Complements of verbs of utterance and thought in Brazilian Portuguese narratives

This paper focuses on reported speech in Brazilian Portuguese oral narratives, examining the status of direct and indirect quotations. In particular, I address the following questions: Can the clause that carries the quotation be analyzed as an objet complement of the verb of saying and thought? What syntactic, semantic, and prosodic properties of utterance/thought verbs support such treatment? The analysis shows that utterance verbs strongly project what is to come, so that the quote can accurately be taken as the object complement of the matrix verb. The paper also argues for a scalar treatment of complementation in terms of syntactic integration with the utterance verb.

varies cross-linguistically, but there are also differences concerning the patterns of reported speech that are distinguished in a language.
Reported speech can be described as a device used in speech or writing when speakers or writers report the speech (or thoughts) of another person, or when they report their own words (or thoughts) considered at a time other than the time of speaking. Reported speech may be introduced by verba dicendi, or verbs of utterance, whose prototype is the verb 'say', followed by the reported clause. Within the domain of reported speech, we can distinguish direct and indirect quotation. 1 Besides the use of a complementizer in indirect quotes, 2 both constructions differ with respect to their deictic orientation. In direct quotation (DQ), the deictic center is that of the original utterance, i. e., the speaking time and the time when the quote was uttered are considered the same. Conversely, in indirect quotation (IQ), the deictic center is that of the moment of speaking, that is, the quote reflects the real time when it was spoken. This difference in deictic orientation determines the choice of pronoun reference (where 'I' and 'you' refer respectively to the original speaker and addressee, in DQ, and to the current speaker and addressee, respectively, in IQ), place deixis (the use of this, here and come, for instance, in DQ, and that, there and go in IQ) and verb tense (where the tense of the verb in the indirect quote is the same of the 'say' verb, while this must not be the case for the direct quote). In other words, in DQ the speaker adopts the deictic orientation of those whom he/she quotes, while in IQ the speaker uses him/herself as the spatiotemporal point of reference (see Klewitz and Couper--Kuhlen, 1999).
It is widely acknowledged that reported speech is a very special domain of language use (Clark and Gerrig, 1990;Chafe, 1994, inter alia). It raises special issues beyond the scope of the grammar of complementation, such as indexing, affect, and stance. Although several works have focused on quotation, most of them analyze either constructed data or excerpts from written texts, in special novels.
The main aspects discussed in the literature refer to: 1. the syntactic and semantic distinctions between direct and indirect quotation; 2. the relationship between a quotative verb and the quoted material; 3. the discourse functions of quotation. This paper addresses these issues by 1 There is also a third type, the so-called free indirect speech, which contains features of both direct and indirect speech. According to Coulmas (1986: 7), in free indirect speech "the author can freely invade other minds and relate events, utterances and thought as if he were a witness to a scene to which no witness can have access". This type of reported speech will not be analyzed in this paper. It is likely that a broader-based investigation of BP speech would show some other forms of quotation, which is beyond the scope of this paper. 2 The term "quotation" is used here to refer to both direct quote as well as indirect speech, following several authors. Cf. Longacre 1983, Mayes 1990, Li 1986, Coulmas 1986 analyzing direct and indirect quotations in Brazilian Portuguese oral narratives. In particular, I will look into the status of the quotation clause, aiming at answering the following questions: can the clause which carries the quotation be analyzed as an object complement of the verb of saying? Which are the syntactic, semantic and prosodic properties of utterance verbs that support such treatment? I will show that the verbs of speech in Brazilian Portuguese exhibit more transitive than intransitive characteristics. I will also make the point that the distinction between direct and indirect quotation is always clearcut. The facts presented support that quotation shares some of the features of a NP direct object although it is not a prototypical direct object.
I follow the assumption that grammatical relations such as subject and direct object have a prototype nature: the definition of the syntactic function does not depend on a single criterial feature. In Brazilian Portuguese the prototypical NP direct object is usually defined by a cluster of characteristic features: it simultaneously displays the formal property of postverbal position, the semantic property of representing the patient and the pragmatic property of being the focus (new information) of the clause. In the case of quotations, they share with NP direct objects the formal and pragmatic properties; as to the semantic feature of affectedness, quotations are effected rather than affected, as we will see later.
This paper is organized as follows. In Section 2, I briefly review the literature on quotation. Section 3 introduces the data. In Section 4, I present my findings in terms of types of verbs and their frequency of occurrence, tense and coreferentiality. The analysis of quotation clauses continues in Section 5, where I investigate their intransitivity (5.1) and transitivity (5.2) properties, and explore the issue of the size of quotation (5.3). The last section points out my conclusions.

Background
Various studies have focused on the syntactic and semantic distinctions between direct and indirect quotation in English from distinct research traditions. 3 Although dealing with very different kinds of data, and radically different ways of interpreting these, in general, scholars from distinct research traditions have assumed that DQ is used to reproduce the actual words of a speaker uttered in a former occasion, while IQ involves the paraphrase of one's words into the current speaker's speech. This distinction has effects on personal, spatial, and temporal deixis, as seen above. However, merely pointing out the syntactic and semantic differences between direct and indirect quotation is not enough to reveal the way they are used in real discourse.
There seems to be, in the literature, a general tendency to consider verbs of saying as intransitive (Partee, 1973;Munro, 1982). While some languages allow utterance verbs to occur with a direct quotation coded as a clausal object, in a number of languages the verbs of saying exhibit characteristics of intransitive verbs. 4 Munro (1982) argues that in some languages (English, for instance) the intransitivity of verbs of saying is much more apparent when the verb is used with direct quotations than it is when the utterance verb introduces indirect quotations and other complements (NP complements). Munro presents cross-linguistic evidence that verbs of saying "are often less than perfectly transitive", based on the fact that the quotations they specify behave very differently from ordinary direct objects or object complement clauses. She (1982: 316) suggests that "a good way to conceptualize the syntactic role of a 'say' sentence is with the 'say' verb and its complement taken together to form a single complex verb which can be used to characterize someone as well as to report an event." As a final observation, she points out that "the syntax of 'say' verbs is unique" in the sense that they not only behave differently from ordinary transitives, but also from ordinary intransitives.
This position is less than consensual, though. Romaine and Lange (1991) point out a syntactic difference between direct and indirect quotation. They argue that in direct speech the quotation is a syntactically independent utterance, marked in written language by inverted commas. In indirect quotation, however, the report is subordinate to the verb of saying and becomes a dependent clause introduced by the complementizer 'that'. Klewitz and Couper-Kuhlen (1999) claim that in English both direct and indirect quotes can be considered complements of the verb of saying, and thus form a subordinate clause which is the direct object of the verb of utterance. In indirect quotation, syntactic subordination to a verbum dicendi is signaled by the use of 'that' as a complementizer.
Although a number of works have investigated the functions of quotation in different languages (Wierzbicka, 1974;Larson, 1978;Halliday, 1985;Glock, 1986;Haiman, 1989, inter alia), most of them did not analyze natural data. It has been usually assumed that direct quotation is a verbatim reproduction of an actual previous utterance, while indirect quote is merely a rewording of a previous utterance. More recent works dispute this view, claiming that quotation is not verbatim reproduction, and assigning to it specific uses in discourse. Davidson (1968) and Partee (1973) 5 claim that a 'say' sentence with a direct quotation includes a demonstrative notion (as in John said [like] this/that), with the quotation corresponding to the physical gesture that usually accompanies ordinary demonstratives. In the same vein, Clark and Gerrig (1990) contrast direct and indirect quotation in terms of their discourse function. According to them, DQ is a type of demonstration, that is, it is an illustration by exemplification: you can demonstrate what a person did in saying something. On the other hand, IQ is a description of what a person did. So, demonstrations and descriptions perform distinct communicative roles, in that demonstrations depict their referents -what is demonstrated, while descriptions do not. In short, various studies have addressed the issue that what is presented as direct speech in a narrative does not correspond to the exact words uttered by the original speaker. Tannen (1989), for instance, claims that what appears as dialogue in a narrative or in a conversation was never spoken by anyone. Although I agree that DQ and IQ perform different discourse functions, the pursuing of this issue is beyond the scope of this paper.
In addition to the distinctions discussed above, direct and indirect quotations are also said to differ in terms of intonation, when language produced in interaction is analyzed. It is assumed that DQ is always separated from the verb of utterance by a pause, whereas IQ is not. Klewitz and Couper -Kuhlen (1999), however, point out that prosodic marking is not restricted to direct quotation, but occurs in indirect quotation as well. They show that prosodic marking of indirect quotation in conversational data is not rare, particularly in high-involvement talk.

Data
The database for this study comes from sixteen conversational narratives, extracted from the Corpus Discurso & Gramática (Furtado da Cunha, 1998). This corpus does not represent spontaneous conversation, but recorded speech as produced by the informants; turntaking by the interlocutor was limited to stimulating the speaker or changing the subject. It is, nevertheless, interactional, in that there is turntaking, although the narrator holds the floor much of the time.
The Corpus consists of spoken and written material produced by students who were homogeneously distributed across the social variables level of schooling, sex, and kind of school (public or private). Each informant was asked to produce five spoken texts and, based on those, five written texts, as follows: personal experience narratives, retold narratives, descriptions, procedural discourse, and opinion discourse. The subjects were four students from each of the following grades: 8th (from 13 to 16 years old), 12th (from 18 to 20 years old) and university seniors (above 23 years old). Their social class is controlled by the kind of school (public and private) they attend to.
This work is based on clauses collected from personal experience and retold narratives produced by students from the 8 th and the 12 th grades and university seniors, totaling about sixteen hours of recording. My database consists of a total of 341 quotations, from which 161 (47%) occurrences represent direct quotation, and 180 (53%) exemplify indirect quotation. Although some studies show a preference for the use of DQ over IQ, my data do not support this preference (see Mayes, 1990).

Findings
In Brazilian Portuguese, IQ is easily distinguishable from DQ: while the use of the complementizer que ('that') is obligatory in the former, it is absent in the latter. In fact, although rare, the complementizer que can be omitted in indirect quotes when the recipient of the verb of saying is introduced by the preposition para/pra ('to') and the verb of the quoted clause is in the infinitive form, as in (1). The other three occurrences of this type of indirect quotation in my data set are introduced by pedir 'ask for'.
(1) e ele não admite você é:: dizer pra ele <Q não beber and he NEG admits you um say to he NEG drink-INF numa ocasião dessa Q>. 6 in.an occasion of.this 'And he doesn't allow you to tell him not to drink on an occasion like that' 6 All the examples in this paper were taken from Furtado da Cunha (1998 There is a variety of verbs which can introduce reported speech, bringing about some semantic differentiation to the quote. Table 1 shows the occurrences of these verbs in direct and indirect quotation. As seen in Table 1, dizer ('say') is by far the most frequent verb used to introduce a quotation in my database (70% of the total), with few occurrences of the other verbs. Dizer corresponds to 81% of all occurrences of verbs of saying/thinking in direct quotation and to 60% in indirect quotation. It represents the prototypical verb of utterance in the class of verbs which frame others' speech, the class of quotation-ascribing verbs, as suggested by Munro (1982). 7 One reason for the unmarkedness of dizer and its prevalence is that it can be considered pragmatically neutral since prosody is usually used to indicate the way in which the utterance was spoken, whether it was a question, a request, a command, or any other speech act, as can be seen in the following example: 7 The high frequency of dizer in discourse is reflected in lexicalization patterns or formulaic expressions, which function as discourse markers. (2) minha mãe trouxe, trouxe, biquini e essas coisa, maiô, my mother brought brought bikini and these stuff swimsuit num sabe? ela trouxe maiô.
Aí disse: <Q ei, você vai Q>? NEG DM she brought swimsuit then said hey you go aí eu disse: <Q não, num vou não Q>. 8 then I said no NEG go NEG 'My mother brought a bikini and this stuff, swimsuit, you know. She brought a swimsuit. Then she said, "Hey, are you going?" Then I said, "No, I'm not going." In this excerpt, the speaker is talking about an excursion she took. The first quotation is a question, while the second is the answer she gave her mother. Both quotes are introduced by dizer. Occurrences like those are common in my database.
Thus, dizer reports speech without contributing some particular pragmatic effect, as it does not specify the mode of saying. In other words, it involves quotation without commitment (see Givón, 1980). Conversely, the prototypical quotation is a report of a speech event (see Clark and Gerrig, 1990). Other quotative verbs are more semantically and pragmatically specific, and perform different rhetorical functions, denoting a lexicalized speech act. For instance, the function of questioning is associated with pedir 'ask for'. In that sense, the choice of the utterance verb represents an important communicative option. Table 1 also allows interesting observations. First, notice that some verbs did not occur in indirect reported speech (reclamar, chamar, indicar). As to chamar ('call') and indicar ('nominate'), their absence from indirect quotation has semantic-pragmatic motivation. With these meanings, these verbs cannot be followed by a clause introduced by the complementizer que ('that'). Since these verbs denote a lexicalized speech act, they can only be used in DQ, where the quote represents the rhetorical functions performed: ( from.the church of.the Satélite 'Then, soon after, a guy arrived, who didn't ask me whether I wanted it or not, and he nominated, "I nominate the name of Gerson, and all, from the church of Satélite." In addition, pensar ('think') has a higher frequency of occurrence in IQ (n = 28) than in DQ (n = 2), while contar ('tell') only appears in IQ (n = 17). The use of the verb pensar shows that the act of reporting includes not only speech activity, but also unspoken mental activity (for a discussion of reported thought, see Chafe, 1994).
It is interesting to notice that when a question is reproduced in indirect quotation the intonation is lost, but the use of the verb perguntar ('ask'), together with an interrogative pronoun in the quote, compensates for this loss (Jespersen, 1965). In my data the 7 occurrences of quotation introduced by perguntar are preceded by different complementizers: se ('whether'), por que ('why'), onde ('where'), como ('how'), and quem ('who'): (5) se alguém pergunta pra gente <Q se você viu aquele filme Q>, if someone asks to people whether you saw that movie 'If someone asks us whether you saw that movie' From the syntactic point of view, these utterance verbs tend to select the same kind of so-called complement clause. In Brazilian Portuguese, reported speech has three complement types: a simple clause, i.e., the complement has the same syntactic form as a main clause (direct quotation), que-clause and infinitive clause (indirect quotation).
My data set also register several occurrences of direct quotation in which the verb of saying is missing: 9 (6) aí vinha um caminhão e descarregava lá, na calçada. Aí then came a truck and unloaded there in.the sidewalk then maínha: <Q ei, tire esse negócio daqui Q>. 10 mom hey take this thing from.here 'Then there came a truck and it unloaded there, on the sidewalk, and mom, "Hey, take this thing away from here' (7) o morador tinha saído. aí ela ofereceu o apartamento. Aí the tenant had left then she offered the apartment then mãe: <Q ah, tudo bem. eu fico com o apartamento. mother ah everything well I keep with the apartment fico pagando o aluguel pra você Q>. keep paying the rent to you 'The tenant had left. Then she offered the apartment [to my mother]. Then mother, "Ah, it's all right. I'll keep the apartment. I'll keep paying you the rent." (6) and (7) show some features that help to identify the bracketed utterances as direct quotation. First, there is subject change: from 3 rd person (maínha 'mom') to 2 nd person singular ('you', omitted but recoverable from the verbal morphology) in (6), and from 3 rd person (mãe 'mother) to 1 st person singular (eu 'I') in (7). Second, there is also a change in verb tense: from past tense to present tense in both cases. Third, the use of particles such as ei and ah at the beginning of the reported utterances indicate that the current speaker is now quoting, and that these particles were supposedly uttered by the original speaker in the event being reported. Notice also the use of the narrative marker aí ('then') before the reference to the speaker being quoted. In addition, prosody helps to indicate that a quotation is meant, since both quotes are preceded by continuing intonation. Thus, the report verb can be omitted, but pronouns and tense are used in a manner appropriate to DQ. This kind of construction seems to be associated with a performative or enacting function of reported speech (see Golato, 2000).
In quotation, the choice of verb tense, like the choice of spatial and temporal orientation, is determined by the point of reference of the quote. As to the tense in which the verb of saying appears in my data, Table 2 exhibits the number of tokens for this feature: The verb which introduces the quotation is predominantly in the past tense (282/341 = 83%), both in DQ and in IQ, but it can also be in the present tense (53/341 = 16%), with very few occurrences in the future. The high frequency of the past tense has to do with the function of reporting a previous situation or event. 11 The use of the present tense in reported speech is an attempt to transfer the immediacy of the present to the narrative. Instead of reporting a past event in the past tense, the speaker uses the present tense as a device to pretend that the event is taking place at the moment of speaking. This strategy is mainly used at points in the narrative where the speaker is highly involved in what he/she is recounting, and has the effect of dramatizing what is being described, making it seem more vivid (see Jespersen, 1924;Schiffrin, 1981;Chafe, 1994;Croft, 1998), as in (8) below, where the speaker reproduces her dialogue with a former boyfriend. The use of time adverbials and the context in which the clause is uttered also make it clear that the reported event took place in the past.
(8) aí ele: <Q num vai comer não, você? digo: <Q não, é then he NEG go eat NEG you say no is porque eu tô sem fome Q>. because I am without hunger e eu morrendo de fome, sabe? Tinha saído do colégio. and I dying of hunger DM had left from. the school 'Then he, "Aren't you going to eat?" I say, "No, it's because I'm not hungry." And I was starving, you know, I had just left school' The example in (9) illustrates the use of the habitual (or generic) present tense, which conveys a timeless situation or state of affairs: (9) a moral da história é, quando o povo diz: <Q ah, tenha the moral of.the story is when the people say ah have paciência de Jó Q>, é porque Jó era o nome do cara. patience of Job is because Job was the name of.the guy 'The moral of the story is that when people say, "Ah, have Job's patience", it's because Job was the guy's name' As for the future tense, in the data examined it is always expressed by means of a phrase formed by the auxiliary verb ir ('go'), inflected for present tense, followed by the main verb dizer in the infinitive. 12 In the excerpt in (10) the narrator is talking about a plan she worked out to deceive the principal of the school where she studied at that time. This fragment is interesting in that it contains direct quotes within another direct quotation, introduced by the verb combinei ('agreed'). I will come back to this issue later. Notice that all the verbs within the extended quotation are in the immediate future: (10)  the others who are in.the auditorium because é pra me levar pro hospital Q>. is to me take to.the hospital 'Then I arranged with my friends, "Look, two of you are going to the classroom to attend the commemorative date. I'm going to begin crying and say that I am sick. One of you is going to arrive and call the principal. You are going to say that I am sick, so that … You're going to call the others who are in the auditorium, so that I can be taken to the hospital." It is usually assumed that the tense of the verb in IQ, but not in DQ, is the same of the verb of utterance (Li, 1986). This can be accounted for by the fact that in DQ the deictic center is that of the reported utterance, while in IQ the deictic center is that of the moment of speaking. Thus, in the first the verb tense is the same as it would have been in the original utterance, while in the latter the verb in the quote is often in the past, as the verb of saying (predominantly past, but also present and even future, in my corpus, for both types of quotation. Cf. Table 2). My data corroborate this assumption: in direct quotation, in 121 out of 161 occurrences (75%) the verb of saying and the verb in the quote have different tenses, whereas in indirect quotation in 160 out of 180 tokens (89%) the verbum dicendi and the reported verb are in the same tense.
In DQ we registered 6 occurrences (4%) in which the quote is not a clause, but an NP, (5 instances, of which 4 are either a vocative or an exclamation, as in ( (11), bellow) and an interrogative pronoun (por que 'why'). In these occurrences, the utterance verb is used as a regular transitive verb, followed by a NP which functions as its direct object. The ability of an utterance verb to also occur with a NP object is evidence that the quote is a clause in a grammatical relation with its predicate. In IQ, only two quotes (1%) are expressed by the complementizer que followed by sim 'yes', and by não 'no', as in (12): (11) era o filho dele, muito sujo, aí chamando: <Q papai Q>, né?
was the son of.his very dirty then calling daddy DM 'It was his son, very dirty, then calling, "Daddy", right.' (12) o cara disse que ela podia sair de lá, mas ela disse the guy said that she could leave from there but she said <Q que não Q>. that no 'The guy said that she could leave, but she said that no." Example (12) is interesting in that the complementizer que is followed by the general Portuguese negator (não) 'not', with elision of the contextually predictable elements of the embedded clause (ela podia sair de lá).
As to verbal mood, although the great majority of quotes are in the indicative, the subjunctive does occur in nine indirect quotations. The preference for the subjunctive seems to be conditioned by the verb, but some verbs admit both. With the exception of dizer (n=5), all the other occurrences of subjunctive are required by the meaning of the quotative verb, pensar 'think' and pedir 'ask for', inflected in the past tense. All these clauses are irrealis, reporting a future event, as in (13) and (14).
(13) eu pensei <Q que eu fosse morrer Q>, sabe, quando o I thought that I go-SUBJ die DM when the colégio inteiro correu pro laboratório pra ver o que tinha sido. school whole ran to.the laboratory to see the what had been 'I thought I was going to die, you know, when the whole school ran to see what had happened' (14) dá esse almanaque e diz para ele <Q que aposte em tudo Q>.
gives this almanac and says to he that bet-SUBJ in everything 'He gives him this almanac and tells him to bet in everything' Some studies have pointed out the differences in the distribution of pronominal expressions, either null or explicit, as a criterion for distinguishing between direct and indirect quotation (Munro, 1982;Li, 1986;Mayes, 1990). In DQ, the 1 st and 2 nd person pronouns refer to the original speaker and original addressee, respectively, while the 3 rd person pronoun refers neither to the original speaker nor to the original addressee. On the other hand, in IQ the 1 st and 2 nd person pronouns refer to the current speaker and current addressee, respectively, whereas the 3 rd person pronoun is used according to the general rule of coreferentiality in the language in question. In Brazilian Portuguese, subject personal pronouns are usually omitted, since the verb is inflected for person. Thus, even if the pronoun is not expressed, the verbal morphology indicates which person is being referred to. My data support the criterion of pronoun distribution, as the following examples show: (15) aí seu Carrilho disse: <Q não, ainda não fui atendido Q>. then mister Carrilho said no yet NEG was served 'Then mister Carrilho said, "No, I wasn't served yet." (16) aí minha tia disse: <Q o que que você tava conversando com then my aunt said the what that you were talking with ele Q>? aí eu disse: <Q nada, ele tava me contando uma história Q>. he then I said nothing he was me telling a story 'Then my aunt said, "What was that you were talking with him? Then I said, "Nothing. He was telling me a story." In (15), the verbal form fui ('was') is inflected for 1 st person singular, and refers to the original speaker, Mr. Carrilho. In (16), você ('you') is co--referential with the original addressee, which happens to be the current speaker, and ele ('he') refers to a third person, who is neither the original speaker nor the original addressee. making an experiment 'When the teacher arrived, he saw that it was me who had done the job. Then he said that it was him who was making an experiment.' In (17), the object pronoun me ('me') is co-referential with the current speaker. Example (18) shows that the second person pronoun você ('you') refers to the current addressee. In (19), the pronoun ele ('he') refers to the NP o professor ('the teacher'), who is not the current speaker or addressee.
Quotation seems to be predominantly used to report other's utterances, both in direct and indirect quotation. In DQ, only 10% (16/161) of the occurrences refer to the current speaker (that is, 1 st person singular), while in IQ 9% of the quotations (17/180) refer to the current speaker. It is worth noting that in direct quotation only when the subject of the quotative verb is the 1 st person singular, we do have the same subject in the quoted clause. Consequently, in direct quotation the subject of the quotation clause is different from the subject of the quote in 90% (145/161) of the cases. As to indirect quotation, we have a slight different picture: the subject of the quoted clause is different from the subject of the quote in 71% (128/180) of the tokens. When the quotative verb and the quote do have the same subject, they may be the 3 rd person singular (32/180 = 18%), the 1 st person singular (16/180 = 9%), or even the 2 nd person singular (only one occurrence).
To summarize, as to the tense of the quotation vis-à-vis the tense of the verb of utterance in my database, we can say that higher temporal cohesion, as in IQ, is correlated to the grade of integration between the clauses. As to coreferentiality, the more coreferential is the quote subject to the main clause subject, the more integrated are the clauses, so that the subject of indirect quotation tends to be grammatically and pragmatically controlled by the subject of the dicendi verb. This control is an expression of conceptual integration, in terms of Givón's (1990Givón's ( , 1995 iconical principle of proximity: concepts that are closer together conceptually or cognitively are coded closer together. The greater the level of conceptual integration between the verb and the complement clause, the more advanced the process of clause combining. The focus on the status of the verb in the main clause, examining the range of tenses and moods it can take, the occurrence or absence of the complementizer, the absence of a main clause verb in some constructions with direct speech, is relevant because these are some of the properties usually pointed out as indicators of the (in)transitive nature of reported speech, as we will see in Section 5.

Transitivity versus intransitivity of quotation clauses
The claim that verbs of utterance and thought are intransitive is generally based on the fact that the complements of these verbs behave differently from other types of direct object arguments. The deviant behavior of quotation clauses is presented as evidence that the clause introduced by a verb of saying plays some role other than that of the direct object, although this role is never identified in the literature. In this section, I will survey both the intransitive (5.1) and transitive (5.2) characteristics of quoted clauses. I will make reference only to those characteristics which apply to direct objects in Brazilian Portuguese. 13 Before we proceed, a brief review of the notions "complementation" and "embedding" is in order here. Complementation refers to a set or series of complements that a verb may or must take; on the other hand, embedding indicates the inclusion of one clause or sentence in another. A prototypical complement clause is usually defined as a clause that functions as an argument of a complement-taking predicate (CTP), so that complements exhibit a grammatical relation with their predicates, as their subject or object (see Noonan, 1985;Payne, 1997;Dixon, 2002). In other words, a complement clause is embedded within the main clause. If a complement clause functions as the direct object of a predicate, as it seems to happen with quoted clauses, it is supposed to occupy the same position within the main clause as the direct object argument. Since Brazilian Portuguese is a SVO language, the complement clauses of verbs of utterance are expected to immediately follow the verb. In the data analyzed here, there was no instance of inversion. Complement clauses are usually described in terms of the properties they exhibit. Thus, Payne (1997: 314) ascribes two characteristics to prototypical finite complements: (i) they carry their own tense and aspect; (ii) they express their subjects directly; subject reference is not restricted to that of the matrix clause. As Payne puts it, verbs of utterance and cognition represent the typical matrix verbs for finite complements, in that, except for the presence of the complementizer, the complement clause can stand alone as a complete and understandable utterance. The mention to the complementizer makes it clear that Payne is referring to indirect quotation, as evidenced by the examples he provides. He does not discuss the grammatical status of direct quotation. 13 Since lexical direct objects in Brazilian Portuguese are not morphologically marked, I will not take marking into consideration here, although this aspect is relevant to distinguish between NP direct objects and complements of verbs of saying in other languages (see Munro, 1982). In languages where the direct object is morphologically marked, the quotation of a 'say' verb does not always show this mark (see Hintz, 2004).
According to Noonan (1985: 43-4), complements can vary in form, even within a single language. A complement type can be identified by (i) the morphology of the predicate; (ii) the syntactic relation it has with its arguments; and (iii) the external syntactic relations of the complement construction as a whole. He also addresses the issue of the independence of complement clauses, stating that a sentence-like complement type is "one that without its complementizer has roughly the same syntactic form as a main clause." (p. 49). Further, Noonan (1985: 55) argues that parataxis may be used in complementation: in this case, "the matrix clause and the paratactic complement each constitute clauses which could stand by themselves as independent sentences with approximately the same meaning." From Noonan's observations, we may conclude that he would consider both indirect quotation (complement type with a complementizer) and direct quotation (paratactic complement type) as complement clauses.
In his paper on the typology of complements, Givón (1980) proposes a hierarchic semantic dimension called "binding" to account for the form of the complements of CTPs. The purpose of his study is to show that the syntactic structure of these complements is largely predictable from the position of CTPs on the binding scale, reflecting a systematic correlation between the semantics and the syntax of complementation. Among the most important parameters "involved in determining what is coded like an independent/main clause" cross-linguistically, he points out: (i) the degree of independent marking of the embedded clause subject from the subject of the main clause; (ii) the degree of independent marking of the embedded verb tense-aspect--modality from the verb of the main clause (p. 337). According to Givón's classification, the verb 'say' occupies the very lowest point on the binding scale, involving "quotation with no emotional commitment whatever" (p. 346). As to the direct or indirect complement quote of 'say', he argues that the first is the lowest point of the scale, "exhibiting semantically the least dependency between main and subordinate clause, and syntactically the least deviation of the subordinate clause from the neutral main-clause pattern." In short, two aspects are generally mentioned as central to the evaluation of the dependent/independent nature of complement clauses: the possibility of expressing their own tense and aspect, on one hand, and the possibility of expressing their own subject. The notion of independence ("can stand alone") seems to be crucial in determining the syntactic function of quoted clauses, either direct or indirect. However, in the domain of quotation, there seems to be a mismatch between the concepts of complementation/embedding, on one hand, and semantic-syntactic independence, on the other hand. From the viewpoint of embedding, verbs of utterance are largely regarded as typical matrix verbs for complement-clauses, which means that the quote is the direct object of the verb of saying. By the same token, one can claim that the reported clause is semantically linked to the matrix verb. From the perspective of syntactic-semantic independence, the quote is considered to be independent from the main clause (it can "stand alone"), and so does not appear to bear any semantic and grammatical relation with the predicate of the matrix clause.
We can say that the major distinction between direct and indirect quotation has to do with the nature of the linkage that binds the matrix and the complement clause. In the case of direct quotation, the clauses are combined through parataxis: they are interdependent, although there is minimal syntactic integration between them, as they do not show any overt linkage markers. On the other hand, indirect quotation reflects an instance of subordination, where there is dependence and maximal integration. Thus, following Hopper and Traugott (2003), in terms of combinations of the features +/-dependent, +/-embedded, in DQ (parataxis) the clauses aredependent and -embedded, whereas in IQ (subordination) they are +dependent and +embedded. In this respect Givón (1990: 826) has suggested that there is a cognitive form-function parallelism expressed as: "The more two events/states are integrated semantically or pragmatically, the more will the clauses that code them be integrated grammatically."

Intransitivity properties of quotation clauses
In this section, I examine some of the properties usually put forward as indicators of the intransitive character of quotation: subordinating marking, intonation, and affectedness. Munro (1982) shows that in several languages the quotation related to a verb of saying appears completely unmarked (except as it would be in isolation), even in cases where all other complement-taking verbs require overt subordination marking, or at least the possibility of it, as in English. Noonan (1985) notices that the function of a complementizer is to identify the entity it is associated with as a complement, but he also admits that some complement types may have no complementizer at all, as non-finite complement clauses, for example. As mentioned above, in Brazilian Portuguese, the complementizer que 'that' cannot appear before a direct quotation (20), but it is obligatory in indirect quotations (21) (cf. example (1)): (20) aí ela disse: <Q quer sair comigo Q>? then she said want go.out with.me 'Then she said, "Do you want to go out with me?"' (21) o médico disse <Q que ela não podia se machucar Q>. the doctor said that she NEG could REFL hurt 'The doctor said that she couldn't get hurt' Thus, the presence of a complementizer before indirect quotations challenges a unified treatment of both direct and indirect quotation as far as their grammatical nature is concerned. 14 Intonation is held as an important factor in assessing the integration of clauses, so that the absence of an intonation break between two clauses is a reliable indicator that one of them is integrated with the other (see Lehmann, 1988). Several linguists have claimed that the presence of pause between a verb of utterance and the quoted clause is evidence that the quotation is not dependent of the quotative verb. Noonan (1985: 112) remarks that almost all languages distinguish direct from indirect quotation by means of intonation: while there is typically a pause before and/or after the DQ, from the viewpoint of intonation ID is treated like any other complement. Here, again, DQ behaves differently from IQ. As Hopper and Traugott (2003) puts it, the different intonation contours involved in paratactic clause combining, as in the case of direct quotation, may function exactly like overt grammatical markers.
However, the criterion of intonation is not useful to decide on the (in)transitive behavior of Brazilian Portuguese quotes, since in spoken discourse while a pause separating dizer and the direct quotation is often present, it is also allowed between dizer and the indirect quotation (22), on one hand, and between any transitive verb and its nominal direct object, on the other hand (23): (22) e disse: <Q que ela era uma prostituta Q>. and said that she was a prostitute 'And he said that she was a prostitute' (23) eles descobriram: o local lá, né? they found the place there DM 'They found out the place there, right?' Thus, in Brazilian Portuguese verbs of utterance/thought and their nominal/clausal complements may be in the same intonation unit, or they may be separated, for no apparent syntactic reason. We have here the two types of situation for complement-taking verbs and their complements described in Ono and Thompson (1995: 239): those that are realized as one intonation unit, and those that are not. Notice, however, that the pause separating a verb of saying and its quote seems to perform different discourse functions, whether the quotation is a direct or an indirect quote. In the first case, the intonation break introduces the original speaker's speech, orienting the hearer to the shifting of characters in the narrative: from the narrator to the speaker being quoted. In IQ, on the other hand, the pause functions as a kind of discourse filler, reflecting the original speaker's attempt to elaborate what he is going to say, as can be seen in (24) below. In this example, the presence of the discourse marker né ('right') immediately after the verb of utterance reinforces this interpretation. Thus, we could say that in DQ the intonation break plays a textual role, whereas in IQ it serves a cognitive function.
(24) aí ele disse, né: <Q que a mulher também dele foi, then he said, DM that the woman also of.he went tava correndo com medo do cachorro Q>. was running with fear of.the dog 'Then he said that his wife also went, was running because she was afraid of the dog' It turns out, then, that intonation alone cannot be used as a reliable criterion to determine the syntactic nature of quotations. Hopper and Thompson (1980, p. 251) propose to "isolate the component parts of the transitivity notion, and study the ways they are typically encoded by languages." According to them, each component involves a different facet of the effectiveness with which the action described is transferred from one participant -an agent -to another -a patient. In their formulation of transitivity, two parameters characterize a high transitive object: affectedness and individuation. Affectedness refers to the degree to which a patient is affected by the transfer of the action; the component of individuation refers both to the distinctness of the patient from the agent and to its distinctness from its own background. Hopper (1985) highlights the importance of the degree of affectedness of the object in the measurement of prototypical transitivity. He acknowledges, moreover, that it is impossible to "assign a unitary, invariant semantic property to the verb-object relationship", because some objects "are not changed but are instead brought about by the action of the verb" (pp. 67-8). Following established tradition (Jespersen, 1924;Quirk and Greenbaum, 1973), Hopper calls these cases effected objects. In this sense, the direct object of a verb of saying is not changed but created by the action of the verb, and the component that applies to it is effectedness.
Affected and effected objects share the grammatical property of occupying the syntactic slot reserved for objects, but they can be distinguished by other grammatical features, such as: (i) their contribution to the degree of clausal transitivity (effected objects being less transitive than affected objects), and (ii) the referentiality of the object (effected objects usually being non-referential). According to Hopper, verbs of utterance form the largest group of effected verbs in English and presumably in any language. He assumes, then, a correlation between Effectivity and Intransitivity: the clause with an effected object (E-clause) will always be less transitive than one with an affected object (A-clause).
As seen, the direct object parameters 'affectedness' and 'individuation' do not fully apply to complements of verbs of utterance. 15 On one hand, as propositions, quotations lack individuation since they are not proper, animate, concrete, singular and referential; on the other hand, although they are not strictly affected, they are effected by the act of saying performed by a volitional agentive subject, exhibiting, therefore, some degree of transitivity. In short, the verbs of saying/thought are not prototypically transitive, to the extent that an effected object is lower in transitive than an affected one.

Transitivity properties of quotation clauses
This section looks into some of the transitive features that apply to verbs of utterance: morphological type of object, word order, and number of arguments. Munro (1982) cites the restrictions on possible objects of utterance verbs as an intransitive characteristic of quotation. Thus, some languages allow 'say' to take a quotation as a clausal object, but not a pronominal object like 'it' or a more concrete NP. In Brazilian Portuguese, however, dizer can occur with either a pronominal (25) or a NP (26) object 16 , as in: (25) foi a única palavra que eu disse.
was the only word that I said 'It was the only word that I said' (26) a gente não disse o nome dela. the people NEG said the name of.her 'We didn't say her name' Word order differences between quotations and other object clauses are often pointed out as an intransitive characteristic of quotation-ascribing verbs: (i) in many languages, quoted clauses may be both preceded and followed by the verb of utterance, while other verbs do not allow this kind of "frame" construction; (ii) some languages allow extraposition of object clauses, but not of quotation; (iii) in the Yuman languages, a direct object clause may appear center-embedded between the subject and the verb, while a complement of 'say' never appears in this position.
In Brazilian Portuguese, verbs of utterance may be used as a three--argument predicate, with the recipient of the speech act expressed in an oblique phrase in the matrix clause. This coding is possible whether the direct object is a NP (27), or a complement clause (28).
(27) você diz uma coisa pra mim, você acha que eu vou acreditar? you say one thing to me you think that I go believe 'You say something to me, do you think I'm going to believe it?' (28) às vezes eu dizia pra minha mãe <Q que tinha aula no some times I said to my mother that had class on.the sábado Q>. Saturday 'Sometimes I said to my mother that I had classes on Saturday' The overt expression of the addressee is more common in indirect quotation (n=21) than in direct quotation (only one occurrence). In 8 examples, the referent of the indirect object of the matrix clause also plays the role of subject in the quoted clause, which shows the integration between the main clause and the quote: (29) ele sempre disse pra mim <Q que eu era muito fria, assim, he always said to me that I was very cold like calculista Q>. emotionless 'He always said to me that I was very cold, like, emotionless' The sharing of elements -predicates, tense and aspect, participantsbetween the main clause and the quotation reflects their semantic and syntactic interlacing (see Lehmann, 1988). Furthermore, object complement clauses that have the same subject as the main clause are considered more dependent than the ones with different subjects. As seen in Section 4, direct quotation is opposed to indirect quotation concerning identity of tense. In DQ, in 25% of the occurrences the main clause and the quote are in the same tense, whereas in IQ the percentage is as high as 89%. With respect to subject identity, however, both types of quotation show the same pattern: in DQ, the subject of the main clause and the quote is identical in 10% of the cases, whereas in IQ the percentage of same subject is 28%. It turns out, then, that neither of these parameters is useful to determine the transitivity of utterance verbs in Brazilian Portuguese, if unified treatment of both direct and indirect quotation is in order.

The length of quotation clauses
This section aims at addressing the issue of the "length" of a quotation. In this sense, some syntactic as well as pragmatic aspects must be looked into, such as: when does a quote start and stop? Or, how does the addressee know the quote has finished?
In the narratives examined here, some quotes, both direct and indirect, are syntactically complex, as long stretches of narration can be presented as the complement of an utterance verb. Let us see some examples of IQ, first: (30) aí ele também correu, disse <Q que num queria mais ela, que then he also ran said that NEG wanted more she that ela num podia casar com ele, que num queria saber mais dela she NEG could marry with he that NEG wanted know more of.she que sumisse da vida dele Q>. that disappear-SUBJ of.the life of.he 'Then he ran too, he said that he didn't want her any more, that she couldn't marry him, that he didn't want to know about her any more, that she disappeared from his life' (31) Batman se apaixona pela Mulher Gato e os dois, assim, Batman himself fell.in.love by.the Woman Cat and the two like vivem um romance, ela sem saber, ela pensando live a romance she without know she thinking <Q que era um rico milionário, que era assim um sujeito normal Q>.
that was a rich millionaire that was like a guy normal 'Batman falls in love with the Cat Woman and the two, like, live a romance. She doesn't know, she thought that he was a millionaire, that he was like a normal guy' In the excerpts above, the verb of utterance/thought is followed by several complete clauses with one finite verb each (4 in the first example and 2 in the second), all of them introduced by the complementizer que, which syntactically and semantically links them to the verb of the matrix clausedizer in (30) and pensar in (31). Instead of repeating the same verb, the speaker economically employs it once, and then attaches to it, by means of que, all the utterances that are being reported. In terms of prosody, the quotation clauses are separated from each other by continuing (or non-final) intonation, which signals the relation between them; only the last quote has final intonation. Besides syntactic and prosodic completion, notice that the complex quotation constitutes a complete pragmatic action (see Ford and Thompson, 1996), in that it constitutes a whole reporting act. This way, prosody, syntax and pragmatic work together to help identify each clause within the complex quotation as the object complement of the speech verb in question. Both production and processing considerations are involved here. From the viewpoint of narrative production, the speaker signals the end of a reporting event by the convergence of syntactic, prosodic and pragmatic cues; from the perspective of narrative processing, the hearer is able to recognize that such cues indicate the completion of significant units.
In direct quotation, the complement of the quotative verb can be even longer and more complex than in indirect quotation, as the examples below demonstrate.
(32) eu disse na cara dele: <Q eu não posso dizer que é mentira I said in.the face of.his I NEG can say that is lie uma coisa que aconteceu. depois você vai saber a verdade a thing that happened later you go know the truth e vai ser pior, vai doer mais. Eu num vou dizer and goes be worse goes hurt more I NEG go say só pra aliviar Q>. eu sei que ele chegou de oito horas. only to relieve I know that he arrived at eight o'clock 'I said in his face: "I can't say that something that happened is a lie, because later you're going to know the truth and it's going to be worse, it's going to hurt more, I'm not going to say anything just to make you feel better". I know that he arrived there at 8 o'clock.' In (32) the speaker reproduces a dialogue she had with her boyfriend. She begins the passage with the verb dizer, and then produces a sequence of complex clauses that represents the quoted material. In this excerpt there is an indirect quotation introduced by dizer, a relative clause, some coordinate and paratactic clauses. As far as prosody is concerned, the clauses are set apart by continuing intonation, except for the last one, which stands out with final intonation, and is a kind of resumption of the dialogue. because you are just a salesman then the doutor Carrilho era muito é:: um homem muito também, polêmico. doctor Carrilho was very um a man very also polemic 'Then doctor Carrilho introduced himself and said "look, 'cause I am doctor Carrilho and you get to know that is not going to be like that, see, my friend? You are just a salesman here. You should treat people equally, without discrimination. As a salesman, this was not a reasonable attitude of yours. If you knew who I was, I would have had another treatment, a differentiated treatment, but as you" how can I say? "you have compared me to any other customer, the treatment was different. I want to speak with the owner of this store here, the manager, the actual representative, because you are just a salesman". Then doctor Carrilho was also a very polemic man.' In the fragment in (33) the current speaker quotes a third person original speaker, adopting his deictic orientation. In this case we have different types of clauses, as well as different patterns of intonation, so that neither syntax nor prosody is sufficient to predict the size of the quote, although there is thematic continuity within the complex quotation. This fragment presents some characteristic features of DQ, such as: (i) the use of vocatives and discourse markers at the beginning and at the end of the complement clauses (olhe 'look', viu, meu amigo? 'see, my friend?'), reflecting the speaker's attempt to capture how people really talk. Such particles help to indicate to the addressee that the speaker is quoting; (ii) the interruption of the quotation (como é que se diz? 'how can I say?'), so that the current speaker may look for the right word to accurately convey his meaning. In this regard, notice the repetition of the pronoun (me, me, me) before the suspension of the quote, and the truncation after it (qualifi/comparou 'qualify/compared'). Thus, in direct quotation the speaker's involvement is more clearly manifested (see Chafe, 1982).
With respect to the size of the quotation, then, the comparison between direct and indirect quotation demonstrates that, once again, both kinds of quotation show different behavior. In IQ, the complement clauses of the utterance verb in the matrix clause are more easily recognizable, syntactically and prosodically as well: they are introduced by the complementizer que, and are constrained by continuing intonation. Besides, the length of the quotation seems to be shorter. In DQ, on the other hand, the degree of complexity is remarkable, as long stretches of quotation can follow the utterance verb. There are no such neat clues to where each clause finishes: the quotation may have complex sentences, with embedded material of their own, and it may exhibit different intonation types within it. Nevertheless, the addressee has no difficulty to identify the end of the quotes due to semantic-pragmatic and prosodic factors. The clause after the quotation introduces another topic, exhibiting some changes in the deictic orientation (from 1 st to 3 rd person singular, and from present to past tense). Notice also the occurrence of some elements that help to mark the end of the quotation: in (32), the use of the epistemic expression at the beginning of the next clause (eu sei que 'I know that'), and in (33), the narrative marker aí 'then'. On the pragmatic level, in both fragments the complex quotation, as a whole, represents a pragmatically complete action.
As to prosody, final intonation signals the end of the long stretch of quotation. Recall that my database is primarily monologic in nature, with little turntaking, so that backchannel reactions are not available to give evidence that the addressee knows the speaker has completed his/her speech. As we have seen, in DQ the amount of reported material can be so high that it does not seem reasonable to consider such stretches as a single object complement of a verb of saying (for a similar position, see Genetti and Slater, 2004). Instead, we have an utterance verb in the matrix clause and several paratactic object complements attached to it (see Noonan, 1985). Each quotation clause contains a fully inflected verb with its own subject. So, we can say that each complete clause in this long stretch of discourse after the verb disse in (33) functions as a direct object of this verb.
The occurrence of several complement-clauses after a single verb of utterance is not restricted to Brazilian Portuguese. Quirk et al. (1985Quirk et al. ( : 1022 assert that "direct speech may extend over many sentences. The reporting clause is then usually positioned within the first sentence." Then, it seems likely that other languages show the same behavior, if we examine natural spoken data.

Conclusion
This section provides a summary of key points raised in this paper.
The structural relationship between the reporting clause and the direct/indirect quote poses some interesting analytical issues. An important difference between direct and indirect quotation has to do with their independent status. To what extent can we assume that indirect quotes can stand alone, since they are obligatory introduced by a complementizer in Brazilian Portuguese? Only exclamatory clauses (mostly formulaic expressions) in the imperative are allowed to begin with the complementizer, as Que Deus te abençoe! 'God bless you!', where the matrix clause (something like 'I hope') can be regarded as omitted. Therefore, it does not seem convincing to treat IQs as independent clauses, since they cannot "stand alone". Moreover, it is a part of matrix clause, functioning as one of its elements. With respect to DQs, on the other hand, they do have all the features of complete, structurally independent clauses (such as an expressed subject and a finite verb inflected in a tense different from the tense of the speech verb), that is, they can stand alone. Differently from IQ, in DQ there is no syntactic marking in the projected material to show its dependency, or its constituency, to the verb of utterance: it is not coded as a constituent. However, from a semantic and pragmatic perspective the direct quotes are closely linked to the utterance verb, representing the events reported. As we have seen, from a syntactic perspective the presence/absence of a complementizer renders difficult a uniform analysis of both direct and indirect quotation.
From a discourse perspective, the matrix clause and the direct/indirect complement clause can be seen to represent increasing degrees of connectedness between states of affairs. Following Hopper and Traugott (2003), the hypothesis is that the more overt and independent devices for signaling clause linkage are correlated with maximal semantic-pragmatic integration.
I suggest that "embedding" and "independent status" are not the only decisive criteria in the evaluation of the syntactic-semantic roles played by quotation clauses. Instead, the nature of these clauses may also be pragmatically accounted for by the notion of "projectability" (Ford and Thompson, 1996;Tanaka, 1999). This notion is mainly applied in the investigation of how conversationalists are able to assess potential points of turn completion before any actual completion or pause has been reached. Three aspects seem to be involved in projecting the end of a turn unity: intonational completion, syntactic completion, and pragmatic completion.
The production of a verbum dicendi in a narrative or conversation provides a strong indication of what is to follow. In other words, it early projects or anticipates, from its utterance, a "quotation format", i.e., a constructional schema that emerges from its frequency of use, and which participants in the conversation orient to. In this sense, the quoted clause, either a direct or an indirect quote, is somewhat dependent from the matrix verb. Thus, although direct quotes have all the features of syntactically complete, independent clauses, they show, nevertheless, a strong semantic and pragmatic bond with the clause which contains a speech verb. One context that favors the use of long amount of reported material is the kind of discourse in which the speaker is "holding forth" on a topic, as it happens in the conversational narratives analyzed here.
In terms of prosody, my data also exhibit a strong correlation between final intonation and the closure of the quoted material, which gives evidence that speakers use prosody functionally. The speaker signals the boundary of the quotation by the convergence of finality at the syntactic, pragmatic and prosodic levels, and the addressee recognizes that such cues point to the completion of meaningful units. Even when the amount of material in the quote is long and complex, the entire quotation is regarded by the addressee as a pragmatic complete action.
In many languages quotations do not have the same status as NP objects of verbs of saying, and therefore they do not fit into the function of 'object' or 'complement'. Furthermore, several languages restrict the complementation of verbs of utterance to direct quotation, which places them further from the transitive center (see Hopper, 1985). As shown above, on one hand, Brazilian Portuguese does not distinguish syntactically between clauses and nominal direct objects as complements of verbs of utterance: both occupy the post-verbal syntactic slot reserved for objects; on the other hand, both direct and indirect quotations are used. Moreover, some linguists have claimed that in many languages the boundary between direct and indirect quotation is fuzzy (Tannen, 1989;Holt, 1996). This does not seem to be the case in Brazilian Portuguese, given the specific characteristics of both types of quotation.
Like many other linguistic categories, reported speech is not a discrete entity. If we adopt a scalar treatment of complementation in terms of syntactic integration with the utterance verb (see Givón, 2001), we can distribute the types of quotation along a scale, with indirect quotation in the highest point on the scale, and direct quotation on the lowest. As seen, direct quotes exhibit the least syntactic dependence to the matrix verb, representing structurally independent clauses that can stand alone. Indirect quotes, on the other hand, cannot stand alone, because they are obligatory introduced by a complementizer, which establishes a relation of dependence between the matrix and the complement clause. In the case of reported speech, then, the criterion of transitivity does not correspond to a clear binary division: indirect and direct quotation occupies each extreme of the scale, with various intermediate points between them, represented by quotation forms of varying degrees of integration. Thus, DQ and IQ are just two schemas, from which several variants can be produced. On the semantic level, however, both quotations are strongly related to the verbum dicendi in the matrix clause.
The facts presented so far support that in Brazilian Portuguese, although the quotation is not a prototypical direct object, it is effected by the act of saying. In other words, verbs of utterance create their object (a semantic property), which are produced by the saying itself (a discourse property). This means that they play a presentational role, introducing new participants into discourse, elements which are "produced" through an action.
From a discourse perspective the quotative verb syntactically project the possible organization of what is to follow: it anticipates how the turn might develop. It projects, from its utterance, "aspects of its planned shape and style", it projects a "quotation format". The production of a verb of saying can provide a strong indication for how the turn might evolve. It is one of the syntactic practices/cues that participants orient to. Since quotative verbs strongly project what is to come, the quote can accurately be taken as the object complement of the matrix verb. If the term 'complement' implies completion (Payne, 1997), an utterance verb does not express a complete proposition until the quote is produced. Notice that in Brazilian Portuguese quotations display some of the features that characterize the prototypical direct object: they occupy the direct object slot and they represent the focus of the sentence, although they are semantically effected, instead of affected. In this sense, we can say that Brazilian Portuguese speech verbs show more transitive than intransitive characteristics.
In short, although the discourse function of the speech verbs is clearascribing a quote to its source -the investigation of the reported clause syntactic role raises interesting questions about the grammar of complementation. More specifically, it highlights the complexity and malleability of the linguistic system in real situations of language use, and the inaccuracy of categorically labeling a relation that is continuous and gradual in nature, as the linguistic constructions are at the service of the grammar of discourse relations.