Low-Level versus High-Level Processing

Four studies were in favor of a distinction between low-level and high-level processing in social cognition. Three of them (Buck et al., Reference Buck, Healey, Gagen, Roberts and Penn2016; Mancuso et al., Reference Mancuso, Horan, Kern and Green2011; Thaler et al., Reference Thaler, Allen, Sutton, Vertinski and Ringdahl2013) reported a similar structure: (i) a low-level factor composed of tasks assessing emotion recognition through facial expressions, vocal intonations or gestures as well as lie recognition. These tasks require answering basic social stimuli or low-level mental representations; (ii) a high-level factor composed of tasks assessing jumping to conclusions, the ability to solve interpersonal problems, to manage emotions as well as theory of mind skills, especially the ability to detect sarcasm and infer an intention. These tasks require inferring complex mental states, taking another perspective or managing subjective emotional states.

In the fourth study, Bliksted et al. (Reference Bliksted, Fagerlund, Weed, Frith and Videbech2014) did not include facial or prosodic emotion recognition tasks but provided evidence for a low versus high process distinction within theory of mind. Indeed, the ability to infer simple mental states is considered as low-level processing. High-level processing is reflected by tasks requiring understanding and accurately describing more complex mental states. The authors established links with the model of Apperly and Butterfill (Reference Apperly and Butterfill2009).

As previously highlighted by Mancuso et al. (Reference Mancuso, Horan, Kern and Green2011), this distinction between low- and high-level processing has also been hypothesized in the social neurosciences. Indeed, on the basis of knowledge about brain structure functioning, Ochsner (Reference Ochsner2008) differentiated a low-level construct, characterized by embodied simulation supported by mirror neurons, and a high-level construct, including integration of context such as in the understanding of false beliefs. Nevertheless, the role of mirror neurons in the low-level construct remains controversial. For example, in their meta-analysis, Van Overwalle and Baetens (Reference van Overwalle and Baetens2009) reported that mirror neurons are not systematically activated in tasks assessing theory of mind.

The literature provides a strong conceptual framework for this dichotomy. Several authors (Adolphs, Reference Adolphs2009, Reference Adolphs2010; Frith & Frith, Reference Frith and Frith2008) differentiate a low-level processing in social cognition that is implicit, fast, automatic, unconscious from a high-level processing, which is explicit, slow, flexible, effortful, and conscious. This corresponds to classic dual-process theories (for a recent review, see Happé, Cook, & Bird, Reference Happé, Cook and Bird2017), also known in the literature as “system 1 and system 2” (Evans, Reference Evans2008), “reflexive and reflective processing” (Lieberman, Reference Lieberman2007), “implicit and explicit processing” (Frith & Frith, Reference Frith and Frith2008), or “controlled and automatic processing” (Schneider & Shiffrin, Reference Schneider and Shiffrin1977).

The dual process system is supported by several types of evidence. Developmental research provides strong evidence for the hierarchical development of low and high processes. First, children begin to understand desires, intentions and false beliefs (Bartsch & Wellman, Reference Bartsch and Wellman1989). In middle childhood, they first acquire the ability to interpret white lies correctly, then ironic criticism, deceptive intent and finally sarcasm (Demorest, Meyer, Phelps, Gardner, & Winner, Reference Demorest, Meyer, Phelps, Gardner and Winner1984). Onishi (Reference Onishi2005) demonstrated that 18-month-old children are able to succeed in implicit false-belief tasks but not in explicit false-belief tasks until the age of 4 years.

According to Apperly and Butterfill (Reference Apperly and Butterfill2009), high-level processing requires heavy cognitive resources like language and executive functions which are acquired later, and is partially underwritten by low-level processing. Both clinical research and neuroimaging studies provide evidence for the partial separability of low- and high-level processes. For example, adults with Asperger syndrome can successfully perform explicit theory of mind tasks but fail to attribute mental states (implicit processing) spontaneously (Senju, Southgate, White, & Frith, Reference Senju, Southgate, White and Frith2009). Moreover, various patterns of activation have been reported for explicit and implicit theory of mind tasks (for a meta-analysis, see Molenberghs, Johnson, Henry, & Mattingley, Reference Molenberghs, Johnson, Henry and Mattingley2016). However, the results of this meta-analysis also suggest some common networks between these processes. Indeed, some authors are in favor of a hierarchy between low and high processing (Apperly & Butterfill, Reference Apperly and Butterfill2009; Mancuso et al., Reference Mancuso, Horan, Kern and Green2011; Onishi, Reference Onishi2005), others not (Senju et al., Reference Senju, Southgate, White and Frith2009).

Affective versus Cognitive Processing

Four studies are in favor of a distinction between the affective and cognitive components in social cognition (Bell et al., Reference Bell, Tsang, Greig and Bryson2009; Etchepare et al., Reference Etchepare, Merceron, Amieva, Cady, Roux and Prouteau2014; Mehta, Thirthalli, et al., Reference Mehta, Bhagyavathi, Thirthalli, Kumar and Gangadhar2014; Ziv et al., Reference Ziv, Leiser and Levine2011). On the one hand, the affective factor is composed of tasks requiring “emotion-driven” processes, participants having to process emotional information. On the other, the cognitive factor is always composed of at least one cognitive theory of mind task, requiring participants to infer intention or understand mental states. In cognitive theory of mind tasks, no emotional information processing is required.

This distinction is reminiscent of Adolphs (Reference Adolphs2010) model, adapted by McDonald (Reference McDonald2013). Both are theoretical models that still lack empirical evidence. In the second of the three components of this model, named “evaluation and interpretation”, the authors distinguish “hot” and “cold” social cognition. Hot social cognition refers to emotion processing, whereas cold social cognition corresponds to the ability of thinking about things from another’s point of view, including theory of mind. Hot versus cold social cognition is equivalent to affective versus cognitive components of social cognition.

Similarly, other authors have differentiated the affective and cognitive components in studies on theory of mind and on empathy. In clinical research, brain impairment studies provide strong evidence for a dissociation between affective and cognitive impairments in the two components (Shamay-Tsoory, Reference Shamay-Tsoory2015; Shamay-Tsoory & Aharon-Peretz, Reference Shamay-Tsoory and Aharon-Peretz2007; Shamay-Tsoory, Tibi-Elhanany, & Aharon-Peretz, Reference Shamay-Tsoory, Tibi-Elhanany and Aharon-Peretz2006). These dissociations are also described in autism spectrum disorder (impaired cognitive but preserved affective empathy) and psychopathy (impaired affective but preserved cognitive empathy) (Blair, Reference Blair2005; Jones, Happé, Gilbert, Burnett, & Viding, Reference Jones, Happé, Gilbert, Burnett and Viding2010; Shamay-Tsoory, Harari, Aharon-Peretz, & Levkovitz, Reference Shamay-Tsoory, Harari, Aharon-Peretz and Levkovitz2010). Neuroimaging research provides strong evidence that dissociated networks support affective and cognitive theory of mind (for a recent review, see Abu-Akel & Shamay-Tsoory, Reference Abu-Akel and Shamay-Tsoory2011), or affective and cognitive empathy (Dvash & Shamay-Tsoory, Reference Dvash and Shamay-Tsoory2014; Fan, Duncan, de Greck, & Northoff, Reference Fan, Duncan, de Greck and Northoff2011; Shamay-Tsoory, Aharon-Peretz, & Perry, Reference Shamay-Tsoory, Aharon-Peretz and Perry2009).

Of the 11 studies included in the present review, it should be underscored that three reported inconsistent results regarding the above-mentioned distinctions (low- vs. high-level processing and affective vs. cognitive processing). Indeed, Browne et al. (Reference Browne, Penn, Raykov, Pinkham, Kelsven, Buck and Harvey2016), Corbera et al. (Reference Corbera, Wexler, Ikezawa and Bell2013) and Stouten et al. (Reference Stouten, Veling, Laan, van der Helm and van der Gaag2015) reported no distinction between theory of mind and emotion perception, tasks loading on the same factor. Regarding Browne et al. (Reference Browne, Penn, Raykov, Pinkham, Kelsven, Buck and Harvey2016), this discrepancy in results may be explained by the statistical analyses that the authors used. Indeed, the study by Browne et al. (Reference Browne, Penn, Raykov, Pinkham, Kelsven, Buck and Harvey2016) was the only one to use a confirmatory factor analysis, in the assumption that measures assessing theory of mind and emotion perception would load on a unidimensional model.

Attributional Style

When included in studies, attributional style is regularly reported to be an isolated factor in social cognition (Buck et al., Reference Buck, Healey, Gagen, Roberts and Penn2016; Mancuso et al., Reference Mancuso, Horan, Kern and Green2011; Mehta, Thirthalli, et al., Reference Mehta, Thirthalli, Bhagyavathi, Keshav Kumar, Subbakrishna, Gangadhar and Keshavan2014; Stouten et al., Reference Stouten, Veling, Laan, van der Helm and van der Gaag2015). This result raises questions about the links between attributional style and other social cognition dimensions. Conceptually, attributional style describes a cognitive style related to personality, rather than a type of performance (Buck et al., Reference Buck, Healey, Gagen, Roberts and Penn2016; Healey, Bartholomeusz, & Penn, Reference Healey, Bartholomeusz and Penn2016). However, some models, such as the one suggested by the Social Cognition Psychometric Evaluation (SCOPE) study (Green et al., Reference Green, Penn, Bentall, Carpenter, Gaebel, Gur and Heinssen2008; Green, Olivier, Crawley, Penn, & Silverstein, Reference Green, Olivier, Crawley, Penn and Silverstein2005; Pinkham et al., Reference Pinkham, Penn, Green, Buck, Healey and Harvey2014), include attributional style in social cognition without explaining how it is related to other dimensions. The SCOPE model is a list of potentially relevant social cognition dimensions based on expert surveys. Nevertheless, consensus among recognized experts is considered as the least compelling level of evidence with regard to interventional research (American Psychological Association, 2006). Further research is thus needed to better understand attributional style as a component of social cognition in neuropsychological assessment, and how it is related to other dimensions.

Crossing Perspectives: Proposition of a Global Model of Social Cognition in Neuropsychology

The above-mentioned models of social cognition (e.g., low- vs. high-level, affective vs. cognitive social cognition) have received strong support from lesion studies, neuroimaging, developmental and clinical research. Some recent studies have tried to bridge the gap between them by partially merging low-level and high-level processing with affective and cognitive processing. For example, following their factor analysis, Mehta, Thirthalli, et al. (Reference Mehta, Bhagyavathi, Thirthalli, Kumar and Gangadhar2014) hypothesized that their cognitive factor of social cognition (i.e., “social–inferential ability”) could be divided into two components: a low-level (understanding goals of actions) and high-level (processing more complex judgements of others) mental state inference. Thaler et al. (Reference Thaler, Allen, Sutton, Vertinski and Ringdahl2013) also claim that their two social cognition factors (i.e., “social/emotional processing” and “theory of mind”) both support the low- versus high-level distinction of Mancuso et al. (Reference Mancuso, Horan, Kern and Green2011) and the affective versus cognitive distinction of Ziv et al. (Reference Ziv, Leiser and Levine2011).

Our review provides further support for the relevance of crossing perspectives. Indeed, in most of the studies reviewed, factor analyses consistently differentiated emotional perception (e.g., through prosody, facial emotions) and cognitive theory of mind (Bell et al., Reference Bell, Tsang, Greig and Bryson2009; Browne et al., Reference Browne, Penn, Raykov, Pinkham, Kelsven, Buck and Harvey2016; Etchepare et al., Reference Etchepare, Merceron, Amieva, Cady, Roux and Prouteau2014; Thaler et al., Reference Thaler, Allen, Sutton, Vertinski and Ringdahl2013; Ziv et al., Reference Ziv, Leiser and Levine2011). This difference could be understood as a distinction between affective and cognitive processing, but also as a distinction between low- and high-level processing.

Thus, among the social cognition tests used in clinical neuropsychology, two dimensions can be distinguished: (i) the nature of processed information: affective processing, which is mainly characterized by the processing of emotional information, and cognitive processing, which requires subjects to process information like intentions, beliefs, and/or knowledge, in which emotion is absent or secondary; (ii) the level of processed information: low-level processing, which is characterized by automatic, implicit processing, and high-level processing, which encompasses more complex cognitive processes (e.g., executive functioning), thus corresponding to more controlled, explicit and effortful processing.

Figure 2 synthesizes these assumptions in a two-dimensional model of social cognition for clinical neuropsychology, with examples of corresponding domains of assessment. For each domain, tasks are proposed. They were chosen according to the following criteria: (1) tasks were included in reviewed studies; (2) tasks are commonly used in clinical practice. In addition, we have added the Situational Features Recognition Test (SFRT) to exemplify social knowledge and the Theory Of Mind-15 (TOM-15), a task simultaneously providing assessment of first- and second-order false beliefs. These tasks were not used in the reviewed studies but are usually cited in the literature. Though based on the review results, the model needs further validation to ensure that the tasks actually correspond to these particular domains.

Fig. 2 A two-dimensional model of social cognition.Note.¹ Baron-Cohen et al. (Reference Baron-Cohen, Leslie and Frith1985); Wimmer & Perner (Reference Wimmer and Perner1983). ² Perner & Wimmer (Reference Perner and Wimmer1985). ³ Desgranges et al. (Reference Desgranges2012). ⁴ Corcoran, Mercer, & Frith (Reference Corcoran, Mercer and Frith1995). ⁵ Brunet, Sarfati, & Hardy-Baylé (Reference Brunet, Sarfati and Hardy-Baylé2003); Sarfati, Hardy-Baylé, Besche, & Widlöcher (Reference Sarfati, Hardy-Baylé, Besche and Widlöcher1997). ⁶ Baron-Cohen, O’Riordan, Stone, Jones, & Plaisted (Reference Baron-Cohen, O’Riordan, Stone, Jones and Plaisted1999); Stone, Baron-Cohen, & Knight (Reference Stone, Baron-Cohen and Knight1988). ⁷ McDonald, Flanagan, Rollins, & Kinch (Reference McDonald, Flanagan, Rollins and Kinch2003). ⁸ Lane, Quinlan, Schwartz, Walker, & Zeitlin (Reference Lane, Quinlan, Schwartz, Walker and Zeitlin1990). ⁹ Corrigan & Green (Reference Corrigan and Green1993). ¹⁰ Bell, Bryson, & Lysaker (Reference Bell, Bryson and Lysaker1997). ¹¹ Baron-Cohen, Wheelwright, Hill, Raste, & Plumb (Reference Baron-Cohen, Wheelwright, Hill, Raste and Plumb2001).

Implications for the Assessment of Social Cognition in Clinical Neuropsychology

Our results also provide new insights into the assessment and interventions that target social cognition, which is a topic of great interest in neuropsychology (Kelly, McDonald, & Frith, Reference Kelly, McDonald and Frith2017). Given the structure of social cognition, clinical assessment should ideally cover the two dimensions of the above-described model. In a treatment perspective, it is indeed crucial to distinguish between impairment tendencies in low-level or high-level processing and affective or cognitive processing. According to clinical neuropsychology guidelines, results should be crossed and performances compared to define and isolate difficulties in specific social cognitive domains (Brown, Del Dotto, Fisk, Taylor, & Breslau, Reference Brown, Del Dotto, Fisk, Taylor and Breslau1993; Lezak, Reference Lezak2004; Seron & Van der Linden, Reference Seron and Van der Linden2000).

Similarly, performances in social cognition also need to be crossed with other domains of cognitive performance. There is indeed strong evidence that neurocognition and social cognition are both related and separate constructs (Addington & Piskulic, Reference Addington and Piskulic2013; van Hooren et al., Reference van Hooren, Versmissen, Janssen, Myin-Germeys, à Campo, Mengelers and Krabbendam2008; Ventura et al., Reference Ventura, Wood and Hellemann2013). For example, Mehta, Bhagyavathi, Thirthalli, Kumar, and Gangadhar (Reference Mehta, Bhagyavathi, Thirthalli, Kumar and Gangadhar2014) reported that cognitive flexibility, as well as verbal and visual memories, predicted faux-pas recognition in a schizophrenia sample. This point has major implications for the methodology of social cognition assessment. First, neurocognitive deficits should be identified before assessment of social cognition. If severe neurocognitive impairments are identified, social cognition assessment should be adapted by proposing tasks with less cognitive load. If more discrete neurocognitive dysfunctions are identified, then caution is required when making conclusions about the assessment of social cognition. In any case, assessing neurocognition before social cognition avoids confounding social cognition difficulties with more basic neurocognitive ones.

In the same therapeutic perspective, the assessment of social cognition can be advantageously complemented by a self-rated questionnaire like the OSCARS (Observable Social Cognition: A Rating Scale) (Healey et al., Reference Healey, Combs, Gibson, Keefe, Roberts and Penn2015). Few tools have received satisfactory validation to date. Self-questionnaires allow subjective perception about one’s social cognitive abilities to be measured and may go beyond the lack of sensitivity of objective testing when exploring potential targets for cognitive intervention.

Limitations and Perspectives

The first limitation of this study is the heterogeneity in the factor analyses in the studies reviewed, which limits their comparability. Indeed, of the 11 studies included, three different factor analyses (principal components analysis, exploratory or confirmatory factor analysis), two various extraction methods (maximum likelihood, principal axis factor), and four different rotation methods (quartimax, oblique, promax, varimax) were used. Moreover, several studies did not report the extraction and/or rotation method used, even though discrepancies in extraction or rotation methods can lead to discrepancies in results (Tabachnick & Fidell, Reference Tabachnick and Fidell2013).

The present review is also indirectly limited by several characteristics of the literature targeted, in which the results obtained should be considered with caution. First, heterogeneity in tasks used and constructs assessed limits the comparability of the results. Second, social cognition is often measured by tasks that have inadequate or unknown psychometric properties. In particular, studies investigating the validity of these tools in the general population remain rare. Further research is necessary to provide adapted norms and evidence-based interpretation of the difficulties encountered. Some tasks were included more often than others, such as the Reading the Mind in the Eyes Test (RMET)/Eyes Test (Browne et al., Reference Browne, Penn, Raykov, Pinkham, Kelsven, Buck and Harvey2016; Etchepare et al., Reference Etchepare, Merceron, Amieva, Cady, Roux and Prouteau2014; Thaler et al., Reference Thaler, Allen, Sutton, Vertinski and Ringdahl2013) or the Bell Lysaker Emotion Recognition Test (BLERT) (Bell et al., Reference Bell, Tsang, Greig and Bryson2009; Browne et al., Reference Browne, Penn, Raykov, Pinkham, Kelsven, Buck and Harvey2016; Corbera et al., Reference Corbera, Wexler, Ikezawa and Bell2013; Thaler et al., Reference Thaler, Allen, Sutton, Vertinski and Ringdahl2013) for assessing emotion perception, and The Awareness of Social Inferences Test (TASIT) (Bliksted et al., Reference Bliksted, Fagerlund, Weed, Frith and Videbech2014; Browne et al., Reference Browne, Penn, Raykov, Pinkham, Kelsven, Buck and Harvey2016; Buck et al., Reference Buck, Healey, Gagen, Roberts and Penn2016; Mancuso et al., Reference Mancuso, Horan, Kern and Green2011) or the Hinting Task (Bell et al., Reference Bell, Tsang, Greig and Bryson2009; Bliksted et al., Reference Bliksted, Fagerlund, Weed, Frith and Videbech2014; Browne et al., Reference Browne, Penn, Raykov, Pinkham, Kelsven, Buck and Harvey2016; Buck et al., Reference Buck, Healey, Gagen, Roberts and Penn2016; Corbera et al., Reference Corbera, Wexler, Ikezawa and Bell2013; Stouten et al., Reference Stouten, Veling, Laan, van der Helm and van der Gaag2015; Thaler et al., Reference Thaler, Allen, Sutton, Vertinski and Ringdahl2013) for assessing theory of mind.

Recently, Pinkham, Penn, Green, and Harvey (Reference Pinkham, Penn, Green and Harvey2016) and Ludwig, Pinkham, Harvey, Kelsven, and Penn (Reference Ludwig, Pinkham, Harvey, Kelsven and Penn2017) investigated several psychometric properties of eight social cognition tasks in 179 and 38 schizophrenia participants and 104 and 39 control participants, respectively. Currently, only the Hinting Task has displayed satisfactory psychometric properties so the validity of these tasks should be further investigated. A third limitation stems from small sample sizes, thereby limiting the generalizability of the results obtained and generating confounding effects of sociodemographic variables. As Bakker, Hartgerink, Wicherts, & van der Maas (Reference Bakker, Hartgerink, Wicherts and van der Maas2016) recently recalled, a formal power analysis should be more systematically performed before planning any study to enhance the robustness of results. Moreover, the heterogeneity of study samples suffering from different conditions (i.e., bipolar disorder, schizophrenia spectrum disorders, first-episode vs. chronic psychosis) could lead to discrepancies in results. Such discrepancies could reflect pathological processes rather than the structure of social cognition.

Similarly, several assumptions can be made about differences in the structure of social cognition between clinical populations and healthy controls. In our review, two studies highlighted a difference between a clinical population (schizophrenia) and healthy controls. The factor analysis of Buck et al. (Reference Buck, Healey, Gagen, Roberts and Penn2016) distinguished low-level and high-level processing in the healthy control group contrary to the schizophrenia group where these two factors loaded on the same one. According to these authors, social cognition abilities were more differentiated in healthy controls, schizophrenia participants suffering from a general performance deficit.

On the contrary, Mehta, Thirthalli, et al. (Reference Mehta, Bhagyavathi, Thirthalli, Kumar and Gangadhar2014) found a different factor structure in their study: three factors for the schizophrenia group and two for healthy controls. Indeed, the component “externalizing bias” loaded as a distinct factor in the schizophrenia group. According to these authors, social cognition abilities in schizophrenia and in healthy controls share more similarities than differences. Further research on this issue is, therefore, required. Finally, most of the studies reviewed were conducted in schizophrenia spectrum disorders. Further research remains necessary before extending the conclusions to other clinical populations.