Forty-two volunteers (25 men and 17 women) diagnosed with schizophrenia based on the 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10) participated in this study. They were inpatients recruited from the Center of Mental Health of Anhui province. The patients had a mean age of 26.40 (standard deviation [SD] = 5.31), with illness duration ≤ 5 years. Thirty-three patients were taking antipsychotic medications as follows: clozapine (28.57%), risperidone (16.67%), olanzapine (9.52%), chlorpromazine (9.52%), aripiprazole (7.14%), haloperidol (4.76%), and sulpiride (2.38%). A control group of 42 healthy participants were recruited through advertisement. The patient and control groups were matched for age, sex, and education level. The controls with a past or present psychiatric illness were excluded according to a semi-structured clinical interview. The exclusion criteria for patients and controls were neurological problems and visual or auditory dysfunction as assessed by chart review and consultation with the clinical staff. The study protocol was approved by the Human Ethics Committee of Anhui Medical University. Informed written consent in accordance with the Declaration of Helsinki were obtained from all participants. The patients’ clinical features and the demographic background of the two groups are summarized in Table 1.

The pragmatic ability was assessed by using irony tasks (Varga et al., 2013), with two conditions, namely, irony and control conditions. In the irony condition, the tasks consisted of a short phase describing a social situation, which was followed by an ironic statement. The control condition included tasks based on physical causality, which was not related to irony. There were 10 scenarios in each condition. The number of sentences, number of words, and grammatical complexity did not differ between the two scenarios. The experimenter read the scenarios, and the participant was asked to answer ironic or control questions after each scenario. The participant scored 1 point for a correct answer on each question; the total score for ironic questions and control questions ranged from 0 to 10.

The cognitive ToM was assessed by using the false-belief task based on our previous study (Li et al., 2014). The task comprised four second-order false-belief stories. Two of the stories were used by Perner and Wimmer (1985), and the other two were from the study by Sullivan et al. (1994). During the task, the experimenter read the stories and the participants were asked to answer two questions after each story. One test question was based on cognitive ToM and the other was a control question that was not related to ToM. Participants scored 1 point for a correct answer on each question; the total score for test questions and control questions ranged from 0 to 4.

The affective ToM was assessed by using an Eyes test, which comprised 30 eye photos (Baron-Cohen et al., 1997, 2001). The photos that were taken from our previous study were adapted for Chinese participants (Li et al., 2013). Each photo was accompanied by four terms, and the participant was subsequently asked to choose the term that best describes the mental state of the character in the photo. In the control condition, the participant was asked to determine the sex of each person in the photo. The participant scored 1 point for a correct response on each photo and the total score for the test questions and control questions ranged from 0 to 30.

Inhibition was assessed by using a numerical Stroop task (Crespo et al., 2009). The task was generated using E-Prime 1.1. The task stimuli were the Arabic digits ‘2,’ ‘3,’ ‘4,’ and ‘5’ with variable length composed of two, three, four, or five digits. The stimuli were presented in black on a white background and were displayed on a 17-inch monitor of a personal computer. Two conditions were included: (1) a congruent condition in which the number of digits and identity of digits are matched, such as 22, 333, 4444, or 55555; (2) an incongruent condition in which the number of digits and identity of digits are mismatched, such as 222, 33, 44, and 555. The task consisted of two blocks, and each block contained 12 practice trials followed by 48 test trials. The congruent and incongruent trials were presented randomly in each block. The participants were instructed to report the number of digits while attempting to ignore the identity of the digit, and they were asked to respond as fast and accurately as possible. The difference in the mean reaction times between the two conditions was regarded as the index of inhibition, with lower scores indicative of superior inhibitory ability.

Updating was assessed by using a running memory paradigm (Miyake et al., 2000). The task was generated using E-Prime 1.1. The task stimuli were digits ranging from 1 to 9. Each digit was presented for 1,000 ms and a series of digits was presented at increasing length (up to 12) in the center of the screen. The participant was instructed to remember the last four digits in the sequence and to report verbally the digits when the sequence stopped. The participant was unaware of the number of digits that would be presented and of the varied length of sequences. The task consisted of 20 trials and the total score ranged between 0 and 20.

Shifting was assessed by using a digit-switching task (Salthouse et al., 1998). The task was generated using E-Prime 1.1. The task stimuli were digits ranging from 1 to 9 (excluding digit 5), which were presented in the center of the screen. The participant performed three sets of tasks in the following order: odd–even, more–less, switching. The odd–even task required the participant to press the F key if the digit was odd and to press the J key if the digit was even. The more–less task required the participants to press the F key if the digit was more than 5 and the J key if it was less than 5. The switching task required the participant to perform either the odd–even task or the more–less task according to the color of the presented digits. A red digit signaled the odd–even task, while a green digit signaled the more–less task. The participant would rely on transformation of color and task rules to complete the task. The index of shifting is the D-value, which is the reaction time of shifting condition minus the average reaction time for the odd–even and more–less conditions. A small score indicated strong shifting ability.

The participants’ clinical status was assessed using the Positive and Negative Symptoms of Schizophrenia (PANSS) (Kay et al., 1987). The PANSS is a 30-item rating instrument where each item is scored on a 7-point scale ranging from 1 (absent) to 7 (extreme). The PANSS included three sub-scales, i.e., positive symptom scale, negative symptom scale, and general psychopathology scale. Two clinical psychiatrists who were blind to the treatment status of the patients performed the assessment; their inter-rater reliability was good (Kappa > 90%).

The statistical analyses were carried out using SPSS 18.0 software. Group differences were assessed by using analysis of variance (ANOVA). The relationship between pragmatic ability, ToM, and EF was investigated by using the Pearson’s correlation. Hierarchical regression analyses were performed to predict the variables contributing to the pragmatic ability. The predictor variables were entered into the equations in three successive steps. In the first step, the variables of age and level of education were entered. The components of EF (inhibition, updating, and shifting) were entered in step two, and the components of ToM (cognitive ToM, affective ToM) were entered in step three.

The scores of performance on the irony tasks were examined using a 2 (group: patient, control) × 2 (question type: test question, control question) mixed-design ANOVA. There were significant effects of group, F(1,82) = 17.16, p < 0.001, ηp2 = 0.17, and question type, F(1,82) = 63.77, p < 0.001, ηp2 = 0.44, with a significant interaction between them, F(1,82) = 60.13, p < 0.001, ηp2 = 0.42. To investigate the interaction further, one-way ANOVAs were conducted on each question type. For the test question of the irony tasks, the group effect was significant, F(1,82) = 32.97, p < 0.001, ηp2 = 0.29. The patient group had a lower score (M = 7.21, SD = 1.88) than the control group (M = 9.12, SD = 1.04). For the control question, the group effect was not significant, F(1,82) = 1.77, p > 0.05, ηp2 = 0.02 (Figure 1A).

The scores of the cognitive ToM were examined using a 2 (group: patient, control) × 2 (question type: test question, control question) mixed-design ANOVA. There were significant effects of group, F(1,82) = 195.61, p < 0.001, ηp2 = 0.71, and question type, F(1,82) = 126.80, p < 0.001, ηp2 = 0.61, with a significant interaction between them, F(1,82) = 126.80, p < 0.001, ηp2 = 0.61. To investigate the interaction further, one-way ANOVAs were conducted on each question type. For the test question of the cognitive ToM, the group effect was significant, F(1,82) = 220.90, p < 0.001, ηp2 = 0.73. The patient group had a lower score (M = 1.62; SD = 0.91) than the control group (M = 3.86; SD = 0.35). For the control question, the group effect was not significant, F(1,82) = 2.24, p > 0.05, ηp2 = 0.03 (Figure 1B).

The scores of the affective ToM were explored using a 2 (group: patient, control) × 2 (question type: test question, control question) mixed-design ANOVA. There were significant effects of group, F(1,82) = 44.78, p < 0.001, ηp2 = 0.35, and question type, F(1,82) = 635.69, p < 0.001, ηp2 = 0.89, with a significant interaction between them, F(1,82) = 44.91, p < 0.001, ηp2 = 0.35. To investigate the interaction further, one-way ANOVAs were conducted on each question type. For the test question of the affective ToM, the group effect was significant, F(1,82) = 48.24, p < 0.001, ηp2 = 0.37. The patient group had a lower score (M = 16.05, SD = 3.94) than the control group (M = 21.64, SD = 3.43). For the control question, the group effect was not significant, F(1,82) = 1.60, p > 0.05, ηp2 = 0.02 (Figure 1C).

The scores of performance on the EF tasks were measured using a 2 (group: patient, control) × 3 (task type: inhibition, updating, shifting) mixed-design ANOVA. There were significant effects of group, F(1,82) = 7.95, p < 0.006, ηp2 = 0.09, and task type, F(2,164) = 222.89, p < 0.001, ηp2 = 0.73, with a significant interaction between them, F(2,164) = 3.35, p < 0.05, ηp2 = 0.04. To investigate the interaction further, one-way ANOVAs were conducted on each type of task. For the inhibition task, the effect group was significant, F(1,82) = 6.53, p = 0.012, ηp2 = 0.07. The patient group had a poorer performance (M = 498.53, SD = 256.18) than the control group (M = 384.75, SD = 132.96). For the updating task, the group effect was significant, F(1,82) = 300.79, p < 0.001, ηp2 = 0.79. The patient group scored less (M = 6.17, SD = 3.03) than the control group (M = 15.88, SD = 2.00). For the shifting task, the group effect was significant, F(1,82) = 4.64, p = 0.034, ηp2 = 0.05. The patient group had a poorer performance (M = 1,216.69, SD = 509.51) than the control group (M = 962.00, SD = 572.72; Figure 2).

Table 2 summarizes the Pearson correlation coefficients for pragmatic, ToM, and EF performances in the patient and control groups. For the schizophrenic group, performance on the pragmatic task was significantly correlated with the cognitive ToM, affective ToM, inhibition, and updating. For the healthy control group, performance on the pragmatic task was significantly correlated with the cognitive ToM, inhibition, updating, and shifting.

The hierarchical regression analyses on the patient group revealed that age and educational level accounted for a largely significant proportion of the variance in pragmatic ability (adjusted R²= 0.102). In the first step, the educational level (β = 0.377, t = 2.229) revealed a significant influence on pragmatic ability, while age (β = 0.122, t = 0.898) was not significant. After controlling for the educational level, the EF subcomponents accounted for a largely significant proportion of the unique variance of pragmatic ability (ΔR² = 0.299) in the second step. Inhibition (β = -0.627, t = -3,990) revealed a significant influence on the pragmatic ability, while updating (β = 0.158, t = 1.052) was not significant. After controlling for the educational level and inhibition, ToM explained a large significant proportion of the unique variance of pragmatic ability (ΔR² = 0.092) in the third step. Cognitive ToM (β = 0.301, t = 2.331) showed a significant influence on pragmatic ability, while affective ToM (β = 0.158, t = 1.128) was not significant (Table 3).

The hierarchical regression analyses on the control group revealed that age and educational level explained a significant proportion of the variance in pragmatic ability (adjusted R² = 0.166). In the first step, educational level (β = 0.428, t = 2,864) showed a significant influence on pragmatic ability, while age (β = 0.072, t = 0.484) was not significant. After controlling for educational level, EF explained a large significant proportion of the unique variance of pragmatic ability (ΔR² = 0.299) in the second step. Inhibition (β = -0.317, t = -2.532) and updating (β = 0.321, t = 2.588) showed a significant influence on pragmatic ability, while shifting (β = -0.230, t = -1.715) was not significant. After controlling for educational level, inhibition, and updating, ToM explained a significant proportion of the unique variance of pragmatic ability (ΔR² = 0.092) in the third step. Cognitive ToM (β = 0.317, t = 2.592) showed a significant influence on pragmatic ability (Table 3).

Table 4 summarizes the Pearson correlation coefficients for ToM and EF in the patient and control groups. Among the patients with schizophrenia, the results indicated that the inhibition was significantly correlated with both cognitive and affective ToM. No significant correlations were found between updating and the two subcomponents of ToM. The correlations between shifting and the two subcomponents of ToM also failed to reach statistical significance.

Among the healthy participants, the results found that inhibition was significantly correlated with both cognitive and affective ToM. Updating was significantly correlated with the cognitive ToM, but not with the affective ToM. Shifting was significantly correlated with the cognitive ToM, but not with the affective ToM.

The Pearson correlation coefficients revealed that the pragmatic ability was significantly correlated with the positive symptom (r = -0.428, p = 0.005), but not with the negative symptom (r = -0.282, p = 0.070).

The results of the Pearson correlation analyses indicated a significant correlation between impaired understanding of irony and deficits in cognitive and affective ToM in the patient group. Similarly, cognitive and affective ToM both significantly predicted the irony understanding in patients with schizophrenia. The finding of the relation between cognitive ToM and irony understanding is consistent with previous studies that reported a relationship between cognitive ToM and pragmatic ability in schizophrenia (Langdon and Coltheart, 2004; Mazza et al., 2008; Champagne-Lavau and Stip, 2010) and other clinical situations such as right-hemisphere lesions, autism, and traumatic brain injury (Happé, 1993; Winner et al., 1998; Champagne-Lavau and Joanette, 2009). Nevertheless, one study has previously reported that pragmatic deficits were not related to cognitive ToM in schizophrenia (Mo et al., 2008). Information underlying ironic remarks is not directly stated, but merely implied. Hence, irony understanding requires first-order inference about the speaker’s belief as well as second-order inference about the speaker’s beliefs about the listener’s beliefs (Marjoram et al., 2005). Therefore, impaired understanding of irony has been associated with the cognitive ToM.

The affective ToM seems to contribute to irony understanding in schizophrenia. Verbal irony often conveys the speaker’s emotions and attitudes that are of opposite valence to the literal meaning of the words spoken. Indeed, different types of irony evoke different emotions. For instance, ironic criticism or sarcasm would elicit negative emotions such as anger, disgust, and contempt. One example is that you said to a friend “you are really good at that” when the friend failed an examination, which involves a positively worded statement that is meant to be taken negatively. The speaker’s attitudes and emotions could be inferred from their paralinguistic features, such as tone of voice, facial expression, and gestures. The ability to infer other people’s emotions and feelings is designated affective ToM (Shamay-Tsoory et al., 2009, 2010). Despite the theoretical relationship between comprehension of irony and affective ToM, only one study reported a mediation effect of affective ToM on the relation between age and understanding ironic criticism in school-aged children (Agostino et al., 2017). Although we found an association between affective ToM and irony understanding in patients with schizophrenia, the regression results suggested that affective ToM no longer predict irony performance when controlling the effect of EF. This finding shed new light on the mixed findings on the association between affective ToM and irony.

We investigated the effect of EF on pragmatic ability in patients affected by schizophrenia. We found that patients were impaired in the three EF components (inhibition, shifting, and updating). Correlation analysis revealed that the ability to understand irony was correlated with the scores of inhibition, updating, but not with shifting in the patient group.

According to the psycholinguistic model of non-literal language understanding (Champagne and Joanette, 2004), irony comprehension seems to depend on the inhibition of the automatic processing and interpretation of literal meaning of an utterance. As a consequence, impaired inhibition may result in deficits in suppressing the more readily accessed literal meanings (McDonald and Pearce, 1996). Previous researchers have also reported associations between inhibition and pragmatic functions after traumatic brain injury (Channon and Watts, 2003). However, Langdon et al. (2002) found that difficulties in understanding metaphors and irony on a story-comprehension task in patients with schizophrenia were independent of inhibition.

In addition to inhibition, impaired updating was also associated with the pragmatic performance in schizophrenia. Ironic statements typically provide a direct literal meaning and an indirect non-literal message. Updating enables the coding of stimuli and replacing non-relevant information with relevant information in working memory (Morris and Jones, 1990). The present results suggested that updating plays a role in understanding irony. However, updating has been rarely examined previously, even though it is one of the most frequently studied EFs (Miyake and Friedman, 2012). To our knowledge, this is the first study that has attempted to examine the relation between irony and updating.

The present results did not find a relation between shifting and irony understanding in schizophrenia. Shifting is regarded as a form of flexibility in response to changing contexts. Previous studies suggested that flexibility did not play a role in pragmatic performance in patients with schizophrenia (Langdon et al., 2002; Champagne-Lavau and Stip, 2010), right-hemisphere lesion (Champagne-Lavau and Joanette, 2009), and traumatic brain injury (Martin and McDonald, 2005). In contrast, some studies reported a relation between reduced flexibility and impaired pragmatic performance (Sponheim et al., 2003; Brune and Bodenstein, 2005). These inconsistencies could be attributed to the varied tasks and different types of pragmatic language.

The regression analysis indicated that the cognitive ToM explained 9.2% of the unique variance of pragmatic ability in patients with schizophrenia, while only one of the three EF components (inhibition) explained 29.9% of the unique variance. These results suggested that cognitive ToM could predict impaired irony comprehension, and that inhibition was the best predictor of the patients’ ability to comprehend irony. Channon and Watts (2003) found that three predictors of inhibitory control gave rise to a significant regression equation accounting for 36% of the variance of pragmatic comprehension in patients with traumatic brain injury. These results suggested a critical role of the inhibition abilities in irony comprehension in patients with schizophrenia. It also suggested a central role of inhibition among the three EF components (Westerhausen et al., 2011). According to the three-factor model of EF (Miyake et al., 2000), inhibition is regarded as a central sub-component of EF, which is the ability to inhibit deliberately dominant, automatic, or prepotent responses in the presence of competing sources. These inhibition features enable patients with schizophrenia to understand pragmatic language, suggesting a particular target of inhibition for rehabilitation and intervention programs.

The regression results indicated significant correlations between inhibition and the two ToM subcomponents in both schizophrenia and healthy groups. This finding suggested that reasoning about the beliefs and emotions of others relies, at least partly, on the inhibitory processing. Two types of evidence from developmental studies on preschool children support this inhibitory account. First, the difficulty level of ToM tasks could be manipulated by modifying the inhibitory demands (Leslie et al., 2005; Yazdi et al., 2006). Second, previous studies demonstrated that children’s performance on ToM tasks is correlated with their performance on tasks assessing inhibition (Carlson and Moses, 2001; Rakoczy, 2010). In addition to inhibition, updating and shifting were correlated with cognitive ToM in healthy controls, but not in patients with schizophrenia. The finding of correlation between the two EF subcomponents (updating and shifting) and cognitive ToM in healthy individuals is consistent with previous studies (Austin et al., 2014; Li et al., 2014). Importantly, the regression results in both patients and healthy controls indicated that ToM ability continued to predict the performance on irony tasks once EF was controlled for. Thus, ToM and EF may independently contribute to the ability of understanding irony.

The present findings suggested that irony understanding was related to positive symptoms, but not to negative symptoms. Misunderstanding of irony may contribute to developing positive symptoms such as delusions. According to the theoretical model proposed by Salvatore et al. (2012), an alteration in pragmatic understanding of the mind of others is one of several triggers for delusional experience in schizophrenia. Furthermore, several positive symptoms such as disorganized thought and difficulty in following a conversation could also contribute to misunderstanding irony. A previous study reported a correlation between positive symptoms and pragmatic performance (Stratta et al., 2007). In contrast, another study reported a correlation between negative symptoms and pragmatic performance in schizophrenia (Saban-Bezalel and Mashal, 2017). Those inconsistencies call for future studies to clarify the relation between irony understanding and specific clinical symptoms in schizophrenia.

Several limitations of the present study should be considered. First, as most patients were medicated in the present study, our results may be influenced by antipsychotic effects. Second, our sample size was moderate, leading to limited statistical power for the mediation effect. Thus, the findings in this study should be considered preliminary. Moreover, there is a lack of comparison between medicated and unmediated patients due to our sample size. Third, we recruited only patients with early-stage schizophrenia. Patients with chronic schizophrenia should be evaluated in future studies. Finally, an intelligence quotient (IQ) test or a test of general cognitive ability should have been used to rule out the possibility that the worse performance of the patients compared to the controls might be explained just in terms of their different general cognitive competence; however, such tests were not included in the present study.