One can see in the examples below that the interrogative locative pronoun tihoori (‘where’) is absent in embedded clauses of (28) and (29):

(27)	Karitiana (Vivanco 2018)
	Tihoori	i-pytim’adna	Ivan?
	where	3-work	Ivan
	“Where does Ivan work?”

(28)	Karitiana (Storto 2022)
	A-so’oot-o	mini	an	[‘õrom	heredni-pa-ty],	y-syp’et?
	2s-see-epen	neg	2s	black.monkey	live-loc-obl	1s-uncle
	“Do you know where the black monkeys live, my uncle?”
	(Literally: “Didn’t you know (the place) of the black monkeys, my uncle?”)

(29)	Karitiana (Vivanco 2018)
	Ana	i-sondyp-y-wak-∅	[Ivan	pytim’adn-i-pa]-ty.
	Ana	nmz-know-epen-des-cop.agr	Ivan	work-epen-nmz-obl
	“Ana wants to know where Ivan works.”
	(Literally: “Ana wants to know about Ivan working.”)

The reason why interrogative pronouns are absent in embedded clauses is that there would be no place to host them.7 Without C, there is no projection CP to make room for interrogative pronouns in its specifier position.8

In sum, all the observations in (23i-iii) are easily captured by a structure without C (and its projection CP).

3.3. Presence of Nominal Morphology

The last section showed that embedded clauses do not exhibit full clausal status, since C and its projection CP are absent in subordinate environments. In this section, we show that embedded clauses exhibit nominal morphology and behavior, which brings them closer to noun phrases. The nominal features to be described are the following:

(30)	Nominal features of embedded clauses
i Presence of the oblique marker -ty ii Presence of the postposition -p iii Presence of the overt nominalizer -pa iv Pied-piping in WH- questions

Regarding (30i), Karitiana allows the oblique marker -ty both on noun phrases and on embedded clauses whenever the matrix verb requires it. The verb pyting, for example, takes an argument marked with -ty, as can be seen in the simple sentence (31) (Rocha 2011). Crucially, when the complement of pyting is a clause as in (32), -ty must also be present.

(31)	Karitiana (Rocha 2011)
	∅-py-pyting-yn	$\tilde{j}$onso	opi-ty
	3-ASS-want-nfut	woman	earring-obl
	“The woman wants the earring.”

(32)	Karitiana (Vivanco 2018)
	Yn	∅-na-aka-t	i-pyting-∅	[gijo	Luciana	ti-tak]-a-ty.
	I	3-decl-cop-nfut	nmz-want-cop.agr	corn	Luciana	inv-grind-epen-obl
	“I want the corn that Luciana ground.”
	(Context: the speaker had to choose between two types of corn)

Another suffix occuring both in noun phrases and embedded clauses is the locative postposition -p (the nominal feature (30ii)):

(33)	Karitiana (Everett 2006, adapted)
	Yn	∅-naka-m-tat-∅	him	pisyp	ambi-p
	I	3-decl-caus-go-nfut	meat		house-loc
	“I sent meat to the house.”

(34)	Karitiana (Rocha 2016)
	Inacio	ø-na-aka-t	i-tat-ø	akan	pip	[’irip	Inácio	oky]-p
	Inacio	3-decl-cop-nfut	nmz-go-cop.agr	village	in	tapir	Inácio	kill-loc
	“Inácio went to the village to kill tapirs.”

(35)	Karitiana (Ferreira 2017)
	Antonio,	[carro-tyyt	y-aki-p]	yn	a-$\tilde{j}$y-atot-∅
	Antonio	car-with	1s-cop-loc	I	2s-cond-take-nfut
	“Antônio, if I had a car, I would take you.”

When present in embedded clauses, the suffix -p gives rise to different (but related) interpretations when compared to noun phrases. In (34), it is a purpose clause ‘to kill tapirs’, whereas it marks the clause expressing the condition of a conditional construction in (35). On the basis of examples such as (34), Rocha (2016) calls this morpheme ‘infinitival’, but we prefer to see it as the same postposition found in noun phrases.

Overt nominalizers may also be present in embedded clauses (the property (30iii)). This is the case of the suffix -pa, which attaches to a base to create a locative or an instrumental noun:

(36)	Karitiana (Storto 1999)
	Kookot-o-pa
	pass-epen-nmz
	“bridge”

This suffix is found in locative relative clauses as in (37) and (38):9

(37)	Karitiana
	Y-pyr-amynt-yn	[cama	taso	kat-a-pa-]ty
	1s-ASS-buy-nfut	bed	man	sleep-epen-nmz-obl
	“I bought the bed where the man slept.”
	(Literally: “I bought the man-sleeping bed.”)

(38)	Karitiana (Vivanco 2018)
	Ana	i-sondyp-y-wak-∅	Ivan	pytim’adn-i-pa-ty.
	Ana	nmz-know-epen-des-cop.agr	Ivan	work-epen-nmz-obl
	“Ana wants to know where Ivan works.”
	(Literally: “Ana wants to know about Ivan working.”)

The fact that -pa is also found in noun phrases constitutes evidence that it is not a complementizer, but truly a nominalizer.

Finally, another similarity between noun phrases and embedded clauses is that both are opaque domains for extraction and require pied-piping (property (30iv)). It is ungrammatical to move a possessor out of an NP, as well as extracting an argument out of an embedded clause. In the example (40), one can see that the genitive WH-word morã cannot be fronted alone. It has to pied-pipe the whole noun phrase ‘whose manioc’ as in (40):

(39)	Karitiana (Vivanco 2018)
	*Morãi-mon	Ivan	ti-’y-t	[ti	gok]?
	wh-int.cop.	Ivan	inv-eat-cop.agr		manioc

(40)	Karitiana (Vivanco 2018)
	[Morã	gok]i-o-mon	Ivan	ti-’y-t?
	wh	manioc-epen-int.cop	Ivan	inv-eat-cop.agr.
	“Whose manioc did Ivan eat?”

The data below show that the very same pattern is found with embedded clauses. Morã cannot undergo WH-movement without pied-piping the whole clause [morã Karin tiko] (Vivanco 2018):

(41)	Karitiana
	*Morã	i-kãra	Karin	[ti	Ana	mĩ]-ty?
	wh	3-suspect/think	Karin		Ana	hit-obl
	“(Intended) What does Karin think that Ana killed?”

(42)	Karitiana (Vivanco 2018)
	[Morã	pop	’it]-i-ty	i-kãra	Karin?
	wh	die	make-epen-obl	3-suspect/think	Karin
	“Who does Karin think fainted?”

All in all, just like noun phrases, embedded clauses are islands for WH-movement and require pied-piping. In the next section, the proposed structure for embedded clauses in Karitiana is presented, and we discuss how it precludes movement while capturing all the morphosyntactic facts described in (23) and (30).

3.4. Structure of Embedded Clauses

In the previous sections, several features of embedded clauses were presented and discussed. We saw that they exhibit two types of features: one that distinguishes them from full clauses (properties in (23)) and another that brings them closer to noun phrases (properties (30)). Therefore, the structure of embedded clauses in Karitiana must be somewhat different from full clauses, while exhibiting some sort of nominal structure at the same time.

In Section 3.2, we saw that embedded clauses do not contain a CP projection. The lack of tense and mood also led Storto (1999) to propose that the embedded Infl is much poorer than its matrix counterpart. Based on that, Storto proposed the following structure for all embedded clauses:10

(43)

Due to the lack of IP, the structure in (43) is non-finite (see also Rocha 2016). This cannot be the whole story, though. Traditionally, non-finite clauses usually tend to make extraction out of them easier because there are fewer projections to interfere between the target and the source positions of movement (see, for example, the phenomenon of raising). This is not what is found in Karitiana. As seen in Section 3.3, extraction out of embedded clauses is ruled out, requiring pied-piping of the whole embedded clause containing the WH-word. Therefore, there must be something in the structure of embedding that interferes with WH-movement, and crucially, this is probably the very same thing that bans extraction out of noun phrases.

We then follow Vivanco’s (2018, 2022) proposal that embedded clauses in Karitiana are actually derived nouns, which include a nominal head N:

(44)

In most cases, N would be null in embedded clauses. Nonetheless, there is at least one possibility for the overt realization of N: subordinate clauses with -pa, as examples (37) and (38) show. In these cases, we propose that the derivational suffix -pa is the overt realization of N.

This N head would turn an embedded clause in Karitiana into a strong island. As proposed by Vivanco (2018), extraction out of embedded clauses can be understood as a violation of the Complex NP Constraint, which was proposed by Ross (1967) to account for data such as (46):11

(45)	Complex NP Constraint12
	No element contained in a sentence dominated by a noun phrase with a lexical head
	noun may be moved out of that noun phrase by a transformation	(Ross 1967)

(46)	English (Ross 1967)
	*The hati which I believed [the claim that Otto was wearing ti] is red.

Summing up, embedded clauses in Karitiana display a nominal structure due to the presence of a nominal head N. In the next sections, we investigate whether clausal complements of attitude verbs, which would be nominal as any other embedded clause in Karitiana, impact the interpretation of matrix verbs.

4. Semantic Interpretation of Complement Clauses in Karitiana

4.1. The Nature of the Factive Inference Tested

The defining characteristic of factive verbs remains a matter of debate.13 Kiparsky and Kiparsky (1971) and Karttunen (1971) assume that presupposition projection is the only defining property shared by factive predicates. Projection in this context is the ability of the truth of the complement surviving negation of the main clause (among other tests). For example, Karttunen (1971) shows that both the sentence in (47a) and its negated counterpart in (47b) still presuppose (47c):

(47)	a.	John realized that he had no money.
	b.	John didn’t realize that he had no money.
	c.	John had no money.	(Karttunen 1971)

Nonetheless, other authors (such as Anand and Hacquard 2014) point out that what defines a factive verb is both presupposition and entailment of their complement. A statement A semantically entails a statement B here if the truth of A (as in (48a)) consequently implicates the truth of B (as in (48b)) and the truth of B cannot be negated.

(48)	a.	Lee kissed Kirn passionately.
	b.	Lee kissed Kirn.	(Chierchia and McConnell-Ginet 1990)

More notably, more recent experimental studies show a diverse spectrum of judgments of presupposition projection, even for canonical factive verbs (e.g., see Degen and Tonhauser 2022), which, as a consequence, casts doubt on how crucial presupposition is for this class of verbs.

In this study, we circumvent this debate and test the factive inference of these types of verbs in terms of entailment only14 and we follow a definition of entailment à la Chierchia and McConnell-Ginet (1990), wherein the truth of the complement clause of a factive verb is indefeasible because the whole sentence cannot be followed up with another sentence contradicting its truth value. In other words, an utterance of the sentence containing the factive verb followed by another utterance that negates the truth of the proposition of the factive complement should result in contradiction. For the purposes of this study, we are thus using the definition of factive verbs that assumes that the content of their clausal complement must be both entailed and presupposed (e.g., Anand and Hacquard 2014).

4.2. Methodology

As described in Section 2, factive attitude verbs in many languages select for clausal complements with nominal traits. Section 3 defended the position that all embedded clauses in Karitiana are nominalized, and one may wonder how this nominal status could impact the interpretation of attitude verbs.

In order to address this question, fieldwork was conducted with two Karitiana consultants through online elicitation sessions. Both speakers are fluent both in Karitiana and Portuguese, and their ages range between 30–50 years old. Throughout this section, they are referred to as Consultants 1 and 2 to assure their anonymity.

Table 1. Verbs tested.

Verb	Possible Translations
diwyt	forget
sodyp	know
sikina	remember
koro’op ’yra	lament
kybawa	doubt, not believe
kywytit	thinks “for sure”, believe
pyting	want
(koro’op) kãra	suspect, think

shows the attitude verbs tested in this study with the most common translations provided by our consultants.15

All verbs in Table 1 can take an embedded clause as its complement.

In order to assess entailment, the method used was elicitation, which can be defined as “the collection of responses to linguistic or non-linguistic stimuli designed to study the respondents’ linguistic competence and/or their practices of language use” (Bohnemeyer 2015). The response in this case was judgments of contradiction. Specifically, two sentences were presented to the consultants, who had to verify whether uttering both would be contradictory or not. More specifically, a second sentence following the primary sentence under evaluation could deny or not the content of the first, as exemplified by the filler context below (background information: the city of Rio de Janeiro belongs to the Rio de Janeiro state. Rondônia is another state, whose capital is called Porto Velho):

(49)	Karitiana	∅-na-aka-t		i-ki-t			Rondônia		pip.	Karitiana
	Karitiana	3-decl-cop-nfut		part-live.pl-cop.agr			Rondônia		in	Karitiana
	∅-na-aka-t		i-ki-t		Rio	de	Janeiro	pip.
	3-decl-cop-nfut		part-live.pl-cop.agr		Rio	de	Janeiro	in
	“The Karitiana live in Rondônia. The Karitiana live in Rio de Janeiro.”

It is important here to draw a distinction between contradiction and judgments of contradiction. Contradiction is a theoretical concept: a (manufactured) status of truth-value clash between two sentences. Judgments of contradiction are a speaker’s stance in relation to this possible clash of truth-values. Judgments of contradiction are therefore an indirect way to access contradiction, and as such, they are subject to many other intervening factors and must be taken into account carefully.

Simple sentences were added as fillers, and these were presented both at the beginning of the session as a training or interspersed with the target contexts throughout the elicitation session. All fillers were affirmative sentences. Both speakers succeeded in all fillers in the sense that they both rightly classified the fillers as contradictory or non-contradictory when that was actually the case.

For target contexts, two types of two-sentence sequences were presented to the speakers, the difference between them being the nature of the second sentence (either negative or affirmative). These types are named here as Test 1 and Test 2. Test 1 is the most paradigmatic method to detect entailment and has frequently been used in the literature to assess factivity. Test 2 was also added because the follow-up sentence does not contain an explicit negation that could induce speakers to offer more negative judgments. An example of each test is depicted below in English:

• TEST 1: João remembered that Luciana killed the jaguar, but Luciana didn’t kill the jaguar.
• TEST 2: João remembered that Francisco was born in [the city of] Porto Velho, but Francisco was born in [the city of] São Paulo.

The Karitiana equivalents of these sentences are as follows:

• TEST 1: João remembered that Luciana killed the jaguar, but Luciana didn’t kill the jaguar.

  (50)	João		∅-na-aka-t			i-sikina-t	[Luciana	ombaky	oky]-ty.	I-oky
  	João		3-decl-cop-nfut			part-remember-cop.agr	Luciana	jaguar	kill-obl	3-kill
  		padni		ombaky	Luciana.
  		neg		jaguar	Luciana
  	“João remembered that Luciana killed the jaguar, (but) Luciana didn’t kill the jaguar.”

• TEST 2: João remembered that Francisco was born in Porto Velho, but Francisco was born in São Paulo.

  (51)

  João

  ∅-na-aka-t

  i-sikina-t

  [Porto

  Velho

  pip

  Francisco

  João

  3-decl-cop-nfut

  part-remember-cop.agr

  Porto

  Velho

  in

  Francisco

  yt]-y-ty.

  Francisco

  ∅-na-aka-t

  i-yt-∅

  São

  be.born-epen-obl

  Francisco

  3-decl-cop-nfut

  part-be.born-cop.agr

  São

  Paulo

  pip.

  Paulo

  in

  “João remembered that Francisco was born in Porto Velho, but Francisco was born in São Paulo.”

In the next section, we discuss some challenges that arose when eliciting the judgments for the aforementioned types of sequences. On the one hand, we found that Test 1 was especially challenging, that is, when the first sentence under evaluation is expressively followed by the negation of the proposition of the complement of the factive verb. On the other hand, Test 2, which also denies the truth of the complement clause yet without directly using negation in front of the content of the embedded proposition, might have been, to some extent, easier to grasp by our consultants. As a consequence, the results had to be interpreted as a whole, taking into account both the results from Test 1 and Test 2 in conjunction.

4.3. Results

In the last section, a distinction was drawn between contradiction and judgments of contradiction. Contradiction is a binary concept: a sequence of sentences is either a contradiction or not. Judgments, on the other hand, are a little more complicated. As pointed out by Bohnemeyer (2015): “Judgments are almost always of a graded nature. That is, even if a speaker gives a categorical response to a simple polar question, this response can be ranked in relative strength with respect to the same speaker’s responses to other stimuli.” Therefore, even though contradiction is binary, judgments of contradiction do not necessarily have to be—and in our study, they were not in some cases.

Generally speaking, the main result is that the tested verbs can be divided into two subcategories as in in

Table 2. Two subcategories of verbs tested.

Group 1	Group 2
diwyt (forget)	kybawa (not believe)
sodyp (know)	kywytit (think “for sure”/believe)
sikina (remember)	pyting (want)
koro’op ’yra (lament)	(koro’op) kãra (suspect/think)

: one class of verbs that yields a very consistent response (Group 1) and another whose judgments are much more complex to interpret (Group 2). The latter will require us to take a closer look at them as their behavior cannot be categorically classified:

The criteria for inclusion in each group will be justified in detail. We start with verbs of Group 1:

(52)	Criterion for inclusion in Group 1
	The sequence of sentences was consistently labeled as a contradiction in Test 1 and Test 2 by both speakers and the clarification given by our consultants was compatible with the use of such a label.

The results for Group 1 verbs are depicted in

Table 3. Responses of Group 1 for Test 1: “…but Luciana didn’t kill a jaguar”.

Verb	Consultant 1	Consultant 2
diwyt (forget)	contradiction	contradiction
sodyp (know)	contradiction	contradiction
sikina (remember)	contradiction	contradiction
koro’op ’yra (lament)	contradiction	contradiction

and

Table 4. Responses of Group 1 for Test 2: “…but Francisco was born in São Paulo”.

Verb	Consultant 1	Consultant 2
diwyt (forget)	contradiction	contradiction
sodyp (know)	contradiction	contradiction
sikina (remember)	contradiction	contradiction
koro’op ’yra (lament)	contradiction	contradiction

:

The second group, on the other hand, comprises verbs whose judgments were not consistent across speakers and/or tests. In some cases, the sequences of sentences with Group 2 verbs were judged as pure contradictions, but in other cases, a variety of responses was obtained. Therefore, the criterion for inclusion in Group 2 is the occurrence of at least one of these different responses, as described below:

(53)	Criterion for inclusion in Group 2
	A verb will belong to Group 2 if at least one of the following responses was obtained:
A The judgment given by at least one consultant was “it is not a contradiction”; B The judgment given by at least one consultant was “it can be a contradiction or not”; C The judgment given by at least one consultant was “it is a contradiction” but the justification for this response was not in line with a linguistic contradiction. For example, the justification given was “because the second sentence contradicts the [subject of the previously stated matrix clause’s] opinions/beliefs.” In this case, the explanation does not necessarily justify a situation that would contradict the occurrence of the entailed event.16

Example (54) shows a verb classified into Group 2 by Criterion (53A): Consultant 2 stated that the two sentences are not contradictory and presented a justification.

(54)	João		∅-na-aka-t				i-koro’op kãra-t	[Luciana	ombaky	oky]-ty.
	João		3-decl-cop-nfut				part-suspect/think-cop.agr	Luciana	jaguar	kill-obl
		I-oky		padni	ombaky	Luciana.
		3-kill		neg	jaguar	Luciana
	Possible translation: “João thought that Luciana killed a jaguar, (but) Luciana didn’t kill a jaguar.”
	Judgment and justification of Consultant 2: “It is not a contradiction, because João was afraid that the jaguar would kill Luciana, but the jaguar didn’t kill her.”

Next, (55) illustrates a response (53B) given by Consultant 1:

(55)

João

∅-na-aka-t

i-kywytit-∅

[Porto

Velho

pip

Francisco

João

3-decl-cop-nfut

part-believe-cop.agr

Porto

Velho

in

Francisco

yt]-y-ty.

Francisco

∅-na-aka-t

i-yt-∅

São

be.born-epen-obl

Francisco

3-decl-cop-nfu

part-be.born-cop.agr

São

Paulo

pip.

Paulo

in

Possible translation: “João thought that Francisco was born in [the city of] Porto Velho, (but) Francisco was born in [the city of] São Paulo.”

Judgment and justification of Consultant 1: “It can be a contradiction or not.”

Below, an example fitting Criterion (53C) is provided:

(56)

João

∅-na-aka-t

i-kãra

[Porto

Velho

pip

Francisco

João

3-decl-cop-nfut

part-suspect/think-cop.agr

Porto

Velho

in

Francisco

yt]-y-ty.

Francisco

∅-na-aka-t

i-yt-∅

São

be.born-epen-obl

Francisco

3-decl-cop-nfu

part-be.born-cop.agr

São

Paulo

pip.

Paulo

in

Possible translation: “João thought that Francisco was born in [the city of] Porto Velho, (but) Francisco was born in [the city of] São Paulo.”

Judgment and justification of Consultant 2: “It is a contradiction, because it contradicts John’s expectation, but it is a possible situation.”

Finally, in example (57), the Consultant 2’s response also fits into Criterion (53C).

(57)

João

∅-na-aka-t

i-koro’op kãra-t

[Porto

Velho

pip

Francisco

João

3-decl-cop-nfut

part-suspect/think-cop.agr

Porto

Velho

in

Francisco

yt]-y-ty.

Francisco

∅-na-aka-t

i-yt-∅

São

be.bornepen-obl

Francisco

3-decl-cop-nfut

part-be.born-cop.agr

São

Paulo

pip.

Paulo

in

Possible translation: “João thought that Francisco was born in [the city of] Porto Velho, (but) Francisco was born in [the city of] São Paulo.”

Judgment and justification of Consultant 2: “It is a contradiction, because João was feeling that Francisco would be born in Porto Velho, but then Francisco was born in São Paulo.”

Summing up, the key difference between Group 2 in relation to Group 1 is that the use of the same verbs in non-factive contexts are not consistently judged as contradictory like with Group 1. The variation of results can be seen in the tables below for Test 1 (

Table 5. Responses of Group 2 for Test 1: “…but Luciana didn’t kill a jaguar”.

Verb	Consultant 1	Consultant 2
kybawa (not believe)	contradiction	not contradiction (53A)
kywytit (believe)	contradiction	contradiction
pyting(want)	contradiction	contradiction
(koro’op) kãra (suspect/think)	contradiction	not contradiction (53A)

) and for Test 2 (

Table 6. Responses of Group 2 for Test 2: “…but Francisco was born in São Paulo”.

Verb	Consultant 1	Consultant 2
kybawa (not believe)	not contradiction (53A)	not contradiction (53A)
kywytit (believe)	it can be a contradiction (53B)	contradiction
pyting (want)	contradiction (53C)	contradiction (53C)
(koro’op) kãra (suspect/think)	contradiction (53C)	contradiction (53C)

):

When mixed results across speakers and/or tests are obtained as in Group 2, we cannot conclude that the complement of the factive verb is entailed in the same way as it is for the factive group (Group 1).17 However, it is also important to stress that even though judgments of non-contradiction are found with these verbs (or something along the lines like it could be contradictory and non-contradictory), there are a certain number of judgments that are labeled as truly contradictions by the consultants, even for the verbs listed in Group 2. This will be important for our classification and understanding of these Group 2 verbs, which is the topic of the next section.

4.4. Discussion

Whenever a sequence of sentences with a certain verb is consistently judged as a contradiction as in Group 1, then we consider this a very strong indication of true contradiction. If a contradiction arises, then we can conclude that there is indeed a factive entailment. Therefore, we conclude that verbs in Group 1 are truly obligatorily factive.

The picture with verbs of Group 2 is less clear, though. The judgments were less consistent in comparison with Group 1, so they do not seem to be obligatorily factive. Nonetheless, the tests containing these verbs were not constantly judged as non-contradictions either, as it would be the case if they were obligatorily or truly non-factive. One possible interpretation of the results in Table 5 and Table 6 is that they can have a factive reading, but this is not obligatory and depends on several factors. In other words, factivity is, for verbs of Group 2, optional to some extent.

Importantly, it has been observed that a binary distinction opposing factive and non-factive predicates may not be universal. For instance, Degen and Tonhauser (2022) conducted an empirical study where they compare twenty canonically and non-canonically factive verbs in English and concluded that there is no evidence for a categorical distinction; rather, a more fine-grained scale might be necessary. Along these lines, in another recent empirical study, Dahlman and van de Weijer (2022) observe different degrees of cross-linguistic acceptance of canonically factive verbs like know used in non-factive contexts across nine different languages. Therefore, variability in judgments is found in other languages as well, and in Karitiana specifically, this would be the situation with verbs of Group 2, which may have a factive reading—or even a high tendency/strong preference to be interpreted factively—but do not require it as verbs of Group 1.

Turning now to the research questions discussed in Section 2, we are now in a position to evaluate the impact that the structure of the embedded clauses in Karitiana has on the interpretation of attitude verbs. The judgments of sentences with verbs of Group 2 prove that nominalization alone is not sufficient to force a factive reading of the matrix attitude verb, as there are cases in which a non-factive reading was possible, i.e., when the two-sentence sequence was not judged as contradiction (Criteria (53A) and (53B)).

What we would like to suggest at this point is that nominalized embedded clauses do have an important contribution to factive readings, but that they alone may be insufficient to generate them. This is what has been proposed by Özyildiz (2017). According to the author, the factive meaning emerges in the composition between certain verbs and a specific type of clause (a nominalized one).18 This view is supported by the fact that in many languages, nominal structure is a necessity, but not a sufficient condition to the presuppositional/factive reading of an attitude verb. In Turkish, for example, certain verbs such as bil- ‘know’ are interpreted as factive if their clausal complement is nominalized (see examples (58a) and (58b), repeated below). However, other verbs such as düşün ‘think’ may take a nominalized clause (as in (59a)) without having any kind of factive interpretation.

(58)

Turkish (Özyildiz 2017)

a.

Tunç

[Hillary’nin

kazan-dığın-ı]

biliyor

Tunç

Hillary

win-nmz-acc

knows

“Tunç knows that Hillary won.’ → Hillary won.”

(factive)

b.

Tunç

[Hillary

kazan-dı

diye]

biliyor.

Tunç

Hillary

win-pst

diye

knows

“Tunç believes that Hillary won.” ↛ Hillary won.

(non-factive)

(59)	Turkish (Özyildiz 2017)
	a.	Tunç	[Hillary’nin			kazan-dığın-ı]				düşünüyor.
		Tunç	Hillary			win-nmz-acc				thinks
	“Tunç thinks that Hillary won.”										(non-factive)
	b.	Tunç	[Hilari		kazan-dı			diye]		düşünüyor.
		Tunç	Hillary		win-pst.pfv			diye		thinks
	“Tunç thinks that Hillary won.”										(non-factive)

The same has been observed for Buryat. The verb hanaxa can receive a factive reading when complemented by a nominalized clause (example (60a)) or a non-factive reading when it selects for a CP (60b). Nonetheless, this is not the case for all verbs: as example (61) shows, the verb ətigəxə (‘believe’) has a non-factive reading regardless of the structure of the complement:

(60)	Buryat (Bondarenko 2019)
	a.	Dugar	[NMNmi:sgəi-n	zagaha	ədj-ə:ʃ-i:jə-n’]			han-a:
		Dugar.nom	cat-gen	fish	eat-part-acc-3sg			think-pst
	“Dugar remembered a cat’s eating fish.”
	b.	Dugar	[CPmi:sgəi	zagaha	ədj-ə:	gəžə]	han-a:
		Dugar.nom	cat.nom	fish	eat-pst	comp	think-pst
	“Dugar thought that a cat ate fish.”

(61)	Buryat (Bondarenko 2019)
	Sajana	badm-i:n	tərgə	əmdəl-ə:ʃ-tə-n’	ətig-ə:,	xarin	badma	tərgə
	Sajana	Badma-gen	cart	break-part-dat-3	believe-pst	but	Badma	cart
	əmdəl-ə:-güi
	break-pst-neg
	“Sajana believes that Badma broke the cart (lit. ‘in Badma’s breaking the cart’), but Badma didn’t break the cart.”

Finally, the factive interpretation of verbs in Group 2 is not fully understood. It could be that these verbs are optionally factive (or say semifactive), in the sense that they strongly implicate the factive reading but still leave open the possibility of finding the right context that would allow for a non-factive interpretation. Nonetheless, even then, there seems to be a strong preference for factive readings of these verbs (specially in Test 1, as can be seen in Table 5). We hypothesize that this could be due to the nominalization of the embedded clause, because nominalized complements do exhibit a factive entailment in other languages. This can be seen in the Portuguese examples below, in which the nominalized complement in (62) has a strong preference for a factive reading (i.e., it strongly implies that John cheated), as opposed to the non-nominalized version in (63):

(62)	Maria	suspeita	da	traição	do	João.
	Maria	suspects	of	cheating	of	João
	“Maria suspected of João’s cheating.”
	(strongly implies that the cheating occurred)

(63)	Maria	suspeita	que	o	João	traiu	a	namorada.
	Maria	suspects	that	the	João	cheated	the	girlfriend
	“Maria suspected that João cheated on (his) girlfriend.”

The same contrast in judgments on factivity for nominalized vs. non-nominalized complements of some attitude verbs like suspect also applies to other Romance languages such as Catalan and Spanish.

All in all, moving from a categorical dichotomy of factive vs. non-factive verbs towards a more gradient scale in terms of factivity might be a more adequate approach in order to capture all the fine-grained intermediate interpretations that appear to occur not only in Karitiana but also cross-linguistically. It is, however, complex to pinpoint the reason behind this, as it appears to be multi-factored: at least both syntactic- (e.g., nominalized vs. non-nominalized clauses) and lexico/context-dependent (e.g., to which degree the factive implication gets accommodated in the discourse, which, in turn, might be influenced by the semantics of the verb). Regarding the latter, our results slightly suggest that the verbs in Group 2 could be further categorized along a scalar degree of factivity depending on how much their use is accepted in non-factive contexts. For instance, on the basis of the consultants’ clarifications to their answers, a three-tier scale of attitude verbs with a denotation similar to ‘know’ could be posited. Under such an approach, the verbs would be placed along a scale of factivity—sodyp (know) » kywytit (think "for sure", believe) » koro’op kãra (think/suspect)—from exhibiting a stronger factive behavior (left) to a weaker one (right). If this idea holds water, something along those lines seems to be the right direction for moving forward. A deeper understanding of how such a scalar reasoning of attitude verbs works remains for future research.

5. Final Remarks

This paper presents an exploratory description of attitude verbs and their clausal complements in Karitiana. Starting from the literature on these verbs—and specifically, on the impact of the clausal structure on factive readings—we proceeded to investigate whether the structure of subordinate clauses impacts the interpretation of these verbs.

Embedded clauses are not full clauses in Karitiana, lacking many clausal features. Moreover, they exhibit a morphological and syntactic behavior that brings them closer to noun phrases. A nominalized structure with an N layer was proposed to account for these facts. We also claimed that this nominalized structure is tied to the status of embedded clauses as strong islands. Lastly, this syntactic structure of the embedded clauses, in which an N head is present, would be similar to factive sentences with “…the fact that…” in English and Hebrew (see Kastner 2015). These constructions, as in Karitiana, behave as strong islands.

Nominalization of embedded clauses does seem to impact the interpretation of attitude verbs: in this language, verbs that are read as non-factive in other languages (e.g., ‘believe’, ‘think’, ‘want’) are what we could categorize as optionally factive or semifactive since by default they are interpreted factively, similarly to nominalized embedded clauses of these verbs in other languages like in Portuguese.

It remains to be answered to what extent the matrix attitude verb contributes to this factive or semifactive interpretation. Up to this point, what we can answer now is that obligatory nominalization of the embedded clause is not enough and the lexical meaning of the verb must contribute to a certain degree for its complement to be taken factively. Nevertheless, how the structure of the clausal complement and the lexical meaning of the verb combine is a matter for future research.