1. Introduction
At the forefront of syntactic theory is the question of what is the inventory of meaningful (i.e., interpretable) morphosyntactic features that provide the semantically relevant building blocks for natural language grammars.Footnote 1 One of those features is tense, which is central to the computation of temporality in many languages. Yet there are languages known to lack tense: one of these is Paraguayan Guaraní; see Tonhauser (Reference Tonhauser2006, Reference Tonhauser2011a,Reference Tonhauser, Musan and Rathertb), who argues against the presence of covert morphological tense in this language.Footnote 2 According to Tonhauser, temporal reference is determined in accordance with a global semantic rule that introduces a topic-time variable in the semantic component, in combination with the usual contributions of aspectual morphemes, temporal adverbials, and context.
Our point of departure is the Minimalist hypothesis that grammatical meaning is constructed by assembling lexical items. Importantly, grammatical meaning cannot be added in the course of the derivation, where “derivation” is construed to include LF and the semantic component (see Chomsky's Reference Chomsky1995 Inclusiveness Principle). Thus, if tense (i.e., a pronoun referring to the topic time) is absent in the syntactic derivation, it is also absent in the semantics. The question then arises as to how Paraguayan Guaraní computes time – a complex question that we do not fully address here. In this article we are concerned with one particular aspect of the grammar of temporal reference in Paraguayan Guaraní, namely the role of two evidential morphemes: the stressless particle ra'e (an indirect evidential that requires that the evidence-acquisition event be close to or at evaluation time) and raka'e (a reportative evidential of distal past events). We develop two main ideas: first, that features from the domain of person and deixis encode and partially determine evidential contrasts associated with ra'e and raka'e; and second, that the temporal contribution of these morphemes is derived on the basis of the interaction of their person and deictic features.
A brief and selective discussion of some mechanisms for resolving temporality in Paraguayan Guaraní is provided in section 2.1. In section 2.2, we provide a brief overview of evidentiality, and introduce the general aspects of our syntactic analysis; in section 2.3, we enrich the proposed structure with the person features [participant] and [author] to capture the main types of evidential distinctions. In section 3, we discuss the basic evidential properties of ra'e and its analysis in terms of the proposed morphosyntax. In section 4, we turn to the temporal–evidential connection, providing a brief background in 4.1 and introducing the role of deictic features [proximate] and [distal] in 4.2. In section 5, we discuss the temporal properties of ra'e and propose an analysis that links the temporal and evidential properties of this morpheme through the interaction of the person and deictic features. In section 6, we examine the behavior of ra'e in embedded clauses, which provides an argument in favor of the proposed syntax of evidentiality. In section 7, we summarize the properties of ra'e and its analysis. In section 8, we turn to the properties of evidential raka'e and its morphosyntactic analysis. Section 9 provides an explicit semantics for the proposed morphosyntax for ra'e and raka'e, and section 10 concludes the paper.Footnote 3
2. Preliminaries
This section sets the stage for the proposals to follow, beginning with a brief discussion of some mechanisms for resolving temporality in Paraguayan Guaraní, followed by a short overview of evidentiality, and finally a proposal to extend the person features [participant] and [author] to the domain of evidentiality.
2.1 On temporal reference
The aktionsart of morphologically bare predicates is known to play a central role in the present vs. past interpretation of clauses in tenseless languages (e.g., Chinese). As discussed by Tonhauser (Reference Tonhauser2006, Reference Tonhauser2011a), the aktionsart of bare predicates also plays an important role in the present vs. past interpretation of sentences in Paraguayan Guaraní. Consider the following context: Maria's friend Kalo goes to Maria's house looking for her. Upon opening the door and seeing Kalo, Maria's mother utters the sentence in (1). Given the bounded nature of the event description, the past interpretation imposes itself: Maria went to the market. On the other hand, if Kalo's mother were to reply with the sentence in (2), given the unbounded nature of the event description, a present interpretation is obtained: Maria is bathing. In order to obtain a present/ongoing event interpretation in the case of (1) (Maria is on her way to the market), the progressive marker hina would need to be added, while such a marker is optional in (2).Footnote 4
Such examples indicate that aktionsart plays a role in determining temporal interpretation in the absence of overt aspectual morphemes: a bounded event must be interpreted perfectively, giving rise to a past interpretation, while unbounded events can be interpreted imperfectively, giving rise to a present temporal interpretation.Footnote 5
A temporal indexical adverb like kuehe ‘yesterday’ can also provide a past interpretation, by restricting the time of the eventuality to those times that are located in the day before the speaker's now, as illustrated below.
On the other hand, future-oriented meanings are constructed with modals, regardless of aktionsart or aspectual morpheme. One of these is the suffix -ta, which is thought to have originated from the modal verb pota ‘want’, as illustrated in (4) (Ayala Reference Ayala1996, Kallfell Reference Kallfell2016). Tonhauser (Reference Tonhauser2006, Reference Tonhauser2011a) treats -ta as a prospective aspect/modal hybrid category.Footnote 6
As Tonhauser notes, indexical adverbs cannot be used to construct a future-in-the-past meaning in the presence of -ta, as illustrated in (5a,b). Consultants describe these sentences as a contradiction: kuehe locates the event in the past, while -ta gives rise to a meaning that locates the same event in the future with respect to the speech time. Similarly, in (5c), in the presence of the indexical adverb ko pyhareve ‘this morning’, the sentence can only be interpreted as locating the described event in the future part of the morning relative to the speech time.
The above examples suggest that -ta is interpreted relative to the speech time. In this -ta is similar to epistemic modals, which are typically interpreted above tense, directly with respect to the speech time (although there are known exceptions to this). Because of this property of -ta, the described event is interpreted as future-oriented with respect to the speech time, which is clearly incompatible with an indexical adverb that locates the described event to a time prior to the speaker's now, as in (5a–b).
Importantly, in the context of narration, the evaluation time of the -ta-containing clause can be past-shifted, as illustrated in (6). The example in (6a) was provided spontaneously by our consultant. In these examples, the optional particle hina, which generally functions as a progressive marker, might not have a purely aspectual function but rather may (also) function as an assertive marker (Kallfell Reference Kallfell2016).
We suggest that in cases like these, Paraguayan Guaraní makes use of a mechanism that is available in other languages in the more restricted context of the Historical Present. The Evaluation (Eval) time of a sentence (= Speech time in the unmarked case) can be shifted to the Narrative time, represented by the speaker as the speech time for the purpose of the narration, which precedes the actual Speech time. As shown by Schlenker (Reference Schlenker2004), in such cases of narrative shift, indexical temporal adverbs are still interpreted relative to the actual speech event. On the other hand, the past-shifted Evaluation time allows a tenseless predicate to be interpreted as past relative to the actual Speech time.
We propose that in Paraguayan Guaraní, the dissociation between the actual speech context and the narrative context is not used merely as a special stylistic device, but rather as a general way of resolving temporal interpretation in narrative sequences. More precisely, our working hypothesis is that in Paraguayan Guaraní, even in ordinary speech, the Eval-time is not rigidly set to the actual speech time. In the unmarked case, it is indeed set to the actual speech time, as in English, but it can also be shifted to a narrative time as in (6) above, yielding a past interpretation relative to the speech time.
In the presence of a non-indexical adverb, the Eval-time can also be shifted to a narrative past. For example, the shift occurs with the clausal adverbials in (7), encoded by the temporal conjunctions unstressed ramo (or shortened rõ) for proximate events and ramo guare (or shortened rõguare) for distal events. These discourse-framing adverbials restrict the temporal location of the described event relative to the shifted Eval-time.Footnote 7
To recapitulate, we have proposed that, while Eval-time is the speech time in the unmarked case in Paraguayan Guaraní, it can be reset to a narrative (past) time in the context of a narrative sequence of events, or in the presence of certain discourse-framing temporal adverbial clauses. As we have seen, in such cases, a past reading is available in the presence of a future-oriented modal element, such as -ta. Footnote 8 Indexical adverbs, on the other hand, are always interpreted with respect to the actual speech context, and in the absence of tense, another mechanism is required in order to obtain a past-shifted interpretation. One such mechanism is Eval-time shift to narrative time.Footnote 9 As we will see later, the notion of narrative time is also relevant in understanding the temporal interpretation of the indirect evidential ra'e.
How is Eval-time represented in the clause? We assume that Eval-time and the Attitude Holder (Att-H), identified as the Speaker in a matrix clause, are represented as a pair of pronouns introduced by a Comp node in the outer layer of the clause (there is also the world coordinate that we will ignore here); see Bianchi (Reference Bianchi, Guéron and Tasmowski2003), Landau (Reference Landau2015), Giorgi (Reference Giorgi2010), a.o. We notate the Att-H as pros and its temporal coordinate as prot (=Eval-time). We will refer to this projection as the Attitude Center (or Att-C). (We emphasize that proT is the Eval-time and not the topic time introduced by T in tensed languages.)
(8) [pros prot C [ … IP
Looking ahead, the Att-C will be mapped onto an attitude event in the semantics. In matrix clauses, the attitude event is the speech event, and in clausal complements of attitude verbs, for example, think and say, the attitude event is the thought event or the reported speech event.
The question arises as to where modals and viewpoint aspect markers are located in the Paraguayan Guaraní clausal spine.Footnote 10 Modals, and -ta in particular, are projected high in the structure, that is, above IP, where IP is an extended predicate. Whether IP is a predicate of events or of times depends on the exact position of Viewpoint Aspect in the clausal spine. The head of IP agrees with the most prominent argument, giving direct–inverse alignment).Footnote 11 As for Viewpoint Aspect, in Paraguayan Guaraní it may be located immediately above IP, or alternatively, immediately above vP. We leave this question open here and indicate the two options below. The choice will determine whether the IP is a predicate of events, as in (9a), or a predicate of times, as in (9b). We assume that indexical temporal adverbs (Index Adv) are modifiers of Viewpoint Aspect and that the discourse-framing temporal adverbials (D-Adv) are located in the outer layer of the clausal structure.
Since modals and viewpoint aspect markers will not play a role in the discussion of the evidential/temporal connection, we will not discuss them any further, and will omit their functional projection from the clausal representations.
2.2 On evidentiality
Evidentials are broadly understood as specifying the source of information for a speaker's utterance. The two most general categories found in evidential systems are direct and indirect evidence: the first involves direct perceptual experience of the described event, or general/authoritative knowledge; the second is either reportative (third party information) or inferential (logical conclusion based on some premises). Languages may make further, finer-grained distinctions within these two broad categories; for example, a common distinction within the inferential category is inference from results (an abductive inference based on the perception of an event that is judged to be causally related to the described event), versus a purely conjectural inference (see Chafe and Nichols Reference Chafe and Nichols1986, Willett Reference Willett1988, Aikhenvald Reference Aikhenvald2004, Reference Aikhenvald2018, among others, for a detailed discussion of typological variation in evidential systems).
We propose to encode evidentiality syntactically, by means of a dedicated functional projection located below the Attitude-Center in the left periphery of the clause, and interpretable person features on the head of this projection and its associated nominal argument. We call the projection Evidence-acquisition event (EA), for continuity with the literature (e.g., Chung Reference Chung2007, Lee Reference Lee2013, Smirnova Reference Smirnova2013, Koev Reference Koev2017). The EA head encodes aspects of the event of acquiring evidence for the information expressed by its sister node – here IP, setting aside other possible intervening categories encoding modality and aspect. IP here corresponds to what is referred to as the prejacent in the evidential literature.
The EA projection is, strictly speaking, a semi-functional verbal projection. It has lexical content akin to that found in basic attitude and perception verbs, like say, think, see, and hear. Speas (Reference Speas2004: 264) proposes that “we might think of evidential morphemes as syntactically reduced logophoric (propositional attitude) predicates.” EA in a direct evidential has solely perceptual content. In indirect evidentials, EA has the content of an attitude predicate, either thought (inferentials) or speech (reportative evidentials). The content of thought-based EA may be more complex, incorporating the notion of event perception or not. Inferential evidentials that are perceptually based yield an inference-from-result (i.e., perception of an event is inferred to be the result of the event described in the clause); those that are not perceptually based are based solely on belief (conjectures). Languages differ in how they morphologically express these categories (see Aikhenvald Reference Aikhenvald2004 for an extensive review). Some examples of perceptually-based evidentials are indirect evidentials in Bulgarian and Turkish (Izvorski Reference Izvorski and Lawson1997, Şener Reference Şener2011, Smirnova Reference Smirnova2013, Koev Reference Koev2017), the clitic combination chu-sina in Cuzco Quechua (Faller Reference Faller2011), and an’ in St’át'imcets (Rullmann et al. Reference Rullmann, Matthewson and Davis2008). The evidential k'a in St’át'imcets (Rullmann et al. Reference Rullmann, Matthewson and Davis2008) and the clitic chá in Cuzco Quechua (Faller Reference Faller2002, Reference Faller2011) have been claimed to be general indirect evidentials, covering both inference from results and conjectures.
In accordance with this view of the content of EA, we propose that EA introduces an (experiential) pro argument, realized in its specifier. This pro argument is referentially controlled (possibly via binding) by the closest Attitude Holder (pro s) above it. We represent this referential relation with underlines. This captures the fact that the evidence is attributed to the speaker in root clauses (or more generally, to the Attitude-holder immediately above it, as we shall see when we look at evidentials in embedded clause). Our account bears key similarities with Speas (Reference Speas2004), who also posits a speech-center projection and an evidential projection, and instantiates the role introduced by each with a pro argument: “speaker” for the speech-centre projection, and “the one who has the evidence regarding the truth of the proposition” for the evidential projection. For Speas, the identity of the two instances of pro is achieved through binding; see (10) below. We emphasize that the syntactic structure makes explicit the fact that evidential statements are based on personal experience on the part of the attitude holder in a perceptual event or an attitude event that serves as the evidential justification for the prejacent, represented by the IP in the structure in (10). In that sense, the EA can be thought of as an extension of the Attitude Center.Footnote 12
(10) [pros prot C [pro [EA […IP ]]]]
We turn next to a discussion of the person-based morphosyntactic features on pro and their relation to the semantic content of EA. This, in conjunction with the contents of the EA, characterizes the type of evidentials.
2.3 Person-based morphosyntactic features: extension to the evidential system
The person features that characterize DP event participants are determined by the relation that the referents of these DP arguments bear to the speech event: 1P and 2P arguments are participants in the speech event, while 3P arguments are not. Nevins (Reference Nevins.2007) furthermore distinguishes [+participant] in terms of the feature [+author] (the speaker) and [–author] (the recipient) of the speech event to characterize the distinction between 1P and 2P.
We propose to extend the person features [participant] and [author] to define the role that the experiencer argument has in the EA event. Within the category of indirect evidentials, person features derive the subtype of attitude. In the case of direct evidentials, the pro subject of EA is the author of the perception event; see (11a). On the other hand, indirect evidentials are characterized as attitude events, namely as a thought event in the case of inferentials (whether inferences from result or conjectures) and as a speech event in the case of reportatives. In the former case, the pro subject of EA is the author of the thought (11b.i) and in the latter, it is the non-author (i.e., recipient) in the speech event (11b.ii). However, in many languages, a single evidential morpheme covers both inferential and reportative meanings. We do not treat this as a case of ambiguity but of underspecification: we propose that the EA encodes a generic attitude (rather than the more specific thought or speech attitude) and its pro argument is marked with the more general [+participant] without further specification as [±author], as in (11b.iii).
Paraguayan Guaraní lacks dedicated direct evidentials, but has a series of indirect evidentials. One of these is ra'e, which instantiates the case in (11b.iii).
3. The basic evidential properties of ra'e
The stressless morpheme ra'e, which forms a prosodic unit with the preceding word, is an evidential marker that augments the propositional content of the utterance with information about the nature of the Speaker's evidence, specifically conveying that the evidence is either a perceptually based abductive inference on the part of the Speaker (S), or a third-party report. Ra'e is thus a type of general indirect evidential which, in its inferential use, requires reasoning from results to causes, instantiating Willett's (Reference Willett1988) category of inference from result.
Ra'e is often present when the described event is unexpected by S, giving rise to an element of (slight) surprise, which can be intensified with other morphemes.Footnote 13 Ayala (Reference Ayala1996) gives the following example: someone goes to visit a friend, and upon seeing somebody else's name on the door, he utters to himself the sentence in (12), which carries a (slight) surprise meaning that is not captured by the English translation. Ayala makes it clear that ra'e marks what is new information for the speaker and therefore its usage does not need an interlocutor: the speaker can use it to formulate it as a thought to him/herself.Footnote 14
The meaning of surprise on the part of the speaker, known as mirativity (DeLancey Reference DeLancey1997), is commonly found with indirect evidentials crosslinguistically (see Peterson Reference Peterson2015). Although the meaning of surprise is quite pervasive with ra'e, it is not always present. See Salanova and Carol (Reference Salanova, Carol, Lamont and Tetzloff2017) and Carol and Avellana (Reference Carol and Avellana2019) for discussion of this issue with respect to ra'e.
Further illustration of the use of ra'e is given below (examples are from our own questionnaire). Consider the following contexts. María (=S1) enters the kitchen and sees Kalo's hat on the table (Context 1) or sees Kalo sitting there (Context 2). S1 then says (13) to Luisa (=S2), who is in the next room. S2 can use (14) to report S1’s observation. In (14) the matrix subject is the Att-H to whom the evidential content is attributed. In effect, in many languages, evidentials shift their meaning contribution to the matrix subject in embedded attitude contexts, rather than remaining anchored to the Speaker.Footnote 15 We return to embedded ra'e in section 5.
In both contexts C1 and C2 considered above, there is an inference on the part of S1 that Kalo came, based on the observation of a related event, namely, the presence of Kalo's hat on the table (a non-trivial inference) or based on the observation that Kalo is in the kitchen (a trivial inference, but an inference nonetheless). Interestingly, there is a constraint on the type of evidence that licenses ra'e with punctual events (also independently established by Carol and Avellana Reference Carol and Avellana2019), illustrated in (15), from our questionnaire, with the punctual verb (o)joka ‘to break’. We asked our consultants to compare two scenarios: Context 1, in which S enters the kitchen and sees a broken dish (S does not see Kalo but knows that he is the only one that has been around) and Context 2, in which S enters the kitchen and sees Kalo break the dish. S then says (15) to Luisa, who is in the next room. All eight of our consultants report that (15) can be uttered felicitously only in Context 1, where S did not see Kalo break the dish.Footnote 16
In contrast, it is fine to use ra'e in the case of direct observation of an activity or a state. For example, S can utter (16a) when they look out the window and discover that it is raining. Similarly, S can utter (16b) upon seeing Kalo singing. The first has a progressive interpretation (given the presence of progressive hína) and the second one a habitual or ability interpretation.
We maintain that in such cases ra'e still marks indirect evidence. Indirect evidentials are commonly used crosslinguistically when the speaker directly observes the described event, with a mirative inference (see, for example, Peterson Reference Peterson2015). We think that in cases like (16b), ra'e marks inference from an observed result (Kalo singing) to a general state of affairs: the prejacent proposition here is not about a particular event, but about a characterizing property of Kalo (he habitually sings, he has the ability to sing). Crucially, the observed event needs to be presented as atelic. This is because characterizing sentences rely on atelic predicates, or on explicit imperfective or habitual aspect, crosslinguistically. Example (16a) is a harder case, although here too we see the semantic contribution of ra'e as marking abductive inference: the observed event is inferred to be the result of a temporally bigger episodic event of the same type (it raining now is part of a larger ongoing event of raining). This inference too is only possible with atelic events since only such events license a subset-to-superset relation, in contrast to telic events. The requirement for atelicity is the reason we only see ra'e used with directly observed events in cases like (16), with atelic sing and rain, but not (15), with telic break a dish. Because the semantic inference in (16a) and (16b) is subtle, the mirative reading is particularly salient (see Peterson Reference Peterson2015 for a pragmatic account of mirativity). For a view of ra'e close to the one presented above, see Carol and Avellana (Reference Carol and Avellana2019), and for a different view, see Salanova and Carol (Reference Salanova, Carol, Lamont and Tetzloff2017).
We conclude that in all of the cases examined above, S acquires information that p via indirect perceptual evidence. However, ra'e is also compatible with situations in which S obtains information from a third-party source, as was illustrated in (12) for embedded ra'e. Furthermore, as noticed by Velázquez-Castillo (Reference Velázquez-Castillo, Estigarribia and Pinta2017), ra'e is compatible with the reportative particle ndaje (often shortened as -je); for example, Kalo o.purahéi ndaje ra'e ‘It is said that Kalo sings, it turns out’.Footnote 17 What we retain as grammatically significant is the fact that in the case of ra'e, the Att-H can be either the inferential author of the evidence-acquiring (EA) event (S infers that p based on an observable event) or the Att-H can be the recipient in the EA event (S acquires the information via a third party rather than via inference of an observable event). As mentioned in the previous section, we propose to unify these two cases by identifying the Att-H as a participant of an attitude EA event. The featural specification of ra'e is as below (note again that the [+participant] feature defines the role of the attitude holder in the EA event, which is not to be confused with the role in the speech event or matrix attitude event):
(17) Ra'e requires that EA be specified as [+participant] (to be modified)
We note that stressless ra'e has mobility and it forms a prosodic word with the preceding lexical item. While it typically appears after the verb phrase (with wide vP focus or narrow focus on the verb), it can also appear after a focused argument. Thus, an alternative to (13) is (18), where stressless ra'e follows the focused subject, forming a prosodic word with it. Because of its mobility and prosodic properties, we refer to ra'e as a clitic-like particle.
Based on the above, we arrived at the syntactic characterization of ra'e as given in (19).
(19) The particle ra'e originates within the IP, either adjoined to the vP or the focused constituent, and it cliticizes to the EA head at LF.
The above is captured by the LF representation given in (20), where XPf represents the focused constituent. The clitic ra'e moves from its surface position within IP, where it is adjoined to the vP containing the focused constituent or to the focused constituent itself, and adjoins to the EA head at LF. The clitic ra'e requires that the EA to which it is attached carry the interpretable feature [+participant]; see (17). This must match the feature composition of the pro in [Spec, EA]. This means that the pro subject of EA (to which ra'e is adjoined) is specified with the [+participant] feature, as shown in (20).
(20) [pros prot [C [pro[+Part] [ra'e [EA[+Part] [IPXPf (ra'e)] (to be modified)
What is the interpretation of the proposed feature on EA? As mentioned earlier, [+participant] encodes the role that the Att-H bears with respect to EA: the Att-H pro can be the author of an inferential reasoning based on perceptual evidence, or the recipient of a report, either of which leads the speaker (coreferential with pro) to introduce p to the discourse. The [+participant] feature captures the indirect evidential nature of ra'e, to the extent that only in the case of an indirect evidential can pro be interpreted as the author of an inference or the recipient of a report. We elaborate further on the meaning of ra'e in section 9.
4. The evidential-temporal connection
In this section, we turn to the temporal–evidential connection, first providing a brief background and then introducing the role of deictic features [proximate] and [distal].
4.1 A brief background
Ra'e has temporal import. Velázquez-Castillo (Reference Velázquez-Castillo2009: 2) has suggested that the temporal properties of such evidential markers follow from a more basic meaning of deixis: “temporal values are natural ramifications of the proximal and distal relations that hold between the information source and the speaker”. Although she does not provide a precise analysis of the connection between evidentiality and temporal deixis, we adopt here the general spirit of the above quote.
We interpret Velázquez-Castillo's claim as having three components: evidential interpretation in Paraguayan Guaraní relies on the representation of an evidence-acquisition event (the information source); evidential morphemes in Paraguayan Guaraní encode a deictic relation between the speech event and either the EA event (ra'e) or the described event (raka'e); and finally, a temporal interpretation is inferred from this deictic relation. All three components have partial analogues in the literature on evidentiality, which we briefly discuss below. We do not intend to give a comprehensive review; rather we aim, first, to highlight the viability of the deixis approach, a version of which we pursue; and second, to illustrate a type of analysis of the link between evidentiality and temporality, which we do not in fact follow, at least not for Paraguayan Guaraní. Specifically, whereas prior studies have grammatically encoded temporal (and, on occasion, basic evidential) properties and then derived full evidential meanings as an inference, we suggest that the evidential-temporal link in Paraguayan Guaraní goes in the other direction, in line with Velázquez Castillo's general view: evidentiality is grammatically encoded and temporal interpretation is derived from the interaction of the person and deictic features of these evidentials.
A number of formal semantic accounts have relied on the representation of an evidence acquisition time or event in the lexical semantics of evidential markers (see Chung Reference Chung2007, Lee Reference Lee2013, Smirnova Reference Smirnova2013, Koev Reference Koev2017). In that literature, the (spatio-)temporal relations of the EA event to the speech event and the described event are typically used to derive the type of evidentiality (direct or indirect) without specifically encoding it in the lexical meaning of evidential morphemes. As mentioned in section 3, the EA event has also been represented in structural terms in formal syntactic accounts (Speas Reference Speas2004). Our account builds on these works; in particular, we acknowledge the grammatical relevance of an evidence-acquisition event (see the structure in (10) above).
The second component of Velázquez-Castillo (Reference Velázquez-Castillo2009)’s suggestion – the relevance of deixis – is a fairly common idea in the more functionally and typologically oriented literature (e.g., de Haan Reference De Haan, Frajzyngier, Hodges and Rood2005), building on the observation that evidential markers sometimes develop from spatial deictic expressions (Aikhenvald Reference Aikhenvald2004). A link between spatial deixis and evidentiality has been proposed in the formal semantic literature as well. Faller (Reference Faller2004) and Chung (Reference Chung2007) argue that certain evidential markers in Quechua (-sqa) and Korean (-te, -ney), respectively, are spatio-temporal deictics that encode a relation between the speaker's perceptual field at the topic time and the described event (-sqa) or the EA event (-te, -ney). The contribution of the temporal deictics as direct or indirect evidentials follows as an inference. A somewhat different spatio-temporal relationship is entertained by Koev (Reference Koev2017) for the Bulgarian evidential marker, which is argued to grammatically encode a disjoint spatio-temporal relation between the EA event and the described event. Apart from their reliance on the grammatical representation of the spatio-temporal properties of an evidence-acquisition event, the accounts discussed above share another aspect: they all deal with evidential markers that have temporal import, and they all encode the temporal meaning fully and derive (aspects of) the evidential meaning on the basis of the temporal semantics. To achieve this, however, some of the accounts need to posit special spatio-temporal tenses alongside regular tenses (Chung Reference Chung2007, Lee Reference Lee2013, Smirnova Reference Smirnova2013, Koev Reference Koev2017).Footnote 18 Since we maintain that Paraguayan Guaraní has no tense, evidential meanings cannot be derived on the basis of tense. Rather, we suggest that evidential meanings themselves contribute to temporal interpretation, in a way we make precise below.
4.2. The role of deictic features
In the person domain, beyond the use of [participant] and [author] features, numerous languages mark distinctions between third persons in terms of a [proximate] feature; for example, the proximate/obviative marking found in the Algonquian languages (DeLancey Reference DeLancey1981, Bliss Reference Bliss2005, Reference Bliss2013, Ritter and Wiltschko Reference Ritter and Wiltschko2014, Wiltschko Reference Wiltschko2014, among others). We interpret the [±proximate] distinction, when incorporated into the system of person features, as encoding a general notion of cognitive proximity/distance (i.e., a person-based perspective), allowing for potential cross-linguistic variation in its specific manifestation. Thus, we arrive at the distinctive features [±participant] and [±proximate]: 1P and 2P are [+participant], while 3P is [–participant], and either [+proximate] or [–proximate].Footnote 19
We now propose to add the feature [±distal] to the person system. This feature allows us to characterize the distinction between 1P and 2P not only with respect to their role as participants in the speech event, but also along the spatial dimension. More precisely, the speaker is defined as [+proximate, –distal] and the hearer as [+proximate, +distal] with respect to the source of the speech event. In other words, the speaker defines the point of origin and the hearer the point of reception of the speech act.Footnote 20 Thus defined, these features are particularly useful to characterize a demonstrative system that defines the spatial relation between the referent of the DP and the speech participants, as in Spanish: este ([+proximate]), ese ([–proximate], [–distal]), aquel ([–proximate, +distal]).
In Paraguayan Guaraní, the [±proximate] feature, and its related [±distal] feature, are useful in the characterization of the four-way demonstrative system described by the grammarian Ayala (Reference Ayala1996), shown in (21), where R stands for the referent of the DP. Our proposed feature system can characterize this four-way distinction, as the features in (21) show. The feature [±proximate] is interpreted relative to both participants of the speech event: a [+proximate] DP is one that is close to both speaker and hearer, and a [–proximate] DP is one that is not close to either speaker or hearer. On the other hand, the feature [±distal] is interpreted relative to the source of speech (i.e., the speaker): a [+proximate, –distal] DP is closer to the speaker than the hearer, and a [+proximate, +distal] DP is closer to the hearer than the speaker. While the spatial features do not appear to be fully active in the characterization of event participants in Paraguayan Guaraní, they play a central role in the characterization of demonstratives, including the feature [±distal], which distinguishes (b) from (c) below.Footnote 21 The spatial relations described below apply with respect to an object within the visual field of the speech participants.Footnote 22 (The demonstrative pronouns in parentheses below are nominals formed by adding the relativizer -va to the demonstrative form.)
Paraguayan Guaraní has a determiner ‘ako’, apparently falling out of use, which appears on a (overt or covert) time-denoting nominal phrase restricted by the temporal clause headed by ramoguare, which can be characterized as [–proximate, +distal]. According to Ayala (Reference Ayala1996), this temporal deictic phrase evokes a distal event in the addressee's memory, for example those in (22) and (23) from Ayala. Demonstrative ako will become particularly relevant when we discuss the temporal properties of raka'e.
Given what we have said earlier about the relations between the speech event and its participants (namely, that the participants are proximate to the speech event, with the author as [–distal] to the source of speech and the addressee as [+distal] to it), it is natural that such relations constrain the possible combinations of person and deixis. First, a [+participant] feature can combine with a [+proximate] but not with a [–proximate] feature. Second, a [–author] feature can combine with a [+distal], but not with a [–distal] feature. Finally, a [+author] feature can combine with a [–distal], but not with a [+distal] feature. We will refer to these combinatorial constraints as the Homorganic Condition (HC) on features. The combination of features in (24a, b) are homorganic, but those in (24c) are not.
While the HC is based on the spatial relations that hold in the demonstrative system, we propose to generalize it to apply to temporal relations between events. This is indeed the most parsimonious assumption. We will see that the HC on feature combinations correctly captures the evidential–temporal connection in the case of the two evidentials ra'e and raka'e in Paraguayan Guaraní.
5. The temporal properties of ra'e
As discussed by Salanova and Carol (Reference Salanova, Carol, Lamont and Tetzloff2017), Carol and Avellana (Reference Carol and Avellana2019), and Velázquez-Castillo (Reference Velázquez-Castillo2009), and corroborated by our own fieldwork, ra'e does not directly contribute to the temporality of the described event. The use of ra'e is thus compatible with a described event that occurred in a distal past (25a) or one that has not yet occurred, as in (25b).Footnote 23
What then is the temporal import of ra'e? In the examples discussed until now, with ra'e in the root clause, the evidence-acquiring event is anchored at or close to speech time. We may assume that this is the unmarked case. Yet, as noted by Salanova and Carol (Reference Salanova, Carol, Lamont and Tetzloff2017) and Carol and Avellana (Reference Carol and Avellanain press), the EA event can be dissociated from speech time. In particular, it can be shifted to a past time by temporal adverbials that we identified in section 2 as discourse-framing adverbs, and also in narrative contexts. The examples in (26) and (27), from our own questionnaire, illustrate these observations. In (26), the evaluation time for ra'e is determined by the temporal adverbial introduced by ramoguare (a-je-juhu ramoguare yma ‘when I met María long time ago’), which we assume is adjoined to the CP that contains ra'e. In (27), the EA time introduced by ra'e in the second sentence of the narrative is determined by the preceding sentence, which is temporally anchored by the temporal-framing adverbial introduced by the conjunction -vo. This conjunction indicates temporal simultaneity with the clause to which it is attached (che a-guahẽ-vo Kalo rogá-pe ‘at the time when I arrived at Kalo's house’). As seen in the discussion of -ta in section 2.1, these temporal discourse-framing adverbials, as well as narrative contexts, can shift the Eval-time from speaker's time to a narrative time that precedes speaker's time. In this case, the EA is understood as located close to or at narrative time. In the case of (26), S found out that María had married when S met her, which was a long time ago relative to the actual speech time. In the case of (27), S realized that Kalo had moved at the time he went to Kalo's house, namely three months prior to speech time. This data further confirms that in Paraguayan Guaraní, the temporal coordinate of the speech event can be dissociated from speaker's now, and shifted to the narrative past.
We turn next to the temporal interpretation of ra'e in embedded clauses, like the one in (14). As mentioned earlier, in such cases, we see a shift in the orientation of ra'e: it is not speaker-oriented, but rather matrix-subject-oriented. Furthermore, we observe that a shift in the referential value of the Att-H brings about a concomitant shift in the temporal value of its temporal coordinate (or Eval-time). This is illustrated by the example below from our questionnaire. In this example, Kalo (the matrix subject) acquired evidence that María had married at a time close to the time that Kalo relayed the information to the speaker. Such data suggests to us that the complement of a verb of saying is a CP with a structure like (20), that is, with a fully specified left periphery, including an Att-C and an EA projection.
The generalization that emerges regarding the temporal interpretation of ra'e is as follows:
(29)
a. Ra'e in root CP clauses: the subject of EA is bound to the speaker, and the temporal interpretation of EA associated with ra'e is close to or at speaker's time (the unmarked case) or close to or at narrative time (in shifted cases).
b. Ra'e in embedded complement CP clauses: the subject of EA is bound to the external argument of the matrix predicate, and the temporal location of the EA event is at or close to the matrix event time.
It is clear from the above data that ra'e has a temporal effect on the EA event. It is also clear that whatever the Eval-time (speaker's now or narrative time for matrix ra'e, or matrix event time for embedded ra'e), the relevant temporal notion is at or close to Eval time; in other words, the EA event is temporally proximate to the Eval Time. We further note that the temporal proximity can only be in the past or at the Eval Time, not in the future. This restriction has a pragmatic source in that the EA event, whether reportative or inferential, cannot be located after the speech time.
Since we want to maintain the strong thesis that Paraguayan Guaraní does not have tense, we propose to model this temporal proximity on the basis of deixis, specifically through the feature [+proximate] encoded in the meaning of ra'e. We already know from the study of person-based alignment that [±proximate] can play a non-spatial role. Similarly, we propose here that these deictic features can be interpreted temporally. These grammatical features are always anchored to the speech event (more generally, the local attitude event), but they may be deployed along different dimensions: cognitive or physical space, and time.
(30) If an EA event is marked as [+proximate], then it is temporally proximate to the Attitude event.
We propose to update our analysis of ra'e accordingly. To the feature [+participant] of ra'e we add the feature [+proximate]. Note that this feature is congruent with the person feature [+participant], as required by the Homorganic Condition in (24). We update the morphosyntactic analysis of EA in combination with ra'e in (31a) (compare with (17)), and its associated syntactic structure in (31b) (compare with (20)). Note that the pro subject of EA is not itself marked [+proximate] because in Paraguayan Guaraní personal pronominal arguments, unlike demonstratives, do not encode deictic features.
(31)
a. Ra'e requires that EA be specified as [+participant, +proximate]
b. [pros prot [C [pro[+part] [ra'e [EA[+part, +prox] [IPXPf (ra'e)]…]
In the next section, we provide data based on preliminary research on the nature of the embedding verb in cases where ra'e appears in the complement clause, and show how the facts strongly support a syntactic representation of the left periphery of CP complement clauses along the lines of (31).Footnote 24
6. The nature of the embedding verb and its theoretical significance
As we have seen above, ra'e can appear within the clausal complement of verbs of saying, such as -e’ ‘say’, -mombe'u ‘tell’, -porandu ‘ask’.Footnote 25 Although ra'e is part of the embedded clause, its prejacent is not a textual transmission of what was said. This is illustrated in (32). After María went to Kalo's house, she said (literally) (32a) to S. The content of what María said can then be reported by S the next day as in (32b), with the indexical kuehe pyhareve ‘yesterday morning’ and possibly with the epithet pe tavyrai ‘that idiot’ in place of the name Kalo. Both of these are speaker-oriented: the embedded temporal adverb is interpreted with respect to the speaker's now, and the epithet expresses the speaker's thoughts about Kalo.
The propositional attitude verb (oi)mo’ã ‘to believe’ allows a complement either with or without a subordinator ha, and there is an interesting contrast between the two. With this verb, it is not possible to embed ra'e in the presence of the subordinator ha, as shown by the ungrammaticality of the examples in (33a).Footnote 26 On the other hand, in the absence of ha, the presence of embedded ra'e is felicitous, as shown in (34). Crucially, in these examples, the evidential content is not attributed to the matrix subject but to the speaker, as in root clauses.Footnote 27 In that case, ra'e has scope over the entire proposition (S infers that María believes that Kalo moved).
It appears that the ungrammaticality of the sentences in (33) is semantic: the degree of certainty regarding the prejacent on the part of the attitude holder expressed by (oi)mo'a is too weak to sustain an inference on the part of the matrix subject. This view finds some support in the contrast between verb (oi)mo'a and the verb (o).rovia; see (35) below. The former is a thought close to a suspicion, while in the latter, there is a higher degree of certainty on the part of the Att-H.Footnote 28 In this case, ra'e is compatible with ha and is matrix-subject oriented.
The emerging generalization is given in (36):
(36) There is a correlation between the presence/absence of subordinator ha and the orientation of ra'e in the embedded clause.
a. Embedded ra'e is speaker-oriented in the absence of subordinator ha
b. Embedded ra'e is matrix-subject oriented in the presence of subordinator ha.
The above generalization suggests that in the presence of subordinator ha, there is a CP layer in the embedded complement, with an Att-C and EA projection represented in its left periphery (i.e., the subordinator ha is selected by C). In such cases, embedded evidential ra'e is matrix-subject oriented and a semantic incongruence appears to arise, namely between the evidentiality and the epistemic state of the Att-H (i.e., the matrix subject) with respect to the embedded proposition. Perhaps the attitude verb is too weak (in terms of degree of certainty) to support ra'e; speakers’ intuitions appear to confirm this conjecture. On the other hand, in the absence of an overt subordinator ha, the embedded C layer is analyzed as absent.Footnote 29 In the absence of an embedded C layer where Att-C and EA are located, embedded ra'e will cliticize onto the closest (i.e., the matrix) EA, giving rise to a speaker-oriented evidential, exactly as in root clauses. The above correlation thus provides strong support for an articulated left periphery in the embedded CP, along the lines proposed in (20).Footnote 30
Note that ra'e can appear in questions and in conditional clauses, with interesting effects on meaning,Footnote 31 but a full discussion and analy`sis of these facts is outside the scope of this article.
7. Summary of proposed analysis of ra'e
Our analysis of ra'e is based on a syntactic representation of evidentiality in the C layer of the clause, along the lines of Speas (Reference Speas2004). Specifically, there is an evidence-acquisition (EA) event, located immediately below the Attitude Center (Att-C). Att-C consists of a pair of pronouns, an Attitude holder pro s and its temporal coordinate pro T, which are interpreted as coordinates of the speech event.
The particle ra'e attaches in the overt syntax to the focused constituent (i.e., the new information in the prejacent IP) and is related to the closest EA projection above it via a covert LF adjunction rule. Ra'e requires that the EA to which it is adjoined be specified as [+participant, +proximate]; a set of features congruent with the Homorganic Condition stated in (24). This feature specification gives rise to an interpretation where (i) the subject of EA is interpreted as [+participant] of the EA event (author of inference or recipient of speech) (ii) the EA event is temporally proximate to the Attitude-event immediately above it.
Importantly, the proposed analysis allows us to model the link between the interpretation of the subject of EA and the temporal interpretation of the EA, thus accounting for the contrast between root and embedded ra'e. This analysis crucially relies on the syntactic representation of the Att-C and related EA event at the left edge of full CP clauses. In the absence of a CP layer in a clausal complement, embedded ra'e is correctly predicted to behave just like root ra'e, having a speaker-oriented rather than a matrix-subject-oriented interpretation.
8. The case of raka'e
In this section, we turn to the second evidential marker, raka'e, first describing its grammatical properties and then proposing a morphosyntactic analysis.
8.1 The grammatical properties of raka'e
The evidential particle raka'e differs from ra'e in that it constrains the temporal interpretation of the described event: it is located in a remote past. On the other hand, unlike ra'e, raka'e does not constrain the temporality of the evidence-acquisition event (EA). Furthermore, the source of evidence is based on third-party knowledge; in other words, raka'e is an unambigously reportative evidential. We illustrate these properties below with examples from our questionnaire.
The sentence in (38) is appropriate if the described event (my grandfather's death) is located in a distal past, but not if it is a recent event. Furthermore, the source of knowledge must be reportative.Footnote 32 Our consultants accepted (38) as a continuation of C2 in (37), where María heard from her mother, in a distal past, that her grandfather had been killed, but not as a continuation of C1, where María was a witness in a distal past to her grandfather's killing.Footnote 33 (A context where María has recently heard about her grandfather's distal death would also be appropriate.)
The particle raka'e can appear in an embedded sentence. Thus, Luisa can report the event in (38) by uttering (39), the Attitude-holder now identified with the matrix subject, to whom the reportative-based evidence is attributed.
We note furthermore that embedded raka'e, like embedded ra'e, is unacceptable with the embedding verb (oi)mõ’a if the subordinator ha is present, but acceptable if ha is absent, in which case raka'e is speaker-oriented, not matrix-subject oriented). See the contrast in (40), compared with ra'e in (33) and (34). As in the case of embedded ra'e, the data with embedded raka'e support a syntactic representation of the Att-C and EA projection in the left periphery of the CP complement of attitudinal embedding verbs, where the presence of a C-domain is flagged by the presence of the subordinator ha in Paraguayan Guaraní. The account provided for the paradigm with ra'e in (33) and (34) applies directly to the paradigm in (40).
Raka'e, like ra'e, has a certain mobility within the IP; it is stressless and forms a prosodic unit with the preceding stressed item. Thus, raka'e can also appear right-adjacent to the subject, identifying it as the focused constituent of the sentence.
Like ra'e (see footnote 31), raka'e can appear in questions. The excerpt in (42) was found in the newspaper ABC Color, July 26, 2016. The mother's (implicit) answer to (42a) justifies the followup questions in (42b), as well as in (42c). The use of raka'e in (42b) and (42c) is licensed because the speaker has reportative information for the presupposed content of the questions (the source being the mother). This naturally produced discourse sample is particularly interesting because the speaker asking the question was a participant in an event that took place a long time ago, but has no memory of it (42b), or only a distant memory of it (42c).
This example shows that not only does raka'e contribute to the distal temporal interpretation of the described event, but it can, at least in certain contexts, also serve the function of evoking a distal event in memory, very much like what Ayala Reference Ayala1996 describes for the demonstrative ako, as in examples (22)–(23).Footnote 34 It is thus very tempting to analyze raka'e as arising, perhaps historically, from the morphological composition of the distal demonstrative ak(o) and an evidential morpheme ra'e. With this in mind, we turn to the analysis of raka'e.
8.2 The analysis of raka'e
We propose that the following two semantic properties of raka'e are intimately related:
(43)
a. The described event is temporally distal with respect to Eval-Time.
b. The Att-H's source of evidence is reportative.
Recall that in the case of ra'e, the deictic feature [+proximate] and the person feature [+participant] are realized on one and the same functional category, namely EA, and EA establishes a relation between itself and the Att-C above it. In the case of raka'e, we propose that the deictic and the event role features are realized on two distinct functional projections: the deictic features a deixis projection (DX) that establishes a temporal relation between the described event and the attitude event, and the event-role features on an evidential projection (EA), with a syntactically realized experiencer of EA whose event role is constrained by the person feature on EA. The two relations are constrained by the Homorganic Condition on features in (24): if DX is specified as [+distal], then EA must be specified as [–author].
(44) Raka'e requires that DX be specified as [+distal] and the EA event as [–author], a combination congruent with the Homorganic Condition (24).
We propose that the DX projection combines with the vP, which semantically contributes an event predicate. In other words, DX is located below Viewpoint Aspect, whatever its precise location in the Paraguayan Guaraní clausal spine is. This allows for the semantic content of DX to act as a modifier of the predicate of events denoted by the vP. This is not unlike the demonstrative ako discussed in section 2.2. Demonstrative ako is specified as [+distal] and modifies a predicate of events, giving an interpretation of a distal event in the addressee's memory. As mentioned earlier, we speculate that the -ak piece in raka'e is possibly historically related to the demonstrative ako.Footnote 35
Like ra'e, raka'e is a stressless particle that right-adjoins to vP or to the focused constituent within IP in the overt syntax. At LF, raka'e adjoins first to the DX above it and then to the EA above DX, licensing and restricting the interpretation of both elements. The syntactic derivation is summarized in (45).
(45)
a. Raka'e, like ra'e, is contained within the IP in the overt syntax (adjoined to vP or to the focused constituent).
b. vP merges with DX, DX merges with IP, IP merges with EA, and EA merges with Att-C.
c. At LF, raka'e adjoins first to DX, forcing DX to be specified with the interpretable [+distal] feature. It then adjoins to EA, forcing EA (and therefore its pro subject) to be specified with the interpretable feature [–author].
d. the pro subject of EA is bound by pros in the Att-C above it.
This gives the structure in (46). The pro subject of EA is referentially controlled by the Att-H pro above it. (As usual, we notate the referential dependency with underlines.)
(46) [pros prot [pro[–author] [raka'e EA [I [raka'e DX[+distal] [vP …XPf… raka'e ]…]
This structure constrains the interpretation of sentences with raka'e as stated in (47).
(47)
a. Given the [–author] feature on the subject of EA, the Att-H pros (the speaker, in case of root raka'e and the matrix subject, in the case of embedded raka'e) is interpreted as the recipient of the evidence. The reportative evidential nature of raka'e is therefore captured.
b. Given the [+distal] feature of DX, the described event is interpreted as temporally distal to the Attitude event above it (i.e., distal with respect to the speech time in the case of root clauses, and distal to the matrix event time in the case of embedded clauses). The remote past interpretation of the described event with respect to the Attitude event above it follows as an inference from this relation (as in the case of ra'e).
To recapitulate, in the proposed analysis, the evidential–temporal connection is captured by two homorganic features [+distal, –author], which, in the case of raka'e (unlike the case of ra'e), is spread across two distinct functional projections: the [+distal] feature on a DX projection that combines with the vP (the described event) and the [–author] feature on EA, giving rise to the interpretations in (47). Since the deictic feature in the case of raka'e is low in the clausal structure, it restricts the temporal relation between the attitude event and the described event. Thus, raka'e’s temporal import is distinct from that of ra'e.
9. A sketch of a formal semantics for ra'e and raka'e
We have discussed two properties of the evidential morphemes ra'e and raka'e: the type of indirect evidence they encode and their effect on the temporal interpretation of sentences. We now consider how the syntax we have proposed encodes these meanings. Recall that according to our earlier proposal, the evidential and temporal meaning is distributed among the functional head EA in the CP domain of the clause, the DX projection (when present), and the features on the particles themselves.
We propose that the core semantics of EA in indirect evidentials – at least its not-at-issue contribution – is as shown in (48). EA is a null, semi-functional lexical item, lexically encoding the meaning that an attitude holder (x, syntactically represented by the pro in the specifier of EA) has evidence for the prejacent proposition (p, syntactically represented by the IP complement of EA). This is a general meaning covering both reportative and inferential evidentials, and within the latter type, both inference from result and conjectures. As we suggested in (11), in purely reportative or purely inferential evidentials, the attitude event (eʹ in (48)) could be further restricted to a speech event or a thought event, respectively.Footnote 36 To distinguish the two types of inferential evidentials, the attitude event (eʹ in (48)) could be further restricted to being perception-based or purely belief-based, something not further discussed here. We similarly do not discuss the semantics of direct evidentials, beyond noting that, in accordance with (11), instead of an attitude event (eʹ), EA introduces a perception event.
(48) EA(p)(x): ∃e ∃eʹ [x is a participant in eʹ & eʹ is an attitude event & eʹ causes e & x-has-evidence-for-p(e)]
We postulate that an EA head, that is, a lexical item with the lexical semantics in (48), is present in the syntax of indirect evidentials. This departs from previous semantic accounts, which introduce an EA event only in the lexical semantics of individual evidential morphemes (Chung Reference Chung2007, Lee Reference Lee2013, Smirnova Reference Smirnova2013, Koev Reference Koev2017; see Speas Reference Speas and Aikhenvald2018 for a recent review.) We see two advantages in encoding the EA syntactically and separately from the evidential morphemes themselves. The first advantage is conceptual: this move allows us to separate the common meaning of indirect evidentials – in our particular case, the common meaning of ra'e and raka'e – from the more specific meanings that these morphemes contribute, simplifying their lexical semantics and isolating the more general grammatical contribution in the lexical semantics of EA. For this reason, we consider EA to be a lexical item itself, realized in the syntax independently of the evidential morphemes. The second advantage of syntactically representing EA is empirical, having to do with the correlation between the orientation of embedded ra'e and the presence or absence of the subordinator ha; this was discussed in detail in section 7.
Finally, note that the meaning of EA in (48) is not-at-issue. The claim that evidence type is not-at-issue conforms to virtually every account of the semantics of evidentials starting with Izvorski (Reference Izvorski and Lawson1997), whether the account is modal (Matthewson et al. Reference Matthewson, Davis and Rullmann2007, Rullmann et al. Reference Rullmann, Matthewson and Davis2008, Faller Reference Faller2011, Smirnova Reference Smirnova2013) or not (Murray Reference Murray and Lima2010, Koev Reference Koev2017).Footnote 37
We next develop an explicit syntax–semantics mapping for sentences with ra'e and raka'e, at least as far as the properties under consideration in this paper are concerned. We discuss in turn for each morpheme how the type of evidence is determined, and how the temporal contribution is derived. Finally, we turn briefly to the question of what the at-issue meaning contribution of evidentials is.
9.1 Ra'e: type of evidence and temporal interpretation
Recall that ra'e is a general indirect evidential, of a type commonly found crosslinguistically, that covers both reportative and inferential evidence. We proposed to model this type of indirect evidential through the person feature [+participant]. Ra'e adjoins to EA and provides it with a [+participant] feature. The content of the EA functional head as in (48), and the interpretable [+participant] feature it now bears, restrict the type of evidence that sentences with ra'e admit. The feature [+participant] requires that the evidence be obtained on the basis of a speech or a thought event in which the pro subject of EA is a participant (see (11b.iii)). This meaning component contributed by ra'e is added to the core not-at-issue meaning of EA, as shown in (49) with the contribution of ra'e in boldface. Note that person features on nominals are commonly treated as contributing not-at-issue meanings (e.g., Schlenker Reference Schlenker1999, Reference Schlenker2003, Heim Reference Heim, Harbour, Adger and Bejar2008, Sudo Reference Sudo2012). Here we extend this semantic role of person features – the role of contributing not-at-issue meanings – to the interpretable [+participant] person feature on EA.
(49) [ra'e EA [+part]](p)(x): ∃e ∃eʹ [x is a participant in eʹ & eʹ is an attitude event & eʹ causes e & x-has-evidence-for-p(e)] (to be modified)
Informally: as the result of an attitude event (speech or thought) in which x is a participant, x has evidence for p; or, in short: x acquires evidence for p.
Whether eʹ is a speech or a thought event, and whether [+participant] is interpreted as a recipient or an author, determines whether the type of evidence the pro subject has is reportative or inferential. If eʹ is a speech event in which the speaker is a recipient (e.g., eʹ is the event of María saying to the speaker that Kalo is at home), then the evidence is reportative. If eʹ is a thought event in which the speaker is an author, then the evidence is inferential. We emphasize that the inferential and reportative readings associated with ra'e are directly linked to the person feature [+participant]. This line of analysis opens up the possibility that indirect evidentials that are lexically restricted to only inferential or only reportative meanings are restricted in this way because the interpretable person feature on the EA projection is further specified as [+author] or as [–author], respectively. In effect, this is what we proposed in (11).
Another question arises at this point. Why is it that when eʹ is a speech event, the pro subject is interpreted as a recipient rather than as an author? The answer derives from what unifies the reportative and inferential readings so that they are commonly expressed crosslinguistically with the same element (e.g., ra'e in Paraguayan Guaraní, apparently in English). As (49) states, eʹ causes e – the event of speaker's having evidence for p. As a participant in eʹ, the speaker must therefore not already have evidence for p. This requires that with a speech event, the pro subject must be a recipient rather than an author.
Another point about (49) requires clarification. The descriptive content of eʹ (the speech or thought event that forms the basis for the evidential attitude) and the proposition p need to be related, but we leave the precise relation unspecified in the lexical entries here. If eʹ is a speech event in which the speaker hears that q, or eʹ, is a thought event in which the speaker infers that q, then either q = p or q ⇒ p. For instance, if María tells the speaker that Kalo is in Paris, the speaker can acceptably say both ra'e (Kalo is in Paris) and ra'e (Kalo is in Europe). Similarly, if the speaker infers that Kalo has broken the vase in the morning, both ra'e (Kalo broke the vase in the morning) and ra'e (Kalo broke the vase) are appropriate uses of the indirect evidential.
Finally, note that besides the [+participant] feature, ra'e contributes another aspect of meaning: the inference must be from result. What this suggests is that the content of EA is even more elaborate, making reference to a perceptual event in which the attitude holder perceives another event (e.g., the speaker sees Kalo's hat on the table), and concludes that the perceived event (Kalo's hat being on the table) is a plausible result of the event described in the prejacent proposition p (e.g., Kalo's being home), see (13). Thus, ra'e’s restriction to perceptually obtained evidence is not universal.Footnote 38
Turning to the temporal contribution of ra'e, recall that Paraguayan Guaraní has no tense morphemes and that we also suggested that it has no tense in the semantic component either. Nevertheless, ra'e contributes to the temporal reference of sentences in which it appears. Specifically, it requires that the acquisition of evidence be recent to, or at, the evaluation time. We propose that this temporal interpretation is an inference from the person feature on the EA head. In particular, the interpretable feature [+participant] on EA is grammatically associated with a [+proximate] interpretation of the relation between the attitude event eʹ introduced in EA and the speech event. The feature [+proximate] contributes the additional meaning given in boldface in (50).
(50) [ra'e EA [+part, +proximate]] (p)(x): ∃e ∃eʹ [x is a participant in eʹ & eʹ is an attitude event
& eʹ causes e & x-has-evidence-for-p(e) & eʹ is proximate to e0],
where e0 is bound by the local evaluation event (the speech event in main clauses or the attitude event in complements to attitude predicate).
We emphasize that the two features [+participant] and [+proximate] are interpreted with respect to different aspects of the structure. The [+participant] feature specifies the role pro has with respect to EA, and thus it determines that the evidence is inferential or reportative, while the [+proximate] feature specifies the relation between the EA attitude event and the speech event (or more generally the matrix attitude event), thus determining, indirectly, and without any tense features, the temporal location of the EA event.
As we noted earlier, there have been accounts of evidentials in other languages that have attributed tense semantics to evidential morphemes. Unlike the proposal here, which derives the temporal meaning as an inference from the evidential meaning, these semantic accounts suggest that evidentials are special tense markers that exist alongside regular tenses and encode in their lexical semantics reference to the (space-)time of an EA event. In fact, if EA is a semi-lexical head, there is no principled reason why in a language with tense, EA could not combine with tense. A plausible example of such a system is Matses (a Panoan language spoken in Amazonian Peru and Brazil), as described by Fleck (Reference Fleck2007) (but see Arregui et al. Reference Arregui, Rivero, Salanova, Arregui, Rivero and Salanova2017 for an alternative view).
9.2 Raka'e: type of evidence and temporal interpretation
We propose that the evidential marker raka'e incorporates the semantic contribution of ra'e and the distal demonstrative ako seen elsewhere in the grammar of Paraguayan Guaraní (see section 4.2, examples (22) and (23)). The use of raka'e is associated with a deictic projection (DX) with a [+distal] feature. The feature itself contributes the presupposition that the described event is spatially distal from the local attitude event.Footnote 39
(51) [[DX [+distal] ]] (P)(e) is defined iff e is distal to e 0
When defined, [[ DX [+distal] ]] (P)(e) =1 iff P(e) = 1
The DX [+distal] aspect of present in sentences with raka'e is in some ways similar to Faller's (Reference Faller2004) analysis of the -sqa evidential in Cuzco Quechua. On that account, -sqa is a spatio-temporal marker that encodes a temporal and spatial deictic relation between the described event and the perceptual space of the attitude holder. This differs from our account in that the temporal contribution is part of the lexical semantics of -sqa (i.e., it is a special past tense marker), and in the case of Paraguayan Guaraní, DX establishes a temporal relation only between events, rather than between times.
The rest of the semantic composition proceeds as expected. The only difference between the contribution of the [raka'e EA] complex and the [ra'e EA] complex is in the person features, with consequences for the evidential content. [raka'e EA] is marked [–author], due to the homorganic restriction in (24), and the evidence type can only be reportative (since inferential evidence requires the pro subject of EA to be [+author]). There is no temporal inference concerning the EA attitude event, since (24) prevents the [+proximate] feature from appearing on EA. The boldface part of (52) is raka'e’s contribution to the interpretation of EA, a restriction of the not-at-issue meaning.
(52) [raka'e EA [+part][ –author]] (p)(x):
∃e ∃eʹ [x is a recipient in eʹ & eʹ is a speech event & eʹ causes e & x-has-evidence-for-p(e)]
9.3. Ra'e and raka'e: Further issues in evidential interpretation
So far we have discussed the not-at-issue contribution of evidential ra'e and raka'e. As for the assertive meaning, we surmise – though we have not examined this issue extensively – that a modal account for these evidentials might be appropriate, following one of the general approaches to evidentiality, as in Izvorski (Reference Izvorski and Lawson1997), Matthewson et al. (Reference Matthewson, Davis and Rullmann2007), Faller (Reference Faller2011), Smirnova (Reference Smirnova2013), Arregui et al. (Reference Arregui, Rivero, Salanova, Arregui, Rivero and Salanova2017), among others. On such an approach, evidential ra'e and raka'e would, like typical epistemic modals, contribute meaning concerning the likelihood of the prejacent proposition p in view of what the attitude holder knows and considers evidence for p. As a particular subtype of an epistemic modal, evidential ra'e and raka'e would restrict the relevant body of knowledge to information obtained through a perceptual experience on the part of the attitude holder (with further differences between the two particles, as discussed above).
The Paraguayan Guaraní data are consistent with an epistemic modal account: ra'e can be embedded under attitude predicates and it shifts its orientation to that of the matrix subject, it scopes over clause-mate negation, and its semantic contribution cannot be challenged by the addressee. These facts do not necessarily rule out a non-modal alternative (except an illocutionary operator as in Faller Reference Faller2002, because of its problems with embeddability), perhaps as a not-at-issue content similar to appositives, as in the approach to evidentiality in Murray (Reference Murray and Lima2010), Koev (Reference Koev2017), among others. Such an alternative, however, would have particular difficulty with the obligatory shift of ra'e in attitude contexts, since appositives tend to either remain speaker-oriented in such cases, or at least only allow optional shift. But most importantly, such alternatives require that the speaker be committed to the truth of the prejacent, which does not appear to be the intuition that native speakers have about the reportative use of ra'e and reportative raka'e. In particular, these evidentials can combine with the reportative ndaje, in which case the speaker is clearly not committed to the truth of the prejacent proposition p (on ndaje, see footnote 17). On a modal account, there is no guarantee that the world of evaluation is among the worlds quantified over by the evidential, and correspondingly the prejacent proposition p is not asserted to be true. This captures well the variability in speaker commitment to p, relative to the strength of the evidence. On balance, the modal account of ra'e seems to fare better.
We do not formalize the assertive modal component of ra'e and raka'e, since this aspect of meaning is not where our main concern and contribution lie. The focus-sensitivity of these evidentials should also be captured in a full account of their meaning, but this is beyond the scope of the present article.
10. Conclusion
We have provided an account of the evidential–temporal connection in Paraguayan Guaraní, a language without tense. The model appeals to independently established morphosyntactic person features: the event-role features [± participant] and [± author], and the general deictic features [± proximate] and [± distal]. We have argued that the evidentials ra'e and raka'e can carry both features: an event-role feature and a deictic feature interpreted along the temporal dimension, and that the two features are constrained by the Homorganic Condition on feature combinations. In the case of ra'e, both the event-role feature [+participant] and the temporal deictic feature [+proximate] are realized on the same functional category, namely EA. These features ultimately constrain the evidential type encoded by the EA morpheme in association with ra'e and the temporal relation between the EA event and either the speech event (in the case of matrix ra'e) or the matrix attitude event (in the case of embedded ra'e). On the other hand, in the case of raka'e, the features are scattered. The [+distal] feature is realized on a demonstrative functional head (DX) that combines with vP (akin to the demonstrative ako), while the event role feature is realized on EA, which is predicted to be [–author] given the Homorganic Condition. It follows that raka'e constrains the temporal relation between the described event and either the speech event (in the case of matrix raka'e) or the matrix attitude event (in the case of subordinate raka'e). The proposed analyses correctly capture the fact that in case of embedded ra'e and raka'e, the evidential orientation shifts to the matrix subject concomitantly with the temporal orientation.
We recognize EA as an abstract morphosyntactic morpheme that projects below the Attitude-Center (Att-C) at the left edge of the clausal periphery. We have argued that, in addition to the conceptual advantage of postulating such a morpheme, there is a syntactic argument for it in Paraguayan Guaraní, namely that its presence/absence in subordinate clauses is governed by the presence or absence of the subordinator ha, which in turn licenses the Comp domain where EA and its associated Att-C are located.
Finally, we note that the analysis of EA as a semi-lexical projection does not preclude that in a language with tense, EA can combine with T. The Matses language, described by Fleck (Reference Fleck2007), might very well be a case in point. As expected, and unlike Paraguayan Guaraní, in Matses the proximate/distal temporal relations between events are not restricted by the evidential type.
