Individualized Theory of Mind (iToM): When Memory Modulates Empathy.

Functional neuroimaging studies have noted that brain regions supporting theory of mind (ToM) overlap remarkably with those underlying episodic memory, suggesting a link between the two processes. The present study shows that memory for others' past experiences modulates significantly our appraisal of, and reaction to, what is happening to them currently. Participants read the life story of two characters; one had experienced a long series of love-related failures, the other a long series of work-related failures. In a later faux pas recognition task, participants reported more empathy for the character unlucky in love in love-related faux pas scenarios, and for the character unlucky at work in work-related faux pas scenarios. The memory-based modulation of empathy correlated with the number of details remembered from the characters' life story. These results suggest that individuals use memory for other people's past experiences to simulate how they feel in similar situations they are currently facing. The integration of ToM and memory processes allows adjusting mental state inferences to fit unique social targets, constructing an individualized ToM.

Introduction

Humans have the ability and the need to interpret the mental states of other people, including their thoughts, feelings, and intentions. This ability, called “theory of mind” (ToM) is crucial for empathy, which includes representing other people’s thoughts (referred to as “cognitive empathy”) and experiencing affective states aligned to theirs (referred to as “affective empathy”; Amodio and Frith, 2006; de Vignemont and Singer, 2006; Zaki and Ochsner, 2012). Understanding the neural bases and the cognitive mechanisms of ToM is an important goal of research in cognitive neuroscience.

Recent reviews of functional neuroimaging (fMRI) studies have noted that brain regions supporting ToM and cognitive empathy overlap remarkably with those underlying autobiographical memory. The regions of overlap include the medial prefrontal and posterior cingulate cortex, frontal pole, inferior frontal gyrus, regions within the medial temporal lobe, superior temporal sulcus and middle temporal gyrus, and the angular gyrus (Buckner and Carroll, 2007; Rabin et al., 2010; Spreng and Grady, 2010; Zaki and Ochsner, 2012). These regions take part in the “default network,” a set of interconnected brain regions whose activity typically is suppressed by stimulus-driven attention, and enhanced by internally focused activities (Buckner et al., 2008; Spreng and Grady, 2010).

What might be the functional meaning of this overlap? Some authors have suggested that, in order to infer others’ mental states, individuals draw on past experience (Corcoran and Frith, 2003; Gallagher and Frith, 2003; Buckner and Carroll, 2007). Thus, individuals may understand how others feel because they recall having experienced similar episodes personally, and how they felt at that time. Batson et al. (1996), for example, found that participants who had experienced the same electrical shock as their confederates displayed a greater empathic response toward them relative to participants who never had experienced the shock. Likewise, participants who went through a similar experience as the protagonist in a story showed a greater empathic response toward the protagonist relative to participants who had never experienced such an event (Batson et al., 1996). Dimaggio et al. (2008) have shown that the more individuals are able to retrieve and reflect on episodes from their own life narratives, the more likely they are to decipher others’ thoughts and emotions. According to the “episodic simulation hypothesis” (Buckner et al., 2008; Schacter et al., 2008), indeed, one fundamental function of episodic memory, beyond recalling the past, is enabling the formulation of flexible models of the future to inform choice. In the social domain, this may entail retrieving fragments of past events rich in experiential detail, recombining them to construct new scenarios suited to represent the situation currently faced, and pre-experiencing how we, or others, might feel in such situations. The hypothesis of a close relation between episodic memory and ToM processes is supported by evidence that episodic memory and ToM emerge close in time during development (Perner and Ruffman, 1995). Moreover, patient populations with ToM impairments, such as high-functioning autism (Adler et al., 2010) and schizophrenia (Corcoran and Frith, 2003), as well as patients with damage to the ventromedial prefrontal cortex (Stone et al., 1998; Stuss et al., 2001; Shamay-Tsoory et al., 2005, 2009; Ciaramelli et al., 2012) also show a (possibly parallel) impairment in autobiographical recollection (see Gilboa and Moscovitch, 2002; Dimaggio et al., 2012; for reviews), as well as functional abnormalities in the brain default network (e.g., Kennedy et al., 2006; Harrison et al., 2007).

On the other hand, a neuropsychological study of ToM in two amnesic individuals provided contradictory evidence. Rosenbaum et al. (2007) showed that, despite severely impaired autobiographical memory, amnesic patients were normally able to perform standard laboratory ToM tests that required them, for example, to identify whether a character unintentionally said something hurtful to another character, committing what is called a “faux pas” (Faux Pas Test; Stone et al., 1998). The finding that ToM may be intact in amnesic patients with impaired autobiographical memory poses constraints on the purported relation between episodic memory and ToM, indicating that, at least under some circumstances, ToM is independent of episodic memory. As Rosenbaum et al. (2007) speculate, indeed, detection of a faux pas may be achieved through the retrieval of semantic knowledge of how a person might stereotypically feel in a given situation, and of social etiquette (see also Spreng and Mar, 2012). For example, we know that people do not like being told that they look older than they are and we avoid making comments like that. To do so, we do not need to resort to episodic retrieval or simulation. The question remains, therefore, as to whether and why ToM would need the support of episodic simulation processes.

As much as semantic memory is about scripts and general knowledge about the world, it is not suited to capture the characteristics of unique individuals and unique situations. Exclusive use of semantic representations may be sufficient to infer what the average person is likely to experience in a given situation, which is what is required by most standardized ToM tests that employ strangers as the social targets (e.g., Stone et al., 1998; Baron-Cohen et al., 2001; Rabin et al., 2010; Spreng and Grady, 2010; St. Jacques et al., 2011). However, we typically do not reason about average strangers, but about distinct individuals, some of whom we know well, and with whom we interact. In this case, semantic representations, which only allow for stereotyped interpretations of others’ behavior, are likely to be insufficient to make adaptive mental state inferences. In this case, episodic memory for shared experience may be necessary to tailor ToM processes on the social target we are interpreting, to construct what we call an “individualized” ToM (iToM). Suppose, for example, you have a friend that for all of his life has been said to look young for his age, and this was a problem in trying to get positions of responsibility. In this situation, one would abandon (semantic) social etiquette, and create instead a new, ad hoc rule of avoiding commenting on his youthful appearance.

The present study

Our hypothesis is that episodic memory is necessary to retrieve previously acquired information in order to derive tailored interpretations of others’ behavior, allowing one to adjust ToM to fit unique social targets (iToM). To test this hypothesis, in the present study we had participants read about the lives of two individuals, one extremely unlucky in love, whose life was punctuated with episodes involving failures in intimate and romantic relationships, and the other extremely unlucky in professional life, whose life was punctuated with failures at school and at work. The two life narratives were rich in episodic and experiential detail. Participants’ immediate free recall of the narratives provided us with a measure of recollection. Next, participants considered social scenarios that could or could not contain a faux pas, that is, a situation in which a character unintentionally hurt a second character (the victim), and participants had to detect the faux pas and report how much empathy they felt for the victim. We assessed both cognitive empathy and affective empathy. Critically, in some scenarios the faux pas concerned the victim’s intimate/romantic relationships (Love scenarios), whereas in other scenarios it was about the victim’s professional relationships (Work scenarios). For comparison purposes, generic violations of social norms not specifically involving intimate/romantic or professional relationships were also used. The type of social violation and the identity of the victim were experimentally crossed, such that the victim in the scenarios could be the characters participants had read about – one unlucky at work, one unlucky in love – as well as another character about whom participants knew nothing.

Our main prediction was that empathy toward the victim of a faux pas should be modulated by memory for the victim’s life story. Thus, we expected more empathy for the victim unlucky in love in Love scenarios, and for the victim unlucky at work in Work scenarios. If this memory-driven modulation of empathy is based on the recollection of the victim’s story, then a correlation should be expected between the entity of the modulation and the amount of detail recalled from victim’s life story. We also investigated whether memory for the victim’s life story would predict a better ability to detect social violations, i.e., ToM accuracy. To the extent that episodic memory and ToM recruit overlapping neural networks (Buckner and Carroll, 2007), one would expect a correlation between recall accuracy and faux pas recognition accuracy. On the other hand, it has been reported that trait empathy may be related to ToM and social cognitive functioning (Zaki and Ochsner, 2012). For example, individual differences in self-report measures of empathy track with activity in brain regions associated with mentalizing and ToM (Wagner et al., 2011). Moreover, patients with lesions in the ventromedial prefrontal cortex exhibit both low self-reported cognitive empathy (Shamay-Tsoory et al., 2009) and impaired understanding of other people’s thoughts and intentions (Stone et al., 1998; Shamay-Tsoory et al., 2005; Ciaramelli et al., 2012). On this view, one may expect that participants with high levels of trait empathy, especially cognitive empathy, would be better able to adopt the perspective of the victims in the scenarios, and therefore adjust their empathic response depending on the identity of the victim.

Materials and Methods

Participants

Thirty-one individuals were recruited for the study. Participants were tested at the Department of Psychology of the University of Toronto, Canada, and at the Department of Psychology of Northwestern University, USA. Two subjects were excluded from the study because part of the data was lost as a result of technical problems. The experimental group, therefore, was composed of 29 participants (11 males), with mean age of 22.38 years (range 18–37) and a mean education of 15.03 years (range 13–19). For four of the participants, cognitive empathy scores were not registered due to a malfunctioning of the computer program. Therefore, all the analyses involving cognitive empathy were run on 25 of the 29 participants. Participants were not taking psychoactive drugs, and were free from current or past psychiatric or neurological illness as determined by history. Participants gave written informed consent for the study, which was approved by the ethics committees of the University of Toronto and of Northwestern University. They received course credit or $10.

Materials

The materials included two “life-stories” narrating excerpts from the lives of two fictitious characters, a faux pas recognition task, and the Interpersonal Reactivity Index Questionnaire (IRI; Davis, 1980).

Life-stories recall task

We created two stories narrating the lives of two fictitious characters, in the form of a series of episodes narrated in first person (see Life-stories in Appendix). The life story of one character concerned a series of events mainly involving failures in intimate and romantic relationships. For male participants in the experiment, this character, unlucky in love, was called Mike, whereas for female participants the character was called Susan. Life-stories were slightly adapted depending on the gender of the protagonist (see Life-stories in Appendix). The life story of the other character, on the other hand, concerned a series of events mainly involving failures in professional life. This character, unlucky at work, was called Adam and Jean for male and female participants, respectively. Thus, while Mike/Susan’s life was punctuated by unsuccessful love-related events (e.g., I just asked her for her name and she answered: “I am sorry, I am very busy and cannot waste my time!”), Adam/Jean’s life mainly involved work-related failures (e.g., I have been hired recently, but I already had a hard time with my boss. He said that I had mistakenly filed some invoices, and that was a disaster). Life-stories were rich in spatial, perceptual, and emotional details (e.g., […] the flat is on Queen West, on the second floor of a Victorian house painted in red with a blue ceiling It’s just that sometimes I can feel like such a nothing). Life-stories were subdivided a priori in conceptual units, each of which conveyed a bit of information, such as a unique occurrence, fact, statement, thought, etc. (see also Levine et al., 2002 for a similar segmentation procedure). In the first version of the life-stories, which was administered to participants tested at University of Toronto (N = 14), the love-related story contained 199 conceptual units, and the work-related story contained 233 conceptual units. Participants subsequently tested at Northwestern University (N = 15) received a slightly shortened version of the love-related story and the work-related story, which both contained 171 conceptual units.

Faux pas recognition task

Thirty-seven social scenarios were selected and adapted from the “Faux Pas Recognition Test” (Baron-Cohen et al., 1999) or the “Social Stories Questionnaire” (Lawson et al., 2004), or created ad hoc (see Faux Pas Scenarios in Appendix for the list of scenarios used). Twenty-seven scenarios contained a violation of accepted social norms (a faux pas): an individual said something awkward that may hurt someone (the victim). The remaining 10 scenarios involved neutral social interactions that did not contain any faux pas (Neutral scenarios). In eight scenarios, the faux pas hit the victim on his/her intimate/romantic relationships (Love scenarios; e.g., […] Susan wondered whether her boyfriend also remembered their anniversary. Then he came out of the shower. “What were you saying honey? I couldn’t hear from the shower,” asked the boy. Susan said: “I thought it would be nice to go out for dinner since today…” But he interrupted her: “Oh sorry honey, not today: I promised some colleagues I would join them for a drink! What about next Saturday?”). In nine scenarios, the faux pas was about the victim’s professional life (Work scenarios; e.g., Susan went to say bye to the director of her office, because she was planning to leave for her holidays. […]. “Don’t worry,” answered the director, “I don’t think your absence will cause huge problems. Have fun!”). Ten scenarios involved generic social violations that did not pertain specifically to the victim’s intimate/romantic or professional life (Generic scenarios; e.g., Susan had just met an Italian colleague, Paola. […] Susan wanted to invite Paola for dinner to get to know her better. “Hey Paola, why don’t you come over for dinner some time, we just bought a pasta machine and you could help us figure how to use it!” Paola answered: “Oh why not… but I guess you should be able to get it to work. Wasn’t there an instruction manual?”).

All participants considered 20 scenarios: five Love scenarios, five Work scenarios, five Social scenarios, and five Neutral scenarios. Participants received slightly different sets of scenarios. This was done as part of a pilot research project testing the generalization of our effects to different types of materials, which was later discontinued. In all our analyses, we assessed (and controlled for) the effect of the different set of scenarios used. Twenty participants received Set A (Love scenarios #: 1, 2, 3, 4, 5; Work scenarios #: 1, 2, 3, 4, 5; Generic scenarios #: 1, 2, 3, 4, 5; Neutral scenarios #: 1, 2, 3, 4, 5; see Faux pas scenarios in Appendix), 4 participants received Set B (Love scenarios #: 1, 4, 6, 7, 8; Work scenarios #: 4, 6, 7, 8, 9; Generic scenarios #: 6, 7, 8, 9, 10; Neutral scenarios #: 3, 4, 6, 7, 8), and five participants received Set C (Love scenarios #: 1, 4, 6, 7, 8; Work scenarios #: 4, 6, 7, 8, 9; Generic scenarios #: 6, 7, 8, 9, 10; Neutral scenarios #: 4, 6, 8, 9, 10). Each scenario was presented three times. One time the victim of the faux pas was Mike/Susan (the character unlucky in love), one time it was Adam/Jean (the character unlucky at work), and one time it was Jason/Patricia, a character participants did not know anything about. The order of the scenarios was randomized for each participant.

The participants’ task was to consider each scenario, and judge whether it did or did not contain a faux pas by pressing one of two buttons. For affirmative responses, participants additionally answered two questions regarding the amount of cognitive empathy (i.e., “How bad do you think [victim] felt from one (not bad) to seven (very bad)?”) and affective empathy (i.e., “How bad do you feel for [victim] from one (not bad) to seven (very bad)?”) they felt for the victim.

Interpersonal reactivity index

The IRI is a self-report scale assessing four main aspects of empathy. The “perspective taking” (PT) scale assesses the ability to adopt spontaneously the perspective of other people; the “fantasy” (F) scale assesses the tendency to identify with characters in movies, novels, and other fictional situations; the “empathic concern” (EC) scale assesses the tendency to experience feelings of warmth, compassion, and concern for others undergoing negative experiences; the “personal distress” (PD) scale assesses feelings of anxiety and distress for people going through sufferings and afflictions (Davis, 1980).

Procedure

Each participant was scheduled for an individual experimental session. First, participants were asked to read the two life-stories and were informed that, after reading each story, they would have to recall it in as much detail as they could. The order of presentation of the two life-stories was counterbalanced across participants. There was no time limit to read and recall the life-stories, which took about 10 min in all cases. Participants’ reports were recorded. All data were collected and scored by author FB. Scoring involved counting the number of conceptual units correctly recalled, a procedure similar to the one used for the Logical Memory subtest of the Wechsler Memory Scale (Wechsler, 1987). At the time of scoring, FB was blind to the results participants had attained in the faux pas recognition task. An additional rater, blind to the experimental hypotheses, rated 50% of the recall reports. Inter-rater agreement was high (r = 0.96).

Participants then underwent the faux pas recognition task. They were told that the task consisted in reading social scenarios, and deciding, for each scenario, whether it contained a faux pas, that is, a situation in which someone said something inconvenient or awkward that may hurt another person, and how they felt for the victim. Participants were encouraged to “put themselves in the victim’s shoes” while considering the scenarios. They were also informed that the scenarios may involve the characters they had read about previously. After completing the faux pas recognition task, participants completed the IRI questionnaire.

Results

Faux pas recognition accuracy

Table 1 reports participants’ accuracy in detecting faux pas, by type of scenario and character. An analysis of variance (ANOVA) on accuracy scores with Scenario (Love scenarios, Work scenarios, Generic scenarios, and Neutral scenarios) and Character (unlucky in love, unlucky at work, unknown) as within-subject factors, and Set (A, B, C) as between-subject factor, revealed a significant effect of Scenario [F(3,78) = 3.24; p < 0.05, ]. Post hoc Duncan comparisons showed that subjects attained a lower recognition accuracy in detecting work-related faux pas compared to love-related faux pas, generic faux pas, or the absence of faux pas (Neutral scenarios; p < 0.05 in all cases). The effect of Set was significant [F(2,26) = 4.32; p < 0.05,], such that a lower recognition accuracy was associated to Set A compared to Set C scenarios (0.79 vs. 0.94; p < 0.05). The factor Set, however, did not interact in a significant way with any of the other variables (p > 0.47, ) in all cases. The Scenario × Character interaction was not significant (p = 0.79, ), indicating that the identity of the victim did not influence participants’ ability to detect faux pas in social interactions (or the lack thereof).

Table 1

Mean faux pas recognition accuracy by type of scenario and character.

	Love scenarios	Work scenarios	Generic scenarios	Neutral scenarios
Unlucky in love	0.90 (0.03)	0.70 (0.04)	0.81 (0.04)	0.90 (0.02)
Unlucky at work	0.91 (0.02)	0.75 (0.04)	0.80 (0.04)	0.91 (0.03)
Unknown	0.88 (0.03)	0.69 (0.04)	0.80 (0.04)	0.90 (0.03)

Values in parentheses represent the standard error of the mean.

Cognitive and affective empathy

We next investigated whether individuals’ empathy toward the victim of a faux pas depended on the identity (and story) of the victim. We first analyzed affective empathy scores. An ANOVA on affective empathy scores with Scenario (Love, Work, Generic), Character, and Set as factors revealed a significant effect of Scenario [F(2,52) = 15.71; p < 0.0001, ], and a significant effect of Character [F(2,52) = 5.41; p < 0.01,], which were qualified by a significant Scenario × Character interaction [F(4,104) = 6.40; p < 0.0001,]. Post hoc Duncan comparisons showed that affective empathy scores in Work scenarios were higher for the character unlucky at work than for the character unlucky in love (p < 0.001) or the unknown character (p < 0.0005), with no difference between the character unlucky at work and the unknown character (p = 0.10). Conversely, affective empathy scores in Love scenarios were higher for the character unlucky in love than for the character unlucky at work (p < 0.05), and, marginally, for the unknown character (p = 0.07), with no difference between the character unlucky at work and the unknown character (p = 0.41). In contrast, affective empathy scores were not modulated by the character’s identity in Generic scenarios (p > 0.09 in all cases; see Figure 1). The factor Set was not significant, and did not interact in a significant way with any of the other variables (p > 0.61, in all cases).

Figure 1

Mean affective empathy scores (TOP panel) and cognitive empathy scores (BOTTOM panel) by type of scenario and character. Error bars indicate the standard error of the mean. *p < 0.05.

Conceptually similar results were obtained on cognitive empathy scores. The ANOVA on cognitive empathy scores evinced a significant effect of Scenario [F(2,44) = 12.14; p < 0.0001, ], and a marginal effect of Character [F(2,44) = 3.02; p = 0.058, ]. The Scenario × Character interaction just failed to reach conventional levels of statistical significance [F(4,88) = 2.45; p = 0.051, ], likely due to the fact that this analysis was run on 25 subjects only. Because the effect size for the Scenario × Character interaction is in the medium range (), and for completeness, we ran post hoc Duncan comparisons. Cognitive empathy scores in Work scenarios were higher for the character unlucky at work than for the character unlucky in love (p < 0.05) or the unknown character (p < 0.01), with no difference between the latter two (p = 0.29). Conversely, cognitive empathy scores in Love scenarios were higher for the character unlucky in love than for the character unlucky at work or the unknown character (p < 0.05 in both cases), with no difference between the latter two (p = 0.76). In contrast, cognitive empathy scores were not modulated by the character’s identity in Generic scenarios (p > 0.06 in all cases; see Figure 1). Again, the factor Set was not significant, and did not interact in a significant way with any of the other variables (p > 0.53, in all cases).

These findings indicate that empathic responses are influenced by the victim’s identity and type of social violation s/he is facing, consistent with the hypothesis that retrieval of past episodes about others influences our appraisal of similar social situations involving the same individuals. This hypothesis is explored further in the next section.

Empathy modulation, episodic memory, and empathy scales

We next investigated whether the degree to which participants modulated their empathic responses depending on the victim’s identity correlated with their ability to recall the victim’s story in detail, and to standard self-report measures of empathy as assessed in the four subscales of the IRI (F, PT, EC, PD).

We calculated a recall accuracy score as the number of conceptual units recalled out of the total number of conceptual units (collapsed across the two life-stories). Recall accuracy was highly variable across participants, ranging from 0.07 for individuals who merely reported the gist of the stories to 0.36 for individuals who reported numerous vivid qualitative details (Mean = 0.21; SD = 0.08). We also calculated an index of memory-based empathy modulation. For Love scenarios, the index was calculated as (Empathyunlucky in love − Empathyunlucky at work)/Empathyunknown, and it indicated the difference between the amount of empathy felt in Love scenarios toward the victim unlucky in love minus that toward the victim unlucky at work, adjusted for the participant’s general tendency to empathize, i.e., empathy toward the unknown victim. The same index was calculated for empathy scores in Work scenarios as (Empathyunlucky at work − Empathyunlucky in love)/Empathyunknown. The two indices were summed to get a general empathy modulation index. The empathy modulation index was calculated for cognitive empathy scores and affective empathy scores separately.

In order to investigate the relation between the empathy modulation indices and recall accuracy and self-report measures of empathy, while controlling for the effect of Set, we ran an Analysis of Covariance (ANCOVA) on affective and cognitive empathy modulation scores, with Set as between-subject factor, and recall accuracy, and the scores at the F, PT, EC, and PD subscales of the IRI as covariates. The ANCOVA on affective empathy modulation scores showed that the affective empathy modulation index correlated with recall accuracy [β = 0.70, F(1,21) = 8.69, p < 0.01, ; see Figure 2], but not with the scores at the IRI subscales (p < 0.52, in all cases). There was no significant effect of Set (p = 0.27, ). The same ANCOVA on cognitive empathy modulation scores showed that the cognitive empathy modulation index correlated positively with recall accuracy [β = 0.79, F(1,17) = 10.69, p < 0.005, ] (see Figure 2), but not with the scores at the IRI subscales (p > 0.37, in all cases). There was no significant effect of Set (p = 0.14, ).

Figure 2

Scatter plot of the bivariate correlation between recall accuracy for the life-stories and the affective empathy modulation index (TOP panel) and the cognitive empathy modulation index (BOTTOM panel).

Faux pas recognition accuracy, episodic memory, and empathy scales

Finally, we investigated whether the ability to detect faux pas was related to recall accuracy for the victims’ life-stories, and to standard self-report measures of empathy. An ANCOVA on faux pas recognition accuracy with Set as between-subject factor, and recall accuracy and the scores at the F, PT, EC, and PD subscales of the IRI as covariates, yielded no significant effect of recall accuracy (p = 0.52; ), IRI scores (p > 0.44; in all cases), or Set (p = 0.14; ). Similar results were obtained if we focused on recognition accuracy for Work scenarios and Love scenarios, whose contents are related to the main themes of the victims’ life-stories, and recognition accuracy for Generic and Neutral scenarios, whose contents are not related to memory contents (p > 0.07, in all cases).

Discussion

Understanding other people’s thoughts and feelings in a given situation is adaptive if we have a distinctive representation of who they are, and how they have behaved in similar situations. This enables us to anticipate their reactions to the current situation as accurately as possible. The present study shows that episodic memory for others’ past experiences modulates significantly our appraisal of, and reaction to, what is happening to them currently.

Participants read the life-stories of two characters; one had experienced a long series of love-related failures, the other a long series of work-related failures. Consistent with the hypotheses, in a later faux pas recognition task, they reported more empathy for the character unlucky in love if she or he was the victim of an additional love-related faux pas than they did for the character unlucky at work or for an unknown character. Analogously, more empathy was reported for the character unlucky at work receiving yet another work-related social violation than it was reported for the character unlucky in love or an unknown character. It is worth emphasizing that participants did not appear to be generally more empathic toward the two familiar characters compared to the unknown character (e.g., Stinson and Ickes, 1992). The increase in empathy for the familiar characters, indeed, was highly situation-specific: participants felt more empathy toward the character unlucky in love in love-related scenarios, but treated this character normally in work-related scenarios, and vice versa for the character unlucky at work. Additionally, all characters attained comparable levels of empathy in Generic scenarios, whose contents were not relevant to the life story of either character. Rather, the present results suggest that while contemplating faux pas scenarios, individuals recollected previous episodes involving the victim analogous in content to the situation she currently faces, and how the victim had felt in those situations. Retrieval of past episodes involving the victim allowed a better simulation of how she felt in response to the current faux pas, or, in other words, the construction of an iToM, shaping individuals’ empathic responses accordingly.

The present findings are in line with the “episodic simulation hypothesis” (Schacter et al., 2008; see also Buckner and Carroll, 2007), according to which retrieval of past experiences is needed to envisage fictitious events, and make decisions based on such simulations. In line with this, the amount of detail recalled from the victims’ life-stories correlated positively with the degree to which individuals modulated their empathic responses depending of the identity of the victim, suggesting that a greater availability of episodic memories from the victims’ life helped them to envisage their feelings in similar situations more vividly and faithfully (see also Spreng and Mar, 2012, for a Discussion). These results make contact with findings from a recent fMRI study (Perry et al., 2011), in which hippocampal activity was detected while subjects made emotional judgments about people deemed similar to themselves and facing events that had occurred in their own life (e.g., How would Joe feel about losing his wallet?), suggesting that individuals may resort to personal memories in order to understand other people better and empathize with them. Of course, different from Perry et al.’s (2011) paradigm, in ours subjects did not need to use the self (and personal memories) as a proxy to empathize with the victims (Mitchell et al., 2006), as they had the original memories of the victims’ past experiences available.

Our findings fit nicely with the results of a recent fMRI study investigating the relation between autobiographical memory and ToM (i.e., imagining the thoughts and feelings of another person) for personally known versus unfamiliar others (Rabin and Rosenbaum, 2012). It was found that brain regions supporting ToM for personally known others overlapped more closely with those supporting autobiographical memory than did regions supporting ToM for unknown others, and the overlap was maximal in midline regions, including the hippocampus. This finding suggests that in order to imagine the mental states of known people individuals rely, to some extent, on shared past experience (Rabin and Rosenbaum, 2012; see also Rabin et al., 2012b). Consistently, a previous fMRI study had shown that the medial prefrontal cortex responded more strongly when participants made judgments about the personal preferences of friends and close others, which arguably are likely to trigger relevant past experiences, relative to strangers, regardless of whether the other person is perceived as similar to oneself (Krienen et al., 2010). On the other hand, ToM for unfamiliar others was associated with activity in more lateral frontal and temporal regions associated with accessing semantic knowledge (e.g., Martin and Chao, 2001), consistent with the idea that to infer strangers’ mental states reliance on social scripts and general knowledge may be sufficient (Rosenbaum et al., 2007).

Previous research has shown that episodic memory retrieval is related to simulation of future events, and functional to social cognition. Amnesic patients with problems at remembering specific episodes from the past, indeed, may also exhibit problems at imagining specific episodes in the future (Tulving, 1985; Rosenbaum et al., 2005; Race et al., 2011; but see Squire et al., 2010), and their constructions of fictitious events appears significantly reduced in richness and content compared with those of controls (Hassabis et al., 2007). Crucially, it has been demonstrated that problems in simulating fictitious events has an impact on social problem solving. Sheldon et al. (2011) had patients with unilateral temporal lobe epilepsy and excisions (TLE), older adults, and control participants describe detailed solutions to various open-ended, social scenarios. TLE patients and older adults, both having deficits in episodic memory, provided fewer steps relevant to the given solution than their comparison group, and their descriptions of the step was made of fewer internal (episodic) details but a similar number of external (semantic) details compared to their control groups. Thus, even though amnesic patients can solve standardized ToM tasks normally (Rosenbaum et al., 2007), memory problems may in fact hinder performance in ill-defined, real world social situation that require evaluating the outcome of multiple, alternative mental simulations of the situation being considered, and integrating context-specific with person-specific information, tapping episodic simulation processes to a greater extent. Consistent with this interpretation, Levine et al. (1998) described a patient, ML, who had sustained damage to the right uncinate fasciculus, a band of fibers which connects the medial temporal lobe with ventral frontal cortex, and who had severe autobiographical memory deficits associated with problems in real world social interactions. Interestingly, ML had difficulty knowing how to behave around family members and friends and, despite being able to re-learn socially acceptable behavior under structured routines, he remained unable to self-regulate his behavior in unstructured situations (Levine et al., 1998).

It has been noted that both re-experiencing the past and inferring others’ mental states require the ability to consider alternatives to events in the immediate environment, or self-projection (Buckner and Carroll, 2007; Mitchell, 2009), be this toward another time (for episodic memory) or another person’s perspective (for ToM). One way to interpret the overlap in brain activity between episodic memory and ToM, therefore, is as structural in nature: both activities are supported by the same neural circuitry, the one that enables self-projection. Were the relation between episodic memory and ToM merely structural, however, one would expect a correlation between episodic memory and ToM performance. However, in the present study free recall (of the life-stories) was not related to faux pas recognition accuracy, and this held even if we focused on Love and Work scenarios, whose contents resonated with memory contents. This result is compatible with previous evidence showing that patients with significant episodic memory problems can attain normal accuracy in ToM tasks, including faux pas recognition tasks (Rosenbaum et al., 2007; Rabin et al., 2012a). Additionally, faux pas recognition accuracy was not related to “PT” scores in the IRI, as the self-projection hypothesis would predict. Our results, therefore, are more consistent with the view that ToM systems, though inherently sufficient to decipher social situation/violations, may co-opt episodic memory systems to integrate flexibly the characteristics of the situation with those of the victim, modulating empathic responses accordingly. This suggests a functional relation between episodic memory and ToM that is more in line with the episodic simulation hypothesis.

The “functional” (as opposed to “structural”) interpretation proposed is also in line with the fact that we found largely parallel effect of episodic memory on cognitive empathy and affective empathy, while only the brain regions supporting cognitive empathy overlap with those supporting autobiographical memory (de Waal, 2008; Shamay-Tsoory et al., 2009; Zaki and Ochsner, 2012). In contrast, affective empathy is related to the ability to share others’ emotional experiences through mirroring neural mechanisms (Preston and de Waal, 2002; Gallese et al., 2004; Singer and Lamm, 2009). Note, however, that mirroring occurs (and has been investigated) typically when perceivers make use of observable cues about what another person is feeling, whereas self-projection is mostly engaged when inferring the mental states of individuals that are not physically present (Zaki and Ochsner, 2012). Because in the present study participants made both cognitive and affective empathy judgments for individuals who were removed from their current experience, both judgments likely relied on, and were modulated by, the same type of (memory) cues (see de Vignemont and Singer, 2006, for other evidence for the contextual modulation of affective empathy). Indeed, the cognitive and the affective modulation indices were highly correlated in our sample (r = 0.83). An additional reason why cognitive empathy and affective empathy may have been aligned in our study is that participants were young individuals, likely struggling with similar love- and work-related issues as the protagonists in the two stories. Thus, while reading the faux pas stories, participants may not only have inferred what the characters unlucky in love and the character unlucky at work felt, but also shared their feelings because, to some extent, the saw bits of their own life in the lives of the fictitious characters. Future studies should investigate whether the degree to which memory for others’ life resonates with one’s own biography modulates the relation between cognitive and affective empathy (see also Batson et al., 1996).

A number of alternative interpretations to our data deserve consideration. Since in the present study recall accuracy was not experimentally manipulated, it is possible that a third variable, for example an initial empathic response while reading the life-stories, may have influenced both subsequent recall and empathic responses to social violations. In support to this hypothesis, recall accuracy correlated positively, though not significantly, with empathy scales (F scale: r = 0.35; p = 0.056; PD scale: r = 0.36; p = 0.051). However, the fact that the memory-based empathy modulation correlated with recall accuracy but not with measures of empathy suggests that it was episodic memory, not empathy, that drove the situation-specific adjustments in empathy for known individuals.

Another possibility is that, instead of episodic memory, a semantic labeling of the characters as “the unlucky in love” and “the unlucky at work,” or implicit emotional associations (see Lieberman et al., 2001) supported situation-specific empathic responses in the current study. Lieberman et al. (2001), for example, have shown that, in a choice paradigm, amnesic patients show a normal tendency to revise their attitudes to fit a counter-attitudinal behavior, in the absence of explicit memory for that behavior. Although we cannot exclude that semantic or implicit memory contributed to our results, the fact that the modulation of empathic responses tracked the amount of detail in participants’ recollection makes it unlikely that it derived merely from semantic or implicit memory.

This study has a number of limitations. First, our conclusions need to be confirmed with different materials. As Stone et al. (1998) noted, detecting a faux pas requires two things: (1) understanding that one person has knowledge that the other person is unaware of, or a mistaken belief, and (2) the empathic understanding of what kind of things someone (the victim) would find upsetting. In our experimental paradigm, episodic memory had an impact on this latter factor, tuning participants to the victims’ inner motives. One may expect, then, that if the faux pas is subtle, or it depends relatively more on the victim’s idiosyncrasies (point #2) rather then on “cold” aspects of ToM (point #1), then an impairment in episodic memory may prevent one from detecting a faux pas in the first place, having an impact on faux pas recognition accuracy. In the extreme case of the example we made in the Introduction, one would not call telling someone that he looks tremendously young for his age a faux pas without having memory for his life. Thus, future studies using more subtle social scenarios that cannot be deciphered completely within ToM systems or through abstract social knowledge would be important to test the relation between episodic memory and ToM accuracy further.

Moreover, it should be noted that the present results are limited to the healthy population we tested. In clinical populations (e.g., autistic patients, schizophrenic patients, patients with personality disorders), impairments in autobiographical memory, and ToM may co-occur and be related to each other (Corcoran and Frith, 2003; Adler et al., 2010; see Dimaggio et al., 2012 for a review). Interestingly, some therapeutic approaches for personality disorders and schizophrenia (e.g., Lysaker et al., 2007, 2011; Dimaggio and Attinà, 2012; Dimaggio et al., 2012) insist on the importance of eliciting patients’ specific memories of relevant social interactions (as opposed to resorting to overgeneralized memories), to help patients appreciate psychological causalities, and track down the mental states of the individuals involved more accurately. For example, Dimaggio and Attinà (2012) described a patient with a narcissistic personality disorder who arrived demoralized at one session reporting that he had been socially rejected by two peers at a party. By re-exploring the original event with the therapist, he was able to recall additional contextual details (“There were more of us, and we were more familiar with each other. And to tell the truth, we weren’t paying them much attention”), and consider the alternative possibility that his peers ignored him because they felt uncomfortable at the party, not because something was wrong with him (Dimaggio and Attinà, 2012, p. 931). The fact that, in clinical populations, the retrieval of past episodes promotes the explicit consideration of others’ mental states is in line with the results of the present study. One question for future research is whether such memory-driven improvements in ToM depend on accessing specific contents about past episodes, or on the repeated activation of the neural network that supports both episodic memory and cognitive empathy.

Finally, we believe that an important step toward specifying the relation between autobiographical memory and ToM, would be to adopt both subjective and objective measures of ToM (see Zaki et al., 2008, for a measure of empathic accuracy). For example, most studies, including the present study, require participants to explicitly consider the mental states of other people, but do not assess whether participants did infer these mental states correctly (Rabin et al., 2010; Spreng and Grady, 2010; Rabin and Rosenbaum, 2012). As well, self-report measures of empathy may not track with the actual ability to read, and resonate with, others’ mental states, especially in clinical populations (see Ritter et al., 2011; Dimaggio et al., 2012). This factor may also explain the lack of correlation between self-report empathy and ToM performance in the present experiment. Investigating the relation between episodic memory, and subjective as well as objective measure of ToM will reveal whether there is a causal relation between the two processes, or whether they represent two instances, not necessarily intertwined, of imagining a perspective removed from current experience.

To conclude, the present study shows that the retrieval of memories of previous episodes influences participants’ current social processing significantly, such that empathy toward the victim of a social violation is modulated by memories involving the victim that bear a resemblance to the situation she or he is currently facing. These findings suggest that understanding others’ thoughts and feelings entails integrating flexibly information about past experience and more contingent information, constructing detailed simulations of social targets and situations that preserve their uniqueness, a function we call individualized ToM.

Conflict of Interest Statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.