The verb ’a usually means ‘to do/make’ as illustrated in (14). Since it is an intransitive verb, it must be causativized with the morpheme -m in order to have a second argument as illustrated in (14).

This same verb means ‘to say’ whenever it is in a reportative context as illustrated in (13) above. Thus, one possible hypothesis to explain the facts would be to say that there are two homophonous verbs ’a in Karitiana. The first meaning ‘to do’ as illustrated in (15a) and the second meaning ‘to say’ as illustrated in (15b).

We might think that an accidental process in the history of this language was responsible for marking these two different verbs resulted in their having the same form. However, this analysis does not seem to be correct. If that was the case, we would observe this phenomenon only in Karitiana or in languages of the Tupian family. However, the exact same phenomenon occurs in languages from other families, such as Australian Aboriginal languages (Merlan 1983; Rumsey 1982). In Ngalakan, the verb yiniñ ‘to do’ can also be used in reportative constructions, as illustrated in (16). The same holds for another Australian Aboriginal language, Ungarinjin, as shown in (17).

The verb ‘to do’ is also used in reportative contexts in Modern Hebrew (Zuckermann 2006). Thus, the fact that the exact same phenomenon occurs in languages from different families suggests that we are not dealing with an ambiguity caused by accidental homophony, but one single verb which is polysemic as illustrated in (18).

If this fact is common in different languages from different families, there must be something in the semantics of ‘to do’ which makes it a good candidate to be used to report someone else’s speech. We assume that the actions of doing something and of saying something are related following the speech act theory in which utterances are also actions (Austin 1975). When you say something, you are also doing something. In this sense, the semantics of ‘to do’ somehow encompasses the semantics of ‘to say’. Therefore, one could use the broader term to refer to the stricter meaning.

Despite the verb ’a not referring specifically to the “saying” action, it is the most productive reporting strategy. This is puzzling considering that the language has other strategies that specifically refer to a “saying” action. Our analysis is that ‘to do’ is a good candidate to denote a speech act because it can refer not only to the content of the sentence, but to the utterance situation as a whole (the person’s intonation and gestures when he/she says something). That seems to be a correct analysis for the Australian Aboriginal languages Ungarinjin and Ngalakan as described by Merlan (1983: 152):

We therefore adopt a similar analysis for Karitiana as the one given by Rumsey and Merlan for the two Australian Aboriginal languages they study, and we believe this may be the correct analysis for other Tupian languages as well, since all Tupian language families have a cognate of this verb that goes back to Proto-Tupi (Galucio et al. 2015; Rodrigues 2005, 2007; Rodrigues & Cabral 2012). The verb ‘to do’ is not the only example of this phenomenon. For instance, the English verb ‘to go’ can also mean ‘to say’ in certain contexts, as illustrated in (19). In this example, the quotative use of verb ‘to go’ in English is accompanied by an imitation of a person’s voice and gestures; this corroborates the analysis that those verbs are being used to refer not only to the content of the utterance, but to the utterance action as a whole.

When ’a is interpreted as ‘to say’, it does not seem to take what is being said as a syntactic complement. The first evidence we can give is that this is an intransitive verb, therefore, it only takes one argument. In order to have a second argument, this verb needs to be causativized as previously illustrated in (14). However, when this verb is used in a reporting context, no causativization is necessary, as illustrated in (13).

Apparently, the reason why causativization is not necessary is because what is being reported is not a syntactic argument of the verb. For example, in (13), Cizino naakat iotam ‘Cizino has arrived’ is not an argument for naka’at ‘said’. If it were a syntactic argument, it would be a subordinate clause. However, as we showed in Section 2.2, in this language subordinate clauses do not have tense and mood morphology. Since both verbs appear fully inflected for tense and mood in (13), there is no subordination between them.

A phenomenon that could occur in quotative constructions if there was an embedding relation is “sequence of tense”, as illustrated in (20). In formal semantics, it is assumed that tense refers to a topic time (TopT), the time interval being discussed, occurring previous, after, or concomitant to the utterance time (UT), which is the time interval in which the utterance is made. Therefore, tense anchors a time interval in the utterance time (Klein 1994). However, there are some instances in which this does not seem to be the case. For example, the sentence in (20) can be used in a context in which John said yesterday, “I will buy a fish that is alive”, meaning that the fish must be alive at the time he buys it. This can be reported as in (20), with the verb ‘is’ shifted to ‘was’ in the subordinate clause, but this does not mean that he would buy a fish that was alive in some interval prior to the speaker’s utterance time. Rather, this shift occurs because ‘was’ is in the scope of a higher tense which is also past. In other words, since the tense in the matrix is past tense, the tense in the subordinate is also past tense even though the event described can be interpreted as occuring before or after the utterance time. This phenomenon is called “sequence of tense” (SOT) (see Abusch 1988, 1997; Kratzer 1998; Ogihara 1989, 1994; among others).4

In order for the SOT to occur, it is necessary for a tense to be in scope of a higher tense. Then, SOT is restricted to languages that have tense in the subordinate environments. In Karitiana quotative constructions, the verbs from both clauses have tense morphemes, but they are interpreted independently from one another, as in (21). Tense in each verb is interpreted according to the context and UT. This is a second argument suggesting that the reporting structure with ’a is formed by two matrix clauses and that the quotation is not syntactically subordinated to the main verb.5 The interpretations in (21a–d) are represented with logical forms representing the events (e) of ‘Milena saying’ (say (e)) and ‘fish being good’ (good (fish, e)) with respect to the utterance time (UT). In each example, logical forms are also translated in English sentences.

Sentence (21a) was selected by speakers in the following contexts:

The interpretations of (21a) show that when both verbs are marked with non-future morphology -t, the event denoted by those verbs are interpreted as being simultaneous or before the utterance time. This shows that non-future tense in those examples is being determined solely by the utterance time and not by a higher tense.

The interpretations of (21b) show that when only the copular verb is marked with future morphology -j, the fish being good must be in the future (after the utterance time) whereas the saying event can be either in the present or past (non-future). Again, if there was SOT, we could expect the copular verb to be marked with non-future -t, since it would be under the scope of a higher tense of ’a in the same way it occurs in the English example in (20).

Sentence (21c) was selected by speakers in the following contexts:

The interpretations of (21c) show that when the verb ’a is marked by future, the denoted saying event occurs after the utterance time, further illustrating that the verb morphology in both verbs is determined solely by the time to which the verb is anchored, rather than by the utterance time.

Sentence (21d) was selected by speakers in the following contexts:

The interpretations of (21d) show that when both verbs are marked with future morphology -j, both events denoted by the verbs must occur after the utterance time. This further illustrates that the verb morphology in both verbs is not determined by a higher tense.

The fact that the clause which reports is not a complement does not necessarily mean they are completely independent syntactically. It is common for verbs in Karitiana which take a semantic complement to be intransitive and for the complement to be introduced in a postpositional phrase which, syntactically, is an adjunct of the verb. One could argue that this is the case for the ’a reporting structure. However, we argue against this analysis in what follows.

We have seen that the verb ’a is intransitive, and as such, it does not select an embedded clause as its complement. However, there are often structures in Karitiana in which intransitive verbs may have postpositional phrases (PP) as non-obligatory complements. In fact, this is a common strategy for making complex sentence in Karitiana, and they are used in quotation structures as well.

For instance, with the intransitive verb hadn ‘to talk’, note that the postposition tyyt ‘with’ is repeated twice in the examples (22–23). Our interpretation of these structures is that the PP complement of the verb ‘to talk’, in this case, is headed by the postposition ‘with’ that selects as its complement the head-internal relative ‘the woman who is/was/will be a nurse’.

Storto, Vivanco & Rocha (2018) explain this type of construction as the extraction of the head of a relative, in (22–23) ‘a woman’, to the left periphery of that relative (‘who is/was/will be a nurse’). The examples are always formed with relatives marked by the oblique postposition -ty as shown in (24–25). Despite the similarities between tyyt and -ty, they are different items. The postposition tyyt is always used to indicate a relationship between two elements, which can be possession or other types of relations roughly translated as ‘with’ in the contexts in which tyyt appears, whereas the oblique -ty marks an extra argument for the verb and does not have a translation.

The verb so’oot ‘to see’, illustrated in (24–25), is intransitive by all tests and may have a PP complement formed by the oblique postposition and a relative clause. Sentences with oblique relative clauses, such as the above examples, may alternatively be realized with an oblique suffix on the head of the relative co-occurring with the oblique suffix at the end of the clause, as demonstrated in (26):

One might think that this case is a juxtaposition or conjunction between a phrasal oblique argument (kinda’o) and a clausal oblique argument (Maria tiamangã) with a null object. However, if the juxtaposition or conjunction analysis were correct for Karitiana, the order between the two oblique noun phrases could be exchanged, and this is not the case as shown by the ungrammaticality of (27).

We therefore propose that (22) and (23) have the same structure as relatives with the oblique postposition seen in (26); that is, they are formed by an intransitive verb which has a postpositional phrase complement in which a relative clause is the complement of the postposition and its head has moved to the beginning of the relative clause accompanied by the postposition.

We will see below that the intransitive verb ’a may introduce quotations or other phrases (ideophones, for instance) in which the quotations or ideophone phrases are not PP complements of the verb.

The verb ’a can also be used to report on constituents smaller than clauses, as illustrated in (28).

Elicitation context: A jaguar appears in the village. Julenilza is the first person to see it and screams ‘jaguar!’ to warn everybody. But one person does not hear what she screamed and does not understand why everybody is running. So, you say to this person, “Julenilza said, ‘Jaguar!’”.

In a similar way, ideophones or verbal roots can be used as “direct speech” (in fact, “direct enacting”) reported by the verb ’a as illustrated in (29–30).

Both data from our elicitation and from narratives show that the speaker does not need to switch the indexical elements of the sentence. In example (31) there are the indexical elements hak ‘here’, yn ‘I’, and kabm ‘now’. When asked to report this sentence, the speaker used the same indexical elements as illustrated in (32). The same can be observed in data from narratives (33–35).

However, the indexical elements may be switched. Our consultants accepted both structures in which indexicals were switched as in (36) and in which they were not switched as in (37):

We conclude that reportative constructions with ’a are not direct speech because indexical elements can be switched. However, as the sentences’ interpretation is independent and there is no proper subordination, they are not indirect speech either. Thus, there is no direct/indirect speech difference in Karitiana when the verb ’a is used. The same has been argued for reportative constructions with the verb ‘to do’ in Ngalakan (Merlan 1983) and Ungarinjin (Rumsey 1982).

Syntactically, the phrase introduced by the intransitive verb ’a, which describes what is said or done, is not a direct object of the verb, nor a PP complement, because there is no postposition marking the phrase in any of the cases. However, although there is no syntactic embedding, there must be semantic embedding since they understand that, although it is a main clause, it serves as a semantic complement for the reporting verb ’a. Whenever two clauses are coordinated, for the entire clause to be true, the sentences being coordinated must also be true. For instance, in order for the clause illustrated in (38) to be true, both sentences being coordinated must be true, thus, it must be true that Mary studied linguistics and that Mary works in Maryland.

This is not the case for reporting constructions. Whenever something is reported, the whole clause is true if and only if the uttering event is true. The content of the utterance being false does not impact the truth conditions of the clause. For instance, the clause in (39) is true if and only if there was an event of Mary saying that she works in Maryland. If she does not work in Maryland, in other words, if she is lying, it does not change the fact that she has said she worked in Maryland, and therefore the clause in (39) is still true.

This happens because what is being reported is not evaluated independently, but in the scope of the reporting verb. Therefore, although there is no syntactic embedding in ’a constructions, there must be semantic embedding because the fact that someone said something does not necessarily entail that what was said is true.