The Animal in Me: Enhancing Emotion Recognition in Adolescents with Autism Using Animal Filters

The data’s distributions (Animal Filter Pictures, SW(15) = .93, p = .273, Original Human Pictures, SW(15) = .943, p = .427, GADS SW(15) = .898, p = .089) did not significantly differ from normality. Data was examined for outliers using box plots in line with recommendations by Field (2017), and none were identified. Matched Samples t tests showed that individuals performed the emotion recognition task significantly more accurately with the Animal Filter (M = 7.33, SD = 2.093), than the Original Human pictures (M = 5.2, SD = 2.455), T(14) = 3.506, p = .003, d = 0.92, (please see Fig. 1). A simple linear regression also showed that GADS scores accounted for around 70% of the variance in emotion recognition with Orginal Human pictures (F(1,13) = 49.183, p < .001, R2 = .791, B = .263, Beta = .889, t = 7.01, p < .001), compared to only around 30% with the Animal Filter pictures. (F(1,13) = 6.392, p = .025, R2 = .33, B = .145, Beta = .574, t = 2.528, p = .025.

We also performed exploratory analyses to identify whether the above effects were driven by particular emotions. We performed five separate Fischer’s exact tests analysing hits and misses across the two presentation types (human and animal) for each emotion separately. These analyses suggested that these results were likely driven by greater performance with the animal filtered pictures compared to human for the happy and angry emotions, but not sad, surprised or afraid (see Table 3 for total number of hits and Fischer’s exact values for each emotion type).

Our results indicate that participants had a 70% success rate when identifying the emotional expressions of faces in the Animal Filter condition, and in contrast had only a 50% success rate in the Original Human condition. Participants’ GADS scores proved to be a much greater indicator of emotion recognition abilities for human rather than animal faces, though both were significantly predictive. Previous research on this topic shows that autistic individuals with mean IQ scores in the average range have improved ToM ability when social agents are non-human rather than human (Atherton and Cross 2018). This study indicates that this pattern persists throughout the wider spectrum, as individuals with autism considered to be low functioning, many of whom also possessed a co-occurring ID also show improved ToM when evaluating non-human versus human agents. Such findings are important as they allow for greater understanding of the similarities of individuals with varying levels of functional ability.

Of interest in this study is to situate results with previous work looking at performance on emotion recognition in individuals with ASD, particularly as those with more severe autism are rarely studied in the context of ToM. For instance, as previously discussed, Sucksmith et al. (2013) found that adults with ASD were impaired when identifying emotions happy, angry, on the KDEF, while those without ASD showed no difference between emotions. Of interest is that in the present study the two emotions driving the effect were also happy and angry, not afraid. In relation to the ability to recognize the animal version of ‘happy’ in our sample, Silva et al. (2015) found that children with ASD showed avoidance towards happy human faces, while approaching happy cartoon faces. Thus, it may be that positive emotions in human faces have different associations for those with ASD which affect identification, and this can be improved when changing characteristics of the stimulus, such as anthropomorphizing them, which makes them more approachable to those with ASD.

Thus, it is clear, in line with the finding that 70% of GADS score variation was explained by the human KDEF measures, that ToM is an important predictor of symptom severity and development (Jones et al. 2018). As this pattern has now been found in autistic people with co-occuring ID, as well as those with average IQ, it would be of future interest to conduct further research into the wider spectrum, and examine whether a similar pattern can be found in those with sub-clinical autistic traits levels. Exploring the stability of this effect independent of IQ may help uncover underlying, shared characteristics of an undeniably diverse group of individuals, thus improving our understanding of the core features of the autism phenotype.

While still speculative, there are several theories as to why autistic people may have differential ToM skills when decoding non-human rather than human faces. While it has been theorized that autistic people may have comparatively reduced social interests (Chevallier et al. 2012), studies such as this one and others instead suggest that social interest in autistic people may fluctuate depending on whether an agent is typically human or non-human. For instance, research shows typical face exploration, including increased attention to eye regions, in autistic samples when faces are presented as animal rather than human (Grandgeorge et al. 2016), and typical neural activation patterns when faces are presented as cartoons rather than humans (Whyte et al. 2016).

The role of motivation in ToM connects with a broader implication of this research, which is that non-human stimuli may be especially rewarding to the autistic population. A large body of work suggests that autistic people have a particular affinity and increased social responsiveness towards animals, and while experiencing anxiety during human contact, this can be ameliorated through animal contact (O’Haire et al. 2013). Examination of neural reward responses patterns indicate that autistic people find it more rewarding to view animal rather than human faces (Whyte et al. 2016).

While autistic people have been theorized to be in less need of social contact than their typically developed (TD) peers, a number of studies rebut this characterization, showing that autistic people report the same degree of social interest as those with TD (Cowart et al. 2004). However, stemming from social differences, autistic people can suffer from exclusion in peer settings (Kasari et al. 2011), and report feeling more lonely than their TD peers (Lasgaard et al. 2010). Thus, it may be possible that a resulting decrease in social self-efficacy causes autistic people to find solace and social reward in non-human agents, which may result in stronger ToM performance when the agent in question is non-human rather than human. With this in mind, it becomes clear that in order to fully understand ToM in autism it is necessary to move beyond the framework of ‘intact’ or ‘impaired.’ Instead, researchers must consider the context in which an autistic person has learned ToM, and how this has shaped their desires or beliefs about mental state processing.

It is important to note the limitations to this study. First, individuals in this study were identified by occupational therapists as having need for significant supports, and were thus educated in special educational settings befitting individuals with severe disability, and exact measures of cognitive ability such as IQ were not obtained. While this opens the interpretation of the analysis to possible confounds regarding heterogeneity in cognitive ability between participants, this is in many ways an unavoidable issue when conducting research on this portion of the ASD population. For instance, research indicates that particularly in individuals with more severe ASD related symptoms, cognitive performance shows peaks and troughs across domains, and there is no specific cognitive profile associated with ASD (Charman et al. 2011). Furthermore, individuals with ASD who also show communicative difficulties are significantly more likely to show poor IQ test performance (Hoekstra et al. 2009). Thus, it can be difficult to directly test individuals with severe ASD symptoms, such as those in this sample, in a way that reveals a true IQ score independent of communicative impairment and reflects the variability in performance across various subscales. For this reason, it was decided to instead utilize the assessments of on-site specialists with an in-depth knowledge of the participants to situate their developmental functioning within the broader autism spectrum. Additionally, it would be of interest to explore visual saccades and obtain more detailed data from eye tracking in relation to face recognition in ASD, specifically comparing visual saccade patterns when viewing human and animal faces.

Despite these limitations, this study examined ToM ability not only with regard to stimulus type, but also to devise a means to test a portion of the autism population that is not typically assessed on ToM. For instance, as discussed by Jarrold and Brock (2004), in order to isolate the specific domain of impairment relative to controls, the majority of autism research is conducted on people with autism who also have average to above average IQ and whose developmental age matches their chronological age. The limitation therein is that those with lower developmental ages, IQ scores and who possess co-occurring intellectual disabilities are not represented in autism research, and findings can not necessarily be generalized to such individuals. Thus, in this study we demonstrate a way to simplify the KDEF testing process, thus allowing for greater inclusion in facial recognition tests for those with autism, and shedding light on possible ToM mechanisms for a greater proportion of the autism spectrum.