The discourse commitments of illocutionary reportatives

This article develops an account of the discourse updates contributed by utterances of declarative sentences with the Cuzco Quechua reportative. The challenge posed by such utterances is that the speaker does not need to be committed to the at-issue proposition φ and may even deny its truth. They are therefore not assertions. Yet φ can behave in many ways like an asserted proposition in discourse: it can be used to answer questions, link to the discourse with veridical rhetorical relations, and, if accepted by the interlocutors, be subsequently presupposed. The proposed semantics for the reportative assigns the commitment to φ to a third-party principal instead of to the discourse participant producing the utterance, leaving them free to disagree with φ . However, if they do not disagree, they will be understood as intending to propose it to the common ground. This, it is argued, is due to the Collaborative Principle, a pragmatic principle that requires discourse participants to provide evidence of any discrepancy in commitments. The analysis is implemented in a modified version of the discourse framework of Farkas & Bruce 2010.


Word count: 18525 including abstract 1 Introduction
This article proposes a discourse-based solution to the puzzle in (1).* I would like to thank the audience and workshop participants of the following events for the very lively discussions of and insightful feedback on some of the material presented here: Semantics and Philosophy in Europe 8, Cambridge, September 2015, Backgrounded Reports, Nijmegen, January 2016, Linguistic Colloquium, Konstanz, January 2016, Semantics Lab, Manchester, March 2016, Pragmatics and Philosophy of Language Workshop, Cambridge, June 2016, and SPINFest, York, July 2016.Judith Tonhauser, the editor for this article, and four reviewers, including Sven Lauer, who lifted his anonymity, have provided me with extremely helpful comments and I cannot thank them enough for engaging with this work in such depth.My deepest thanks also as always to my consultants over the years, including in particular Inés Callalli, Natalia Pumayalli and the late Edith Zevallos.
(1) a. Absence of Commitment to φ : In Cuzco Quechua (CQ), a speaker uttering a declarative sentence with the reportative evidential does not need to be committed to the reported proposition φ in its scope.Such utterances are therefore not assertions of φ .b. Intention to resolve the QUD with φ : Despite (a), a speaker will often intend a reported proposition φ to resolve the Question under Discussion (QUD), that is, propose φ for the common ground.
The first part of the puzzle, Absence of Commitment, is illustrated in (2). 1 Similar examples have been described for reportative evidentials in a variety of other languages (AnderBois 2014).

Mana=má, no=IMPR
ni not un one sol-ta Sol-ACC saqi-sha-wa-n=chu.leave-PROG-1O-3=NEG 'They left me money (I was told)'.(But) no, they didn't leave me one sol.' (Faller 2002: 191) The definition of assertion of φ adopted in this paper is that the speaker proposes φ for the common ground CG (Stalnaker 1978).The CG is the set of joint discourse commitments of the participants and proposing φ for the CG therefore requires commitment to φ .The speaker of (2) is neither proposing φ for the CG nor committed to it. 2The reported proposition is, however, at-issue (see Section 2.3).At-issue content can therefore not be identified with asserted content.
The second part of the puzzle, Intention to Resolve the QUD, is the observation that a speaker may nevertheless intend to resolve the QUD with a reported proposition, as illustrated in (3).A proposition (partially) resolves the QUD if it addresses the QUD and all discourse participants accept it into the CG.The intention to resolve the QUD with φ requires the speaker's commitment to φ .
(3) B: May-pi=taq where-LOC=CONTR ka-sha-n be-PROG-3 chay this wawa-yki.child-2 'And where is this child of yours?' A: San San Salvadur-pi=s Salvador-LOC=REP ka-sha-n.be-PROG-3 'He is in San Salvador (I was told).' (Radio program Warmikuna Rimanchis, 15 May 2002) The fact that reportative utterances are often intended to resolve the QUD has not been as widely discussed as Absence of Commitment, though Murray (2014: 5) observes that Cheyenne speakers may be fully committed to a proposition presented with a reportative, and then presumably may intend to resolve the QUD with it.
The key to resolving the apparent tension between (1a) and (1b) lies in recognizing that reportative utterances involve two agents: The current speaker only animates the utterance of another agent, the principal (Goffman 1979), and it is the principal's commitment that is being conveyed.Nevertheless, the animator typically pursues their own conversational goals by bringing a third party's commitment into the discourse, and this may include making φ common ground.The proposed solution to the puzzle has the following components: (i) Declarative sentences introduce three defaults: a commitment by the principal to the truth of the at-issue proposition, a commitment by the animator to having adequate evidence, and the identification of the principal with the animator.When these defaults apply, the resulting speech act is assertion.
(ii) The semantics3 of illocutionary reportatives introduce a commitment by the animator to having reportative evidence and specify that the animator is distinct from the principal.The latter accounts for the absence of commitment by the animator.
(iii) A pragmatic discourse rule, the so-called Collaborative Principle (Walker 1996), requires discourse participants to voice any disagreement immediately.
This ensures that the animator will be understood as proposing φ for the CG unless they signal disagreement.
The commitment incurred by a reportative utterance is however weaker than that of standard assertion.I propose that this is due to the animator having committed to φ only on the basis of reportative evidence, resulting in a dependent commitment (Gunlogson 2008).The account, which will be formulated in Farkas & Bruce's (2010) discourse framework, is intended to generalize to illocutionary reportatives in other languages.
Because the CQ reportative and reportatives like it in other languages have the properties just described, their study provides insights into more general issues at the semantics/pragmatics interface.Specifically, the paper argues that a sharp distinction between at-issue and asserted content should be drawn; that declarative sentence type should be associated only with the speech act of presentation, from which assertion is derived by default, and that evidential commitments should be added to the speech act type of assertion.These aspects of the analysis are needed independently of reportatives, for example in the analysis of propositional attitudes and hedges.
The discussion proceeds as follows.Section 2 presents the CQ reportative and illustrates the key phenomena to be accounted for; it also argues that at-issue content cannot be identified with asserted content.Section 3 evaluates to what extent previous accounts of reportatives capture the relevant properties.Section 4 introduces Farkas & Bruce's (2010) account of default assertion to which evidential commitments are added as a distinct type of discourse commitment.Section 5 introduces Goffman's (1979) distinctions within the speaker role.In particular, the distinction between the roles of principal and animator will be central in the analysis of the CQ reportative developed in Section 6. Section 6.1 presents the semantics of the CQ reportative and explains how it accounts for the absence of the animator's commitment to the reported proposition φ .Section 6.2 develops the pragmatic account of how the animator may nevertheless intend for φ to resolve the QUD, and Section 6.3 relates the analysis to Gunlogson's (2008) ideas about source and dependent commitments.Finally, Section 7 concludes with a summary and issues for future research.
For convenience, I will continue to use the term speaker to refer to the person producing an utterance until the distinction between animator and principal has been properly introduced in Section 5.

Discourse properties of the proposition reported with CQ =si
Reportatives are a type of evidential, that is, markers that indicate the speaker's type of source of information (Anderson 1986, Willett 1988, Aikhenvald 2004).With reportatives, this source is the report of a third party.The CQ reportative =si (allomorphs =s or =sis) is part of a paradigm of evidential focus enclitics, illustrated in (4), which includes the best possible grounds marker =mi (allomorph -n) and the conjectural =chá in addition to the reportative =si (Cusihuaman 2001, Muysken 1995, Faller 2002). 44) Para-mu-sha-n=mi/=chá/=si. The CQ evidentials occur in both declarative and interrogative main clauses,5 but are ungrammatical with imperatives.This paper only deals with reportative declarative sentences, abbreviated to RDS for convenience.

News story
(5) a. Hinaspa then chay-pi=s, this-LOC=REP qonqay, suddenly hap'i-ra-pu-sqa-ku grab-EXH-BEN-NX.PST-PL kinsa-manta.three-ABL 'And then, there, all of a sudden, they grabbed him, between three.The next two subsections illustrate the puzzle that is the topic of this article in more detail.Note: I use the term reportative condition to refer to the conventional contribution of the reportative and the term reported proposition for its prejacent φ .Section 2.3 argues that this proposition is at-issue.

Absence of commitment to reported proposition
As shown in (2), a speaker may follow a RDS with an explicit claim that the reported proposition φ is false.While contradicting φ , the speaker is not understood as contradicting herself.This, I claim, is due to the fact that a RDS does not encode speaker commitment to φ .To give another example, the speaker of (8) expresses his doubts that the reported proposition (highlighted in bold) is true, but does not go as far as asserting its falsity. 6There is again no perception that the speaker is being inconsistent.'In all streets there was talk, the gringos, it is said, went to the Moon, travelling one whole week in an aeroplane, saying.For myself, this seems just talk.' (Valderrama & Escalante 1982: 25) There are also examples where the context does not provide any clues as to whether the speaker, for example, the messenger in (32) below, is committed to φ .Thus, RDSs in CQ do not conventionally encode any degree of commitment by the speaker (Faller 2002(Faller , 2007)).Similar observations have been made for RDSs in other languages (AnderBois 2014 and references cited therein).
6 This example is complicated by the fact that the relevant utterance is embedded as a direct quotation under nispa 'saying'.This raises the possibility that the reportative is used in concord with this verb of saying.Given the nature of the example, however, it seems to me a more natural interpretation that the reportative is part of what people said.In either case, the narrator has doubts about the veracity of the reported proposition that the gringos went to the moon.

Reported proposition may enter the CG
Despite the speaker not publicly committing to the reported proposition, it can, in many ways, behave like a proposition that has been asserted (with the best possible grounds evidential or no evidential) and become accepted into the CG.In this example, too, the addressee accepts (9) as true and the conversation moves on to what Mario could eat.
The reported proposition can moreover connect with the previous discourse with the full spectrum of rhetorical relations (Faller 2007), including veridical relations, that is, relations that entail the truth of the related propositions (Asher & Lascarides 2003: 157).( 5) and ( 6 'It cannot fly, but they take it to the lake, to the top of a small mountain.'(Conversation, Faller 2007) I use the phrase Intention to Resolve the QUD to refer to the speaker's intention to answer an overt question as well as to other ways of linking the reported proposition to the previous discourse, based on the assumption that all contributions address an (implicit) QUD (Roberts 1996(Roberts /2012)).When the addressee recognizes this intention, they have the same options as with asserted propositions, namely to accept or reject it (or ask clarification questions).If they accept it, I will say that the QUD has been resolved.
Once a reported proposition is in the CG, it can be presupposed in subsequent discourse, by the same speaker or by their interlocutors.For example, the request of speaker A in (11) to her husband B to buy certain items presupposes that the two women referred to brought them, and so does his utterance accepting the request.7  exchange-EXH-CISL-1FUT 'I will go, then I will quickly buy the medicine.'(Loaiza 2010: 25) The above examples show that a reported proposition can update the discourse in ways similar to regularly asserted propositions, but only if the speaker intends this and if the addressee recognizes this intention.In cases such as (2), the speaker introduces the reported proposition only to immediately refute it and so does not intend for it to become common ground.
While this is not often explicitly discussed, I assume that RDSs in other languages can also be used with the intention of resolving the QUD.Thus, Murray (2014: 5) mentions that Cheyenne speakers can be fully committed to reported propositions, and so presumably then intend them to resolve the QUD.There might be crosslinguistic variation with regards to whether languages allow this.In both Cheyenne and CQ, the reportative has to be used if that is the speaker's only kind of evidence regardless of the strength of their own commitment.In other languages, however, speakers may use bare assertions even with reportative evidence, and propositions introduced with a reportative are then perhaps less likely to be intended to enter the CG.More research is needed to establish the parameters of cross-linguistic variation in this respect.(2005) coined the term at-issue to distinguish what Grice called what is said from conventionally encoded meanings that provide new information but that are not at-issue.At-issue entailments contribute "controversial propositions" or carry the "main themes" of a discourse.As such, these are the propositions that speakers "are most expecting to have to negotiate with their interlocutors before [they are] accepted into the common ground" (Potts 2007: 666).This characterization directly fits the reported propositions in the examples presented: It is the reported propositions, not the fact that the speaker acquired them via reports, that constitute the main theme in the news report in (5) and the folktale in (6), and it is the reported proposition that answers the question in (3).This also holds for reported propositions that the speaker is not committed to.Thus, while the proposition that they left me money is not proposed by the speaker of (2) for the CG, it is still the main theme and controversial.Murray (2010Murray ( , 2014) also analyzes the prejacent of the Cheyenne reportative as at-issue and this is presumably the case for reportatives across a variety of languages. 8his section provides support for the claim that the reported proposition is atissue by applying two standard tests.It is also shown that the reportative condition is not at-issue.The QUD test probes for the main theme status of at-issue propositions by checking which of the propositions conveyed by a sentence address an overtly asked QUD (Simons et al. 2010, Tonhauser 2012).The direct response test probes for the negotiability of at-issue propositions by checking whether discourse participants can directly (dis)agree with them (Papafragou 2000, Faller 2002, Tonhauser 2012, AnderBois et al. 2015).The tests are first illustrated with propositional attitudes and epistemic modals as they support the more general claim that at-issue content is not necessarily asserted.

Potts
With propositional attitude reports, either the prejacent or the attitude can be at-issue (Simons 2007).The most likely interpretation of B's utterance in ( 12) is that Louise is tired because she did all the work, not because Henry said so.B's response in (13), however, conveys that Louise is upset because Henry said that he did all the work.
(12) A: Why is Louise so tired?B: Henry said that she did all the work on their renovation project.
(13) A: Why is Louise so upset?B: Henry said that he did all the work on their renovation project, when it was in fact her who did the lion share.
The direct response test in ( 14) confirms that both propositions can be at-issue.(14B) denies the truth of the prejacent in (14A), but (14B ) denies the truth of the main clause proposition.
(14) A: Henry said that he did all the work.B: That's not true.He did at most a third of it.B : That's not true.He didn't say that.
Epistemic modal statements also allow both candidate propositions to be at-issue (von Fintel & Gillies 2008).It is the prejacent of must in (15B) that potentially answers A's question, and that is targeted by C's subsequent denial in (15C).
(15) A: Why has Louise not been coming to our meetings recently?
B: She must have left town.C: No, she hasn't.
While rarer, there are also examples of epistemic modals themselves being atissue.Assume that A and B are playing Mastermind, where A has to work out a pattern of pegs that only B can see.At some point, A reasons out loud that there might be two red pegs, to which B responds with (16B) (von Fintel & Gillies 2008: 83).
(16) A: There might be two red pegs.
B: That's right.There might be.
B's answer accesses the epistemic modal itself, not the prejacent.It would be felicitous even if B knew that there were not two reds.
In contrast, with CQ RDSs, only the prejacent can be at-issue.In (17) it is the proposition that it crawls like a bug that answers the question of how the train might be, not that this is said (see also (3)).
(17) A: Tren, train, tren, train, imayna=chá?how=CONJ 'The train, the train, how might it be?' B: Kuru bug hina=s like=REP suchu-n.crawl-3 'It crawls like a bug (they say).' (Valderrama & Escalante 1982: 30) The reportative condition cannot answer the QUD, even if the QUD asks about it.Thus, (18B) is not a coherent answer to (18A).( 18 Thus, with Quechua RDSs, only the reported proposition is at-issue, while the reportative condition can never be.Murray (2014: 16) provides comparable data for the Cheyenne reportative and for English parenthetical I hear.
Having argued that the reported proposition φ is at-issue, it bears emphasizing that it is not asserted, given that utterances of RDSs neither require speaker commitment to φ nor necessarily propose φ for the CG.In cases like (2) and ( 13), the speaker is precisely not proposing to add φ to the CG.The same holds for the prejacents of propositional attitudes and epistemic modals.Definitions that characterize at-issue content as being proposed to the CG (in distinction to not-at-issue content which is imposed) (Murray 2014, AnderBois et al. 2015) therefore fail to classify these prejacents as at-issue.I therefore adopt the definition proposed by Simons et al. (2010: 323) in (20), in which the central notion is whether or not a proposition addresses the QUD.
(20) a.A proposition φ is at-issue iff the speaker intends to address the QUD via φ .b.An intention to address the QUD via φ is felicitous only if: i. φ is relevant to the QUD, and ii. the speaker can reasonably expect the addressee to recognize this intention.
(20) correctly classifies the prejacents of reportatives, propositional attitudes, and epistemic modals as at-issue when these are relevant to the QUD and so directly captures the results of the QUD test.Given that these prejacents are potential answers to the QUD, they are expected to be negotiable, and (20) therefore also accounts for the fact that they are accessible to direct responses such as yes/no/that's (not) true.If we make the additional assumption that the CQ reportative is conventionally specified as not-at-issue, (20) also accounts for the observation that the reportative condition cannot be at-issue even when relevant (see example ( 18)), since a speaker who uses a marker that is conventionally specified as not-at-issue indicates that they do not intend to address the QUD with it (Simons et al. 2010).
In sum, the prejacents of reportatives, propositional attitudes and epistemic modals can be at-issue but are not necessarily proposed for the CG.

Previous accounts of reportatives
There are then three properties of reported propositions φ that we need to capture: (i) the absence of speaker commitment to φ in some cases, (ii) the intention to resolve the QUD with φ in other cases, and (iii) the at-issueness of φ .This section reviews whether and how previous accounts of reportatives account for these, and motivates the need for another account.In doing so, this section also provides an overview of the different options available to researchers approaching this complex issue.With regard to the puzzle that is the focus of this paper, the main choice to be made is whether or not to capture absence of speaker commitment in the semantics of RDSs.

Reportatives as informational modals
This paper assumes the existence of evidentials that affect the illocutionary force with which their prejacent φ is put into the discourse.The CQ and Cheyenne reportatives are of this kind (Faller 2002, Murray 2014).Other evidentials have been analyzed as modals with an evidential presupposition.9The diagnostics that have been used to draw this distinction include the ability of modal but not illocutionary evidentials to be at-issue, the ability of modal but not illocutionary evidentials to take narrow scope with respect to propositional operators, and the ability of illocutionary but not modal evidentials to take scope over speech acts (Faller 2002, 2011, Matthewson et al. 2007, Murray 2010). 10hile this paper is primarily concerned with illocutionary reportatives, it is nevertheless useful to review whether and how the reportative-as-modal approach captures the properties of reported propositions.This approach is typically formulated within Kratzer's framework for modality (1981,1987), and Kratzer (1987) herself proposed such an analysis for the reportative use of the German modal verb sollen (see also Ehrich 2001).Early work in this tradition (e.g., Izvorski 1997, Matthewson et al. 2007) assumes that evidentials are a subtype of epistemic necessity modal.However, this requires that φ be compatible with the speaker's knowledge and so is not applicable to reportatives that allow the speaker to know that φ is false.An alternative modal account associates reportatives with an informational conversational background which contains the contents of reports (Kratzer 2012: 34) rather than taking reports to constitute evidence for φ .A sentence of the form Rep(φ ) then asserts that φ is true in all worlds that are compatible with what has been said.This account predicts that the reportative condition itself should be able to be at-issue (and is therefore not applicable to the CQ and Cheyenne reportatives).While it says nothing in and of itself about the ability of φ to be at-issue, it is compatible with the definition of at-issueness in terms of relevance to the QUD and so allows for this.The informational modal account also captures Absence of Commitment given that it does not require φ to be consistent with the speaker's knowledge or beliefs.It does, however, not account for cases where φ is intended to resolve the QUD, because without the speaker's commitment, it cannot be taken to be proposed for the CG.

Reportatives as illocutionary modifiers
Faller (2002) analyzes the CQ evidentials as illocutionary modifiers (see also Faller (2007Faller ( , 2011Faller ( , 2014))).The analysis is couched in the speech act theory of Searle and Vanderveken (Searle 1969, Searle & Vanderveken 1985, a. o.) which captures the intentions and attitudes of the speaker towards the propositional content φ by means of an illocutionary force operator F. F is constituted of several components, Korotkova (2016) also acknowledges that the fact that the CQ reportative can take a question act in its scope, as in (i) (Faller 2002), supports an illocutionary account of at least this evidential.
(i) Context: Martina asked the mother-in-law of her consultant how she was.The mother-in-law didn't hear her, so the consultant asks: Imayna=s how=REP ka-sha-nki.be-PROG-2 '(She says) How are you?' (Conversation) Since Korotkova (2016) by her own acknowledgement does not aim to capture Absence of Commitment, her account of evidentials as markers of subjectivity is not reviewed here.
including preparatory and sincerity conditions and the illocutionary point.For basic assertion, these are (Searle & Vanderveken 1985: 37-45): (21) a. Illocutionary point: speaker presents φ as representing an actual state of affairs in the world of utterance b.Preparatory condition: speaker has reasons or evidence for the truth of φ c.Sincerity condition: speaker believes that φ This basic force can be made more specific by adding illocutionary modifiers; for example, the CQ reportative adds the sincerity condition that the speaker has heard reports that φ (Faller 2002).To account for Absence of Commitment, Faller (2002) suggests that the reportative furthermore removes the sincerity condition that the speaker believes that φ , resulting in the overall weaker illocutionary force of presentation.11In this type of analysis, the illocutionary and propositional meanings belong to distinct levels, and the assumption is that only propositional-level meanings contribute to at-issue meaning.We can relate this to the definition of at-issueness in (20), as briefly discussed in Section 2.3, by requiring that illocutionary elements are conventionally marked as not at-issue.It then follows that only the reported proposition can be at-issue.However, due to the sincerity condition in (21c) having been removed, we do again not have an account of cases where the speaker intends to resolve the QUD.
The upshot so far is this: any analysis that encodes Absence of Commitment in the semantics is not able to simultaneously claim, in the semantics, that the speaker proposes φ for the CG.(This also applies to Murray's (2014) account discussed in the next section.)In order to account for cases where the speaker intends to resolve the QUD with φ , an additional pragmatic account is needed that brings the speaker's commitment back into the equation.Faller 2007 is an earlier attempt to develop such an account.It recasts the basic insights from Faller 2002 in the framework of Segmented Discourse Representation Theory, which augments DRT with rhetorical relations and introduces a separate cognitive module for modeling beliefs and intentions.Discourse content is built up independently from the speaker's beliefs and intentions.This allows the reported proposition to be integrated into the discourse via rhetorical relations just as asserted at-issue propositions.For example, (3A) can be analyzed as standing in an answer-relation to (3B), without the need to refer to the speaker's belief that he is in San Salvador at this level.Sincerity only comes into play in the cognitive module.In SDRT, sincerity is, in the first instance, defined on rhetorical relations: In asserting a proposition φ , the speaker conveys the belief that φ stands in a particular rhetorical relation to the preceding discourse, for example, an answer-relation.Standard sincerity is entailed for veridical relations (Asher & Lascarides 2003: 398).
The reportative contributes the condition that the speaker has reportative evidence for φ to the cognitive module, which captures its not-at-issue nature.In addition, it overrides default sincerity and triggers instead conditional sincerity, whereby the speaker conveys that they believe that φ stands in the relevant rhetorical relation to the previous discourse if it is true.Thus, the speaker of (3A) believes that He is in San Salvador answers the question if it is true.12Conditional sincerity does not entail that the speaker believes φ and therefore accounts for Absence of Commitment.Now, if the speaker does not intend for φ to be part of the discourse content (and therefore CG), they can deny its truth, as in (2).Faller (2007) proposes that the contradicting proposition (But no, they didn't leave me one sol) connects to the preceding one with the non-veridical relation Correction, which makes the discourse overall coherent.But in the absence of Correction, φ becomes part of the discourse content.This accounts for the fact that a speaker may present φ with the intention of resolving the QUD, even though conditional sincerity does not require them to believe it.
I consider the basic insights of this account to be essentially correct, in particular the idea that it is the absence of disagreement that triggers the inference that the speaker intends to resolve the QUD with the reported proposition.But it also has some problems, which the account developed in Section 6.2 aims to avoid.Most importantly, the idea of conditional belief towards a rhetorical relation is problematic.It requires the postulation of a specialized, conditional sincerity axiom that is triggered by the reportative.However, sincerity is a concept that should not be dependent on the particular linguistic expressions used.This solution is ad hoc and stipulative.Moreover, it simply does not seem to be the case that the speaker of (3A) is only conditionally committed to her utterance providing an answer to the preceding question.She believes it to be true.
The account in Section 6.2 goes beyond Faller 2007 by being explicit about the discourse pragmatic reasoning that leads to the reported proposition being added to the CG, without the stipulation of an additional sincerity axiom.The speaker's commitment, where intended, is a full, non-conditional, commitment, resulting from Walker's (1996) Collaborative Principle, which is needed independently.It is moreover couched in a more mainstream discourse framework, sidestepping the potential SDRT-internal technical issues raised by Portner (2006) and develops in detail the idea of associating declaratives with the speech act of presentation.

Reportatives as hedges in update semantics
Murray (2014) analyzes reportatives within an update semantics in which sentence type plays a central role.Declaratives contribute three kinds of update: (i) the proposal to add a proposition, usually the at-issue proposition, to the CG; (ii) the association of the at-issue proposition with a discourse referent that makes it available for anaphoric reference and accessible to expressions of (dis-)agreement by other participants; (iii) the imposition of any not-at-issue propositions directly on the CG.In the case of RDSs, the reported proposition is at-issue, and therefore, by (ii), associated with a discourse referent, and the reportative condition is not-at-issue and therefore, by (iii), imposed on the CG.The reported proposition is however not proposed for the CG, and this accounts for Absence of Commitment.But again, precisely because RDSs do not propose the reported proposition, there is no account of those cases where it is nevertheless presented with the intention of resolving the QUD.
Since Murray 2014 is the main semantic competitor for the account of Absence of Commitment proposed in the current paper, let us look in more detail at what it involves.Note that there is an apparent conflict between the assumption in (i), that declarative sentences propose a proposition to the CG, and the claim that RDSs do not propose their at-issue proposition.This is resolved in Murray 2014 by allowing hedges in general, and reportative evidentials and parentheticals such as I hear in particular, to alter the argument of the proposal update.Instead of proposing the at-issue proposition, RDSs propose "the current context set" (Murray 2014: 13).As a result the only updates taking place are (ii) and (iii).While it technically resolves the conflict, I find this move conceptually problematic.The idea that RDSs propose a proposition that is not even derived from (e.g., by modalizing) the at-issue proposition is a significant departure from the standard view that sentence type determines how the at-issue proposition enters the discourse structure.Moreover, proposing the current context set is in fact not a proposal in Murray's sense, as the main distinction between proposed and imposed content is that interlocutors may object to the former and thereby prevent it from entering the CG.A proposal of the current context set cannot be objected to, however, as it reflects the status of the CG at the point of update.On the assumption that the current context set is the intersection of the propositions taken for granted by the participants (Murray 2014: 8) this move also blurs the distinction between presupposed and proposed content.
The accounts reviewed so far assume that RDSs do not propose the reported proposition φ for the CG.What needs to be explained is how a speaker can nevertheless intend to resolve the QUD with φ .One can, however, also approach the puzzle from the opposite direction and start with the assumption that RDSs, like other declaratives, conventionally convey that the speaker proposes φ for the CG.What needs to be explained then are cases where the speaker goes on to contradict or doubt φ .The account by AnderBois (2014) reviewed next is of this sort.

Absence of Commitment as pragmatic perspective shift
AnderBois (2014) takes issue with accounts that do not associate RDSs with the speech act of assertion because they contravene the Baseline Conception of Evidentials in ( 22) (adapted from his (2)).( 22) A speaker who sincerely utters a declarative sentence with propositional content φ and an evidential of type EVID typically: a. Performs an assertion with content φ (or a modalized version thereof).b.Conveys in some way that the speaker has EVID-type evidence that φ .(AnderBois 2014: 235) (22a) is a common assumption in the literature, though not always formulated explicitly, which has its roots in the more general assumption that declarative sentences are typically used to make assertions.If we maintain that RDSs are first and foremost used to make assertions, and therefore, to propose φ to the CG, we have an immediate explanation of the observation that a speaker may intend to resolve the QUD with φ .What needs to be explained are then those cases where the speaker does not have this intention, such as (2).

Mana=má, no=IMPR
ni not un one sol-ta Sol-ACC saqi-sha-wa-n=chu.leave-PROG-1O-3=NEG 'They left me money (I was told)'.(But) no, they didn't leave me one sol.' (Faller 2002: 191) AnderBois (2014) proposes that in such cases a pragmatic perspective shift in the sense of Harris & Potts 2009 applies: the reportative condition makes salient the perspective of the original speaker, and, in sufficiently rich contexts, 13 the proposal to add φ to the CG is shifted to them. 14This explains why the current speaker can deny φ .
Empirical support for this approach is provided by the fact that, cross-linguistically, most, if not all, reportatives, and only reportatives, allow for the speaker to deny the truth of φ , despite being otherwise syntactically and semantically diverse (Ander-Bois 2014: 237ff).A pragmatic account of this phenomenon is to be preferred over one that stipulates this in the semantics of individual reportatives, especially if this allows us to give the same kind of semantics to reportatives as to other evidentials.Moreover, when speakers explicitly contradict φ , they "typically make use of words translatable with English really, actually, or true, first person attitude reports, negative polarity items, and other kinds of evaluative language" (AnderBois 2014: 242).In (2) this is the surprise enclitic =má.
These are important empirical observations and the perspective shift account explains them both: all, and only, reportative evidentials make a third party salient to which the proposal can be shifted, and the use of evaluative and other perspectival markers is expected as this signals the speaker's intention to shift the perspective back to themselves.
AnderBois's ( 2014) account is at first sight very appealing, as it offers an account of denial cases like (2) while simultaneously maintaining a tight connection between declarative sentence type and assertion.On closer inspection, however, it does in fact not achieve the latter, or only rather superficially.First note that AnderBois (2014) postulates two distinct types of assertion.Ordinary assertions propose the at-issue proposition φ for the CG and commit the speaker to φ , whereas evidential assertions only propose, but do not require speaker commitment to, φ . 15Second, as indicated in (22a), some evidential declarative sentences (presumably those containing an AnderBois (2014: 243) clarifies that "context" here is to be understood "in the broadest sense, including world knowledge about the speaker's beliefs, the speaker's presumptions about the addressee's perspective on these, as well as knowledge about the perspective of the contextually salient reporter including his/her reliability."(By reporter, AnderBois (2014) means the original speaker.)More precisely, in a perspective shift context, "[the] speaker points out that the reporter would propose to add [φ ] to CG {A,B} [ . . .]" (AnderBois 2014: 251).To make this work, AnderBois defines the CG as a set of propositions that speaker and addressee mutually agree on being true as a basis for joint action "even if neither is entirely committed to this being so, publicly or privately" (p.250).That is, unlike in previous work that tracks the individual commitments of participants separately from the joint commitments in the CG (e.g., Gunlogson 2001, Farkas & Bruce 2010), the intersection of the discourse participants' sets of commitments is not a subset of the CG.Proposing a proposition for the CG does therefore not entail a commitment.While not entailed, commitment is usually implicated also with evidential assertions, because "a cooperative, rational speaker nonetheless should not make such an assertion if they believe p to be false (i.e., proposing to add false information to the CG would violate the Maxim of Quality)" (AnderBois 2014: 250).inferential) are not used to propose their prejacent φ , but rather a modalized φ .Maintaining the baseline in ( 22) thus comes at the cost of significantly weakening the standard understanding of what it means to assert φ .
It is also not the case that the empirical facts unequivocally support a pragmatic perspective shift account.While such an account explains the presence of evaluative markers in denials of a reported proposition (AnderBois 2014: 244), it is not the only possible explanation.Moreover, at least in some languages "bare" denials are also possible. 16 The possibility of such bare denials, which is expected on a semantic account of Absence of Commitment (AnderBois 2014: 243), weakens the empirical support for the pragmatic perspective shift account.
Both Murray (2014) and AnderBois (2014) aim to keep the contribution of declarative sentence type constant across all uses, namely as proposing a proposition to the CG, normally the at-issue proposition φ .Additional mechanisms are then put in place to ensure that RDSs may be used without proposing φ and that conjecturals and other hedges weaken the commitment to or modalize φ .The account developed in the following takes the opposite approach: it also keeps the contribution of declarative sentence type constant but associates it only with the weaker force of presenting φ .This is strengthened to full, ordinary assertion by default, unless there is an illocutionary marker that overrides the default (Section 4).The CQ reportative is such a marker, attributing the commitment to φ not to the current speaker but to their reportative source of information (Section 6.1).This semantic account of Absence of Commitment also explains the frequent use of perspectival or evaluative markers in denial cases, but does not require it (Section 6.3).

Default Assertion and Acceptance
Since we are interested in how RDSs address the QUD and restrict the ways in which a discourse can develop, a dynamic model is to be preferred over a static one.I continue to use the term speech act and use it to refer both to discourse moves, that is, proper acts carried out by speakers in using a sentence, as well as to the formal object that models such acts in the discourse representation.In the latter usage, I will, for example, say that speech acts are functions from discourse structures to discourse structures.
In Stalnaker's influential account, assertions are made against the background of a context set, the set of worlds obtained by intersecting the propositions in the CG and that the discourse participants have agreed are "live options" for the possible ways things may be (Stalnaker 1978: 85).The (highly idealized) purpose of conversation is to reduce the set of live options.Assertions of φ serve this purpose by proposing to add φ to the CG.Current theories of discourse have added to and modified Stalnaker's basic set-up in various ways, and I found the tools developed by Farkas & Bruce (2010) particularly useful.Before presenting a modified version of their account of default assertion in Section 4.2, I review the notion of illocutionary commitment in 4.1.

Illocutionary commitments
Speech act theories fall into two broad groups.The first categorizes speech acts on the basis of the mental states or attitudes expressed by sincere utterances.For example, as shown in (21c), Searle & Vanderveken (1985) associate the sincerity condition that the speaker believe that φ with the assertion of φ .The second categorizes speech acts in terms of public commitments, in the case of assertion, to the truth of φ (Harnish 2005).I adopt the commitment approach which is currently prevalent in leading linguistic theories of discourse (Gunlogson 2001, Farkas & Bruce 2010, Krifka 2014, Murray 2014, Portner 2018, amongst others), because we are interested in how certain linguistic expressions update the conversational record, and not directly in the participants' mental states.Discourse commitments are "propositions [ . . .] publicly taken by the participants in a conversation as being true of the world of the conversation, from the perspective of the conversation" (Farkas & Bruce 2010: 84), but they "need not be in fact true of the world in which the conversation takes place" (Farkas & Bruce 2010: 86) and are "independent of participants' actual beliefs and intentions" (Farkas & Bruce 2010: 93).Thus, a speaker who falsely asserts φ is nevertheless publicly committed to the truth of φ .
The term commitment in this linguistic tradition is however slightly different from the way it is used in the normative approach to speech acts in philosophy (see Harnish 2005 for an overview), according to which assertion has consequences beyond the speaker's public commitment being registered in the conversational record.Speakers also subject themselves to relevant societal norms and lay themselves "open to the possibility of censure, correction, or the like in case the conditions in question are not satisfied (Alston 2000: 71, cited in Harnish 2005).When challenged, a speaker needs to be prepared to offer justification or evidence for their claim (Krifka 2014).Thus, in asserting φ , a speaker not only commits to its truth but also to having adequate evidence for it.I will track both types of commitment.
While a speaker's commitments do not directly equate to actual mental attitudes, these can be inferred on the presumption of sincerity.Thus, if a speaker commits to the truth of φ , their interlocutors are justified in inferring that they believe φ .Indeed, successful lying relies on this inference.In the same vein, evidential commitments give rise to the inference that the speaker has adequate evidence.Thus, truth and evidential discourse commitments indirectly correspond to the sincerity and preparatory condition of assertion in a Searlian framework (21), and to Grice's (1989) first and second Quality maxims.

Default assertion
I adopt Farkas & Bruce's (2010) discourse framework as the basis for developing my analysis of CQ RDSs, as it allows for easy integration of the additional components necessary to account for the observations made so far.This section introduces their account of default assertion, augmented with evidential commitments.
Building on insights from Hamblin (1971), Gunlogson (2001), Ginzburg (1996), Roberts (1996Roberts ( /2012)), Krifka (2001) and others, Farkas & Bruce (2010) model assertion and other discourse moves in terms of the effects they have on the discourse context, represented in so-called discourse structures (DSs).In addition to Stalnaker's CG, a DS consists of separate sets of discourse commitments for each individual discourse participant and a Table .Tracking the individual participants' commitments separately enables Farkas & Bruce (2010) to account for the fact that two participants may have contradictory public commitments, a crisis which can be resolved for example by "agreeing to disagree" (see also Farkas & Roelofsen 2015).The Table is a stack of questions/issues under discussion.It is related to the Question Set in other theories (Portner 2018), but is conceived more broadly as accepting questions as well as answers.Both questions and assertions put their at-issue content onto the Table (a proposition in the case of assertion, a set of propositions in the case of questions) and the expectation is for the discourse participants to attempt to resolve these issues, that is, to answer any questions or to accept or reject the propositions on the Table .The Table is thus a direct implementation of the idea that propositions are at-issue relative to a QUD.
As discussed in the preceding section, in order to fully capture the nature of assertion, we should track the evidential commitments of participants in addition to their truth commitments.I therefore rename Farkas & Bruce's (2010) set of a participant x's discourse commitments DC x to truth commitments TC x , and add a set of evidential commitments.In the absence of any indication as to the specific type of source of information, these are required to be adequate in Grice's (1989: 29) sense, that is, strong enough to support the truth of φ .This is summarized in (24).
(24) a. TC x = the set of propositions the truth of which x is committed to b. AeC x = the set of propositions for which x is committed to having adequate evidence Evidentials contribute commitments of a specific type which are tracked in distinct sets.For example, the CQ reportative adds its prejacent to RepC, whereas the best possible grounds evidential adds its to BpgC.As we will see below (in the discussion of ( 31)) the context may also provide clues as to the specific type of evidence.
(25) a. RepC x = the set of propositions for which x is committed to having reportative evidence b.BpgC x = the set of propositions for which x is committed to having best possible grounds For the purposes of this paper, I treat the truth and evidential commitment sets as formally independent of each other.They are however connected in a fairly obvious way: a speaker's type of evidential support for φ may determine the strength of their commitment to the truth of φ .17Thus, a commitment to having best possible grounds for φ entails a strong commitment to the truth of φ , an inferential commitment entails a weak truth commitment to φ , and a reportative commitment entails that someone else is committed to the truth of φ .I cannot fully explore these interactions here, but will relate some of them to Gunlogson's (2008) distinction between source and dependent commitments in Section 6.3.
Speech acts are modeled as functions from DSs (and speakers and sentence denotations) to DSs.An assertion of φ by speaker A has the effect that φ is put on the Table and added to both A's truth and evidential commitments.Following Farkas & Bruce (2010), I represent DSs in the form of tableaux such as Figure 1, which shows the output DS of Amy's assertion of φ = that the server is down to Ben (example (40) in Section 6.3). 18 Output DS after A's assertion of φ = that the server is down The CG contains the propositions in the intersection of the participants' sets of truth commitments.φ therefore only enters the CG after the addressee has accepted it, see Section 4.3.The CG in addition contains records of "obviously observable" events occurring in the utterance situation (Stalnaker 1978: 86).Speech acts are also such manifest events and therefore give rise to a number of secondary effects recorded directly in the CG: "the fact that a speaker is speaking, saying the words he is saying in the way he is saying them", etc. (Stalnaker 1978: 86).This includes records of the illocutionary commitments, that is, that the speaker has committed to the truth of and to having evidence for φ . 19Since such secondary effects are not relevant for the current paper, I do not include them in the representations, but for completeness, the update to CG with these is shown in (26).
(26) CG ∪ {A is committed to the truth of φ , A has adequate evidence for φ } 18 I adopt the following conventions for the tableaux: Commitment sets, including the set CG that tracks the joint commitments of the discourse participants, are represented with italics.The (subscripted) upper case letters A and B refer to the particular speaker and addressee of an utterance.All tableaux include TC, RepC and AeC for both discourse participants as the most relevant components, even when only one of them is updated.Other types of evidential commitment sets such as BpgC, while also tracked, are only included when specifically relevant for an example, e.g., in Figure 5.The bold faced labels A, B, Table , Common Ground serve to structure the tableaux, but play no formal role in the analysis.19 I would like to thank Sven Lauer for suggesting that records of illocutionary commitments enter the CG directly because they cannot be disagreed with, directly or indirectly.For example, while a speaker asserting The Earth is flat can be challenged on their underlying belief, You don't really believe that!, this does not challenge the fact that they have committed to it.
Following Farkas & Bruce (2010) and others, I employ speech act operators to model an utterance's effects on the discourse structure.PRESENT in ( 27), which is to be revised in Section 6.1, takes the at-issue content φ of a declarative sentence, the speaker a and the input DS K i as its arguments.This operator corresponds to Farkas & Bruce's (2010) A(ssertion) operator (somewhat simplified) plus evidential commitments.The label PRESENT is intended to capture the idea that a speaker using a declarative minimally presents φ for consideration by the discourse participants (Faller 2002).Its only hard-wired discourse effect is to put φ on the Table T as captured in (27i).In addition, it is associated with two defaults (indicated with parentheses in ( 27)) that update the speaker's truth and evidential commitments.When these apply, the overall force is that of default assertion (Farkas & Bruce 2010), that is, that of proposing φ for the CG, as in Figure 1.To be very clear on this point, proposing φ for the CG involves both (i) putting φ on the Table and (ii) being committed to its truth. 20This is further justification for tracking the individual participants' TCs, as it allows us to distinguish between declaratives that are used to propose φ and those that are not.( 27) PRESENT(φ , a, K i ) = K i+1 such that (to be revised) add φ to the truth commitments of a (iii) (AeC a,i+1 = AeC a,i ∪ {φ }) add φ to the evidential commitments of a A default update applies unless there is a linguistic element or contextual clue that overrides it.Reportatives, as we will see in Section 6.1, shift the truth commitment to a third party.Other markers have other effects.Thus, PRESENT serves as the starting point for the derivation of a variety of illocutionary forces, ranging from pure presentations to assertions.An example of a pure presentation, whose only effect is to put φ on the Table, would be a teacher presenting a proposition for debate by the class as in (28a).Evidential hedges override (27iii) but may leave (27ii) intact, as illustrated in (28b).Rising intonation as in (28c), discourse particles such as English maybe or Dutch wel (Zeevat 2006), and conjecturals override the default in (27ii) and replace it with a weaker commitment.
(28) a. Eating chocolate is unethical.Discuss.20 Farkas & Bruce (2010) include an additional component, the set of projected common grounds, to capture the fact that an assertion creates a bias for the asserted proposition.This component is however not independent and can "always be calculated from the current [CG] and the items on the Table" (Farkas & Bruce 2010: 89).This assumes that all at-issue propositions on the Table are proposed for the CG, but, as argued in Section 2.3, some at-issue propositions are not.I therefore do not include this component.

Assertion acceptance
Because the (idealized) goal of conversation is to resolve the QUD, there is "conversational pressure in the direction of increasing the [CG] by turning publicized commitments into joint commitments" (Farkas & Bruce 2010: 85).That is, an assertion of φ creates a bias for φ , making acceptance of φ the default response (Walker 1996, Gunlogson 2001, Farkas & Bruce 2010)).
In accepting φ , the addressee B publicly commits to φ .Since φ is now in the intersection of TC A and TC B , it is also in CG as shown in Figure 2. The QUD is taken to be resolved and φ is removed from the Table .B moreover acquires a reportative commitment, RepC B , given that their evidence is A's assertion.These updates are shown in Figure 2.
Output DS after the acceptance of φ = that the server is down by B Acceptance can be signalled in a variety of ways (one can nod, utter agreement phrases such as uh huh, sure, right, you bet, yup, repeat or paraphrase part of the original assertion, etc. (Walker 1996), but its default nature becomes most apparent in the fact that silence (that is, absence of disagreement) counts as acceptance (Stalnaker 1978, Walker 1996, Farkas & Bruce 2010, Asher & Lascarides 2003: 362).Walker (1996) attributes this to the Collaborative Principle in (29), which discourse participants are expected to observe.
(29) COLLABORATIVE PRINCIPLE: Discourse participants must provide evidence of a detected discrepancy in commitment as soon as possible.(adapted from Walker (1996: 269)) 21The Collaborative Principle comes into play whenever a participant conveys a (new) truth commitment to a proposition, as, for example, in Figure 1.It then falls onto their interlocutor to express any disagreement at the first point at which they can express their attitude.Failure to do so will be understood as implicit acceptance and their sets of commitments will be updated accordingly, as in Figure 2.
In Section 6.2, I will argue that the Collaborative Principle also plays a crucial role in deriving speaker commitment to a reported proposition.

Speech act operators and sentence types
In this section, I briefly discuss the choices made regarding the speech act operator PRESENT, though a full discussion of the issues raised is not possible.
It is a fairly standard assumption that declarative sentence type correlates, more or less directly, with the illocutionary force of assertion.( 27) however, associates declarative sentences with the force of presenting φ , from which assertion is derived only by defaults.Does this mean that assertion is no longer considered a basic illocutionary force? 22My answer to this is 'no'.While I claim that the only operation encoded by the declarative sentence type is putting the at-issue proposition φ on the Table, the fact that the assertive illocutionary commitments are added by default means that assertion is still the most basic force.Any other illocutionary force, including pure presentation such as (28a), has to prevent the defaults from applying or replace them with other conditions.
It might be argued that speech acts with weak commitments such as (28c) still count as assertions, and that the postulation of a separate type of presentation is not warranted.However, the label no longer fits, in my view, when speaker commitment is completely underspecified, and nothing is proposed for the CG, as is the case with RDSs.A tempting hypothesis to consider in this context is that the CQ reportative is a marker of sentence type.This would allow us to maintain the correlation between declarative and assertion and introduce a new one-to-one correlation between reportative sentence type and a reportative speech act.However, there is no evidence (such as the evidentials morphologically being part of the mood system) to suggest that this is the case for the CQ reportative.On the contrary, the fact that the reportative also occurs in interrogatives is evidence against this hypothesis, as one would have to postulate two distinct sentence types (and possibly two more for each of the other evidentials). 23I therefore assume that RDSs in CQ are indeed declaratives. 24 If one wants to maintain that declarative sentence type is correlated with a unique basic illocutionary force, one can adopt one of two strategies.First, one can associate declarative mood with a weaker speech act such as presentation, which only puts φ on the Table and from which more specific forces are derived, or one can broaden the meaning of assertion so as to encompass speech acts such as the ones performed with RDSs.(AnderBois's (2014) account discussed in Section 3.4 may be considered of this latter type.)I adopt the first approach, because I find it conceptually cleaner and because the term assertion is strongly associated in the literature with the idea of proposing the at-issue proposition for the CG (or, in non-dynamic approaches, with the idea of believing/knowing/comitting to φ (MacFarlane 2011)).
A related question is whether to use speech act operators at all, and not all authors do.Gunlogson (2001), for example, encodes discourse effects directly in the semantics of sentence type.Another approach is Portner's (2004), who postulates a generalized update function which matches types of sentence denotations with the corresponding discourse component: declaratives denote propositions which are of the right semantic type to be added to the CG, interrogatives denote sets of propositions, which are suitable for being added to the Question Set.Yet another approach is to appeal to mechanisms that do not participate in the compositional process.For example, Lauer (2013: 42, 90) postulates so-called extra-compositional conventions of use that constrain how sentence types can be used.An in-depth discussion of the best way of associating force with sentence type is beyond the scope of this paper, but I have here chosen to follow Farkas & Bruce (2010), Krifka (2014) and others in using speech act operators because having illocutionary force explicitly represented in the compositional apparatus is convenient for certain purposes.Thus, the PRESENT operator provides a hook on which to hang the assertive defaults.More importantly, having it explicitly represented makes it available as an argument for illocutionary modifiers such as the CQ reportative.More generally, conceiving of discourse moves/speech acts as consisting of a force element and a content explicitly recognizes the existence of an illocutionary level of meaning.
The tools introduced in this section will be used to develop the semantics of the CQ reportative in Section 6.1.To recap, I make use of truth (TC) as well as evidential commitment sets, where the latter come in a variety of different flavors depending on whether and what kind of evidential is being used (AeC, RepC, BpgC, etc.).Declarative sentence type is associated with the speech act type of presentation, which involves putting the at-issue proposition φ on the Table T. From this, assertion is derived by two default updates that add φ to the truth and evidential commitment sets.The addressee can accept or reject φ .If they accept it, φ becomes common ground.
it may be possible to treat evidential sentences as a different sentence type, thereby allowing the correlation between declarative and assertion to be maintained.

Goffman's speaker roles
Farkas & Bruce 2010, like every other formal account of assertion I am aware of, assume that assertive commitments are the speaker's, where the notion of speaker is taken as basic and not in need of further explication.However, the term speaker conflates distinct roles that can come apart in certain contexts.I argue in this section that the CQ reportative encodes a distinction between two such roles.I adopt Goffman's (1979) scheme in (30) as the best known. 2530) a. Animator: individual active in the role of utterance production b.Author: someone who has selected the sentiments that are being expressed and the words in which they are encoded c.Principal: someone whose position is established by the words that are spoken, someone whose beliefs have been told, someone who has committed himself to what the words say.(Goffman 1979: 17) While in most contexts all three roles are performed by the discourse participant standardly referred to as speaker, in some contexts they are distributed across two or even three people.For example, marriage vows or citizenship oaths are typically authored by one person (who is likely not present at the ceremony), but animated by another.The latter also purports to express their own sentiments and beliefs and is therefore also principal.When someone delivers a speech written by someone else and reflecting that person's thoughts and beliefs, we have a separation of animator on the one hand and author=principal26 on the other (Goffman 1979: 18, McCawley 1999: 601f).Examples of animator=author =principal include spokespersons or messengers.For example in (31), Thomas has a message for his sister Ali from their mother.(The relevant utterance is boldfaced).This, I argue below, is also the right constellation for reportatives.
(31) "Hey Ali" I yelled.She looked at me. "Thomas, why are you here" She said.
"Mom sent me, you never texted her" I said. (http://pokesonicfanthomas.deviantart.com/) Thomas might not have any evidence of his own in support of the proposition that Ali never texted their mother, only their mother's claim.While he likely believes it, by making it clear that he was sent by her, he transfers the assertive truth commitment to their mother.
The separation of roles in the examples above comes about exclusively through context.What I would like to suggest is that the CQ reportative and reportatives of the illocutionary type in other languages conventionally effect a separation of the principal role on the one hand and the animator and author roles on the other.This is most easily illustrated with messenger scenarios such as the one in (32).
(32) Context: Martina, the addressee's daughter, has disappeared, and nothing is known about her whereabouts for three days.On the fourth day, a messenger delivers the following message from Martina's father to the addressee: Martina-qa Martina-TOP pay-wan=si he-COM=REP ka-sha-n be-PROG-3 'Your child Martina is with him.' (Valderrama & Escalante 1982: 108) The message was sent by Martina's father, and it is his commitment that is conveyed, not the messenger's.That is, the father is the principal.The messenger is the animator, as they are acting as the "sounding box" (Goffman 1974(Goffman /1986: 517): 517), as well as the author, as they are choosing the words. 27There is no indication in the context whether the messenger is also committed to the claim that Martina is at her father's or not.Since the discourse participant physically producing the utterance is, by definition, always the animator, I will refer to them as animator from now on.
Reportative utterances are often made with the intention of resolving the QUD, as already shown with (3).( 33) is another example.It was uttered in reply to the addressee having asked where his son was.He correctly understood that the animator of (33) intended to resolve his question, went to the indicated puna and there caught up with his son.
(33) Context: A father is looking for his son and asks around for his whereabouts.
A man answers: Willermo Willermo Wich'i=s Wich'i=REP pusa-ra-n take-PST-3 wawa-yki-ta child-2-ACC ahay there kharu far Ollantaytampu-q Ollantaytambo-GEN puna-n-man.puna-3-ILLA 'Willermo Wich'i took him to the far puna of Ollantaytambo (I heard).' (Espinoza 1997: 18) In such cases, too, the reportative still encodes that a principal other than the animator has publicly committed to φ . 28The principal's commitment is in fact 27 At a minimum, a messenger needs to adjust any indexicals to the current speech situation.Thus, the addressee in both ( 31) and ( 32) is referred to with a second person pronoun.28 Evidence in support of this claim is the fact that the reportative cannot be used to report the content of utterances that do not involve commitment, for example, those of a learner of Quechua who practices pronunciation.According to McCawley (1999: 596) such utterances lack a principal.
primary and the source of the animator's own commitment.That is, all RDSs convey that a third party principal is committed to φ .The idea that the distinction between principal and animator is useful in the analysis of reportatives and reported speech more generally is not new.In Frame Analysis, Goffman himself already refers to reportatives when he writes: And in some languages -American Indian languages provide examples -the source29 of a reported action is established not by a pronoun-verb link but by a verb suffix.(Goffman 1974(Goffman /1986: 522) : 522) He mentions Wintu and Tonkawa as languages that mark the source with a verb suffix in footnote 21 (Goffman 1974(Goffman /1986: 522): 522).Wintu is well-known to have a complex evidential system, including a reported category (Schlichter 1986), and the relevant suffixes in Tonkawa are labelled "quotative" and "narrative" by Hoijer (1946).Levinson (1987: 185ff) elaborates this idea for reportative evidentials, and other authors (e.g., Grenoble 1998, Fetzer 2015) have used it in the analysis of (indirect) quotation more generally.
6 The account: CQ reportative encodes distinctness of animator and principal This section develops an account of the CQ reportative that accounts for the discourse properties of the reported proposition φ .In Section 6.1, the speech act operator associated with declarative sentence type PRESENT is revised so as to assign the truth commitment to a principal that may be distinct from the animator.The semantics of the reportative then requires that they are distinct.This accounts for Absence of Commitment.However, as will be discussed in Section 6.2, because the animator has introduced the principal's commitment into the discourse structure, the animator is subject to the Collaborative Principle and so will become committed to φ unless they immediately express disagreement with φ .This commitment is weaker than if they had committed as the principal, and Section 6.3 accounts for this in terms of Gunlogson's (2008) distinction between dependent and source commitments.

Absence of Commitment
As discussed in Section 5, in certain contexts, for example, messenger contexts such as (31), the roles of principal and animator come apart.We can capture this by tracking the truth commitments of the principal in addition to those of the animator.This is a fairly direct way of implementing Krifka's (2014) idea that the effect of certain constructions is "as if the speaker invites another person into the communication; the speaker acts as a proxy for that other source."Krifka was concerned with propositional attitude reports such as The weather report said there will be rain, where the prejacent is the main point and the main clause provides evidential support, but this characterization applies equally well, if not better, to reportative utterances.
The DS for the message in (31), you never texted her, is presented in Figure 3.It includes a set of truth commitments for the principal P, TC P , 30 as well as a reportative evidential commitment for A. The introductory clause in (31), Mom sent me, provides the contextual clue that P is distinct from A.
Output DS of A's message φ = that you never texted her A's commitment to the truth of φ is irrelevant and indeed not known and it can therefore not be presumed that A proposes φ for the CG.To capture the potential separation of the animator and principal roles, I propose the revision to PRESENT in (34).
30 Since it is A who introduces P's commitment into the discourse, I include TC P under A in the tableaux.
(iv) (a i+1 = p i+1 ) (34ii) assigns the truth commitment to the principal, 31 but (34iv) identifies the principal with the animator by default.This results in standard assertion because an animator who is also principal commits to φ and is therefore understood as proposing φ for the CG.In contrast, when a context such as (31) establishes that the two roles are played by distinct individuals, the animator does not publicly commit to φ and does not, in the first instance, propose φ for the CG. 32 We are now in a position to present the semantics of the CQ reportative in (35).Following Faller (2002), I analyze it as an illocutionary modifier, that is, a function from speech acts to speech acts.( 35) =si(PRESENT)(φ , a, K i ) = PRESENT(φ , a, K i ) such that (i) RepC a,i+1 = RepC a,i ∪ {φ } add φ to the reportative commitments of A (ii) a i+1 = p i+1 require A and P to be distinct 33   Illocutionary modifiers can override the defaults associated with speech act operators.(35ii) overrides the default in (34iii) and instead requires that the principal and the animator be distinct.As a result the truth commitment is attributed to the principal only, and the animator is not understood as proposing φ for the CG.In addition, the CQ reportative adds φ to the animator's set of reportative commitments RepC a .Assuming that each sentence can only introduce one type of evidential commitment, this prevents φ from being added to AeC, that is, (35i) overrides the default in (34ii).
Putting (34) and ( 35) together, the update contributed by (36) (= first sentence of (2)) is represented in Figure 4.The principal is not an element of the context in the same way a discourse participant is (it cannot be picked up by indexicals).I treat p as a free variable in (34) which can get instantiated in different ways.When it is not identified with the animator, the context may give a clue as to their identity as in ( 31) and ( 32), its value may be anaphorically dependent on a previously introduced principal (Murray 2010), or it may be existentially quantified.Goffman's (1979: 18) cases of "speaking for someone else", for example, by reading a deposition, or McCawley's (1999) example of a spokesperson are also covered by PRESENT: the contextual knowledge that the animator is officially speaking on someone else's behalf overrides (34iv).Note that, in these as well as messenger scenarios, it would be relevant to track the joint commitments of P and B rather than A and B, though I will not attempt to model this here.An anonymous reviewer points out that the principal is usually also not the addressee.We might therefore want to add this to clause (ii), though I will not do this here.This DS is nearly identical to the one in Figure 3, with the difference that the condition P =A is contributed conventionally and that A has a reportative commitment that is conventionally specified as such.This captures Absence of Commitment, as A does not express a truth commitment to φ when using the reportative.That φ is nevertheless at-issue and accessible for direct expressions of (dis)agreement by A herself or other participants is captured by φ being on the Table .Now, because A put φ on the Table with someone else's truth commitment, not her own (Figure 4), I hold that A is subject to the Collaborative Principle in (29), just like an addressee.That is, if A disagrees with φ , she needs to express this immediately.As we know, this is what she does.Interestingly, she uses the particle mana 'no', just like an addressee would, to introduce her denial in (37) (= second sentence of (2)).'(But) no, they didn't leave me one sol.'

A
(37) has the effect of replacing φ with ¬φ on the Table, but φ remains in TC P and RepC A .Assuming that A has best possible grounds for her denial (Faller (2012) argues that sentences without an evidential implicate best possible grounds), the resulting DS is Figure 5.As shown, ¬φ is on the Table and A committed to its truth, resulting overall in a proposal of ¬φ for the CG.
As discussed in Section 3.3, Murray 2014 also captures Absence of Commitment of RDSs semantically, namely by associating them with a vacuous proposal.The current account instead shifts the truth commitment to the principal.The two accounts are otherwise very similar.In both the reportative condition is directly imposed on the CG (by virtue of being not-at-issue truth-conditional content in Murray 2014, as a secondary effect in the current account, as shown in ( 26)) and the reported We are now faced with the question of how to bring the animator's truth commitment into the equation in such cases.The solution I propose is pragmatic and based on the Collaborative Principle in (29).
Let us first think through a scenario to justify the claim that the animator can sincerely commit to φ despite only having reportative evidence.Assume that A had a conversation with a reliable friend P who tells her, based on direct evidence, that A's son is now living in San Salvador.In the normal course of events, the proposition that A's son is living in San Salvador will have become a joint commitment of P and A (see Sections 4.2 and 4.3).Assuming that A was sincere in accepting φ into the CG of that conversation, she will as a result have acquired the belief that φ .She can therefore sincerely commit to the truth of φ herself in subsequent discourses.The question is how this commitment enters the DS.
A in (3) puts φ on the Table on the strength of somebody else's truth commitment (the DS is like Figure 4, but with φ = that he is in Salvador).This again evokes the Collaborative Principle.In this case, however, A does not deny the truth of φ .As a result, she is understood as accepting φ and becomes publicly committed to it, resulting in the update in Figure 6. 34 Output DS of reportative presentation and acceptance of φ = that he is in San Salvador by A 34 A prediction of this account is that an animator who knows that the principal was lying should not present φ with the intention of resolving the QUD.This is so because, on the presumption of sincerity, a truth commitment to φ implies a belief that φ , see Section 4.1, even if that truth commitment was incurred via the Collaborative Principle.Intuitively, this seems correct.A speaker who answered (3B) with He lives in San Salvador, I hear knowing that their source had lied and not making this clear, would rightfully be considered insincere.I thank an anonymous reviewer for raising this question.Note also that I assume that the Collaborative Principle does not apply to pure messengers, given that, as noted in footnote 32, their utterance is intended to update the CG between P and B. The question of their sincerity does therefore not arise.
but not if he answered with oh.Ben would not be held to the same standards as Amy, after indicating only a dependent commitment.I would like to suggest that the current set-up of tracking different types of evidential commitments in addition to truth commitments captures this difference between source and dependent commitments.An animator has a dependent truth commitment to φ if they have reportative evidence for φ , that is, if φ is in the intersection of TC A and RepC A .They have a source commitment if they have unspecified but adequate evidence for φ , that is, if φ is in the intersection of TC A and AeC A , or if they have a specified first-hand type of evidence such as best possible grounds, that is, if φ is the intersection of TC A and BpgC A .
To illustrate, in English one can assert Juan has a tractor even if one only has reportative evidence.Since this is a standard assertion, however, the adequate evidence default applies and puts φ in AeC, resulting in a source commitment.However, if the reportative evidence is overtly specified, as in Juan has a tractor, I hear, φ is in RepC, resulting in a dependent commitment.Should it turn out that Juan does not have a tractor after all, pointing to one's reportative evidence would only be a valid defense in the latter case.
In sum, the speaker's intention to resolve the QUD with a reported proposition is accounted for by assuming that an animator who brings a third-party principal's truth commitments into the discourse is subject to the Collaborative Principle and acquires a dependent truth commitment unless they immediately make it clear that they are not committed to φ .
An anonymous reviewer suggests that assigning the animator a truth commitment to φ in such cases is too strong, as they may only offer φ as the best answer available without necessarily fully believing it themselves.Instead, we might want to say that they are committed only to φ being more likely than ¬φ .This is a valid point, though I think it is (partly) addressed by the notion of a dependent commitment.Recall that truth commitments do not directly map onto beliefs, but only allow inferences about a participant's beliefs on the presumption of sincerity (Section 4.1).It seems reasonable to assume that a dependent commitment gives rise only to an inference of weak belief, given that it is supported only by reportative evidence.What is important from a discourse perspective is, however, not the animator's actual belief, but their intention to resolve the QUD with φ .That is, for the purposes of the discourse, unless an animator makes explicit that they only want to commit to a weaker proposition, they incur a truth commitment to φ .
Having presented my own account of RDSs, let me briefly return to a comparison with AnderBois's (2014) perspective shift account.The two accounts are similar in that both shift a core feature of assertion (commitment in the present account, proposal in AnderBois 2014) to a third party.The main difference is that this shift is pragmatic in AnderBois 2014 but semantic here.In the present account, RDSs entail that a third party is committed to φ in all its uses, not only in pragmatically rich shifting contexts.This seems intuitively correct to me, and I have no examples of reportatives that do not involve a third party principal's commitment.It is however difficult to test this.The English example Juan has a tractor, I hear, but nobody has publicly committed to this seems odd, but this may be due to the fact that explicit talk about discourse commitments is odd. 37In the absence of convincing data, I have to leave this for future research.
Regarding the two accounts' empirical coverage, recall that denials of RDS are typically accompanied by perspectival language (AnderBois 2014).In (2) ('They left me money (I was told), but they did not leave me one sol'), this is the surprise enclitic =má.The current account explains this as follows.A initially presents φ as a possible answer to the QUD but then denies that it is in fact the correct answer.In Goffman's terms, a shift in footing occurred between the two utterances: A positions herself as only the animator in the first utterance, but then speaks as principal in the second.It is expected that this change in footing is signalled with additional linguistic marking.
AnderBois's (2014) second empirical observation is that most and only reportatives allow for the overt denial of their prejacent.A pragmatic account avoids having to stipulate this in the lexical entries of reportatives.On the current semantic view, no extra stipulation is required either.Declaratives associate the truth commitment with the principal in all cases, but only reportatives require the principal to be distinct from the animator.I consider this to be part of the core-meaning of a reportative which needs to be specified independently of the denial cases.Apart from this, the semantics of evidentials is uniform in that all contribute a specific evidential commitment.Thus, while the current proposal may not adhere to AnderBois's (2014) Baseline Conception of evidentials, it nevertheless meets his other desideratum of a uniform semantics of evidentials.
The current account moreover accounts for the fact that the speaker commitment found with RDSs is weaker than the commitment associated with standard assertions or with assertions containing a direct evidential.

Conclusion
This paper addressed the puzzle in (1), namely that RDSs are in some cases used with Absence of Commitment and in others with the Intention to Resolve the QUD.The proposed solution relies on the following two independently motivated insights: Goffman's (1979) division of the concept of speaker into animator and principal (as well as author), and the observation that the absence of disagreement with an at-issue proposition typically signals its acceptance (Walker 1996).The original contribution of this paper lies in applying these ideas to the CQ reportative (and illocutionary reportatives in other languages).In particular, I proposed that the reportative conventionally encodes that the principal is distinct from the animator, as a result of which only the principal is committed to the truth of its prejacent.This accounts for the first part of the puzzle, that the animator need not be committed to the reported proposition φ .
Utterances of RDSs are not assertions and I therefore proposed that declarative sentence type is instead associated with the speech act of presentation.From this, standard assertion is derived via a set of defaults.The first default assigns a truth commitment to φ to the principal, the second requires that the animator have adequate evidence for φ , the third identifies the principal and animator.The latter two are overridden by the semantics of the CQ reportative.A key insight captured by this approach is that at-issue propositions are not necessarily asserted.This is not only true for reported propositions, but also the prejacents of propositional attitudes and other epistemic hedges.The assertive defaults may be overridden in a variety of ways, resulting in a range of illocutionary forces being associated with declaratives, from pure presentation to full assertion.Approaches that hard-wire the association between declarative sentence type and the speech act of assertion instead have to introduce mechanisms to weaken the assertive force to account for non-assertive uses.AnderBois (2014) achieves this for RDSs by introducing a pragmatic perspective shift, but this will not work for non-reportative hedges.Murray's (2014) account works for all types of hedges, but is problematic for the reasons discussed in Section 3.3.More research is needed to decide which of these two ways of approaching the association of illocutionary force with sentence type is ultimately to be preferred.
The key innovation of the proposed account that distinguishes it from previous semantic accounts of Absence of Commitment, such as Faller's (2002), Faller's (2007), and Murray's (2014), is the introduction of a truth commitment of a third party principal into the discourse structure.It is the presence of this commitment that enables us to account for the second part of the puzzle, the fact that the animator may nevertheless intend for φ to enter the CG.The animator's introduction of the commitment of a principal that is distinct from themselves triggers Walker's (1996) Collaborative Principle, requiring them to voice any disagreement with φ immediately.If they do not, the animator incurs a commitment of their own.This is, however, a dependent commitment, and the animator can therefore not be held accountable for the truth of φ in the same way as if they had made a standard assertion.
Of course, there is a host of remaining issues.I will only briefly mention two that seem fruitful for cross-linguistic study.First, recall that some reportatives have been analyzed as modals and as such their reportative condition should be able to be at-issue itself.We cannot assume that their prejacents have the same discourse properties as those of illocutionary reportatives.Thus, it is not obvious that they can be presented with the intention of resolving the QUD.If this is possible, then it seems reasonable to hypothesize that this may also be accounted for by the Collaborative Principle.Faller (2017) explores this issue for declarative sentences containing the German reportative modal sollen and suggests that they indirectly perform the speech act of presenting φ with another principal's truth commitment.This indirect speech act results from having asserted the sincerity condition of reportative presentations, that the speaker has reportative evidence, in the same way that asserting the sincerity condition of requests, that the speaker wants the requested event to happen (e.g., I want some tea) results in the indirect speech act of a request (Gordon & Lakoff 1975).The indirect speech act of a reportative presentation in turn triggers the Collaborative Principle.
Secondly, mirroring Searle's account of assertion and Grice's Quality maxim, I have treated truth and evidential commitments as distinct and unrelated sets, which are both needed in the performance of sincere assertions.As briefly discussed in Section 4.2, however, the two types of commitment are intuitively related in that a participant's type of evidence for φ determines the nature of their commitment to the truth of φ .Thus, having best possible grounds entails a strong commitment to φ , inferential grounds a weak commitment, and reportative grounds that a thirdparty principal is committed to φ .Conversely, in the absence of evidentials, it nevertheless seems to be possible to infer something about a participant's supporting type of evidence from their truth commitments.Thus, the indication of a strong truth commitment entails a commitment to at least adequate evidence, whereas a weak truth commitment (as, for example, indicated by epistemic modals) suggests a type of evidence that only weakly supports φ .Ultimately, it might be possible to make a typological distinction such that in evidential languages the speaker's evidential commitments are primary and their truth commitments derived, and in non-evidential languages the other way round.Much more research is needed to test this idea, and so I will leave it at just putting it on the Table.

'
-mu-sha-n-ku=ña.take-CISL-PROG-3-PL=DISC 'Then some of them hit him and kicked him, while others, right, were taking (his stuff).' (Radio program Warmikuna Rimanchis, 15 May 2002) of the bear arrived.He knocked on a door, silence.He knocked on another doorHe wants to be a congressman (I am told).'(Conversation,Faller 2007) ) are examples of Narration and (10) illustrates Contrast.

( 11 )
Context: Speaker A had seen two women arrive in their village the day before.She did not see what they brought to trade.'The women who arrived yesterday brought alcohol, balsam and arnica to trade, run and buy (them).'B: Ri-saq go-1FUT chay-hina=qa this-like=TOP hampi-ta cure-ACC ranti-rqa-mu-saq.

'
They left me money (I was told)'.
Output DS of reportative presentation of φ = that they left me money by A