Autism is a neurodevelopmental disorder involving impairments in social interaction and communication, and restricted, repetitive patterns of behaviour, interests or activities (DSM-5, American Psychiatric Association
; ICD-10, World Health Organization
). One common explanation for the difficulties in social interaction is a deficit in the ability to mentalize, or Theory of Mind (ToM), the aptitude for inferring other people’s states of mind, such as intentions, beliefs, desires and wishes (Frith and Frith
). While older and more high-functioning individuals on the autism spectrum tend to pass traditional ToM tasks used with lower-functioning children, they display persistent daily difficulties understanding other people’s states of mind (Frith et al.
In order to reveal these individuals’ persistent ToM deficits, Baron-Cohen and colleagues developed an advanced ToM task based on the ability to recognise faux pas (Baron-Cohen et al.
). A faux pas was defined as a situation where “a speaker says something without considering if it is something that the listener might not want to hear or know, and which typically has negative consequences that the speaker never intended” (Baron-Cohen et al.
, p. 408). The recognition of a faux pas is considered an advanced test of ToM ability as it requires subtle social reasoning: one must be able to appreciate (a) that two protagonists might have different knowledge states and also (b) the emotional impact a statement can have on the listener (Baron-Cohen et al.
; Lee et al.
). In this way, recognition of a faux pas committed by others is closely related to recognition of embarrassment; the Oxford Dictionary defines a faux pas as “an embarrassing or tactless act or remark in a social situation” (Faux pas [Def. 1]
). By definition therefore, all faux pas statements lead to an awkward situation where one or more character is embarrassed. Indeed, all those involved in a particular situation who realise a faux pas has occurred tend to feel embarrassed: the person who committed the faux pas, the person who was affected by it, and any witnesses.
High-functioning children and adults with autism spectrum disorder (ASD) who perform well on first- and second-order false-belief tasks consistently display difficulties in recognising faux pas situations (Baron-Cohen et al.
; Zalla et al.
). In Baron-Cohen’s study, the group of children with ASD showed a tendency to under-detect faux pas in comparison to a control group (Baron-Cohen et al.
). Interestingly, adults with medial prefrontal cortex (mPFC) lesions also tend to under-detect faux pas (Stone et al.
); this region is both widely associated with ToM (Frith and Frith
) and known to be abnormally recruited by people with ASD (Gilbert et al.
). Surprisingly however, a recent study testing adults with ASD on an adapted adult version of the same faux pas task (Stone et al.
) found the opposite pattern (Zalla et al.
): adults with ASD tended to over-detect faux pas, thinking that they had occurred when in fact they had not.
In this study, we examine four possible explanations for this apparent “over-detection” of faux pas in adults. The first possibility (Hypothesis 1) is that a feature of autism might be excessive attribution of mental states, i.e. to over-mentalizing. This seems prima facie unlikely since ToM impairments are widely attributed to a
of attribution and understanding of others’ mental states, i.e. to under-mentalizing (Frith
). However, it is a possibility that should nevertheless be examined in a study of this kind. If this account were true, then the ASD participants would be likely, when presented with faux pas test-like formats, to always detect faux pas when they are present, but also over-detect, perhaps, when they are not. Moreover, their performances should be well predicted by the difficulty of the items in controls—i.e. if typically developing (TD) adults find a faux pas is easy to detect then they will detect it easily too, and similarly, if a faux pas is hard to detect, then they should also find it hard, even if they then may adopt a liberal criterion for deciding that one is present (i.e. require less evidence before deciding that a faux pas has been committed). However, their performance overall would be expected to be good, and in line with TD adults who similarly adopt a liberal criterion.
A second, putative account (Hypothesis 2) is that, as a consequence of poor mentalizing skills, adults with ASD compensate by becoming over-sensitive to embarrassment; adults with ASD are certainly capable of experiencing vicarious social pain (Paulus et al.
), although their affective responses to vicarious embarrassment may be modulated and reduced by their difficulties in understanding and integrating another person’s mental state. Even children with ASD seem to have a rather good conceptual understanding of embarrassment (Capps et al.
; Hillier and Allinson
). It is possible therefore that the combination of poor mentalizing ability plus intact awareness of embarrassment might lead an individual with ASD to be over-sensitive to potentially embarrassing situations (Hypothesis 2a). A strongly related version of this (Hypothesis 2b) is that, having been told, or having learnt through experience, that they are poor in such situations, people with ASD deliberately adopt a strategy of suspecting embarrassment potential when in doubt, but this is not due to mentalizing difficulties. These accounts both predict adoption of a liberal criterion for saying that a faux pas has been committed, but (2a) also predicts poor ability to detect faux pas when they are present. An account of these “increased sensitivity” types may also predict relatively fast reaction times (RTs) when faux pas are presented, because the ASD participants are, in effect, primed to see them.
A third possible explanation (Hypothesis 3) for “false positives” in adult ASD participants’ faux pas responses encompasses a variety of hypotheses that can be loosely grouped together as all involving social cognitive processes. One example is that knowing when a faux pas has
not been committed is a harder form of social judgement than detecting a faux pas when it has been committed. This may occur for instance if detection of faux pas proceeds through a trial-and-error process of attempting to fit a set of experience-based social schemas of “embarrassing situations” to the stimuli. Where no faux pas is depicted in the stimulus materials, the fitting or search process will on average be more extensive (i.e. because it will have to run until exhaustion) than where a faux pas is shown. If the problem that people with ASD have with performance of faux pas tests is because they have a decrement in a “social cognition” mental resource, and individual variation in this same resource is also the cause of performance differences between TD individuals, then the test items that TD adults find hard (or easy) should also be found relatively hard (or easy) by people with ASD. In other words, the mean performance or intercept may change, but the relative difficulties (as measured by accuracy) of different test items should be similar across the two populations. This hypothesis predicts that RTs should be slower when stimuli are being shown that contain no faux pas, since the exhaustive searching and problem-solving that will be required to decide that there was no faux pas will be reflected in response times.
A fourth possible explanation (Hypothesis 4) concerns non-social decision-making processes and makes a very different prediction. Specific cognitive processes are recruited when dealing with “open-ended” (or in the jargon of the field “ill-structured”) situations that are not also involved when one is dealing with well-structured problems (Burgess et al.
). Open-ended problems have a typical set of characteristics, for example, (a) there may be many ways to achieve a given aim; (b) participants have to decide for themselves what constitutes success; (c) success or failure is not clearly signalled at the time of problem-solving. It is easy to see how a test item that asks whether a faux pas has been committed when it does not actually depict one (a “non-faux pas item”) may differ along this dimension of “open-endedness” compared with an item that requires detection of a faux pas when one is depicted. For instance, when one has detected a faux pas, one can be fairly sure of the correctness of one’s response. However, when one responds that no faux pas is depicted, there will always be the possibility that one exists but it was not detected (and therefore that one should have carried on looking). The participant has to set the criteria for their decision point themselves, which is one of the characteristics of “open-endedness”. People with a diagnosis of ASD tend to be poorer at open-ended neuropsychological tasks compared with well-structured ones (White et al.
). So it is plausible that non-faux pas items may be harder for people with ASD than faux pas ones quite independently of their social content. This possibility predicts a specific pattern of results on non-faux pas items by individuals with ASD: if the problem with these items is independent of the social content of the items, then item-by-item variability in performance will not match that of TD adults who will have no difficulties dealing with open-ended situations. In this circumstance, most of the variance in TD data will reflect the difficulty of the social processing of the items, but the variance in the ASD non-faux pas items will reflect individual differences in ability to deal with open-ended situations. Thus item-by-item accuracies should be similar between different samples of the TD population, but these values should not be well predictive of item-by-item accuracies in an ASD population. Furthermore, on the faux pas items (i.e. where a faux pas is depicted) ASD participants’ item-by-item performance should be relatively closer to the TD population, since they should find items that are less “open-ended” but require a considerable depth of social processing hard in a similar way to TD populations. This account should also predict that RTs to the non-faux pas items should be slower in ASD participants than they are for faux-pas items, since it is implausible that problems dealing with the open-endedness of the non-faux pas items would not have a consequence for processing and decision speed.
These four possibilities were investigated in this study. We used a newly created version of the faux pas test. In Zalla et al.’s (
) study, the examiner sat in front of the participant and read each story aloud. The story was also placed in front of the participant so they could read along themselves and remained there throughout the reading and questioning. This procedure places a large demand upon verbal auditory skill and reading comprehension. It also perhaps places a substantial demand upon imagination and imagery. We sought to try to reduce these potentially confounding variables by using a simple cartoon-like presentation of the social scenarios.
This study investigated the understanding of vicarious embarrassment by using a novel task, the cartoon faux pas test. High-functioning adults with clinical diagnoses on the autism spectrum were compared to typically-developed adults carefully matched on age, gender, and IQ. The key findings were the following. First, the ASD participants’ performance in answering the comprehension items (that were presented in the same cartoon format as the other items) was no different from the controls either in terms of accuracy or RT. Second, if one considers the faux pas and non-faux pas items separately the ASD participants were not significantly poorer than the IQ- and age-matched controls at detecting a faux pas when one was present (a “faux pas item”), although their mean performance was slightly below that of the controls. But when the cartoon depicted a scenario where a faux pas had
not been committed, the ASD participants responded significantly more often that a faux pas had been. Third, for both groups of participants, RTs were correlated across the three different conditions (comprehension, faux pas and no-faux pas). Fourth, RTs for both groups were faster for the non faux-pas items than the faux pas items. Fifth, the ASD participants took significantly longer to give their responses in both faux pas and non-faux pas conditions, but there was no group by condition interaction. Sixth, signal detection analysis demonstrated that the ASD participants have both significantly greater difficulty knowing if a cartoon depicts a faux pas, and are also more likely to display a liberal confirmatory bias (i.e. say “yes” regardless of how good they are at detecting a faux pas if it is there). Finally, we found that while the ASD group were in good agreement with the reference control group about which comprehension and faux pas items were most difficult, they were in significantly less agreement for the non faux-pas items.
The tendency towards false positives in the non-faux pas condition found here replicates previous studies showing over-detection of embarrassment in ASD (Zalla et al.
; Hillier and Allinson
), and shows that this effect is not an artefact of the particular type of material or method used to assess faux pas detection, since the material presented here was in a different format and most of the scenarios were different from previous versions. However, this pattern of results provides constraint upon theorising about the cause of this apparent “over-detection” of faux pas in ASD participants. Since the TD participants were carefully matched for age and IQ, and the ASD participants performed as well as they did on questions testing comprehension of events depicted in the cartoons, we can dismiss any explanation that the faux pas test differences here were an artefact of differences in general cognitive ability, or ability to attend to, or perceive, the test materials.
Instead, we consider here how well the data fits four potential explanations for these results. Hypothesis 1 was that ASD participants may show excessive attributions of mental states. This explanation would be supported by a pattern of results where ASD participants show good performance on faux pas items, but poor performance on non-faux pas items, with good agreement in terms of items difficulties with TD controls. This pattern was not observed here. While the ASD participants did perform similarly to the matched controls on the faux pas items and worse on the non-faux pas items, the items difficulty analysis showed poor agreement with the reference controls for the non-faux pas items.
Hypothesis 2 proposed an over-sensitivity to embarrassment, due either to compensation for poor mentalizing (Hypothesis 2a) or as a learned or compensatory strategy without the context of poor mentalizing (Hypothesis 2b). These accounts would be most consistent with a pattern of a liberal criterion for responding “yes” as assessed by the signal detection analysis, with relatively fast RTs to faux pas items, and either poor faux pas detection in the case of Hypothesis 2a or good faux pas detection in the case of Hypothesis 2b. The ASD participants did indeed show a liberal criterion for responding yes, as shown by the signal detection analysis. However, the SDT analysis also showed poor sensitivity to faux pas, which excludes Hypothesis 2b. In terms of RT, the ASD participants were slower with both faux pas and non-faux pas items than the TD controls. However, although there was no significant group by condition interaction, it was the case that the mean difference in overall speeds between faux pas and non-faux pas items was slightly smaller for the ASD participants (Table
). So, while this result does not strongly support Hypothesis 2a, it is not inconsistent with it. In general, therefore, Hypothesis 2a is a reasonable fit to the results here.
Hypothesis 3 proposed that knowing that a faux pas has not been committed requires more subtle or detailed social judgement than detecting a faux pas when it is present. This account would be supported by good agreement between the ASD group and reference controls in terms of relative item difficulties for both faux pas and non-faux pas items, and slower RTs to non-faux pas items. This hypothesis is not supported by the results here. There was significantly less agreement between the ASD and reference controls as to the difficulty of the non-faux pas items than was found between the matched controls and the reference controls; this was not the case for the faux pas items. A typical explanation in psychometrics for such a circumstance would be that the non faux pas items are not measuring the same thing to the same degree in the two groups, and across the two sets of items. Moreover, RTs to non-faux pas items were actually faster than to the faux pas items. Thus Hypothesis 3 is not supported.
Hypothesis 4 was that deciding that something is not present is a more “open-ended” decision-making task than detecting something that is present, and that ASD participants may have a particular problem with open-ended tasks. This predicted slower RTs to the non-faux pas items and lack of agreement between the ASD participants and controls as to item difficulty for the non faux pas items. The latter finding was found here, but not the former: the ASD groups’ RTs to non-faux pas items were significantly faster than to the faux pas items. Thus this hypothesis receives only partial support.
In summary, this study has demonstrated that the phenomenon of apparent “over-detection” of embarrassment in faux pas tests by adults with diagnoses of autism spectrum conditions is found even when the materials being used differ greatly from the original versions of the test in terms of how they are presented, and even where the content of individual situations depicted is different. This phenomenon can be seen independently of issues to do with age, gender, or IQ. The hypothesis that best fits the patterns of data in this study is that the ASD participants have both a problem with detecting faux pas, and also that they adopt a liberal criterion for detection. In other words, in this study they appeared to default to saying “yes” when non-faux pas items were presented. It is noteworthy in this respect that the ASD participants did not say “yes” significantly more frequently than matched controls to the faux pas items. This suggests that it is a less likely explanation that they are showing a confirmation bias to the question presented in this format of test (“embarrassing?”) than it is that they are adopting a more liberal criterion for answering “yes”. The reason for the ASD adoption of a liberal threshold is beyond the scope of this paper. However one possibility may be that it is an adaptation to their relative inability to detect embarrassment or in response to having learned, or being told, that their differences in faux pas appreciation are likely to lead to false negatives. Thus this study may add to the debate concerning “compensatory” behaviours when considering psychometric test results in autism (see e.g. Brunsdon and Happé
Having a better understanding of the one’s own social difficulties may lead to a greater awareness of potentially embarrassing situations and, while this may lead to over-compensation, this may help individuals with autism to avoid making faux pas.