Much literature has debated the syntactic status of quotations when they occur in structures with speech verbs (Buchstaller 2014; D’Arcy 2015; Munro 1982; Partee 1972 among many others). Cross-linguistic data shows that the verb say in quotative constructions can be analyzed as an intransitive verb in many languages. Munro (1982) discusses several features of quotatives that can be used to argue that speech verbs do not pattern as canonical transitive verbs.

The first feature discussed by Munro (1982) is the lack of morphological marking on quotatives in comparison with transitive structures. For example, in Chickasaw, objects are suffixed with the object marker -ã. Quotations, however, are incompatible with this marker as illustrated in examples (1) and (2) (Munro 1982: 303):1

A similar pattern is observed in the Uto-Aztecan language Cahuilla, where an agreement prefix occurs in transitive clauses (pe-) for both non-clausal (3a) and clausal complements (3b), but not in quotations (4). The central contrast is observed between (3b) and (4): in (3b) the clause henhichika ‘I want to go/I am going’ is a clausal object—the argument of the verb ’ayaw ‘want’—and the verb is marked with the third person singular object prefix pe-; contrariwise, in (4), the same clause is unmarked because it is a quotation.

Munro also discusses the restrictions on objects in constructions with speech verbs. In Godié, a Kru language, the speech verb lʌ ‘say’ introduces quotations (5) (Munro 1982: 304). But this verb is incompatible with non-clausal complements, such as pronouns and noun phrases.

Other languages discussed by Munro (1982) have restrictions on which speech verb can occur with quotations. In Hausa, the verb céè ‘say’ (6) is only compatible with quotative clauses while fàdi ‘say’ (7) can be used in non-quotative clauses. In examples (6–7), we see the same phrase (kalmà biyu ‘two words’) following these verbs, but only in (6), with the verb céè ‘say’, will this phrase be interpreted as a quotation:

It has also been shown that in English there are syntactic differences between objects of transitive sentences and quotative clauses in sentences with speech verbs (see Partee 1973, Munro 1982). For example, while quotations and objects of transitive sentences both allow for clefting (8), passivization (9) is not available for quotations (Vandelanotte 2008, 2012, as cited in Buchstaller 2014: 38–39). Conversely, while inversion is possible in quotative structures (especially in written genres), it is disallowed in transitive sentences (10):

Munro’s work was instrumental in providing cross-linguistic evidence and tests to argue that speech verbs may not display features of transitive verbs when they occur in quotative constructions. Two of these diagnostics are presented above: the absence of object marking in quotations in languages where objects are morphologically marked, as in Chickasaw and Cahuilla (Munro 1982, examples 1–4), and restrictions on the types of objects a speech verb might take as in Godié and Hausa (Munro 1982, examples 5–7). The other diagnostics discussed by Munro (1982) include: absence of agreement in languages where there is object marking on verbs; alignment of a ‘say’ verb with intransitive verbs in ergative languages such as Samoan and Yup’ik Eskimo (Munro 1982: 307); restrictions on passivization, as discussed in example 10 above for English; treating quoted materials as obliques as in Mojave (Munro 1982: 308); word order differences between quotative structures and transitive sentences (Munro 1982: 311); and overall similarities between quotative structures and intransitive clauses.

The goal of this paper is to discuss the status of the speech verbs in Yudja (Juruna branch, Tupi family; Brazil), and more specifically, to draw hypotheses about their argument structure based on their distribution in quotative and non-quotative structures. Tests to explore the transitivity of a verb are language specific. As such, I will first introduce some morphosyntactic characteristics of transitive verbs in the language, focusing on word order and encoding of objects (Section 2). After a presentation of materials and methods (Section 3), I will discuss three characteristics of quotative constructions that can be used to argue that they are not parallel to canonical transitive sentences in Yudja (Section 4). Then, I will discuss constructions with the reportative particles hi/ti, since hi often will co-occur with speech verbs (Section 5). I also discuss quotative frame constructions, when a quote occurs between two speech verbs (Section 6). I conclude the paper by summarizing the patterns observed and discussing how Partee’s (1973) analysis of speech verbs can account for speech verbs in Yudja (Section 7). By comparing morphosyntactic features of speech verbs and transitive verbs such as word order and object marking, I argue that quotations are not complements of speech verbs in Yudja.

Yudja belongs to the Juruna branch of the Tupian family. The Yudja number around 950 people, most of whom live in the Xingu Indigenous Territory.2 In this Section, I will present a brief overview of transitivity in Yudja by focusing on word order and on how objects are marked in transitive sentences.

As highlighted in the introduction, in many languages speech verbs do not present features of canonical transitive verbs. As such, in this section, I will provide features that characterize transitive verbs in Yudja to motivate the discussion about the argument structure of speech verbs, presented in Section 4. Most of the literature review presented in Section 2 is based on Fargetti’s PhD dissertation that presents a comprehensive overview of phonological and morphosyntatic features of the Yudja language (Fargetti 2001). Examples will be presented in the language’s orthography. (A guide on the orthography’s correspondence to the IPA can be found in Fargetti (2001: 53)). Examples whose author is not identified were elicited by me. Finally, examples are not marked for tone in this paper (tone is not marked in the Yudja orthography). However, for examples from other authors, please consult the original source for tone marking.

The data presented in this paper is from texts and elicitation tasks. Data from elicitation tasks were obtained via a mix of context-based translations (where a context is presented, and then a sentence is provided for translation) and grammaticality judgment tasks (where a grammatical sentence is modified and the speaker has to evaluate whether the modified version is possible in the language or not). The text data comes from three previously published texts by Fargetti (2001) and one storyboard elicited by the author, presented in Table 3.

The storyboard referred to in this paper was created by the Totem Field Storyboards Working Group.6 Storyboards are coherent stories that use overarching storylines to explore a target question/hypothesis (Burton & Matthewson 2015). I followed Burton and Matthewson’s (2015: 146) guide on how to work with storyboards in the field. First, I presented the story to the language consultant in Portuguese, with the support of pictures. The language consultant then took a few minutes to think about how they would tell the story in Yudja. After that, they told the story to another speaker of the language who did not hear the story in Portuguese; this was important to make the task communicatively relevant to the language consultants who participated in this task. The story was then transcribed, glossed, and translated with the help of the language consultant.

In the following sections, I will indicate the source of the examples by including the method or source below each example.

Yudja uses different strategies to encode quotations. We may encounter both examples of quotations introduced by speech verbs, as well as quotations that are not introduced by them. Consider (27):

In this example, we have two quotations: one that is introduced by a speech verb, hae ‘say’ (line b) as well as a quotation that occurs independently in the text, without a speech verb (line d). My focus here will be on the first strategy, since my goal is to explore the argument structure of speech verbs.

Four verbs will be considered: abï ‘tell’ and endu ‘ask’, hae ‘say/talk’ and a ‘say/speak’. As suggested by their translations, some of these verbs are translatable in more than one possible way by the language consultants and/or in the published sources. The verb endu, for example, was translated as ‘ask’ in some cases, but it can also be used to describe auditive perception (hearing, listening) and certain forms of somesthesis (e.g. feeling a burn).

Here I argue that quotations do not occupy the object position in constructions with speech verbs in Yudja. Three pieces of evidence will be provided towards this analysis: 1) the use of some speech verbs (hae ‘say/talk’, a ‘say/speak’) as intranstive verbs; 2) the presence of pronominal objects co-occuring with quotations in constructions with transitive speech verbs (abï ‘tell’ and endu ‘ask’); 3) the absence of dative marking on quotations occurring after speech verbs.

To facilitate the presentation and discussion of the data in this section, as in other parts of the paper, quotations are italicized in the first line of the example. In the translation line, quotations are presented inside quotation marks.

Among the common features that characterize quotative clauses, Munro (1982: 311) identifies the use of what I will label here as quotative frames. In Munro’s words (1982: 311), “in a variety of languages, for example, quotation clauses (direct or indirect) may be framed – both preceded and followed by an inflected form of ‘say (…)”. That is, when quotations occur between two speech verbs.11 This is illustrated with Chickasaw in (47):

In Yudja, quotative frames seem to be used as a strategy to delimit direct speech in the discourse. This type of strategy is used both in narratives and elicited data. Consider (48):

In (48), the speech verb hae ‘say’ is used to introduce the quotation. The quotation’s end is indicated by the verb a ‘say’.

Other speech verbs can also serve as quotation framers. Consider examples (49) where the quotation frame includes the verb abï ‘to tell’ before the quote and a hi ‘say + reportative’ after the quote:

The difference of frames between hae ‘say’ and abï ‘tell’ is that abï has the structure of a transitive verb and, as such, an object marker may occur after it in constructions with quotations, similarly to what we saw for endu ‘ask’ (34–37).

In elicitation sessions, speakers have also produced sentences with such frames. In examples (50) and (51), we observe the frame: hae ‘say’ + quote + a ‘say’. In Yudja, this type of lexical verb ‘framing’ is a specific feature of quotations, not observed in other constructions.

Example (51) also illustrates hae ‘say’ being inflected for mood; hala ‘will say’ is the irrealis version of hae (unmarked). The verb a ‘say’ does not alternate morphologically for realis/irrealis and, as such, it remains the same in (51). As expected, when one of the speech verbs that is part of the frame is negated, both of them are (52):

During elicitation sessions, it was also observed that, when quotative frames are used in Yudja, the addressee may also appear on both sides of the frame (in 53 u-be ‘1S-DAT’):

Similarly to hae ‘say’ and abï ‘tell’, endu ‘ask’ may also occur as part of a quotative frame (54):

In (54) we observe that endu ‘ask’ + quote + a ‘say’ forms a quotative frame; as observed in examples (53) with hae ‘say’ + quote + a ‘say’, here the addressee is also repeated.

Cross-linguistic evidence suggests that quotations are not necessarily complements of speech verbs. In this paper, I provided evidence that this is also the case for Yudja. The morphosyntactic evidence includes word order and absence of object marking. The materials analyzed provided overwhelming evidence that whenever speech verbs such as hae ‘say/talk’, endu ‘ask’, and abï ‘tell’ are used in quotatives, they precede the quotation. As discussed in Section 1, objects appearing after the verb (non-canonical position) are marked as dative in constructions with transitive verbs, regardless of whether they are non-clausal (15) or clausal complements (17). If quotations were objects of the speech verbs that precede them, we would expect that they would be marked as dative, like other objects in the language, which is not the case.

With transitive verbs, such as endu ‘ask’ and abï ‘tell’ it becomes very clear that the quotation has an independent status of the verb that introduces it. As illustrated for endu (34–37, 54) and abï (28, 32, 42, and 49), the argument of the verb is encoded by the third person dative morpheme te/he; as such, the quotation may not occupy the object position in the structure since it is already occupied.

Out of the analyzed speech verbs, the only one that appears after the quotation is a ‘say.’ This verb is most frequently followed by the reportative/evidential marker hi. Evidence from non-quotative structures suggests that this verb is intransitive (42). In quotative constructions, it is often used as part of a quotative frame with another speech verb (speech verb + quote + a or speech verb + quote + a hi, Section 6). As discussed in Section 6, a seems to be a delimitator that indicates the end of quotes (48). Additional evidence for this analysis comes from Gavião (-á followed by an evidential marker). Moore (2019) analyzes á as a clitic that occurs in the language to mark right hand boundaries. In Karitiana (Ferreira & Storto, this volume), ‘a ‘to do/to make’ is also analyzed as an intransitive verb that reports speech acts. As such, not only internal evidence from Yudja but cognates of this verb in other Tupian languages provide support to the hypothesis that this verb is best analyzed as an intransitive verb.

We can analyze Yudja quotatives in light of Partee’s (1973) analysis of demonstratives (see also Davidson 1968). To understand Partee’s proposal, we first need to discuss how demonstratives work. Partee (1973: 416) points out that this, that, here, etc.:

Partee illustrates this idea with examples such as (55):

In such examples, the demonstrative is referring to the gesture, but the gesture per se is not part of the sentence. From there, Partee argues that a similar pattern is observed in sentences such as (56):

The vocal noise is not part of the sentence in the same way the gesture is not.

Based on these observations, Partee argues that a similar analysis can be proposed for speech verbs such as say in English:

I propose that in Yudja, verbs such as endu ‘ask’ and abï ‘ask/tell’ exemplify something similar, since a third person dative object appears as the object of these speech verbs in quotatives and refers to the quotation. The quotation is, however, syntactically and semantically independent of the speech verb – in the same way the gestures and speech verbs are independent of each other in (56) and (57). As predicted by Partee (1973), quotations do not seem to have complement status in Yudja.