Exploring Self-Defining Memories in Old Age and their Digital Cues

Self-defining memories represent significant emotional events capturing the most important concerns in our lives. While much HCI work on memory technologies has focused on autobiographic memories and lifelogging technologies for capturing them, there has been little exploration of self-defining memories and how they may be supported by appropriate cues. This is important as such memories are key in the development and maintenance of the sense of self, particularly in old age. We report on interviews with 8 older adults in their homes. Findings advance the understanding of self-defining memories and their possible cues with new insights into their relationship with self-identities and cues' specific qualities supporting richer emotional recall. Our findings led to several design implications such as novel technologies for curating self-defining memories and their cues, for embedding layered meaning in such cues across the lifespan, for crafting them, and sensitive design for their physical handling.


INTRODUCTION
Sense of self is crucial for our perception of continuity through time, and invaluable for wellbeing [33]. It captures perceptions and attitudes about oneself, often as a set of identities such as personal, relational or collective self [17,42,45]. Sense of self is actively supported by personal memories [57] and in particular by self-defining ones. The latter represent a specific type of autobiographical memories capturing lasting concerns or unresolved conflicts [4,20,28] around significant events, which help people explain how they have become who they are [8]. When memory weakens because of cognitive decline of healthy aging or dementia, the sense of self also weakens, with findings showing the painful impact of the disintegrating self on people living with dementia and their families [5]. Much HCI research on memory technologies has focused on the value of lifelogging systems for capturing and archiving large amount of digital memories to support total recall [44]. Such emphasis on capture technologies however does not account for the significance that people attach to their memories and their specific cues. In addition, despite the importance of supporting selfdefining memories, and of their much exploration in memory research [4,20,45], we know little about if and how such memories may be cued in healthy ageing and dementia. Relevant HCI work on cueing has focused mostly on autobiographical memories in general [10,14,24,37,38,43,58] albeit with less focused on those representing self-defining memories.
We argue that if the aim is to strengthen the sense of self in old age, then we must move beyond the main emphasis on capture technologies. This paper aims to investigate selfdefining memories in older people, and in particular the possible cues that they identify for recalling them. The idea that people can identify memory cues, before they cue, has been previously argued based on meta-memory judgments [56], and on emerging findings on self-chosen cues supporting recall of self-defining memories of people living with dementia [20]. An understanding of what can cue selfdefining memories, the content, and format of such cues, where they are stored, and how they are retrieved should support the design of technologies to capture, store, and retrieve digital memory across the lifespan, and to strengthen the sense of self in old age. We report on interviews with 8 older adults in their homes about such memories and how they can be cued. Because of the exploratory nature of our study and the novelty of research topic, we selected participants without signs of dementia; it is arguably easier to explore self-defining memories and their potential cues in healthy older people before the onset of dementia. Our study addresses the following questions: • What types of self-defining memories do people identify?
What are their affect, content, and reflected self-identities? • What types of possible cues do people associate with self-defining memories, where do they store them, and what challenges do they face in identifying such cues?
• What qualities do self-defining memory cues have?

RELATED WORK
The study outlined in this paper draws from previous HCI research on memory cues, and memory technologies for aging, and of the growing body of work in psychology of memory and in particular self-defining memories.

Memory Cues in HCI
Most HCI research has focused on technologies for capturing and supporting recall of episodic memories [43,58]. There are two main types of cues supporting this recall namely internal ones consisting of sensorial experiences and feelings, and external cues present in physical environment [56]. The latter may include objects, people, activities and places, and their capture through photos, physical and tangible objects has been particularly explored in interaction design [56].
Much work has looked into the impact on recall of different modalities of memory cues, mostly visual [43] and audio cues [14], although the investigation of other cues such as location-based [24], or crafted cues [37,39] has also started to emerge. Beyond the exploration of cue modalities in isolation, other scholars have looked at combining them. One such example is HP memory box [18] which integrates sound-based narrative with stored memorabilia to support episodic recall. A few studies have explored the enduring impact of cue types on retrieval, such as photos, sounds, videos, odors, or physical objects on one-month old autobiographical memories [55]. Findings indicate that no cues condition led to more memory-details. However, with the exception of physical objects crafted by participants, the rest of the cues were generated by researchers, and hence may not have been typical or unique; characteristics considered key for successful cueing. Other HCI work has explored physical or digital possessions and their value for remembering [30,31], building on the link between possessions and sense of identity, and in particular Belk's [2] theory that possessions are components and determinants of sense of self consisting of: collections, money, pets, other people, and body parts. Later work has identified alternative categorizations of possessions in objects of power, self-continuity, and relationships [11], or, for their role in recall, in mementos (reminders of special events), souvenirs (reminders of special places) or heirlooms (family or inherited objects) [12]. Souvenirs relate mostly to holiday memories, are on display in house's public rooms and used for reminiscing [54], while mementos were categorized in artwork, photos or symbols of achievements often on display in public spaces [34]. Memorabilia (serving specific function and occasionally used), and idiosyncratic deeply personal objects tend to be privately stored [34]. A few studies have focused on how older adults' memory and identity are challenged by moving in later life to residential care or downsized homes. Findings suggest that to address limited space and fewer cherished possessions they can be bring with, people document and engage in tagging or telling stories of lost objects [27]. An emerging body of work has looked into how possessions may be questioned when cueing strong negative emotional events, no longer relevant for the current self [40,41].
To summarize, much work has focused on cueing autobiographical memories to support recall [10,14,24,37,43,58] or explore forgetting [40,41]. Most cues are printed photos or physical objects [56] created by researchers or belonging to participants' possessions. Although few prior studies have touched upon the relationship between possessions and self [27,34,40], there has been however limited explicit exploration of selfdefining memories and how they may be cued.

Memory Technologies for Aging in HCI
A particular strand of memory research in HCI has focused on older adults to support their memory impairment due to aging or dementia [21,43,59]. Memento [60] creates hybrid physical-digital scrapbooks allowing older adults to reminisce. A much used technology is SenseCam, a wearable device which automatically capture photos in daily life. Consistent findings have shown its value in supporting the recall of episodic memories [21], particularly those of personal relevance to people experiencing memory impairment [25]. Other work integrated SenseCam photos with ambient displays for showing both past and recent photos, which has supported the sense of self of people living with Alzheimer's disease [10]. These authors also describe multimedia biographies co-designed as collages of personal photos, videos, and narratives on life stages and themes by people with memory impairment. Findings suggest that this practice supports the reminiscence of positive events, but its value for strengthening the sense of self has been less investigated.
To conclude, most HCI work on memory technologies has focused on curating and capturing life events to support older people's recall and share of their episodic memories. The emphasis here is on digital caption of past or very recent life events, but with a few exceptions [10], their value for sense of self and in particular for remembering self-defining memories has been less explored.

Memory Research on Self-Defining Memories
Autobiographical memories contain crucial information about the self [9,35,36] in the form of personal semantics comprising general knowledge and facts about oneself [28], and episodic memories of personal events located in a specific time and space and re-experienced through mental travel of the self [51]. Self-defining memories share characteristics with both episodic memories such as rich phenomenological details and specific spatio-temporal context, and with personal semantics in the form of more generic memories of personal events clustering in patterns over time [45]. What sets self-defining memories apart is one's enduring concerns reflected in events focused on selfdiscovery, self-understanding and identity [4,20]. Findings have also shown that self-defining memories related to negative emotional events are more challenging to process albeit crucial for one's wellbeing [33]. With respect to content, prior work identified different types of self-defining events such as narratives of threat, disrupted relationships, conflict-free relationships, achievement, failure, exploration, or guilt/shame [4]. Additional studies have also shown that the content of self-defining memories is structured around concerns, unresolved conflicts, and current goals [3].
In terms of healthy aging's impact on self-defining memories, previous findings showed that when compared to college students, older people recall less detailed and more summarized and positive self-defining memories encoded across the lifespan and focusing on meaningful themes such as achievements (32%), relationships (27.5%), lifethreatening events (21%), recreation (8%), or guilt (5.5%) [45]. De Vries and colleagues [13] also showed that in life review, older people focus more on career and illness than younger people, suggesting that the former have more selfdefining memories around life-threatening events. In terms of the endurance of sense of self in old age, following interviews with people living with dementia in nursing home, Cohen-Mansfield and colleagues [5,6] have shown that all roles deteriorate (professional, family, leisure and personal), while the family role is best maintained.
There is also a wealth of research on self-defining memories in dementia, with consistent findings showing their decline due to the required high level of specificity [28]. Such decline is made responsible for the loss of sense of identity in people living with dementia [4,20]. Unfortunately there is a paucity of studies exploring the retrieval of self-defining memories in dementia. A noticeable exception is Haj and colleagues' work [20]. They showed that exposure to music supports better retrieval of self-defining memories than autobiographical memories, when the music was self-chosen by participants instead of selected by researcher. This important finding can be explained by active involvement in the selection of favorite music of people living with dementia.
To summarize, self-defining memories are key in maintaining the sense of self, but prone to aggressive deterioration in old age and dementia. With a few exceptions there is a lack of studies exploring the possible cues for retrieval of self-defining memories in old age.

METHOD
For our exploratory study, we recruited a convenience sample of 8 older adults (age range 65-82), 7 female and 1 male (P4). Participants were recruited with help from the local branch of the UK University of the Third Age (U3A), a volunteer charitable organization focusing on learning. The rationale for this choice is threefold: older adults have more self-defining memories to report, arguably larger collections of possessions which may be used to cue recall, and are more inclined to contemplate and reflect on their lives in order to develop coherent narratives [15]. Participants are from the UK so the empirical data is drawn from a similar Western context as most of reviewed work.
We employed contextual interviews in participants' homes [23] to observe people interacting with the identified cues and where they are usually stored. Prior to the interviews, they were asked to think of three self-defining memories described according to Singer and Salovey [46] as remembered clearly, important to them, leading to strong feelings, and helping understand who they are and how they come to be the person they currently are. We sensitively focused only on positive self-defining memories in order to limit the risks of revisiting previous trauma which may be associated with negative self-defining memories [47]. Participants were also asked to give a house tour to identify among their possessions those cueing self-defining memories, leaving it open to them to select physical or digital artifacts.
Interviews took about an hour, were audio recorded and fully transcribed. Data analysis involved a hybrid approach with existing concepts informing the deductive coding, while new ones, grounded in the empirical data, contributed to the inductive coding [16]. The deductive coding included concepts from autobiographical and self-defining memory research [28,45,51], with codes capturing the content and types of such memories and their cues [14,24,37,43], and different self-identities [6,17,45]. Participants' memories were identified as self-defined if they contained specific spatio-temporal and phenomenological details such as thoughts and feelings, as well as integrative meaning as lessons extracted beyond the recalled event [4] contributing to their identity construction, enduring concerns or unresolved conflicts [28]. The coding list was iteratively refined as new codes emerged such as redemption memories, crafted cues, and cues' qualities.

FINDINGS
This section describes the identified self-defining memories in terms affect and content, and how they reflect different self -identities. Then we focus on the possible cues for these memories, and the challenges of identifying them.

Self-defining Memories: Retrieval Process
Only two participants (P2 and P8) reported directly their top three self-defining memories. The others started their free recall with life story narratives identifying self-defining memories throughout life periods. Because of this approach, participants were able recollect the sequence of events leading up to their self-defining memories, demonstrating also how those continued to affect them long after the initial event.
While previous work also indicated that older people's selfdefining memories are spread across the lifespan [45], our findings provided new insights into the life periods when such memories were encoded and their weight: childhood (25%), adolescence (15%), young adulthood (15%), adulthood (41%), and old age (4%). Such emphasis on the early life, providing over half of self-defining memories, extends similar findings on episodic memories [28] and reminiscence bump theory [35]. Our findings also contrast those on the life periods from where mementos emerge, i.e., less from formative years and more from recent past and adulthood [34]. This underscores critical distinctions between self-defining memories' cues and mementos, as the former may be shaped by earlier life periods when selfidentities are emerging.

Self-Defining Memories: Affect and Content
Participants reported 28 self-defining memories; slightly more than 3 per person. Over three quarters of self-defining memories elicited positive emotions related to events such as achievements and loving relationships, or to negative ones transformed into redemption narratives (Table 1, column 2,3). The latter could be either relationship-focused: taking responsibility for less fortunate others; or selffocused: overcoming personal challenges through selfmastery. These outcomes confirm those on positive selfdefining memories of older people [45] underpinned by successful relationships and personal achievements [3,5,6,45]. The content of negative self-defining memories relates to loss or transgression, confirming previous findings on unresolved conflicts [3,28] and disrupted relationships [4] albeit not mortality and guilt [3,4,45,48].

Self-Defining Memories: Identities
Sedikides and Marilynn [42] conceptualized self-identity across three levels: personal self (aspects of the self that differentiate the self from all others), relational self (derived from intimate relationships), and collective self (derived from membership in large social groups).
Our findings show the predominance of self-defining memories focused on relational self, followed by those focused on personal self , and to less extent on collective self (Table 1 column 1), confirming outcomes on older adults' emphasis on relational self, followed by personal, and collective self [1]. We now describe the identified selfdefined memories integrating self-identities, with their content and affect.

Relational Self: Loving Relationships
All participants recalled self-defining memories capturing key relationships with loved ones. These include memorable events of meeting the spouse-to-be, the first pregnancy, or family leisure time. P8 reports the memory of acquiring a handmade toy (Fig. 2a)  This quote illustrates that the relational identity wraps around multiple roles, both of wife and mother, and the toy's ability to connect in a nexus, key self-defining events about both roles, signifying romance as well as pregnancy, unknown at the time, followed by the birth of her son. . While this memory is about relational self, it also touches upon the personal self because its spatio-temporal context intersects that of apprenticeship stage with its value for personal self.

Relational Self: Challenging Relationships-Transgression & Loss
Three participants mentioned self-defining negative memories in relation to significant hard to reconcile losses, such as the loss of a child through accident or stillbirth, or transgressions such as physical abuse: "My stepfather was such a bully: sometimes when he was whacking me, the only way I could make him stop was to fall on the floor and pretend to be unconscious. So, I became very afraid of confrontation, disagreement and for lots and lots of years was very much a doormat, and wouldn't hold my ground for anything really" [P3]. This quote indicates the enduring impact of such experiences on the current self. The fewer episodic details of such memories confirm prior findings on repression [3] as a defensive mechanism intrinsic to selfmemory system [7]. Such self-defining memories share the same negative emotions as the original even, but unlike redemption-based memories, they remained unprocessed into a coherent positive resolution [33,59]. HCI work has also started to explore social loss such as the one due to death or separation [41] and its challenging digital traces consisting of poignant evocative and ubiquitous reminders that people feel compelled to delete [40].

Personal Self: Personal Achievements
Five participants mentioned self-defining memories reflecting achievements such as academic success or professional recognition, which shaped their personal self. One such example is from P5 who attended a convent school where she won a handwriting prize (Fig. 2c) [42]. In this light, redemption narratives represent specific types of personal achievements. To summarize, an important outcome is our integration of self-defining memories with self-identities which prior work has limitedly addressed. More specifically, findings indicate that personal self is reflected in memories of achievements and redemption; relational self captures the significance of both loving and challenging relationships, while collective self reflects predominantly negative events with large social impact.

Possible Cues for Self-Defining Memories
The findings from the home tour data show that the majority of cues for self-defining memories are physical objects (Table 1 column 4). Most of these are ready-made objects followed by crafted objects. The other two types of cues are place-based and body-based cues. The emphasis on physical objects, be them ready-made or crafted, extends findings on the prevalence of physical and tangible objects as external cues for autobiographical memories [56] to selfdefining memories. An important finding is the absence of digital artifacts as cues for such memories. Even though non digital printed photos are mentioned, participants have limited preference using them as possible cues. This is  [31,40,55]. To explore the preference for different types of cues, we looked at their link with self-identities, content and affect (Table 1).

Personal Self: Achievement Objects as Possible Cues
Five participants mentioned achievement-related objects usually in paper or object format, including for example old school notebooks with high marks (P1,P2), certificates of academic or professional exams subsequently framed (P2), prizes from school competitions (P5) (Fig. 2c), and objects of professional recognition (P4) (Fig. 3). Such objects are deeply valued for their ability to represent positive emotions related to self-mastery, as key aspect of personal self [17] and content of self-defining memories [3,48,49]. For example, P5's recall of being awarded a handwriting prize in a school competition became richer when the Mary statue was fetched and handled (Fig. 2c).

Relational Self: Possessions and Crafted Objects as Cues
An important finding is that one third of positive selfdefining memories of loving relationships have been linked to handmade crafted objects. Their ability to facilitate rich emotional recall suggests their value as possible cues. Three participants mentioned such possible cues as crafted objects made by themselves through embroidery (P1) or woodcraft (P4), or by others through sewing or weaving (P2). P2's appreciation for her elephant toy has been earlier described and two other such cues are further detailed. One is the wooden banister (Fig. 2b) mentioned by P4 with respect to the self-defining memory of meeting his wife. He crafted the banister for a church and kept because it represented not only the formative time of his joiner apprenticeship, but more importantly the event at which he met his wife organized by the same church: "I can go back even now to the churches that I've worked at, and it gives a great deal of satisfaction to see it" [P4]. Another such cue is a doll cloth handmade by P1 for her daughter to celebrate their bond and for passing on the sewing skills ( Fig. 1 left).

Collective Self: Places as Possible Cues
Given the limited number of identified self-defining memories related to collective self, it is not surprising that we have found only a few such cues. However, given the predominantly negative affect characterizing such memories, it is surprising that such possible cues were even suggested. Interestingly, this place in Normandy has not been photographed, although the family took pictures during this commemoration trip. Even more striking is the crafting of a cue through embroidery (Fig. 1 right): "My husband made that stitching of the unit he was with in Normandy to help his fingers after he was wounded" [P6]. Although the choice of craft was pragmatic, it allows not just the capture but also the processing of negative emotions. This may open up the design space of exploring craft for cueing selfdefining memories related to personal self. This finding also indicates that significant others are in position to identify possible cues which may scaffold the recall of loved ones' self-defining memories.

Personal Self: Body-Based Cues for Redemption Narratives
Another interesting cue type is provided by P2 in relation to her physical disability (Fig. 4) [40,41], but because people are not much motivated to cue the recall of painful memories. The limited reference to cues for negative self-defining memories focused on relational self may also be the reason for finding fewer cues for relational as opposed to personal self. Together, these findings suggest that people have a strong bias towards identifying possible cues for positive rather than negative self-defining memories. Possible Cues for Self-Defining Memory: Challenges We now discuss the challenges of identifying possible cues for self-defining memories. They should be identifiable: cues that people can think of, with or without the ability to reach them physically; retrievable, i.e., easy to be physically found; and accessible, i.e., physically seen or touched.

Difficult to Identify
An important finding is that for almost one third of selfdefining memories, participants experienced difficulties in identifying specific cues. Most of these memories relate to negative events, either as redemption-based memories, or as memories of unresolved issues pertaining to challenging relationships. While people may lack incentives for cueing the recall of memories of unresolved issues, such cues may be valuable in developing coherent life story once successfully processed [33]. It is surprising however, that some redemption-based memories also lack cues. Apart from their negative affect, it is possible that this lack of cues reflects people's challenges of identifying them among their cherished possessions. We do not argue that such cues cannot be found or created, just that during the interviews, participants did not succeed identifying them. . This is an important finding which may be due either to the fact that people do not usually reflect on how specific artefacts may cue their self-defining memories, or because such artefacts need to meet specific qualities which are not commonly embodied by cherished possessions.

Hidden among Cherished Possessions
Another challenge is that even when participants succeed identifying possible cues, they sometimes could not physically retrieve them. This can occur because such selfdefining memory cues have been unwillingly disposed of:

Lost Due to Moving in Later Life
HCI work on ageing and memories has shown that older adults' sense of self is challenged by moving to residential care or downsized homes, which forces them to part with cherished possessions [27]. Our outcomes provide support for this argument by highlighting the negative impact of losing during downsizing or moving house (P1,P3) not only cherished possessions but also possible cues for selfdefining memories hidden among them.
Although participants were able to fetch and show some of self-defining memory cues, they were often able to recall other object-based cues which they were no longer able to physically locate. For example, in relation to her selfdefining memory of a fast learner, P1 mentioned a second memory box with cherished possessions such as books, newspaper clippings, or recipes from her mother, as well as a special schoolbook that she has kept since she was four years old. It had pages of sums written in different color crayons, and most memorable: the sums were all marked correctly which relates to her self-defining memory of fast learner. However, after moving house, she was no longer able to remember where she has put it. Memory boxes were mentioned by other three participants (P2,P3,P5) who indicated the importance of safeguarding their content. It is in such private places where most possible cues for selfdefining memories tend to be stored, with less than one quarter being on display.
To conclude, identifying possible cues for self-defining memories is not trivial, particularly for those with negative emotional content. Possible cues consisting of readyavailable objects are often unassuming objects, hidden among cherished possessions, for which people may lack the sensitivity to recognize their recall value. Moreover, collections of cherished possessions are difficult to maintain in old age, becoming lost, scattered or downsized.

THEORETICAL IMPLICATIONS
We now discuss the findings and how they address the research questions outlined in the introduction. The outcomes advance the understanding of self-defining memories and their cues with new insights into their relationship with self-identities. We also discuss the types of such cues and particularly detail their specific qualities supporting richer emotional recall. Suh qualities are important as they can open up novel design opportunities for memory technologies in old age. We also acknowledge the limitation of the sample consisting mainly of female participants and its possible impact on cues generated through the gendered practice of craft. Despite the relationship between gender and emotional aspects of autobiographical memories [19], previous findings  [4]. Future work however is needed for exploring the role of self-defining memories and their cues for older males, and for experimentally assessing the impact of identified cues on the richer recall of people living with dementia.

New Insights on Self-Defining Memories and Identities
Our approach to link identities with self-defining memories is innovative, offering additional insights into their relationship between self-identities. For example, P8's memory of meeting his future wife, P3 and P6's experience of transgression in relational and social self indicate how personal self is also affected. Our findings also show an emerging relationship between self-identities on the one hand, and the affect and content of their self-defining memories on the other hand. Personal self is reflected predominantly in positive memories of achievement or mastery in front of adversities; relational self on positive loving relationships and less on negative challenging ones, while collective self on negative large scale events. Our outcomes on positive affect, characterizing in decreasing order personal, relational, and collective self, echo those showing that people protect and invest significantly more efforts in personal self [42].

Types of Possible Cues for Self-Defining Memories
Possible cues for self-defining memories include mostly objects tailored to the self-identities they aim to support. Cues as ready-made achievement and crafted objects extend the function of mementos from supporting recall of autobiographical memories [34] to self-defining memories. However, our findings suggest that unlike mementos, cues for self-defining memories consist less of artwork, photos, or memorabilia [34]. This may be because such cues are self-relevant:on display when symbolizing achievements, and seldom interacted with, unlike memorabilia.
We now reflect on the distinction between ready-made and crafted objects as possible cues. For the ready-made cues, the encoding of the memory and the identification of its cue tend to take place simultaneously. Often such objects become part of collections of cherished possessions. The value of crafted objects as possible cues emerges when ready-made ones are neither available nor recognized as important. Crafted objects can be identified as powerful cues either during ( Fig. 2a and 2b), or after the memory has been encoded (Fig. 1 right). Due to the personal involvement in the generation of these cues, they are arguably more effective in supporting recall. As consistent findings have also shown the impact on retrieval of selfgenerated encoding cues [26], we argue for the importance of directly involving the self in the making of cues for selfdefining memories. The craft practice for making such objects is gendered, with women favoring female-oriented crafting such as embroidery, knitting or sewing, and male engaging in woodcraft. These findings extend Csikszentmihalyi's [11] categories of household objects relevant for the self with the dimension of gendered craft.
We also argue for the novelty of body-based cues which received limited attention in cueing episodic memories. Despite that only one participant identified such a cue, our argument for their value is supported by Belk's [2] theory on components of self including body parts as a distinct category. In addition, as our findings indicate, body-based cues have particular value for self-defining memories of redemption narratives related to personal self. The identified place-based cues also build on previous work on location-based cues [24], extending them with insights emphasizing less the tracking of the place but the construction of its meaning. Moreover, for cueing negative memories with large social impact such as battlefields or riots, photos may be too evocative [37].
We also argue that place-based cues have distinct characteristics from object-or body-based cues pertaining to positive emotions of personal and relational self and their defining memories. Instead, the place-based cues tend to be of a larger scale: cities or battle fields versus handheld objects, and public rather than private thus able to safely contain the recall of and sharing with others negative and collective self-defining memories.

Qualities of Self-Defining Memories' Cues
Although, many identified cues are physical artefacts, we argue for the importance of digital cues. The rationale for this argument is threefold. First, our participants between 65 and 82 years old did not live digitally rich lives, but millennials do [32,40], and in their old age much of their identities will be reflected in digital artefacts. Second, the tangibility of many physical cues is particularly important for recall, and interactive tangible cues could better support it. Third, some physical cues are difficult to identify or locate unlike their potential digital counterparts. We also argue that because of their content and qualities, the identified cues for self-defining memories have the potential of supporting richer recall. Their qualities include tangibility, layered meaning, challenging to identify and access, privately stored and fragile.

Tangibility
Despite the exploratory nature of our study, findings indicate that when physical cues are handled, the mere act of holding and often gently caressing them opens up participants' access to richer emotional content in their recall of associated self-defining memories. This is an important finding, given the considerable decline of selfdefining memories in dementia [28]. The richer emotional content triggered by such cues can facilitate access to more specific details required by successful recall of self-defining memories, which in turn can help people with dementia strengthen their sense of identity [4,20]. Arguably, tangible embodied interaction through its ability to support multisensorial recall could open up new design opportunities. Another quality strengthening their value is the ability of such cues to embody layered meaning.

Layered Meanings
An interesting characteristic which makes for strong selfdefining memory cues is layered meanings, which we define as relating to more than one personal goal or theme embodied in a self-defining memory, to a cluster of selfdefining memories occurring close in time, or multiple selfidentities involved in the event. For example, the elephant mentioned by P8 (Fig. 2a) cues a powerful memory capturing two crucial themes around love and motherhood and the object imparts its powerful meaning with three family members. Such cue has been acquired at a life transition when the participant was pregnant, in a spatiotemporal context symbolizing romance, abundance and fecundity. Another example is the wooden banister (Fig.  2b), described by P4 capturing both self-mastery as a young apprentice, and the romance of meeting his future wife. This characteristic is also supported by findings showing that some mementos are also rooted in more than one reason [34]. Lifelogging technologies already capture digital traces of self-identities across the lifespan, and tailoring them to identify the self-defining clusters of events or goals can open up valuable design opportunities.

Challenging to Identify or Access
Cues for positive self-defining memories tend to focus on achievements or relationships, and take mostly the form of ready-made or crafted objects. Cues for negative selfdefining memories which focus on significant personal or social losses, rather than mortality and guilt as previously suggested [3,48,49], tend to be less available. Interestingly, when available, they are not directly accessible taking the form of place-based cues such as outdoor places seldom visited. Findings also indicate that once identified, selfdefining memory cues are valuable in supporting rich recall. However, such object-based cues can become inaccessible at times, as described by almost one third of participants. This can be due to the cues being disposed of by others, misplaced or lost during downsizing or moving house, or being parted with for the benefit of the adult children. Again, lifelogging technologies tailored to identify the self-defining memories across the lifespan could offer easier access to their digital cues. Another surprising finding is the lack of crafted objects for cueing negative self-defining memories, despite the value of craft for processing negative emotions [50]. Here, the value of digital craft can open up new design opportunities.

Privately Stored and Fragile
An important finding is that the majority of accessible selfdefining memory cues are stored in private spaces, with fewer being on display. Most of the objects in the public spaces of the house are achievement-related, i.e., framed certificates. In contrast, other achievement-related objects, usually smaller and linked to early school years, or relationship-related objects tend to be kept in private rooms. Such outcomes support previous ones showing that private spaces store mementos as idiosyncratic objects related to intense emotions, and in particular negative ones which are never on display in public spaces [34]. We found that this insight is even more relevant to objects cuing self-defining memories, as there is a clear distinction between those kept on display, and those kept privately. This is reflected in participants' emotions whilst recalling such memories. While the objects on display were talked about and showed with enthusiasm, those stored privately such as the school prize of Mary statue (Fig. 2c) were more emotionally handled; both however facilitated emotionally richer recall.
An important quality of such cues is their unique material fragility which strongly limits people's interaction with them. For example, participants P5 expressed concerns of even taking out from her memory box the small booklet where she was practicing her handwriting prior to wining the school prize. Such findings confirm those showing that people tend to have boxes of memories with strong reminiscing power, and that fragility of some mementos requires careful storage [34]. In contrast, their digital counterparts, if sensitively designed could allow for more interaction while still engendering a sense of preciousness.

IMPLICATIONS FOR DESIGN
Study findings offer several implications for the design of memory technology to support self-defining memories. They address the challenge of curating self-defining memories and their cues, the value of embedded layered meaning in such cues, and of crafting them when people cannot identify existing cues for self-defining memories.

Curating Self-Defining Memories and their Cues
Findings indicate that people experience difficulties in identifying their self-defining memories. It may be that from the richness of autobiographical memories, they are not used to discriminate those which are self-defining from other episodic memories or personal semantics.
We suggest novel memory technologies which may scaffold people's effort to identify their self-defining memories. For example, such technologies may integrate life review with a variant of card sorting method to iteratively sort such memories, relate them to major goals within life periods, and in particular to themes of achievements, relationships, redemption, and negative unresolved issues as suggested by our findings.
People also have many self-defining memories without identifiable cues. This is surprising given the vast collections of physical and digital possessions that people acquire throughout life. Such challenge can be addressed through novel tools supporting reflective reminiscing on self-defining memories in order to explicitly identify their cues among personal collections. For example, these may support engagement with cherished possessions, stored both digitally and in private memory boxes, with the aim of creating links between such artefacts and specific selfdefining memories. Metadata of digital collections can be used for selecting such cues from digital trails captured through lifelogging or social media.

Embedding Cues with Layered Meaning across Lifespan
Our findings show that richer recall is particularly supported by cues able to prompt retrieval of multiple themes associated with a self-defining memory, multiple such memories or self-identities related to them. We can think of novel memory technologies to support not just the identification of self-defining memories' cues but also the evolving construction of the meaning embedded in them. This may exploit cues acquired or crafted at life transitions when multiple identities intersect, offsetting the risk of not recognizing their associated memories as self-defining at the time. The emerging HCI interest in life transitions offers a useful lens to support the identification of cues for such self-defining memories even before they are encoded [40].
The lifespan approach to identifying and developing these cues is also valuable in addressing the challenge of them being disposed of, lost during downsizing, or unwillingly parted with.
Augmenting cherished objects with digital memories has been previously suggested [34], particularly for tracking such cues during the moving in later life in residential or care home [27]. Arguably, such curatorial acts can be even more valuable when performed across the entire lifespan. Stories attached to self-defining memory cues could mitigate the risk of having them disposed by others, as such stories uncover the deep personal meaning hidden within otherwise unassuming objects. They can also help communicate to adult children the value of such objects for supporting the personal self of their old parents, and mitigate the risk of older people unwillingly parting with them. In addition, digital stories could mitigate the risk of misplacing or losing these cues across the lifespan.

Crafting Self-Defining Memories' Cues
An important finding is that almost one third of selfdefining memories have unidentifiable cues. This may be because identifying possible cues for self-defining memories is challenging particularly for those with negative emotional content. Such challenge opens up design opportunities shifting the focus away from personal possessions and possible cues buried in them, to novel technologies for crafting new cues [39]. Moreover, crafting is particularly valuable in facilitating self-involvement whose role in recall has been consistently shown [20,26].
Our study methodology asked participants to actively identify and select their cues, an additional factor which may be responsible for their success. This is supported by findings showing that self-chosen music can cue selfdefining memories of people living with dementia [20].
Building on this argument, we could consider extending people's involvement from cue selection to cue generation, since consistent findings have also shown the impact of self-generated encoding cues on recall of autobiographical memories [26]. We argue that precisely this involvement is provided by crafted objects, which may explain their emphasis in our findings. We can imagine novel interactive systems integrating tangible interfaces with elements of art therapy to support not just capturing the event, but also reflecting on and processing of negative self-defining memories and their progression into redemption narratives.

Sensitive Design for Physical Handling of Fragile Cues
An interesting outcome is the value of physical handling of cues for self-defining memories. The mere gentle touch of their surfaces allows for accessing richer emotional content. Such cues tend to be fragile as they stood the test of time accompanying people through their life trajectories. We can imagine new memory technologies consisting of hybrid objects. The physical part could be crafted objects that people will value such jewelry. Such artifacts have the potential to embody both preciousness and fragility. The digital part of these hybrid objects could capture, store, and play short digital stories of its meaning, or even multisensorial digital stories further emphasizing touch. To support their fragility, sensitive design can explore ways in which such hybrid objects could embody the signs of wear of both its physical and digital parts. For example, the physical object could get patina [29], while the digital part through subsequent use may show signs of faded images or crackle sound similar to those of old vinyl records.

CONCLUSION
This study investigates older adults' identification of selfdefining memories and their possible cues. We advance the understanding of such memories and cues with new insights into their relationship with self-identities. Personal self is reflected in memories of achievements and redemption; relational self captures loving or challenging relationships, while collective self reflects negative events with large social impact. These tend to be cued by achievement objects, body-based cues, crafted objects, and place-based objects, respectively. The key qualities of such cues include being ordinarily unassuming, difficult to identify and access as they are hidden among cherished possessions, often fragile and privately stored, and rich in layered meanings which allow for strong emotional recall. Our findings suggest novel memory technologies with potential to better support self-defining memories and their cueing than lifelogging technologies. The shift of focus from capture technologies towards those emphasizing the curation of self-defining memories and their cues, the embedding of layered meaning in such cues across lifespan, and of crafting these cues could better support such memories and the fragile sense of self in old age.