Abstract
Linguistic studies of evidentiality, the coding of source of knowledge, have often appeared divided into two camps: those whose focus is the semantic, morphological and typological characteristics of grammaticalized morphological evidential systems (e.g. Aikhenvald, 2004), and those whose focus is on the social functions of non-grammaticalized evidential constructions as markers of epistemic authority and responsibility (e.g. Fox, 2001; Sidnell, 2012). While interest in the discourse functions of all evidential systems has been growing as seen in the recent special issue of the journal Pragmatics and Society on ‘Evidentiality in Interaction’, there has been little direct attention on whether the deployment of evidential strategies in discourse varies according to the grammatical status of the grammatical resources available to the speaker. This article examines the nature of both grammaticalized and non-grammaticalized evidential systems in a number of languages to show that while the underlying pragmatics of evidentiality is the same regardless of grammatical system, nonetheless grammaticalized evidential systems provide important evidence of the particular features of knowledge sources that are used in routine ways in discourse sufficiently to motivate their development into grammatical systems.
Introduction
Recent work in conversation analysis has argued that the coordination of knowledge and the management of asymmetries of knowledge are fundamental drivers of conversation (e.g. the ‘epistemic engine’; Heritage, 2012a, 2012b). In order to do this people require linguistic resources to account for their knowledge (Sidnell, 2012). One set of such resources comprises those that draw attention to how knowledge was acquired – how one comes to know what one knows. The term ‘evidentiality’ is generally used by linguists to talk about this linguistic coding of source of knowledge (e.g. Chafe and Nichols, 1986; Mushin, 2001). Evidentiality is thus a way of making the status of one’s knowledge ‘visible’ in discourse as it draws attention to the fact that what is being talked about is knowledge – a body of information filtered, synthesized, analysed by someone’s mind – and not omnipresent information. 1 In this domain linguists have largely focused on a subset of evidential resources that have been grammaticalized as bound morphemes, but the significance of grammaticalized evidentiality for understanding ways of managing knowledge is still poorly understood. Patterns of grammaticalization of evidentiality are windows into the management of knowledge in discourse because the very fact that meanings become so systematized formally points to routine and frequent ways of using language (Bybee and Hopper, 2001). So understanding the nature of grammaticalized systems and their deployment in discourse, and how they may compare with non-grammaticalized systems, helps us develop a deeper understanding of the aspects of knowledge that appear salient and routinely expressed in human social interaction.
In this article I evaluate ways that typological and discourse linguistics have taken the study of evidentiality over the past 20 years or so, especially in terms of the nature of grammaticalized evidential systems and the themes that have emerged concerning their use in discourse. I then use data from a range of languages to show that while grammaticalized and non-grammaticalized evidential resources may be utilized for the same range of discourse functions, the fact of grammaticalized evidentiality and the paths that lead to their development lend important support for a model of language as emergent and ultimately subject to the local needs of the interactive context.
Linguistic studies of evidentiality have mostly centred around two themes: typological and semantic. Typological studies (e.g. Aikhenvald, 2004; Aikhenvald and Dixon, 2003; Chafe and Nicholls, 1986; Willett, 1988) have focused on the description of morphological systems that express sources of knowledge, the range of evidential meanings expressed by these morphemes, and their origins and paths of grammaticalization. Much of this work also focuses on the question of the semantic boundaries of evidential meanings – what constitutes the coding of ‘source of knowledge’ as opposed to knowledge status (e.g. as known, unknown, partially known, etc.) (e.g. Chafe, 1986; Mushin, 2001; Palmer, 1986).
Discourse studies of evidentiality have mostly focused on social motivations for evoking source of knowledge, especially in terms of the speaker’s authority over knowledge and the use of evidentials to make claims about whether one is or is not taking responsibility for the validity or factual content of what they say (e.g. Fox, 2001; Hill and Irvine, 1992; Sidnell, 2012). These studies have tended to focus on data from languages lacking morphological systems of evidentiality, where evidential coding is seen as obligatory in every utterance, in favour of languages for which the coding of evidentiality is seen as an optional context-driven choice. While there have been some recent discourse-based studies of morphological evidentiality (e.g. Michael, 2012; Nuckolls, 2012), the areas of focal interest have remained divided, with scholars of grammaticalized evidentiality remaining more focused on typology and semantics while scholars of languages lacking in grammatical evidential resources have tended to focus on their discourse functions. There are, however, at least aspects of evidential use and meaning that raise questions about the usefulness of this historical division of labour in the study of evidentiality.
First, we must consider whether, aside from linguistic form, there are essential differences in evidential pragmatics between grammaticalized and non-grammaticalized evidentials. Mushin’s (2001) comparison of Macedonian, Japanese and English evidential strategies sought to clarify whether there were any differences in the discourse functions of evidential choices according to the grammatical status of evidentiality. That study showed that speakers of languages with so-called ‘obligatory’ grammaticalized evidential systems also made context-driven choices in not only the selection of evidential category (i.e. choosing category A or category B) according to actual source of knowledge, but also in the selection of linguistic resource for expressing that category (i.e. selecting an evidential morpheme or a periphrastic or adverbial evidential expression to express a particular evidential category). That study nonetheless showed that speakers of languages with more grammaticalized evidential resources (Macedonian and Japanese) made more use of evidential resources, grammaticalized or otherwise, than English speakers. This suggests that speakers of languages which have developed grammaticalized evidential forms are more sensitive to making their knowledge status explicit than speakers of languages that lack these resources. 2 I survey similarities and differences in evidential pragmatics across these systems in the next section.
Second, we must consider the types of evidence that are utilized by speakers of languages with grammaticalized and non-grammaticalized evidentials, and their relationship with assertions of validity and degrees of epistemic authority. The typological research on grammaticalized evidentiality shows that while there is some variation in the number and range of meanings associated with grammaticalized evidential systems (Aikhenvald, 2004; Willett, 1988), there is also remarkable consistency in the kinds of sources that have a greater tendency to form the basis for grammaticalized evidential systems. For example, it is relatively common for languages to have grammaticalized ways of indicating that knowledge is acquired from what someone else has said (‘reportive’ evidentiality). However, if we examine corpora of languages that lack grammaticalized reportive evidentiality, we can see that it is a common discourse strategy to attribute knowledge to what someone else has said using a range of strategies (e.g. direct speech, adverbials, etc.) and that these mirror the contexts in which we find the use of grammatical evidential forms. Indeed, for some languages it is possible to trace the development of grammaticalized reportive evidential forms from reported speech strategies (e.g. Mushin, 1997; Travis, 2006). Furthermore, we find regularities in the association of such sources with degrees of validity, so that reportive evidential forms are often associated with a speaker claiming a lack of epistemic authority (i.e. does not take responsibility for the validity of the information). These conventional relationships between purported source of information, validity and degree of epistemic authority suggests that what counts as ‘evidence’ is a feature of human interaction, and independent of the grammatical status of the evidential form. I examine the systematic relationships between source and validity in grammaticalized and non-grammaticalized systems in the following section.
Finally, we must account for the use of particular evidential categories, grammaticalized or not, in contexts where the evoked source of information is not the actual source of information, for example, the use of a direct (witness) evidential in cases where knowledge could not have been gained through direct experience. Again we see very similar conventions of use regardless of the grammatical status of the evidential system, a use related to the adoption of epistemic stances in discourse. The relationship between evidentiality and epistemic stance is the focus of sections 3 and 4.
Taken together these findings show that the motivation to make knowledge visible in discourse through the utilization of evidential resources transcends any particular linguistic system and therefore has something to contribute to our understanding of human social interaction through language. This article thus examines all kinds of evidential coding, grammatical or otherwise, as linguistic resources for drawing attention to the status of one’s knowledge vis-a-vis the way in which such knowledge was acquired. However, I show here that while an understanding of how evidentials are used in discourse must examine all kinds of evidential resources, grammaticalized or otherwise, what becomes grammaticalized in languages should be of particular interest. I return to this point in the concluding section.
Since the linguistic tradition in evidential scholarship has been to separate the grammaticalized from the non-grammaticalized, I begin the detailed discussion next with a summary of recent typological work on grammaticalized and non-grammaticalized evidentiality, focusing especially on what has been observed on the use of such evidentials in discourse. In section 3, I introduce the notion of epistemic stance and show how the pragmatics of stance-taking transcend the grammatical status of evidential systems. In section 4, I argue that despite many similarities in discourse functions between grammaticalized and non-grammaticalized evidential resources, the nature of grammaticalized evidential systems reveals particular features of the ‘epistemic engine’ (Heritage 2012b), especially in terms of what counts as ‘evidence’ when accounting for the status of one’s knowledge.
Grammaticalized and non-grammaticalized evidentiality
According to Aikhenvald (2004), about a quarter of the world’s languages express evidentiality grammatically through morphological forms that directly point to source of knowledge and are considered by both speakers and linguists to be an obligatory category, analogous to tense systems in European languages 3 ; that is, an utterance sounds incomplete or ungrammatical without the use of a form that points to source of knowledge (Aikhenvald, 2004: 6). In most cases, such forms are bound morphemes – suffixes or clitics. In such languages, the sources of knowledge are restricted to a small number of fairly schematic meanings.
Common meanings of grammatical evidentials include witnessed or firsthand knowledge, hearsay or reported knowledge, indirect or inferred knowledge, and sensory (but non-visual) knowledge. Aikhenvald’s (2004) typological survey classifies grammatical evidential systems in terms of their number and type of distinctions. For example, Quechua languages (a widespread group of languages primarily of the Andean region of Colombia, Ecuador, Bolivia and Peru) are classified as languages with a three-way distinction between direct, inferred (or conjectural) and reported, marked by obligatory clitics, as in the following examples: 4
(1) (a) Cha-chruu- this- alma-ku-lkaa-ña bath- ‘Many girls and boys were swimming.’ (I saw them) (b) Daañu pawa-shra-si ka-ya-n- field finish- ‘It (the field) might be completely destroyed.’ (I infer) (c) Ancha-p- too.much- ‘You make my daughter cry too much.’ (They tell me)
Aikhenvald is very clear in her study that the primary meanings of grammaticalized evidentials must be the specification of source of knowledge, with no necessary connection with a speaker’s evaluation of the reliability or validity of what they say. Under this interpretation the evidential form simply indexes the propositional information in an utterance with an actual type of source. For Quechua this would be either acquiring knowledge through direct sensory experience, acquiring knowledge through what someone else has said, or acquiring knowledge by inference from current evidence and knowledge.
Nonetheless, as noted in the Introduction, there are clear relationships between the way one acquires knowledge and the resulting certainty one can have concerning its validity. These often emerge in the use of evidentials in discourse. For example, in Wanka Quechua the direct evidence (witness) evidential is used in contexts where it is clear that the speaker did not acquire knowledge through direct experience. An example of this is in (2), from Floyd (1999: 73), reported in Aikhenvald (2004: 318). This utterance occurred in a context where it is claimed that a certain man has fathered a child. Since he has no direct evidence that the child is not his, but can only infer whether he is the father or not, an inferred evidential would be the most likely option, based on actual source of knowledge. However, a direct evidential was used because the man is claiming a high degree of validity for the child not being his. According to Floyd, using the inferential evidential in this context would leave the claim open to dispute. Claiming direct evidence here is the equivalent of claiming this as indisputable fact.
(2) mana- not- ‘That (the child) is not mine.’
Conventional associations between source of knowledge and validity of knowledge are common in other languages that have grammatical evidential systems (e.g. Macedonian – Friedman, 1986, 2003; Quechua – Weber, 1986; Abhkaz – Chirikba, 2003; Eastern Pomo – McLendon, 2003; and other languages described in Aikhenvald and Dixon, 2003).
Irony may also be expressed with non-firsthand evidentials used in contexts where a speaker clearly has direct evidence, since the marked source of knowledge is at variance with the actual source of knowledge. In Wanka Quechua, the inferential form is used in ironic contexts. In (3), for example, a father makes fun of his daughter’s assumption that she might learn by not going to school (note: Floyd does not give the daughter’s original utterance so we don’t know exactly what (3) is responding to).
(3) chay-nuu-pa- that- ‘This is how you are supposed to learn!’ (Floyd, 1999: 115, cited in Aikhenvald, 2004: 166)
So for languages with grammaticalized evidential systems, like Quechua, questions of discourse function tend to focus less on whether an evidential form is chosen, and more about what evidential category is selected in what kind of discourse context. We can see already though that ‘source of knowledge’ is not an adequate characterization of the use of evidential forms in discourse, in the same way that ‘time relative to time of speaking’ is not an adequate characterization of the use of tense systems in discourse.
Non-grammatical evidential strategies, following Aikhenvald (2004), include (non-exhaustively) the use of adverbs and particles (e.g. reportedly), extensions of other grammatical categories like mood and modality (e.g. must, German subjunctive), reported speech, and complement clauses of perception and cognition. Included among these strategies, for example, are the conjunct–disjunct systems of some Tibeto-Burman languages (e.g. Newari; Bendix, 1992) where verb suffixes indicate whether a situation has been personally experienced or not by the locutor, and the Japanese/Korean systems of marking the private states of other people differently from personal experience through processes of nominalization (e.g. Aoki, 1986; Chun and Zubin, 1990; Kamio, 1998). While many of these linguistic resources are indeed grammatical forms, they are not grammaticalized evidentials of the Quechua kind because their evidential meanings are considered implicated in their use, or extensions of their use, rather than being part of their core semantics. 5
Discourse studies of evidential strategies include discourse functions of particular forms in particular languages. For example, Travis (2006) shows how Colombian Spanish has developed an evidential particle dizque from reported speech marking (from decir ‘say’ plus the complementizer que), and how this particle has attained more general evidential meanings in different discourse contexts. But reportive evidential forms need not be transparently derived from verbs of speaking. Wilkins (1986) reports that many Australian Aboriginal languages have a particle or clitic used to frame an utterance as one the speaker has no direct experience of (usually interpreted as hearsay). In (4), from the Central Australian language Mparntwe Arrernte, the particle kwele is used to frame a story as one that the speaker only knows from hearing it from others. In (5), kwele is used to indicate that the speaker has no direct experience of killing someone – hence deniability of a crime he is accused of. 6
(4) pmere arrwerle Camp long-ago KWELE be-pc, Dreamtime-ABL. (5) the 1sgA KWELE 3sg-0 hit/kill-PC. ‘I’m supposed to have killed him.’ [But I should know if I did, and I didn’t.] (Wilkins, 1986: 586)
The important point for this article is that evidential strategies appear to express the same range of evidential meanings as those found in grammatical systems. For example, languages often draw attention to knowledge acquired directly or indirectly through what other people say (e.g. using quotation, reported speech and/or reportive particles/adverbs), and this may in turn generalize to indirectly acquired information. Examples of the grammaticalization of new evidential forms from reported speech verbs reinforce the salience of information acquired through the speech of others as a source of knowledge. Regular associations of reportive evidential strategies with equivocal validity or certainty point reflect perhaps the understanding that someone else’s words can only ever be mediated information.
Additionally, as for grammaticalized evidentials, evidential strategies need not be a reflection of the actual way that knowledge was acquired. For example, we saw in (3) earlier that reportive or indirect grammaticalized evidential forms are used in contexts where the speaker may well have acquired knowledge through more direct meanings in order to express irony, disbelief or incredulity. We find very similar evidential uses in languages lacking grammaticalized evidentials. For example, the Arrernte particle kwele can also be used in ironic contexts, such as (6), where the speaker addresses a friend with kwele in order to claim that perhaps this person is not (behaving) as a friend.
(6) tyewe ngkwinhe imerte friend 2sgPOSS then KWELE 1sgS. ‘You’re supposed to be my friend.’ [But what you’re doing now makes me question that.] (Wilkins, 1986: 587)
Note that the English translations of the Arrernte examples above show that English adverbials like ‘supposedly’ may share functions with kwele.
So discourse-based studies of non-grammaticalized evidential strategies are consistent with what we know about the uses of evidentials in languages with grammaticalized evidentials: their use does not necessarily correspond with actual source of knowledge, and there are conventional associations between source of knowledge and status of knowledge.
The close association of evidentiality with epistemic modality (which is fundamentally about certainty of knowledge, rather than source of knowledge) among evidential strategies mirrors the associations with grammatical evidentials between source of knowledge and validity of knowledge. 7 It is unsurprising therefore that evidentiality has at times been considered a sub-type of epistemic modality (e.g. Palmer, 1986), or that epistemic modality is a part of evidentiality, broadly defined (e.g. Chafe, 1986).
The intersections between source, validity and modality among both grammaticalized and non-grammaticalized evidentials supports Sidnell’s (2012) casting of evidentiality in terms of resources for managing epistemic asymmetry between conversation participants, particularly in terms of who is claiming greater authority at a particular moment in an interaction. Such conversational work will result in forms that contribute not only information about the source, but also what that source means for whatever the speaker is claiming to be the epistemic status of the information. If we are to understand evidentiality as a linguistic category, and understand more precisely where speakers adopt conventionalized, grammaticalized strategies and where they do not, as well as the degree to which generalizations can be made, then we need a theoretical framework for understanding the discourse functions of evidentials that accounts for the evidential choices we see in language use. In the next section, I argue that if we understand evidentials as deictic markers of stance, then this provides the underpinning for understanding evidential pragmatics more generally, but also the ways in which new evidential systems may emerge.
Bringing it together: Evidentiality, deixis and stance
While there is no logical or necessary connection between source of knowledge and evaluation of that source as valid or reliable or certain, it is well established that this is what speakers often do by virtue of selecting evidential forms. For this reason, evidential marking is considered a part of the more general discourse function of ‘evaluation’. Anything evaluative, however, has to have an ‘evaluator’, just as any information concerning knowledge must imply a ‘knower’. The link between what is being expressed (the propositional information) and the originator of that knowledge highlights what I consider to be a key feature of evidential meaning and usage – its deictic properties. These deictic properties are what enable the systematization of evidential meanings using particular forms, leading to grammaticalization of evidential systems.
Evidentials have been considered a type of deictic expression at least since Jakobson (1957/1971), who classified them as a type of shifter, a verbal category which takes into account three events: a narrated event (En), a speech event (Es) and a narrated speech event (Ens) – EnEns/Es. That is, an evidential category expresses that the speaker is speaking on the basis of someone else’s report, or an earlier experience represented in speech (the ‘narrated event’ must crucially be a speech event), and the narrated speech event is removed from the speech event proper. Jakobson’s formulation here recognizes that the use of evidentials, however they may be expressed, may mediate between two situations, the one pointing to the other.
Crucially, evidential usage points back to some origo, which could be the actual speaker, but could equally be shifted to a represented cognizer. The acquisition of knowledge must be expressed with reference to the individual who has acquired the knowledge. This is seen acutely in quotation, where all deixis must shift to the reported speaker as origo in the content of quote. In such cases any evidential strategy must be ascribed to the quoted speaker and their particular knowledge status, and not the actual speaker. This is illustrated in (7) and (8) for English and Pastaza Quichua, an Ecuadorian variety of Quechua (Nuckolls, 2012) which has essentially the same three-way evidential distinction described for Wanka Quechua (in (1) and (2) earlier).
(7) She said, “I heard that John was going to Barcelona tomorrow.” (8) kimsa tiru-i- three shot- “With three shots he dies” he said.’ (Nuckolls, 2012, ex. 7)
In both (7) and (8) the content of speech is evidentially marked as ‘reportive’ – that knowledge of John going to Barcelona and knowledge of the cause of death (three shots) has been gained from a report by someone other than the quoted speaker. In the Quichua example in (8), the verb of speaking in the matrix clause is also evidentially marked as firsthand information, expressing that the actual speaker was present when the reporting speaker repeated the content of (8). Both examples demonstrate that evidential meanings must be interpreted in terms of the overall deictic orientation (or perspective) of the utterance.
Hanks (2012) rightly points out that evidentials, grammaticalized or otherwise, are not classically deictic in the same way that pronouns or spatial deictics are because of the added layer of speaker evaluation/modality that is associated with their use. Furthermore, as the examples above demonstrate, despite the claim that the core meaning of evidentials is source of knowledge, selection of an evidential category is not simply a matter of indexing an actual source. Rather, evidential categories, like many other linguistic categories, must be understood in terms of a speaker’s motivations (rhetorical, interactional, intentional, etc.) in making an evidential choice in the particular situation which is selected. In other words, grammatical evidentials are a resource speakers use to express their stance towards their knowledge. 8
The analysis of evidentials, grammaticalized or otherwise, in terms of stance-taking avoids the notion that at the core of grammatical evidential systems is an objective designation of knowledge source. All evidential usage must be understood in terms of how speakers position themselves with respect to the content of their talk. Creative uses of evidentials in non-conventional language use (e.g. the use of non-firsthand evidentials in irony) become less about marked or manipulative uses of ‘core’ evidential meanings, and more about the kind of stance a speaker takes with respect to information. 9
Stance-taking is at the core of Nuckolls’s (2008, 2012) (re)analysis of the grammatical evidential system in Pastaza Quichua. Recall that the Quechua system has long been analysed in terms of information source, with three clitics meaning ‘direct experience’, ‘report/hearsay’ and ‘inference/conjecture’ respectively. 10 Nuckolls (2008) argues that the distinction between -mi ‘direct evidence’ and -shi ‘reported’ rather about the representation of the information as emanating from a ‘speaking self’. The ‘speaking self’ in this framework is one who claims authority or responsibility for the utterance, regardless of the actual way information was acquired. Under this analysis, -mi is used when the origo (usually the speaker) is being represented as the speaking self, while -shi is used when someone other than the origo is being represented as the speaking self. The speaking self versus other contrast results in conventional associations with validity. By marking an utterance with -mi, the speaking self, a speaker is taking responsibility for the validity of its contents. By marking an utterance with -shi, an ‘other’ speaking self, the speaker is not taking responsibility for the validity of the contents and may be interpreted as disclaiming or distancing him/herself from contents. The example in (9) comes from a personal experience story about an anaconda. The utterance in (9) is a response to a question by the recipient of the story about the anaconda’s value for love magic. The response asserts that body parts are used for love charms. Nuckolls (2008: 79) accounts for the use of -shi because love charms is an embarrassing topic, and the speaker is here avoiding the impression that she is involved in the making of such love magic.
(9) chi raigu- that reason- a-n be-3 ‘That’s why his brain and his fat are (used for) love charms.’ (Nuckolls, 2008: 79)
The analysis of what has been considered a classic grammaticalized evidential system in terms of its deictic and stance-related properties raises the question of whether other evidential systems should be reconsidered in this way. More elaborate evidential systems (e.g. those with more than three grammatical evidential markers) do indeed appear to genuinely point to source of knowledge rather than to some more general deictic configuration (as with Pastaza Quichua earlier). For example, Makah (a Wakashan language that used to be spoken in the northwest coast of Washington State) has a series of six ‘evidential’ clitics. These are illustrated in (10).
(10) (a) wiki.c’axaw ‘It’s bad weather’ (seen or experienced directly) (b) wiki.c’ (c) wiki.c’ ‘It sounds like bad weather’ (on the evidence of hearing) (d) wiki.c’ ‘I’m told there’s bad weather’ (the quotative) (e) wiki.c’ ‘It’s probably bad weather’ (general inference from unspecified evidence) (f) wiki.c’ ‘My goodness! It’s bad weather!’ (surprise/sudden realization) (From Jacobsen, 1986: 10, 19, and Davidson, personal communication)
Languages with relatively large numbers of evidential morphemes, like Makah, tend also to be the ones for which little has been written on discourse functions. 11 Schlichter (1986) commented on the deictic properties of Wintu (California), which also has a more elaborated set of evidential morphemes. It seems clear that such evidentials do express an origo’s stance towards knowledge, which may emanate from a specific type of knowledge source (as in (10a–c)), or it may more generally express the origo’s evaluation of the status of knowledge (as uncertain, or unexpected, as in (10e) and (10f)). 12
Grammaticalized forms arise from the ways people use language. Grammaticalized evidential forms have many sources, but at least some of the evidential markers in Makah and Wintu systems derive from verbs of speaking and perception. This suggests that at least the origins of such evidential systems lay in actual specification of knowledge source. We can assume that even prior to the grammaticalization of the system, speakers took overt stances towards their knowledge that over time conventionalized into grammatical systems. In theory, this practice should emerge in an environment where knowledge is never treated as objectively acquired (e.g. facts floating about the world waiting to be picked up) and interlocutors must position themselves with respect to the knowledge they have. I will return to this notion in the Conclusion.
Adopting evidential stances with grammatical and non-grammatical evidential strategies
In languages with ‘obligatory’ evidential systems there also are times where speakers do not use an evidential form, even when an evidential would be normally required by the grammar. For example, Floyd (1993) notes that speakers often omit evidential clitics in ‘free-flowing’ conversation, but does not develop an analysis of where such omissions tend to occur. 13 Aikhenvald (2004: 78) reports that evidentials may be omitted in some languages where the same evidential category is sustained over a number of utterances. For example in Tariana, evidential forms ‘… can be omitted in narrative if the time and evidence frame is set in the previous or in the following clause, or is clear from the context’. 14
This suggests that the stance projected by the adoption of an evidential form (e.g. a reportive form in some kinds of storytelling) may be understood as sustained until a new evidential marker is used. While further study of this phenomenon is required, we can say that use of evidentials to frame a particular stance is consistent with its deictic properties and is consistent with the behaviour of other deictic categories (e.g. the omission of tense marking in clause chains).
Even languages that have rich morphological evidential systems may also have resources for signalling evidential meanings in other ways. All languages have the means to represent the content of speech (and thought), for example, and may do so to attribute information to someone else. This is called an ‘evidential’ use of reported speech (Mushin, 2001: 112; see also Aikhenvald, 2004: 132ff). The Pastaza Quichua example in (8) earlier illustrates the use of evidential reported speech, as the content of speech reports what the actual speaker heard (directly) from someone else. In this example, reported speech appears to operate in tandem with the grammatical system. For example, the speaker uses the ‘speaking self’ form -mi form on the verb of speaking to reflect her direct experience of hearing the story from the reported speaker.
In Mushin (2001), direct reported speech was shown to be a regular evidential strategy in Macedonian narrative retelling, where retellers elected to retell a story they had just heard to a third person by directly quoting the original storyteller, as in (11).
(11) I veli “Tatko mi beshe posle na krajot” veli and say(3sg) father 1sg “A listatsijata pochna od pet ilijadi. I nagore” veli and auction( “krenavme do kaj sedum” veli “i pol ilijadi” veli rise(1pl And she [previous teller] says, “My father was last in the end.” She [previous teller] says, “and the auction started from $5000 and up” she [previous teller] says. “We raised (it) to seven,” she [previous teller] says, “and a half thousand.” she [previous teller] says. (Mushin, 2001: 112)
What makes the use of evidential direct speech particularly interesting in Macedonian is that it appears to be used as an alternative strategy to a grammaticalized evidential system. Macedonian past tense distinguishes between ‘simple past’, used when speakers claim to have direct experience of their knowledge or are prepared to ‘vouch for’ their knowledge, and the ‘l-form past’, used when speakers claim not to have direct experience of their knowledge. This contrast is illustrated in (12), and also in (11) where the content of direct speech (what the original teller reported) is presented as the direct experience of the original teller, using simple past.
(12) (a) Taa go 3sg 3sg make.bread(3sg ‘She baked the bread,’ (I saw her do it/I vouch for it) (b) Taa go 3sg 3sg make.bread( ‘She baked the bread.’ (that’s what she told me/supposedly/that’s what I heard)
The existence of the contrast between simple past and l-form past in Macedonian means that to use a past tense, the tense typically associated with narration, speakers must take a stance with respect to the acquisition of knowledge. The regular use of evidential direct speech in contexts where speakers are retelling someone else’s story is one way to take the conventional (reportive) stance, but without utilizing the grammatical system. This suggests that it is the stance, rather than the choice of linguistic resource used to express that stance, that is of primary concern to Macedonian speakers, at least as far as evidential strategies are concerned.
It may be that Macedonian speakers were motivated to adopt a reportive stance towards the story they retold because their language makes a grammatical distinction between firsthand and non-firsthand knowledge, leading to speakers’ sensitivity to taking an evidential stance at all. Or it may be that the consistent adoption of a reportive stance is reflective of the kind of pragmatic circumstances that led to the grammaticalization of evidentiality in Macedonian in the first place. It is well documented that while grammatical evidentiality in the Balkans may have been emerging at the time of the Ottoman conquest, it fully emerged under the influence of Turkish, which makes a similar firsthand/non-firsthand distinction (Aikhenvald, 2004: 288). This suggests that while language contact may have nudged the process along, already the pragmatic conditions made it an easy transition for past tenses to attain regular evidential contrasts.
Whether by evidential framing or by the adoption of an alternative evidential strategy, like reported speech, speakers of languages which have grammaticalized evidential systems do appear to consistently adopt explicit stances towards what they talk about in terms of the resulting status of their knowledge. So while we can still view the use of evidentiality as a resource for managing epistemic asymmetry, we can also see scope for understanding cultural difference in the role evidential claims may serve in doing this conversational work.
It should be noted that for many other languages, adoption of an evidential stance does not appear to be such a high priority, at least for many discourse contexts. In contrast with Macedonian retellers, American English narrative retellers did not consistently indicate that the story they retold came from another source, and in some cases did not indicate that the story was another person’s story at all. The hypothesis developed in Mushin (2001) for this difference suggested that the focus for the English speaking retellers was to tell as ‘good’ a story as possible, rather than to represent the story as a retelling. Macedonian (and Japanese) speakers, in contrast, appeared concerned to both tell a good story and to represent the story as a retelling, taking the appropriate evidential stance towards it.
There are, however, many discourse contexts where English speakers do indeed pay attention to the source and status of knowledge. As pointed out in Fox (2001), and more recently in Sidnell (2012), these tend to be contexts where epistemic authority must be established and/or negotiated. They come to the fore in public domains such as the press and broadcast media, and in legal and forensic contexts, where veracity must be justified.
If we move away from English, we can see how variable adoption of evidential stances can be, but also more evidence of where aspects of evidential usage appear to be a feature of human communicative behaviour. When I began fieldwork on the Aboriginal Australian language Garrwa, I was struck by the lack of evidential marking in discourse, and indeed the overall lack of epistemic modification. Stories were told with no evidential framing (see Arrernte dreamtime stories reported earlier, which are marked as distanced from the teller with the particle kwele). 15
While there does not appear to be a Garrwa form that points to how knowledge of past events was acquired, an epistemic particle junu ‘don’t know/perhaps’ may be used to indicate a lack of knowledge without implication of how knowledge was acquired. In (13) the speaker indicates that her knowledge of birds being in the east is not certain. This was uttered as part of a conversation about the best places to find particular birds.
(13) julaki nayi karri bird this east ‘There might be birds here eastwards.’ (3.9.03.4.DG)
While it is also possible to formulate evidential strategies using existing linguistic resources in Garrwa (e.g. direct speech, perceptual framing), I have found very few examples in the extensive corpus of conversations and narratives I have recorded over the last 10 years. The main context in which evidential strategies are found in my corpus is where knowledge is directly being disputed, but as argued in Mushin (2012), even under these circumstances there are relatively few recourses to source or status of information.
The extract in (14) illustrates a lack of evidential marking, even when there is a dispute about (factual) knowledge.
(14) Porch Conversation
OlSi-
waterlily mix food ‘Her sister used to take the hot waterlilies mixed with food.’
‘Give’
‘But, no they didn’t.’
old.woman ‘They didn’t eat it like that, old woman.’
Grind 3Pl ‘They grind it.’
Grind 3Pl ‘They grind it’.
Grind 3Pl ‘They grind it up with a stone and put it in a billycan.’
see 1sg ‘I used to watch.’
Really?
In (14), Kate is describing a way of preparing bujuwan, which are water lily roots, a local delicacy. In line (f), Daphne disputes Kate’s account with a direct contradiction ‘no they didn’t’, repeated before offering her own account (‘They grind it’). This account seems to be accepted by Kate because she repeats it and elaborates, continuing the account. Note that Daphne provides no modification when she asserts her contradiction, even though Kate is the one telling the account (and therefore presumably with epistemic rights). In (q), Hilda, who had hitherto not been part of the dispute sequence, adopts an evidential strategy when she supports Daphne’s characterization of the preparation of bujuwan by claiming direct sensory (visual) evidence of grinding. While it is difficult to analyse why Hilda uses an evidential strategy when Daphne and Kate do not, it is possible that it is because she has initiated her own entry into this discussion that Hilda must establish her own epistemic credentials, hence the evidential strategy. Note the newsworthiness of the assertion of visual evidence in Daphne’s response ngarri really?
Conclusion: Evidentiality and territories of knowledge
Clearly, there is far more to say about the discourse functions of linguistic evidentiality, especially in how such functions relate to the range of evidential resources speakers of any given language may have to express evidential meanings. In this article, I have used data from a broad range of languages – Quechua, Makah, Macedonian, Garrwa and English – to explore whether the grammatical status of evidentiality in any given language affects the way it is used.
This exploration has shown that pragmatically speaking, evidential strategies are remarkably similar regardless of grammatical status. We see very similar kinds of sources being evoked when evidential strategies are adopted. We see the same range of conventional relationships between source of knowledge and claims of validity, responsibility and authority. We see the same range of extensions to marked pragmatic functions such as irony and traditional storytelling. I have argued here and elsewhere that this pattern of evidential usage is more consistent with the analysis of evidentiality as a deictic phenomenon associated with the adoption of epistemic stances more generally, than with an indexical relationship between speaker and source of information.
However, the results of this exploration have also shown that the existence of grammaticalized evidentiality in a language does make a difference to the extent that speakers are motivated to regularly focus on knowledge source in designing their utterances. Aikhenvald (2004) reports that speakers of languages with obligatory evidentiality notice its lack in other languages. One does not have to be deterministic to show that obligatory evidential systems emerged from systematic expression of knowledge sources in everyday discourses, and thus must reflect a cultural practice of drawing recipients’ attention to the status of knowledge with respect to source.
For languages that lack an obligatory system, evidentials may still play a significant discourse role, marking particular epistemic stances and negotiating epistemic authority. While the extent to which explicit evidential resources are utilized is highly contingent on the discourse context, we do see examples, such as the Australian Aboriginal language Garrwa, where explicit evidential (and indeed epistemic) resources are rarely used. To understand this better we need further analysis of the role knowledge (both individual and communal) plays in social interaction in these cultures. This is the subject of ongoing research.
As noted throughout this article, evidential resources, whatever their linguistic status, tend to reflect the same kinds of knowledge sources: direct sensory experience, report from others, inferential processes based on external evidence. Variation exists such that some languages specifically separate visual evidence from other kinds of sensory perception, or other languages distinguish inferences depending on the kind of evidence one uses to draw the inference. However, some kinds of knowledge sources are never explicitly coded and never grammaticalized. I know of no language that distinguishes reported information depending on who the reported speaker is (e.g. a reliable or unreliable source), or knowledge acquired from a dream or hallucination (languages typically apply other evidential categories conventionally to such information), or knowledge claimed to come from pure imagination.
What can we infer from these patterns of grammaticalization and language use? The kinds of knowledge that get packaged into linguistic units relate to our sensory experiences and our experiences of social interaction. Inferential evidential forms ultimately must point to knowledge acquired from sensory evidence or report, but mediated through additional thought processes. As noted earlier, evoking knowledge as reported draws attention to the facts that a) much of what we know is gained through social interaction and b) much of what we know is mediated by other people’s knowledge and experience and must be evaluated in those terms. Evoking knowledge as sensory experience highlights that a) much of what we know is gained by perception and b) as humans we assume shared perceptual and cognitive abilities so that recipients will be able to evaluate knowledge claimed to come from others’ direct perception.
To cite sensory or linguistic evidence as the basis for one’s knowledge is thus to draw on (potentially) publicly verifiable evidence – what anyone might have seen or heard if they had been in the speaker’s place. Knowledge gained from purely internalized sources, such as dreams, hallucinations or conscious creative thinking, does not draw on some kind of external evidence or assume that others could come to a similar knowledge status in the same situation, and this may account for their absence in evidential systems.
Footnotes
Acknowledgements
Many thanks to Teun Van Dijk for organizing the Knowledge in Discourse symposium in April 2012 out of which this paper has emerged. I thank the audience members at that symposium and at a University of Queensland Linguistics Seminar for feedback on earlier versions of this paper.
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.