The role of linguistic and cognitive factors in emotion recognition difficulties in children with ASD, ADHD or DLD

Background: Many children with neurodevelopmental disorders such as ASD (Autism Spectrum Disorder), ADHD (Attention Deficit and Hyperactivity Disorder) or DLD (Developmental Language Disorder) have difficulty recognising and understanding emotions. However, the reasons for these difficulties are currently not well understood. Aims: The objective of this study was to compare the emotion recognition skills of children with neurodevelopmental disorders and compare these children’s skills to those of their typically developing age peers. The second aim was to identify the role of underlying factors in predicting emotion recognition skills. Methods & Procedures: The 6–10-year-old children (N=50) who participated in this study had either ASD, ADHD or DLD and difficulties recognising emotions from face and/or in voice. Typically developing age peers (N=106) served as controls. Children’s skills were tested using six forced-choice tasks with emotional nonsense words, meaningful emotional sentences, the FEFA 2 test, photographs, video clips and a task in which facial expressions and tones of voice had to be matched. Expressive vocabulary, rapid serial naming, auditory and visual working memory and Theory of Mind skills were explored as possible explanatory factors of the emotion recognition difficulties of the diagnosed children. Outcomes & Results: Children with ASD, ADHD or DLD did not significantly differ from each other in their linguistic or cognitive skills. Moreover, there were only minor differences between children with these diagnoses in recognising facial expressions and emotional tone of voice and matching the two. The only significant difference was that children with ADHD recognised facial expressions in photographs better than children with DLD. The participants with diagnoses scored significantly lower than the controls in all but one emotion recognition tasks presented. According to the linear regression analysis, first order Theory of Mind skills predicted the delay relative to typical development in the recognition of facial expressions in the FEFA 2 test, and expressive vocabulary and working memory skills together predicted the delay in the recognition of emotions in the matching task. Conclusions & Implications: Children with ASD, ADHD or DLD showed very similar emotion recognition skills and were also found to be significantly delayed in their development of these skills. Some predictive factors related to linguistic and cognitive skills were found for these difficulties. Information about impaired emotion recognition and underlying linguistic and cognitive skills helps to select intervention procedures. Without this information, therapy might unnecessarily focus on only symptoms.


Introduction
Emotion recognition skills are important parts of communication and children's socialemotional development.In contrast to typically developing children, children with neurodevelopmental disorders often face difficulties in the development of emotion recognition skills which are linked to the development of social competence, peer-relations and self-esteem (Evers et al., 2015).Emotion recognition abilities are complex set of skills, which are also affected by an individuals linguistic and cognitive abilities (Keltner et al., 2014).
The present study explores the emotion recognition abilities of children with ASD (Autism Spectrum Disorders), ADHD (Attention Deficit Hyperactivity Disorder) and DLD (Developmental Language Disorder).These disorders have similar etiological background consisting of genetic or hereditary factors (Onnis, Truzzi & Ma, 2018;Simpson et al., 2015;Smoller et al., 2013), and they quite often occur together, which may suggest at least partly shared neurobiological and etiological background (Rommelse et al., 2010; see also Bishop, 2010).The growing understanding of the similarities and differences in these three diagnostic groups has also affected the diagnostic criteria of ASD, ADHD and DLD (or Specific Language Disorder, SLI) as they have been undergoing change in both the coming ICD-11 and the recently published DSM-5 (APA, 2013) classifications.Additionally, increased attention has recently been paid to the occurrence of a broad spectrum of other difficulties, such as attention problems, social impairment and behavioural and emotional disorders which sometimes accompany DLD (Loucas et al. 2008;Taylor, et al., 2015).As groups, children with ASD, ADHD and DLD have all been found to experience overlapping difficulties not only in neurodevelopmental domains (e.g., Geurts & Embrechts, 2008) but also in emotion recognition skills.For example, Waddington and colleagues (2018) found that children with ADHD had problems in recognition of facial expressions and tones of voice to the same extent as children with ASD.Studies where overlaps in emotion recognition difficulties between DLD and ASD or ADHD have been demonstrated are still sparse, although descriptions of difficulties within these diagnostic groups exist.For example, a fairly recent study by Taylor and colleagues (2015) showed that children with ASD (N=29) and DLD (N=18) both performed poorly in recognising emotions from face or voice.They also stress the importance of language ability in affective understanding.These cross-domain relationships are important to be explored further to create a more comprehensive picture of neurodevelopmental disorders.

Emotion recognition difficulties in children with ASD, ADHD and DLD
Recognising emotions from face, voice or social context is difficult for many children with ASD, and evidence of poor facial emotion recognition is found in several studies (e.g.Golan et al., 2008;Leung et al., 2013).Fewer studies have focused on children with ASD recognising emotions from voice, but some researchers have found impairment in their ability to process emotional prosody (e.g.Demopoulos et al., 2013) and interpreting them in social contexts.
Like children with ASD, children with ADHD have demonstrated difficulties with emotion recognition (Demopoulos et al., 2013), although contrary findings also exist (see a review of Borhani & Nejati, 2018).In ADHD, inattention, impulsivity and hyperactivity hamper a child's development.Differing views have been presented favouring either impaired executive functions (Sinzig et al., 2008) or linguistic problems (Geurts & Embrechts, 2008).Of the studies reviewed by Borhani and Nejati (2018), 18 out of 26 had found facial emotion recognition problems in children and adults with ADHD.Moreover, children with DLD have also been demonstrated to have emotion recognition difficulties (Boucher, Lewis & Collis, 2000;Taylor et al., 2015).Compared to children with ASD, their skills have been far less researched, but some findings have been presented on the nature of their social-emotional skills (Botting & Conti-Ramsden, 2008).There is evidence that children and adolescents with DLD have difficulty recognising both simple (such as joy or anger) and complex (such as embarrassment) emotions (Boucher, Lewis & Collis, 2000;Taylor et al., 2015).Children with DLD often have difficulties with social competence, and this has usually been thought to result from poor communication skills (Spackman et al. (2005).Spackman et al. (2005), however, also argue that difficulties the children with DLD have with social competence, are partly due to emotion recognition difficulties.
In the recent study of Taylor et al. (2015), typically developing children (N=61) were compared to children with ASD (N=29) and children with DLD (N=18).They found that the 4 to 11-year-old children with ASD and DLD performed significantly worse than typically developing children at recognising emotions from both face and voice.The authors concluded that emotion recognition difficulties were specifically due to poor linguistic skills in both of these groups.Boucher, Lewis and Collis (2000) found that 9-year-old children with DLD (N=19) recognised emotions from faces and named emotions even worse than same-age children with ASD (N=19).

Linguistic and cognitive abilities and emotion recognition
Based on what is already known about emotion recognition skills, the strongest evidence that currently exists concerns associations with Theory of Mind (ToM), language and working memory skills.ToM skills have been found to be delayed in children with ASD and some children with ADHD and DLD (Loukusa et al., 2014).Furthermore, ToM skills are strongly and directly associated with emotion recognition problems, without mediating factors (Baron-Cohen et al., 1985).Language has been shown to be an important mediating factor in emotion recognition skills (Loucas et al. 2008).Alloway and colleagues (2009) noticed that there is also evidence of shared deficits among children with ASD, ADHD and DLD in, for example, attention skills, short-term memory and self-regulation skills.They found that working memory difficulties were extremely prevalent in all these diagnostic groups, with the children with DLD having the most severe problems.
In sum, it is still unclear which underlying factors can explain the emotion recognition difficulties in these three disorders and to what extent different language and cognitive functions are needed for emotion recognition even in typical development.To the best of our knowledge, no other study has included both ASD, ADHD and DLD and explored the interrelationships between the linguistic and cognitive factors and emotion recognition as extensively as we aim to do in this study.

Aims of the study
We aimed to determine the level of emotion recognition skills, their differences from those of typically developing children and the possible underlying linguistic and cognitive factors of these difficulties in children with ASD, ADHD and DLD.

Participants
Children who had ASD, ADHD or DLD (N=50) and were 6 to 10 years old took part this study (Table 1).They were recruited from both northern and southern Finland through hospitals, privately practising speech and language and occupational therapists, psychologists, schools and parent organisations.From here on, these three diagnostic groups are together called the clinical group.Additionally, a group of typically developing (TD) children 6 to 10 years old (N=106), 20 to 22 children in each age group, were recruited to serve as controls.They were recruited from day-care centres and schools and were judged to be typically developing based on the report of their parents (questionnaire filled out).The study protocol was approved by the Ethical Committee of the Northern Ostrobothnia Hospital District, and written informed consent was obtained from the parents.Written consent was also obtained from those children who were able to read and sign the consent form.
The inclusion criteria for the clinical group were: 1) diagnosis of ASD, ADHD or DLD, 2) difficulties recognising emotions from face and/or in voice as reported by parents or a professional (speech and language therapist, occupational therapist, psychologist, kindergarten teacher or teacher), 3) age between 6 and 10 years, 4) non-verbal IQ over 85, 5) monolingual Finnish-speaking family and 6) child's vision, hearing, motor and attention skills sufficient for test situations.Some participants had the diagnostic label of SLI as they were diagnosed according to the ICD-10 (National Institute for Health and Welfare, 2011;WHO, 1992), which is still in use.However, the up-to-date label of DLD will be used here, as this term does not exclude cognitive, motor or emotional difficulties and will also be included in the upcoming ICD-11 classification (Bishop et al., 2017).Per the exclusion criteria, the children in the clinical group were not allowed to have any psychiatric diagnoses, such as depression.Information needed to fulfil the inclusion criteria and detailed information about their child's diagnosis and other issues was gathered from the parents using a questionnaire.Demographic characteristics of the children can be found in Table 1.The mean age of the children with one of the diagnoses was 8.02 years, which matched the mean age of the TD children.Neither the age of the children nor the male-female ratio differed significantly between the three diagnostic groups.In the whole clinical group (N=50), the proportion of males was significantly higher than in the TD 106 age peers (Fisher's Exact test, p<0.001).
Of the children with neurodevelopmental disorders, 30 had a single diagnosis but 20 children had at least one comorbid diagnosis in addition to that which we judged to be the primary one (see description later).Of these 20 children, six had been diagnosed with both ASD and ADHD, seven had ADHD and DLD, three had ASD and DLD and four had all three diagnoses.Therefore, as many as 48% of the participants had comorbid diagnoses despite the diagnostic criteria of SLI (nowadays increasingly called DLD) in ICD-10 (National Institute of Health and Welfare Finland, 2011;WHO, 1992).According to ICD-10 a child's language difficulties cannot be explained by other disorders (see also Rapin & Allen, 1988).
Most (N=17) of the 20 children with ASD had AS (Asperger's Syndrome, code F84.5 in ICD-10) as their diagnosis, and three had diagnosis of Pervasive developmental disorders, unspecified (F84.9).Three of the children grouped under ADHD had ADD (Attention Deficit Disorder), but since ADD is categorised under the diagnostic code of ADHD (F90.0),we will use ADHD to cover these cases.In the case of comorbid diagnoses, the primary diagnosis was determined based on a child's symptom profile in the medical records provided by parents.If there was no additional information available, the decision was based on scientific and clinical literature.For example, if diagnoses of both ASD and ADHD had been given to the same child, complying with the clinical routine used in Finland (Moilanen, 2011), ASD was determined to be the primary one.If a child had medication for ADHD and he or she also had DLD, ADHD was judged to be more severe and therefore the primary diagnosis.Additionally, seven of the participants also had subsidiary diagnoses such as a motor function disorder or Tourette´s syndrome.Out of the children with ADHD, 12 out of 17 had medication for their ADHD symptoms.There were 35 children (70%) in the clinical group who had received or had ongoing speech therapy during the time of data collection, and 15 children (30%), who did not.

Measurements
Data was collected by formal testing using vocabulary, rapid serial naming, working memory, ToM and emotion recognition tasks.All children with a diagnosis were tested individually, and the assessments were both audio-and video-recorded for purposes of scoring.To avoid testing fatigue, the test sessions were usually scheduled to take place in two consecutive days.
The same was applied to the typically developing 7-to 8-year-olds who served as controls, but for the sake of time-efficiency, all the typically developing 9-to 10-year-olds were tested in groups.This procedure was judged to give valid and unbiased results as we checked that the emotion recognition results of the 7-and 8-year-old TD children (N=42) tested individually in our study did not significantly differ from those of 7-and 8-year-old TD children (N=43) tested in groups (these data have been collected earlier and are reported elsewhere).During group testing, stimuli were presented using a computer, loudspeakers and a data projector, and children marked down their responses on paper.

Linguistic and cognitive tasks
The linguistic skills of the participants with neurodevelopmental disorders were explored in the domain of expressive vocabulary by using the validated Finnish version (Laine et al., 1997) of the Boston Naming Test (Kaplan et al., 1983).Rapid automatised naming was tested by using two subtests of the validated Rapid serial naming test (Ahonen et al., 1999), that is, for the most part, based on the Rapid Automatized Naming Test (RAN; Denckla & Rudel, 1974).Delay relative to typical development was calculated by using the results of the validation samples documented in the test manuals.Two-category delay variables were formed for the scores of the Boston Naming Test and the two RAN subtests to separate children performing at their age level from those who performed more than 1 SD below it.This categorization of the data was done because no closer information for relating the test results obtained to typical development were given; only mean and SDs for different age groups were available in the test manual of the Finnish version of the Boston Naming Test and RAN.The same -1 SD criterion has also been used by Taylor and colleagues (2015) and we obtained comparability by using the same criterion in our study.
Memory skills were assessed using both the auditory and visual short-term sequential memory tasks in the two subtests of the validated Finnish version (Kuusinen & Blåfield, 1974) of the Illinois Test of Psycholinguistic Abilities (ITPA) (Kirk et al., 1968).Delay in months relative to age peers was determined based on the mean scaled scores representing typical development at each age level documented in the test manual (Kuusinen & Blåfield, 1974).
Theory of Mind skills were assessed with first-and second-order false belief tasks.The Sally-Anne Test (Baron-Cohen et al., 1985) was used as the first-order false belief task.To help to diminish the memory load, a girl and a boy doll were used, as suggested by Doherty (2009), as opposed to the original set-up in which two girl dolls are used.The Ice Cream Van story based on Perner and Wimmer (1985) was used as the second-order ToM false belief task, with some modifications made to shorten and simplify it.A justification question was also presented requiring the child to provide reasons for his or her answer to the second-order ToM question.All ToM tasks were scored as pass or fail.

Emotion recognition tasks
Emotion recognition skills were assessed by using forced-choice tasks of facial expression and tone of voice recognition, and matching them.
In the emotional nonsense word task, children listened to 18 items comprising nonsense words "paappa", "piippi" and "paippi", either as single words (in 9 items) or (also in 9 items) embedded in a linguistically neutral carrier phrase ("Now I say…"), with the prosody of the carrier phrase matching the target tone of voice.Nonsense words representing joy, anger and sadness all randomly occurred 6 times among the 18 items.In the next task, children listened to three-to four-word sentences in which the linguistic content complied with the emotion with which the sentences were spoken (e.g."Don't come here!"with an angry voice).With its 11 items, this task contained joy, anger and sadness, all occurring twice as targets, and fear, surprise, disgust, shame and neutral tones of voice occurring once as targets.
The  (Bölte et al., 2002).Additionaly, a set of eight photographs depicting eight different emotions (the six basic emotions, and ashamed and neutral expressions), posed by two children and two adult professional actors, was constructed and shown with four verbal labels spoken by the test administrator as response choices.To test facial emotion recognition skills using dynamic input, a set of eight video clips of four seconds each (again, comprising the six basic emotions, and ashamed and neutral expressions) was created and shown with four verbal labels as response choices.In all clips but the one illustrating a neutral facial expression, the expression developed from neutral into the target emotion.
A task in which facial expressions and tone of voice had to be matched comprised 11 different items (the six basic emotions, and ashamed and neutral expressions).Children needed to point at a facial expression with which they thought the sentence they heard matched.All the emotions were posed and the sentences were spoken by two speech and language therapists.All the emotion recognition tasks were also conducted on the controls, 106 TD children.

Statistical analyses
SPSS for Windows (versions 24 and 25) was used for the statistical analyses.Because tasks of different lengths and maximum scores were used in emotion recognition testing, their results were expressed as percent performance to allow comparability.Additionally, for the emotion recognition tasks, a delay in relation to TD age peers was also calculated and used in scoring.This was done to further obtain comparability between the different tests and tasks and children of varying ages and uncover the emotion recognition profile of the children in different diagnostic groups.Delay variables were also needed in the linear regression analysis.In all emotion recognition tasks, the TD sample was used in forming the delay variables of the clinical group.
Reliability of the measurements was explored with Pearson's correlation coefficient and Cronbach's alpha.Differences in the ages between different child groups and whether emotion recognition ability differed between children with single diagnosis and children with at least one comorbid diagnosis were investigated with Independent Sample T Test.One-way ANOVA, Independent Sample and Paired Sample T Tests, Chi Square and Fisher's Exact Test were used to compare linguistic and cognitive test results and the emotion recognition results between the different diagnostic groups and between children with diagnoses and typical development.
Linear regression analysis was used to explore which combinations of linguistic and cognitive factors would predict the level of children's emotion recognition skills.Three tasksemotional nonsense words, FEFA 2 test of facial expressions and matching facial expressions and tone of voice-were chosen as dependents.All measured linguistic and cognitive factors served as independent variables to find the best possible model for predicting emotion recognition skills.Because of the sample size (N=50), only models with a maximum of two independent variables were used.The diagnosis of DLD was used as a reference group because they performed most poorly.Preliminary analysis was performed to ensure that the assumptions of normality were met.A factor was left in the regression model if it had p<0.05 or a significant impact on the model's R 2 value (>10% change).Given the fairly large number of statistical tests performed, p-values should be interpreted with caution.

Results
The linguistic, cognitive and emotion recognition skills in children with ASD, ADHD and DLD (N=50) were explored.Depending on the task or test, between 39 and 50 children accomplished each task.Missing data was caused by children's co-operation skills and children not yet knowing numbers and letters to be named in one RAN subtest.However, we perceive our results to be valid without any data imputation because the few (at maximum, four per diagnostic group) missing results were evenly or randomly spread across the three diagnostic groups.

Linguistic and cognitive skills
The mean raw scores and the number of children performing more poorly than 1 SD below age level in the linguistic and cognitive tasks can be found in Table 2. Notes: Boston Naming Test = expressive vocabulary naming test (Laine et al., 1997;Kaplan et al., 1983); RAN = Rapid Automatised Naming Test (Ahonen et al., 1999;Denckla & Rudel, 1974); ITPA = The Illinois Test of Psycholinguistic Abilities (Kuusinen & Blåfield, 1974;Kirk et al., 1968) We wanted to know if the linguistic and cognitive skills in the three diagnostic groups would differ from the group of TD children.Because the vocabulary and rapid serial naming skills of the TD children serving as controls in this study were not tested, no statistical testing could be performed using their results.Instead, we used age norms from the test manuals or references from Loukusa (2007).Overall, ANOVA showed that there was no difference between the diagnostic groups in the raw scores compared to age norms in RAN time subtest "Objects" (F (2,38)= 0.776, p=0.467), in RAN time "Colours, numbers and letters" (F (2,36)= 1.504, p=0.236), or in ITPA auditory (F (2,44)= 1.750, p=0.186) or visual (F (2,44)= 0.627, p=0.539) short term memory.Additionally, raw scores of the Boston Naming Test (F (2,47)= 3.207, p=0.049) were not, after Bonferroni correction, significantly different between the three groups of children with neurodevelopmental disorders.
Compared to age norms or age references, children with DLD were significantly delayed in their expressive vocabulary and in their ability to rapidly name objects (Table 3).All the three diagnostic groups were significantly delayed in their auditory short term memory skills and children with DLD also in their visual short term memory skills.
For the ToM tasks, the clinical group was found to both pass the ToM 1 task significantly less often than the 106 TD controls (χ² (3)=12.631;p=0.006), and to less often give the right answer to the ToM 2 question (χ² (3)=8.277;p=0.041).Specifically, children in both the group of children with ADHD (Fisher's Exact Test, p=0.027) and DLD (Fisher's Exact Test, p=0.007) passed the ToM 1 task less often than their TD age peers and children with ADHD could also less often give the correct ToM 2 explanation (Fisher's Exact Test, p=0.028).

Emotion recognition skills
The tasks of recognising tone of voice in nonsense words and in meaningful sentences were found to correlate significantly (Pearson's correlation coefficient 0.334, p=0.018) in the clinical group.However, Cronbach's alpha was low (0.498).Low alpha level was probably due to low variability in the task of meaningful sentences; 17 out of 50 children scored at maximum, the mean accuracy of the clinical group was 85% (SD 16.837) and median 91% (Interquartile Range 20).Instead, recognition of facial expressions correlated significantly between the FEFA 2 test, the task containing photographs (Pearson's correlation coefficient 0.634, p<0.001) and the task containing video clips (r=0.590,p<0.001), and the Cronbach's alpha indicated good reliability (0.834).
ANOVA showed that the performance of the children with a diagnosis was significantly below that of the controls in all emotion recognition tasks except in the ability to match emotion input from face and voice (Table 4), with the recognition ability profile across different tasks being similar in all diagnostic groups.When looking at the mean percent performance (raw scores), one difference was found between the diagnostic groups (F (2,46)= 4.407, p=0.024); children with ADHD (mean 77.4%, SD 15) performed significantly better than children with DLD (mean 60.9%, SD 13) in the recognition of facial expressions from photographs task (p=0.020).
The largest mean difference between the clinical group compared to typically developing children was found in recognition of emotions from photographs (11 percent units, F (3,150)= 8.470, p<0.001), and the smallest difference was found in recognition of emotions from meaningful sentences (five percent units, F (3,149)= 4.044, p=0.008).Notes: Nonsense words = recognition of an emotion in a tone of voice in nonsense words or embedded in a carrier sentence; meaningful sentences = recognition of an emotion from a meaningful sentence; FEFA 2 test = The Frankfurt Test and Training of Facial Affect Recognition, (Bölte et al., 2013); photographs = eight photographs depicting eight different emotions; video clips = eight video clips depicting the same emotions as the photographs; matching task = matching facial expressions with the same tone of voice.
Because delay variables were needed to be formed for the linear regression analysis (see later), all the emotion recognition variables presented from hereon are delay variables (clinical group was compared to the group of TD controls), except in the raw score results shown in Table 4.Although there were proportionally clearly more females in the TD control group (56%) than in the clinical group (18%), comparisons in emotion recognition were seen to be relevant because there was no significant difference in the results in any of the emotion recognition tasks between the TD females and males (p-values ranged from 0.659 to 0.877).
A subgroup analysis using Independent Sample T Test did not reveal any significant differences (t(48)= -1.249, p=0.223 to t(47)= 0.112, p=0.911) in the emotion recognition skills between the children with a single diagnosis (N=30) and those with comorbid diagnoses (N=20).Then we looked at the possible differences in emotion recognition abilities as the function of the diagnosis of ASD, ADHD and DLD.When the diagnostic groups were compared to the group of TD controls (using delay variables), ANOVA with Bonferroni correction as the post hoc analysis did not indicate any differences between the three diagnostic groups in the six emotion recognition tasks (F (2,47)= 2.519, p=0.91 to F (2,47)= 0.048, p=0.953).

Interrelationships between linguistic, cognitive and emotion recognition skills
Linear regression analysis was applied to determine which linguistic and cognitive factors of their combinations would best predict the selected emotion recognition abilities.Since there were neither differences between the children with single or comorbid diagnoses nor widescale differences between the diagnoses of ASD, ADHD or DLD in emotion recognition skills, the whole clinical group (N=50) was used as one in the linear regression analysis.
The nonsense word task, the FEFA 2 test, and the matching task were chosen as the emotion recognition variables in the regression analyses, because they had the largest number of items and they represent both facial and vocal emotion recognition skills.The variables were entered as delay variables (in comparison to the TD children) in the regression models.As the children with DLD performed slightly more poorly (although non-significantly) than children with ASD or ADHD in the above mentioned three tasks, the DLD group was used as the constant (reference group) in the regression analyses.
No model had significant predictive value for the nonsense word task.The best predictive models for the FEFA 2 test and the matching task are found in Table 5. Passing the ToM 1 task was a significant predictor of the FEFA 2 test results contributing to, on average, a smaller delay by eight percent units compared to those not passing the task.In this same model, children with ADHD performed significantly better in the FEFA 2 test than the children with DLD.Children with ADHD and ASD had, on average, nine and five percent units smaller delays in FEFA 2 scores compared to children with DLD, respectively.The delay was three percent units larger in males than in females, and it decreased by two percent units for every increasing age year, but neither sex nor age were significant predictors of FEFA 2. This model predicted 17.9% (R 2 =0.179) of the variation in the recognition of facial expressions in the FEFA 2 test.
The model in Table 5 predicted 15.9% of the variation in the matching task (R 2 =0.159), with the Boston Naming Test being a marginally significant predictor (p=0.05).Those children who scored more than 1 SD below their age level in the expressive vocabulary test had a mean of 10 percent units larger delay in the matching task than children with age-appropriate expressive vocabulary.The ITPA auditory and visual working memory subtests themselves were not significant predictors, but when entered into the model, the delay in the Boston Naming Test scores and either of the ITPA subtests together predicted the scores of the matching task.Age was not a significant predictor in the matching task.

Discussion
The main findings of this study indicate that children with ASD, ADHD or DLD had shared difficulties in all emotion recognition tasks.They were significantly delayed in emotion recognition skills compared to TD age peers in all tasks measured.No significant difference was found between the three diagnostic groups other than that of the children with ADHD, who had better percent scores than children with DLD in the recognition of facial expressions from photographs task.Furthermore, there was no significant difference in children's emotion recognition skills depending on whether the child had a single diagnosis or comorbid diagnoses.
Our regression analysis showed that the diagnosis and the first-order ToM false belief skill predicted children's delays in facial emotion recognition in the FEFA 2 "Faces" subtest, with children with ADHD having the smallest delay.Additionally, the delay in expressive vocabulary measured with the Boston Naming Test and delay in either ITPA auditory or visual working memory subtest scores together predicted the degree of delay in the task in which facial expressions and tone of voice were matched with each other.Similar findings have been found when the emotion recognition skills of children with ASD and DLD have been compared with each other (e.g.Boucher, Lewis & Collis, 2000;Golan et al., 2008;Taylor et al., 2015), but prior to our study, no research has compared these difficulties in ASD, ADHD and DLD together.
There were far fewer predictive factors between linguistic and cognitive factors and emotion recognition abilities than we expected to find.There could be several reasons for this.First, of the factors we could test statistically, there were no striking differences between the children with neurodevelopmental disorders and the TD children in the linguistic and cognitive tasks except in the ToM 1 task and in answering the ToM 2 justification question.On that basis, linguistic and cognitive skills cannot be expected to be strong explanatory factors of difficulties in emotion recognition skills in the present data.Second, differences in emotion recognition skills between children with diagnoses and TD children were not large; their means differed only from five to eleven percent units.It is possible that additional explanatory factors could be found in children with more severe symptom profiles.The diagnostic groups were also relatively small in size (N=13-20).Thus, the regression analysis only allowed two independent variables to be entered into each model at one time.The best predictive models explained up to 17.9% of the variation in emotion recognition skills, which is a moderate figure and typical of studies on human behaviour.
Language has been found to be an important factor in emotion recognition skills (Boucher, Lewis & Collis, 2000;Spackman et al., 2005;Taylor et al., 2015).In the present study, 70% of the children in the clinical group had received prior or had on-going speech therapy, and this was not limited only to the children with DLD.However, only 28% of the children had a delay of -1 SD or more in the vocabulary task which may suggest that other aspects of language than only expressive vocabulary are needed in emotion recognition.For example, some children with ASD, ADHD and DLD have been shown to have difficulties in the use of language, that is to say, pragmatics (e.g., Green, Johnson & Bretherton, 2014;Helland & Helland, 2017).Ideally in this study too, the language assessments should also have covered more than just expressive vocabulary and rapid serial naming.
Research suggests that ToM skills are strongly associated with social-emotional skills (Golan et al., 2008).According to a recent meta-analysis by Bora and Pantelis (2016), both facial and vocal emotion recognition and ToM skills are significantly impaired in individuals with ADHD with large effect sizes of 0.40-0.44.In the present study, the ADHD group did not differ significantly from the ASD and DLD groups in ToM skills.However, in the regression analysis model in which children with ADHD had smaller delays in expressive vocabulary, passing the ToM 1 task predicted emotion recognition from facial expressions.In contrast, the ASD group did not differ from the other groups in ToM and emotion recognition skills, though, based on earlier research findings (e.g.Loukusa et al., 2014), we expected to see a difference.This may be due to the small sample size, and because our sample of children may have been somewhat biased; according to our inclusion criterion, the children's nonverbal IQ had to be over 85, and therefore our participants did not represent the whole range of children, especially with ASD.

Study limitations
The biggest limitations restricting the generalisability of our results are the fairly small number of participants and the unequally distributed number of children between the three diagnosis groups.Especially the group of children with DLD was small (N=13), and these children also often scored lower in different tasks than children with ASD and ADHD.The recruitment of children depended on the parents' and children's interest in participating.It may be that children with a more severe symptom profile and their families were reluctant to volunteer and participate in the study.
Excluding the FEFA 2 test, all the emotion recognition tasks used were self-constructed.They have, however, high face-validity because the task types they represent are typically used when emotion recognition is assessed (e.g.Taylor et al., 2015).Additionally, although discerning emotions from voice had low reliability, emotion recognition from facial expressions had high reliability when FEFA 2, and tasks containing photographs and video clips were explored with Cronbach's alpha.By using results obtained from 106 TD children as a reference, we could, however, determine the typical performance in the tasks at each age level.However, due to time constraints during the data collection, we were not able to test the TD children's expressive vocabulary, rapid serial naming, and auditory and visual short-term memory skills.Instead, we had to base the delay variables of the clinical group on the test norms documented in the test manuals and age references published by Loukusa (2007).This is a clear limitation in our study.

Clinical implications
Despite the rather limited number of participants (N=50), our findings suggest shared emotion recognition difficulties in children with ASD, ADHD and DLD.This means that the interventions these children need to improve their emotion recognition skills could also be similar across these diagnostic groups.Our study also revealed that language, ToM and working memory skills may be fundamental in the process of recognising emotions.It is therefore important for clinicians to focus therapy on improving emotional vocabulary and ToM abilities, as well as keep in mind that these children may also have difficulties with working memory skills, which may further hamper emotion recognition skill development.
However, focusing intervention only on linguistic or cognitive skills may not suffice; it is also important that therapy encompasses social-emotional skills in social situations and encourages generalisation of newly learned emotion recognition skills to everyday life.The challenge is how exactly to help the children transfer emotion recognition skills to the peer-relations and social situations.Research on this area has thus far been contradictory, at least concerning children with considerable challenges, such as those with ASD (e.g.Golan et al., 2006).

Conclusions
Overall, this study suggests that, compared to each other and to TD age peers, children with ASD, ADHD and DLD have similar difficulties in emotion recognition.The findings also included the predictive role of the first-order ToM, expressive vocabulary and working memory skills in facial expression recognition and matching facial expressions and tone of voice.Further research with larger samples and a wider set of predictive variables is needed to explore the shared basis of the emotion recognition skills in these disorders, since only a few studies focused on this topic prior to ours.
; ToM 1 task= Theory of Mind first-order false belief task(Baron-Cohen et al., 1985); ToM 2 = Theory of Mind second-order false belief task(Wimmer & Perner, 1985); ToM 2 task: justification = can a child provide a justification for the answer in the ToM 2 task.*Age references fromLoukusa (2007), **Results of TD children are based on the 106 children (mean age 8.02 years, min 6, max 10 years) tested in the present study.

Table 1 .
Demographic characteristics of the children with neurodevelopmental disorders and the TD children CHR age = chronological age; ASD = Autism Spectrum Disorders; ADHD = Attention Deficit Hyperactivity Disorder (including three with ADD); DLD = Developmental Language Disorder; TD = typically developing; N/A = not applicable "Faces" submodule of the Finnish version of the FEFA 2 test(TheBölte et al., 2013;Bölte & Poustka, 2003)ffect Recognition;Bölte et al., 2013;Bölte & Poustka, 2003)was used as a standardised task to assess children's facial emotion recognition skills.This computerised test consists of 50 photographs depicting seven different emotions and their labels as response choices (joy, anger, sadness, fear, surprise, disgust and neutral).The test administrator read aloud the possible answers for illiterate children.The original version of the "Faces" submodule has been reported to have excellent psychometric properties: internal consistency is 0.95 as measured with Cronbach's alpha and stability r=0.92 based on test-retest measurements

Table 2 .
The results of the linguistic and cognitive tasks of the children with neurodevelopmental disorders with age norms or references

Table 3 .
Comparison of the results of linguistic and cognitive tasks of the children with neurodevelopmental disorders with age norms or references with p-values from the Paired Sample T test or Fisher's Exact Test.

RAN time, subtest "Colours, Numbers and Letters"
*Compared to age references ofLoukusa (2007), ** Compared to the 106 TD children tested in the present study

Table 4 .
The mean percent performance in emotion recognition tasks of the different diagnostic groups compared to typically developing children (N=106) in the emotion recognition tasks with p-values from One-

Table 5 .
The best models found through linear regression analysis for the FEFA 2 test and the matching task as dependent variables.The variables were entered as delay variables (in comparison to the TD children).