(De-)constructing evidentiality

Available online at www.sciencedirect.com

ScienceDirect Lingua 186--187 (2017) 21--54 www.elsevier.com/locate/lingua

(De-)constructing evidentiality Rose-Marie Déchaine a,*, Clare Cook b, Jeffrey Muehlbauer c, Ryan Waldie c a

Department of Linguistics, University of British Columbia, 2613 West Mall, Vancouver, BC, Canada V6T 1Z4 b Northern Ontario School of Medicine, Canada c Verbulous, Inc., Canada Received 12 December 2013; received in revised form 19 September 2016; accepted 1 October 2016 Available online 6 December 2016

Abstract Extending Faller (2002), we analyze clauses with evidential marking as presenting, but not asserting, a proposition p. Crucial to this analysis is the distinction between common ground and origo ground. The common ground regulates p’s to which interlocutors have made a commitment and is subject to the logic of contradiction: p and NOT-p cannot hold at the same time. The origo ground regulates p’s that depend on a perspective-holder’s experience and is subject to the logic of faultless disagreement: p and NOT-p can hold concurrently, as long as they live in two distinct origo gounds. This has two consequences. First, languages differ in default illocutionary force: assertion versus presentation. Second, languages differ in how they code presentational force: lexically (English), morphologically (Nuu-chahnulth), or syntactically (Plains Cree). © 2016 Published by Elsevier B.V.

Keywords: Evidential; Assertion; Presentation; English; Plains Cree; Nuu-chah-nulth

1. Introduction Recent treatments of evidentials focus on the question of whether they are best analyzed as illocutionary operators or epistemic modals.1 In line with Murray (2010), we argue that the tension between these two approaches dissolves if one acknowledges that propositions introduced by evidentials may, but need not, update the common ground. This is best illustrated with an example. Speakers use evidentials to flag information they have access to, as distinct from other

* Corresponding author. E-mail addresses: [email protected] (R.-M. Déchaine), [email protected] (C. Cook), [email protected] (J. Muehlbauer), [email protected] (R. Waldie). 1 This work has had a long maturational process. Thanks to audiences at the University of Alberta (WSCLA 6), University of British Columbia (WSCLA 11), University of Ottawa (NELS xx), Leiden University (The Nature of Evidentiality Workshop), University of Arizona (WSCLA 20), as well as to E. Blain, S. Burton, T. Cardinal, yB. Crane, J. Crippen, H. Davis, J. Deschamps, D. Gerdts, J. Guéron, H. Harley, D. James, yD. Lightning, M. Keewatin, M. MacKenzie, L. Matthewson, A. Osa-Gomez, M.A. Palmer, D. Pentland, H. Rullmann, M. Rochemont, J.R. Ross, M. Speas, J. Tonhauser, M. Wiltschko, and H.C. Wolfart. http://dx.doi.org/10.1016/j.lingua.2016.10.001 0024-3841/© 2016 Published by Elsevier B.V.

22

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

commitments. An example of how evidentials achieve this is given in (1), which reports an exchange between a linguist and a Plains Cree language consultant.2 (1)

Linguist: How do you say ‘‘John saw Mary’’? Consultant: John ês ê-wâpamât Mary-wa Linguist parses sentence and is unable to identify ês: a. John ês ê-wâpam-â-t Mary-wa John ?? C-see.VTA-DIR-3>30 Mary-OBV Linguist: Can you say ‘‘John ê-wâpamât Mary-wa’’ b. John ê-wâpam-â-t Mary-wa John C-see.VTA-DIR-3>30 Mary-OBV Consultant: Well [YOUFOCUS] could Linguist is puzzled, and turns to other matters; later finds out that ês is the Plains Cree c. John ês ê-wâpam-â-t Mary-wa John REPORT C-see.VTA-DIR-3>30 Mary-OBV ‘Reportedly/supposedly/apparently, John saw Mary’

REPORTATIVE

evidential.

In (1), the linguist requests a translation of an English declarative sentence, and the consultant volunteers a sentence that has a reportative evidential. The inclusion of an evidential is appropriate, as the speaker does not have first-hand knowledge of the veracity of the proposition. The only felicitous update is one where the speaker indicates how they acquired the propositional content, here via hearsay. (1) highlights that speakers deploy evidentials to indicate how they acquire knowledge of p, as distinct from truth-claims about p. This is a defining feature of evidentials: they present, but need not assert, p (cf. Faller, 2002). Our presentative analysis, which claims that evidentials constitute a distinct presentative clause-type, makes the following contributions:  It models how participants present p without committing to the truth of p by distinguishing the common ground from the origo ground (Section 2). The common ground --- which regulates p’s to which participants have made a public discourse commitment --- is subject to the logic of contradiction: p and NOT-p cannot hold at the same time. The origo ground (og) --which regulates p’s that depend on participants’ experiential knowledge --- is subject to the logic of faultless disagreement: p and NOT-p can hold concurrently, as long as they inhabit distinct origo gounds, with p in ogA, and NOT-p in ogB.  It enriches the set of form-force pairings: in addition to requiring, asking, and asserting, speakers also present. We argue in Section 3 that while presenting p updates the origo ground, asserting p updates the common ground.  It establishes that, in some languages, the basic clause type is presentative. Languages differ according to whether they mark presentational force lexically (Section 4), morphologically (Section 5), or structurally (Section 6).  It contributes to research on the syntax--pragmatics interface relative to how participants’ commitments are integrated into discourse contexts (Section 7). 2. The problem Direct evidentials present propositions that are based on evidence directly accessible to the speaker through sensory modalities such as vision, hearing, and smell. Indirect evidentials present propositions based on evidence indirectly accessible to the speaker via inference, hearsay or quotation. Aikhenvald’s (2004) survey of evidentials identifies one locus of variation as being the number of terms that code direct versus indirect information sources. To see this, consider (1), adapted from Aikhenvald (2004:65). In some languages, e.g. Cherokee, there is a two-way contrast between DIRECT

Abbreviations used: 0 = 3rd person inanimate, 1 = 1st person, 2 = 2nd person, 3 = 3rd person, 30 = 3rd person obviative, A = addressee, = agreement, ANIM = animate, AUD = auditory, C = complementizer, cg = common ground, CONJ = conjunct mode, CONJR = conjectural, CONN = connective, CONT = continuative, DEM = demonstrative, DUBIT = dubitative, DIR = direct.evidence, EMPH = emphatic, EV = evidential force, EVID = evidential, FUT = future, ILL = illocutionary force, IMP = imperative, IND = indicative, INDF = indefinite, INDEP = independent mode, INDIR = indirect inter = interrogative, INV = inverse, LOC = locative, MOM = momentary, NA = noun animate, NEG = negation, NON.AFFIRM = non-affirmative, NUM = number, o = origo, og = origo ground, p = at-issue proposition, np = not-at-issue proposition, P = property, PERS = person, PL = plural, POSS = possessive, PROG = progressive, PROX = proximate, ps = presented set (Portner); ps = projected set (Farkas & Bruce), q = set of propositions, Q = interrogative marker, QS = question set, QUOT = quotative, REL = relative clause, REPORT = reportative RESUM = resumptive, SD = declarative sentence, SG = singular, SI = interrogative sentence, SINC = sincerity condition, SP = presentative sentence, TDL = to-do-list, TOP = topic, OBV = obviative, SAP = speech act participant (1st or 2nd person), SG = singular, SUB = subject, VAI = verb intransitive animate, VII = verb intransitive inanimate, VIS = visual, VTA = transitive verb animate VTI = transitive verb inanimate. 2

AGR

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

23

versus INDIRECT evidence; this is Aikhenvald’s Type A. But many languages have more than two evidentials; these correspond to Aikhenvald’s Types B (three evidentials), C (four evidentials), and D (five evidentials). Relevant to our concerns is that fact that both DIRECT and INDIRECT evidentials can be further partitioned. For non-sensory indirect information, there may be one INDIRECT form (Cherokee, Washo), a 2-way distinction between INFERRED/REPORTED (Quechua, Tucano), or a 3-way distinction between INFERRED/ASSUMED/REPORTED (Tsafiki, Tariana). And for direct sensory information, there may be one DIRECT form (Cherokee, Quechua, Tsafiki) or a 2-way VISUAL/NON-VISUAL distinction (Washo, Tucano, Tariana). Another widely cited classification is that of Willett (1988), whose ATTESTED evidentials correspond to Aikhenvald’s DIRECT evidentials. Willett also recognizes two other-types, INFERRING and REPORTED, but unlike Aikhenvald, he does not group them together. In other words, Willett’s classification does not recognize as a superset anything akin to Aikhenvald’s INDIRECT, which groups together INFERRING and REPORTED evidentials. (1)

TYPOLOGY OF DIRECT VERSUS INDIRECT EVIDENTIALS

(Aikhenvald, 2004; Willett, 1988)

(from Blain and Déchaine (2006), (4))

Though the direct/indirect partition is not accepted by everyone (Waldie, 2012), it allows us to talk about how evidentials can be challenged. To see how the problem presents itself, consider the Quechua paradigm, which has a fourway contrast between reportative -si, conjectural -cha, inferential -mi, and a zero-marked direct evidential3:

3 For the variety of Quechua discussed by Faller (2002), relative to Aikhenvald’s (2004) classification, -si presents REPORTED evidence, conjectural -cha and inferential -mi present INFERRED evidence, and zero-marked clauses present DIRECT evidence. As discussed by Faller, both inferential -mi and ‘direct’ zero-marking present eyewitness evidence; in Aikhenvald’s scheme, they present ‘‘direct’’ evidence. This is one of the ways in which there is a mismatch between Aikhenvald’s (2004) typological survey --- which focuses on morphological contrasts --- and analyses such as Faller’s that focus on the felicity conditions for evidentials. For related discussion and critique, see Blain and Déchaine (2006) and Waldie (2012).

24

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

Faller (2002) analyzes reportative -si as having presentative force: the speaker presents p by putting it forth for consideration. The sincerity condition associated with the reportative is that someone (s2) other than the current speaker or hearer asserts p. Faller analyzes the other two overt Quechua evidentials --- conjectural -chá and inferential -mi --- as having assertive force, with the speaker committed to the truth of p. The sincerity condition for conjectural -chá is that the speaker believes that p is an epistemic possibility and that this belief is based on reasoning. The sincerity condition for the inferential evidential -mi is that the speaker has the best possible grounds for believing p, which is eyewitness evidence. It is also possible for a clause to have no overt marking of evidentiality. Because this contrast is paradigmatically conditioned in Quechua, we treat it as a zero-marked evidential, and gloss it as DIRECT. Although Faller does not posit a zero-marked evidential, she does argue that the evidential value of such clauses arises via Gricean implicature: the speaker has the best possible evidence for p, namely eyewitness evidence. In Faller’s analysis, a sentence with a reportative has presentative force, while sentences with other evidentials have assertive force. The presentative/assertive contrast is based on whether p can be denied or canceled. The prejacent p of a sentence with presentative force (reportative -si) can be denied/canceled (3)a, but that of a sentence with assertive force (conjectural -cha, inferential -mi, zero-marked direct evidential) cannot be, (3)b--d.

Following Murray (2010), we take deniability/cancelation to be a diagnostic for which propositions are committed to: if the speaker is not committed to (the truth of) p, then the speaker can deny/cancel p. Assertions cannot be felicitously denied/canceled; thus, in uttering a declarative, the speaker is committed to p, and so denial of p yields a contradiction (4) a. The fact that a speaker can deny the prejacent p of a reportative evidential or adverb, as in (4)b, shows that the speaker is not committed to p. However, lack of speaker commitment does not entail that denial/cancelation is always possible. For example, the infelicity of denying the prejacent p of epistemic modal must, as in (4)c --- treated as an evidential in some analyses (Roberts, 2015; von Fintel and Gillies, 2011; Westmoreland, 1995, 1998) --- does not mean that the speaker is committed to p. Similarly, the infelicity of denying the propositional complement of the visual inferential look like (4)d does not entail speaker commitment to that proposition. We will argue that factors other than speaker commitment to p can make it infelicitous to deny/cancel p. (4)

a. a. b. c.

# It’s raining, but I don’t believe it. Reportedly, it’s raining, but I don’t believe it. # The keys must be in my backpack, but they’re not there. # It looks like it’s raining, but I don’t believe it.

There remains the question of why the prejacent p of reportative evidentials can be denied/canceled, but that of epistemic and inferential evidentials cannot be. In Faller’s analysis, this arises via a difference in illocutionary force. If the prejacent p associated with an evidential is not deniable, this indicates that p is asserted. If the prejacent p associated with

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

25

an evidential is deniable, this indicates that p is presented; the latter includes reportatives. But this begs the question of why reportatives differ from other evidentials, a problem that Faller (2002:200) acknowledges but does not solve. Though we adopt the deniability diagnostic, we claim that this test does not show that reportatives are presentational while other evidentials are assertive. Rather, we argue that all evidentials are presentational. In our analysis, the deniability test distinguishes propositions introduced into only the origo ground (reportatives, predicates of internal state and evidentials of internal state) from those introduced into the common ground (assertions) or both the common ground and the origo ground (direct and inferential evidentials). That some evidentials do a double update --- i.e. update the common ground and origo ground --- is confirmed by data from Nuu-chah-nulth, discussed below (see Section 5.2). A thread of analyses approaches evidentials through the lens of epistemic modality, and Matthewson (2011) goes so far as to claim that all evidentials are epistemic modals. In this approach, reportative evidentials are integrated into a modal analysis by distinguishing two kinds of conversational backgrounds (Kratzer, 2012), namely realistic versus informational. A realistic conversational background requires that there be evidence available to support p, while an informational conversational background requires a salient content source --- such as a report --- to support p. An informational conversational background makes it possible for a speaker to introduce a report of p that the speaker knows to be false. Whether framed in terms of illocutionary force or epistemic modality, extant analyses invoke special mechanisms for the analysis of reportatives. In this context, our goals converge with those of Murray (2010), namely: (i) to develop a general, crosslinguistic syntax and semantics for evidentials of all types; (ii) to do away with a special mechanism for the reportative. Rather than stipulating the divergent behavior of reports, we take them to be a diagnostic for the difference between presentation and assertion. In asserting p, a speaker introduces p into the common ground and commits to a truth-claim. But it is also possible for a speaker to present p without committing to a truth-claim. To allow for this possibility, we distinguish the common ground (cg) from the origo ground (og). The cg is the set of p’s --- both uttered p’s and background p’s --- to which participants make commitments. The origo is the individual from whose perspective p is evaluated, and the og is the set of uttered p’s for which the origo has experientially mediated information.4 Assertion updates the cg by introducing a p to which discourse participants make commitments; such a p can be endorsed or contested. But presentation updates the og by introducing a p for which the origo has an experiential basis. Because discourse participants do not make commitments to a presented p, such a p does not make it into the cg, and so can neither be contested nor endorsed. Consequently, when a speaker uses a reportative evidential to present p, though the addressee cannot endorse or contest p, they can acknowledge p. (And, as discussed above, a speaker can contest/ endorse their own presented p.) Before showing how the cg/og contrast accounts for the lexical, morphological, and syntactic correlates of evidentiality, we introduce the mechanics of our proposal. 3. The proposal Our main idea is that evidential clauses present p. Specificially, evidentials perform a speech act that presents p (Déchaine, 2007; Faller, 2002; Portner, 2006; von Fintel, 2003), and presentation has different kinds of commitments than assertion. We take as a starting point Portner’s (2006) idea that presenting is the most basic kind of conversational update. (5)

PRESENTED SET, ASSERTING AND PUTTING FORTH

(based on Portner, 2006)

4 See Garrett (2001) for an ORIGO-based analysis of Tibetan evidentials, which draws upon earlier work by Agha (1993). Space precludes discussion of the points of convergence and divergence.

26

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

The simplest construct is the ‘presented set’ ( ps), which is the set of p’s of which participants are mutually aware, (5)a. The common ground (cg) is a subset of the presented set, (5)b, and is the set of p’s to which participants have made commitments to; for example, they are committed to the truth-value of p.5 But other subsets are possible, in particular the set of p’s for which the participants have additional information --- e.g. relating to their evidential status --- these are p’s that are put forward ( put). These two sets, cg(ps) and put(ps), can be related to each other in a number of ways. For example, the set of p’s put forward can be a subset of the cg (5)c-i. Or the set of p’s put forward can be disjoint from the cg (5)c-ii.6 In the ensuing discussion, we equate Portner’s put( ps) with our og. Portner (2006) suggests that Quechua evidentials exemplify (5)c-i, where the put(ps) = og is a subset of the cg. This predicts that unmarked clauses default to the cg, i.e. they are assertive. But another possibility is for the put(ps) = og to be disjoint from the cg, as in (5)c-ii; Portner (2006:11) conjectures that in this case languages would have ‘‘a grammaticalized way of presenting a proposition -- of making one’s interlocutors aware of this proposition, without saying anything more about it.’’ This makes two predictions. First, some unmarked clauses default to the put(ps) = og and are purely presentative. Second, presented p’s can be incorporated into discourse structure without contributing to the cg. We integrate Portner’s insight (one can present p without asserting p) within a model of context structure. As a first step, we define assertion and presentation as follows: (6)

(7)

in asserting p, a discourse participant X (i) publicly commits to the content of p; (ii) proposes that p be added to the common ground7 in presenting p, a discourse participant X (i) publicly commits to the experiential grounding of p8; (ii) adds p to the origo ground

For assertion, following Farkas and Bruce (2010), we take a discourse structure K to represent the public discourse commitments of participants (Gunlogson, 2001; Hamblin, 1971) which are organized into two sets, DCX and cg, defined as in (8). (We henceforth refer to Farkas & Bruce as F&B.) We extend the F&B treatment of assertion to presentation. Accordingly, a discourse structure K can represent the public origo commitments (OC) of participants, which are organized into two sets, OCX and og, defined as in (9). (8)

(9)

DC p’s publicly taken as true by participants DCX (possibly empty) set of p’s for each participant X that X has publicly committed to, and which are not shared by the other participants cg set of p’s that have been confirmed by all participants (adapted from Farkas and Bruce, 2010) OC p’s publicly taken as experienced by participants OCX (possibly empty) set of p’s for each participant X that X has publicly presented evidence for, and which are not shared by the other participants og set of sets of p’s that have been presented by all participants, i.e. {{OCX1}, {OCX2}, . . .}}

Sentences differ in clause-type; this is indicated by the sentential features D (declarative), P (presentative), and I (interrogative). Following Hamblin (1971), F&B treat the denotation of a declarative sentence SD as the singleton set 5 Portner’s (2006) definition of COMMON GROUND as a subset of the PRESENTED SET departs from the view of the COMMON GROUND which includes background and contextually available information (Stalnaker, 2002). For Portner, background information is in the PRESENTED SET and includes the set of p’s of which participants are mutually aware. As we do not adopt Portner’s specific formulation, we put this matter aside. 6 A reviewer asks about a fourth possibility, namely having cg( ps) being a subset of the p’s put forward; this is not discussed by Portner. We believe that Plains Cree approximates this possibility in contexts of ‘‘mutual thinking’’; see Section 6.3 for discussion. 7 In discussing the relation of discourse commitments to the common ground (cg), Farkas & Bruce (2010) identify the following approaches:

(i) (ii) (iii) (iv) (v)

cg cg cg cg cg

= = = = =

a single set of shared public commitments (Stalnaker, 1978) union of the participants’ commitment sets (Gunlogson, 2001) set of FACTS (Ginzburg, 1996) ‘settled’ contents (Asher and Lascarides, 2003) set of propositions that have been agreed upon by all participants together with propositions that represent the shared background knowledge of the participants (Farkas and Bruce, 2010) We adopt the Farkas & Bruce conception of cg. 8 Experiential grounding includes both direct sensory experience and the experience of being told (i.e., hearsay). For relevant discussion, see Waldie (2012).

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

27

containing the at-issue proposition denoted by its sentence radical, (10)a. Extending F&B’s proposal, we treat the denotation of a presentative sentence SP as the singleton set containing the not-at-issue proposition denoted by its sentence radical, (10)b. Following Karttunen (1977), F&B treat the denotation of an interrogative sentence SI as the set of its complete answers, (10)c. (10)

a. declarative b. presentative c. interrogative

A novel feature of F&B’s analysis is that participants’ discourse commitments are distinct from the cg. This allows for an account of disagreement, in particular for cases where participants agree to disagree. Consider the following scenario: (11)

A: The French do sauces the best. B: No, they don’t; the Italians do. A: Well, we’ll just have to agree to disagree on that one.

Adopting F&B’s formalism, this exchange can be modeled as in (12). ‘The Table’ contains syntactic objects paired with their denotations, e.g. SD is a declarative sentence; we adopt the convention of shading the Table in gray. The items on the Table form a stack, the Table records what is at issue in the conversation, and a conversation is in a stable state when the Table is empty. A conversational move that tables a proposition projects a set of future cgs --- called the projected set ( ps) --- relative to which the issue on the Table is decided. Distinguishing the cg from the ps is crucial in resolving disagreement. In the initial context state K1, (12)a, no propositions are tabled. A tables p, (12)b; as a result A’s public discourse commitments (DCA) contain p, and the projected set also contains p. B then tables NOT-p (12)c; B’s public discourse commitments (DCB) contain NOT-p, and thereby eliminate p from the projected set. In addition, B also tables q, (12)d, so DCB contains {NOT-p, q}. In response, A tables NOT-q, (12)e; DCA now contains {p, NOT-q}, DCB contains {NOT-p, q}, and the Table contains {p, NOT-p, q, NOT-q}. This creates what Farkas and Bruce (2010:83) call a conversational crisis, which can only be resolved by retracting a commitment. In this scenario, A resolves the crisis with the ‘agreeing-to-disagree’ strategy, which removes all commitments from the Table, (12)f, thereby achieving a stable state. (12) a.

Successive context states for ‘agreeing to disagree’ K1: Initial context state

b.

K2: A puts p on Table relative to K1

c.

K3: B puts

d.

K4: B puts q on Table relative to K3

e.

K5: A puts

NOT-p

NOT-q

on Table relative to K2

on Table relative to K4

28

f.

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

K6: conversation achieves a stable state

In the AGREE-TO-DISAGREE scenario, no proposition makes it into the cg. However, propositions are added to the public discourse commitments of each participant, and they are sustained even if propositions are removed from the Table. This reflects the logic of the F&B analysis, which holds that conversations move forward if either the common ground is increased or a stable state is reached. We suggest that one additional factor is in play, namely the need to increase the origo ground. Accordingly, in our analysis, there are three (not two) engines that drive conversations: (i) reaching a stable state; (ii) increasing the cg; (iii) increasing the og. The latter is most easily detected in contexts of faultless disagreement (Kölbel, 2003) where what is at stake is the presentation of someone’s perspective or point-of-view. Consider the following scenario: (13)

Amy: For me, Montreal is the best city to live in. Beth: For me, Vancouver is the best city to live in.

Extending F&B’s formalism, this exchange can be modeled as in (14). The conversational goal of presenting a not-atissue p is to provide perspectival information. Such p’s are put in View by the origo, who is the perspective-holder. In the initial context state K1, (14)a, no propositions are asserted or presented: there are no public discourse commitments (DC), nor are there any public origo commitments (OC). A utters a presentative sentence, (14)b. As a result, A’s public origo commitments (OCA) contain np. B then presents nq (14)c, so B’s public origo commitments (OCB) contain nq. Although the origo commitments of A and B differ, no contradiction arises. Presentation of a proposition updates the og, and this update removes propositions from View. This allows the conversation to achieve a stable state: no propositions are on the Table, and none are in View, (14)d. (14) Successive context states for ‘faultless disagreement’ a. K1: Initial context state

b.

K2: A presents np relative to K1

c.

K3: B presents qNAI relative to K2

d.

K4: conversation achieves a stable state

3.1. How origo commitments interact with discourse commitments Distinguishing assertion from presentation allows us to model two kinds of disagreement: agreeing-to-disagree arises with assertion; faultless disagreement arises with presentation. With this in place, we now turn to the question of how discourse commitments and origo commitments interact. It is instructive to consider in more detail how assertions are

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

29

contested or endorsed. In (15), A puts p on the Table. In contesting p, B rejects A’s attempt to put p on the Table. Relevant is the fact that the form of the challenge can vary. In (15)a, A puts p on the Table, B removes p from the Table and replaces it with q. This is identical to the agreeing-to-disagree scenario discussed above except that the conversational crisis is resolved by removing only one of the commitments from the Table (rather than all of them). More interesting is (15)b. As before, B removes p from the Table, and puts q on the Table. But B does something else too: she presents evidence for q. And in (15)c, B requests that A present evidence for p. Finally, observe that while the attempt to Table p can be challenged, the act of asserting p cannot be, hence the infelicity of (15)d--e. (15)

a. b. c. d. e.

CONTESTATION PARADIGM A: Barack Obama was born in Kenya. B: No, that’s wrong, he wasn’t born in Kenya; he was born in Hawaii. No, you’re wrong, he wasn’t born in Kenya; I’ve seen his birth certificate, and it says he was born in Hawaii. No way! What makes you say that? #No, you didn’t say that. #No, that wasn’t said.

MOVE

(i) (ii) (i) (ii) (i) (ii)

contest p assert q contest p; present evidence for q contest p enquire about p contest illocutionary act

In presenting evidence, and in requesting that A present evidence, B is tracking public origo commitments: hers (when presenting evidence) or A’s (when requesting that A present evidence). Consider how to model the presentation of evidence, namely (15)b. The initial context state is given in (16)a. A puts p on the Table, (16)b, so A’s discourse commitments (DCA) now contain {p}. In (16)c, B removes p from the Table (accompanied by the retraction of p from the projected set), and brings nq into View (by presenting evidence for nq). In (16)d, the conversation returns to a stable state, as both the Table and the View are empty. (16) a.

Context states for (15b) K1: Initial context state

b.

K2: A puts p on Table relative to K1

c.

K3: B removes p from Table & puts qn in View, relative to K2

d.

K4: conversation achieves a stable state

The contestation paradigm shows that a challenge can involve removing p from the Table as well as putting other propositions in View. The latter can take two forms: (i) presenting evidence for an alternative to p; (ii) requesting that

30

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

evidence be presented in support of p. The same pattern is found with endorsement.9 Consider (17). Anne tables p. In English at least, as argued at length by F&B, endorsement of p is the unmarked option, and so p can be received in silence, (17)a. But as shown in (17)b, it is also possible for a participant to endorse p by committing to its evidential basis. In (17)c, B indicates that p is not in her origo commitments (That’s news to me!) and requests supporting evidence for p (What’s makes you say that?). And while the attempt to put p into the cg can be endorsed, the act of asserting p cannot be, hence the infelicity of (17)d--e. (17)

a. b. c. d. e.

ENDORSEMENT PARADIGM A: Barack Obama was born in Kenya. B: [Silence] Yes that’s right; I’ve even seen his birth certificate, and it says he was born in Hawaii. That’s news to me! What makes you say that? #No, you didn’t say that. #No, that wasn’t said.

MOVE

(i) (i) (ii) (i) (ii)

endorse p endorse p; present evidence for p flag p enquire about p contest illocutionary act

The analytic tools of F&B allow us to model assertion and presentation as updates to cg and og respectively. This approach also captures a difference in the logic of assertion and presentation. To see this, consider (18), an informal representation of how cg and og interact. With assertion, (18)a, participants attend to cg, and updates are endorsed or contested. In the parlance of F&B, a declarative sentence tables p, and if it is accepted by participants, then they are publicly committed to p. With presentation, (18)b, participants attend to og, and faultless disagreement holds sway, We adopt the convention of annotating the participants’ origo commitments as ORIGO GROUNDX. (In our implementation of F&S, ORIGO GROUNDX is equivalent to the origo commitments of X; i.e. OCX.) (18)

The difference between cg and og updates has consequences for how p and NOT-p are resolved. With assertion, p and leads to contradiction: one (or both) propositions are retracted from the cg. With presentation, p and NOT-p leads to faultless disagreement, with p and NOT-p sustained in their respective ORIGO GROUNDS.10 NOT-p

9 Analyses of cg updates often assume that silence is equivalent to strong endorsement of p, i.e. agreement. However, this is only one analytic possibility, and is not universally adopted as a cultural norm. As discussed in the ethnographic literature (Braithwaite, 1990; Tannen, 1984; Thomas, 1983), in some speech communities silence, rather than signaling endorsement, signals faultless disagreement. We also draw attention to the fact that, as diagnostic tests, the contestation and endorsement paradigms span utterances, while the deniability test is limited to a single utterance. This is because they are probing for different properties: contestation/endorsement probes for whether p is accepted into the common ground; deniabilitiy probes for whether the speaker is committed to the truth of p. 10 A reviewer suggests deriving commitment to the truth of p from evidential commitment. If presentative force is strengthened to the speaker ‘‘committing to having direct evidence for p’’, then the speaker would be committed to the truth of p. This (correctly) predicts that we would not find data of the form ‘‘p (direct evidence), but not p’’. Such a solution maintains truth-commitment, but does not capture the logic of faultless disagreement that arises with presentation: in presenting p, the speaker presents her experience of p without commiting to the truth of p. For related discussion on the ‘‘weak assertiveness’’ of English epistemic modal auxiliaries such as might and must, see Roberts (2015).

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

31

Drilling down further, there remains the question of which contexts permit faultless disagreement. We show below that evaluative predicates denoting personal taste (tasty, delicious) and internalized states (hungry, tired) are lexically specified to provide og updates, and so give rise to faultless disagreement. More generally, presentative force gives rise to faultless disagreement. Relevant is the fact that for many sentences marked with evidentials, the prejacent p is not part of the cg. Such p’s are not-at-issue: p cannot be challenged or endorsed, but the evidential basis for p can be. This can be shown in spontaneous utterances involving grammaticized evidentials. When a clause marked by an evidential is challenged, speakers present an alternative p as well as an alternative source of evidence for p. This subtle, but systematic, difference in how evidentials are challenged indicates two things: (i) speakers monitor the og; (ii) speakers are not committed to the truth-value of the p they present. Indeed, Faller (2002) presents a spontaneous example from Quechua that works in this way. Consider (19), where the speaker presents a reported p, and then presents NOT-p framed by an alternative (visual) source of evidence. The fact that speakers can switch from p to NOT-p without contradicting themselves confirms that p itself is not-at-issue (more on this below). In Plains Cree and Nuu-chah-nulth, we find the same pattern. An example from Plains Cree is given in (20), which shows that when rejecting a p marked with an evidential, speakers can present an alternative source of evidence for NOT-p. In both the Quechua and Plains Cree examples, in presenting p, the speaker is clearly not committed to the truth of p, as she also entertains NOT-p (on the basis of evidence). (19)

QUECHUA

They left me a lot of money. They didn’t leave me a lot of money.

Pay-kun-s noqa-man-qa [qulqi-ta muntu-ntin-pi saqiy-wa-n], 3SG-PL-EVID 1sg-ILLA-TOP money-ACC lot-INCL-LOC leave-1.OBJ-3 ‘They reportedly left me a lot of money, . . .’ mana-má riki riku-sqa-yki [ni un sol-ta centavo-ta-pis saqi-sha-wa-n-chu] not-SURP right see-PP-2 NEG one sol-ACC cent-ACC-ADD leave-PROG-1.OBJ-3-NEG ‘. . . but, as you have seen, they didn’t leave me one sol, not one cent’ (spontaneous; Faller, 2002:191, (152)) (20)

PLAINS CREE SCENARIO:

Somebody has told Ann that there are rumors that she has been given a lot of money by a recently deceased relative. In response to this, she says the following: A relative left me a lot of money. A relative didn’t leave me a lot of money. mistahi sôniyas êsa ê-miy-kawi-yân, pêyak awa ni-cisan ê-nipi-t much money REPORT C-give-PASS-1SG one D.PRX.ANIM 1-blood.relative C-die-3 ‘Reportedly, I was left with a lot of money by a relative that passed away; ‘ êkwa ôma ê-miyi-t êsa; and this C-give-3 REPORT ‘they left me this, reportedly’ mâka ki-wapâhtê-n ma-kîkway ôta, kâ-miy-kawi-yân but 2-see-SAP(VTI INDEP) NEG-thing here C-give-PASS-1SG ‘but as you can see here, they left me nothing’ 3.2. Form-force pairings Our analysis enriches form-force pairings to include clause-types that update the og, and whose sentential force is that of presenting p. This corresponds to the first row in (21), shaded in gray. Following Portner (2004), we assume a closed universal set of clause-types: in addition to declaratives, interrogatives, and imperatives, we posit the existence of presentatives. In addition to the discourse components of COMMON GROUND, QUESTION SET, and TO-DO-LIST, we posit an ORIGO GROUND. Besides asserting, asking, and requesting, presenting is a distinct sentential force. We predict the existence of clauses whose propositional content is not-at-issue, whose discourse component is the og, and whose sentential force is that of presentation. Evidentials represent the convergence of these properties.11

11 Portner (2004) acknowledges that morpho-syntax can contribute to meaning in a compositional way, but claims that illocutionary force is not one of the aspects of meaning it contributes. Nuu-chah-nulth and Plains Cree challenge this claim, as in these languages, morpho-syntax determines illocutionary force.

32

(21)

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54 ASSOCIATION OF CLAUSE-TYPE, DENOTATION, DISCOURSE COMPONENT AND SENTENTIAL FORCE

(adapted from Portner, 2004:238)

In Portner (2004), each clause type is associated with a different semantic type. For example, imperatives denote properties, and the to-do-list is a (function from an individual to a) set of properties. But it is not part of the semantics of imperatives to add a property to this set; rather, imperative force is deduced from its semantic type via pragmatic implicature. Similarly, according to Portner, the assertive force of declaratives is deduced from the fact that they denote atissue p’s, ‘‘and thus are suitable for being added to the Common Ground by the universal update function F.’’ (Portner, 2004:238f). Likewise, we take the presentative force of evidential clauses to be deduced from the fact that they denote notat-issue np’s, and so are suitable for being added to the origo ground.12 A central feature of our analysis is the bifurcation between at-issue versus not-at-issue content. Following Roberts (2015), we adopt the following definition of at-issueness and RELEVANCE: (22)

(23)

at-issuenessDEF For any proposition p, let ?p denote the question whether p, i.e. the partition on the set of worlds with members p and not-p. Then: A preferred proposition p is at-issue relative to a question Q iff ?p is RELEVANT to Q. (Roberts, 2015:9)13 RELEVANCE to the QUESTION-UNDER-DISCUSSION: Felicity of an utterance requires relevance to the QUD, where a. An assertion is RELEVANT to a QUD iff it contextually entails a partial or complete answer to the QUD. b. A question is RELEVANT to a QUD iff it has an answer which contextually entails a partial or complete answer to the QUD c. A directive is RELEVANT to a QUD iff its realization promises to play a role in resolving the QUD (Roberts, 2015:9)

In asserting p, an interlocutor drives the conversation forward by resolving a question-under-discussion. In our view, the presentation of p has a different goal, as it does not resolve a QUD. This is confirmed by the fact that p and NOT-p can be simultaneously entertained --- and shown in (19) and (20) above --- as long as their evidentiary base is distinct. While we concur with Roberts (2015) that the felicity conditions of at-issue utterances are adjudicated by relevance to the QUD, we maintain that what distinguishes an asserted p from a presented p is that the latter is not-at-issue. While most previous accounts of evidentials have addressed whether or not EV( p) is at issue (with different conclusions drawn for different languages/evidentials), we claim that p itself is not at issue (in addition to claiming that EV( p) is not at issue.) Specifically, in our analysis, the origo presents p for consideration by putting p into View. When this happens, a speech act is performed, namely that of presenting p. On this view, languages do not differ in whether they have grammaticized evidential morphemes. Rather, they differ in default illocutionary force and in whether they have morphemes that mark presentational force. With this in mind, consider (24). All languages have evaluative predicates that update the og; in Section 4 we illustrate this with English data. By hypothesis, such lexicalized presentatives are a property of Universal Grammar, and are likely the basis for other types of presentatives. For Nuu-chah-nulth, which exemplifies morphologized 12 See Murray (2010) for another way of analyzing form-force pairings: clause types are assigned a semantics, and structure the context so as to constrain (but not determine) the force of an utterance. 13 Roberts’ (2015) definition of at-issueness is adapted from Simons et al. (2010). For discussion of how to apply (not)-as-issue diagnostics, see Tonhauser (2012).

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

33

presentatives, we argue in Section 5 that clause-typing evidentials are root-bound and update the og. For Plains Cree, which exemplifies syntacticized presentatives, we argue that it has a dedicated clause-type with presentative force. (24)

FORM-FORCE PAIRINGS RELATIVE TO ORIGO GROUND

&

COMMON GROUND

4. English evaluative predicates as lexicalized presentatives Consider the conventional form-force pairings of English, illustrated in (25). In English, presentative force is lexically encoded with evaluative predicates such as tasty, hungry, or tired which update the og, (25)a. One indication that such predicates are origo-centered is the fact that, when they occur in root clauses, they are egocentric and restricted to a 1st person subject (more on this immediately below). Treating such predicates as having lexicalized presentative force captures the origo effects that characterize them (Kölbel, 2003; Lasersohn, 2005; Roberts, 2015; Snyder, 2013; Stephenson, 2007; Tenny, 2006).14 Elsewhere, English root clauses are declarative, have assertive force, and update the cg, (25)b. As for interrogative clauses, they update the question set (25)c, while imperative clauses update the to-do-list (25)d. (25)

ENGLISH

CONVENTIONAL FORM-FORCE PAIRINGS

The activity of the og with evaluative predicates is revealed by endorsement and contestation paradigms, origo shifts, and faultless disagreement. Let’s have a look. 4.1. Grounds for endorsing or challenging an evaluative predicate In all languages, evaluative predicates show the hallmark properties of evidentials. This can be illustrated with a predicate of internal state such as tired, which is lexically specified to introduce the origo’s perspective. For example, by virtue of volunteering information about her internal state, as in (26), A puts np in View and thereby updates the og. (26)

Anna: I’m tired.

og [ (np)

That an utterance such as I’m tired is an og update is confirmed by the fact that B’s acknowledgment of A’s contribution cannot take the form of agreement (27)a. This is because, although np has been put in View, it has not been put on the Table; in other words, np is not at-issue. Although B cannot endorse np (as it is not at-issue), B can acknowledge the origo of np, as in (27)b, or the evidentiary basis for np, as in (27)c. (27)

14

Beth: a. #Yeah, you’re tired. b. Well [YOUFOC] might be tired, but [IFOC] most certainly am not. c. Well you certainly [{LOOK, SOUND, SEEM}FOC] tired.

For related discussion see Egan et al. (2005), Egan (2007), MacFarlane (2005, 2011), and Moss (2015).

34

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

And just as np cannot be endorsed, neither can it be challenged, (28). This contrasts with non-evaluative predicates, (29). This difference reflects the distinction between not-at-issue presented np’s (which are put in View and added to og), and at-issue asserted p’s (which are put on the Table and added to the cg).15 (28) (29)

a. b. a. b.

Anna: Beth: Anna: Beth:

I’m tired. og [ {np} #No, you’re not! Lucy watched the football game. cg [ {p} No, she didn’t!

4.2. Default origo can shift The origo of evaluative predicates is syntactically conditioned. Consider first unmarked affirmative root clauses, where the default origo is the speaker. Observe that a 2nd person subject is unfelicitous (30)a, but that this can be repaired with an evidential (30)b, a diagnostic for egocentric evidentiality (Garrett, 2001). That this ‘‘1st person effect’’ reflects an og update is confirmed by the fact that imperatives, which are inherently 2nd person, pattern in the same way: an evaluative predicate is unfelicitous with an imperative (31)a, but can be repaired with an evidential, (31)b. With interrogative clauses, the origo is the addressee, so there is an flip in the felicity conditions, and now it is a 1st person subject which is anomalous, and which can be repaired with overt evidential marking (32). (30)

(31)

(32)

a. #You’re tired. og [ {np] b. You look/sound/seem tired. = ‘Given the sensory evidence available to me, You’re tired’ a. #Quick, be tired!16 b. Quick, look tired! = ‘Given the visual evidence available to me, tdlA {tired} a. #Am I tired? b. Do I look/sound/seem tired? = ‘Given the sensory evidence available to you, qs {I am tired; I am not tired}

In root clauses, the origo of an evaluative predicate is the speaker (33)a. But in embedded clauses, the origo is the subject of the matrix cause, (33)b--c. (Following Stephenson (2007), we indicate a link between the evaluative predicate and the origo by underlining.) Such origo shifts indicate that the origo index, even when not overt, is syntactically active. (33)

a. [SpeakerORIGO] The roller coaster was fun. b. [SamORIGO] thinks that the roller coaster was fun. c. Mary thinks that [SamORIGO] thinks that the roller coaster was fun. (adapted from Stephenson, 2007:24f.)

Evaluative predicates are sometimes accompanied by an overt origo index (Lasersohn, 2005; Stephenson, 2007). In English, this takes the form of a PP modifier ( for x, to x) that introduces a 1st, 2nd or 3rd person argument. It is always possible to introduce a 1st person origo (34)a, but a 2nd person requires modalization and contrast (34)a, and a 3rd person requires an external information source (34)c. (As we shall see, languages with a grammaticized origo index regiment such contrasts via syntactic clause-typing.) (34)

a. That ice-cream tasted good (to me). That hike was easy ( for me). b. That ice-cream might have tasted good to you, but to me it tasted awful. That hike might have been easy for you, but for me it was a nightmare. c. According to Beth, the ice-cream tasted good to Anne. According to Beth, the hike was easy for Anne.

The possibility of introducing an overt origo index is criterial, as it is the basis for the expression of faultless disagreement, to which we now turn.

15 16

For a different view of the challengeability of evaluative predicates, see Stephenson (2007). The infelicity of #Quick, be tired! cannot be attributed to a ban on stative imperatives, cf. Quick, be rude!

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

35

4.3. Faultless disagreement with evaluative predicates Evaluative predicates are the locus of faultless disagreement, which is a diagnostic for og updates. Consider the following examples, adapted from Stephenson (2007:27). In (35), A puts np in View, B puts nq in View, and participants sustain (potentially incompatible) origo commitments. This is more dramatic in (36), where A puts np in View and B puts n n NOT- p in View. This leads to faultless disagreement, with A maintaining p in her set of public origo commitments (OCA), n and B maintaining NOT- p in her set of public origo commitments (OCB). (35)

(36)

a. b. c. a. b.

Cathy: Ann: Beth: Cathy: Ann:

c. Beth:

How’s the cake? It tastes good to me. og [ ( pn} It’s a bit too sweet for my taste. og [ (qn} How was the party? It was fun for me, og [ ( pn} because I was able to catch up with old friends. Well it may have been fun for you, but for me, it wasn’t fun at all! og [ (:pn} In fact, it was boring because I didn’t know anyone there.

Evaluative predicates show the following cluster of properties: (i) they put p in View without tabling it, so p cannot be contested or accepted; (ii) they participate in origo shifts; (iii) they lead to faultless disagreement. We take this to indicate that they are lexicalized evidentials, and we expect to find this cluster of properties in all languages. 5. Nuu-chah-nulth C-typing evidentials are morphologized presentatives We now consider a language where presentative force is expressed via dedicated clause-typing morphology, namely Nuu-chah-nulth. In particular, there is a paradigmatic contrast between clause-types, which we argue codes a difference between presentative and declarative sentences (Section 5.1), which respectively provide og and cg updates (Section 5.2), and which participate in origo shifts (Section 5.3). 5.1. Nuu-chah-nulth contrasts clause-types Nuu-chah-nulth evidentials can be in the C-, I-, or V-domain (Waldie, 2012). C-domain evidentials --- quotative (37)a, indirect interrogative (37)b, dubitative (37)c --- are all bi-morphemic and part of the C-typing mood paradigm.17 I-domain evidentials --- inferential (38)a and past inferential (38)b --- are part of the mode paradigm that expresses modality/tense contrasts and occur to the left of C suffixes. V-domain evidentials are found with sensory percept evidentials that include the visual evidential (39)a, and the auditory evidential (39)b. (37)

a. [[m̓iƛ-aa IP]-wa-Ɂisˇ -Ø CP] rain-CONT-QUOTE-C-3 ‘It’s raining, according to what I’ve been told’ b. [[m̓iƛ-aa IP]-ḥ-ač-Ø CP] rain-CONT-C-INDIR-3 ‘Is it raining, according to what you’ve been told?’ c. [[m̓iƛ-aa IP]-q-ač̓-a CP] rain-CONT-C-INDIR-3 ‘It must be raining’

17 We analyze Nuu-chah-nulth clause-typing suffixes as C. A possible objection to this is that Nuu-chah-nulth has a non-affixal C in the form of Ɂin, and this might be the only authentic C in the language. However, non-affixal C is in complementary distribution with suffixal C-typing morphology; this is predicted if the latter are C-heads. While suffixal C is found only in root clauses, non-affixal C is found in two contexts: (i) in matrix clauses that lack suffixal C-typing (zero-marked absolutive mood); (ii) in complement clauses (again with zero-marked absolutive mood). For these reasons, we are confident that C-typing suffixes spell out C.

(i)

. . . Ɂin kw ̓ aḥsit . . . Ɂin ink̓w-ȧḥs-(m)it C fire-in.vessel-PAST ‘(Because) his lights were on’ (Waldie, 2012:41, (68b))

(ii) huḥtaks [Ɂin wiksuuk Ɂunaak ʕiniiƛ] huḥtak-s Ɂin wik-su:k Ɂu-nȧk ʕiniiƛ know-1SG C NEG-2SG TRANS-have dog ‘I know you don’t own a dog’ (Waldie, 2012:42, (69))

36

(38)

(39)

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

a. [[m̓iƛ-aa-matak IP]-Ɂisˇ CP] rain-CONT-INFL.IND.EVID-C.3 ‘Maybe it’s raining’ b. [[m̓iƛ-aa-ckwi IP]-Ɂisˇ CP] rain-CONT-INFL.IND.PST.EVID-C.3 ‘It must have rained’ a. [[m̓iƛ-aa-k̓uk IP]-Ɂisˇ CP] rain-CONT-PRED.VIS.EVID-C.3 ‘It looks like it’s raining’ b. [m̓iƛ-aa-Ɂisˇ CP] naɁaat rain-CONT-C.3 PRED.AUD.EVID ‘It sounds like it’s raining’

As shown in (40), Nuu-chah-nulth evidentials occupy different syntactic positions. C-domain evidentials are part of the mood paradigm and do not co-occur with each other. I-domain evidentials have the distribution of adverbs: they co-occur with each other, and they combine with C-domain evidentials. V-domain evidentials are verbs, and co-occur with both Iand C-domain evidentials.18 (40)

NUU-CHAH-NULTH SYNTACTIC DOMAINS FOR EVIDENTIALS

Nuu-chah-nulth C-domain evidentials are derived by adding a suffix to an already existing mood suffix. To see this, consider (41). The quotative evidential in (41)a is derived by adding -wa ‘say’ to the indicative suffix -Ɂisˇ , to derive -wa-Ɂisˇ . ̇ to form -ḥ-ač ̇ . And The indirect interrogative evidential in (41)b is derived by supplementing the interrogative suffix -ḥ with -ač ̇ to form q-ač-. ̇ 19 the dubitative evidential in (41)c is likewise derived by supplementing the subordinative suffix -q with -ač, (41)

ADDITIVE EVIDENTIAL MORPHOLOGY IN AHOUSAHT NUU-CHAH-NULTH

(from Nakayama, 2001; Waldie, 2012)

18 Nuu-chah-nulth visual and auditory evidentials differ in how they interact with other evidential classes (Waldie, 2012). The visual inferential does not co-occur with I-domain evidentials, and does not co-occur with the dubitative C-domain evidential. The auditory evidential has a lexically specified ORIGOSPEAKER, and does not occur with the past inferential I-domain evidential, or with the indirect interrogative C-domain evidential. 19 The form of the dubitative evidential is irregular in that it contains a supplementary final vowel; instead of the expected -q-ȧč we have -q-ȧč-a. Nevertheless, the basic generalization --- that Nuu-chah-nulth C-typing evidential morphemes are all minimally bi-morphemic --- still holds.

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

37

Nuu-chah-nulth presentative clauses are built from their counterpart assertive and interrogative clauses. Concretely, we analyze the C-typing morphemes as in (42), which group into pairs, with each member having a presentative and nonpresentative form. In terms of the present analysis, while ‘quotative’ -waɁisˇ puts np into view, indicative -Ɂisˇ puts p on the table, (42)a. And while the indirect interrogative -ḥ queries whether np or NOT-np are in View, interrogative -ḥ queries whether p or NOT-p are on the Table, (42)b. Finally, dubitative -q-ȧč-a does a double update --- by putting np into view (in the og) and p on the table (in the cg) --- while -q is the elsewhere case, and is found only in non-root contexts, (42)c.20 (42) a.i a.ii b.i b.ii c.i c.ii

SPECIFICATION



SPELL-OUT

$ $ $ $ $ $

-wa-Ɂisˇ -Ɂisˇ ̇ -ḥ-ač -ḥ -q-ȧč-a -q

C-TYPE ‘quotative’ ‘indicative’ ‘indirect interrogative’ ‘interrogative’ ‘dubitative’ ‘subordinative’

Crucially, Nuu-chah-nulth C evidentials only occur in root clauses, and contrast with I-domain evidentials, which occur in root or embedded clauses. Moreover, Nuu-chah-nulth provides a counter-example to the claim that only one evidential marker can be introduced per clause (Nuyts, 2009), as it permits evidential stacking.21 Examples are given below, showing that quotative wa-Ɂisˇ , a C-domain evidential, can combine with I-domain evidentials (43), or V-domain evidentials (44).22 We follow Waldie (2012) in taking evidential stacking as confirming that evidentials contribute not-atissue content; in such cases, they jointly restrict the same np. (43)

(44)

a. [[[haw̓iiqƛVP] -matak IP] -waɁisˇ CP] hungry -IND.EVID -3.QUOT ‘It is said that he must be hungry’ (adapted from Waldie, 2012:76 (116a)) b. [[[haw̓iiqƛVP] -ckwi IP] -waɁisˇ CP] hungry -PAST.EVID -3.QUOT ‘It is said that he must have been hungry’ (adapted from Waldie, 2012:77, (117a)) a. [[[haw̓iiqƛVP] -k̓uk.IP] -waɁisˇ CP] hungry VIS.EVID -3.QUOT ‘It is said that he looks hungry’ (adapted from Waldie, 2012:80, (124b)) b. [[[ɁuḥVP]-it IP]-waGiS CP Ken naʕaat maƛ-piƛ FOC-PAST-3.QUOT Ken AUD.EVID tied-in.house.MOM ‘It is said it was Ken who ended up in jail’ (Waldie, 2012:83, (131))

In sum, Nuu-chah-nulth C-typing evidentials have the following distribution: they are part of the mood paradigm, they are formed with additive morphology, they are restricted to root clauses, and they stack with other non-C-typing evidentials. 5.2. Origo ground and common ground updates in Nuu-chah-nulth Nuu-chah-nulth C-typing evidentials have additive morphology that supplements already existing forms of the mood paradigm. We take this to indicate that sentences default to cg updates, and it is only in the presence of overt evidential marking that a sentence updates the og. In other words, in Nuu-chah-nulth, put p on the Table is a default strategy, while

20 There are (at least) two ways of understanding the contrast between Nuu-chah-nulth simplex and complex Cs in (42): it could be that nonevidential ‘‘simplex’’ Cs are basic, and that evidential ‘‘compound’’ Cs are derived by additive morphology. Consequently, presentative force would be derived by suppressing non-presentative force; i.e. unless otherwise indicated, utterances would be evaluated as being at-issue. But the opposite could just as well be the case: namely evidential ‘‘compound’’ Cs are basic, and non-evidential ‘‘simplex’’ Cs are derived by truncation. Accordingly, non-presentative force would be derived by suppressing presentative force; i.e. unless otherwise indicated, utterances would be evaluated as being not-at-issue. In the main text, we adopt the hypothesis that evidential compound Cs are derived from non-evidential simplex Cs. We leave the resolution of this question to future research. 21 Also see Cornillie (2009) for a challenge to the claim that there is only one evidential per clause. 22 In Nuu-chah-nulth, the possibility of combining evidentials is subject to two constraints. First, C-domain evidentials don’t co-occur. We take this to be not only a paradigmatic constraint, but also a hallmark of evidentials that have a specified origo. Second, specific combinations of evidentials are blocked; for example the visual inferential k̓uk (introduced in the V-domain) does not co-occur with other inferentials; this blocking seems to be semantically conditioned.

38

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

putting p in View is a marked option, both pragmatically and morphologically. In (45), indicative -Ɂisˇ puts p on the Table and updates the cg; its evidential counterpart (reportative -wa-Ɂisˇ ) puts p in View and updates the og.

Just as there is a pairwise relation between declaratives and presentatives, there is also a pairwise relation between interrogatives and presentative-interrogatives (46).

As shown in (47), a polar question creates an inquisitive context by placing p and NOT-p on the Table; the ps contains both future cg’s to which p was added and future cg’s to which NOT-p was added. The most direct way of returning an inquisitive context to a stable state is for addressee B to publicly commit to p or NOT-p, and for this discourse commitment (DCB) to be accepted into the cg. (For reasons of space, context states corresponding to those moves are not shown.) (47) Context state for (46a): ?p a. K1: Initial context state

b.

K2: A puts {p, :p} on Table relative to K1

With the mechanics of polar questions in place, consider what happens with presentative-interrogatives. As shown in (48), a question with presentative force creates an inquisitive context. (The initial context is the same as (46a).) A places np and NOT-np in the View of B; the og contains B’s origo commitments to np and to NOT-np. The most direct way of returning such an inquisitive context to a stable state is for B to make a public origo commitment to np or NOT-np; i.e. to provide evidence for np or NOT-np. (48) Context state for (46b): Evid(?p) K2: A puts {p, :p} in View relative to K1

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

39

Consider the contrast that arises with the default C -q, (49): in non-presentative contexts it marks subordination; when supplemented with evidential morphology, it updates the og and cg, as in (50). A puts p on the Table and in View, so A has a discourse commitment (DCA) to p and an origo commitment (OCA) to np. p is resolved by being accepted or retracted from the Table (if challenged); np is resolved by being added to the og.

We now have a solution to the problem we started with. Recall that the prejacent p of reportatives, but not other evidentials, can be denied. In the present analysis, this contrast reflects the fact that presentatives must update the og, and in some cases, may update the cg. Not surprisingly, the class of evidentials that update both the og and cg are the ones that behave most like epistemic modals, and these are the ones whose prejacent p cannot be denied. (We return to this in Section 7.) Our og update analysis correctly predicts that a presented np cannot be directly challenged (as it is not put on the Table). Rather, what can be challenged is the evidential basis for presenting np; in other words, the evidence type can be denied. This is illustrated for quotative wa-Ɂisˇ in (51), for the inferential evidentials -q-ač̓a (dubitative) and -matak (indirect evidence) in (52), and for the past evidential -ckwi in (53). In each case, the participant challenges the evidentiary basis of the presented np. For example, in (51), Ken presents reportative evidence for np. Ann’s response implies that Ken should have direct evidence for np. Ann does not deny np, but she challenges Ken’s evidentiary basis for np. Similarly, in (52) and (53), Ken presents inferential evidence for np. In her response, Ann implies that Ken has direct evidence for np, and again challenges the evidentiary basis without challenging np.23 (51)

(52)

Ken: yaacpanačitwaɁisˇ Kay yac-panač-(m)it-wa-Ɂisˇ Kay walk-randomly-PAST-say-3.IND Kay ‘(Reportedly) Kay went for a walk’ Ann: wikɁiick t̓aaquaḥl ̴i Ɂin n̓aacsamitsa suw̓a Ɂukwink Kay ̇ wik-Ɂick t̓aqu-(q)ḥl ̴i[L] Ɂin n̓a:csa-(m)it-sa suw̓a Ɂu-čink[R] Kay NEG-2SG.IND truth-tell C see-PAST-1SG.ABS 2SG.PRO TRANS-with ‘You’re not telling the truth, I saw you with Kay’ Ken: wal ̴yuuqač̓a Kay OR wal ̴yuumatakɁisˇ Kay wal ̴yu:-qač̓a Kay walyu:-matak-Ɂisˇ Kay home-DUB Kay home-IND.EVID-IND Kay ‘Kay must be at home’ ‘Kay must be at home’

23 A reviewer asks whether the fact that an information source can be challenged indicates that the evidential itself is at-issue. Answering this turns on whether Ann’s response --- the Nuu-chah-nulth equivalent to ‘You’re not telling the truth’ --- is more akin to English That’s not true (which challenges at-issue content) or English Hey wait a minute! (which challenges not-at-issue content). Our analysis predicts that challenging the information source is akin to the Hey wait a minute! test. Confirmation awaits future research.

40

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

Ann: wikɁiick t̓aaquaḥl ̴i Ɂin n̓aacsamitsa suw̓a wal ̴aak ̇ wik-Ɂick t̓aqu-(q)ḥl ̴i[L] Ɂin n̓a:csa-(m)it-sa suw̓a wal ̴-ȧkw NEG-2SG.IND truth-tell c see-PAST-1SG.ABS 2SG.PRO go-DUR ‘You’re not telling the truth, I saw you go over there’ Ken: yaacwaasckwiɁisˇ Kay qwiyuuyii n̓iƛaak w yac-wa:s-ck i-Ɂisˇ kay qwiyu:-(y)ii n̓iƛaak walk-go.out-PAST.EVID-3.IND Kay when-3.INDF fight younger.sibling-KIN-KIN.POSS ‘Kay must have walked out when she was arguing with her younger sibling’ Ann: wikɁiick t̓aaquaḥl ̴i Ɂin n̓aacsamitsa suw̓a yaaɁal ̴ wik-Ɂi̇ck t̓aqu-(q)ḥl ̴i[L] Ɂin n̓a:csa-(m)it- sa suw̓a yaɁal ̴-[L] NEG-2SG.IND truth-tell c see-PAST-1SG.ABS 2SG.PRO watch-GRAD ‘You’re not telling the truth, I saw you watching’

(53)

5.3. Tracking origo shifts in Nuu-chah-nulth In (54)a, the report that it is raining is presented from the perspective of the origo qua speaker. And (54)b is a request for a report to be presented from the perspective of the origo qua hearer. This shift from speaker to hearer is called an origo shift.24 Origo shifts are a diagnostic for the origo argument (Garrett, 2001; Tenny, 2006), and in Nuu-chah-nulth they are syntactically conditioned in the same way as we saw for evaluatives.25 As summarized in (1), the origo index can be the speaker, the hearer, or be unspecified. (54)

(55)

NUU-CHAH-NULTH a. [[m̓iƛ-aa IP]-waɁisˇ CP] rain-CONT-3.QUOT ‘It’s raining, according to what I’ve been told’ b. [[m̓iƛ-aa IP]-ḥač CP] rain-CONT-3.INDIR.INTER ‘Is it raining, according to what you’ve been told?’ ORIGO ASSIGNMENT AND EVIDENTIALITY IN

ORIGO

= speaker

ORIGO

= hearer

NUU-CHAH-NULTH

(adapted from Waldie, 2012)

If an evidential lacks a lexically specified origo, we observe syntactically conditioned origo shifts. In Nuu-chah-nulth, this is found with inferential evidentials: the general inferential -matak, the past inferential -ckwi, and the visual evidential -k̓uk. The logic of the argument is as follows. With evidentials whose origo is not specified, the value of the origo is conditioned by syntactic environment. In matrix clauses, the default origo is the speaker; in dependent clauses, the default origo is the subject of the matrix clause. We illustrate this with the visual inferential -ckwi. Consider (56), where Kay

24

index also accounts for the viewpoint shifts associated with evidentiality (Speas, 2010). shifts are not specific to evidentials, but are also seen in the contrast between declaratives and interrogatives: a declarative has a speaker ORIGO (i), an interrogative has a hearer ORIGO (ii). For related discussion see Garrett (2001) and Roberts (2015). 25

(i) (ii)

The

ORIGO

ORIGO

The professor is Johan. Is the professor Johan?

ORIGO ORIGO

= speaker = hearer

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

41

(the speaker) presents np to Bill on the basis of visual inference. As expected, the origo for np is the speaker. (We use underlining to indicate a link between the origo and evidential.) (56)

ORIGO:

KaySPEAKER [SpeakerORIGO] [CP waɬyuu-k̓uk-Ɂisˇ Ken]] be.home-VIS.EVID-3.IND Ken ‘It looks like Ken is at home’ (Waldie, 2012:147 (243b))

That the speaker is the default origo is confirmed by the fact that if a syntactic antecedent is available --- in the form of the subject of a matrix clause --- the origo is the subject. This is illustrated in (57) and (58), where the presented np is embedded under the propositional attitude verbs ‘think’ and ‘say’ respectively, and the origo of np is the subject of the matrix clause.26 (57)

(58)

ORIGO:

LindaSUBJECT Ken]] [CP Ɂuql ̴aa-mit-Ɂisˇ Linda [CP waɬyuu-k̓uk-q think-PAST-3.IND Linda be.home-VIS.EVID-3.SUB Ken ‘Linda thought that it looks like Ken is at home’ (Waldie, 2012:147 (243a)) ORIGO: LindaSUBJECT [CP wawaa-mit-Ɂisˇ Linda [CP Ɂin waɬyuu-k̓uk Ken]] say-PAST-3.IND Linda COMP be.home-VIS.EVID Ken ‘Linda said that it looks like Ken is home’ (Waldie, 2012:149 (246))

For evidentials that lack a specified origo, when they occur in dependent clauses, the origo is never the speaker. This holds not only of complement clauses, as in (57)--(58), but also of adjunct clauses. Consider (59), where the visual inferential occurs in an adjunct reason clause: the origo of the presented np (Linda’s daughter knows how to make bread now) is the subject. (59)

ORIGO:

LindaSUBJECT [CP nučḥak-Ɂisˇ Linda [CP Ɂin ʕac̓ik-k̓uk-’aƛ-uk t̓an̓a saapn-q-iil ̴]] proud-3.IND LindaC know.how-VIS.EVID-now-POSS child bread-STEM-make ‘‘Linda is proud that it appears her daughter knows how to make bread now’ (Waldie, 2012:152 (251))

When the visual inferential -k̓uk occurs in a relative clause, as in (60), observe that the origo of the np (‘They are broke’) is, once again, the subject of the matrix clause. (60)

ORIGO:

LindaSUBJECT [CP Ɂan-aḥta-mit-Ɂisˇ Linda taana-q-ayi [CP yaq-k̓uk-ii ƛ̓icx ̣-Ɂakƛi]] only-do.to-PAST-3.IND Linda money-STEM-give REL-VIS.EVID-3.INDF faded-buttocks ‘Linda only gave money to those that appeared broke’

26 Embedded evidentials are generally introduced by propositional attitude verbs whose subject is the origo of the presented p in the embedded clause, so the evidential holds of the embedded utterance situation. This is not specific to our analysis, as some such mechanism is needed for the analysis of any embedded indexical, of which the origo is a special case. For related discussion, see Anand and Nevins (2004).

42

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

Nuu-chah-nulth clauses marked with evidentials pattern differently according to whether they occur in root or embedded clauses. Three factors govern their distribution: whether they are C-typing evidentials, whether they are restricted to root clauses, and whether they have a lexically specified origo. The same factors that determine origo resolution with evaluative predicates --- which we analyze as lexicalized presentatives --- also obtain with morphologically marked evidentials. This is consistent with our hypothesis that evidentiality is first and foremost a mechanism for updating the og. 6. Plains Cree indexical clauses are syntacticized presentatives Our analysis predicts that, in some languages, the unnmarked clause type is presentative, and the default illocutionary force is that of presenting. Plains Cree is such a language: it has unmarked root clauses with presentative force that update the og and introduce the direct experience of the origo. This is of some interest because it challenges the widelyheld assumption that assertion is the default illocutionary force. Before diving into Plains Cree evidentiality, we introduce nomenclature relating to morphology, syntax, and context-of-use. Consider (61). Plains Cree distinguishes clause-types in terms of two paradigms, traditionally called the independent mode and conjunct mode. These two paradigms show a syntactic partition (Cook, 2008, 2014): forms in the independent mode are indexical clauses (restricted to root contexts); forms in the conjunct mode are anaphoric clauses, and appear in a wide range of syntactic contexts. In terms of context-ofuse, the default illocutionary force of indexical clauses is presentative, that of anaphoric clauses is (weakly) assertive.27 (61)

PLAINS CREE ROOT CLAUSES

The two verb paradigms found in Plains Cree are illustrated in (62) with the intransitive verb stem miyosi- ‘be.good’. Indexical clauses combine prefixal 1st and 2nd person agreement, with suffixal person/number agreement. Anaphoric clauses have prefixal C-typing (null, ê-, or kâ-) and suffixal person/number agreement. (62)

PLAINS CREE INTRANSITIVE VERB PARADIGM

(cf. (Wolfart, 1973)

The distribution of these two paradigms is conditioned by a combination of syntactic, semantic, and pragmatic factors. Indexical clauses can only be root clauses, but anaphoric clauses can be root or embedded clauses (Wolfart, 1996). This means that Plains Cree has two kinds of root clauses. This raises the question of what distinguishes them from each other. The contrast between indexical and anaphoric clauses reflects a structural difference in the C-domain (Cook, 2008, 2014; Déchaine and Wiltschko, 2015). While indexical clauses introduce indexical 1st and 2nd person agreement in Spec,CP (63), anaphoric clauses introduce a null operator in Spec,CP and clause-typing elements in C, (64). 27 Plains Cree discourse-linked indexicals are only found in root clauses; elsewhere they are anaphoric. This has consequences for the deployment of 1st and 2nd person pronominal agreement. English has no (surface contrast) between a discourse-initial indexical mention of a first person: the same form occurs in root clause and embedded clauses; e.g., I said that I like beans. In Plains Cree, the independent mode only introduces a truly indexical 1st person; an anaphoric 1st person is introduced by the conjunct mode. (Same thing for 2nd person.) This is the motivation for Cook’s (2008, 2014) characterization of these clause types as indexical versus anaphoric. For discussion and analysis of how the indexical/anaphoric contrast plays out in other areas of the grammar, see Cook (2008, 2014) and Déchaine and Wiltschko (2015).

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

(63)

INDEXICAL

(INDEPENDENT (64)

ANAPHORIC

(CONJUNCT

[CP

PERSON-

[CP

nikiOP

MODE)

MODE)

[ CØ

VSTEM. . .]

43

-AGRPERS/NUM -AGRNUM]

[ C- [VSTEM. . .] -AGRPERS/NUM -AGRNUM] Ø êkâ-

With this background information in place, we introduce the evidential morphemes attested in Plains Cree, examine Plains Cree unmarked root clauses, and discuss the broader implications of the presentative analysis. 6.1. Plains Cree evidentials and their morpho-syntax We argue that the indexical/anaphoric contrast found in Plains Cree reflects a general contrast between clauses with presentative versus assertive force. In particular, indexical clauses behave like direct evidentials, in that they put into View propositions related to the origo’s direct experience. Curiously, no specific morpheme indicates that they are evidential in nature. Moreover, one canot appeal to a null morpheme analysis, as there is no paradigmatically conditioned alternation. This is especially striking if one compares Plains Cree to Cheyenne: both are Algonquian languages, but only the latter has a full-fledged evidential paradigm (Leman, 1980), which contrasts a zero direct evidential with three overt evidential morphemes, namely reportative -nėse, quotative -neho, and inferential mo´. . .hané-he: (65) a.

b.

c.

d.

CHEYENNE EVIDENTIAL PARADIGM28 (adapted from Murray, 2010:21, (2.15)) É-hoo’koho-Ø 3-rain-DIR ‘It’s raining, I’m sure’ (Leman, 1980:43) É-hoo’kȯho´-nėse 3-rain-REPORT.SG.ANIM ‘It’s raining, I hear’ (Leman, 1980:51) É-hoo’kȯ’ho´-neho 3-rain-NARRATIVE.SG.ANIM ‘It rained, it is said’ (Leman, 1980:53) Mo´-hoo’kȯho´-hané-he CNJ-rain-NON.AFFIRM-Q ‘It’s raining, I gather’ (Leman, 1980:50)

Plains Cree is innovative in that it has re-structured the C-typing system that prevails in other Algonquian languages relative to evidential marking. In particular, evidential mood suffixes have been re-analyzed as clitics and are introduced in the C- or I-domain (Blain and Déchaine, 2007) While quotative itwê (66)a and reportative êsa (66)b are C-domain evidentials, dubitative êtokwê (67) is an I-domain evidential. Plains Cree also has V-domain evidentials (68) that introduce (visual and auditory) inference.29 (66)

(67)

a. ê-kimiwahk itwêw C-rain(VII CONJ 3) thus(VTA INDEP3) ‘S/he said it’s raining’ b. ôkimâhkan ê-kîpikskwâtât Jeff-a, êsa chief C-PAST-speak(VTA CONJ3) Jeff-OBV REPORT ‘Reportedly the chief spoke to Jeff’ êtokwê môy ê-kimiwahk Vancouverihk DUBIT NEG C-rain(VII CONJ3) Vancouver.LOC ‘Apparently it’s not raining in Vancouver.’

28 The glossing of the Cheyenne paradigm, which for consistency we have adapted from Murray (2010), is based on data from Leman (1980); the latter does not provide a morphological analysis. 29 Plains Cree êsa is described as a reportative in the literature, but in some contexts it functions as an inferential or mirative. The construal of êsa is sensitive to clause-typing, in particular to the contrast between independent and conjunct mode clauses. Space considerations preclude further discussion here; see Cook (2008, 2014) for details.

44

(68)

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

a. isi-nâ-kw-an ê-kimowahk thus-by.vision-INV-VII(INDEP) C-rain(VII CONJ 3) ‘It looks like it’s raining.’ b. it-ihtâ-kw-an ê-kimowahk thus-by.hearing-INV-VII(INDEP) C-rain(VII CONJ 3) ‘It sounds likes it’s raining.’

Except for the direct evidential, Plains Cree maintains the same contrasts as Cheyenne. Indeed, these morphemes are historically cognate: (69)

EVIDENTIAL MORPHEMES IN PLAINS CREE

&

CHEYENNE

Does Plains Cree have the equivalent of the zero-marked ‘‘direct evidence’’ morpheme that Cheyenne has? We think not. Rather, a radical shift has occurred in the organization of the Plains Cree evidential system. Consider (70), which shows how the two clause-types (indexical and anaphoric) combine with evidential morphemes. For each clause-type, there are four possibilities: the clause by itself, and in combination with êsa, itwê, and êtokwê. What emerges is that indexical clauses resist combining with evidential morphemes in their canonical usage. Instead, they impose an additional restriction. The presented np --- whether reported, quoted or inferred --- must have been directly experienced by the origo. This is the first of many clues that Plains Cree has shifted over to a system where the default illocutionary force is presentative. (70)

PLAINS CREE EVIDENTIAL MORPHEMES

&

CLAUSE-TYPES

6.2. Plains Cree clause-typing and the origo ground We propose that Plains Cree distinguishes presentative and assertive force via syntactic clause-typing. Indexical clauses present the direct unmediated experience of the origo, and are felicitous only in contexts where the og is updated, (71)a. Anaphoric clauses assert the propositional content of the clause that that they introduce and so (by default) update the cg, (71)b. In addition, interrogative clauses are marked by a yes/no question particle cî, and have the force of questioning (71)c. Notably, such yes/no questions are formed with indexical clauses. This indicates that the speaker is requesting that the hearer select np from the og and so requires that np be based on direct experience. Finally, imperative clauses, with the force of requiring, are marked by a distinct morpho-syntactic paradigm (71)d.

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

45

Even in the absence of overt evidential marking, Plains Cree indexical clauses are presentative: in other words, they always put np in View. Several diagnostics support our claim that Plains Cree indexical clauses have presentative force: they are utterance-bound, they are always and only root clauses, they do not support anaphoric dependencies. 6.2.1. Plains Cree indexical clauses anchor to the utterance situation Indexical clauses anchor to the utterance situation: such clauses are evaluated relative to utterance time, utterance location, and speaker. In the following data sets, the target clause is bold-faced, with the (a) example being an indexical clause, and the (b) example being an anaphoric clause. Consider the interaction with utterance time. The temporal resolution of root clauses that are not overtly marked for tense differs according to whether they are indexical or anaphoric. As shown in (72), while indexical clauses index the utterance time, anaphoric clauses index the utterance time or a time preceding the utterance time. (72)

a. ni-pêhtawâ-w nisîmis wayawihtam-ihk ê-mêtawêt 1-hear-3(VTAINDEP) 1-sibling outside-LOC C-play-3(VAI CONJ) = (i) ‘I hear my little brother playing outside’ ≠ (ii) ‘I heard my little brother playing outside’ COMMENT: . . .like I’m on the phone with you, and he’s making noise, and I’m telling you about it (Cook, 2008:95, (71a)) b. &ê-pêhtaw-ak nisîmis wayawihtam-ihk ê-mêtawêt 1-hear-1>3(VTACONJ) 1-sibling outside-LOC C-play-3 (VAI CONJ) = (i) ‘I hear my little brother playing outside’ = (ii) ‘I heard my little brother playing outside’ COMMENT: This one could mean any time; it could mean before, or it could mean I’m hearing him now (Cook, 2008:95, (71a))

A similar contrast holds for anchoring to the utterance location. Indexical clauses anchor to the utterance location; as such they are compatible with the deictic locative êkotê ‘there’ (73)a, but not with the indefinite locative ita ‘(some)where’ (73)b. In contrast, anaphoric clauses are compatible with both deictic locatives, (74)a, and indefinite locatives, (74)b. (73)

(74)

a itohtê-w êkotê go-3(VAIINDEP) there ‘S/he went over there’ (cf. Cook, 2008:75, (33a) b. *ita itohtê-w LOC go-3(VAIINDEP) [INTENDED: where s/he went’] a. ê-itohtê-t êkotê C-go-3(VAICONJ) there ‘S/he went over there’ b. ita kâ-itohtê-t LOC C-go-3(VAICONJ) ‘where s/he went’

In addition to anchoring to the utterance time and utterance location, indexical clauses also anchor to the utterance speaker. Because the role of the utterance speaker is grammaticized by 1st person agreement, this gives rises to a number of subtle, but systematic, origo-centered effects. There are three cases to consider: predicates whose lexical semantics require the inclusion of the origo’s perspective, those that exclude the origo’s perspective, and those that are neutral with respect to the origo’s perspective. In each case, there is a contrast in the deployment of indexical versus anaphoric clauses. Recall that we analyze evaluative predicates --- which include predicates of internal state such as hungry and tired --- as lexicalized presentatives that provide an og update. Consequently, for English, the contrast between (75)a and (75)b is a difference between an og update and a cg update. Our analysis of Plains Cree (correctly) predicts that the og/cg contrast will correlate with a contrast between indexical clauses (that update the og) and anaphoric clauses (that update the cg). We this in mind, we illustrate how different predicate types condition the typing of root clauses in Plains Cree. (75)

a. I’m tired. og [ (np) b. I’m the first speaker tomorrow. cg [ {p}

In Plains Cree, a predicate of internal state such as nêstosi- ‘be tired’ presents the origo’s perspective, and can only be uttered by a conscious participant. As shown in (76), with a clause that is inflected for 1st person, predicates of internal

46

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

state are felicitous only with indexical clauses. This is consistent with our claim that indexical clauses have presentative force and update the og. (76)

a.

ni-nêstosi-n 1-tired-SAP(VAI INDEP) ‘I’m tired.’ (i.e., ME, HERE, NOW) b. # ê-nêstosi-yân C-tired-1(VAI INDEP) ≠ ‘I’m tired.’ = ‘I was tired at that time.’ COMMENT: It’s referring to when you were tired. It seems like if you were talking about it, and not being tired.’ (Cook, 2007, (1))

The Plains Cree counterpart to ‘‘I’ll be the first speaker tomorrow’’ also shows a sensitivity to clause-typing. An indexical clause is felicitous in a context where the speaker is co-present in the situation upon which the np being presented is grounded, (77)a. And an anaphoric clause can be felicitously uttered if the speaker presents a proposition that is grounded on a situation where she was not co-present, (77)b. (77)

a. n-îya, nîkân n-ka-pikswâ-n wâpahk-i 1-PRO first 1-IRR-speak-SAP(VAIINDEP) be.next.day-C(VII SUBJUNCTIVE) ‘I will be speaking first tomorrow’ b. n-îya, nîkân ê-wî-pikswa-yân wâpahk-i 1-PRO first 1-FUT-speak-1SG(VAICONJ) be.next.day-C(VII SUBJUNCTIVE) ‘I will be speaking first tomorrow’

Since indexical clauses anchor to the utterance situation, and in particular, to the utterance speaker, this predicts that indexical clauses will not be felicitous with predicates like wantipskinê- ‘lose consciousness’ that exclude the origo’s perspective. As shown in (78), this prediction is confirmed. (78)

a. # ni-wantipskinê-n 1- lose.consciousness-SAP(VTI INDEP) b. ê-wantipskinam-ân C-lose.consciousness-1(VTICONJ) ‘I lost consciousness.’ (Cook, 2007, (4))

Even with predicates that are neutral with respect to the value of the origo, for example, pahksini- ‘fall’, we see that their context-of-use is conditioned by clause-type. With a 1st person subject, while an indexical clause presents the direct experience of the speaker as conscious participant (79)a, an anaphoric clause makes no claim about direct experience and so the speaker is not a conscious participant, (79)b.30 This contrast is also found with the 3rd person: indexical clauses are felicitous if the speaker is co-present (e.g. as an eyewitness) (80)a; anaphoric clauses are felicitous if the speaker is not co-present (80)b. (79)

30

a. ni-pahksini-n 1-fall-SAP(VAIINDEP) ‘I fell.’ b. ê-pahksini-yân C-fall-1(VAICNJ) ‘I fell.’

Blain and Déchaine (2006) discuss such effects in terms of ‘‘accessibility to consciousness’’.

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

(80)

47

a. miyomâciho-w Anna feel.well-3(VAI INDEP) Anna ‘Anna’s feeling well.’ COMMENT: seems like you’re getting that from seeing her and looking at her b. ê-miyomâciho-t Anna. c-feel.well-3(VAI CONJ 3) Anna ‘Anna’s feeling well.’ COMMENT: ê-miyomâcihot is more like you’re hearing about it (Cook, 2007, (8))

6.2.2. Plains Cree indexical clauses are always and only root clauses If independent mode clauses are indexical, and if their indexicality requires them to anchor to the utterance situation, this predicts that they will occur only in root clauses. To see why, consider (81). An indexical clause anchors to the utterance context ({ ) (81)a, while an anaphoric clause is anaphoric relative to another clause (81)b. If a language has dedicated root-clause morphology, as Plains Cree does, then we expect to have a ‘‘root-bound’’ clause-type; that is, a clause-type that is licit only as a root clause. (81)

a. {i [CPi . . .] b. {i [CPi . . . [CPi . . .]

Indeed, the independent mode, as an indexical clause, is ‘‘root-bound’’: it can be a root clause (82)a, but not a subordinate clause (82)b. This contrasts with the conjunct mode, which occurs as a root clause (83)a or as a subordinate clause (83)b. (82)

(83)

a. [CP kimiwan] rain.3(VIIINDEP) ‘It’s raining’ b. *ni-wîhtamawâ-w [CP kimiwan] 1-tell-3(VTA INDEP) rain.3(VIIINDEP) [INTENDED: ‘I told him/her that it’s raining’] a. [CP ê-kimiwah-k] C-rain-3(VIICONJ) ‘It’s raining’ b. ni-wîhtamawâ-w [CP ê-kimiwah-k] 1-tell-3(VTA INDEP) C-rain-3(VIICONJ) ‘I told him/her that it’s raining’

ROOT CLAUSE

SUBORDINATE CLAUSE

ROOT CLAUSE

SUBORDINATE CLAUSE

6.2.3. Plains Cree indexical clauses don’t support anaphoric dependencies Indexical clauses anchor to a discourse referent in the utterance situation, and so cannot support anaphora of any sort. This is confirmed by the interaction of clause-typing with A’-binding and bound variable anaphora. As established by Blain (1997), while indexical clauses do not support A’-binding (84)a, anaphoric clauses do (84)b. And, as established by Cook (2008, 2014), while indexical clauses do not support bound variable anaphora (85)a, anaphoric clauses do (85)b. (84)

(85)

a. * awîna ana ocêmê-w John-a who DEM.NA kiss-3(VTA INDEP)) John-OBV [INTENDED: ‘Who is it that kissed John’] awîna ana kâ-ocêmâ-t John-a b. who DEM.NA C-kiss-3(VTA CONJ)) John-OBV ‘Who is it that kissed John’ (Blain, 1997:68) a. * niya ni-wâpamâ-w atim [CP ni-nitonâ-w], mâka môya Jeff 1.EMPH 1-see-3(VTA INDEP) dog 1-look.for-3(VTA INDEP) but NEG Jeff [INTENDED: ‘I saw the dog I was looking for, but Jeff didn’t see the dog he was looking for] (Cook, 2008:79, (41b)) niya ni-wâpamâ-w atim [CP kâ-nitona-k], mâka môya Jeff b. 1.EMPH 1-see-3(VTA INDEP) dog C-look.for-1>3(VTA CONJ) but NEG Jeff ‘I saw the dog I was looking for, but Jeff didn’t see the dog he was looking for’ (Cook, 2008:79, (41a))

48

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

6.3. Plains Cree rules of affirmation: faultless disagreement is the norm Plains Cree indexical clauses only update the og. This confirms Portner’s (2006) conjecture that some languages will have a grammaticized clause-type that simply presents np. Pushing this a step further, in Plains Cree, presentation is the default, and origo-centric discourse-level strategies are pervasive. Interlocutors establish the author-ity, rather than veracity, of what they are presenting. This accords with Plains Cree ‘‘rules of affirmation’’ (Wolfart, 2000:148) which require participants to track information source: speakers must situate themselves relative to the np’s they present. Two strategies achieve this: (i) publicly endorsing the reliability of the information source; (ii) publicly commenting on the speaker’s relation to the information source. We consider each in turn. A speaker’s public endorsement of information source often takes the form of comments that frame the reliability of what is presented. For example, âtot- ‘give account’ in (86)a and âcimowin ‘factual account’ in (86)b commit the speaker to presenting factual information. A speaker can also affirm the reliability of np by invoking co-presence (Cook and Muehlbauer, 2008; Muehlbauer, 2008) as an eyewitness (87)a, as a participant (87)b, via landmark realia (Wolfart, 2000) (87)c, or via a relation with someone who has a connection to the presented np (87)d--e. (86)

(87)

a. . . ., kotak êkwa nik-âtotên,. . . other and FUT-give.account(IND VAI 1) ‘. . ., now I will tell another [factual] story,. . .’ (AA12-1, cited in Wolfart, 2000:147) b. . . ., êwako wiy ôma tâpwê-âcimowin,. . . RESUMP 3SG this true-factual.account ‘. . .., and this one is a true [factual] story,. . . (AA12-1, cited in Wolfart, 2000:147) a. . . .,êkota nîst ê-kî-wîcihiwêyân ôma kâ-âtotamân kotak. there 1.too PAST-be.along(VAI 1 CONJ) this give.account(CNJ VTI 1) other ‘I myself was present at the time when this second story I am about to tell took place.’ (AA12-1, cited in Wolfart, 2000:148) b. nikâh-koskwâpisininân wiy âta wiya kîkway êkotê surprised.at.sight(IND VAI 1PL) 3SG although 3SG thing there ‘we were amazed by what we saw there’ (AA2000:58, line 5) c. little red schoolhouse mân êkota kî-cimatêw; little red schoolhouse usually there PAST-stand(VAI 3 INDEP) kayâs ôma k-âtotamân, long.ago this give.account(VAI 1 CONJ) ‘a little red school-house stood there; what I am telling about happened long ago,’ (AA2000:122, lines 23--24) d. êwakw âna, kiwâpamâw êtikwê mâna, Lawrence James RESUMP that.one see(INDEP VTA 2-3) DUBIT usually L.J. ‘You have probably seen that one around, Lawrence James’ (AA2000:126, lines 32--33) e. « wiya nanâtohk ê-itahkamikisit, 3SG variously behave.thus(VAI 3 CONJ) ahpô êtikwê nik-ôh-pakamahok anima masinahikan, » ê-itêyimak. even DUBIT hit(VTA 3-1 INDEP) that book think.of(VTA 1-3 CONJ) ‘‘‘He behaves oddly, he might even hit me with the book,’’ I thought of him.’ (AA2000:124, lines 3--5)

Rules of affirmation dovetail with and supplement presentative force: (86)a and (87)b--c--d are indexical clauses with presentative force; (87)d--e have overt evidential marking (dubitative êtikwê, quotative itêyim-). The convergence of rules of affirmation with presentative force reflects a general feature of Plains Cree discourse structure: what interlocutors monitor, and what the grammar regulates, is the og. Rules of affirmation and presentative force are not redundant: the former are legislated by discourse, the latter by syntax. This does not mean that there are no cg effects in Plains Cree; rather, the og plays a more prominent role in this language. In other words, affirmation of author-ity outranks assertion of verac-ity. Rules of affirmation also show up in elicitation, where challenging a presented p requires careful and judicious monitoring of the og. Consider (88): A puts np in View. B1--B3 illustrate strategies for challenging np. (88)

A: nipêhtên ê-nihtâ-atoskwêt 1-hear(VAI SAPINDEP) C-able-work-3(VAI 3 CONJ) ‘I heard that he is a good worker’

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

49

B1 a. kiyâskowin êwkwan’ôma lie.NMLZ DEM ‘That’s a lie’ b. môy ni-tâpwêhtê-n êwkwan’ôma NEG 1-true.by.mind(VAI LOCINDEP) DEM ‘I don’t believe that’ B2 a. mwê ni-tâpwê-kisk-êyihtê-n êwkwan’ôma NEG 1-true-cognize-by.mind(VAI SAPINDEP) DEM ‘I don’t’ know that for sure,. . . mâka ni-kî-pêhtê-n mwê tonih ê-nihtâ-atoskêt 1-PAST-hear-LOC(VAIINDEP) NEG very C-able-work-3(VAI CONJ) but but I heard that he or she isn’t a good worker’ b. mâka n-îya, ni-kî-pêhtê-n 1-PRO 1-PAST-hear-LOC(VAIINDEP) but môy tonih ê-nihtâ-atoskê-t, êsa NEG very C-able-work-3(VAICONJ) REPORT ‘As for me, I hear that he or she is not a very good worker’ B3 a. môy nihtâ-atoskê-w êsa NEG able-work-3(VAIINDEP) REPORT ‘Apparently, she/he is not a good worker’ b. mwê tonih êsa mwê tonih êsa NEG very REPORT ‘Apparently not’ In B1, np is strongly challenged; this is a request to remove np from the og. B2 and B3 are instances of faultless disagreement, and a source of evidence for NOT-np is presented; in B2 the evidence is 1st-hand; in B3, the evidence is 2nd hand. In both cases, the public origo commitments of A and B contrast. Such indirect challenges provide og updates by putting NOT-np in View. Such faultless disagreement, which is the norm in Plains Cree, results in the og retaining the public origo commitments of A and B. If this model of (faultless) disagreement is correct, we expect Plains Cree speakers to refrain from positing a cg. Rather, if the origo commitments of participants intersect, this gives the appearance of a cg. This accounts for the discursive style characteristic of Plains Cree counseling speeches which create what Lightning (1992), calls mutual thinking: It is assumed that there will be an effort to think mutually with the Elder. The assumption is that active attention, humility of the hearer, and respect for the Elder, will put one in the frame of mind where the minds can meet. (Lightning, 1992:62) In our analysis, mutual thinking arises when there is extensive overlap between origo commitments, as in (89)a. In such contexts, something approximating a truth-claim arises, but it is mediated by shared experiential knowledge (i.e. authority), rather than shared discourse commitments (i.e. verac-ity). If there is no intersection, then faultless disagreement persists (89)b.31

31 This has implications for rhetorical analysis. While Aristotelian rhetoric is oriented toward persuasion, Plains Cree rhetoric is oriented toward mediation. A Plains Cree speaker does not seek to persuade the audience by inculcating a belief; rather, the goal is to facilitate transmission of knowledge. For an example of how this applies to a Cree counseling speech (kakêskihkêmowin) see Cook et al. (2006).

50

(89)

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54 PLAINS CREE ORIGO GROUND

7. Comparison with other analyses We briefly compare our proposal to alternatives, and point out areas of convergence and divergence. All formal analyses of evidentials must contend with the fact that reports are adjudicated differently than other kinds of evidence. We take this to be a central fact about evidentiality. Our analysis combines two main ideas: (i) some propositions are presented rather than asserted; (ii) presented p’s update the og. Our claim that some propositions are presented converges with Murray’s (2010) proposal that the evidential commitment (speaker has x-type evidence for p) enters the cg directly, without first being proposed. In contrast, assertions of p propose p and are entered into the cg only if p is accepted by the addressee, possibly implicitly. In contrast to our proposal, Murray’s approach is more conservative in that it does not require a separate og. For example, for Murray, all sentences present a proposition and some also add propositions --- directly or via proposal --- to the common ground. Thus, a declarative with a direct evidential ( p-DIRECT) presents p, adds p to the cg, and adds DIRECT(i,p) to the cg, where i is the speaker. In contrast, a declarative with a reportative ( p-REPORT) presents p and directly adds REPORT (i,p) to the cg; so with reports there is no ‘‘assertion’’ of p, just presentation. Especially with reportatives, there are many parallels between our treatment and Murray’s. Crucially, Murray (2010) does not appeal to additional discourse structure, but we do (in the form of origo commitments), so her proposal is more parsimonious. And all things being equal, a simpler system is preferable. However, we persist in our belief that the og is distinct from the cg, and this for two reasons. One, something like the og is independently needed for evaluative predicates (Malamud and Stephenson, 2011). Two, the fact that Plains Cree has a dedicated clause type with presentative force that only ever updates the og would remain unaccounted for. Our claim that some evidentials --- namely the ones with epistemic force --- update both the og and cg converges with Matthewson’s (2010) analysis of evidentials as epistemic modals. However, while Matthewson claims that all evidentials have epistemic force, we hold that only a subset of evidentials behave in this way. Our double update analysis, which we take from Portner (2006), replicates Kratzer’s (2012) insight that epistemic modals have an evidential base. The double update approach also captures the contrast made between ‘‘objective’’ versus ‘‘subjective’’ uses of epistemic modals (Papafragou, 2006); in the present analysis, the latter update both the og (contributing subjectivity) and the cg (contributing a truth-claim). This converges with Roberts (2015) who argues that the English epistemic modal auxiliary must ‘‘presupposes that it is anchored to one of the contextually salient discourse centers and thereby to the beliefs of a particular agent whose doxastic state is relevant’’. As far as we can determine, Roberts’ ‘‘contextually salient discourse center’’ is our origo. The distinction that we introduce between public discourse commitments (DCX) and public origo commitments (OCX) converges with many of the concerns discussed in Gunlogson (2008:114) relative to how to determine the source set (ss), defined as ‘‘all commitments of agent X in a discourse d for which the agent X is a source’’. Gunlogson’s source seems to be equivalent to our origo. However, the two approaches differ in where they locate public origo commitments. In our view, they have an existence of their own; for Gunlogson, they are subordinate to discourse commitments. This relates to the figure introduced in (5) above; Gunlogson’s analysis seems to be akin to (5)c.i, ours approximates (5)c.ii.

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

51

For obvious reasons, the logic of our analysis shares much in common with the proposal of Farkas and Bruce (2010). The major difference in the two analyses lies in the enriched set of conversational goals that we work with. For F&B, participants move a conversation forward in two ways: (i) by increasing the cg; (ii) by reaching a stable state. If our analysis is tenable, there is a third way for participants to move a conversation forward, namely by increasing the og. Our extension of the F&B algorithm also provides a novel way of modeling the difference between ‘‘agreeing-to-disagree’’ and ‘‘faultless disagreement’’ which, to our knowledge, has not been attempted before. More broadly, in treating the proposition introduced by an evidential as not-at-issue, our analysis converges with Potts (2005), who argues that what he calls supplements and expressives --- boldfaced in (90) and (91) respectively --- fail to contribute at-issue content. (90)

(Potts, 2005:90, (4.2)) Ames, a successful spy, is now behind bars. Ames was, as the press reported, a successful spy. Ames, who was a successful spy, is now behind bars. Amazingly, they refused our offer. Thoughtfully, Ed destroyed the evidence for us. Just between you and me, Aldo is a dangerous spy. EXPRESSIVES (Potts, 2005:162, (5.24)) a. The damn Republicans are aggressively cutting taxes. b. We saw that bastard Charlie at the pool hall. SUPPLEMENTS

a. b. c. d. e. f. (91)

NOMINAL APPOSITIVE AS-PARENTHETICAL SUPPLEMENTARY RELATIVE SPEAKER-ORIENTED ADVERB TOPIC-ORIENTED ADVERB UTTERANCE-MODIFYING ADV. EXPRESSIVE ADJECTIVE EPITHET

In our terms, both supplements (qua parentheticals) and expressives can be analyzed as having an origo index and introducing a not-at-issue proposition. Indeed these two properties are key features of Pott’s analysis (2005:89, (4.1)): the commitment is made by the speaker of the utterance, and it is both logically and compositionally independent of the atissue entailments. Like supplements and expressives, the prejacent p of reportative evidentials is not deniable. It remains to be seen whether the other diagnostics that we have developed to identify the og generalize to Pott’s cases. However, one crucial property of the data sets considered here --- and most dramatically with Plains Cree indexical clauses --- is that they are not-at-issue root clauses. This contrasts with the cases that Potts considers, where not-at-issue contents can supplement or evaluate* the at-issue content provided by the main clause; in Pott’s words, they ‘‘comment upon an atissue core’’ (Potts, 2005:42). However, Potts (2005:24) himself entertains the possibility that main clauses can be at notat-issue: ‘‘[P]eople sometimes use main clauses to say things that are not at issue in the sense they are unresolved in the discourse.’’ Although he does not develop an analysis of such cases, he does view them as falling under the general purview of his analysis of not-at-issue content. Regarding the relation of supplements/expressives --- which correspond to our presented nps --- Potts (2005) comments as follows (where CI = conventional implicature, a cover term for not-at-issue supplements and expressives): [I]t is quite generally appropriate to think of CIs as comments upon an asserted core. [. . .] When a CI item is used, it invariably contributes a new proposition, one that is separable from the main clause. [. . .] However, this is not a definitional property of CIs. We will see, [. . .] that it is sometimes the case that CI propositions arrive fully saturated. (Potts, 2005:43) We suggest that Plains Cree indexical clauses --- which we analyze as not-at-issue root clauses --- instantiate Potts’ ‘‘fully saturated CI propositions’’. This is by no means specific to Plains Cree. Potts (2005) speculates that English discourse adjectives in root clauses (Barker and Taranto, 2003; Taranto, 2006) contribute not-at-issue content (92)a. Their adverbial counterparts (92)b would presumably be treated as not-at-issue supplements in a Potts-style analysis. In fact, extending Potts’ (2005) analysis, McCready (2010) analyzes evaluative predicates in root clauses such as (93) as contributing not-at-issue content. He further speculates that expressive small clauses such as (94) should likewise be amenable to a not-at-issue analysis. In a similar vein, Simons (2006) argues that certain instances of propositional attitude verbs contribute not-at-issue content. In (95)b, the matrix verbs (think, believe, etc.) contribute not-at-issue content, and so can be used to frame an at-issue p that is being offered as a response to a question. In contrast, the matrix verbs in (95)c (hope, wish, dream) contribute at-issue content and so cannot felicitously frame an at-issue p. (92) (93)

a. b. a. b.

It is {clear, apparent, evident, obvious} that Briscoe is a detective. {Clearly, apparently, evidently, obviously}, Briscoe is a detective. Lucy is a genius. Bill is a jerk.

52

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

(94)

a. You idiot! b. Me argumentative? (Potts and Roeper, 2006) A: Where was Louise last night? B: a. She was with Bill. b. Henry {thinks, believes, said, suggested, hinted, imagines, supposes, heard, is convinced} that she was with Bill. c. ?Henry {hopes, wishes, dreamt} that she was with Bill. (cf. Simons, 2006:1036, (2))

(95)

8. Conclusion Regarding the empirical motivation for og updates, the most compelling evidence comes from Plains Cree, which we have argued has a clause-type whose default illocutionary force is presentative. More broadly, we have argued that otherwise puzzling properties relating to Plains Cree discursive practice fall into place once we recognize that Plains Cree interlocutors closely monitor the og, rather than the cg. If we are correct, then in a language such as English, cg updates are frequent, but in a language like Plains Cree, they are relatively infrequent. This accounts for the difficulty in making the logic of one system apparent to speakers of another system. Anecdotally, it has been our experience that audiences familiar with Plains Cree oratory and rhetoric take it to be self-evident that presentative illocutionary force is the default in that language, that speakers are guided by the desire to affirm author-ity, and that faultless disagreement is the norm. But when the same material is presented to audiences more familiar with English-style discursive practice --- where assertive illocutionary force is the default, speakers seek to establish verac-ity, and (dis)agreement is the norm --- our claims appear exotic and far-fetched. This incommensurability relates to a point made by McCready (2010), relative to the difficulty, and perhaps even the impossibility, of providing a translation for not-at-issue content: [T]ranslation of these kinds of non-truth-conditional content might be difficult or impossible, if there is no term in the target language with the same semantics. For example, it is not at all obvious how one might translate a sentence containing honorifics, or (certain) evidentials [. . .] into a language wihtout similar constructions, in a way that preserves meaning. (McCready, 2010:50) We hope that the rudimentary formal tools that we have deployed --- namely ORIGO GROUND (as distinct from the COMMON and presentative force (as distinct from assertive force) --- provide a way of investigating discursive styles in a detailed and explicit fashion. More broadly, this is no different than cross-linguistic analysis of phonological, syntactic, or semantic patterning. In this sense then, this paper can be seen as a contribution to cross-linguistic pragmatics. GROUND)

References Agha, A., 1993. Structural Form and Utterance Context in Lhasa Tibetan. Peter Lang, New York, NY, USA. Aikhenvald, A.Y., 2004. Evidentiality. Oxford University Press, Oxford, UK. Anand, P., Nevins, A., 2004. Shifty operators in changing contexts. Paper presented at the SALT 14. Asher, N., Lascarides, A., 2003. Logics of Conversation. Cambridge University Press, Cambridge. Barker, C., Taranto, G., 2003. The paradox of asserting clarity. In: Proceedings of the Western Conference on Linguistics (WECOL), vol. 24. pp. 10--21. Blain, E., 1997. Wh-questions in nêhiyawêwin (Plains Cree) (Ph.D. thesis). University of British Columbia, Vancouver, BC, Canada. Blain, E., Déchaine, R.-M., 2006. The evidential domain hypothesis. Paper presented at the Proceedings of the Workshop on Structure & Constituency in the Languages of the Americas (WSCLA 6). University of British Columbia. Blain, E., Déchaine, R.-M., 2007. Evidential types: evidence from Cree dialects. Int. J. Am. Linguist. 73, 257--291. Braithwaite, C.A., 1990. Communicative silence: a cross-cultural study of Basso’s hypothesis. In: Carbaugh, D. (Ed.), Cultural Communication and Intercultural Contact. Lawrence Erlbaum Associates Inc., Hillsdale, NJ, USA, pp. 321--327. Cook, C., 2007. Clause-typing in Plains Cree: The internal and external syntax of CP. Given at the annual meeting of the Canadian Linguistics Association; Saskatoon, Saskatchewan. Cook, C., 2008. The Syntax and Semantics of Clause-typing in Plains Cree (Ph.D. thesis). University of British Columbia, Vancouver, BC, Canada. Cook, C., 2014. The Clause-Typing System of Plains Cree: Indexicality, Anaphoricity, and Contrast. OUP. Cook, C., Muehlbauer, J., 2008. Constructing Presence in Plains Cree. In Solveiga Armoskaite and James Thompson (eds.), UBCWPL Volume 21: Proceedings of the Workshop on Structure and Constituency in Languages of the Americas X, pp. 15--28, Vancouver: UBC Working Papers in Linguistics.

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

53

Cook, C., Déchaine, R.-M., Muehlbauer, J., 2006. The form of rhetoric: an analysis of a Plains Cree kakêskihkkêmowin (counseling speech). Paper presented at the Workshop on Linguistics and Literature. University of Texas at Austin. Cornillie, B., 2009. Evidentiality and epistemic modality: on the close relationship between two different categories. Funct. Lang. 16 (1), 44--62. Déchaine, R.-M., 2007. The evidential base. Paper presented at NELS 38, 26 October 2007. University of Ottawa. Déchaine, R.-M., Wiltschko, M., 2015. Micro-variation in agreement, clause-typing and tiniteness: comparative evidence from Plains Cree and Blackfoot. PAC 42, 69--101. Egan, A., 2007. Epistemic modals, relativism, and assertion. Phil. Stud. 133, 1--22. Egan, A., Hawthorne, J., Weatherson, B., 2005. Epistemic modals in context. In: Preyer, G., Peter, G. (Eds.), Contextualism in Philosophy: Knowledge, Meaning and Truth. Oxford University Press, pp. 131--170. Faller, M., 2002. Semantics and Pragmatics of Evidentials in Cuzco Quechua (Ph.D. thesis). Stanford University. Farkas, D.F., Bruce, K.B., 2010. On reacting to assertion and polar questions. J. Semant. 27 (1), 81--118. Garrett, E.J., 2001. Evidentiality and Assertion in Tibetan (Ph.D. thesis). University of California, Los Angeles. Ginzburg, J., 1996. Dynamics and the semantics of dialogue. In: Seligman, J., Westerstahl, D. (Eds.), Language, Logic, and Computation. CSLI Lecture Notes, vol. 1. CSLI, Stanford, pp. 221--237. Gopnik, A., Choi, S., 1995. Names, relational words, and cognitive development in English and Korean speakers: nouns are not always learned before verbs. In: Tomasello, M., Merriman, W.E. (Eds.), Beyond Names for Things: Young Children’s Acquisition of Verbs. Erlbaum, Hillsdale, NJ, pp. 63--80. Gunlogson, C., 2001. True to Form: Rising and Falling Declaratives as Questions in English. Routledge, London. Gunlogson, C., 2008. A question of commitment. Belgian J. Linguist. 22 (1), 101--136. http://dx.doi.org/10.1075/bjl.22.06gun Hamblin, C.I., 1971. Mathematical models of dialogue. Theoria 37, 130--155. Karttunen, L., 1977. Syntax and semantics of questions. Linguist. Phil. 1, 3--44. Kölbel, M., 2003. Faultless disagreement. Proc. Aristotelian Soc. 104, 53--73. Kratzer, A., 2012. Modals and Conditionals: New and Revised Perspectives. Oxford University Press, Oxford, UK. Lasersohn, P., 2005. Context dependence, disagreement, and predicates of personal taste. Linguist. Phil. 28, 643--686. Leman, W., 1980. A Reference Grammar of the Cheyenne Language. Museum of Anhropology, University of Northern Colorado, Greely, Colorado. Lightning, W., 1992. Compassionate Mind: Implications of A Text Written by Elder Louis Sunchild (M.Ed. thesis). University of Alberta. MacFarlane, J., 2005. The assessment sensitivity of knowledge attributions. In: Szabo´ Gendler, T., Hawthorne, J. (Eds.), Oxford Studies in Epistemology 1. Oxford University Press, pp. 197--233. MacFarlane, J., 2011. Epistemic modals are assessment-sensitive. In: Weatherson, B., Egan, A. (Eds.), Epistemic Modality. Oxford University Press, pp. 144--178. Malamud, S., Stephenson, T., 2011. Three ways to avoid commitments: declarative force modifiers in the conversational scoreboard. Paper presented at the Proceedings of the 15th Workshop on the Semantics and Pragmatics of Dialogue. Matthewson, L., 2010. Cross-linguistic variation in modality systems: The role of mood. Semant. Pragmat. 3, 1--74. Matthewson, L., 2011. On apparently non-modal evidentials. In: Bonami, O., Cabredo Hofherr, P. (Eds.), Emprical Issues in Syntax and Semantics, vol. 8. pp. 333--357. McCready, E., 2010. Varieties of conventional implicature. Semant. Pragmat. 3, 1--57. Moss, S., 2015. On the semantics and pragmatics of epistemic vocabulary. Semant. Pragmat. 8 (5), 1--81. Muehlbauer, J., 2008. kâ-yôskâtahk ôma nêhiyawêwin: The representation of intentionality in Plains Cree (Ph.D. thesis). University of British Columbia, Vancouver, BC, Canada. Murray, S.E., 2010. Evidentiality and the Structure of Speech Acts (Ph.D. thesis). Rutgers University. Nakayama, T., 2001. Nuuchahnulth (Nootka) Morphosyntax. University of California Press, Berkeley. Nuyts, J., 2009. The ‘‘one-commitment-per-clause’’ principle and the cognitive status of qualificational categories. Linguistics 47 (1), 141--171. Papafragou, A., 2006. Epistemic modality and truth conditions. Lingua 116, 1688--1702. Portner, P., 2004. The semantics of imperatives within a theory of clause types. Paper presented at the SALT XIV. Portner, P., 2006. Comments on Faller’s paper. Paper presented at the Workshop on Philosophy and Linguistics. University of Michigan. Potts, C., 2005. The Logic of Conventional Implicatures. Oxford University Press, Oxford, UK. Potts, C., Roeper, T., 2006. The narrowing acquisition path: from expressive small clauses to declaratives. In: Progovac, L., Paesani, K., Casielles, E., Barton, E. (Eds.), The Syntax of Nonsententials. Benjamins, pp. 183--201. Roberts, C., 2015. The Character of Epistemic Modality: Evidentiality, Indexicality, and What’s at Issue (Ms. thesis). Ohio State University. Simons, M., 2006. Observations on embedding verbs, evidentiality, and presupposition. Lingua 117 (6), 1034--1056. Simons, M., Tonhauser, J., Beaver, D., Roberts, C., 2010. What projects and why? Paper presented at the Proceedings of SALT 20. Snyder, E., 2013. Binding, genericity, and predicates of personal taste. Inquiry 56 (2--3), 278--306. Speas, M., 2010. Evidentials as generalized functional heads. In: Di Sciullo, A.-M., Hill, V. (Eds.), Edges, Heads, and Projections: Interface Properties, vol. 156. pp. 127--150. Stalnaker, R., 1978. Assertion. In: Cole, P. (Ed.), Synax and semantics, vol. 9. Academic Press, New York, pp. 315--332. Stalnaker, R., 2002. Common ground. Linguist. Phil. 25, 701--721. Stephenson, T.C., 2007. Towards a Theory of Subjective Meaning (Ph.D. thesis). MIT. Tannen, D., 1984. The pragmatics of cross-cultural communication. Appl. Linguist. 5 (3), 189--195. Taranto, G., 2006. Discourse Adjectives. Routledge, New York. Tenny, C.T., 2006. Evidentiality, experiencers, and the syntax of sentence in Japanese. J. East Asian Linguist. 15 (3), 245--288. Thomas, J., 1983. Cross-cultural pragmatic failure. Appl. Linguist. 4 (2), 91--112. Tonhauser, J., 2012. Diagnosing (not-)at-issue content. In: Proceedings of Semantics of Under-represented Languages in the Americas (SULA), vol. 6. pp. 239--254. von Fintel, K., 2003. Epistemic modals and conditionals revisited. Paper presented at the UMass Linguistics Colloquium, 12 December 2003.

54

R.-M. Déchaine et al. / Lingua 186--187 (2017) 21--54

von Fintel, K., Gillies, A.S., 2011. Might made right. In: Egan, A., Weatherson, B. (Eds.), Epistemic Modality. Oxford University Press, Oxford, UK, pp. 108--130. Waldie, R., 2012. Evidentiality in Nuu-chah-nulth (Ph.D. thesis). University of British Columbia, Vancouver, BC. Westmoreland, R.R., 1995. Epistemic must as evidential. Paper presented at the Proceedings of the 10th Amsterdam Colloquium, ILLC. Westmoreland, R.R., 1998. Information and Intonation in Natural Language Modality (Ph.D. thesis). Indiana University. Willett, T., 1988. A cross-linguistic survey of the grammaticisation of evidentiality. Stud. Lang. 12, 51--97. Wolfart, H.C., 1973. Plains Cree: A Grammatical Study. Smithsonian Institution, Washington, DC, USA. Wolfart, H.C., 1996. Sketch of Cree, an Algonquian language. In: Sturtevant, W.C. (Ed.), Handbook of North American Indians, 17. Smithsonia Institution, Washington, DC, USA, pp. 390--439. Wolfart, H.C., 2000. Introduction & Notes. In âh-âyîtaw isi ê-kî-kiskêyihtahkik maskihkiy / They knew both sides of medicine: Cree tales of curing and cursing told by Alice Ahenakew, ed. by H.C. Wolfart and Freda Ahenakew, ed. and tr. by H.C. Wolfart and Freda Ahenakew. Winnipeg: University of Manitoba Press.