• 沒有找到結果。

臉部表情的分類判斷是否受之前的情境脈絡影響

N/A
N/A
Protected

Academic year: 2021

Share "臉部表情的分類判斷是否受之前的情境脈絡影響"

Copied!
63
0
0

加載中.... (立即查看全文)

全文

(1)

行政院國家科學委員會專題研究計畫 期末報告

臉部表情的分類判斷是否受之前的情境脈絡影響(第 2 年)

計 畫 類 別 : 個別型 計 畫 編 號 : NSC 100-2410-H-004-005-MY2 執 行 期 間 : 100 年 08 月 01 日至 102 年 07 月 31 日 執 行 單 位 : 國立政治大學心智、大腦與學習研究中心 計 畫 主 持 人 : 徐慎謀 計畫參與人員: 此計畫無其他參與人員 報 告 附 件 : 出席國際會議研究心得報告及發表論文 公 開 資 訊 : 本計畫可公開查詢

中 華 民 國 102 年 10 月 30 日

(2)

中 文 摘 要 : 過往的研究主要著重在人類如何對單一呈現的臉部表情進行 分類,而忽略表情分類的過程中是否受到各種情境脈絡的影 響。本計劃的研究重點放在當我們對連續呈現的臉部表情進 行分類時,之前的情境是否會影響分類的結果。研究結果發 現之前的情境是會影響對當下臉部表情的判斷。這些影響可 分為兩類:一.對比效應,也就是受試者傾向認為當下的臉部表 情和之前的臉部表情屬於不同的類別;二.同化效應,也就是 受試者傾向認為當下的臉部表情和之前的臉部表情屬於相同 的類別。何者效應會產生端賴之前與當下臉部表情的相對距 離,以及其他相關因素例如兩張人臉都是否都屬同一人。此 研究顯示臉部表情分類為動態的過程,會依之前的情境發生變 化。 中文關鍵詞: 臉部表情,分類,順序效應

英 文 摘 要 : Facial expressions are highly dynamic signals that are rarely categorized as static, isolated displays. However, the role of sequential context in facial expression categorization is poorly understood. This study examined the fine temporal structure of

expression-based categorization on a trial-to-trial basis as participants categorized a sequence of

facial expressions. The results showed that the local sequential context provided by preceding facial

expressions could bias the categorical judgments of current facial expressions. Two types of

categorization biases were found: 1) assimilation effects - current expressions were categorized as close to the category of the preceding expressions, and 2) contrast effects - current expressions were categorized as away from the category of the

preceding expressions. The effects of such

categorization biases were modulated by the relative distance between the preceding and current

expressions, as well as by the different experimental contexts, possibly including the factors of face identity and the range effect. Thus, the present study suggests that facial expression categorization is not a static process. Rather, the temporal

relation between the preceding and current

expressions could inform categorization, revealing a more dynamic and adaptive aspect of facial expression

(3)

processing.

(4)

In daily social life, humans continuously decipher the emotional cues provided by others to respond properly. A particularly important source of such cues is facial expressions. To rapidly and effortlessly make sense of multifarious and fast-changing facial expressions, perceptual categorization is critical in order to simplify the task of interpretation. Previously, two major opposing theories have been proposed to account for categorical processing of facial expressions. According to the discrete-category view (Calder, Young, Perrett, Etcoff, & Rowland, 1996; Ekman, 1992; Young, et al., 1997), facial expressions are perceived as belonging to qualitatively discrete categories. Those categories comprise the innate “basic” emotions that are found universally in humans, including anger, fear, sadness, happiness, disgust and surprise. The main alternative to the discrete-category view is the dimensional account (Russell, 1980, 1997), which posits that facial expressions are perceived as varying continuously along two continuous underlying dimensions - valence and arousal.

Facial expressions are highly dynamic signals that are rarely categorized as static, isolated displays. Rather, a facial expression is encountered as part of a sequence in which faces of various facial configurations and emotional categories are juxtaposed in a temporal order. Although the discrete-category and dimensional accounts hold different views about whether faces convey qualitatively distinct emotions or dimensional information regarding emotions, both focus on how categorical decisions are reached on single percepts, largely ignoring the role of sequential context in facial expression categorization. This neglect is not surprising

(5)

given that our ability to process facial expressions is thought to reflect part of our functional and neurobiological heritage (Darwin, 1872). It has been argued that facial expression processing, especially for threat-related expressions, is obligatory and independent of contextual modulation (Ekman, 1992; Lane & Nadel, 2000; Luo, et al., 2010; Pourtois, Spinelli, Seeck, & Vuilleumier, 2010; Russell, 1997; Vuilleumier, Armony, Driver, & Dolan, 2001).

A large body of evidence suggests that sequential context, or trial-to-trial transition more specifically, plays an important part in shaping behavioral responses. In absolute

identification tasks, it has been found that current stimuli are judged as closer to immediately preceding stimuli than they actually are – a bias called assimilation effects (Garner, 1953; Holland & Lockhead, 1968; Lacouture, 1997; Mori, 1989; Ward & Lockhead, 1970). For example, a neutral tone is judged as louder than it actually is if preceded by a loud tone. Intriguingly, such assimilation effects are reversed when the preceding stimuli are presented earlier in the trial sequence. Instead, contrast effects are observed in which the current stimuli are judged as away from the previous stimuli (Holland & Lockhead, 1968; Lacouture, 1997; Ward & Lockhead, 1970). For example, a neutral tone is judged as quieter than it actually is if preceded by a loud tone presented two or more trials earlier in the trial sequence.

More recently, evidence of contrast effects has been reported in categorization (Hampton, Estes, & Simmons, 2005; Stewart & Brown, 2004; Stewart, Brown, & Chater, 2002). In the

(6)

tasks, participants learn to categorize a group of stimuli that vary continuously along a certain dimension, yet are divided into two categories, such as a continuum of equally spaced 10 tones with the 5 lowest frequency tones in Category A (tones 1 to 5) and the 5 highest frequency tones in Category B (tones 6-10). The results show that categorization of the current stimuli near the category borderline (tone 5) is more accurate following the distant stimuli from the opposite category (tone 10) than following the distant stimuli from the same category (tone 1). Thus, participants seem to be biased to categorize current stimuli,

particularly those whose absolute magnitude information is less readily available, as away from the category of the previous stimuli. In more intuitive terms, the participants tend to believe that the preceding and current stimuli belong to different categories. Moreover, Stewart and Brown (2004) have found that the preceding stimuli that are presented two trials back in the trial sequence could also produce such contrast effects. In contrast, evidence of assimilation effects in categorization remains obscure. The tendency for participants to be biased to categorize the current stimulus as close to the category of the preceding stimulus has either not been found (Hampton, et al., 2005), or has occurred only when the preceding and current stimuli are from different categories (Zotov, Jones, & Mewhort, 2011). One study (Jones, Love, & Maddox, 2006) has reported that assimilation effects seem to depend on performance feedback and are evident only when successive stimuli are similar.

(7)

averaged over trials, thereby possibly removing potential sequential effects. The current study aimed to investigate the fine temporal structure of the data on a trial-to-trial basis to gain a new source of insight into the mechanisms of facial expression processing: Does sequential context affect the categorization of complex stimuli, such as facial expressions? If the influence of sequential context is observed, what is the underlying mechanism and do the observed sequential effects exhibit in the same manner as those that have been previously reported? Two aspects of sequential effects were specifically investigated to probe their presence in facial expression categorization: (1) contrast effects in which current stimuli are categorized as further from the category of the preceding stimuli than they actually are and (2) assimilation effects in which current stimuli are categorized as closer to the category of the preceding stimuli than they actually are. Participants in this study performed a binary categorization task in which the physical features of the facial expression stimuli morphed continuously between two emotion categories – fear and disgust. In line with prior literature, the stimulus continua used here also seem to be uni-dimensional, as evidence (Calder, et al., 2000) has shown that different levels of morphed facial expressions vary linearly with

participants’ ratings of emotional intensities on those morphed faces as opposed to the ratings of other dimensional properties. The emotion categories of fear and disgust were chosen given that their neural mechanisms have been explored thoroughly by neuropsychological and fMRI investigations, which show that the neural processes of fear and disgust are

(8)

represented in dissociable areas of the brain (Adolphs, Tranel, Damasio, & Damasio, 1994; Morris, et al., 1996; Phillips, et al., 1997). Lastly, the expressive faces were randomly presented and no performance feedback was provided to mimic typical experiments in facial expression categorization.

Experiment 1 – same-identity condition

Method

Participants. Fifteen right-handed participants without past neurological or psychiatric

history participated in this experiment (13 females, mean age = 21.13 years, range = 18 – 35). All had normal or corrected-to-normal vision and provided their written informed consent.

Stimuli. Ten continua of morphed facial expressions from fear to disgust were

created using FantaMorph (Abrosoft). In each continuum, a disgusted prototype was morphed 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80% and 90% of the physical distance to an identity-matched fearful prototype, resulting in 11 face images (i.e., fearful and disgusted prototypes, 90:10, 80:20, 70:30, 60:40, 50:50, 40:60, 70:30, 80:20 and 10:90 fear-disgust morphed faces). The prototypical expressions of fear and disgust were selected from FEEST (Young, Perrett, Calder, Sprengelmeyer, & Ekman, 2002). A total of 110 face stimuli were used (10 continua of different identities × 11

(9)

stimuli per continuum). The face images subtended a horizontal visual angle of 6.8° and a vertical angle of 8.6° around the center of the screen. The stimulus presentation was controlled by Psychtoolbox (Brainard, 1997; Pelli, 1997), and the viewing distance was 60 cm.

Procedure. Each trial began with a 600-ms fixation cross located in the center of the

screen, followed by a 400-ms presentation of a facial expression. A blank screen was presented at the offset of the face stimulus and participants were instructed to categorize whether the face was fearful or disgusted via a key press with no time limitation. Performance feedback was not provided during the experiment. The key press initiated a new trial after a 500-ms inter-trial interval. Trials were blocked by continua. In other words, participants had to complete 10 blocks, with a break between blocks. Within each block, the order of face stimuli from the same continuum was randomized. Each face was repeated 9 times, resulting in a total of 99 trials in each block (9 repetitions x 11 expressions per continuum). To acquaint the participants with the procedure, the experiment began with 1-2 blocks of practice trials, with different sets of face continua not being used in the experiment.

Results and discussion

Categorization data of expression continua. For each expression continuum,

(10)

or “disgust” emotion category for each morphed face (Fig. 1). Responses to stimuli at the same morph steps were averaged, irrespective of their sequential context. Although the exact data patterns varied across continua and participants, a highly consistent picture emerged. In accord with previous evidence (Calder, et al., 1996; Etcoff & Magee, 1992), the

categorization data of each continuum fell into two clear regions with an abrupt category shift, and each region belonged to the emotion category that corresponded to the prototype at that end. In general, a morphed face blended with more elements of fear or disgust from the prototypes, i.e., a smaller distance between the prototype and the morphed face, was more likely to be categorized as “fear” or “disgust”, respectively.

General sequential effects. To control for the variability in the locations of the

fear-disgust borderline across continua and participants so as to properly examine the effect of sequential context, we chose to analyze three types of face images from each emotion category: the prototypes (P-face), the morphs close to the category boundary (B-face), and the morphs lying at the mid-points of the P-and B-faces (M-face). These faces were all highly recognizable and were judged as belonging to a distinct emotion category with categorization rates above 77.78% for each continuum and each individual (P-face: mean ± SEM = 95.59 ± 0.88%, after collapsing across continua, emotion categories, and participants; M-face: 94.56 ± 0.66%; B-face: 87.95 ± 0.73%). The grey bars in Fig. 1 illustrate an example of how these target faces were selected.

(11)

Fig. 2A shows how the “accurate” (dominant) categorization responses to the three types of current faces varied as a function of the six different preceding stimuli: P-, M-, B-faces from the same emotion category (white zone), and P-, M-, B-faces from the opposite category (grey zone). It is important to note that categorization performance was averaged across the two emotion categories. The results showed that categorization of the current facial

expressions differed according to the preceding stimulus types (current P-face: one-way repeated measures ANOVA, F(5, 70) = 2.87, p < 0.05; M-face: F(5, 70) = 2.34, p = 0.05; B-face: F(5, 70) = 9.99, p < 0.001), indicating that categorical judgments of the current expressions depended on the local sequential context provided by the immediately preceding trials.

Additionally, we explored whether such sequential effects were limited to the effects of the immediately preceding stimuli, or whether the stimuli presented earlier in the trial sequence, such as two trials back, still had an impact on the categorization responses to the current expressions. As shown in Fig. 3A, the sequential effects derived from the preceding expressions presented up to two trials back were diminished for all three types of current faces (P-face: one-way repeated measures ANOVA, F(5, 70) = 0.54, p = 0.75; M-face: F(5, 70) = 1.71, p = 0.33; B-face: F(5, 70) = 0.81, p = 0.55). Taken together, general sequential effects were only observed between two successively presented expressions, with the effects significantly reduced when there was a two-trial gap between the current and preceding

(12)

expressions.

In an additional analysis, we investigated whether the general sequential effects could be observed individually for fearful and disgusted current faces. Similar to the results obtained when both emotion categories were combined, categorical responses to either fearful or disgusted current faces were affected by the immediately preceding stimuli (fearful current P-face: one-way repeated measures ANOVA, F(5, 70) = 2.78, p < 0.05; M-face: F(5, 70) = 1.06, p = 0.39; B-face: F(5, 70) = 4.29, p < 0.01; disgusted current P-face: F(5, 70) = 2.01, p = 0.09; M-face: F(5, 70) = 4.73, p < 0.001; B-face: F(5, 70) = 3.02, p < 0.05), suggesting that the sequential effects were not driven by any particular emotion category. Furthermore, the sequential effects observed in the two emotion categories were comparable in general, given that no significant interaction effects between categorization performance of fearful and disgusted current expressions were found when the current stimuli were the P-faces (2x6 repeated measures ANOVA, F(5, 70) = 0.38, p = 0.86) or the B-faces (F(5, 70) = 1.55, p = 0.19). However, a significant interaction effect was found for the M-faces (F(5, 70) = 3.8, p < 0.01). Additionally, the preceding stimuli from two trials back did not significantly bias the judgments of fearful and disgusted current expressions (fearful P-face: one-way repeated measures ANOVA, F(5, 70) = 0.70, p = 0.62; M-face: F(5, 70) = 1.93, p = 0.10; B-face: F(5, 70) = 0.95, p = 0.45; disgusted P-face: F(5, 70) = 0.74, p = 0.59; M-face: F(5, 70) = 0.24, p = 0.95; B-face: F(5, 70) = 0.35, p = 0.88). It should be emphasized here that some of the data

(13)

points were based on as few as two trials when the fearful and disgusted faces were studied individually. As a consequence, we restricted the following analyses to the data collapsed across emotion categories.

Same- and different-category transitions. To further understand the nature of the

sequential effects in facial expression categorization, we investigated how the different types of preceding stimuli affected categorization performance. More specifically, we assessed whether the sequential effects were different depending on whether the preceding and the current stimuli had the same or different emotion category memberships. In addition, trend analyses were conducted separately for each type of current expressions to qualitatively estimate whether the relative distance between the preceding and current expressions contributed to the effects in a linear manner. To this end, the accurate categorization responses to the current expressions were reorganized according to their distance from the preceding expressions, as illustrated in Fig. 4.

When the preceding and current expressions were from the same emotion category (Fig. 4A, white zone), categorization responses to the current B-faces were less accurate, with increasing relative distances between the preceding and current expressions (F(1, 28) = 6.70, p < 0.05). For the current M- and P-faces, no significant linear relationship between

categorization responses and relative distances was found (M-face: F(1, 28) = 0.10, p > 0.05; P-face: F(1, 28) = 4.12, p > 0.05). When the preceding and current expressions were from

(14)

different categories (Fig. 4A, grey zone), the categorical judgments of the current B-faces

were more accurate, with increasing relative distances between the preceding and current expressions (F(1, 28) = 36.18, p < 0.001). There was not a significant fit to a linear function for the current M-faces (F(1, 28) = 4.16, p > 0.05) or P-faces (F(1, 28) = 0.22, p > 0.05).

Taken together, the data showed that there was decreased accuracy to the current B-faces after more distant expressions from the same category or increased accuracy to the current B-faces after more distant expressions from the opposite category. This pattern provides evidence for contrast effects when the relative distance between the preceding and current expressions was increasingly large, given that the current stimulus was more likely to be judged as away from the category of the distant preceding stimulus. However, a

complementary indication of the present results is that responses to the current B-faces were more accurate after more nearby preceding expressions from the same category, or more errors were induced after more nearby preceding expressions from the opposite category. In this view, when the relative distance was increasingly small, the current stimulus was judged as close to the category of the nearby preceding stimulus. This suggests that the findings of this experiment could be alternatively explained in terms of assimilation effects. Which explanation is more compatible with our overall findings will be discussed below.

Between-category comparisons. To better understand the role of the relative distance in the

(15)

assimilation effects, contrast effects or both, we examined responses to the current expressions that were preceded by a member of the opposite category relative to after preceded by a member of the same category. These analyses are in accord with those performed by Stewart et al. (2002).

As was evident from the findings in the previous section, contrast effects could be observed when the relative distance between successive expressions was large. We thus expected that participants would classify current stimuli as further from the category of the

distant preceding stimuli, leading to increased accuracy if preceded by the distant stimuli

from the opposite category but to decreased accuracy if preceded by the distant stimuli from the same category. Indeed, the present results showed higher accuracy on the current B-faces after the distant P-faces from the opposite category (P→B pair: 93.59 ± 1.61%, solid circle in Fig. 4A) than after the distant P-faces from the same category (P→B pair: 82.94 ± 2.80%, dashed circle; paired t-test, t(14) = 4.33, p < 0.001). Given that our previous findings also revealed a possible involvement of assimilation effects when the relative distance was small, we expected that participants would classify current stimuli as close to the category of the

nearby preceding stimuli. As a result, accuracy would be increased if preceded by the nearby

stimuli from the same category, but more errors would be induced if preceded by the nearby stimuli from the opposite category. As expected, responses to the current B-faces were more accurate after the nearby M-faces from the same category (M→B pair: 90.54 ± 1.44%, solid

(16)

square in Fig. 4A) than by the nearby B-faces from the opposite category (B→B pair: 75.5 ± 2.53%, dashed square; t(14) = 4.67, p < 0.001). The between-category comparisons

confirmed previous analyses and further showed that both contrast and assimilation biases seemed equally involved in the sequential effects, depending on whether the relative distance was large or small. The data for the current M- and P-faces were not analyzed because they did not serve as proper candidates for between-category comparisons due to a lack of well-defined distant preceding stimuli of the same category for the M-faces and a lack of well-defined nearby preceding stimuli of the opposite category for both the M- and P-faces.

Summary. During categorization of a sequence of randomly presented expressions with

the same identity, categorization performance regarding current expressions was influenced by the local sequential context provided by previous trials. Such sequential effects were observed not only for the current morphed and highly recognizable expressions, but also for the prototypes to some extent. However, the sequential effects only occurred between successive expressions, as expressions presented up to two trials earlier in the sequence had no significant impact on the responses to current expressions. Intriguingly, the relation between two successive stimuli seemed to determine how current expressions, particularly the B-faces, were categorized. Two types of categorization biases were likely to be equally involved regardless of whether the preceding and current expressions had the same or different category memberships. On one hand, when the relative distance between two

(17)

successive expressions was relatively large, participants were biased to categorize the current expressions as away from the category of the immediately preceding expressions (contrast effects). On the other hand, when the relative distance was relatively small, participants were biased to assimilate responses towards the category of the preceding expressions

(assimilation effects).

Rather than being construed as supporting the influence of the local sequential context, our findings could simply reflect a response bias. For instance, participants might be biased against making two identical responses in a row. On this account, accuracy on current expressions would be the lowest after presenting the preceding P-faces from the same category, given that the P-faces have the highest correct categorization rates. By the same token, accuracy would be the highest after the presentation of the preceding P-faces from the

opposite category. Such predictive patterns are clearly not fully compatible with the

categorization data for the current M- and P-faces (Fig. 2A). Moreover, if the findings could be explained in terms of response-repetition bias, the categorization data of all three types of current expressions should have an identical pattern, irrespective of their perceptual

properties. Our data, however, showed that the patterns of categorization for all three types of current expressions were not identical (3 x 6 repeated measures ANOVA analysis on the interaction effect, F(10, 140) = 5.05, p < 0.001).

(18)

mechanism, such as adaptation effects or emotional priming effects? Although adaptation effects in facial expression recognition involve a biased recognition of a current expression after a period of stimulation from preceding material, evidence has shown that adaptation effects are fleeting (Hsu & Young, 2004). In contrast to the experimental procedure adopted in this study, prolonged presentation of a preceding item is usually required to generate robust adaptation effects. Moreover, during the adaptation times, participants need to attentively inspect the preceding stimulus without performing any cognitive task. Emotional priming effects are also unlikely to be the major source of the sequential effects. If the present data were due to emotional priming effects, improved categorization accuracy regarding current expressions would only have been observed following the presentation of stimuli from the same emotion category (Carroll & Young, 2005). However, we failed to find such a pattern in the current results.

Experiment 2 – different-identity/same-identity condition

The data from Experiment 1 provided evidence for sequential effects between successive expressions with the same identity. Traditionally, the processes of facial identity and facial expression have been thought to involve in separate, independent visual routes (Bruce & Young, 1986; Haxby, Hoffman, & Gobbini, 2000). However, some forms of dependencies between these two facial components have also been reported (Ganel & Goshen-Gottstein,

(19)

2004; Schweinberger & Soukup, 1998). Experiment 2 therefore aimed to investigate whether the dynamic nature of facial expression categorization as captured by the observed sequential effects could be generalized across different identities. To this end, morphed faces from continua of different identities were mixed and randomly presented. However, in the context of this design, a current face would also be preceded by a face with the same identity in some trials. Experiment 2 therefore provided an additional opportunity to test whether the

same-identity sequential effects obtained in Experiment 1 would still be evident when the experimental context differed.

Method

Participants. Fifteen right-handed participants without past neurological or psychiatric

history participated in the experiment (13 females, mean age = 20.13 years, range = 18 – 23). All had normal or corrected-to-normal vision and provided their written informed consent.

Stimuli and procedure. The stimuli and procedure were as in Experiment 1, with the

exception that the faces from two different sets of continua of different identities were

combined within one block. More specifically, eight continua of morphed faces were selected and organized into 4 pairs. In each pair, categorization performance along the two expression continua was highly correlated (all pairs: r >= 0.99, p < 0.001), based on the results from Experiment 1. This indicates that the patterns of categorization in each pair were comparable.

(20)

Participants had to complete 2 sessions on different days. Each session consisted of 4 blocks, one block for each pair. Each block had 2 runs. Participants had a break between runs and blocks. Within a run and a block, a face stimulus was randomly selected from one of the 4 pairs. Each face image was repeated 6 times, resulting in a total of 132 trials in one run (6 repetitions x 22 expressions per continuum pair).

Results and discussion

Categorization data of expression continua. As before, for each continuum,

categorization data were calculated as the percentage of choices corresponding to the “fear” or “disgust” emotion category for the individual morphed faces. The categorization data for each continuum also fell into two clear regions, each of which belonged to the emotion category corresponding to the prototype at that end. Further analyses showed that the same expression continuum, respectively studied in Experiment 1 and 2, had comparable patterns of categorization data, as categorization performance along the two expression continua was highly correlated (all continua: r >= 0.99, p < 0.001). This suggests that categorization performance along each expression continuum, after being averaged over trials without considering their sequential context, did not change in different experimental contexts. Because a current face could have been preceded by a face with either a different identity or with the same identity, the trials for which the preceding and current expressions had the

(21)

same identity or different identities were analyzed separately.

General sequential effects: different identities. Consistent with the analyses performed in

Experiment 1, for every continuum, three types of highly recognizable P-, M-, and B-face stimuli (categorization rates above 75% for each continuum and individual) from each emotion category were selected (P-face: mean ± SEM = 95.67 ± 0.58%; M-face: 92.53 ± 1.02%; B-face: 84.52 ± 0.65%). Fig. 2B depicts how accurate categorization responses to the three types of current faces, after being collapsed across emotion categories, varied as a function of the immediately preceding stimuli with a different identity. The results showed that categorization of the current facial expressions differed according to the preceding

stimulus types (current P-face: one-way repeated measures ANOVA, F(5, 70) = 4.17, p < 0.01; M-face: F(5, 70) = 5.44, p < 0.001; B-face: F(5, 70) = 3.01, p < 0.05). Despite the finding that sequential effects generalized across two successive expressions with different identities, the effects obtained here were distinct from the same-identity effects found in Experiment 1, as revealed by significant interaction effects between the categorization data of the current P-faces (2 x 6 mixed ANOVA, F(5, 140) = 2.21, p = 0.06), the M-faces (F(5, 140) = 5.85, p < 0.001) and the B-faces (F(5, 140) = 7.62, p < 0.001). As shown in Fig. 3B, the

different-identity sequential effects were limited to the two immediately successive stimuli, given that the stimuli presented two trials previously in the sequence did not have significant impact on the categorization responses to the current P-faces (F(5, 70) = 1.32, p = 0.27), the

(22)

M-faces (F(5, 70) = 0.53, p = 0.76) and the B-faces (F(5, 70) = 0.90, p = 0.49). To test whether the different-identity sequential effects were driven by any specific emotion category, the trials that presented fearful and disgusted expressions as the current stimuli were analyzed separately. Although some of the data points were based on as few as 2 trials, in general, both fearful or disgusted current faces were affected by the different types of immediately preceding expressions (fearful current P-face: one-way repeated measures ANOVA, F(5, 70) = 2.34, p = 0.05; M-face: F(5, 70) = 3.98, p < 0.01; B-face: F(5, 70) = 2.1, p = 0.07; disgusted current P-face: F(5, 70) = 2.55, p < 0.05; M-face: F(5, 70) = 1.98, p = 0.09; B-face, F(5, 70) = 1.53, p = 0.19). No significant interaction effect was found for the categorization data of fearful and disgusted current expressions (P-faces: 2x6 repeated measures ANOVA, F(5, 70) = 0.91, p = 0.48); M-faces: F(5, 70) = 1.5, p = 0.2; B-faces: F(5, 70) = 1.3, p = 0.27), suggesting that the different-identity sequential effects observed in the two emotion categories were comparable. Consistent with the results from Experiment 1, the preceding expressions in the two trials back did not significantly bias the judgments of fearful and disgusted current expressions (fearful P-face: one-way repeated measured ANOVA, F(5, 70) = 1.16, p = 0.34; M-face: F(5, 70) = 1.00, p = 0.42; B-face: F(5, 70) = 0.63, p = 0.68; disgusted P-face: F(5, 70) = 0.94, p = 0.46; M-face: F(5, 70) = 0.45, p = 0.81; B-face: F(5, 70) = 1.16, p = 0.34).

(23)

relative distance in the different-identity sequential effects, the trials for which preceding and current stimuli were chosen from the same emotion category were firstly analyzed. As shown in Fig. 4B (white zone), a linear relationship between relative distances and categorization performance was not found for all three types of current expressions (B-face: F(1, 28) = 0.11; M-face: F(1, 28) = 0.81; P-face: F(1, 28) = 1.13, all p’s > 0.05). On the contrary, when the preceding and current expressions had different category memberships (Fig. 4B, grey zone), categorization accuracy on the current P-faces was linearly improved or impaired with

increasing or decreasing relative distances between the current and preceding stimuli (F(1, 28)

= 13.22, p < 0.01). The finding from the different-category transitions could be accounted for by either a contrast effect or an assimilation effect, as explained on pp. 12 and 13. Which effect was the potential candidate will be discussed next. No significant linear trend was observed when the current expressions were the B-faces (F(1, 28) = 0.01, p > 0.05) or the M-faces (F(1, 28) = 2.95, p > 0.05).

Between-category comparisons: different identities. As in Experiment 1, responses to the

current B-faces were more accurate after the distant P-faces from the opposite category (P→B pair: 88.22 ± 2.50%, solid circle in Fig. 4B) than after the distant P-faces from the

same category (P→B pair: 84.82 ±1.92%, dotted circle), although the comparison was only

nearly significantly different (paired t-test, t(14) = 2.10, p = 0.06). This finding suggests that participants were biased to categorize the current expressions as away from the category of

(24)

the distant preceding expressions. In contrast to the results obtained in Experiment 1, the current expressions were also categorized as away from the category of the nearby preceding expressions. Improved accuracy was found when the current B-faces were preceded by the

nearby B-faces from the opposite category (B→B pair: 88.00 ± 2.77%, solid square) than by

the nearby M-faces from the same category (M→B pair: 79.75 ± 1.95%, dashed square; t(14) = 2.56, p < 0.05).

In sum, the overall findings point towards a consistent pattern: When the preceding and current stimuli had different identities as well as different category memberships, participants were likely to respond as if the current expressions were away from the category of the preceding expressions - a contrast effect revealed by the data from both the between-category comparisons and the different-category transitions. This contrast effect was enhanced when the relative distance between the preceding and current expressions was increasingly large, leading to increased accuracy on the current expressions after the more distant expressions of the opposite category, as revealed by the data from the different-category transitions.

General sequential effects: same identity. When the preceding and current expressions

had the same identity (Fig. 2C), the categorical judgments of the current expressions were found to differ as a function of the preceding stimuli when the current expressions were the B-faces (one-way repeated measures ANOVA: F(5, 70) = 17.58, p < 0.001) or the M-faces (F(5, 70) = 3.64, p < 0.001), but not when they were the P-faces (F(5, 70) = 1.59, p = 0.18).

(25)

In contrast to all of the previous findings, Fig. 3C shows that the preceding stimuli from two trials back in the sequence still had a significant impact on the categorization responses to the current P-faces ( F(5, 70) = 3.01, p < 0.05) and the B-faces (F(5, 70) = 2.71, p < 0.05), but not to the M-faces (F(5, 70) = 0.65, p = 0.67). However, this observation could be due to the influence from the immediately preceding expressions instead if it happened that the preceding expressions from two trials back were followed by the same type of immediately preceding expressions. For example, if the preceding stimulus from two trials back was a P-face from the same category, then the immediately preceding stimulus could also be the P-face from the same category. To rule out this possibility, the occurrences of each possible type of immediately preceding expressions were counted with regard to each type of

preceding expressions from two trials back for each participant. The data were then analyzed with Chi-Squared Goodness-of-Fit tests. The results did not support the aforementioned possibility, as no particular type of immediately preceding expression was presented more frequently for a given two-trial-back preceding expression (all p’s > 0.05 for all participants). The analyses suggest that the observed long-range sequential effects were directly affected by the stimuli presented two trials back in the sequence.

Additional analyses showed that when the current stimuli were the B-faces, the same-identity sequential effects observed in Experiment 2 had a distinct pattern of categorization compared to those observed in Experiment 1 (Fig 2C vs. 2A), which was

(26)

supported by a significant interaction effect in a 2 x 6 mixed ANOVA analysis (F(5,140) = 6.29, p < 0.001). No significant interaction was found when the current expressions were the P-faces (F(5,140) = 0.46, p = 0.8) or the M-faces (F(5,140) = 0.87, p = 0.5). As shown in Fig. 2B and 2C, the categorization data also had significantly different patterns for the

same-identity and the different-identity sequential effects obtained in Experiment 2 (current P-face: F(5,70) = 2.89, p < 0.05; M-face: F(5,70) = 6.66, p < 0.001; B-face: F(5,70) = 18.40, p < 0.001). Altogether, the sequential effects obtained in Experiment 1 and 2 were distinct from one another when the experimental contexts differed.

When current expressions of different emotion categories were investigated separately, the results showed that categorization performance on the fearful or disgusted current B-faces was affected by the different types of immediately preceding expressions (fearful current P-face: one-way repeated measures ANOVA, F(5, 70) = 1.16, p = 0.34; M-face: F(5, 70) = 1.71, p = 0.14; B-face: F(5, 70) = 3.62, p < 0.01; disgusted current P-face: F(5, 70) = 2.35, p = 0.05; M-face: F(5, 70) = 2.93, p = 0.05; B-face: F(5, 70) = 9.44, p < 0.001) No significant interaction effect was found between categorization performance of fearful and disgusted current expressions (P-faces: 2x6 repeated measures ANOVA, F(5, 70) = 1.39, p = 0.24; M-faces: F(5, 70) = 0.38, p = 0.86; B-faces: F(5, 70) = 1.22, p = 0.31) , suggesting that the same-identity sequential effects observed in the two emotion categories were comparable.

(27)

trials back had marginal effects on the categorization of fearful and disgusted current

expressions (fearful current P-face: one-way repeated measures ANOVA, F(5, 70) = 0.51, p = 0.77; M-face: F(5, 70) = 1.05, p = 0.40; B-face: F(5, 70) = 2.08, p = 0.08; disgusted current P-face: F(5, 70) = 2.11, p = 0.07; M-face: F(5, 70) = 0.45, p = 0.81; B-face: F(5, 70) = 2.19, p = 0.06). No significant interaction effect was found between categorization performance of fearful and disgusted current expressions when the preceding expressions were from two trials earlier (P-faces: 2x6 repeated measures ANOVA, F(5, 70) = 0.58, p = 0.72; M-faces: F(5, 70) = 1.360, p = 0.17; B-faces: F(5, 70) = 1.80, p = 0.12).

Same- and different-category transitions: same identity. When the preceding and current

expressions had the same category membership (Fig. 4C, white zone), categorization accuracy on all three types of current expressions did not vary linearly with their relative distances from the preceding expressions (P-face: F(1, 28) = 2.23; M-face: F(1, 28) = 1.07; B-face: F(1, 28) = 0.32, all p’s > 0.05). In contrast, when the preceding and current

expressions were from different emotion categories (Fig. 4C, grey zone), categorization accuracy on the current B-faces (F(1, 28) = 16.48, p < 0.001) and the M-faces (F(1, 28) = 4.84, p < 0.05), but not the P-faces (F(1, 28) = 0.65, p > 0.05), was linearly improved or

impaired with increasing or decreasing relative distances. Again, these findings could

indicate evidence either of a contrast effect with increasing relative distances or of an assimilation effect with decreasing relative distances (See below for further discussion).

(28)

As shown in Fig. 4D, a similar pattern of results was observed when the preceding expressions were from two trials back in the sequence. When the preceding and current expressions had the same category membership (Fig. 4D, white zone), no linear relationship between categorization performance and relative distances was found for any type of current expressions (P-face: F(1, 28) = 1.86; M-face: F(1, 28) = 0.01; B-face: F(1, 28) = 3.21, all p’s > 0.05). When the preceding and current expressions were from different emotion categories (Fig. 4D, grey zone), categorization accuracy on the current B-faces (F(1, 28) = 6.79, p < 0.05), but not the M-faces (F(1, 28) = 0.38, p >0.05) or the P-faces (F(1, 28) = 2.04, p > 0.05), was linearly improved or impaired with increasing or decreasing relative distances.

Between-category comparisons: same identity. In contrast with the different-identity

effects, participants seemed to categorize the current expressions as close to the category of both distant and nearby expressions that immediately preceded the current expressions (Fig. 4C). The between-category comparisons showed increased accuracy on the current B-faces following the distant P-faces from the same category (P→B pair: 89.72 ± 2.12%, dashed circle in Fig. 4C) than following the distant P-faces from the opposite category (P→B pair: 82.80 ± 2.56%, solid circle; paired t-test, t(14) = 2.11, p = 0.05). In a similar vein, responses to the current B-faces were more accurate after the nearby M-faces from the same category (M→B pair: 88.16 ± 2.43%, dashed square) than after the nearby B-faces from the opposite category (B→B pair: 70.57 ± 2.68%, solid square; t(14) = 7.06, p < 0.001).

(29)

When preceding expressions were from two trials back, the current B-faces were

categorized more accurately following the nearby M-faces from the same category (M→B pair: 87.48 ± 2.41%, dashed square in Fig. 4D) than following the nearby B-faces from the

opposite category (B→B pair: 79.93 ± 1.75%, solid square; t(14) = 2.16, p < 0.05). However,

analyses failed to show any significant accuracy differences for the current B-faces when they were presented after the distant P-faces from the same category (P→B pair: 82.94 ± 2.41%, dashed circle) compared to when they were presented after the distant P-faces from the

opposite category (P→B pair: 87.96 ± 2.43%, solid circle; paired t-test, t(14) = 1.32, p >

0.05). The overall results suggest that, similar to the effects between successive expressions, the current expressions were also categorized as close to the category of the nearby

expressions that preceded the current expressions in two trials earlier although such assimilation effects were reduced.

To account for the overall results obtained in the same-identity condition in Experiment 2, we suggest that when the preceding and current expressions had different category

memberships, participants were biased to respond as if the current expressions were close to the category of the previous ones - an assimilation effect revealed by the data from both the between-category comparisons and the different-category transitions. As revealed by the results from the different-category transitions, such assimilation effects were enhanced when the relative distance between the preceding and current expressions was increasingly small,

(30)

leading to increasingly impaired performance on current expressions following the presentation of the more nearby expressions of the opposite category.

Summary. Experiment 2 extended our previous findings, showing that sequential effects

may generalize across different identities. However, the different-identity sequential effects in Experiment 2 appeared to involve a categorization process distinct from the same-identity effects in Experiment 1. In Experiment 2, only contrast effects were observed when the preceding and current expressions were from different categories, and the effects were enhanced with increasing relative distances. As in Experiment 1, we also observed the

same-identity sequential effects in Experiment 2; however, the effects exhibited some distinct characteristics. First, only assimilation effects were observed when the preceding and current expressions were from different categories, and the effects were enhanced with decreasing relative distances. Second, the same-identity effects in Experiment 2 were more sustained, as the expressions presented up to two trials back could still exert their influence. This

long-range sequential effects suggest that there is a context-dependent selection of particular previous stimuli presented previously in the sequence, particularly when those stimuli could provide useful information for categorizing current stimuli (Petzold & Haubensak, 2001; Stewart & Brown, 2004).

(31)

The present study demonstrated that facial expression categorization was influenced by the local sequential context provided by previous stimuli. Two types of categorization biases were found: 1) assimilation effects in which current expressions were categorized as close to the category of the preceding expressions and 2) contrast effects in which current expressions were categorized as away from the category of the preceding expressions. However, different experimental contexts determined which bias might be involved. When participants categorized a sequence of expressions from the same identity (Experiment 1), assimilation effects occurred if the relative distance between the preceding and current stimuli was small, whereas contrast effects occurred if the relative distance was large. Both bias occurred regardless of whether the preceding and current expressions had the same or different category memberships. When participants categorized a sequence of expressions from two different identities (Experiment 2), only contrasts effects were observed if the preceding and current expressions had different identities as well as different category memberships, whereas only assimilation effects were observed if the preceding and current expressions had the same identity as well as different category memberships. Furthermore, the contrast or the assimilation effects observed in Experiment 2 could have been enhanced with increasing or decreasing relative distances, respectively.

Relevance to existing accounts of facial expression categorization

(32)

categorization. In contrast to the discrete-category account, we found that facial expressions were not perceived as belonging to qualitatively discrete categories, but instead, the categorization processes could be modulated by the local sequential context. The dimensional account is also not free from difficulties in accounting for our findings. According to this account, fearful expressions are more likely to be judged as “fear” after viewing disgusted expressions and vice versa because of shifts in the values of arousal and valence (Russell & Fehr, 1987). Accordingly, categorization accuracy would always be higher when the preceding and current expressions are from different categories than when the two are from the same category. However, the results of the same-identity sequential effects in both Experiment 1 and 2 clearly contradict this prediction. Moreover, the current study has shown that different experimental contexts may produce differential sequential effects with distinct natures, which is a result that the dimensional account can not fully explain.

Underlying mechanisms of the expression-based sequential effects

Several recent models have been proposed to account for sequential effects during supervised categorization, in which performance feedback is given. Although it remains unclear to what extent those models could be generalized to account for unsupervised categorization as shown in the current study (but see Hampton, et al., 2005), the models may still hint at the potential mechanisms underlying the expression-based sequential effects.

(33)

suggests that sequential effects are the result of a particular decision strategy in which relative difference information between successive items is used by participants to inform their categorization decisions (Stewart & Brown, 2005; Stewart & Morin, 2007). In this view, when two successive stimuli are similar, participants may believe that the current stimulus has the same category membership as that of the preceding stimulus (assimilation effects). When facing two successive stimuli that are quite dissimilar, participants tend to judge these two stimuli as belonging to different categories (contrast effect). Consistent with the model, the results of Experiment 1 showed assimilation effects when the relative distance between the preceding and current expressions was increasingly small (more similar) and showed contrast effects when the relative distance was increasingly large (more dissimilar).

Is the SD-GCM model also compatible with the results of Experiment 2? We suggest that, in facial expression categorization, participants estimate not only the differences in the perceptual attributes between the successive stimuli but also the differences in the emotional attributes of the stimuli. As a result, sequential effects may be observed even when the preceding and current expressions are taken from two sets of continua with different identities, yielding different-identity sequential effects.

In line with previous behavioral evidence showing that face identity interferes with expression categorization (Ganel & Goshen-Gottstein, 2004; Schweinberger & Soukup, 1998), we suggest that face identity information may have been involved in

(34)

similarity/dissimilarity comparisons between successive expressions in Experiment 2. If the preceding and current expressions have different identities, the expressions may appear more dissimilar due to their contrastive facial configurations. As a consequence, only contrast effects would be found, as is evident in the different-identity sequential effects. In contrast, if the preceding and current expressions have the same identity, the preceding and current stimuli may appear more similar when facial expressions are categorized in the context of multiple face identities. As a consequence, only assimilation effects would be observed, as is evident in the same-identity sequential effects. Given that assimilation of stimulus categories might be enhanced for faces of the same identity, this offers a potential explanation why assimilation effects persist for preceding expressions up to two trials back in the same-identity sequential effects found in Experiment 2. From the above perspectives, our findings lend additional support to the view of interdependencies between facial identity and facial expression. We suggest that the contextual information imposed by facial identity could shape the dynamic nature of facial expression processing.

In Experiment 2, the relative distance significantly contributed to the sequential effects only when the preceding and current expressions were from different categories. The nature of the mechanisms regarding why the similarity/dissimilarity comparison strategy did not operate on two successive stimuli from the same category remains unclear. In the different-identity condition, it was likely that the preceding and current expressions were

(35)

from different sets of continua so that facial features supporting expressions of the same emotion category differed. Outweighed by such featural differences, the preceding and current expressions of the same category may have appeared equally dissimilar, irrespective of the relative distance between the two. In the same-identity condition, given that the range along the fear-disgust emotional dimension in Experiment 2 was increased after combining two sets of continua, participants might have perceived the preceding and current expressions of the same category as equally similar, such that the impact of the relative distance was diluted.

Representation-shift account. Alternatively, our current findings could be explained in

terms of shifts in the participants’ internal representations or criteria of categories (Petrov & Anderson, 2005; Treisman & Williams, 1984; Zotov, et al., 2011). As illustrated in Fig. 5, there are two types of shifts. The first type is same-category shifts (Fig. 5, right panel): After presenting a stimulus from Category A, the center of the representation of Category A is shifted towards that stimulus. For example, the category representation is shifted right following the presentation of the P-faces (Fig. 5A, right panel) and shifted left after the B-faces (Fig. 5C, right panel). Because the position of the M-faces corresponds to the center of the category representation, the shift does not occur (Fig. 5B, right panel). The second type is different-category shifts (Fig. 5, left panel): After presenting a stimulus from Category A, the internal representation of Category B is pushed away.

(36)

What are the consequences of these shifts? Due to the same-category shifts, the position of the current B-faces of Category A, for example, is located outside the shifted representation of Category A if preceded by the distant P-faces of the same category (Fig. 5A, right panel). However, the position of the B-faces is still within the shifted representation if preceded by the nearby M- or B-faces of the same category (Fig. 5B and 5C, right panel). Therefore, in agreement with the results of the same-category transitions in Experiment 1, when the preceding and current expressions are from the same category, participants would be more likely to respond as if the current B-faces were away from the category of the more distant preceding expressions (i.e., contrast effects with increasing relative distances). A complementary indication is that participants would be more likely to respond as if the current B-faces were close to the category of the more nearby preceding expressions (i.e., assimilation effects with decreasing relative distances).

The results of the different-category transitions in Experiment 1 could be explained in terms of both the different-category and the same-category shifts. For example, due to the different-category shifts, the position of the current B-faces of Category B is located outside the shifted representation of Category B following the presentation of an item from Category A (Fig. 5, left panel). However, the position of the current B-faces of Category B falls within the shifted representation of Category A if preceded by the nearby B-faces of Category A (Fig. 5C) compared to if preceded by the distant M- or P-faces (Fig. 5A and 5B), due to

(37)

same-category shifts. As a result, when the preceding and current expressions are from different categories, the current expressions would also be categorized as close to or away from the category of the preceding expressions with decreasing or increasing relative distances.

Although the representation-shift account could account for our findings in Experiment 1, it is less clear how this account explains the findings in Experiment 2. It is possible that, in Experiment 2, participants’ internal representations of face identities and the range effect may interact with the category representations of facial expressions, yielding differential patterns of sequential effects.

Characteristics of the expression-based sequential effects

Both the SD-GCM model and the representation-shift account suggest that contrast and assimilation effects reflect a common bias in decision-making. A recent study, however, has argued that the locus of contrast effects operates during the early perceptual stage of stimulus processing (Jones, et al., 2006). Our findings appear to favor the decision-bias view, given that the contrast effects were found in the same-identity condition in Experiment 1, but not in Experiment 2, despite the perceptual information between two successive expressions being the same in both experiment (identical sequential contexts derived from identical sets of facial expression stimuli).

(38)

such as facial expressions. Moreover, Expression-based sequential effects exhibit a number of features that are distinct from those reported in previous research using simple, neutral

stimuli during category learning. First, the effects were labile, in general. The expressions presented two trials back had little impact on the categorical judgments of current expressions, except with regard to the same-identity effects found in Experiment 2. Second, when results across Experiment 1 and 2 are taken into account, the experimental contexts, which include the face identities and the range effect, could completely change the nature of the sequential effects. Third, the sequential effects could be observed not only for the current stimuli close to the category borderline but also for the current stimuli that were the prototypes to some extent (but see (Jones, et al., 2006). Lastly, in contrast to prior literature, assimilation effects could be robustly observed even when the preceding and current stimuli had the same category membership, as demonstrated in Experiment 1. However, it should be pointed out that different from most of prior research (but see Hampton, et al., 2005), no performance feedback was provided during categorization in this study. Therefore, further investigations are needed to clarify whether the aforementioned characteristics of the expression-based sequential effects are due to feedback-related effects or the intrinsic nature of facial expression stimuli.

(39)

The findings from the present study suggest that facial expression categorization is not a static process. Rather, the relation between the preceding and current expressions could provide a basis for categorization, revealing a more dynamic and adaptive aspect of facial expression categorization. Our findings are also in accord with a broader trend in recent research that shows that the perception of emotional faces can be influenced by various forms of contextual information, such as body posture (Aviezer, et al., 2008), language (Lindquist, Barrett, Bliss-Moreau, & Russell, 2006; Roberson, Davidoff, & Braisby, 1999), and scenes (Barrett & Kensinger, 2010; Righart & De Gelder, 2008). With this study, we contribute to this body of research by suggesting that temporal contextual information may interact in guiding our categorical decisions on emotional faces. Acknowledgment and consideration of this phenomenon in future research would provide a better understanding of the nature of emotional processing.

Acknowledgments

This work was supported by the National Science Council of Taiwan, R. O. C. (NSC100-2410-H-004-005-MY2) to SMH.

(40)

References

Adolphs, R., Tranel, D., Damasio, H., & Damasio, A. (1994). Impaired recognition of emotion in facial expressions following bilateral damage to the human amygdala.

Nature, 372(6507), 669-672.

Aviezer, H., Hassin, R. R., Ryan, J., Grady, C., Susskind, J., Anderson, A., et al. (2008). Angry, disgusted, or afraid? Studies on the malleability of emotion perception.

Psychological Science, 19(7), 724-732.

Barrett, L. F., & Kensinger, E. A. (2010). Context is routinely encoded during emotion perception. Psychological Science, 21(4), 595-599.

Brainard, D. H. (1997). The psychophysics toolbox. Spatial Vision, 10, 433-436. Bruce, V., & Young, A. (1986). Understanding face recognition. British Journal of

Psychology, 77, 305-327.

Calder, A. J., Rowland, D., Young, A. W., Nimmo-Smith, I., Keane, J., & Perrett, D. I. (2000). Caricaturing facial expressions. Cognition, 76(2), 105-146.

Calder, A. J., Young, A. W., Perrett, D. I., Etcoff, N. L., & Rowland, D. (1996). Categorical perception of morphed facial expressions. Visual Cognition, 3(2), 81-117.

Carroll, N. C., & Young, A. W. (2005). Priming of emotion recognition. Quarterly Journal of

Experimental Psychology Section a-Human Experimental Psychology, 58(7),

(41)

Darwin, C. (1872). The expression of the emotions in man and animals. London: John Murray.

Ekman, P. (1992). An argument for basic emotions. Cognition and Emotion, 6, 169-200. Etcoff, N. L., & Magee, J. J. (1992). Categorical perception of facial expressions. Cognition,

44(3), 227-240.

Ganel, T., & Goshen-Gottstein, Y. (2004). Effects of familiarity on the perceptual integrality of the identity and expression of faces: The parallel-route hypothesis revisited.

Journal of Experimental Psychology-Human Perception and Performance, 30(3),

583-597.

Garner, W. R. (1953). An informational analysis of absolute judgments of loudness. Journal

of Experimental Psychology, 48, 218-224.

Hampton, J. A., Estes, Z., & Simmons, C. L. (2005). Comparison and contrast in perceptual categorization. Journal of Experimental Psychology: Learning, Memory, and

Cognition, 31, 1459-1476.

Haxby, J. V., Hoffman, E. A., & Gobbini, M. I. (2000). The distributed human neural system for face perception. Trends in Cognitive Sciences, 4(6), 223-233.

Holland, M. K., & Lockhead, G. R. (1968). Sequential effects in absolute judgment of loudness. Perception & Psychophysics, 3, 409-414.

(42)

Visual Cognition, 11, 871-899.

Jones, M., Love, B. C., & Maddox, W. T. (2006). Recency effects as a window to

generalization: separating decisional and perceptual sequential effects in category learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32, 316-332.

Lacouture, Y. (1997). Bow, range, and sequential effects in absolute identification: A response-time analysis. Psychological Research-Psychologische Forschung, 60(3), 121-133.

Lane, R. D., & Nadel, L. (Eds.). (2000). Cognitive neuroscience of emotion. Oxford: Oxford University Press.

Lindquist, K. A., Barrett, L. F., Bliss-Moreau, E., & Russell, J. A. (2006). Language and the perception of emotion. Emotion, 6(1), 125-138.

Luo, Q., Holroyd, T., Majestic, C., Cheng, X., Schechter, J., & Blair, R. J. (2010). Emotional automaticity is a matter of timing. J Neurosci, 30(17), 5825-5829.

Mori, S. (1989). A Limited-Capacity Response Process in Absolute Identification. Perception

& Psychophysics, 46(2), 167-173.

Morris, J. S., Frith, C. D., Perrett, D. I., Rowland, D., Young, A. W., Calder, A. J., et al. (1996). A differential neural response in the human amygdala to fearful and happy facial expressions. Nature, 383(6603), 812-815.

(43)

Pelli, D. G. (1997). The VideoToolBox software for visual psychphysics: transforming numbers into movies. Spatial Vision, 10, 437-442.

Petrov, A. A., & Anderson, J. R. (2005). The dynamics of scaling: A memory-based anchor model of category rating and absolute identification. Psychological Review, 112(2), 383-416.

Petzold, P., & Haubensak, G. (2001). Higher orders sequential effects in psychophysical judgments. Perception & Psychophysics, 63, 969-978.

Phillips, M. L., Young, A. W., Senior, C., Brammer, M., Andrew, C., Calder, A. J., et al.

(1997). A specific neural substrate for perceiving facial expressions of disgust. Nature,

389(6650), 495-498.

Pourtois, G., Spinelli, L., Seeck, M., & Vuilleumier, P. (2010). Temporal precedence of emotion over attention modulations in the lateral amygdala: Intracranial ERP evidence from a patient with temporal lobe epilepsy. Cogn Affect Behav Neurosci,

10(1), 83-93.

Righart, R., & De Gelder, B. (2008). Recognition of facial expressions is influenced by emotional scene gist. Cognitive Affective & Behavioral Neuroscience, 8(3), 264-272. Roberson, D., Davidoff, J., & Braisby, N. (1999). Similarity and categorisation:

neuropsychological evidence for a dissociation in explicit categorisation tasks.

(44)

Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social

Psychology, 39, 1161-1178.

Russell, J. A. (1997). Reading emotions from and into faces. In J. A. Russell & J. M.

Fernandez-Dols (Eds.), The psychology of facial expressions (pp. 295-320). New York: Cambridge University Press.

Russell, J. A., & Fehr, B. (1987). Relativity in the perception of emotion in facial expressions.

Journal of Experimental Psychology: General, 116, 223-237.

Schweinberger, S. R., & Soukup, G. R. (1998). Asymmetric relationships among perceptions of facial identity, emotion, and facial speech. Journal of Experimental

Psychology-Human Perception and Performance, 24(6), 1748-1765.

Stewart, N., & Brown, G. D. A. (2004). Sequence effects in the categorization of tones varying in frequency. Journal of Experimental Psychology: Learning, Memory, and

Cognition, 30, 416-430.

Stewart, N., & Brown, G. D. A. (2005). Similarity and dissimilarity as evidence in perceptual categorization. Journal of Mathematical Psychology, 49(5), 403-409.

Stewart, N., Brown, G. D. A., & Chater, N. (2002). Sequential effects in categorization of simple perceptual stimuli. Journal of Experimental Psychology: Learning, Memory,

and Cognition, 28, 3-11.

(45)

multidimensional perceptual categorization: A test of the similarity-dissimilarity generalized context model. Quarterly Journal of Experimental Psychology, 60(10), 1337-1346.

Treisman, M., & Williams, T. C. (1984). A Theory of Criterion Setting with an Application to Sequential Dependencies. Psychological Review, 91(1), 68-111.

Vuilleumier, P., Armony, J. L., Driver, J., & Dolan, R. J. (2001). Effects of attention and emotion on face processing in the human brain: An event-related fMRI study. Neuron,

30, 829-841.

Ward, L. M., & Lockhead, G. R. (1970). Sequential effects and memory in category judgments. Journal of Experimental Psychology, 84, 27-34.

Young, A. W., Perrett, D. I., Calder, A. J., Sprengelmeyer, R., & Ekman, P. (2002). Facial expression of emotion: stimuli and test (FEEST). Bury St. Edmunds: Thames Valley Test Company.

Young, A. W., Rowland, D., Calder, A. J., Etcoff, N. L., Seth, A., & Perrett, D. I. (1997). Facial expression megamix: Tests of dimensional and category accounts of emotion recognition. Cognition, 63(3), 271-313.

Zotov, V., Jones, M. N., & Mewhort, D. J. K. (2011). Contrast and assimilation in

categorization and exemplar production. Attention Perception & Psychophysics, 73(2), 621-639.

(46)

Figure Captions

Figure 1. An example of the categorization data. The data were taken from one of fear-disgust

continua from a single participant. The expression continuum ranges from the prototypical expression of fear to the prototypical expression of disgust in 10 morphing steps. It should be noted that depending on the continua and individuals, the exact location of a category

boundary was between 70:30 and 30:70 fear-disgust morphs. The grey bars indicate the expression stimuli selected for analyzing sequential effects. The notations P, M and B represent the P-, M- and B-faces, respectively.

Figure 2. Categorization accuracy on current expressions as a function of different types of

immediately preceding expressions in (A) the same-identity condition in Experiment 1, (B) the different-identity condition in Experiment 2, and (C) the same-identity condition in Experiment 2. The data are collapsed across categories. The grey zone indicates that the preceding and current stimuli have different category memberships. Error bars represent ± SEM.

Figure 3. Categorization accuracy on current expressions as a function of different types of

preceding expressions presented two trials back in (A) the same-identity condition in

(47)

condition in Experiment 2. The data are collapsed across categories. The grey zone indicates that the preceding and current stimuli have different category memberships. Error bars represent ±SEM.

Figure 4. Categorization accuracy on current expressions as a function of their distance from

the immediately preceding expressions in (A) the same-identity condition in Experiment 1, (B) the different-identity condition in Experiment 2, (C) the same-identity condition in Experiment 2, and (D) categorization accuracy when the preceding expressions were from two trials back in the same-identity condition in Experiment 2. The data are collapsed across categories. For the same-category preceding-current expression pairs (white zone), a

zero-step distance represents a repetition. A one-step distance includes the M (preceding stimulus) →P (current stimulus), B→M, and M→B pairs. A two-step distance includes the B→P and P→B pairs. To better illustrate that the B→M and the P→M pairs have the same distance with opposite directions in reference to the same M-faces, a negative one-step distance is used to represent the P→M pairs. For the different-category pairs (grey zone), a one-step distance includes the B (preceding stimulus of the opposite category) →B (current stimulus) pairs. A two-step distance includes the B→M and M→B pairs. A three-step distance includes the B→P, M→M, and P→B pairs. A four-step distance includes the M→P and P→M pairs. A five-step distance includes the P→P pairs. For the current B-faces, the

參考文獻

相關文件

z Choose a delivery month that is as close as possible to, but later than, the end of the life of the hedge. z When there is no futures contract on the asset being hedged, choose

These findings suggest that when teaching small classes many teachers were still experimenting with different approaches, even after three years, while in normal classes there

In particular, in the context of folded supersymmetry it is pointed out in Ref.[4] that production of the squirk-antisquirk pair ˜ Q ˜ Q ∗ at the large hadron collider (LHC)

(2007) demonstrated that the minimum β-aberration design tends to be Q B -optimal if there is more weight on linear effects and the prior information leads to a model of small size;

There are three major types of personal finance products: mortgages, personal loans and credit cards. Mortgage is a long-term loan that is used for buying a

The continuity of learning that is produced by the second type of transfer, transfer of principles, is dependent upon mastery of the structure of the subject matter …in order for a

Dynamic programming is a method that in general solves optimization prob- lems that involve making a sequence of decisions by determining, for each decision, subproblems that can

* All rights reserved, Tei-Wei Kuo, National Taiwan University, 2005..