Abstract
This study establishes a sequence of developing mental state understandings in infants. We used three violation-of-expectation paradigms to assess fifty-seven 16-month-olds’ ability to (a) infer an actress’s intention from her prior repeated approaches to an object, (b) recognize her emotion by watching her facial-emotional display, and (c) deduce her false belief by noticing her lack of visual access to a change in the experimental setup. Contingencies between passing the three tasks were analyzed. Results showed that the infants made sense of intention first, followed by emotion, and then false belief. This progressive sequence parallels what has been found with preschoolers using verbal theory-of-mind tasks.
In acquiring a theory of mind (ToM), children learn to understand not only false belief but also other mind states such as desire and emotion. Wellman and Liu (2004) studied 3- to 5-year-olds’ performances on a range of verbal tasks and constructed a ToM scale to assess preschoolers’ mentalizing development. The scale orders five mentalistic understandings in a developmental sequence. Children first judge that two persons can have different desires (diverse desires) and then different beliefs for the same objects (diverse beliefs), followed by the understanding that a lack of perceptual access renders one ignorant about reality (knowledge access). Later children know that people can possess and act in accordance with mistaken representations of reality (content false belief) and, finally, that people can hide their true emotion and display a different one (real-apparent emotion). This scaling of mentalizing abilities has been supported by both cross-sectional and longitudinal findings (Wellman & Liu, 2004; Wellman, Fang, & Peterson, 2011).
Mentalizing is thought to mature at around 4 years of age when children give correct verbal responses to explicit ToM tasks. However, non-verbal methods such as expectancy-violation looking, anticipatory looking, observation, and active helping have revealed that infants and toddlers are not blind to others’ minds. By their first year, infants are able to represent simple motivational states including goals and preferences (Luo, 2011a; Woodward, 1998). Their understanding of emotions develops somewhat later in the second year (Hepach & Westermann, 2013; Phillips, Wellman, & Spelke, 2002). In a classic study, Zahn-Waxler, Radke-Yarrow, Wagner, and Chapman (1992) observed that 1- to 2-year-olds were sensitive to distress feigned by both their mothers and unfamiliar persons and responded to it by expressing concern (affective empathy) and trying to understand the suffering party’s feeling (cognitive empathy). Such empathic responses to perceived negative emotion during the second year of life are believed to be significantly heritable (Knafo, Zahn-Waxler, Van Hulle, Robinson, & Rhee, 2008; Zahn-Waxler, Robinson, & Emde, 1992), related to temperament at 4 months (Young, Fox, & Zahn-Waxler, 1999) and maternal practices (Eisenberg et al., 1992), and independent of self-reflective abilities and self-distress (Davidov, Zahn-Waxler, Roth-Hanania, & Knafo, 2013; Roth-Hanania, Davidov, & Zahn-Waxler, 2011). More recent studies have demonstrated infants’ understanding of the link between emotion and action. First, infants are sensitive to concurrent emotion-action match. Using the violation-of-expectation paradigm, Hepach and Westermann (2013) found that 14-month-olds were surprised when a happy actor thumped a toy (negative action) and when an angry actor patted a toy (positive action). Second, infants form expectations about emotions following actions. Skerry and Spelke (2014) showed that 10-month-olds looked longer when an agent displayed a negative emotion after completing than after failing a demonstrated goal. Infants at 18 months further showed more hypothesis testing and less empathic concern when an agent expressed distress after positive than after negative experiences (Chiarella & Poulin-Dubois, 2013). They also offered less empathic helping to those whose emotional responses to events were incongruent (Chiarella & Poulin-Dubois, 2018). Third, infants also use others’ emotions to predict their subsequent actions. Fourteen-month-olds anticipated that an agent who earlier expressed positive emotion toward a toy would reach for it later (Phillips et al., 2002). In a recent study, Jin, Houston, Baillargeon, Groh, and Roisman (2018) showed that infants in their first year of life were sensitive to other infants’ emotional distress and, furthermore, expected their mothers to comfort them. Hence, understanding of emotion and how it should be addressed in a prosocial way is well in place before 2 years of age and continues to develop into middle childhood (Eisenberg, Betkowski, & Spinrad, 2013).
The first two years also witness infants’ developing understanding of epistemic and counterfactual states. Infants can track what an agent has seen and predict her subsequent action based on what they think she knows about the scene (Luo & Baillargeon, 2007; Luo & Johnson, 2009). Remarkably, infants can also ascribe to an agent false beliefs about the location, properties, and identity of an object (Kovacs, Teglas, & Endress, 2010; Onishi & Baillargeon, 2005; Scott & Baillargeon, 2009; Southgate, Senju, & Csibra, 2007). They use an agent’s mistaken belief not only to predict her action but also to intervene when necessary (Buttelmann, Carpenter, & Tomasello, 2009). Taken together, infants’ spontaneous responses in non-verbal paradigms suggest a much earlier beginning of mentalizing (Baillargeon, Scott, & Bian, 2016; Scott & Baillargeon, 2017; though see Poulin-Dubois et al., 2018 for counter-arguments).
So far attention has been mainly paid to how early infants start to reason about mental states. Across studies different mental state understandings in different age groups are examined. One recent study, however, adopted a within-subjects design and assessed each infant’s understanding of intention, true belief, desire, and false belief via the violation-of-expectation paradigm. Yott and Poulin-Dubois (2016) tested 14- and 18-month-olds using the intention, desire, and belief tasks developed respectively by Phillips and Wellman (2005), Repacholi and Gopnik (1997), and Onishi and Baillargeon (2005). Their results showed that both 14- and 18-month-olds appeared to understand intention and true belief while only the older infants understood desire. Neither age group was sensitive to the difference between the inconsistent and consistent event in the false belief task. The authors thus conclude that mental state understanding development in infancy follows the sequence of intention, true belief, desire, and finally false belief.
Similar to Yott and Poulin-Dubois (2016), the present study seeks to establish a developmental sequence of mental state understandings via a within-subjects design. We used three violation-of-expectation paradigms to test a group of 16-month-olds on their understanding of intention, emotion, and false belief. These paradigms assumed that if infants ascribed a certain mental state to an agent they would expect her to act in accordance with that mental state. Infants would thus be surprised and look longer when the agent’s action was inconsistent with the perceived mental state than when they matched. Understanding of a mental state was therefore assumed if looking times were longer for action-mental state inconsistent than consistent events. We then contrasted infants’ performances on the three tasks and ordered the tasks in terms of developmental ease. There are several differences between the present study and Yott and Poulin-Dubois (2016). First, they define intention as reaching for an object via the best possible path (Phillips & Wellman, 2005) whereas in the present study intention means preference for one object over another (Woodward, 1998). Second, their desire understanding means whether one’s desire, expressed as attraction toward one object in contrast with dislike for another object, is satisfied by the action of another person (Repacholi & Gopnik, 1997). On the other hand, the present emotion task focuses on whether or not the agent acts in accordance with her expressed emotion. The two tasks are thus different, although still comparable because both rely critically on the infant’s sensitivity to the agent’s expressed emotion. Third, Yott and Poulin-Dubois (2016) compare two age groups’ average performances on the different tasks to derive a developmental sequence while we examine the pass–fail ratios for the different tasks in one age group. Because of these differences, the present results complement Yott and Poulin-Dubois’s (2016) findings and reinforce their conclusion rather than only replicating what they have observed. We hypothesize that infants pass the intention task first, followed by the emotion and finally false-belief task. The hypothetical sequence is derived from existing findings showing that (a) infants are sensitive to intention behind action well before their first birthday (Woodward, 1998), (b) emotion and false-belief understanding show a rather protracted development across the first two years (Baillargeon et al., 2016), (c) mental state understanding in infants under 2 years of age progresses from intention and true belief to desire and finally false belief (Yott & Poulin-Dubois, 2016), and (d) preschoolers respond to others’ intentions and desires before false beliefs in explicit verbal tasks (Wellman & Liu, 2004).
Method
Participants
Participants were 57 healthy-term 16-month-olds (Mage = 16.09 months, SD = 0.31; 27 females). Another 11 infants were tested but excluded because of fussiness (3), parental interference (1), experimenter error (1), and inter-observer reliability in looking time coding lower than .80 (6). We recruited the infants from local parent–child forums. All the infants were Chinese born and raised in middle-class families in Hong Kong. Parents provided written informed consent before testing. Each parent-infant dyad received a certificate for their participation and were reimbursed for their travel expenses.
Apparatus, Materials, and Procedure
After a 15-minute warm-up, we tested the infants’ mental state understandings with three violation-of-expectation paradigms. Experimental events across the paradigms were shown on a stage (40 cm high x 108 cm wide) enclosed in a display booth (180 cm high x 116 cm wide x 78 cm deep). An actress sat on a chair centered behind the stage. Her face, upper body, and the objects on the stage were clearly visible to the infant. A hidden camcorder was mounted below the stage to record infant behavior. Two independent observers saw the image on a television screen in an adjacent room and coded the infant’s attention to the display on the spot. The observers were blind to the experimental conditions and could see neither the objects nor the actress. They were well trained to depress a button on a game pad when the infant fixated on the display, and to release it when the infant looked away. Time lengths of these button presses were transferred to a computer program which computed the infant’s aggregate looking times and inter-observer reliabilities in looking time coding. This program was developed by Baillargeon and Barrett (2005) and used in various violation-of-expectation experiments for infants under 2 years of age (e.g., Choi, Mou, & Luo, 2018; Krehm, Onishi, & Vouloumanos, 2014; Luo, 2011b; Onishi & Baillargeon, 2005). Parents were asked to remain silent and avoid interacting with their infants during testing. Short breaks were given in between the paradigms.
Intention
The intention understanding test was based on Woodward’s (1998) paradigm. The infant sat on her parent’s lap and faced the stage 90 cm away. A cylinder and a block were placed on the left and right side of the stage respectively, with a neutral mark between them. The infants first received four familiarization trials, 1 followed by two test trials. When familiarization began, the actress placed her hands on the stage and fixated at the mark. After the infant had looked at the display for 2 seconds, the actress extended her arm to reach for the cylinder and grasped it; she then paused and held this gesture until the infant looked away for 2 consecutive seconds or looked for 30 cumulative seconds without looking away. The curtain was then drawn and the first familiarization trial ended.
After four familiarization trials, the cylinder and the block swapped locations behind the closed curtain. In a consistent test trial, the curtain opened and the actress again reached for the cylinder, now in a different location, which was consistent with her prior goal. Conversely, in an inconsistent test trial the actress grasped the block, an act that was inconsistent with her prior goal. Each infant received one consistent and one inconsistent test trial. In both trials, the actress paused until the infant looked away for 2 consecutive seconds or 60 cumulative seconds had elapsed.
Emotion
We used Phillips, Wellman, and Spelke (2002) paradigm to test the infants’ emotional state understanding. A blue and a yellow toy bear were rested side by side on the stage. There were four familiarization and four subsequent test trials. Familiarization began when the curtain was drawn, revealing the actress with a neutral facial expression. After the infant had looked at the display for 2 seconds, the actress looked and smiled at the blue bear, and exclaimed in delight “Oh! Look at that teddy bear!” She held this facial expression for 5 seconds before the curtain was closed. When the curtain reopened, the actress looked at and held the blue bear, the one she had shown interest in. A familiarization trial ended when the infant looked away from the display for 2 consecutive seconds or 30 cumulative seconds had elapsed.
Two consistent and two inconsistent test trials, shown in alternation, followed familiarization. In the consistent test trials, the actress emoted positively toward the yellow bear, the one not regarded and grasped in familiarization, for 5 seconds before the curtain was closed. When the curtain reopened, she smiled at and reached for the yellow bear. Such grasping was consistent with her prior expressed emotion toward the bears. In the inconsistent events, the actress first looked and smiled at the blue bear, but later grasped the yellow bear. Because the actress did not reach for the bear that she had felt excited about, this event sequence was considered inconsistent. A test trial ended when the infant looked away for 2 consecutive seconds or had looked for 60 cumulative seconds. For each infant, looking times from the two consistent (and inconsistent) trials were averaged.
False belief
We adopted Onishi and Baillargeon’s (2005) paradigm to test the infants’ false belief understanding. 2 In the first familiarization trial, a toy lion was rested on the stage between a green and a yellow box. The boxes’ openings faced each other so that one could hide the content of the boxes from both the infant’s and the actress’s view. When the back curtain (between the actress and the stage) was drawn, the actress appeared and played with the toy for 5 seconds. She then hid the toy in the green box, paused with her hand and the toy inside it, and held this gesture. After the infant looked away from the display for 2 consecutive seconds, the front curtain (between the infant and the stage) was closed and the trial ended. The second and third familiarization trial similarly involved the actress reaching inside the green box to hide the toy after playing with it and paused until the infant looked away for 2 consecutive seconds.
A critical belief-induction trial followed. The infants were randomly assigned to the false-belief-green (FB-green) or false-belief-yellow (FB-yellow) condition. In the FB-green condition, while the back curtain remained closed, the infant saw the toy move from the green to the yellow box. Because only the infant witnessed this location transfer, the actress should have a false belief about the toy’s current location. In the FB-yellow condition, the back curtain was first opened so that the actress saw the toy move from the green to the yellow box. The back curtain was then closed; only the infant saw the toy return to the green box at this point. Hence the actress held the false belief that the toy remained in the yellow box after the first transfer. In both conditions, the trial ended when the infants looked away from the final paused scene for 2 consecutive seconds or 30 cumulative seconds had passed.
The infants received a consistent and an inconsistent test trial after belief induction. In the consistent trial, the actress searched for the toy where she last saw it (i.e., green box in FB-green condition and yellow box in FB-yellow condition). In contrast, in the inconsistent trial, the actress reached into the other box (i.e., yellow box in FB-green condition and green box in FB-yellow condition). A test trial ended when the infants looked away for 2 consecutive seconds or 60 seconds had elapsed.
Administration order of the three tasks and that of the consistent vs. inconsistent test trials within each task were counterbalanced across the infants.
Results
Overall inter-observer reliability in looking time coding was .90. For the infants assigned to the FB-green condition, mean looking times for the inconsistent and consistent test trials were 16.85 s (SD = 12.98 s) and 13.38 s (SD = 8.43 s), respectively. For the infants assigned to the FB-yellow condition, the corresponding means were 17.33 s (SD = 10.44 s) and 14.10 s (SD = 6.82 s). Because the corresponding means from the FB-green and FB-yellow condition were very comparable, data from the two conditions were pooled despite some surface differences in the actual false belief induction procedure. Pooling the data from the FB-green and FB-yellow condition rendered the sample size for the false belief task comparable to that for the other two tasks.
Order of tasks had no effect on total looking times in the familiarization and test trials, all ps > .05; fatigue was therefore not a factor to consider.
Familiarization Trials
Average looking times for the last familiarization trials in the intention, emotion, and false belief tasks were 12.29 s (SD = 7.81 s), 10.65 s (SD = 6.07 s), and 11.75 s (SD = 6.34 s), respectively. There was no overall difference, F(2, 55) = 1.10, p = .341. Average looking times for all familiarization trials in the intention, emotion, and false belief task were 15.34 s (SD = 4.39 s), 16.51 s (SD = 4.68 s), and 15.45 s (SD = 5.62 s), respectively. There was no overall difference, F(2, 55) = 1.92, p = .157. The infants were thus equally attentive to the familiarization trials across the three tasks.
For the intention task, average looking times to the first, second, third, and fourth familiarization trials were 19.48 s (SD = 6.69 s), 16.30 s (SD = 7.34 s), 14.29 s (SD = 7.72 s), and 12.29 s (SD = 7.81 s), respectively. The infants’ looking time during familiarization declined significantly, F(3, 54) = 11.18, p < .001. For the emotion task, the corresponding average looking times were 21.69 s (SD = 7.90 s), 18.59 s (SD = 9.00 s), 15.08 s (SD = 8.22 s), and 10.65 s (SD = 6.07 s). The looking time decrement across the familiarization trials was significant, F(3, 54) = 31.25, p < .001. For the false belief task, the average looking times dropped from 21.50 s (SD = 7.79 s), 17.58 s (SD = 8.35 s), 13.09 s (SD = 8.19 s), to 11.75 s (SD = 6.34 s) during familiarization. There was again a significant decrease in looking time during familiarization, F(3, 54) = 41.12, p < .001. Hence the infants’ attention to the familiarization trials decreased over time in all the three tasks.
Test Trials
Figure 1 shows average looking times in the test trials for the three tasks. In all three tasks looking times were reliably longer in the inconsistent than consistent trials: intention: inconsistent, M = 18.28 s, SD = 7.12 s, consistent, M = 10.83 s, SD = 6.78 s, t(56) = -6.32, p < .001; emotion: inconsistent, M = 21.90 s, SD = 12.21 s, consistent, M = 14.55 s, SD = 9.09 s, t(56) = -4.37, p < .001; false-belief: inconsistent, M = 17.10 s, SD = 11.61 s, consistent, M = 13.76 s, SD = 7.56 s, t(56) = -2.26, p = .028.
Infants’ Mean Looking Times in the Intention, Emotion, and False-belief Understanding Tasks.
A total of 47 (82%), 41 (72%), and 33 (58%) out of the 57 infants looked longer at the inconsistent than consistent test event in the intention, emotion, and false belief task, respectively. The rest of the infants all looked longer at the consistent than inconsistent event. The proportion of infants looking longer at the inconsistent than consistent event differed significantly across the three tasks, Cochran’s Q = 11.67, df = 2, p = .003. Further Sign tests showed that the number of these infants was significantly larger than the number of infants who looked longer at the consistent than inconsistent test trials for the intention, Z = -4.77, p < .001, and emotion task, Z = -3.18, p = .001, but not the false belief task, Z = -1.20, p = .229.
We compared the infants’ performances on the three tasks to examine their relative difficulties. In each task, infants who did not look longer in the inconsistent than consistent trial were considered failing the task. For each remaining infant, we adjusted the looking time difference between the inconsistent and consistent trial by her total looking time, that is, (inconsistent – consistent) / (inconsistent + consistent). This adjusted difference looking time was also calculated for the group by using the average difference and average total looking time. Infants whose adjusted difference looking time was greater than the group average were considered passing the task; otherwise they failed. We acknowledged the difficulty of choosing a reasonable criterion for passing because any such criteria would be necessarily arbitrary. The present criterion, which was rather conservative, qualified performance with both individual differences and group tendency by adjusting each inconsistent–consistent looking time difference with the individual’s total and comparing it to the group mean. We adopted the present passing criterion also because it was conservative. Infant looking time data from the violation-of-expectation paradigm have been shown to be relatively noisy and results are not always easily replicable (e.g., Powell, Hobbs, Bardis, Carey, & Saxe, 2018). Therefore, more stringent criteria demanding large looking time differences should be preferable if such surface differences are to indicate the presence of an ability.
Tables 1 to 3 present the contingency between passing the intention, emotion, and false-belief tasks. Overall difference in the proportion of passing infants among the three tasks was significant, Cochran’s Q = 25.19, df = 2, p < .001. Next, post-hoc pairwise McNemar tests were conducted. First, while 18 infants passed the intention but failed the emotion task, only four infants passed the emotion but failed the intention task. This asymmetry was significant (McNemar binomial test, p = .004; see Table 1). Second, 25 infants passed the intention but failed the false-belief task; only two infants showed the opposite pattern (McNemar binomial test, p < .001; see Table 2). Therefore, a difference in difficulty between intention understanding and the other two understandings was evident. Finally, Table 3 shows that 12 infants passed the emotion but failed the false-belief task, while only three infants did the reverse (McNemar binomial test, p = .035). Hence emotion understanding develops earlier than false-belief understanding.
Number of Infants Failing and Passing the Intention and Emotion Understanding Tasks (N = 57).
Number of Infants Failing and Passing the Intention and False-belief Understanding Tasks (N = 57).
Number of Infants Failing and Passing the Emotion and False-belief Understanding Tasks (N = 57).
Finally, using Yott and Poulin-Dubois’s (2016) method, we calculated inter-task correlations among the inconsistent trial looking times from the three tasks. Results showed that all the tasks were significantly correlated with each other: intention-emotion, r(56) = 0.41, p < .01; intention-false belief, r(56) = 0.28, p < .05; emotion-false belief, r(56) = 0.45, p < .01.
Discussion
Our results provide evidence for a developmental sequence of mental state understandings in infancy: intention → emotion → false-belief understanding. This pattern is consistent with Yott and Poulin-Dubois’s (2016) results. Although in their study desire instead of emotion understanding was examined, we think that the two underlying abilities being assessed are still fairly comparable because both rely critically on the infant’s sensitivity to and interpretation of the agent’s expressed liking (desire) for an object. The fact that infants understand intentions before false beliefs is also consistent with the ToM progression observed in preschoolers (Wellman & Liu, 2004). Westra and Carruthers (2017) reason that young children find it easier to represent desires than beliefs because of the abundance of desire-talk in child-directed conversations. Corpus data show that children are frequently prompted to talk about people’s intentions and wants and also to express their own (MacWhinney, 2014). We argue that in caregiving, infant-directed conversations may be even more centered around desires and wants than beliefs and other abstract thoughts. Hence it is not surprising that infants grasp intentions first.
Our results seem to contradict Wellman and Liu’s scale (2004) with regard to emotion understanding, which is considered the most difficult in the scale. We argue that the emotion concept currently tested is quite different from that in the real-apparent emotion task in the ToM scale, and is much closer to Yott and Poulin-Dubois’s (2016) desire understanding. We chose Phillips et al.’s (2002) emotion paradigm because its design is highly comparable to the present intention and false belief tasks: All the three paradigms similarly involve infants watching an agent act on some objects, and test infants’ ability to represent the agent’s mental states to predict her action. In contrast, the real-apparent emotion task requires children to understand that we occasionally hide our true feeling and display another emotion. Obviously, such emotion understanding is much more advanced and is not what we are currently testing. The present emotion task therefore may not be seen as the non-verbal parallel of Wellman and Liu’s (2004) real-apparent emotion task.
It is not entirely clear from the infant literature whether emotion or false-belief understanding develops earlier. Different studies have reported different “passing ages” for each understanding, depending on the task that they used. Emotion understanding has been shown to develop at around 10 to 18 months (e.g., Hepach & Westermann, 2013; Jin, Houston, Baillargeon, Groh, & Roisman, 2018; Phillips et al., 2002; Skerry & Spelke, 2014) whereas false-belief understanding from 7 to 18 months (Buttelmann et al., 2009; Kovacs et al., 2010; Onishi & Baillargeon, 2005; Scott & Baillargeon, 2009). In the present study we used well-matched paradigms to examine and compare intention, emotion, and false belief understanding so as to mitigate the problem of task variation, since the selection of task is likely to affect our estimation of the developmental timelines of the underlying abilities. Our findings indicate that infants tend to grasp emotion before false belief when tested with paradigms matched on overall structure and sharing the same dependent measure. This parallels Yott and Poulin-Dubois’s (2016) finding that the development of desire understanding precedes that of false belief. We speculate that similar to intention-talk, emotion-talk is likely to be a recurrent topic of conversation in the caregiving process. Hence infants at this age are more experienced in attending to and reasoning about people’s motivational (e.g., intention and emotion) than belief states, and generally find such understanding easier.
Our findings on false belief understanding in infancy were mixed. On the one hand, the mean looking times indicated that the present 16-month-olds were able to discriminate the inconsistent from the consistent event in the false belief task (Figure 1), but, on the other hand, the number of infants who looked longer at the inconsistent than consistent event (33) was not significantly larger than the number of those who did not (24). Such mixed findings may be considered in the light of some previous successful demonstrations of false belief understanding in 1- to 2-year-olds (e.g., Buttelmann et al., 2009; Onishi & Baillargeon, 2005; Scott & Baillargeon, 2009; Scott, Baillargeon, Song, & Leslie, 2010), as well as some subsequent failures in attempted replications (e.g., Powell et al., 2018; Yott & Poulin-Dubois, 2016, see also the special issue Understanding theory of mind in infancy and toddlerhood edited by Sabbagh & Paulus, 2018). At this time a definite answer to the question does not seem available yet. In a recent commentary, Poulin-Dubois et al. (2018) summarized important non-replication results and warned against accepting early, rich false belief understanding prematurely. The authors also called for large-scale cross-laboratory studies to reconcile the largely inconsistent findings in the field. For our present purpose, however, such diverse findings regarding false belief understanding in infancy do provide some support for our argument that this understanding is the last to develop following intention and emotion, which have consistently been shown to be well in place in the second year of life (e.g., Hepach & Westermann, 2013).
To conclude, the present study provides evidence for a developmental sequence of mental state understandings which specifies that intention understanding develops first, followed by emotion and lastly false belief. This pattern is consistent with many previous findings from a variety of test paradigms. Future research may further examine the correspondence, or the lack of it, between mentalizing development in infancy investigated via non-verbal paradigms and that in early and middle childhood often studied through explicit, verbal methods.
Footnotes
Funding
The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: This research was supported by a General Research Fund (441809) to the corresponding author from the Research Grants Council, Hong Kong Government.