Epistemic authority and sociolinguistic stance in an Australian Aboriginal language

Murrinhpatha, an Aboriginal language of northern Australia, has an initial k-alternation in verbs that has hitherto been resistant to grammatical analysis. I argue that kdoes not encode any feature of event structure, but rather signals the speaker’s epistemic primacy over the addressee. This authority may relate to concrete perceptual factors in the field of discourse, or to socially normative authority, where it asserts the speaker’s epistemic rights. These rights are most salient in the domains of kin, country and totems, as opposed to other topics in which speakers are habitually circumspect and co-construct knowledge. My analysis of the k-alternation thus brings together the typology of epistemic grammar (Evans, Bergqvist, & San Roque, 2018a, 2018b), and a sociolinguistic perspective on stance (Jaffe, 2009).


Introduction
The assumption of superior knowledge, and contestation of this assumption, are ubiquitous in conversation (Heritage, 2012;Stivers, Mondala, & Steensig, 2011). But one may also ask to what degree there are culturally specific norms for assuming and contesting knowledge. Much of the work on the assumption of 'asymmetric epistemic access' -i.e. superior knowledge -draws evidence from Englishlanguage conversation. Meanwhile, with respect to Australian Aboriginal societies, it has been argued that consensus, collaborative knowledge and epistemic equality are the preferred modes of interaction (e.g. Liberman, 1982Liberman, , 1985. Claiming epistemic authority is a 'marked' act, and often depends upon kin-and clan-based determinations of who has the right to know (Keen, 1994;Myers, 1986).
In this article I discuss Murrinhpatha, an Aboriginal language for which both consensual interaction and rights to knowledge have been well documented (Blythe, 2009(Blythe, , 2013Stanner, 1936Stanner, , 1966Walsh, 1997). Murrinhpatha verbs exhibit forms where the initial consonant is replaced by k-, and this alternation has hitherto been resistant to grammatical analysis. I argue that the general function of this alternation is the marking of epistemic authority, which can be seen as a pragmatically marked strategy in Murrinhpatha, compared to the unmarked norm of epistemic equality (1, 2). Epistemic authority may be claimed on the grounds of concrete perceptual access, but also on socially normative grounds -i.e. because the speaker claims the right to know. I propose that the Murrinhpatha k-alternation reflects the markedness of epistemic authority in Murrinhpatha culture. (2a) dem-wirnturt (2b) kem-wirnturt pierce.rr.3sg.nfut-get.up pierce.rr.3sg.auth-get.up 'She/he gets up. ' 'She/he gets up.' (speaker authority) The analysis presented here draws on recent work in the typology of 'engagement' -where grammar encodes not just propositional content, but also speakers' and addressees' epistemic relation to the content (Evans et al., 2018a(Evans et al., , 2018bBergqvist & Knuchel, this volume). A canonical engagement system is that of Andoke (Witótoan: Landaburu, 2007), where all declarative and interrogative sentences have an 'auxiliary base' marking a symmetrical paradigm of the four possible configurations of epistemic access for speaker and addressee [+spk+addr, +spk-addr, -spk+addr, -spk+addr]. The concept of engagement has some overlap with the concepts of 'mirativity' (DeLancey, 1997) and 'egophoricity' (San Roque, Floyd, & Norcliffe, 2018), though engagement's focus on speaker/addressee epistemic states and attentional coordination is a closer fit for the Murrinhpatha data. Though engagment is a new typological concept, its applicability to an Australian language has already been shown for Jaminjung (Schultze-Berndt, 2017), which is an immediate neighbour of Murrinhpatha, though the languages are not closely related. A comparison of Murrinhpatha and Jaminjung engagement marking is provided below ( §7). The Murrinhpatha k-alternation is not canonical engagement of the type found in Andoke, but rather provides a single form for marking speaker authority, i.e. [+spk-addr], while engagement is otherwise unmarked in the verb. Furthermore, Murrinhpatha k-is used only with third-person verbs in non-future (nfut) tense, and is much more frequent on certain 'postural' verb stems than others. I will argue that this rather contingent type of engagement marking has probably developed by reanalysis of a person/TAM marker. I will also propose that the optionality of Murrinhpatha k-is expected, given what we know about Aboriginal cultural epistemology. One of the main contributions of this article is to situate the grammar of engagement in its sociocultural context, thus linking this new typological endeavour with anthropological and sociolinguistic perspectives. This article also draws on concepts from conversation analysis, since epistemic stance forms an important part of interactional management (Heritage, 2012;Stivers et al., 2011; though see Lindwall, Lymer, & Ivarsson, 2016 for a dissenting view). I analyse Murrinhpatha epistemic authority using fragments of natural discourse, in which the preceding or subsequent turns often provide crucial evidence. I draw these examples from the Murrinhpatha Morpho-corpus (Mansfield, Blythe, Nordlinger, & Street, 2018), a 100,000-word corpus of spontaneous and elicited Murrinhpatha speech, which is morphologically annotated.
In the next section I review observations of cultural epistemology and interaction in Australian Aboriginal societies generally, and in Murrinhpatha in particular ( §2). I then introduce the formal characteristics of the Murrinhpatha epistemic authority marker ( §3), before providing examples of its attentional coordination function in discourse ( §4). I then contrast 'concrete' epistemic uses with socially normative epistemic authority ( §5), and present corpus frequency evidence to support my claims of social stratification ( §6). I compare Murrinhpatha and Jaminjung engagement marking ( §7), then explore likely grammaticalisation pathways and reflect on connections between linguistic form and social context ( §8).

Consensus and epistemic rights in Murrinhpatha interaction
The Murrinhpatha people of northern Australian were until recently semi-nomadic hunter-gatherers (Falkenberg, 1962;Stanner, 1965), but during the last century have transitioned to sedentary life in the town of Wadeye (Pye, 1972). Several language groups settled in Wadeye, but since the town was founded on Murrinhpatha land, other groups underwent language-shift to Murrinhpatha, in accordance with Aboriginal beliefs about the interconnectedness of language and place (Rumsey, 1993). Thus the Murrinhpatha speaker population has grown over the last century, and now numbers some 3000 people, with children learning it as their first language (Taylor, 2010).
It has been claimed that Australian Aboriginal societies in general prefer an interactional style that is circumspect, collaborative and consensual (e.g. Eades, 1982;Walsh, 1997). For example, it is reported that Pitjantjatjara speakers are reticent about offering personal opinions, that talk is filled with phatic consensus markers palya, uwa, affirming knowledge-held-in-common, and that interlocutors often echo each others' statements (Liberman, 1982(Liberman, , 1985. Statements of personal opinion, as opposed to shared knowledge, are preferably aired in early morning 'broadcast' speech, which is declaimed in the hearing of the stillsleepy group, without any specific interlocutor (Liberman, 1985, p. 3ff.;Walsh, 1997). But while epistemic symmetry may be a general preference in Aboriginal societies, there are certain circumstances in which epistemic authority rests with particular parties, and is carefully controlled (e.g. Keen, 1994;Michaels, 1986;Mushin, 2012;Sansom, 1980). Sometimes this may simply be a matter of differential perceptual access. But it may also involve 'epistemic rights' which are particularly salient in the domains of kinship, country, totems and language. To simplify somewhat, European Enlightenment epistemology grounds knowledge in perceptual evidence and logical induction (Solomon, 1988). The knower faces the world directly, and alone. By contrast, the Aboriginal knower is situated in a kin-based social network, and epistemological access is determined by one's place in this network (cf. Shweder & Bourne, 1984;Spiro, 1993). Totems, country, ancestors and language are known because one is related to them, and this knowledge is treated as an inalienable right (Stanner, 1936(Stanner, , 1966. Murrinhpatha interaction exhibits both the default consensual stance, and the emphatic demarcation of epistemic rights. In the default setting speakers often co-construct information in such a way that no speaker could be accused of assuming epistemic authority, with each speaker providing only partial contributions to a shared body of knowledge, replete with omissions and doubts. Speakers actively seek confirmation from addressees, and recruit co-narrators to provide details. In many cases this is not because the facts are actually in doubt, but because this is a socially normative form of interaction. Blythe (2009) provides conversation analysis of several Murrinhpatha interactions that demonstrate circumspection and collaboration. In one passage (Blythe, 2009, pp. 312-313), two elderly women recollect a mishap that they experienced together many years earlier (3). Other protagonists in the mishap are now deceased and their names must be avoided. Thus the passage is characterised by name avoidance (in lines 4 and 6), but there is also general scarcity of detail. The sequence initiator (EC) provides information in small chunks, while her co-narrator (PB) crafts questions to propel the narrative. The rationing of information is particularly clear in line 6, where a semantically broad noun kura 'liquid' is the only clue given as to the purpose of the ill-fated car trip. In this example, narrator and co-narrator have equal epistemic access to the nature of the trip (bringing beer to a bush camp) and identities of the protagonists (two deceased kin). Speakers respect this epistemic equality by refraining from providing too much information.
(  (Blythe, 2009, pp. 312-313) A further example, this time from a conversation between two young men, illustrates features of both narrative co-construction as well as epistemic non-assertiveness (4). LP initiates the narrative about typical recreational activities of young men in the town, but when BP fails to offer any contribution after four consecutive turns, LP recruits BP to fill in the next piece of information, with a direct question accompanied by a gesture (line 4). The interrogative thangku 'what, why?' in itself would be enough to formulate the question, but it is here accompanied by the enclitic =kama uncertain, which encodes a speaker's uncertainty about propositional content. The =kama clitic is used again by both speakers in lines 5-6. However, as with the previous example (3), the facts of the matter under discussion are quite clear to each speaker, being in the domain of their own peer group. Mi wiwi 'ganja' is a mildly taboo topic, and thus provides socially normative, rather than epistemically concrete, reasons for encoding uncertainty and recruiting a co-narrator. This passage therefore demonstrates a performative aspect to the non-assertive stance (Kiesling, 2009, p. 177).
(4) 1. LP: Purtpurl kirra-dha ini=ka laif ngalla=matha football stand.3st.irr-pst anaph=cst lively big=intens 'When the football is on, things really liven up.' 2. LP: Kardu kigay kani-wat-tha purtpurl person young.man be.3sg.irr-attend-pst football kardu mere nangkal kani-dha tjalput=yu person neg anybody be.3sg.irr-pst house=cls 'The young men go to the football, nobody stays at home.' 3. LP: Kardu purni-dha da ngarra obal da=matha person go.3pl.pst-pst place to oval place=intens 'They all go to the oval.' 4. LP: Pardi-yedhek puy puddini-wurl thangku=kama be.3pl.pst-play continue turn.3pl.pst-return what=unctn 'After playing they go back home, for what?' (Open-handed gesture towards BP) 5. BP: Mi wiwi=kama veg smoking=unctn 'Maybe to smoke ganja.' 6. LP: Mi=kama puddini-wi veg=unctn turn.3pl.pst-blow 'Maybe to smoke.' (2012-06-30_LP-BP) For a contrasting example in which speakers vigorously defend their epistemic rights, we again turn to Blythe (2009, p. 331).1 In this passage, the identity of someone's deceased husband is at issue, and a misunderstanding leads to a tussle over epistemic rights to the deceased (5). In lines 2-3, RI claims to know the identity of the deceased, formulating this as a statement of kin relationship. There is a complex nameavoidance technique in line 3, before RI and EC vigorously dispute each others' claims in lines 4-6. RI uses two linguistic forms to fortify her epistemic stance: firstly the -nimin contrastive suffix in line 2 indicates that her proposal is in opposition to some other possible candidate(s); secondly, she uses k-auth on all verbs (underlined). Examples in the sections below further demonstrate how k-auth is associated with the stance of epistemic authority.  (Blythe, 2009, p. 331)

Murrinhpatha verbs and the epistemic authority inflection
Murrinhpatha is a non-Pama-Nyungan, Australian Aboriginal language, best known in the literature for its highly complex verbal morphology (e.g. Mansfield, forthcoming;Nordlinger, 2015;Walsh, 1976). The epistemic authority marker is part of the complex inflectional system. Murrinhpatha verbs are built on 'finite verb stems'. These encompass highly irregular inflectional exponence, though they are nonetheless referred to as 'stems' in the Murrinhpatha literature because they are not clearly segmentable, and because they form the base for complex wordforms by combining with agglutinative inflectional suffixes and compounded stem elements.2 Finite verb stems are a closed class of 39 members (listed in It is the finite stem in which epistemic authority marking occurs. The finite stem has internal morphological structure, though as in the examples above, this is usually not shown as segmented in Murrinhpatha literature. This is partly because the inflections involve several suprasegmental alternations such as vowel fronting and consonant gemination, and partly because the morphological patterns are highly inconsistent from one finite stem to another (Mansfield, 2016b). For example, the contrast between 3sg and 3pl forms of 'use feet' in (6, 7) above is made by the addition of a pu-prefix and gemination n → nn. The 2sg form in (8) takes a thu-prefix, and being in irr rather than nfut tense, loses the -ngam suffix. However these morphological patterns do not appear consistently with other stems. Table 1 illustrates all 42 forms of the 'use feet' stem, including epistemic authority alternations in third-person, nfut forms. As can be seen in the table, verb stems also have k-in 3.irr forms. While this may be historically related (see §8 below), its synchronic function is modal and temporal (Nordlinger & Caudal, 2012), and does not appear to have a role in engagement marking.

Posture verb stems
Of the 39 Murrinhpatha finite verb stems, seven form a distinct class of 'postural verb stems', e.g. dim 'sit.3sg. nfut', pirrim 'stand.3sg.nfut' and wurran 'go.3sg.nfut'. As we will see below, the auth inflectional form (e.g. kem 'sit.3sg.auth') is much more common with posturals than other verb stems. Postural stems can form simple verbs on their own, in contrast to other verb stems, most of which appear only in compounds. Posturals are also distinguished by their use as imperfective markers when compounded to a main verb, in which case their postural semantics may become a secondary predication (10), or may be lost altogether (11).

The k-alternation
Epistemic authority is encoded by an initial k-consonant on third-person, nfut verb stems, which are the most frequent person/TAM forms in discourse.3 The k-consonant usually replaces the initial consonant of the 3sg.nfut form (15-17), though in a few verb stems a ka-~ ku-syllable is added (18,19). Under some morpho-phonological analyses this may be considered a prefix (Mansfield, forthcoming), but I here refer to it as 'the k-alternation'.4 The k-alternation is thus more irregular, and more contingent on grammatical context, than canonical engagement systems. In the glossing system used by Evans et al (2017a), k-encodes +spkr-addr.engag, but since k-does not contrast with other spkr/addr values, I gloss it simply as 'auth'. Auth forms appear to be possible for all finite verb stems, but they are infrequent for most. They are also infrequent in plural forms. Table 2 illustrates the percentage of 3.nfut stem tokens that use auth forms in the Murrinhpatha Morpho-Corpus. For posture verb stems, auth is relatively common (8-64% of instances), while for other verb stems it is rare or absent in the corpus (≤ 5% of instances), though it may be elsewhere attested in elicitation. There are three verb stems which have k-in all 3sg forms: kanam 'be', kanthin 'carry' and kanthangan 'crouch'. These three verb stems therefore do not encode epistemic stance, and are excluded from examples and token-counts below.5 Table 2. Corpus percentage of 3.nfut verb stems with auth marking. Each stem is illustrated in 3sg.nfut, 3sg.auth forms. Question marks appear where auth is unattested. Stems glossed in small caps are semantically bleached, and their glosses are largely conventional (Mansfield, 2019).

Attentional coordination functions
The function of the Murrinhpatha k-alternation has proven somewhat elusive, as its use does not have obvious translational consequences in English. Nordlinger and Caudal (2012, pp. 101-102) note that its function is 'not well understood', and report that an informant provides both the auth and the epistemically unmarked nfut form to describe the same situation (20).6 (20a) kem-tharl sit.3sg.auth-bathe 'He's sitting in the bath-tub.' (20b) dim-tharl sit.3sg.nfut-bathe 'He's sitting in the bath-tub.' (Nordlinger and Caudal 2012, pp. 101-102) I argue here that auth is not required to encode any grammatical or semantic feature of event structure. Rather, it is a pragmatic stance, which explains why speakers in elicitation generally neither insist on it, nor dismiss it as incorrect. Below I exemplify the usage of auth for 'attentional coordination' -in particular, where the speaker is attending to something, and presumes that the addressee is not. One Murrinhpatha speaker explained auth to me as 'You can see it,' which I interpret as an indicator of this attentional function. As in other forms of grammatical engagement, attentional coordination may relate to entities, or to events (Evans et al., 2017b). Entity attention is coordinated by existential and locative constructions ( §4.1), while event attention is coordinated by verbal predicates ( §4.2).

Existential and locative constructions
Previous studies including Nordlinger and Caudal (2012) and Street and Kulampurut (1978) have labelled the k-alternation 'existential' and described it as 'asserting the existence of an entity, place, state or event' (Nordlinger & Caudal, 2012, p. 101). In fact many of the examples are existential and/or locative, and it is typologically common for these meanings to overlap (Bentley, Ciconte, & Cruschina, 2013). As mentioned above, Murrinhpatha encodes locative and existential predicates using 'posture verbs', and in these functions the posture verbs are indeed very frequently marked with the k-alternation, which has led to its interpretation as an existential marker. However I will argue below that the existential/locative function is best understood as one usage case in which attentional coordination is particularly frequent. Intersubjectivity is implicit in existentials, which presume an entity to be inaccessible to the addressee (Freeze, 2001;Lambrecht, 1994, p. 179). The use of auth in an existential construction is illustrated in (21). EC uses the unmarked nfut to enquire about the presence of a desirable fruit species (line 1). LT somewhat distractedly answers in the affirmative (line 2), but EC indicates ongoing uncertainty (line 3). The failure to coordinate attentional states has now become salient, leading LT to use auth in asserting fruit's existence and location (line 4).
Some existential examples have an 'alert' function, where the speaker draws the addressee's attention to an entity that should concern them (22, 23). Conversely, auth may be used in assertions of non-existence where the addressee falsely assumes the contrary (24). While auth is common in existential/locative constructions, it is not grammatically required. Thus we also find existentials where the epistemically unmarked nfut form is used (25, 26). Here we see nfut used where the interlocutors have already signalled their shared attention to the referent (25), and when an existential is cautiously offered for confirmation (26). These are both co-constructed passages, with no speaker taking an authoritative stance.  (27). This passage is the continuation of the 'drowned man' passage above. BB uses auth to propose a tree as a reference point, but RM does not yet recognise the reference (lines 1, 3). RM then offers his own tree, seeking coordination from his own reference point, for which he also uses auth (line 4). It is only after the interlocutors reach a shared epistemic representation of the tree that BB switches to the nfut form (line 6).

Use with other types of predicates
Examples above show that auth can be used with existentials and locatives, but is not grammatically required for them. Conversely, we also find examples where auth is used in predicates that are not existential or locative. Some of these use postural stems in compound verbs, where the primary predication is no longer postural. Other examples illustrate auth in non-postural verb stems. These examples show that the best generalisation about auth usage involves epistemic pragmatics, rather than existential/ locative semantics.
Examples (28, 29) illustrate 'alert' situations, which are like (22, 23) above in that the speaker draws the addressee's attention to something that should concern them. But whereas the above examples alerted addressees to entities, these two examples alert addressees to events (watching, recording). In (28) three speakers know that they are being recorded and are wondering what they should talk about. In line 2, KM observes that the camera is on, but this is already clear to all participants, so it is encoded with nfut. But in line 3, BB makes the more startling observation that the researcher is looking at them (the researcher was standing some way off, trying to be unobtrusive), so they should start speaking. BB presumes that his addressees are not aware of this, and therefore encodes the 'watching' predicate with auth. Example (29) is quite similar to the previous. In this instance, the addressees had been aware that they were being recorded, but in line 2 LT seems to forget this when she playfully insults DT. SB then reminds her that she is being recorded, using auth to imply her ignorance of this fact. A final example (30) shows a speaker using auth when correcting the addressee's misconceptions about a state of affairs. In line 1, SD makes an incorrect supposition about an aeroplane being the cause of an engine noise. LP provides an alternative guess (line 2), before adducing visual evidence that his proposal is epistemically superior (lines 5-6).

A cognate prefix in Ngan'gi
Murrinhpatha's sister language Ngan'gi has a cognate g-verb prefix, which has also proved somewhat elusive to grammatical analysis (Hoddinott & Kofod, 1988, p. 113;Reid, 1990, pp. 108-111). Reid tentatively proposes that Ngan'gi g-encodes 'states that are remote from the speaker, either spatially distant or conceptually distant', labeling the category 'subject remote' (sr) (p. 110). In terms of engagement theory (Evans et al., 2017a), we may interpret Reid's formulation of 'conceptually distant' as meaning 'epistemically inaccessible'. But whereas Reid focuses on the epistemic state of the speaker, engagement theory highlights the epistemic coordination of speaker and addressee. This leads us to note that in each of Reid's examples, the speaker is attending to something that can be construed as inaccessible to the addressee (31, 32). Thus the Ngan'gi g-prefix may have a similar function to the Murrinhpatha k-alternation. On the other hand, Reid's examples involve more 'out of sight' or 'in the distance' predicates, which raises the possibility that the Ngan'gi cognate is more specifically related to visibility.
Ngan'gi (31) Yu Darwin kine, wembem yaga, yes place.name this house dem buy-gumum-derri-gaganim white-do.3sg.prs.sr-back-go.3sg.prs.sr 'Yes this is Darwin alright, the houses are standing out white in the distance.' (32) Yu wetimbi kinyi gibem yes didgeridoo here lie.3sg.prs.sr 'Yeah OK! (the didgeridu) is here (inside the house / out of sight).' (Reid 1990: 111) Ngan'gi also furnishes examples where the predicate is not existential or locative. In (33) the speaker has asymmetric access to the identity of someone on the other end of the telephone, and the 'talking' predicate is thus marked with auth.

Socially normative authority and 'heavy speech'
The functions of auth described above involve epistemic asymmetry rooted in actual differences of perceptual access, recognition of a referent, or at least presumed differences where the speaker believes the addressee to be unaware of something. But examples of this type do not account for all the corpus instances. There are other instances in which the speaker does not have superior epistemic access via physical perception such as visibility or audibility, but rather on socially normative grounds, because their social status makes it proper for them to have special knowledge of some referent (see §2 above). Enfield characterises this type of epistemic stance as 'source-based authority' versus 'status-based authority' (Enfield, 2011, p. 300). In the typology of engagement, it is notable that some systems like Kogi (Chibchan: Bergqvist, 2016) separate out epistemic authority and epistemic accessibility, whereas Murrinhpatha conflates the two.
In this section I discuss how Murrinhpatha auth is indexical of 'stance' in the sociolinguistic sense (e.g. Jaffe, 2009), which is somewhat different from the sense used in studies of epistemic management (e.g. Heritage, 2012). Each of these literatures draws attention to a different aspect of how speakers situate themselves in discourse. Whereas epistemic studies tend to abstract away from the different social identities of speakers, sociolinguistic stance focuses on culturally specific tropes that speakers draw upon in the course of linguistic interaction. For example, Kiesling (2004) shows how the term dude in American English indexes a stance of 'cool solidarity', which is a highly valued form of youth masculinity in their cultural milieu (Kiesling, 2004). In the case of Murrinhpatha auth, we will see how a culturally specific stance can shape the use of engagement grammar.

Totemic references
In Murrinhpatha and other Aboriginal languages there is a narrative genre in which a speaker describes their totemic spirit beings, and in this genre auth is used extensively. Murrinhpatha people are spiritually connected to tracts of country over which they have ownership, and to spirit forms of the animals and plants that reside there (Falkenberg, 1962;Stanner, 1936). As mentioned above ( §2), kin connections establish knowledge of the landscape and its totemic as epistemic rights, though lived experience can also play a role (Keen, 1994;Myers, 1986). Thus Murrinhpatha totemic narratives exhibit a form of socially normative epistemic authority. Extensive reproduction of such narratives would be inappropriate in this article, but in any case this is not required, as the use of auth in totemic narratives does not depend on preceding discourse context. The auth verbs are used for most references (either first-mention or subsequent) to places and totemic beings (34, 35).
(36) 1. K____ karrim nanthi thay kala place.name stand.3sg.auth thing tree bloodwood 'The bloodwood tree is there at K____.' 2. Nangkal-ngatha kurru da pangu=ka someone-cond go.3sg.irr place dist=cst thigalan=wa=kathu ma-wuy-kurru cyclone=unxp=from do.3sg.irr-exit-go.irr 'If anyone goes there, a cyclone will appear.' […] 3. Bem-ke-dim beremathangu pan-pekarl-dim afct.rr.3sg.nfut -avoid-sit.nfut always slash.3sg.nfut-spin-sit.nfut 'It should be avoided, (that wind) will go round and round.' (2008-09-25_ThNa_Thithay) The frequent use of auth in totemic narratives is partly explained by the prominence of existential and locative constructions in this genre, as in (34, 35) above. However the frequency in totemic existentials goes beyond that of other speech types, becoming almost categorical. The effect may be incremented when addressing an out-group researcher (as in the examples above), as this produces a particularly great asymmetry of sacred knowledge. But use of auth in referring to totems is not solely existential, and is not limited to out-group addressees. In example (37), classificatory sisters AN and CP are looking at the sun. AN uses auth in a locative construction (line 2), and CP again uses it observing that the sun is setting (line 3). CP refers to the sun as 'our mother' because it is a totem for the two women's matrilineal clan. In this instance, speaker and addressee have equal perceptual and attentional access to the setting sun, but the auth form is used as part of the conventionalised usage for totemic references.

Sociolinguistic stance and 'heavy' speech
Socially normative epistemic authority in Murrinhpatha is associated with sacred knowledge, but also has a lot to do with the social role of the speaker. Speakers may take up an authoritative stance in any discourse situation, if they feel that they have the appropriate social status. Differential social rank may be founded in greater age than the addressee, or a senior kinship relation to the addressee. Examples of this authoritative stance can be seen in Murrinhpatha recordings of the Family Problems Picture Task (Barth & Evans, 2017;San Roque et al., 2012), where pairs of speakers collaboratively describe a series of sixteen pictures, then devise a sequence for the pictures to form a coherent narrative. Some Murrinhpatha recordings of this task involve senior/junior kin pairings, and in these recordings we find the senior partner repeatedly using auth, while the junior partner uses it little or not at all. There is no concrete asymmetry in epistemic access here, as both speakers are looking at the same pictures. Neither is the use of auth restricted to existential or locative functions. For example, NT is the grandson of LT, and with his better eyesight, he initiates description of most of the pictures presented. He does so predominantly using the epistemically unmarked nfut form, while LT responds with repetitions or further commentary, making far more frequent use of the auth form (38, 39).7 (38) 1. NT: Kura dim-tha-gurdurduk liquid sit.3sg.nfut-dual-drink.plrct 'The two of them are sitting drinking.' 2. LT: Kardu kay=yu … kibim-ka-gurdurduk-nume awu person prox=cls … stand.3pl.auth-pc.subj-drink.plrct-pc.m no 'These people … they're drinking, oh no!' (39) 1. NT: Ngarra kardu nimi dim-gurdurduk=ka kuy=nukun rel person one sit.3sg.nfut-drink.plrct=cst fight=adverse nimi kardu patha ne another person good right? 'When this one drinks he tries to start a fight, but this other one is a peaceful man, eh?' 2. LT: Yuwu yuwu kardu kem-nintha pana-nimin pepe yes yes person sit.3sg.auth-du.m recog-contr below 'Yes yes, the two men sitting below.' Auth thus contributes to the linguistic construction of a senior, authoritative sociolinguistic stance. This stance is indexed by a recognised speech style, murrinh yitthit 'heavy speech', which is contrasted with murrinh parnturtparn 'light speech'. Murrinh yitthit also involves the use of more archaic vocabulary, which may not be fully understood by all addressees (Mansfield, 2014, p. 129ff.;Walsh, 1976, p. 5). One speaker explains the yitthit style as being 'like the Queen's English', while another describes it as language da nginalmin-kathu, 'from the roots'. Murrinh yitthit is associated with age, place and clan heritage. As mentioned above, the founding of a Catholic mission on Murrinhpatha land led to community-wide language shift among other groups in the area from the 1940s onwards. Thus contemporary Murrinhpatha has around 3000 speakers, but not all of them have Murrinhpatha patrilineal heritage, which confers the right to speak murrinh yitthit. Similar speech styles (sometimes labelled 'big' or 'deep' speech) are recognised in many Australian Aboriginal communities, including Gooniyandi (McGregor, 1990, p. 8), Ndjébbana (McKay, 2000, Nyikina (Stokes, 1982, p. 1) and Ngan'gi (Reid & McTaggart, 2008, p. 321). One of the more carefully documented cases is in the Wik dialect cluster, where 'big' speech is an esoteric style marking seniority, and considered to be beyond the comprehension of young people (Sutton, 1978, pp. 184, 223).

Quantifying auth usage by corpus genre
I observed in the previous section that the propensity to use auth varies substantially according to speaker, topic and speech types. The Murrinhpatha Morpho-Corpus provides further insight on this variation, thanks to its coding of texts for age, genre and subgenre. The 'genre' tag is used to make a coarse division of texts into Conversation, Elicitation and Narrative types. Within the latter two genres there is further 'subgenre' coding, as shown in Table 3. The table illustrates the percentage of eligible verbs in which the auth form is used, given that the marker is used only on 3.nfut forms, and only shows a substantial frequency with the posture verb stems. There are a total of 1414 stem tokens of this type, which is a large enough sample to give a general idea of the trends, though statistical analysis would require taking into account speaker identity, age and the precise mix of verb stems used.8 Conversation and Narrative appear to have similar rates of auth usage, in their broad classification. However the finer-grained subgenre coding of Narrative types reveals substantial variation. Totemic narratives have the most frequent auth usage of all types, reflecting socially normative authority as discussed above. History narratives include a mixture of town history, personal history and some descriptions of cultural traditions, and these also have a fairly high rate of auth usage. Translation narratives are Murrinhpatha translations of English-language sources (the Bible and some education texts). These have a far lower rate of auth usage, reflecting the fact that the speakers have no authority over these texts, either from personal experience or epistemic rights. All Narrative texts are monologues with researchers as the immediate audience, so audience differences are unlikely to account for auth frequency differences among subgenres. Auth is used less frequently in the Elicitation genre, presumably because consultants producing elicited phrases do not vouch for the content of the elicited phrase, but only for its grammaticality. There is less reason for a speaker in elicitation to encode epistemic authority over a state of affairs, when the propositional content is provided by the linguistic researcher. However, within the Elicitation genre, it is evident that Video stimulus produces a significantly higher rate or auth usage. This is somewhat surprising, given that Video stimulus narratives, like other Elicitation texts, describe content provided by the linguistic researcher, rather than knowledge that is in the speaker's domain of authority. However there are some possible explanations for this discrepancy. The Video stimulus subgenre comprises a series of eight recordings in which speakers describe events unfolding in classic silent films by Charlie Chaplin, Mr Bean, and Laurel and Hardy (Mansfield, 2014, pp. 18-21). These provide fast-moving, complex visual material in comparison to the other elicitation subgenres. It is thus the Video material in which the speaker has more scope to assume that the addressee (either an imagined addressee, or the researcher) has not noticed certain elements of the action, and needs to have them pointed out.
It is possible to further analyse the Video subgenre by speaker seniority. The corpus contains eight runs of this elicitation task, four with speakers aged 60 or older (two male, two female), the other four with speakers aged 18-22 (all male). Dividing the text in this way shows that it is the senior speakers who provide the vast majority of the auth tokens, and that the split between senior and junior speakers is quite consistent. As Figure 1 illustrates, the senior speakers use auth in 35-50% of instances, while young speakers use it in 5-10% of instances. One of the 'senior' video elicitation sessions represented above further reinforces the connection between frequent use of auth and the murrinh yitthit speech style. MM (43% auth) is an elderly female speaker of Murrinhpatha clan heritage, who had been introduced to me as one who is able to (has the right to) speak murrinh yitthit. Before making the recording, she confirmed that she would be speaking murrinh yitthit, though we did not discuss any specific linguistic forms. Apart from having a high percentage of auth usage, MM uses the inflection in a wider variety of verb stems than other speakers. As we have seen, auth usage is concentrated in postural verb stems ( §3). But in MM's characterisation of murrinh yitthit, she uses auth on nine different verb stems, including non-posturals such as kam-'pierce', kirran-'watch' and kurdan-'turn' for which the auth form is rarely attested (40, 41, see Table 2 above). There is nothing in the discourse context of these examples that suggests attentional coordination; rather, they appear to be purely stylistic markers.
(40) kirran-nintha-bath-kanam kam-nintha-ngkarda-kanam watch.3sg.auth-du.m-look-be.nfut pierce.3sg.auth-du.m-point-be.nfut 'The two men are looking around, they are pointing at something.' (41) dhi=wangu kurdan-nintha-ngkarl-tharra dist=twd turn.3sg.nfut-du.m-return-move pangu-da=ngu kangkarl=nu dist-loc=twd above=dat 'They are taking it back up there to the top' While different rates of auth usage among older and younger speakers can be clearly demonstrated, it remains unclear to what extent this reflects age-grading, as opposed to change-in-progress. The fact that senior speakers like MM use auth on a wider range of verb stems than most younger speakers may suggest that its distribution is becoming progressively constrained, which might also reduce its overall frequency. On the other hand, auth is clearly being acquired by younger speakers, at least for postural stems. This suggests that although they use auth at a lower frequency, younger speakers have the grammatical competence that would enable them to use it more frequently with advancing age, especially if their social position enables them to take on the heavy speech style. Further sociolinguistic research will be required to test these suppositions.

Comparison with Jaminjung engagement clitics
Jaminjung is a neighbouring though unrelated language that has been shown to have engagement marking (Schultze-Berndt, 2017). Jaminjung verbs may optionally host an engagement clitic, either =ngarndi ego encoding speaker authority, or =mirndi ego+tu encoding epistemic equality. As in Murrinhpatha, verbs may also simply lack any engagement marking. Thus the two systems are similar in their optionality, though Jaminjung is more elaborated in that it encodes two types of marked engagement, asymmetric and symmetric. The Murrinhpatha k-alternation is bound up in a system of irregular verb inflection, while the Jaminjung markers are loosely bound, and do not depend upon particular grammatical features in the event structure.
There is considerable overlap in the speech situations that warrant Jaminjung ego and Murrinhpatha auth. For example, the Jaminjung marker may be used when an assertion contradicts the expectations of the addressee. In (42), Jaminjung ego is used because the addressee believes the car to be functional, but the speaker asserts that it is defective. We have already seen that such counter-expectation assertions may invoke Murrinhpatha auth ( §3.1), and a further example, in this instance describing unexpected and undesirable behaviour, highlights the similarity with the Jaminjung usage (43).
While the examples above show similar discourse contexts for Murrinhpatha auth and Jaminjung ego, we have noted that they have very different morphosyntactic distributions. The Jaminjung engagement clitics are independent of subject and tense on the verb, and thus can occur with verbs over a wide range of participant, temporal and modal features. The Murrinhpatha k-alternation occurs only with third-person subject in non-future tense. Thus several of the use cases for Jaminjung ego require other encoding strategies to mark speaker authority in Murrinhpatha. For example, Jaminjung ego is used for a future declarative when the speaker is imparting knowledge to the researcher about a local foodsource (46). In a comparable Murrinhpatha example, where a quoted speaker makes a future prediction based on superior knowledge, we find the =wa clitic, which has elsewhere been interpreted as a marker of surprisal or emphasis (47) (Blythe, 2009, p. 107;Mansfield, 2019).

Discussion and conclusions
We are accustomed to inflectional alternations encoding obligatory grammatical features, while clitics or particles encode meanings that are more contingent on discourse context. The Murrinhpatha k-alternation is a striking exception to this, where an irregular inflectional alternation in the verb's inner layer is not associated with event encoding, but rather with a cluster of pragmatic and sociolinguistic functions. The functions of k-auth all relate to speaker's epistemic primacy over the addressee, thus making it relevant to the typology of engagement. But its usage is more contingent than a canonical engagement marker. Whereas Andoke speakers must obligatorily select from a set of engagement values to produce a grammatical utterance (Landaburu, 2007), Murrinhpatha speakers use the k-auth only in particular contexts, and otherwise leave engagement parameters unmarked in the verb.
It is likely that the Murrinhpatha k-alternation did not originally grammaticalise with an engagement function, but rather evolved by reanalysis of a marker fusing third-person subject with TAM features. A *ka-prefix marking 3sg person, and perhaps also non-past tense, has been reconstructed for several non-Pama-Nyungan language families (Harvey, 2003, p. 499).9 Its reflexes are present, with various TAM values, in most or all of Murrinhpatha's neighbouring languages, e.g. Jaminjung ga-3sg.r (Schultze-Berndt, 2000, p. 86), Marri Ngarr k-3sg.irr and 3sg.r.ipfv (Tryon, 1974, p. 126), Marrithiyel g-3sg.r and 3sg.irr (Green, 1989, p. 76), Merranunggu k-3sg.fut (Tryon, 1970, p. 18). Murrinhpatha has k-in 3.nfut.auth as well as 3.irr, and this is largely shared with its Southern Daly sister language Ngan'gi, where g-has an engagement function in 3.prs, and also appears in 2.irr and 3.irr with prohibitive and adversative functions (Reid, 1990, pp. 112-114). One possible scenario is that *k-was the 3sg marker for a distinct TAM series, but in Proto Southern Daly this TAM series became obsolete and left only its most frequent member behind, i.e. the 3sg. With the erstwhile TAM category now obsolete, *k-could have taken on an engagement function, arbitrarily restricted to 3sg verb stems. Whatever the actual scenario may have been, Murrinhpatha auth seems likely to have evolved by reanalysis from a person/TAM prefix of great time depth. This has set up the unusual situation of having a highly contingent, 'optional' marker encoded by an irregular verb inflection.
Within its auth function, Murrinhpatha k-has developed conventionalised subpatterns. In particular, it has become particularly frequent with existential and locative predicates. Because these predicates are formed with postural verb stems, auth is much more frequent on these than other verb stems, and some speakers may only use it with a subset of verb stems. It has also developed a conventionalised usage in the semantic domain of totemic reference, as well as a sociolinguistic indexicality for tradition and heritage, as part of the 'heavy' speech style. It is not unknown for inflectional categories to develop domain and sociolinguistic indexicalities in this way -for example, in Australian English the present perfect tense has a conventionalised usage in police media reports (Ritz, 2010). In neighbouring Jaminjung, =ngarndi ego has a similar function in claiming epistemic authority, though it is not known to have any special associations with existential or locatives, particular verb stems, or domains of discourse. But despite their differences of form and distribution, Murrinhpatha and Jaminjung markers share the property of optionality, being used in only a minority of utterances, while for the most part engagement has no dedicated grammatical marker. I propose that this type of engagement marking is precisely what we should expect, given what we know about Aboriginal cultural epistemology.
Two contrasting features of Aboriginal cultural epistemology were introduced above: the circumspect, consensual style of everyday conversation; and the careful monitoring of epistemic rights in the sensitive domains of kinship, country and spirituality. This implies that epistemic symmetry is the default, and need not be marked. But where epistemic authority is claimed, this is a marked stance that might attract dedicated linguistic encoding. The consequence of this is not an obligatory engagement paradigm as in Andoke, but a single, optional marker of epistemic authority, as in Murrinhpatha and Jaminjung. However, this proposal leads to a wider question about Australian languages: if the cultural epistemology described above is a widespread, continental feature, shouldn't we expect to also find a large number of optional epistemic authority markers across the continent? The literature does mention some epistemic markers, such as Guugu Yimidhirr particles encoding whether a statement is presumed to agree or disagree with the addressees' expectations (Haviland, 1987), Warlpiri 'propositional particles' expressing various epistemic stances on the part of the speaker (Laughren, 1982), or Kuuk Thaayorre demonstratives encoding a spkr. prox / addr.prox contrast that can involve epistemic accessibility (Gaby, 2017, p. 185). But epistemic grammar is not a major theme in Australian grammatical descriptions. The solution to this paradox may well lie in descriptive practices. If optional engagement markers are usually clitics and particles, then this puts them in a domain that is often attracts less descriptive and analytical attention. For example, Murrinhpatha has about a dozen clitics that have only the vaguest descriptions (Street, 1996, p. 214ff.). Indeed, the only reason that k-auth has attracted analytical attention is its inflectional form, which impinges on the analysis of verbal paradigms. But with the recent flourishing of epistemic analysis, we may hope that related phenomena will be uncovered in other Australian Aboriginal languages.