Couples living with dementia managing conflicting knowledge claims

This conversation analytic study investigates how couples manage conflicting knowledge claims when one of the persons has dementia (PWD). The data are video-recordings of 16 couples talking with a third party. The analysis focuses on the negotiation of epistemic rights, more precisely how partners initiate repair and correct claims made by the PWD on matters belonging to the latter’s epistemic domain. We identified three main practices for correcting the PWD: (1) correcting the statement, thereby claiming epistemic authority for oneself and denying it to the PWD, (2) inviting the PWD to self-correct, thereby attributing some epistemic authority to the PWD, and (3) disagreeing and providing reasons for one’s alternative claim, establishing a more symmetric epistemic gradient. The PWDs responses to the corrections displayed different degrees of acceptance, ranging from self-denigration to resistance and insistence.


Introduction
Dementia is a term used for a number of progressive diseases that affect the brain and result in a range of cognitive symptoms. Memory problems constitute a central symptom, and a dementia disease usually affects both episodic and semantic memory, as well as working memory (Marcusson et al., 2011;Morris and Becker, 2004). Due to such memory problems, it is often difficult to predict what kind of information a person with dementia (PWD) has access to at a specific moment, as it may also vary from one point in time to another. This can lead to various interactional problems, for instance related to lack of knowledge (Nilsson, 2017), or incorrect knowledge claims (Lindholm, 2015), which interlocutors have to deal with, as the problems unfold turn-by-turn. Thus, dementia interaction unavoidably confronts the persons involved with the (difficult) task of managing the consequences of memory loss, as realized in their everyday life.
The current paper investigates how couples, having had long, intimate relations, deal with potentially incorrect knowledge claims, made by a PWD, about matters within his/ her epistemic domain. Being a partner of a PWD engenders an interactional dilemma of deciding when to interfere and correct the other, thereby making the 'error' public and halt the progressivity of the talk, and if so, how to do it without reducing the other's selfesteem and epistemic authority as a knowledgeable and competent speaker. This may be especially salient in interactions in which PWDs and their partners talk to a third party, where the interlocutors need to manage the opposing concerns of, on the one hand, getting information across, and on the other, managing the sensitivity of correction.
The aim of this study is to explore how couples manage conflicting knowledge claims when one of the spouses has dementia, focusing on the negotiation of epistemic rights. More specifically, we examine the ways in which partners 1 initiate repair on knowledge claims made by the PWDs, belonging to the latter's epistemic domain. Moreover, we explore how the spouses claim rights to knowledge and assign such rights to each other in these repair sequences.

Knowledge in interaction
In relation to how knowledge claims feature in social interaction, Pomerantz (1980) describes two types of 'knowables' constituting different domains: type 1 knowables and type 2 knowables. Type 1 knowables refer to knowledge gained from first-hand experience including for example, thoughts, feelings, beliefs, opinions, experiences and motives, whereas type 2 knowables refer to knowledge acquired indirectly, for example information someone has 'been told' or has 'figured out' (Pomerantz, 1980: 188). For information belonging to type 1, a person is considered to have primary rights, access and responsibility for remembering, something that is reflected in how speakers design their knowledge claims and interlocutors respond to them (Heritage, 2012;Pomerantz, 1980;Stivers et al., 2011). The violation of this order has been described as 'epistemic trespassing' where one person enters another person's epistemic domain. However, in close relationships the trespassing may be less of a violation than in distant relations (Bristol and Rossano, 2020). Also, as people with dementia commonly have trouble remembering specific events and information (Marcusson et al., 2011), their role as 'experts' on their own lives and their authority over this knowledge (their epistemic authority; Heritage and Raymond, 2005) may be challenged or lead to PWD being treated as 'less than full member' (Shakespeare, 1998).
Central to studies of knowledge in interaction is the distinction between epistemic status and epistemic stance. Epistemic status refers to issues of 'what is known, how is it known (with what method, with what degree of definiteness, certainty, recency, etc.) and persons' rights, responsibilities and obligations to know it' (Heritage, 2013: 377). Participants recognize each other as more or less knowable in relation to a current domain of knowledge, that is, speakers would regularly be viewed as having high epistemic status on type 1 knowables. As a consequence of memory problems associated with a dementia disease, the epistemic status of a person living with dementia can be both uncertain and fluctuating (Lindholm and Stevanovic, 2020). Epistemic stance is used to describe how the participants position themselves vis-à-vis some specific piece of information. In-and-through the design of turns, participants can position themselves as more or less knowledgeable in relation to a domain of knowledge, and they can also display their stance in terms of the degree of certainty of their knowledge (Heritage, 2012;Heritage and Raymond, 2005). It is important to note that participants might take an epistemic stance toward a specific domain of knowledge to appear more or less knowing than they in fact are (Goodwin, 1979;Heritage and Raymond, 2005). Claiming forgetfulness or taking a position as being less knowledgeable (i.e. taking an epistemic stance that differs from one's epistemic status) are practices that may, for example, be used to elicit talk and encourage participation by someone else in the conversation (Goodwin, 1987).

Conflicting knowledge claims
A knowledge claim may be treated by the interlocutor as potentially incorrect and thus as an acceptability problem in need of repair. Problems of acceptability involve a participant having a problem with accepting the truthfulness, appropriateness or sincerity of an utterance by another, and are consequently oriented to as sensitive (Svennevig, 2008). One reason for this sensitivity may be that the problem may not just be an unintentional 'error' that may be corrected, but a more severe problem of differences in the assessment of the acceptability of the utterance. Thus, the initial treatment of a knowledge claim as a repairable may be contested by the producer of the initial claim and consequently, the repair sequence may turn into a sequence of disagreement, a dispute (Kangasharju, 2009;Kotthoff, 1993). This is yet another sequence type involving dispreference and social sensitivity (Sacks, 1987).
Problems of acceptability are manifested in conversation as either other-initiation of self-repair or other-initiated other-repair (or other-correction). The first is the preferred option, in that the repair-initiator merely indicates the presence of a problem, leaving it to the producer of the trouble source to correct it. The second option is dispreferred as the repair-initiator corrects the 'error' immediately, without giving the interlocutor the chance to do it. The dispreferred character is evidenced by the fact that they are often mitigated by question intonation, thereby seeking confirmation from the interlocutor (Schegloff et al., 1977). However, other-corrections may also be performed more bluntly, exposing the error of the interlocutor and suggesting a substitution (Haakana and Kurhila, 2009;Jefferson, 1987). When the problem is the polarity of a claim, they may consist in a simple negation. However, more commonly, they take the form of a negation plus the substitution of the problematic item with a candidate correction. Finally, the correction may be performed without any initial negation. This practice minimizes the intrusion into the talk in progress and treats the error as merely an accidental 'slip of the tongue' (Haakana and Kurhila, 2009).
Other-corrections are claimed to be rather uncommon in everyday interaction among peers (Schegloff et al., 1977). They seem more common (and acceptable) in certain contexts involving clear asymmetries of knowledge, such as teacher-student interaction (Macbeth, 2004;McHoul, 1990), conversations between children and adults (Drew, 1981) first and second language speakers (Kurhila, 2001) as well as in conversations involving persons with disabilities (Williams and Porter, 2015;Antaki and Chinn, 2019). Of special relevance to the current study, Haakana and Kurhila (2009) found that othercorrections occur regularly and without modulation in conversations between parties in close relationships, such as spouses and friends, and suggest that such unmitigated othercorrection may be an index of such a relationship. Moreover, they also found such corrections more frequently in multiparty interaction, in order to 'get it right' for the third party (Haakana and Kurhila, 2009: 175).
Other-correction and disagreement are both dispreferred actions, but they differ with respect to epistemic stance. In other-correction, the party carrying out the correction claims epistemic authority and positions the other as less knowledgeable. If the correction is accepted, the epistemic asymmetry is also confirmed. In disagreement about factual statements, by contrast, both parties claim epistemic primacy vis-à-vis the other.

Knowledge in dementia interaction
More recently, a few studies have investigated how interlocutors manage rights and obligations to knowledge in dementia interaction where either a lack of knowledge or conflicting knowledge claims emerge. Focusing on aspects related to lack of knowledge, Hamilton (2019) describes strategies used by PWDs to overcome memory problems when asked about personal facts. One strategy is to use one's episodic memory to try to work out the answer, another is to seek help from the interlocutor, and a third is to seek help from objects in the visible surroundings. Especially relevant for the current study, Hamilton (2019: 75) notes that in conversations with a third party, partners will sometimes contradict the PWD and speak on their behalf, thereby intruding on the PWDs' 'information preserve' (see also Williams et al., 2020). Also Nilsson et al. (2018) have argued that it is common in interactions involving couples living with dementia that the PWD relies on and invite their partner to speak on their behalf, possibly prioritizing progressivity over epistemic rights. The social sensitivity of not knowing has also been explored. Svennevig and Landmark (2019) found that PWDs and their interlocutors oriented to not remembering personal experiences as accountable when asked factual questions by their interlocutor. In their study, the PWDs, or their interlocutors, regularly provided accounts that either normalized the lack of knowledge, or, on the contrary, 'exceptionalized' it by attributing it to the dementia disease. A third type of account was justifying the lack of an answer by claiming that it is not important, interesting or the like. The accounts thereby reduced the social sensitivity of failing to provide the requested knowledge. Nilsson (2017) showed that for couples, different expectations to what the partner with dementia ought to remember (i.e. have epistemic access to) lead to prolonged sequences and communicative problems. When the partners without dementia offered clues or hints, thus fishing for an answer, it did not support the recollection of memories but rather exposed the lack of knowledge, thereby causing embarrassment for the PWD.
Turning to the issue of potentially incorrect knowledge claims, interlocutors have the option of not treating them as problematic and merely 'let them pass' (Jefferson, 2007). This is reported to occur frequently in conversations with PWDs (Lindholm, 2015). Lindholm examined how interlocutors manage confabulations, defined as 'incorrect beliefs presented by a person who is unaware of his/her incorrectness' (p. 178), in interaction between a PWD and his formal carers. In Lindholm's study, noncommittal responses (e.g. minimal responses) were the most common types of responses to confabulations, avoiding confirming or disconfirming the confabulation. Although rare, acquiescent questions and confirming responses, going along with the content of the confabulation were also found, thereby contributing to developing the confabulation. On the other end of the continuum, challenges to the confabulations were also rare, and when they occurred, they were carried out cautiously and indirectly. In a case study from the context of family interaction, Black (2011) described how a daughter challenges her mother's statements and takes an epistemic stance as more knowledgeable in relation to domains of knowledge belonging primarily to the mother with dementia. Black argues that although the daughter in this way threatens the mother's position as an equal participant in the conversation with the capability of knowing about her own life, the mother manages to participate in the conversation 'without overt loss of face in the form of allowing herself to be corrected or contradicted about her own ability, experience, or knowledge' (Black, 2011: 86). Also Hydén and Samuelsson (2019), in their case study of a conversation between a PWD and her two relatives, argue that both the relatives and the PWD work to avoid face threats and confrontation when disagreeing not only on factual matters, but also on more foundational questions about such things as who could possibly be alive at a certain point in time; what the authors call 'reality disjunctions'. Finally, Lindholm and Stevanovic (2020) describe how conversational partners make the dementia diagnosis of their interlocutor relevant by exhibiting a lack of trust in the existence of common ground, for instance by not acknowledging 'the speaker's epistemic status in the matter at hand' (p. 10). Following this line of research, the current study seeks to further our understanding of how epistemic rights are negotiated by couples living with dementia, by looking at instances where the partners initiate repair on a knowledge claim made by their partner with dementia. The relation between a PWD and their partner may involve both a considerable epistemic asymmetry and a high degree of intimacy, but to date corrections have not been studied systematically in such interactions.

Data and method
The data for this study consist of video-recordings of PWDs and their partners talking with a third party, either in research interviews, or in informal conversations with care providers, family and friends. The data are drawn from the Swedish CEDER-corpus (Center for Dementia Research) and the Norwegian MultiLing Dementia-corpus. The CEDER-corpus consists of video-recorded research interviews with 13 couples, in which one of the spouses is diagnosed with dementia. The interviews focused on the couples' past lives together, as well as their current experience of living with dementia. The MultiLing Dementia-corpus consists of video-recordings from various speech contexts involving seven multilingual speakers diagnosed with dementia. For this study, we have included naturally occurring conversations and research interviews in which their partner was also present, which amounts to three of the participants. Taken together, the study draws on approximately 12 hours of video-recorded multi-party interactions involving 16 couples living with dementia. In the CEDER-corpus, the third-party participants were researchers who visited a geriatric clinic for the interview, and for the MultiLing Dementiacorpus they were either friends, health care providers or researchers.
All participants included in this study had mild to moderate dementia. At the time of recording they all lived at home with some support from informal caregivers and/or social or health care services. The recordings were collected with the informed consent of the participants, and the names of persons and places are anonymized. The data are in Swedish, Norwegian and American English. The extracts are supplied with English translations where applicable.
The method used in this study is conversation analysis (Sidnell and Stivers, 2012). The recordings have been transcribed following the conventions developed by Jefferson (2004a) (see Appendix). By taking an emic perspective, the analysts seek to describe the participants' orientations to the knowledge claims being made, grounded in the 'reality' of the participants (Schegloff, 1996: 172;Seedhouse, 2005). For this study, we collected 84 instances in which a partner initiated repair on a knowledge claim produced by their partner diagnosed with dementia. Through detailed sequential analysis, we distinguished three different practices used by partners for correcting the PWD depending on the type of repair organization being used. The interactional characteristics and consequences of these practices in terms of epistemic rights will be presented below.

Analysis
This section presents three practices spouses used for initiating repair of knowledge claims made by the PWD: corrections, invitations to self-correct and explicit disagreement. They differ in terms of epistemic rights allocated to the PWD. The most common practice found was corrections (45 instances), where the partners take on superior epistemic rights. In invitations to self-correct (29 instances), more epistemic authority is allocated to the PWD, whereas in explicit disagreement (10 instances), both parties claim superior epistemic rights to the factual claims made.

Corrections
Other-corrections of knowledge claims made by the PWD (and within the PWD's epistemic domain) appeared in both data sets. These sequences were characterized by mitigating actions, reducing the sensitivity of the correction in some way. However, the first part of this section illustrates a less frequent, albeit not unique practice, in which corrections were carried out without involving the person being corrected.

Concealed corrections
This extract is from a research interview with Elvira (E) (diagnosed with dementia) and her husband Lennie (L). Elvira has been telling about the activities at the day care center that she regularly attends, with some assistance from her husband. In terms of epistemic status, we may in principle consider the wife as having primary epistemic access and rights to the information concerned, having direct personal experience with the activities reported (a type 1 knowable), while the husband has only secondhand access to it (type 2 knowable). In lines 1 and 2, Elvira's telling about helping out at the day care center is modified by the epistemic hedge väl, which marks the reported event as merely likely and not certain. The report gives rise to a follow-up question, in which the interviewer asks about one type of thing they might be helping out with. Elvira answers the question with a clear 'ja/yes' without any hedges or hesitation, thus taking a knowing stance. Immediately, Lennie contradicts Elvira's response by shaking his head while gazing at the interviewers, taking a similarly strong epistemic stance. The couple is seated side by side, so the head shake is clearly addressed to the interviewers and not the wife, although the action may be in her visual space. Neither the interviewers nor the wife shows any uptake of the head shake, and a long silence ensues. Contradicting someone with epistemic primacy is an accountable action and would normally engender an account for one's opposing view (Sacks, 1987). However, no such thing occurs here. This noticeable absence may indicate that the partner treats his epistemic status as trumping the wife's epistemic status, despite the claim in question being within her epistemic domain. Following Lindholm and Stevanovic (2020), this may be a case of 'indirect evidence that the dementia condition is relevant -not for all participants in the interaction -but for the co-participants interacting with the person with dementia' (p. 15). As such, their respective roles in the interview, as a partner without dementia versus the partner with dementia, may be made relevant at this point in the interaction. By contradicting his wife's claim silently to the interviewers, the husband excludes the wife from the participation framework, thereby denying her the opportunity to react to his correction. The interviewers collude in side-stepping the wife by not responding to the contradictory answer. Neither do they follow up the topic of cooking, which would have been an expectable thing to do, given that they have asked a topic initiating question, projecting more talk on the matter than just a minimal confirmation (Schegloff, 2007). Thereby, they seem to accept the husband's, and not the wife's, account. Instead, they leave a long silence, giving the wife the opportunity, but no invitation, to revise or modify her answer. Subsequently, they resume the interview by asking a new question related to other aspects of the day care center, thereby changing the topic.
In the online negotiation of epistemic authority, the husband's contradicting response has different status for the two parties. For the wife, it contributes to respecting her epistemic rights and authority by not making the correction available to her. For the interviewers, however, it denies her not only epistemic access to the event, but also the opportunity to inspect her recollection and modify her account in a repair sequence. He thus seems to orient to two opposing considerations and to 'hit two birds with one stone'. By correcting his wife's 'wrong' answer, he orients to 'getting it right' by providing the interviewers with correct information (Haakana and Kurhila, 2009), and by concealing the correction to his wife, he orients to the social sensitivity of exposing her memory problems to her. The concealed correction also avoids opening an extended repair sequence, disrupting the progressivity of the talk.

Mitigated corrections
A more common practice in the data was to involve the PWDs in the repair sequence. By addressing them with second person singular pronoun, they were given the opportunity to accept or reject the proposed correction. Moreover, the corrections were generally accompanied by mitigation and some type of support for the claims made by the PWDs.
Extract 2 below involves the same couple as in extract 1, and is taken from a telling of how the couple first met over 60 years ago. This knowledge may be characterized as a 'type 1 knowable' for both spouses, and thus give them equal epistemic rights (Heritage, 2013). After a lengthy account by the partner without dementia, the interviewer turns to the PWD, Elvira (E), and addresses a question to her. The interviewer's question is formulated as a candidate answer (Pomerantz, 1988), a declarative statement of the speaker's assumption, making confirmation from the interlocutor the preferred response. The answer conforms to this preference, and Elvira's confirming 'yes' accompanied by a nod in line 3, takes a knowing stance, without signs of hesitation or mitigation. A 1.6 second pause ensues, after which Lennie corrects her statement (lines 5 and 6). While making this correction, Lennie pats Elvira's arm, and this display of affection may be understood as a mitigation (Goodwin, 2017). Also the fact that he leaves a substantial gap before providing the correction displays an orientation to the sensitivity of the correction by giving her the opportunity to self-correct. Lennie produces his turn with a strong epistemic stance and an unmitigated correction, taking the form of an explicit negation and rejection ('no you didn't'), followed by a correction. As Elvira responds (line 8), she produces a turn that simultaneously agrees with Lennie's statement, and sustains the correctness of her previous claim by framing Lennie's statement as an addition ('också' -too) to what she herself had claimed. Lennie accepts this in line 9, and acknowledges that Elvira is partly right, by adding that she was in fact 'home for some vacation' (lines 13 and 14). By doing so, Lennie mitigates his initial correction and formulates an account that is compatible with both of their claims. He thereby ascribes some epistemic authority to Elvira. Nevertheless, through the othercorrection and the acknowledgment of Elvira's response to it, Lennie casts himself as having superior epistemic rights to this piece of information. Extract (3) shows a similar example, with an explicit correction followed by a mitigating concession. The extract is drawn from a mealtime conversation in the home of Laura (L), a woman with dementia, and her spouse Gary (G). Their friends Berta (B) and Anne (A) are over for lunch. All participants are bilingual and code-switch between English and Norwegian; when speaking in Norwegian this is marked in grey in the excerpt and translated to English. Just before the extract starts, Laura has told a story about her father's numerous siblings, and added that two of them were twins.  Berta here asks a direct question about a specific piece of information related to Laura's telling (line 1). Instead of answering, Laura performs a palm up gesture (line 2), signaling either lack of knowledge or inability to produce an answer (possibly due to having food in her mouth). Berta pursues a response by reformulating the question, this time adding a candidate answer (line 3, cf. Svennevig, 2013). However, the candidate answer is immediately rejected with a 'no' in a preemptive answer to her own question. This rejection of the candidate answer thus expresses her assumptions about the father and thereby makes confirmation the preferred response. This is also what Laura does, by means of a minimal token of agreement ('no', line 5). It is at this point the couple's conflicting knowledge claims comes to the surface, as Gary in overlap produces the opposite response (line 6). Although it is produced almost simultaneously, Gary's response may be considered as reacting to Laura's response in progress and correcting it. The question was asked to Laura, so Gary intrudes into her turn space, thereby claiming the right to correct her utterance-in-progress. After a slight delay, in which Laura does not take the opportunity to revise her answer or otherwise self-repair, he continues with his correction, 'your fa:ther was a (.) twin', taking a strong epistemic stance towards knowledge in Laura's epistemic domain (type 1 knowable). The correction is addressed to Laura by the use of the second person pronoun ('your'), thus presenting the information as a reminder to her rather than addressing the interlocutor (Berta) and speaking on Laura's behalf. He thereby gives her the opportunity to revise and correct her claim.
In response to this correction, Laura, in line 10, points up while gazing up and chewing. Her embodied response, with a 'thinking face' (Goodwin and Goodwin, 1986), may signal that she is not fully convinced, and is searching her memory. By doing so, she does not accept the correction immediately but instead claims some epistemic rights to decide for herself whether she agrees. At this point, Gary continues with an increment (lines 11 and 14) that mitigates his correction: He reduces the distance between their opposing claims by providing the concession that they were not monozygotic. Thereby, he also offers a potential account for her incorrect claim by providing an 'excuse' (Scott and Lyman, 1968), a rational explanation for why she might have answered the way she did. After the repair sequence, the other friend, Anne, again takes up the line of inquiry about the twins (lines 15 and 17), this time with a question about the gender of the twin. At this point, Laura displays full epistemic access to the subject matter (lines 19-21), both by responding to the question with an emphatic confirmation token ('oh ye:ah,') and by providing the sister's name, thereby demonstrating her knowledge of the matter at hand.
The two extracts in this section both involve explicit corrections produced with strong epistemic stance. However, they are subsequently mitigated by concessions that give the partner with dementia partly right, thus mitigating the social sensitivity of the correction and ascribing the spouse some epistemic authority. The corrections are designed to address primarily the PWD, as the partners use the singular second person pronoun 'you' rather than 'he' or 'she'. By doing so, the PWDs gets the opportunity to accept or revise the correction. The PWDs, on their side, also claim some epistemic rights by not immediately and unconditionally accepting the corrections. In this way, both parties contribute to establishing a more balanced epistemic gradient.

Invitations to self-correct
Although other-corrections were a common practice for dealing with problematic knowledge claims in our data, there were also instances of other-initiated self-repair, in which the PWDs were invited to self-correct and thereby maintain epistemic primacy concerning the question at hand. The next extract (4) is from a research interview carried out in Laura's first language, English. Her partner, Gary, is also present. Prompted by the interviewer's (H) question in line (1) The interviewer treats the telling as potentially complete in line 22 by producing a positive assessment, and Laura furthermore contributes to closing the sequence by producing a summative evaluation of the facility in line 24. The interviewer produces yet another positive assessment as a sequence closing third, and thereby marks the sequence as completed. However, Gary reopens the sequence by producing a delayed repair initiator: 'a little bit more than just lunch perhaps?'. Gary's strategy of suggesting an elaboration of Laura's telling may be considered a 'fishing device' (Nilsson, 2017;Pomerantz, 1980) enabling Laura to continue. The repair initiation does not contradict Laura's answer but merely suggests a possible addition to it. Thereby, Gary does not challenge Laura's epistemic access, and neither does he claim the right to provide the missing information. He invites her to provide it and thereby to speak for herself. Furthermore, he downgrades his own epistemic stance by using the epistemic hedge 'perhaps' and he mitigates the problematic character of the trouble source by downgrading the magnitude of the missing information ('a little bit more than just lunch'). In sum, he maintains her epistemic primacy by inviting her to self-correct and by downgrading his own epistemic stance. Nevertheless, in her response, Laura refrains from expanding on the matter, producing only a minimal confirmation token (line 28). It is only after the interviewer's upshot formulation 'yeah.=so you stay there for the day?' (Heritage and Watson, 1979), that Laura provides a more elaborate response. However, she disconfirms the upshot and instead seems to stick to her original account about staying there for lunch (lines 30, 31 and 33). Thus, similar to Nilsson's (2017) findings, the invitation to self-repair did not yield any new information.

Explicit disagreement
This last example is slightly different from the previous ones in the sense that the interlocutors have an explicit and prolonged disagreement about the events reported in which they provide evidence for their own versions. It is from a research interview where the interviewer (LC) asks Inga (I) and Ove (O) for information about when they first discovered the husband's dementia symptoms (line 1). They are addressed collectively by the second person plural pronoun 'ni' (you). The initiating question seeks information that is potentially accessible to both parties, given that it concerns their shared experience as a couple. Ove provides a lengthy reply to the question (not shown), in which he tells about a medical study he participated in, and how after that he was in the system somehow (lines 2-6), indicating that it was by coincidence and not due to manifest symptoms that he got into 'the system'. Before he comes to anything recognizable as an answer to the question (when they started to notice dementia symptoms), his wife Inga interrupts his account with the adversative conjunction 'fast' (however), projecting an objection or a contrasting account (line 7). But subsequently, she works to reduce the disaffiliative potential of this misaligning turn initiation. First, she makes an explicit request to take over the floor (which is granted by Ove in line 9), thereby orienting to the interruptive and delicate character of her talk. Second, she starts by a metacommunicative framing device, presenting her account as her own subjective experience (lines 7 and 8). By designing her turn this way, Inga reduces her epistemic authority and opens up for the possibility that there may be several co-existing versions of the story (much like subject-side assessments, cf. Edwards and Potter, 2017). Finally, in the narrative, she addresses Ove by using a second person pronoun ('du' -you) rather than the interviewers (line 10). Thereby, she provides the occasion for him to take the next turn and to confirm or reject her account. Also, she starts off by stressing their shared initial belief that there was nothing wrong ('there is nothing wrong with me and one doesn't think there is and neither did I at first'). Such 'at first I thought'-formulations usually precede contrasting statements of the actual state-ofaffairs, having an extraordinary status (Jefferson, 2004b). Ove, in overlap (line 13), agrees with Inga's initial belief, either indicating that this is still his belief, or merely displaying alignment-so-far in the narrative. Inga, however, continues with the contrastive marker but', but instead of stating explicitly the realization of dementia symptoms (as asked for in the question), she provides a positive assessment of the dementia team (line 14), only indirectly alluding to a dementia diagnosis. The assessment seems to have a double addressee: Both the interviewers (by gaze direction), informing them, while simultaneously appealing to the spouses' and/or potentially the interviewers' shared knowledge on this (with the common ground marker 'ju' in lines 14 and 15). The 'ju' may possibly also orient to a potential discrepancy or epistemic incongruence between the spouses recall or view on the matter (Heinemann et al., 2011). This way of presenting the event may thus also pursue agreement from the PWD, without making the potential disagreement explicit (Asmuß, 2011). However, it does not succeed, and in the next extract, we can observe the subsequent unfolding of the disagreement: Ove once again objects, not to the veracity of the facts, but to the temporal ordering of the events (lines 16 and 18). And he uses the same common ground marker ('ju'), indicating that his version constitutes shared knowledge. Inga's response is a non-committal continuer (line 19), orienting to his right to the turn, but not responding to the objection. After a long pause (line 20), Ove corroborates his claim by providing an alternative narrative (lines 21-24). Inga counters Ove's account by contesting his correction and returning to her original claim that the contact with the dementia team happened first ('but first. . .', line 25). This time the statement is presented without hedges or other mitigation devices. This version is yet again explicitly rejected by Ove (line 27), after which Inga at the face of it resigns by saying 'nehe ne' (no okay) (line 28). But the minimal form of this concession, without any account for the change of position, seems to indicate that she is not sincerely accepting the counter-argument. This may be interpreted as a signal to the interviewers that she is not accepting his claim, and thereby that she does not ascribe him epistemic authority.
In this disagreement, we see the participants taking more symmetric epistemic stances, both of them claiming epistemic primacy and contesting the factual claims made by the other. The wife starts by taking a weak epistemic stance and by presenting her objections in a dispreferred format, but in the face of unmitigated and strong disagreement, she upgrades her epistemic stance and starts disagreeing in a preferred format, without hesitation, mitigation or accounts. The example thus differs from the other extracts analyzed in that the PWD does not accept the correction by the partner but instead maintains his position with strong epistemic stance. Consequently, the sequence develops into a type of dispute that is much is line with what has been described for typical interactions (Kangasharju, 2009;Kotthoff, 1993). Only when the partner 'gives in' without accounting for her withdrawal from the oppositional stance, she may be considered to treat the PWD as epistemically inferior by merely letting the interviewers judge for themselves whose account is most reliable (Lindholm and Stevanovic, 2020).

Discussion
This paper has explored how persons with dementia and their partners claim epistemic rights to knowledge and assign such rights to each other in situations where a PWD's knowledge claim within their own epistemic domain is challenged by their partner. By correcting or questioning the PWD, the partners claim epistemic authority within the PWD's epistemic domain. Although such other-corrections may occur regularly in close relationships, in our data, it is the partners who initiate and carry out most of the corrections, suggesting an asymmetric distribution of epistemic rights. This is in line with what has been found in other asymmetrical relations (Antaki and Chinn, 2019;Drew, 1981;Kurhila, 2001;Macbeth, 2004;McHoul, 1990;Williams and Porter, 2015). Since the spouses here do not display other forms of asymmetry, the relevant source of their asymmetric rights thus seems to be their status (at least with respect to their roles in these interactions) as cognitively healthy versus challenged.
Our analysis illustrated a range of practices for initiating correction, and their consequences for the epistemic rights allocated to the PWD. At the most asymmetrical end of the epistemic gradient, we showed an example of concealed other-correction (extract 1), only addressing the interviewer and side-stepping the PWD. Thus, the partner denied the PWD both epistemic rights and the opportunity to participate in the repair sequence and self-correct. At the other end of the epistemic gradient, we found a more symmetrical distribution of epistemic rights in cases where the PWDs maintained their claim in disagreement with the other (extract 5). In this case, the interlocutor's version was presented as merely a different subjective experience of the events, ascribing the PWD both the right and the accountability to have a contrasting view. As the PWD in this case took the opportunity to maintain his opposing view, a dispute ensued that had many of the characteristics of typical interactions between peers (Kangasharju, 2009;Kotthoff, 1993). Nevertheless, the dispute ends with the partner backing down, while simultaneously signaling to the interviewers that she does not endorse his claim. Similarly to extract 1, the partner without dementia thus leaves it up to the interviewers to decide which version to believe, possibly relying on her epistemic status trumping the epistemic status of the PWD. This resonates with Lindholm and Stevanovic's (2020) findings, where the interlocutors may 'respond in ways that do not acknowledge the first speakers' statuses' (p. 13).
The most common practices in our data were found in between these two extremes. They consisted in other-corrections addressed to both the PWD and the third party (extracts 2 and 3). As already mentioned, these other-corrections ascribed reduced epistemic rights to the PWD, within his or her epistemic domain. A more symmetrical distribution of epistemic rights was achieved in the cases where the partners invited the PWD to correct or revise their claim, thus giving them the opportunity to self-repair and thereby maintain epistemic authority (extract 4). This is in line with Antaki and Chinn's (2019) findings from medical consultations with patients with intellectual disabilities. The study shows how companions adopted a continuum of practices that to a larger or lesser degree displayed epistemic authority over the disabled person when responding to questions directed to the latter. Thus, the companions could 'use their greater epistemic status to simply over-ride the patient, or (. . .) adopt different epistemic stances to manage the matter with more delicacy' (Antaki andChinn, 2019: 2026). In our study, the partners did substantial interactional work to mitigate both their epistemic authority, and the 'incorrectness' of the PWD's problematic claim. These mitigation strategies worked to reduce the sensitivity of correcting the PWDs and trespassing into their epistemic domain. This was done by using features characteristic of dispreferred turn shapes, such as delays, hedges, accounts, highlighting partial agreements, as well as embodied mitigators. A special type of account was illustrated in extracts 2 and 3, namely accounts of how the PWD might have come to provide an incorrect answer. These accounts rationalize the process of arriving at an incorrect answer, thereby normalizing it rather than imputing it to the memory problems associated with dementia (see also Lindholm, 2015). As such, they resemble the type of accounts that have been described to occur when a PWD is unable to answer a question about personal experiences (Svennevig and Landmark, 2019). Our study also revealed that in response to the repair initiations, the PWDs managed to preserve some epistemic rights, either by confirming the correction (Kristiansen et al., 2017;Mikesell, 2010), by integrating it into their own account (Black, 2011), or by opposing the correction and maintaining their original claim. Our findings thus support previous findings that persons with dementia have ways of maintaining epistemic rights, even when challenged openly (e.g. Black, 2011;Hydén and Samuelsson, 2019).
More broadly, our findings differ from Lindholm (2015), who observed that corrections of problematic factual statements were rare in her data from residential care. There may be at least two possible explanations for this difference: First, the participation framework in our data, involving a couple talking to a third party, differs from the carer-PWD dyads in Lindholm's study. Our findings may thus reflect Haakana and Kurhila's (2009: 175) observation that 'in multiparty interactions, then, other-corrections are made not only to correct the producer of the error, but also for the benefit of the audience, for them to get it right'. This may be especially true for the research interview setting, in which it seems that the partners may be oriented to the institutional goals of conveying correct information, and to 'get it right' (see also Shakespeare, 1998;Williams and Porter, 2015). Second, the type of knowledge claim may also be consequential. It may be easier to correct a minor factual detail in an otherwise acceptable account than to correct a more encompassing confabulation, displaying more obvious reality disjunctions (Hydén and Samuelsson, 2019;Lindholm, 2015). In this regard, it may be noted that the explicit dispute in extract 5 concerns the temporal ordering of the events, which is normally not treated as a very sensitive issue in itself. However, judging from the surrounding talk, it may seem that the dispute covers a more fundamental disagreement between the two about whether or not the husband had (and has) dementia symptoms.
When participants manage conflicting knowledge claims, their relational status is made relevant in various ways. Goffman (1971) argued that participants display to each other and others their relationship through tie-signs (or 'with-markers'), such as holding hands or completing each other's utterances (Goffman, 1971: 194). Within intimate relationships such as a marriage, the participants may benefit from 'minor liberties' in terms of entering or trespassing the other person's personal space (Bristol and Rossano, 2020;Goffman, 1971: 203). In our analysis, we show how partners of persons with dementia occasionally claim epistemic rights to knowledge which mainly regard the person with dementia, something that could be regarded as taking a minor liberty and also marking a 'with' (Goffman, 1971: 43) with the person with dementia. This is also in line with the observation by Haakana and Kurhila (2009) that unmitigated other-correction may be an index of a close relationship. Notably, unlike Haakana and Kurhila (2009), the other-corrections in our data are mostly mitigated in some way, indicating that the partners treat the corrections as sensitive. This sensitivity may be related to the fact that exposing a lack of veracity makes the dementia condition salient (Lindholm and Stevanovic, 2020). In addition, the information being corrected is often within the PWD's privileged epistemic domain (type 1 knowables), and thus the correction may be considered as epistemic trespassing (Antaki and Chinn, 2019;Bristol and Rossano, 2020). In Haakana and Kurhila (2009), the unmitigated (or unmodulated) corrections were carried out by speakers with first-hand knowledge of the matter. The current study may thus broaden our understanding of corrections within another persons' epistemic domain. It remains to be seen whether the sensitivity displayed in the mitigated corrections pertain (mainly) to the dementia condition, the type of knowledge being corrected, and/or the participation framework involving a couple interacting with a third party.
Finally, issues of competence may be relevant for whether corrections are carried out or not (Jefferson, 2007), and for how the corrections are designed. On the one hand, exposed corrections make problematic claims publicly observable and relevant, and slow down the progressivity of the talk. In our case, they also draw attention to a lack of knowledge that (competent) speakers normatively ought to have (Stivers et al., 2011). On the other hand, refraining from initiating repair would, according to Shakespeare (1998: 152), have consequences for the PWD's participation role, in that 'if a repair is not required of a participant who has made a noticeable error, then he or she is not being treated as a full participant'. The double orientation in our data, where partners do initiate repair, while simultaneously mitigating the social sensitivity of it, may be a way in which interlocutors seek to strike a balance between these contradictory concerns. Taking the perspective of the persons with dementia, they may be managing issues of competence in multiparty interactions by orienting to and prioritizing fluency and progressivity in the conversation, with the consequence of potentially sacrificing the precise accuracy of their claims.
This study has illustrated a range of practices interlocutors use in order to correct PWDs' knowledge claims, and pointed to ways for ascribing more epistemic authority to the PWD by inviting them to self-correct. Furthermore, the interlocutors (in our data) orient to the balancing act of 'getting it right' for the third party, while mitigating the sensitivity of making corrections within another person's epistemic domain. As such, it may contribute to awareness by carers and family members about how to deal with problematic knowledge claims by PWDs in ways that reduce the sensitivity of initiating corrections, while still treating them as competent participants in the interaction, epistemically accountable for their factual claims.
Many guidelines for communicating with persons with dementia recommend using polar questions, as they are easily answerable. However, this study points to a potential downside of this type of question: Questions expressing a candidate answer, thereby displaying a specific response preference, may induce persons who are unsure of their memory to 'follow the lead' and comply with the response bias rather than disconfirming the assumptions expressed. As shown in extracts 1-3, the PWDs' 'incorrect' answers constitute confirmations of assumptions expressed in the question. Although all question recipients make use of this kind of information conveyed in questions (Heritage and Clayman, 2011), PWDs may possibly use presuppositions and assumptions expressed in questions as an even more prominent resource when faced with uncertainty about factual knowledge. This possible effect would need further exploration in future studies.