2 Categorization and Intensity of Chinese Emotion Words 2.1 Data and Design A total of 374 Chinese emotion words are taken from Xu and Tao ([4], cf. [

Size: px

Start display at page:

Download "2 Categorization and Intensity of Chinese Emotion Words 2.1 Data and Design A total of 374 Chinese emotion words are taken from Xu and Tao ([4], cf. ["

擂佟佳
5 years ago
Views:

1 Categorization and Intensity of Chinese Emotion Words Jingxia Lin 1 and Yao Yao 2 1 Nanyang Technological University, Division of Chinese, Singapore jingxialin@ntu.edu.sg 2 The Hong Kong Polytechnic University, Department of Chinese and Bilingual Studies, Hong Kong ctyaoyao@polyu.edu.hk Abstract. Despite the increasing interest in studying Chinese emotion words, there has been no reliable references in the published literature on the category (e.g. happiness, anger) and intensity (e.g. low, high) of emotion words in Chinese as perceived by native speakers. This study is the first to collect and analyze average language users perception of emotion category and intensity of Chinese emotion words. Results of this study will serve as an important reference for future research on language and emotion. Keywords: Chinese emotion words, emotion category, emotion intensity 1 Introduction In recent years, the research on Chinese emotion words and expressions has enjoyed increasingly high popularity and wide application in the fields of both linguistics and artificial intelligence (especially in automatic sentiment analysis; see [1], [2], among others). However, a fundamental problem in this line of research is the lack of reliable, widely-accepted ratings of emotion category (e.g. happiness, anger) and emotion intensity (e.g. low, high) for Chinese emotion words. Existing studies computed the measures either from a few researchers judgment or automatic computation ([3], [4], [5], [6]), neither of which can accurately represent the perception of average language users (cf. [5], [7], [8]). This study aims to fill the gap by collecting laymen s ratings of emotion category and intensity of Chinese emotion words from a large-scale online survey. In the current paper, we report results of the first stage of the study. 2 Data and Procedure Design of the current study mainly follows Strauss and Allen s [8] survey of English emotion words.

2 2 Categorization and Intensity of Chinese Emotion Words 2.1 Data and Design A total of 374 Chinese emotion words are taken from Xu and Tao ([4], cf. [6], [8]) 3 and randomly divided into four word lists. To control for rating reliability, each list also includes two or three words repeated within the list, two words repeated across all the lists and one pseudoword ( 几几 ), resulting in 98 items on each list. Table 1 shows all the repeated items within or across the lists. 厌倦, 悲痛 Repeated within list 1 担忧, 丧气 Repeated within list 2 惊惧, 惊喜 Repeated within list 3 愤慨, 愁闷, 惦念 Repeated within list 4 开心, 愤怒 Repeated across all lists 几几 Made-up word across all lists Table 1. Repeated items for testing rating reliability 2.2 Participants and Procedure Participants are 92 native Chinese speakers (mostly university students) from Mainland China. Each participant is randomly assigned to work on one word list adminis-tered on the Google Form platform. For each test item, the participant s task is to choose the most appropriate emotion category from seven basic emotion categories ([8], [4], [6]): happiness ( 喜 ), sadness ( 哀 ), anger ( 怒 ), fear ( 惧 ), surprise ( 惊 ), anxiety ( 焦虑 ), disgust ( 厌恶 ). If none of the basic emotion categories is appropriate, the participant may choose from the other three options: other emotions ( 其他情感类型 ), neutral/emotionless ( 中立 / 无情感色彩 ), and unfamiliar with the word ( 不理解词义 ). If the word is recognized as a true emotion word (i.e. one of the seven basic categories or other emotions), the participant also needs to rate its emotion intensity on a 7-point scale (1=basically no emotion; 7=very strong emotion). It should be noted that the 374 test words not only cover typical emotion words (e.g., 开心, 愤怒 ) but also words whose emotional content is arguable (e.g., 对不住, 痒痒 ; cf. [6]). Furthermore, the test materials also include words that are less commonly used or only used in certain dialectal regions (e.g., 背悔, 来劲 ). Thus, it is up to the participants to decide whether or not the test words are truly emotion words. The participants can choose neutral/emotionless ( 中立 / 无情感色彩 ) if they think a certain test word does not express any emotion, or unfamiliar with the word ( 不理解词义 ) if they do not understand the meaning of a test word. The number of participants assigned to work on each word list is given in Table 2. On average, each word is rated by 21.1 participants (s.d. = 3.3). 3 Xu and Tao s (2003) list has 390 words in total. Our study excludes 16 monosyllabic words that are listed as category names in Xu and Tao ([4]), e.g., 喜 and 悲.

3 Categorization and Intensity of Chinese Emotion Words 3 Word list Number of participants List 1 27 List 2 22 List 3 21 List 4 22 Table 2. Number of participants for each list 3 Results 3.1 Reliability Rating reliability within the rater is measured by the rating consistency of words repeated within the word list. For each intra-list word (n = 9), we first calculate the percentage of participants that rate the two occurrences of the word with the same category. On average, each item is rated consistently by 75% of the participants (range: 55% - 100%). Furthermore, correlation tests show that six out of the nine intra-list repetition items have consistent intensity ratings (correlation coefficient > 0.5) between the two occurrences and these six items cover all four lists (see Table 3). Taken together, these results indicate that the participants are in general quite consistent when rating emotional category and intensity. Word repeated twice within a list Correlation coefficient List 1 厌倦 0.12 悲痛 0.82 List 2 担忧 0.78 丧气 0.57 List 3 惊喜 0.40 惊惧 0.62 愁闷 0.74 List 4 愤慨 0.62 惦念 0.41 Table 3. Correlation coefficient of the intensity ratings of items repeated within lists The second type of rating reliability is the consistency across raters. Since each participant only works on one word list, word items repeated across lists ( 开心, 愤怒 ) are used to gauge rating consistency across different rater groups. All four groups of participants unanimously (>85%) categorized 开心 and 愤怒 as words of happiness and anger, respectively. Furthermore, T-test results show that the intensity ratings of 愤怒 did not vary significantly across lists (all p > 0.05). Cross-list differences did exist in intensity ratings of 开心, but the range of variation was small (between 4.3 and 5.5 on the 7-point scale) and in

4 4 Categorization and Intensity of Chinese Emotion Words all four lists, 开心 was recognized as a high-intensity emotion word (see section 3.2 for the classification of emotion intensity). The repeated pseudoword, 几几, is recognized most often as unfamiliar across groups ( %), followed by neutral/emotionless ( %). Altogether these two categories accounted for more than 70% of the ratings in all four groups. Compared with the English study, the Chinese pseudoword was less reliably recognized as a non-word (i.e. unfamiliar), probably because the Chinese pseudoword 几几 is associated with real words such as 磨磨叽叽 and 叽叽喳喳 by the participants. 3.2 Categorization and Intensity Results For each test word, we calculate the emotion categories with the highest and second highest percentage of raters as the primary and secondary emotion categories. Following Strauss and Allen [8], we divide the test words into two groups: representative emotion words and blended emotion words. Representative emotion words are defined as those whose primary emotion category is agreed by more than 70% of the raters and therefore can be considered as highly representative of the primary category, e.g., 开心 (happiness by 100%) and 厌倦 (disgust by 96.3%, sadness by 3.7%). Blended emotion words are those whose primary category is voted by no more than 70% of the voters and therefore express a mixture of more than one emotion type, e.g., 狂乱 (the two most-often-rated categories are other emotions by 31.82% and anxiety by 22.73%) and 懊恼 (the two most-often-rated categories are anger by 33.33% and sadness by 29.63%). Raw intensity ratings for true emotion words (i.e. primary category is one of the seven basic emotion types or other emotions) are averaged across all raters and then broadly classified as high (average intensity rating > 4) and low (average intensity < 4). By using a less fine-grained intensity scale, we reduce potential problems associated with within- and inter-rater inconsistencies. Category and intensity information of all the representative emotion words (n = 139) is listed in the Appendix. 4 Discussion 4.1 Findings of the current study As discussed above, current results from the survey generate a list of words that are highly representative of each emotion category (see Appendix for more). Table 4 below further presents a list of words that are unanimously voted (100%) into one emotion category and thus can be considered as core members of the category. The results of this study also suggest that words with lower intensity tend to be blended with multiple emotions. As shown in Table 5 below, emotion words with high intensity have roughly equal chances to be representative of an emotion category or blended. However, when emotion intensity is low, the odds of being representative against blended is even lower than 1:4.

5 Categorization and Intensity of Chinese Emotion Words 5 Category Words voted by all participants into the category disgust 腻烦, 讨厌 sadness 哀伤, 悲伤, 悲恸 happiness 赞赏, 欢乐, 欢欣, 乐意, 快乐, 愉悦, 喜悦, 欢娱, 欢快, 欢愉, 高兴, 惊喜, 欣喜, 喜欢, 欢 anger 愤怒, 忿怒, 暴怒 surprise 诧异, 吃惊, 惊疑, 惊愕, 震惊 fear, anxiety, NA neutral, unfamiliar Table 4. Highly representative emotion words High intensity Low intensity Representative Blended Table 5. Counts of representative and blended emotion words in different intensity conditions This study also finds six words that are rated as other emotions by more than 70% of the participants: 怜惜 (86.36%), 吝惜 (80.95%), 惭愧 (74.07%), 羞涩 (71.43%), 激昂 (71.43%), 羞怯 (70.37%). These ratings indicate that the seven emotions of this study (happiness 喜, sadness 哀, anger 怒, fear 惧, surprise 惊, anxiety 焦虑, disgust 厌恶 ) may not be sufficient to categorize emotions, and thus serve as a reference for the long-debated question regarding what emotion is primary or basic (cf. 24 categories in Xu and Tao [4], five in Lee [6], and six in Turner [9], [10]). 4.2 Comparison with previous studies Among others, Lee [6] also publishes a Chinese emotion word list, divided into five basic emotions (happiness, sadness, anger, fear, and surprise) and complex emotions as well as three intensity categories (high, moderate, and low). The emotion words in Lee [6] also come from Xu and Tao [4]. However, the categorization of emotion type and intensity in Lee [6] mainly followed Turner s ([11]) emotion classification model for English by mapping Chinese emotion words to the English emotion taxonomy. There are two potential problems with this approach. First, it can be hard (if not impossible) to find precise equivalents of emotion words across languages; second, even if cross-language mapping based on lexical meaning can be established, the corresponding words may carry different emotional connotation, resulting in nuanced differences in the perception of emotional content by language users. When compared with the word list in Lee [6], the current study reveals both similarity and significant differences. Take words of happiness in Lee [6] as an example. Lee [6] lists 37 emotion words in the happiness category, among which four are labeled as low intensity, four as high intensity and 29 as moderate intensity. Our results show that while the majority

6 6 Categorization and Intensity of Chinese Emotion Words of these words are indeed categorized as representative of happiness (i.e. agreed by more than 70% of the raters) in our study, nine items fall into the blended category. Furthermore, four items are rated as happiness by less than half of the raters: 怡和 (27.27%), 晓畅 (36.36%), 闲适 (40.91%), 放松 (45.45%). More differences are revealed when comparing intensity classification. While the words labeled as low-intensity in Lee [6] do receive a low average rating of 2.98 in our study, the high- and moderate-intensity items were rated with 4.78 and 4.72, respectively. A follow-up T-test revealed that the high- and moderate-intensity items are rated with similar intensity in our study (p > 0.1). In other words, the high- and moderate-intensity items in Lee s study are shown to be very similar in intensity in our study. 5 Conclusion and Future Studies To summarize, we report the first study to collect and analyze laymen s ratings of emotion category and emotion intensity of Chinese emotion words. Results of this study will serve as an important reference for future research on language and emotion, for instance, the possible relations among emotion category, emotion intensity, and the tones of the Chinese emotion words ([12]). In the next stage of the study, we will recruit more participants and also take into consideration factors that may affect perception of emotion words, including gender, age, and region (Mainland China, Hong Kong, and Singapore). References 1. Li, S., Wang, Z., Lee, S. Y.-M., Huang, C.-R.: Sentiment Classification with Polarity Shifting Detection. In: 2013 International Conference on Asian Language Processing (IALP), pp IEEE, Urumuqi (2013) 2. Xu, G.: The Construction and Application of Chinese Sentiment Lexicons. PhD dissertation, Peking University (2012) 3. Chang, L.-L., Chen, K.-J., Huang, C.-R.: Alternation across Semantic Field: A Study of Mandarin Verbs of Emotion. In: Yung, O-B. (ed.) Special Issue on Chinese Verbal Semantics. Computational Linguistics and Chinese Language Processing, vol. 5(1), pp (2000) 4. Xu, X., Tao, J.: The Study of Affective Categorization in Chinese. In: 1st Chinese Conference on Affective Computing and Intelligent Interaction, Beijing (2003) 5. Xu, L., Lin, H., Zhao,J: Construction and Analysis of Emotional Corpus. Journal of Chinese Information Processing, 22(1), pp (2008) 6. Lee, S. Y.-M.: A Linguistic Approach to Emotion Detection and Classification. PhD dissertation, the Hong Kong Polytechnic University (2010) 7. Nabi, R. L.: The Theoretical versus the Lay Meaning of Disgust: Implications for Emotion Research. Cognition and Emotion, 16(5), pp (2002) 8. Strauss, G. P., Allen, D. N.: Emotional Intensity and Categorisation Ratings for Emotional and Nonemotional words. Cognition and Emotion 22(1), pp (2008) 9. Turner, J. H.: The Evolution of Emotions in Humans: A Darwinian Durkheimian Analysis. Journal for the Theory of Social Behaviour, 26, pp (1996)

7 Categorization and Intensity of Chinese Emotion Words Turner, J. H.: Human Emotions: a Sociological Theory. Routledge, NewYork (2007) 11. Turner, J. H.: On the Origins of Hu-man Emotions: A Sociological Inquiry into the Evolution of Human Affect. Stanford University Press, California (2000) 12. Yao, Y., Lin, J., Huang, C.-R.: Lexicalized Emotion? - Tonal Patterns of Emotion Words in Chinese. In: 25th North American Conference on Chinese Linguistics, Ann Arbor (to appear) Appendix: Emotion intensity and category of representative emotion words Word N (# of Intensity Primary Category Secondary category raters) (broad) 乐意 27 High 喜乐 27 High 喜喜悦 22 High 喜喜欢 27 High 喜开心 27 High 喜快乐 22 High 喜惊喜 21 High 喜愉悦 21 High 喜欢乐 21 High 喜欢娱 27 High 喜欢快 21 High 喜欢愉 21 High 喜欢欣 21 High 喜欢 27 High 喜欣喜 21 High 喜高兴 22 High 喜热爱 27 High 喜其他情感类型爽心 26 High 喜不理解词义兴奋 22 High 喜其他情感类型欢喜 20 High 喜焦虑欢悦 21 High 喜惊喜爱 21 High 喜得意 21 High 喜其他情感类型畅快 26 High 喜其他情感类型幸福 20 High 喜中立快慰 20 High 喜不理解词义自豪 22 High 喜中立欢畅 22 High 喜其他情感类型满意 21 High 喜中立顺心 27 High 喜其他情感类型 continued on next page...

8 8 Categorization and Intensity of Chinese Emotion Words Table 6 continued from previous page Word N (# of Intensity Primary Category Secondary category raters) (broad) 舒畅 21 High 喜其他情感类型珍视 26 High 喜其他情感类型痛快 21 High 喜其他情感类型得志 27 High 喜其他情感类型舒坦 27 High 喜其他情感类型舒服 26 High 喜其他情感类型快活 22 High 喜其他情感类型来劲 22 High 喜中立赏识 21 High 喜其他情感类型推崇 25 High 喜其他情感类型景慕 24 High 喜其他情感类型珍爱 27 High 喜其他情感类型钟爱 21 High 喜其他情感类型亢奋 25 High 喜其他情感类型赞赏 18 Low 喜舒心 21 Low 喜其他情感类型可心 19 Low 喜不理解词义喜欢 21 Low 喜中立欣慰 20 Low 喜其他情感类型可意 26 Low 喜不理解词义称意 18 Low 喜中立称心 20 Low 喜其他情感类型忿怒 27 High 怒暴怒 21 High 怒愤懑 21 High 怒厌恶生气 22 High 怒厌恶激愤 26 High 怒其他情感类型愤怒 24 High 怒厌恶愤慨 22 High 怒其他情感类型忿恨 21 High 怒厌恶窝火 27 Low 怒焦虑吃惊 21 High 惊惊愕 21 High 惊惊疑 26 High 惊震惊 20 High 惊惊讶 21 High 惊哀愕然 27 High 惊其他情感类型惊骇 21 High 惊不理解词义惊诧 21 High 惊惧惊惶 27 High 惊焦虑 continued on next page...

9 Categorization and Intensity of Chinese Emotion Words 9 Table 6 continued from previous page Word N (# of Intensity Primary Category Secondary category raters) (broad) 惊奇 20 High 惊其他情感类型惊慌 20 High 惊惧诧异 25 Low 惊畏惧 22 High 惧其他情感类型恐惧 22 High 惧惊怕 27 High 惧其他情感类型畏怯 27 High 惧其他情感类型害怕 21 High 惧惊焦急 27 High 焦虑其他情感类型焦虑 22 High 焦虑惧心焦 21 High 焦虑其他情感类型焦炙 26 High 焦虑不理解词义着急 21 High 焦虑惊忧虑 19 High 焦虑哀焦躁 26 High 焦虑其他情感类型心急 21 High 焦虑其他情感类型烦燥 21 High 焦虑厌恶担忧 20 High 焦虑惊烦躁 20 High 焦虑其他情感类型担忧 20 High 焦虑惊烦躁 20 High 焦虑其他情感类型烦 21 High 焦虑厌恶发愁 20 High 焦虑哀不安 21 Low 焦虑惧烦恼 22 Low 焦虑其他情感类型烦闷 20 Low 焦虑厌恶挂虑 19 Low 焦虑其他情感类型烦心 27 Low 焦虑厌恶腻烦 27 High 厌恶讨厌 27 High 厌恶厌恶 21 High 厌恶哀厌烦 22 High 厌恶焦虑反感 22 High 厌恶焦虑鄙视 21 High 厌恶哀厌倦 27 High 厌恶哀鄙薄 25 High 厌恶其他情感类型憎恶 21 High 厌恶怒蔑视 27 High 厌恶其他情感类型轻蔑 21 High 厌恶其他情感类型鄙夷 25 High 厌恶其他情感类型 continued on next page...

10 10 Categorization and Intensity of Chinese Emotion Words Table 6 continued from previous page Word N (# of Intensity Primary Category Secondary category raters) (broad) 瞧不起 21 High 厌恶中立敌视 22 High 厌恶怒哀伤 27 High 哀悲伤 27 High 哀悲恸 22 High 哀哀痛 21 High 哀悲哀 22 High 哀其他情感类型伤心 27 High 哀厌恶哀愁 26 High 哀怒悲怆 20 High 哀不理解词义沉痛 21 High 哀中立苍凉 21 High 哀中立哀怨 21 High 哀厌恶悲痛 26 High 哀厌恶悲切 20 High 哀其他情感类型伤感 20 High 哀中立哀戚 20 High 哀不理解词义悲凄 20 High 哀其他情感类型悲凉 21 High 哀其他情感类型沮丧 22 High 哀焦虑心酸 21 High 哀其他情感类型悲壮 20 High 哀其他情感类型绝望 21 High 哀焦虑感伤 22 Low 哀其他情感类型遗憾 21 Low 哀其他情感类型惭愧 26 High 其他情感类型哀激昂 19 High 其他情感类型喜怜惜 20 Low 其他情感类型中立吝惜 18 Low 其他情感类型中立羞涩 17 Low 其他情感类型中立羞怯 23 Low 其他情感类型中立 Table 6: Emotion intensity and category of representative emotion words

國家圖書館典藏電子全文

國家圖書館典藏電子全文 i ii Abstract The most important task in human resource management is to encourage and help employees to develop their potential so that they can fully contribute to the organization s goals. The main