The goal of this paper is to describe the morphosyntactic properties that characterize reported speech in Sakurabiat. Reported speech in this language is realized through direct quotation embedded in a specific framer composed of a quotative verb and an NP referring to the reported speaker [[RS] QUOT+SBJ].1 In addition, reported speech shows no complementizers, and no deictic category shift inside of speech reports. Hence, inside of speech reports the deictics remain anchored to the context of the reported speaker rather than to that of the reporter. A sentence like ‘John said that he loved me’ is framed as ‘John said: “I love you’”.

In this paper, I use reported speech as a cover term to refer to utterances in which one speaker quotes or reports the speech/thought/perception of another speaker. In this sense, reported speech includes direct reported speech, indirect reported speech and also free indirect (or quasi-direct speech) (Janssen & van der Wurff 1996: 3). The term direct speech is employed here in the sense given by Jespersen (1924: 290 apud Holt & Clift 2007: 3) to refer to the mechanism in which a speaker “gives, or purports to give, the exact words of the speaker.” Hence, we use the terms direct speech, direct quotation, or quote as synonymous. Another important terminological issue concerns the definition of the participants in a reported speech utterance. I use the terms reported speaker (internal speaker) to refer to the speaker responsible for the content of the speech/thought/perception being quoted, and reporter (external speaker) for the speaker that utters the embedding clause in which the reported speech is enclosed (Janssen & van der Wurff 1996: 4). These authors illustrate the definition of reporter/reported speaker with a setting in which Mary utters the sentence ‘John said that he loved me’. In this setting, Mary would be the reporter and John the reported speaker. I also employ in this paper the term quotative construction to refer to the whole utterance which includes the quotative verb and the NP referring to the reported speaker [QUOT+SBJ] plus the speech/thought/perception being quoted (see Section 3).

The Sakurabiat language (ISO skf) is spoken by the Sakurabiat people of the Terra Indígena Rio Mequens in the Brazilian state of Rondônia.2 The language has also been referred to as Mekens in previous works such as Hanke et al. (1958) and Galucio (2001). However, it has been shown that this term was in fact a general ethnonym applied, since the XVIII century, to the indigenous peoples who lived in the area near the headwaters of the Mekens river, which spoke varieties of the same language but self-identified as distinct ethnic groups (Miller 1983, Maldi 1991), such as the Mampi-ape (or Amniapé), Sakurabiat, Guaratira (or Koaratira), and Guarategayat (or Guarategaja), among others. The current population in the Terra Indígena Rio Mekens self-identifies with three of these ethnic groups, the Guaratira, the Guarategayat and the Sakurabiat. The Guarategayat and the Sakurabiat groups speak the same language variety (Sakurabiat) while the Guaratira people speak a distinct variety (Guaratira). The autonym Sakurabiat refers to the language and the people of the Terra Indígena Rio Mekens, including the different clans or sub-ethnic groups and their speech varieties.

Sakurabiat is one of the five languages of the Tuparian branch of the Tupian family, the other four languages being Akuntsu, Makurap, Tupari, and Wayoro (Galucio et al. 2015). All five languages are spoken in the Brazilian state of Rondônia, and they are all considered critically endangered due to their reduced number of speakers. There are currently only about 10–12 fluent speakers of Sakurabiat, including its Guaratira and Sakurabiat varieties. There are about the same number of passive bilinguals.

The Sakurabiat people are currently a small group of less than 100 people, composed of smaller subgroups. These groups were distributed in a large geographical area in the Rondônia region and spoke different but mutually intelligible varieties of the language. Their first contacts with non-Indigenous people are reported to have taken place in the first decades of the 20th century (Galucio 2001), and the continued violence of this contact resulted in the Sakurabiat experiencing a process of loss and linguistic change. This led to marginalization and repression of the language, which in turn reduced its domains of use and interrupted its transmission. Currently their language faces a situation of high vulnerability and low linguistic vitality. The younger generations have not learned Sakurabiat, and Portuguese is dominant. The socio-linguistic situation was made more difficult by the loss of four speakers in the past three years. Even though nowadays only a small percentage of the people speaks or understands the Sakurabiat language, it plays a major role in their cultural, linguistic, and ethnic identities (Galucio 2021).

The data in this paper comes from my own database of Sakurabiat language material, which was collected through extensive fieldwork in past years and which contains natural discourse and elicited data that covers both the Guaratira and Sakurabiat varieties.3 For this specific study on reported speech, the Questionnaire/Considerations for research on quotative constructions (based on Aikhenvald 2008) was used (Veenstra et al. this volume). The speakers were given a context explaining a situation, and then asked to produce a sentence in that specific context. Data collected through application of the questionnaire are marked “Quote-Questionnaire” in the text and referenced by their identifier at the Museu Goeldi Language Archive (ALIM).4 Data extracted from natural discourse are identified as such and given the appropriate identifier in ALIM. Elicited material is otherwise left unmarked. The complete set of Sakurabiat data is archived at ALIM (Galucio 1994–2019). Examples in the text are given in phonemic representation.

The remainder of the paper is organized as follows. Section 2 presents the basics of Sakurabiat morphosyntax. Section 3 describes the main characteristics of reported speech constructions (RSC) in Sakurabiat, and it is further subdivided into two subsections (3.1 Reported speech construction internal structure, and 3.2 Reported speech construction syntax and distribution), which discuss in detail the constituent structure and the syntactic distribution of reported speech constructions, including their interaction with focus and negation. In Section 4, I argue for a reanalysis of quotative constructions as desiderative constructions. It has been shown in the literature (D’Arcy 2015) that reportative verbs tend to grammaticalize into functional categories, such as complementizers, TAME markers, discourse markers, etc. This tendency is also exhibited by Sakurabiat, which shows a crystallization of the direct quote frame ‘I said “I will go”’, and its reanalysis as a desiderative construction, meaning ‘I want to go’. The topics discussed in the paper are then summarized in Section 5.

A comprehensive description of Sakurabiat’s grammar can be found in Galucio (2001). The parts of speech of Sakurabiat are nouns, verbs, adjectives, adverbs, pronouns, demonstratives, auxiliaries, postpositions, and particles. Nouns are further divided into alienable and inalienable, and verbs are subdivided into transitive, intransitive, and uninflectable or particle verbs.5 While transitive and intransitive verbs take person prefixes for argument indexing, and tense and polarity suffixes, particle verbs do not take argument indexing or tense suffixes (Galucio, 2014a). The quotative verbs employed in reported speech constructions belong to the subclass of particle verbs.

Typologically, Sakurabiat is a typical head-final language. The basic constituent order is AOV/SV. However, pronominal subjects, especially monosyllabic ones, cliticize to the verb phrase as enclitics, thus deriving the very common OVA information order. For the most part, tense, aspect, mood, and polarity are expressed by means of postverbal particles, but there are also a few inflectional TAME and polarity suffixes. The language employs two main strategies for coding grammatical relations: word order and person indexing. The core syntactic arguments A, S, and O are not marked for case. In transitive clauses with two nominal arguments, grammatical relations are defined by word order only, as seen in (1–2) below.

In the case of pronominal arguments, grammatical relations are identified through person indexing and through a partially complementary distribution of pronouns versus bound person forms (indexes). The major morphosyntactic alignment pattern is best described as nominative-absolutive (Galucio & Nogueira 2018): the same set of argument indexes (person prefixes) mandatorily expresses the absolutive participant (S/O), whereas the set of pronouns expresses the nominative participant (S/A), optionally for intransitive and for third person transitive subjects in general, and obligatorily for first and second person transitive subjects. A third person transitive subject may be unmarked. This pattern is illustrated by examples (3a–b) for intransitive verbs and examples (4a–c) for transitive verbs.

In addition, absolutive person prefixes—subject (S) and object (O)—show a distinct behavior. Whereas, S indexes may optionally co-occur with nouns or pronouns in subject position, O indexes are in complementary distribution with full nominals. They cannot be doubled by pronouns, that is, they do not co-occur with nominals with the same reference and function.

Galucio (2001) analyzes Sakurabiat subject indexes in intransitive verbs as agreement markers when co-occurring with an NP in subject function, but as grammatical arguments when they occur without an NP. Object indexes, on the other hand, are analyzed as pronominal arguments cliticized to the verb, since they do not co-occur with an NP in the same function (Galucio 2001: 131). This distinction corresponds to the notions of anaphoric vs. grammatical agreement, as described for Chichêwa (Bantu) in the seminal work by Bresnan and Mchombo (1987).8

Indirect objects, such as in (5) and verbal adjuncts such as in (6) and (7), are expressed by postpositional phrases, chosen accordingly to their semantics (dative, locative, ablative).

There is no formal distinction between direct and indirect speech in Sakurabiat. Speech reports used as complements of verbs of utterance or thought, such as ‘say’, ‘ask’, ‘reply’, and ‘think’, are expressed as direct speech or direct quotation. Hence, the language uses direct quotation, i.e., literal repetition of what was said, in order to report speech or thought. When translating into another language, like Portuguese, for instance, the speakers may offer an indirect speech quote, but the actual form produced in Sakurabiat is a direct speech report. The main characteristics of reported speech in Sakurabiat are, thus, the absence of complementizers, the absence of deictic shift inside the quote, and the use of a specific quotative framer. These specific properties will be discussed in the following sections.

It has been observed in various languages that reportative verbs tend to grammaticalize as functional categories, such as complementizers, TAME markers, discourse markers etc. For instance, D’Arcy (2015) discusses the change of lexical dicendi verbs into markers of TAME categories, such as future tense, imperfective aspect, causative, desiderative, imperative, counterfactual, evidential, etc. Similar paths of grammaticalization have been described for some Amazonian languages as well. Two very clear cases of such phenomena are found in van der Voort (2016), wherein the author describes the grammaticalization of quotative constructions as purposive, desiderative, and future constructions in two unrelated languages, Aikanã and Kwazá, spoken in the same region (the Brazilian state of Rondônia) where the Sakurabiat language is spoken.18 In this section, we argue that Sakurabiat quotative constructions can be similarly reanalyzed as desideratives.

As we have seen in the previous sections, direct speech quotes are the main way to report a speech/thought in Sakurabiat, and there is no deictic category shift inside the quote. Hence, the reported speech clause is commonly expressed in the future tense, since it purports to reenact the actual utterance, repeating it exactly as it was uttered. This, in turn, creates a context for the quotative verbs (ke; kaat) to be commonly used within the frame IM.FUT+QUOT+SUBJ, as in (42) below.

Example (42) is a direct speech quote with a future tense marker. The use of the immediate future morpheme entails the meaning that the reported speaker is about to perform the reported event. For instance, the above example expresses an utterance spoken in a context in which the reported speaker (Vilacy’s mother) had finished eating and was heading to her room to lie down. Given that meaning and the typologically common association of immediate future with strong volition or intention, it follows quite straightforwardly that this quotative construction in the immediate future can then be reanalyzed as a desiderative construction, as given in (43).

Thus, a construction like ‘I said “I will go”’ or ‘She said “I will go”’ is reanalyzed as “I want to go” or “She wants to go”, respectively, as seen in Table 1, which indicates the path of reanalysis from quotative to desiderative.

One troublesome point worth noting is that the desiderative reading is based on material from two different levels in the reported speech construction: the future morpheme kot ‘IM.FUT’ is part of the embedded reported speech clause (the direct quotation) and the quotative verb (ke or kaat) is part of the larger quotative framer. I argue that there is a morphosyntactic reanalysis in process, since there are also formal differences between the quotative and the newly reanalyzed desiderative construction. The quotative and the reanalyzed desiderative construction co-occur in the language and show different properties. Unlike quotative constructions, desiderative clauses allow shift of deictic categories. This property is seen in (44), where a quotative reading is no longer possible.

We saw in Section 3 that due to the absence of deictic category shift inside the reported speech, when the reported speaker is third person, there is a difference in the person category between the subject of the reported speech and the subject of the embedding clause (quotative construction). This is illustrated in example (42), where the subject inside the reported speech is first person and the quotative construction subject is third person. However, as we can see in (44), there is total person agreement in the desiderative construction, which is indicated by using the coreferential third person prefix se-. This prefix is used in the object of a transitive (either as the object proper or as its possessor) only when its referent is coreferential with the subject of the sentence. The use of the coreferential prefix also precludes a quotative reading in that example, since a speech report would have a first-person agreement marker as the possessor of the noun, as in o-iko ka kot kaat ‘“I will eat”, (she) said’.

Finally, it is worth mentioning that Sakurabiat also has a lexical verb to express intentions or desires, the transitive verb sobeka ‘to want, to desire’ shown in (44) and (45).

Example (44) above demonstrates that the quotative future tense construction, reanalyzed as a desiderative construction, and the desiderative construction with the lexical verb sobeka ‘to want, to desire’ are equivalent and that they may co-occur in a sentence.

I have described the major morphosyntactic properties that characterize reported speech in Sakurabiat, showing that there is no formal distinction between direct and indirect speech in the language. Reported speech in Sakurabiat is realized through direct speech (or direct quotation), embedded in a specific framer composed of a quotative verb and an NP referring to the reported speaker [[RS] QUOT+SBJ]. Two other characteristics of reported speech in the language are the absence of complementizers and the absence of deictic category shift inside the reported speech clause. The reported speech clause can be a finite clause or a smaller syntactic unit, even just an exclamation.

Direct quotation, even without the quotative framer, is a resource frequently used in narratives and other genres in Sakurabiat, both as structural frame to present supposedly reported utterances and as a mechanism to add vivacity and/or authenticity to the narrative. I argue that at least this use of direct speech quotation, framed as a conversational interaction, fits well the definition of fictive interaction (Pascual & Sandler 2016), since it employs what seems to be a conversation in order to refer to what is not necessarily an actual conversation. The reporting of thought is also in the range of fictive interaction in the language, since reported thought is also realized through the same quotative mechanism of framing it as a direct speech construction, which is presented as conversation with oneself.

Finally, I show that Sakurabiat realizes the widespread tendency observed in the languages of the world for reportative verbs to grammaticalize into other functional categories. Frequency of use, meaning, and the morphosyntactic properties of reported speech constructions in the language provide the context that gives rise to the desiderative interpretation. Due to the absence of deictic category shift inside the quote, the reported speech clause is commonly expressed in the future tense, creating a context for the quotative verbs (ke, kaat) to be commonly used within the frame [IM.FUT+QUOT+SUBJ]. The use of the immediate future morpheme entails the meaning that the reported speaker is about to perform the reported event. Assuming the association of immediate future with strong volition or intention, it follows naturally that the quotative construction in the immediate future can be reanalyzed as a desiderative construction. Hence, from a construction like ‘I said “I will go,”’ we get the new desiderative construction ‘I want to go.’