個人特質差異對不同中文詞類情緒詞處理之影響：事件相關腦電位研究

全文

(1)國立臺灣師範大學英語學系碩. 士. 論. 文. Master Thesis Department of English National Taiwan Normal University. 個人特質差異對不同中文詞類情緒詞處理之影響：事件相關腦電位研究. The Modulation of Personality on Chinese Emotion Word Processing with Different Lexical Categories: An ERP Study. 指導教授：詹曉蕙博士 Advisor: Dr. Shiao-hui Chan 研究生：古李全 Student: Li-Chuan Ku. 中華民國一零三年七月 July, 2014.

(2) 摘要詞彙內含的情緒意義可以影響人類對於字詞的處理歷程。過去研究發現，不同詞彙類別的字詞對於情緒效果有其調節和交互作用。但以一般人的情緒字詞處理而言，可以顯現個體差異原因之一的個人特質因素卻鮮少被納入討論。因此，本研究旨在以事件相關腦電位技術來探討個人特質中的外向性和神經質性是否會影響中文情緒詞處理。本實驗採用詞彙判斷作業並操控中文雙字詞之情緒價性(正向詞、中性詞、負向詞)和詞彙類別(狀態動詞、動作動詞、名詞)特徵。實驗結果顯示，情緒字詞影響早期的 P2 和晚期 N400 及 LPC 振幅特性。另外，受試者的神經質傾向在晚期的 N400 及 LPC 皆有調節作用，顯示此調節作用在不同詞彙類別字詞之情緒意含已被辨識之後才發生。與艾森克人格特質論不同的是，實驗結果似乎並無證據顯示受試者的外向性與字詞情緒價性的處理有關。另外，本實驗亦無觀察到早期的詞彙類別效果或不同詞彙類別間有不同時區情緒效果的早晚差異，推測此結果與中文母語人士存取詞彙類別的認知系統為一分散式架構有關，因而不同詞彙類別的中文情緒詞有相似的處理歷程。. 關鍵詞：情緒、字詞處理、詞彙類別、個人特質、事件相關腦電位. i.

(3) Abstract Emotion connotations embedded in words have been found to impact word processing. Past studies have also showed that words in different lexical categories would modulate and interact with emotion effects. However, personality traits, one of the variables demonstrating individual differences in emotion word processing in healthy populations, have seldom been taken into account. This study aimed to explore whether personality traits, extraversion and neuroticism in particular, influenced the processing of Chinese emotion words in a lexical decision task by manipulating both emotional (valence: positive, neutral, negative) and linguistic (lexical category: stative verbs, eventive verbs, nouns) features of Chinese two-word compounds with the event-related potential technique. The results showed that the emotion effects appeared in P2, N400, and LPC, and that the modulations of participants’ neuroticism level on emotion effects were observed in N400 and LPC, after the emotion meanings of words with different semantic complexity were already recognized. In contrast to Eysenck’s model of personality, there was no evidence of connections between extraversion and the processing of emotion valence. Furthermore, the lack of a lexical category effect in the early time window and the same onsets of emotion effects on words across different lexical categories might reflect similar processing of the emotion contents in these categories due to the shared underlying distributional structures in Chinese speakers’ conceptual system. ii.

(4) Keywords: emotion, word processing, lexical category, personality traits, event-related potentials (ERPs). iii.

(5) Acknowledgements It is really a pleasure that I can finally reach the stage to express my sincere gratitude to the people who have supported and instructed me in conducting my thesis. I have learned a lot through the process with the help from these people. Without them, I would never have been able to finish my thesis successfully. First of all, I would like to give my heartfelt thanks to my advisor, Prof. Shiao-hui Chan. It is Prof. Chan’s neurolinguistics class that made me start to appreciate the field of neuroscience in language which I would not even imagine before. Prof. Chan’s enthusiasm in this field and her tireless teaching always motivated me to better my research and to pursue more treasure in this field. Most importantly, were it not for Prof. Chan’s constant encouragement for both academics and life, I might not have the endless courage to insist on the way of doing research to date. I would also like to show my appreciation to my committee members, Prof. Hsueh-Chih Chen and Prof. Chia-ying Lee. Their professional and insightful advice definitely made the thesis more valuable. My gratitude also goes to the professors who have taught me at NTNU: Prof. Chun-yin Doris Chen, Prof. Charles Chien-jer Lin, Prof. Hsiao-hung Iris Wu, Prof. Hui-Shan Lin, Prof. Jin-lan Joy Wu, Prof. Jen-i Li, Prof. Kwock-ping Tse, Prof. Miao-lin Hsieh, Dr. Shu-Yen Lin, and Prof. Yung-o Biq by the alphabetic order. They not only broadened my horizons but also equipped me with the knowledge in different sub-fields of linguistics. In addition, I would iv.

(6) like to thank Prof. Ching-Po Lin, Prof. Denise Hsien Wu, Prof. Li-Fen Chen, Prof. Shih-Wei Wu, Prof. Wen-Jui Kuo, and Prof. Yawei Cheng at NCU and NYMU by the alphabetic order. Their professional instructions in neuroscience and magnetoencephalography (MEG) technique made me well-prepared not only in doing my research but also in analyzing the data. Furthermore, I would like to send my sincere thanks to all my classmates at NTNU: Alison, Andres, Becky, Carol, Debbie, Eric, Felix, Horace, Irene, Jeccie, Jessica, Jocelyn, and Joy. Their company and support really boosted my confidence in continuing the thesis writing. I would also like to express my appreciation to the members in LOPE lab at NTU led by Prof. Shu-Kai Hsieh: Ajax, Chan-Chia, Chin-Ju, Katherine, Pieere, Simon, Taco, Vicky, and Wallace. Prof. Hsieh and his lab members made me explore the beauty in computational linguistics and corpus linguistics, two fascinating fields in linguistics that can integrate the knowledge of computer science and linguistics I learned so that I could apply it practically. Their creativity and insightful perspectives towards linguistics often made me astonished and pushed me forward continuously. Special thankfulness also goes to all the anonymous participants who were willing to attend my ERP experiments without any complaint. Without their help, the research would not have been completed. Moreover, I am indebted to all the members in Neurolinguistics lab at NTNU: Elvis, v.

(7) Gracie, Helen, Jeff, Julia, Ken, Lillian, and Vivi, for their kind help and support in aiding me to carry out the ERP experiments. I would like to especially express my heartiest thanks to Gracie and Helen for their precious experience sharing in conducting data analysis, and to Ken for his encouragement from time to time when I was struggling with the thesis writing. I really learned a lot and felt lucky to be in such a lovely and warm lab. Finally and most importantly, I am very grateful to my family, especially to my dear mom and dad for their nonstop and generous support and confidence in me when I was writing my thesis. I would also give special thanks to my wonderful grandmother for her upbringing in my childhood, who is suffering from Alzheimer's disease with aphasia due to epilepsy recently. Without you all, I would not go that far on the way of pursuing and appreciating the knowledge in linguistics.. vi.

(8) Table of Contents. 摘要-------------------------------------------------------------------- i Abstract -------------------------------------------------------------------------------------------------- ii Acknowledgements ----------------------------------------------------------------------------------- iv Table of Contents ------------------------------------------------------------------------------------ vii List of Tables ------------------------------------------------------------------------------------------ ix List of Figures ------------------------------------------------------------------------------------------ x Chapter 1 Introduction --------------------------------------------------------------------------------- 1 1.1 Research Question --------------------------------------------------------------------------- 5 Chapter 2 Literature Review -------------------------------------------------------------------------- 6 2.1 Emotions in Psychology -------------------------------------------------------------------- 6 2.2 Emotions in Language ---------------------------------------------------------------------- 9 2.2.1 Emotion Classification in Western Languages ---------------------------------- 9 2.2.2 Emotion Classification in Chinese ----------------------------------------------- 14 2.2.3 Emotion Word Processing -------------------------------------------------------- 16 2.2.4 Trait Differences and Emotion Word Processing ------------------------------ 24 Chapter 3 Methodology ------------------------------------------------------------------------------ 29 3.1 Subjects -------------------------------------------------------------------------------------- 29 3.2 Material -------------------------------------------------------------------------------------- 30 3.3 Procedure ------------------------------------------------------------------------------------ 35 3.4 Data acquisition and analysis ------------------------------------------------------------- 37 Chapter 4 Result --------------------------------------------------------------------------------------- 40 4.1 Behavioral Data ----------------------------------------------------------------------------- 40 4.2 ERP Data ------------------------------------------------------------------------------------- 43 4.2.1 P2 ------------------------------------------------------------------------------------- 48 4.2.2 N400 ---------------------------------------------------------------------------------- 48 4.2.3 LPC ----------------------------------------------------------------------------------- 51 Chapter 5 Discussion --------------------------------------------------------------------------------- 53 5.1 Emotion Effects on Word Processing: P2, N400 and LPC --------------------------- 54 5.2 Lack of the Lexical Category Effect in Early Time Window ------------------------ 57 5.3 Modulation of Neuroticism in Processing Emotion Words in Later Time Windows ----------------------------------------------------------------------------------------------------- 58 5.4 Summary ------------------------------------------------------------------------------------- 60 Chapter 6 Conclusion --------------------------------------------------------------------------------- 61 6.1 Limitations of the Current Study --------------------------------------------------------- 62 References ---------------------------------------------------------------------------------------------- 63 vii.

(9) Appendix A: Short-scale EPQ-R -------------------------------------------------------------------- 69 Appendix B: Beck Depression Inventory-second edition (BDI-II) ---------------------------- 71 Appendix C: The rating questionnaire for stimulus valence ------------------------------------ 73 Appendix D: The rating questionnaire for stimulus arousal ------------------------------------ 74 Appendix E: The rating questionnaire for stimulus imageability ------------------------------ 75 Appendix F: A complete list of stimulus words -------------------------------------------------- 76. viii.

(10) List of Tables. Table 1. Descriptive statistics (mean (M) values with standard deviation (SD)) for controlled variables and rating results of the stimulus set ---------------------------------------------------- 34 Table 2. The EPQ-R Scores of the Participants --------------------------------------------------- 40 Table 3. Descriptive statistics (mean values with standard deviation (SD)) for RTs and Accuracy Rates of Word and Pseudoword Stimuli ----------------------------------------------- 41 Table 4. RTs and Accuracy Rates (mean values with standard deviation (SD)) in Each Experimental Condition for Participants with Different Extraversion Level ----------------- 42 Table 5. RTs and Accuracy Rates (mean values with standard deviation (SD)) in Each Experimental Condition for Participants with Different Neuroticism Level ----------------- 43. ix.

(11) List of Figures. Figure 1. Sequence of events under lexical decision task in this experiment----------------- 36 Figure 2. The Grand Average Waveforms of Positive, Neutral, and Negative Stimuli ------ 44 Figure 3. The Topography of ERP Differences (Unit: uV) between the Emotion (positive and negative) and Non-emotion (neutral) Stimuli at Indicated Time Intervals -------------------- 44 Figure 4. The Grand Average Waveforms of Stative Verb, Eventive Verb, and Noun Stimuli ----------------------------------------------------------------------------------------------------------- 46 Figure 5. The Grand Average Waveforms of Participants with Low vs. High Extraversion ----------------------------------------------------------------------------------------------------------- 47 Figure 6. The Grand Average Waveforms of Participants with Low vs. High Neuroticism ----------------------------------------------------------------------------------------------------------- 47 Figure 7. The Comparison Waveforms of All Participants in Conditions of Different Lexical Categories at the CZ Electrode ---------------------------------------------------------------------- 50 Figure 8. The Grand Average Waveforms of Participants with Low vs. High Neuroticism in Response to Negative Stative Verbs ---------------------------------------------------------------- 52. x.

(12) Chapter 1 Introduction Human emotion has been extensively studied in affective neuroscience. In daily communication, it is often found that personal emotions, moods, and even sentiments can be embedded into languages to express speakers' psychological states and to help listeners comprehend one’s mental feelings. While moods are longer extension of one’s affective states that can generally have multiple causes, emotions are relatively short in duration and often elicited by a particular event, both of which are very different from sentiments that signal one’s attitude and standard toward a particular object and thus need higher cognitive elaboration. In fact, the ability of appraising, expressing, and regulating emotions in the self and others during multimodal communication is a fundamental element in measuring one’s emotional intelligence (Salovey & Mayer, 1989). The accurate recognition of emotion thus increases individuals’ competency of emotion. Although previous psychological research focuses much on emotional stimuli such as facial expressions and pictures, the investigations of the corresponding neural activities in language comprehension with emotion words have just started to bloom recently (see Citron, 2012, for a complete review). It is showed that the emotion effects induced by word stimuli may be weaker due to the complexity of stimuli or subjects’ biological preparedness (Kissler et al., 2006); however, they are still qualitatively similar to the effect elicited by other forms of affective stimuli. Emotion, as a subjective experience resulting from these word stimuli, can trigger different mental, cognitive, or 1.

(13) physiological reactions depending on one’s mood, personality, or motivation in the experiment, and vary with different types of experimental tasks and verbal stimuli. Besides, emotion words in the discourse can shape the perceptual encoding of emotions, supporting the weak Sapir-Whorf hypothesis of linguistic relativity theory, in which language can influence one’s conceptualization of the world (Whorf, 2012). It was reported that in judging the morphed faces with an equal blend of happiness and angry, the face would be reported to be angrier if paired up with the word “angry,” a perceptual shift affecting emotion classification caused by language (Halberstadt & Niedenthal, 2001). However, few attempts have been made to take both linguistic and non-linguistic factors into account to get a clearer picture in emotion word processing researches. Language comprehension relies partly on a variety of lexical and semantic factors such as word frequency, word length, word familiarity, the number of word senses, the neighborhood density, the morphological family size, and so on. Among these factors, it is the semantic ones that help conceptualize the meaning of words and affect the way one perceives other cognitive mechanisms like emotion. According to Yap et al.’s (2012) article, these semantic factors mainly include (1) the number of features associated with the word’s referent, (2) the semantic neighborhood density, (3) the number of distinct associates elicited firstly by the word in a free-association task, (4) imageability, (5) the number of senses, and (6) the body–object interaction, the degree to which a human body can physically interact 2.

(14) with the word’s referent. Due to its computational complexity, few models of visual word recognition have focused on the role of semantic information. However, a potential attempt to reconcile this gap may lie in lexical categories (e.g. stative verbs, event verbs, and nouns), a marker to distinguish lexical-semantic representations by the inherent information these categories carry. For instance, nouns are typically used to denote objects, while verbs can predicate the properties and relationships relative to what is denoted by nouns. In emotion words, this noun–verb dichotomy can generally imply whether a word is to denote or to cause an emotion state, further implying the expressiveness of the emotion word. However, few studies emphasized the lexical category effects on emotion word processing, and the results are often inconsistent (Scott et al., 2009; Palazova et al., 2011). In addition to linguistic factors, personality traits, as markers of the stability and individuality of human behavioral patterns, can steadily describe the individual differences in emotion word processing within a longer period of time. It has been found that two personality traits, extraversion and neuroticism, are highly correlated or associated with human emotion (Bartussek et al., 1996; Canli et al., 2001; Kehoe et al., 2011) and it is common to observe people with these traits demonstrating different stylistic affective patterns in daily speeches or texts such as product reviews, advertising campaigns, political criticism, or other opinion articles. For instance, Pennebaker and King (1999) analyzed written texts and found significant correlations between the linguistic features of the texts and 3.

(15) the writers’ personality traits: neurotics use more negative words but fewer positive words while extroverts show the reversed pattern, exhibiting more agreement, compliment, and self-references compared with introverts. These correlations of personality traits can be seen not only in the production of written texts but also in the comprehension of affective paragraphs. Studies showed that words can be linked to the conceptual knowledge of our emotions which stems from prior experiences and is re-constructed in perception (Gendron et al., 2012), and the emotional percepts can be influenced by perceptual differences in word stimuli modulated by individuals with various traits (Matthews et al., 1995; Canli et al., 2001; Kehoe et al, 2011). Although Pennebaker and King (1999)’s development of Linguistic Inquiry and Word Count (LIWC) program led to a series of text analysis on emotion paragraphs to pinpoint the stylistic differences of language production between individuals based on computational metrics and machine learning, there seems no acknowledged ways to evaluate individuals’ affectivity during one’s comprehending of language. By investigating the influences of extraversion and neuroticism on emotion word comprehension with neurophysiological tools, EEG/MEG, we hope to extend our understanding of how people of different personalities process such words in human brain. Also, by manipulating the expressiveness of emotion words with linguistic features in word recognition paradigm, we can better understand the perceptualization and semantic encoding of emotion words, which can in turn be beneficial for future application on the analysis of 4.

(16) longer discourses or narratives and for other individualized context-awareness and affect-detection systems that can sketch the source of emotions and quantify the affectivity through emotion contextual information in written texts.. 1.1 Research Question This study aims to examine whether one’s personality trait, extraversion and neuroticism in particular, influences his or her processing of emotion words in the word recognition paradigm. The emotion words will be manipulated not only by traditional emotional features (valence and arousal) but also by an important linguistic factor that was proposed by theoretical linguists but rarely explored in empirical studies--lexical categories (e.g. stative verbs, event verbs, and nouns). Different lexical categories lead to different semantic complexity, and it would be informative to examine whether variance in semantic complexity can induce different possessing under the modulation of personality in emotion word processing.. 5.

(17) Chapter 2 Literature Review 2.1 Emotions in Psychology Emotion is a psychological status elicited by specific stimulus, whether it is from the inner body (e.g. the wound), the mind (e.g. the stress), or the outer environment (e.g. the blood), and can cause a series of physiological, behavioral, and cognitive reactions. Always being an interest of psychologists, the origin, the experience, and the regulation of emotion comprise the most of contemporary psychological research of emotions. Adopting the embodiment theory of emotions, James and Lange (1884) argued that emotion is secondary to and the result of physiological response to stimuli. It is the bodily reaction that modulates the experiences of emotion. Cannon-Bard’s (1927) theory of emotion also agreed on the importance of the bodily response, but emphasized the simultaneity of physiological reactions and the recognition of emotions via stimulation of thalamus. Supporting these somatic theories, the two-factor theory of emotion by Schachter and Singer (1962) stated that human’s experiences of emotion contain two stages: physiological arousal and cognitive appraisal of the arousal. It is the latter one that defines people’s subjective emotional experiences and makes the different emotions towards the same physiological response possible. In the meanwhile, there was not a comprehensive approach to illustrate how emotion contents are classified after being conceptualized.. 6.

(18) Ekman’s (1971) emotion model was the first to claim that there are basic and universal emotions based on the observation of the similarity between facial expressions. According to him, emotion can be classified into six basic categories: anger, fear, disgust, happiness, surprise, and sadness, which are fundamental to human’s solution to adaptive problems in the outer world, and it engages certain nervous circuits that are universal in different species or cultures. This discrete emotion approach thus considers certain emotion categories as universal biological states that are triggered by evolutionarily preserved “affect programs”. The “affect programs” are expressed as unambiguous bio-behavioral signals and recognized by the hardwired and universal mental mechanism with which people are born in possession of five to six perceptually grounded emotions suggested (Ekman & Cordaro, 2011). Criticized by lacking neurological evidence, the discrete emotion approach was totally rejected in alternative constructionist models to classify emotions based on the two-dimensional structures of emotions: valence and arousal. Valence generally describes the degree to which the emotion is pleasant or unpleasant, and therefore elicits appetitive or defensive motivational system. Arousal, on the other hand, specifies the extents of activation an affect elicits. Often, stimuli with positive or negative valence elicit a higher arousal level than neutral stimuli. Although whether the two dimensions of emotion are dependent or not may still be under debate (Feldman, 1995; Feldman Barrett & Russell, 1998; Citron et al., 2012), they are both accepted as a common underlying mechanism triggering emotion 7.

(19) evaluation no matter what the stimulus type is (pictures, words etc.). With this dimensional view, Russel and Barrett (1999) proposed their conceptual act model of emotion by stating that emotions are events constructed by core affect and categorization. Core affect is a neurophysiological state measured by the continuous scale of valence and arousal. Human’s experiences of emotion involve using prototypical emotional episode (i.e. prior knowledge) to categorize the affective state, with the aid of language as contexts to stimulate such emotional episode or knowledge. By different combinations of valence and arousal, stimuli can be tagged with an affective connotation in the mental representation. Since arousal represents one’s subjective feeling of elicitation by emotion stimuli, more and more dictionaries and thesauri start to classify emotion words solely by their inherent valence. For instance, in Pennebaker and King (1999)’s later development of LIWC program that aimed to collect a default set of word categories as a dictionary, the affect category (including 915 words) were built based on large textual analysis with the aid of Roget's Thesaurus and standard English dictionaries. Words are classified into Positive and Negative emotion by their valence scales. Huang et al. (2012) followed this dimensional view and recently translated the dictionary into Chinese with fine-grained rating, word segmentation, and category reconfirmation. However, to date, the classification of basic and complex emotions via the discrete or dimensional views still seems unattainable, making the search for the answer to the origins of emotion more difficult. 8.

(20) 2.2 Emotions in Language Language can serve as a link between human’s perception and knowledge of emotion and its semiotic representations can limit speakers’ interpretations of emotional experiences or approximate the representation of emotional experiences to aid emotion conceptualization and categorization (Wierzbicka, 1992). Researchers have been investigating the language of emotion over the past decades to explore the possible and reliable classification of emotion and to help settle the traditional discrete or dimensional view in psychology. A review of studies on emotion classification in both western languages and Chinese would be provided as follows.. 2.2.1 Emotion Classification in Western Languages With the theory of Natural Semantic Metalanguage (NSM), Wierzbicka (1992) was the first to propose a series of semantic primitives such as I, you, someone, this, want, don’t want, think, say, imagine, feel, know, good, and bad, and tried to decompose emotion into complex events involving a cause and a mental state in terms of simple and non-technical terms. The following was her exemplification of the emotion Disappointment: X feels something; sometimes a person thinks something like this: (1) something good will happen; (2) I want this; after this, this person thinks something like this: I know. 9.

(21) now: this will not happen; because of this, this person feels something bad. X feels like this (p. 548). Using these prototypical scripts or scenarios in terms of wants, thoughts, and feelings, even the basic emotion can be rigorously and revealingly portrayed. Furthermore, the differences between apparent synonyms like unhappy and sad can be fully specified according to their conceptual structures, which may first seem “fuzzy” or overlapping. However, the main purpose of NSM is not to classify emotion words but to emphasize the specification of each emotion concept and thus leads to more research on the lexical semantics of emotion words. For example, Semin and Fielder (1991)’s Linguistic Category Model (LCM) distinguished action and stative verbs by several semantic criteria, one of which being positive or negative valence, a hedonic index to emotions. In LCM, verbs are divided into state verbs (e.g., to love, to admire), state action verbs (e.g., to surprise, to amaze), interpretative action verbs (e.g., to cheat, to help), and descriptive action verbs (e.g., to kick, to kiss). While state verbs are used to describe mental and emotional states, state action verbs often express emotion consequences of an action with a beginning and end. Together with interpretative action verbs, which often refer to a general class of behaviors or actions, these three types of verbs are all associated with semantic valence connotations. Although whether descriptive action verbs, which are distinguished from interpretative action verbs by invariant physical features of event types, could be related to affects is still under discussion, Semin and Fielder (1991)’s 10.

(22) model brings the necessity to incorporate the lexical analysis of words to the encoding of our emotion experiences. It is worth noting that the LCM was based on one general verb taxonomy between states and events (Davidson, 1971; Dowty, 1979; Rappaport Hovav & Levin, 1998). Stative verbs describe an enduring situation or a state of being, while eventive verbs entail a series of causal changes throughout the process in event structure (Dowty, 1979), the representation of events and their participants. Eventive verbs can also be differentiated from stative verbs by their entailment of the conceptual units involving CAUSE, BECOME, CHANGE and resulting STATE, while stative verbs do not have this causal chain, which contributes to different lexical complexity to the two types of verbs. The distinction between eventive and stative verbs has psychological reality, which was evidenced in Gennari and Poeppel’s (2003) study. With lexical decision task (LDT), subjects had longer reaction time in processing eventitve verbs compared with stative ones (e.g. vanish vs. exist). The authors concluded that this result cannot be explained by thematic roles or argument structures the verb meanings carry, but should be attributed to different event structure properties activated during processing. Apart from lexical distinctions, some scholars focus more on the semantic and syntactic nature of emotion words. Levin (1993) classified emotion verbs based on the transitivity and the experiencer’s syntactic position in a sentence, labeling them as “Psych-verbs” with four 11.

(23) subcategories. Amuse and Admire Verbs are transitive with object and subject being experiencers respectively, whereas Appeal and Marvel Verbs are intransitive with experiencers as object and subject respectively. Goy (2000) also analyzed Italian adjectives using a frame-based approach integrating case-frame semantics, the generative lexicon and the prototype theories. She observed that there are three interpretations/readings of emotion adjectives: stative (e.g. cheerful boy), manifestative (e.g. affectionate letter) and causative (e.g. amusing movie), and classified Italian emotion adjectives based on the explanation from qualia structures (Pustejovsky, 1995), the specification of how a lexicon’s arguments and events are connected to modify relations in the semantic composition within the affective noun phrases. According to her proposal, the semantic representation of emotion states can be characterized as knowledge structures that also encode prototypical sequences of actions and events, which are very similar to scripts containing the beliefs, the reactions, the causes, and the consequences of the emotion state (Fehr & Russell, 1984; Shaver et al., 1987; Wierzbicka, 1992). However, these linguistic approaches to the classification of emotion are often restricted to specific lexical categories, such as verbs or adjectives, and hence cannot provide a comprehensive model to characterize the emotion connotation across different lexical categories. In contrast to traditional lexical and semantic views, the corpus-based analysis of linguistic expressions of emotion focuses on establishing norms for affective lexicon and 12.

(24) detecting the structures in which emotion concepts may be embedded. Affective Norms for English words (ANEW), WordNet Affect, and Berlin Affective Word List Reloaded (BAWL-R) are the most used materials in research of visual word processing in western languages. In ANEW and BAWL-R, a set of English and German affective words was manually rated on the valence, arousal and dominance/imageability; however, only BAWL-R assigned affective words to three different word classes: nouns, verbs, and adjectives. Unlike the previous resource of affective lexicon, WordNet Affect is not only an extension of WordNet domain to label affective concepts to a subset of synonymous sets (so-called synsets) but also labels a semantic domain to provide conceptual relations among word senses used to group words hierarchically in WordNet. In particular, words in WordNet Affect are categorized into two groups: direct emotion words that denoting emotion states (e.g. happy, fear), and indirect emotion words that eliciting emotions (e.g. snake, monster). Although the interpretations vary among individuals for indirect emotion words due to different causality they evoke, the general affectivity is considered collective imagination and calculated by semantic affinity with affective lexical concepts (Strapparava et al., 2006). In addition to valence tagging, WordNet Affect also annotated stative/causal interpretations of adjectives, a classification similar to Goy’s (2000) analysis. Unfortunately, most previous psycholinguistic experiments used affective words from ANEW and did not tell the (in-) direct emotion words apart, and this could yield qualitatively different results in terms of 13.

(25) processing and information encoding (e.g. different costs of episodic affective memory retrieval).. 2.2.2 Emotion Classification in Chinese Few studies are devoted to emotion classification in Chinese. One of the early researches on Chinese emotion words was corpus-based and conducted by Chang et al. (2000). Following psychologists’ discrete emotion approaches, the authors examined seven sub-fields of Chinese emotion verbs: Happiness, Depression, Sadness, Regret, Anger, Fear and Worry based on the Sinica Corpus, a tagged corpus of mostly news documents containing five million Chinese words in total. Focusing on stative verbs, they demonstrated that the differences between synonymous affective words like kuai4le4 快樂 “to be happy” and gao1xing4 高興 “to be glad” could be motivated by different lexical event types that affect the grammatical functions, occurrence in imperative and evaluative constructions, verb aspects, and transitivity. Lee et al.(2010) adopted emotion keywords representing happiness, sadness, fear, anger, and surprise, to track events outside the word boundary that immediately caused emotions, arguing their keywords had been agreed upon by most scholars as primary emotions (Turner, 2000). In their study, the cause events of emotion were first specified in Chinese texts, and the authors categorized them into nominal events (mainly nouns) and verbal events that comprised nominalization and verbs in Chinese. Most studies on 14.

(26) categorizing Chinese emotion words also followed this discrete emotion view to form a set of universally basic emotions, proposing that emotion is compositional. For example, a culture-specific, complex emotion like “guilt” can be decomposed into basic emotions of joy and fear with different intensities. Unlike the western literature, Cheng et al. (2013) analyzed the structure of Chinese emotions and found Chinese emotion words (mainly stative verbs) can be grouped into seven clusters: love, fear, sad, disgust, pride, apprehension, and yearning. Based on the addictive tree clustering, a graphical representation in which distances along branches reflect similarities among the objects, emotions in the same cluster share higher centrality and reciprocity in terms of conceptual representations. Meanwhile, Zhuo et al. (2013)’s Chinese emotion word database was built according to Ekman’s taxonomy of emotion (Ekman, 1972) that comprises emotion-describing and emotion inducing words. Emotion-describing words can further be categorized into words relating to (1) emotion experiences, (2) behavioral expressions, (3) cognitive status, (4) physiological sensation, and (5) emotion reaction, whereas emotion-inducing words contain (1) objects/concepts, (2) events/actions,. (3). personality/traits,. and. (4). evaluation/judgment.. However,. this. classification also could not justify how emotions are naturally classified in humans since these affective words might contain both Chinese stative verbs (e.g. kuai4le4 快樂 “to be happy”; beishang 悲傷 “to be sorrowful”), and eventive words (e.g. she2 蛇 “snake”; zhong4jiang3 中獎 “to win a prize”) that may conceptually be different in relating to 15.

(27) emotions. Therefore, our study will focus on Chinese emotion words of different lexical category types that can best demonstrate the differences in lexical processing and emotion connotation.. 2.2.3 Emotion Word Processing Adopting the two-dimensional structures of emotion, valence and arousal, previous research comprises a great portion of the “emotion effect” on the modulation of word recognition. Since Kissler’s (2006) study, a series of neuroimaging techniques were applied to explore the temporal and spatial neuronal activities in human brain during emotion word processing. Several typical early components of emotion effects are specified with mixed results in event-related potential (ERP) studies such as P1, N1, and P2. For instance, Herbert et al. (2008) using silent reading task with valenced words (60 pleasant, 60 unpleasant and 60 neutral adjectives) of controlled arousal found no P1 and N1 elicitation while Scott et al. (2009) adopting lexical decision task with arousal-matching words (80 positive, 80 negative, 80 neutral words, half high (HF) and half low (LF) in word frequency) showed greater P1 in positive and neutral stimuli than negative ones and also greater N1 in LF neutral and HF negative stimuli compared with HF neutral and HF positive ones, respectively. On the other hand, with LDT, Kanske and Kotz (2007) presented emotional stimuli (60 positive, 60 negative, 120 neutral nouns, half high and half low in word concreteness) in the right and left 16.

(28) visual field (RVF/LVF), and found greater P2 for positive than neutral stimuli, whereas Carretié et al. (2008) showed no P2 elicitation with 10 insults, 10 compliments, and 10 neutral adjectives. It is thus proposed that the P1 and N1 components in previous studies represent subcortical feed-forward mechanisms (for reviews, see LeDoux, 2003) that link to the acquisition of conditioned responses and the detection of visual emotion stimuli when there are severely limited perceptual resources (e.g. rapid serial visual presentation and subliminal or near-threshold presentation). Also, the emotion effect on P2 amplitude over frontal-central sites (Begleiter & Platz, 1969; Begleiter et al., 1979; Bernat et al., 2001; Schapkin et al., 2000) can reflect automatic processing of emotion when attention is focused on the emotional connotation in passive viewing (Begleiter and Platz, 1969), with subliminal presentation (Bernat et al., 2001), or in an affective evaluation condition (Begleiter et al., 1979) that does not necessarily associate with emotional word processing. According to Kanske and Kotz (2007), the enhanced P2 amplitude for positive stimuli represented the positive offset since under low levels of arousal input (e.g. word stimuli), people would respond more intensively to positive emotion stimuli compared with negative ones. However, the studies reviewed here often adopted different tasks and manipulation of emotion word stimuli; hence, it could produce artifacts and made most of the inconsistent results hard to interpret.. 17.

(29) One of the late commonly seen components is the early posterior negativity (EPN), which peaks between 200 and 300 ms after stimulus onset. It has an occipito-temporal scalp distribution and its amplitude is larger for emotionally valenced words (positive and negative) than neutral words during silent reading (Herbert et al., 2008; Kissler et al., 2009) or lexical decision task (Schacht & Sommer, 2009a; Scott et al., 2009). EPN has been associated with the initial and automatic processing of stimuli with emotional connotations since its effect is not modulated by the emotional nature of the task (Kissler et al., 2006), or the self-referentiality of the emotional stimulus (Herbert et al., 2010). In this time window, source analysis also found that mapping of the visual stimulus with its corresponding lexical representation can occur with higher activities in fusiform gyrus or the visual word form area, which is a part of the ventral visual processing stream in the inferotemporal brain regions (Hinojosa et al., 2001; Schacht & Sommer, 2009a). It is also noted that there is no discrimination between positive and negative valence in this stage, so EPN may be an index of natural selective attention, which signals a general emotion effect without distinguishing relationships between emotional valence and arousal. Another component evoked by emotion effect and yielding more consistent results in literature is late positivity complex (LPC). LPC has been found in response to emotional stimuli including words, faces, and also pictures. It peaks between 500 and 800 ms and has a centro-parietal scalp distribution. A larger LPC is often seen to emotionally valenced stimuli 18.

(30) than neutral ones (Carretié et al. 2008; Herbert et al., 2008; Kanske & Kotz, 2007; Palazova et al., 2011; Schacht & Sommer, 2009a), representing the valence processing, and may extend several hundred milliseconds, forming a sustained slow positivity (SSP) and peaking from 700 to 1000 ms. However, if comparing LPC on pleasant and unpleasant stimuli directly, different experimental designs again render biased results. For example, it is found that LPC was larger when one experienced unpleasant stimuli (Baumeister et al., 2001; Schacht & Sommer, 2009b): a common “negative bias” with which people would respond more intensely to negative emotion stimuli compared with arousal-matching positive ones. Nevertheless, other studies had null results or even a positive bias (Herbert et al., 2008; Kissler et al., 2009; Palazova et al., 2011). LPC is considered to be a part of a larger P3 family, related to reallocation of a more domain-general attentional orientation toward an evaluation or processing of the stimulus. Emotional stimuli will thus trigger LPC due to its motivational salience. Also, in contrary to EPN, LPC is task dependent, often occurring in tasks requiring deep processing such as LDT and semantic judgment, and can be modulated by the arousal of the stimuli and the self-relevance in the emotional contexts (Fields & Kuperberg, 2012). According to Kissler’s (2006), higher cortical response can be observed through visual word processing if the word is related to or contains affective associations. This higher cortical response occurs in visual word form recognition, the access of the mental 19.

(31) representation, the allocation of attention, contextual integration and memory encoding, which may sustain to 500 ms after the word onset. The variable timing of each component may result from subjects’ different motivational states with different tasks applied, and additional contextual factors. Under LDT, it is important to examine whether the emotion effect precedes or follows the lexical effect (e.g. N400-like negativity) elicited by comparing the emotion words and pseudowords/non-words. This can also be discussed with Coltheart (2001)’s dual-route cascaded (DRC) model to explain skilled reading: visual word recognition can bypass two routes, the lexical and non-lexical one through lexicon/semantic system or grapheme-phoneme rule system respectively. If the time locus of emotion effect can be assured, then it is possible to answer the question whether emotion is encoded as a semantic/lexical feature of the visual words by certain connotation mechanisms or it is just an associative conditioned effect of some mnemonic templates. Only a few studies addressed this issue and there were mixed results. In Scott et al.’s (2009) research, lexical frequency of the emotion words from ANEW was considered an index of lexical processing and the authors found the frequency effect on emotion word stimuli occurs earlier than the lexical effect. On the contrary, Palazova et al. (2011) using the same lexical decision task claimed a reversed pattern. In their research, words were half high and half low in frequency and categorized into nouns, verbs, and adjectives. Nouns and adjectives elicited emotion effect (i.e. EPN) approximately at the same time as the onset of the lexical effect (~270 ms), while 20.

(32) verbs’ emotion effect came much later (~350ms). The inconsistency in the stimuli variables controlled across these studies thus shows the relationship between the linguistic features of the stimuli and emotion effect needs further inspection to explore whether there is linguistic confounds modulating emotion effects or even impacting the positive or negative bias of ERP components, such as P2 or LPC. Recently, Zhuo et al. (2013) used their own Chinese emotion word database to observe ERP response by a valence judgment task for emotion-denoting (e.g. kuai4le4 快樂. “to be happy”) and emotion-inducing (e.g.. zhong4jiang3 中獎 “to win a prize”) words and found these two categories show differences in a time window of 460~570 ms in the midline positions (FPz, Fz, Cz, Pz, Oz) and the frontal areas (FP1 and FP2). Their emotion word stimuli were rated on not only emotional features such as arousal, valence, continuance (i.e. the duration of the elicited emotion), and controllability (i.e. the degree to which you can control the elicited emotion), but also linguistic features like frequency (i.e. the word’s frequency of occurrence in daily life), and typicality (i.e. the prototypicality of the word) with the 9-point Likert Type scale. However, both the two groups of stimuli in this experiment contained various lexical categories such as stative and event words that were evidenced to cause different processing loads due to its lexical semantic complexity (Gennari and Poeppel, 2003). It is hence necessary that we observe the emotion-related components in an experiment with more detailed temporal resolution and delicate manipulation of emotion word stimuli. 21.

(33) As an index of semantic complexity, lexical category is one of the candidates for showing different semantic information a word can carry. Lexical category has long been one of linguistic universals widely studied in previous literature (for reviews, see Vigliocco et al., 2010) and verbs can assign thematic roles and impose greater processing demands than nouns at the semantic level. Therefore, research using different methods like EEG/fMRI often aimed to investigate whether processing words with different lexical categories, mainly nouns and verbs, would engage different neural systems. Lexical/semantic categorical differences of word stimuli can be observed through ERP differences, with nouns eliciting a left occipitotemporal-distributed negativity as early as 250~280 ms after word onset, but not verbs or numerals, in a semantic category judgment task (Dehaene, 1995). Also, Federmeier et al. (2000) compared the ERP waveforms of unambiguous nouns, verbs, and category ambiguous words (either to be nouns or verbs) embedded in a sentence context. Their materials were picked without the bias in semantic domains so that nouns and verbs were not restricted to objects and actions respectively. The results showed that verbs elicited a left lateralized anterior positivity while nouns generated a more negative wave between 250~450 ms over central-posterior sites. On the contrary, Pulvermüller et al.’s (1999a, b) study compared action verbs, nouns with strong action associations, and nouns with strong visual associations and found no lexical category effect. The authors found that verbs showed a larger P2 at frontal-central sites than nouns but there was no difference between action verbs 22.

(34) and action nouns, suggesting an effect of semantics rather than lexical category during word processing. In addition, Vigliocco et al. (2006) manipulated both the lexical categories (i.e. verbs and nouns) and the semantic features (i.e. sensory and motion words) of the words and found that nouns and sensory words had a larger N400 effect than verbs and motion words on central-parietal sites. These studies might suggest the effect of lexical category can come from different semantic complexity of words in both the early and late time windows of word recognition; however, few studies focused on this modulation on emotion word processing given that emotion words include both emotion-denoting and emotion-inducing words that may possess different semantic attributes. Besides, previous ERP studies often focused on different onsets of emotion effects on words with different lexical categories. For instance, emotion effects of verbs on EPN (Schacht and Sommer, 2009a, 2009b) started approximately 150 ms later than the effects reported by Kissler et al. (2007) with nouns and by Herbert et al. (2008) with adjectives. It is thus argued that the emotion contents in verbs and/or adjectives may be elaborated more deeply and for a longer time compared to nouns since nouns are easy to process and acquired early so that they can be processed more superficially without the sustained attention (Palazova et al., 2011). However, to date, no comprehensive comparisons among these word categories were conducted on Chinese emotion words to show whether there are latency, duration, or even scalp differences of the emotion effects.. 23.

(35) 2.2.4 Trait Theory and Emotion Word Processing In Eysenck’s (1947) early model of personality, two independent dimensions are specified, namely extraversion and neuroticism. Extraversion is characterized as the sensitivity to positive cues and the tendency to experience positive affects in the environment, whereas neuroticism is associated with emotion instability and the sensitivity to negative emotions, a strong predictor of psychological problems such as depression, anxiety disorder, and schizophrenia. It is further proposed that people with high extraversion (so-called extroverts) seek excitement and social activity in an effort to heighten their reticulothalamic–cortical arousal level, while people with low extraversion (so-called introverts) tend to avoid social situations in an effort to keep such arousal to a minimum. Neuroticism can be expressed by an increase of reactivity to the limbic system that is sensitive to emotion arousal stimuli and its level can affect the tendency of affective reactivity and emotion regulation. Extraversion and Neuroticism were later also specified in a widely accepted model of personality, the Big Five model (for reviews, see Costa & McCrae, 1992), which included five domains of personality: openness, conscientiousness, extraversion, agreeableness, and neuroticism. Although the neuroanatomical correlates of extraversion and neuroticism were commonly investigated (e.g. Wright et al., 2006; DeYoung et al., 2010)--with extraversion positively correlated with volume of medial orbitofrontal cortex, a brain region involved in processing reward information, and neuroticism negatively 24.

(36) correlated with volume of hippocampus systems, brain regions associated with threat, punishment, and negative emotion--their relationship with emotion word stimuli is still unclear in the normal population. These personal traits affect not only the way people categorize emotion words regarding semantic associates but also the degree of valence and arousal focus during perception. For example, behavioral priming (e.g. deciding if a letter string is an English word is speeded by prior presentation of a semantically related prime) data from Matthews et al. (1995) demonstrated that negative prime-target pairs generally possessed greater priming effects than positive ones and that neutral prime-target pairs were more strongly primed in more neurotic subjects, highlighting the bias on valence processing. Borkenau et al.’s (2010) study showed extroverts had faster identiﬁcation of pleasant words than of neutral and of unpleasant words under a go/no-go lexical decision task that included pleasant, unpleasant and neutral words. However, whether these traits postulate different arousablities or activate the limbic system to a different degree between neuroticism and extraversion, as suggested by Eysenck’s proposal of arousal variance, is still an issue worthy of careful examining. By using event-related potential technique in the current study, it is expected that a preliminary examination of the interaction between personality traits and (both emotion and linguistic) features of the stimuli can extend previous research and show the time course of the modulation by personality.. 25.

(37) Several recent studies used functional magnetic resonance imaging (fMRI) to investigate emotion word processing in people with minor or major mood-related disorders, such as subliminal depression, anxiety, or even schizophrenia (Laeger et al., 2012; Pankow et al., 2013). In Laeger et al.’s (2012) research, subclinical measures of anxiety or depression were found to modulate the neural processing of emotional words: emotion valenced words activated the amygdala and there was a positive correlation between the activation of the amygdala and the scores on trait anxiety and subclinical depression during negative word processing. Subjects with high trait anxiety also showed greater functional correlations between left amygdala and left dorsolateral prefrontal cortex (DLPFC) activation. The amygdala’s role in the emotion word processing could thus be more general, regardless of the valence nature of the emotion words. Pankow et al.’s 2013 study, on the other hand, revealed that with affective pictures, schizophrenic patients showed stronger activation of the right amygdala while viewing negative compared to positive stimuli. Among these patients, medication status further influences the degree of activation in ventral anterior cingulate cortex, a region involved in emotion processing and conflict monitoring. Other studies discussing the personality modulation such as extraversion and neuroticism on the emotion effect in human brain were mainly conducted by pictorial stimuli (Kehoe et al., 2011) in passive viewing which cannot truly reflect the reading models, but it is informative to make an analogical comparison between word and pictorial stimuli and clarify the relationship 26.

(38) between emotion valence and arousal. For example, individuals with high neuroticism showed reduced activation in the orbitofrontal cortex and valence processing in the right temporal lobe, but an increased response to emotional arousal in the right medial prefrontal cortex (mPFC) was observed (Kehoe et al., 2011). Canli et al.’s (2001) study showed that extraversion is correlated with the amygdala, caudate nucleus, and putamen response to positive stimuli. The left middle frontal gyrus and temporal gyrus were all observed to be associated with extraversion and neuroticism in emotion processing. However, since the valence difference is only reflected in the LPC or SSP activities during individualized emotion word processing in previous ERP studies and BOLD signals are not able to differentiate different aspects of emotion, using fMRI might not provide sufficient temporal resolution on the time window of interest we want to focus on. Indeed, ERP studies have showed the correlation of the characteristics in late components that reflect motivational significance of stimulus meanings or salience with peculiarities of the personality (for reviews, see Kovalenko & Pavlenko, 2009). For instance, the amplitude and latency of the LPC (mainly P3) in the central/parietal/occipital regions demonstrated a negative correlation with the indices based on the extraversion scale, and this was argued to be related to the level of voluntary attention and the intensity of habituation to stimulation in introverts. Also, in the right hemisphere, the neuroticism level was shown to be positively correlated with the amplitude of all the ERP components, along with the latency of late ERP components. 27.

(39) However, these ERP studies often adopted visual/auditory oddball or attention spatial orienting tasks that focused on individual differences in attention allocation with mostly pictorial stimuli; there was little proof of this modulation by a better temporal description under word recognition paradigm. Even though Laeger et al.’s (2012) fMRI study used linguistic stimuli, it only incorporated German nouns that could cause emotion connotations and left other lexical categories of emotion stimuli untested, such as the emotion-denoting words. No work has been done on the modulation of individual differences in healthy population on the emotion effect during visual word processing with the consideration of semantic complexity of the stimuli. With EEG, we hope our study can fill in the gap in the modulation of emotion effect on early and later components with comparison of individual differences and thus can shed more light on the impact of lexical differences on emotion word perception.. 28.

(40) Chapter 3 Methodology 3.1 Subjects Thirty-five native Chinese speakers (19 males), aged between 20 and 40 year (M=26, SD=4.46), took part in the experiment. They were all right-handed, had normal or corrected-to-normal vision, had no neurological disorders and psychological/mental illness and took no medication for mood disorders. They completed the Eysenck personality questionnaire revised (EPQ-R, revised short-scale edition in Chinese, see Appendix A) (Eysenck et al., 1995) before EEG recording. EPQ-R is a self-report questionnaire which measures levels of extraversion, neuroticism and psychoticism on scales ranging from 0 to 12, with 0 indicating the lowest and 12 the highest level of the trait, and it shows good reliability and validity in the male sample (N=408, Age=38.44±17.67). The coefficient of internal consistency (Cronbach’s α ) for the scale of extraversion is 0.88, and for the scale of neuroticism, 0.84 (Eysenck et al., 1995). We focused on extraversion and neuroticism in this study. The participants were then divided into two groups (high scores, low scores) based on the median scores of extraversion and neuroticism levels in EPQ-R. Accordingly, each participant was categorized as a member of extroverts or introverts (i.e. low extraversion) with high or low neuroticism at the same time. Besides, to ensure participants’ moods are in relative stability before the experiment, participants also filled out the Beck Depression Inventory-second edition in Chinese (BDI-II, see Appendix B). BDI-II is a self-report 29.

(41) questionnaire using 4-point Likert scale which measures levels of depression originally based on Beck et al. (1961). It also shows good reliability and validity in both the college student and patient samples. The internal reliability (Cronbach’s α ) for clinical patients is 0.92, while for college students, 0.93, with the test-retest reliability being 0.93 (Beck et al., 1961). The score ranges from 0 to 63 with the cut-off score for mild depression being 14. The participants’ scores indicated that they were not going through moderate or severe depression during the experiment. All participants were given written informed consent to participate in the experiment according to the guidelines approved by the Research Ethics Office of National Taiwan University, and they were paid for participating in the experiment.. 3.2 Materials Three hundred and twenty-four emotion words were selected from (1) the materials in the study of Zhuo et al. (2013), which consisted of 218 emotion-denoting words and 395 emotion-eliciting words, and (2) the Chinese LIWC dictionary by Huang et al. (2012). In Zhuo et al.’s (2013) database, emotional arousal, valence, along with continuance, controllability, frequency, and typicality were respectively rated by 353 and 1673 participants for each emotion word category using the 9-point Likert Type scale. From Huang et al.’s (2012) dictionary, only a portion of 427 emotion-inducing words labeled as affect were taken to supplement Zhuo et al.’s materials, which added up to 1040 words. The stimuli selection 30.

(42) went through the following steps. First, all the 1040 emotion word candidates were checked for repetition, word frequency, lexical categories, the frequency distribution of its lexical categories, and orthographic neighborhood sizes based on Chinese Word Sketch Engine1. Then only the words tagged with single or nearly single (i.e. unambiguous) lexical category (i.e. the frequency of the word’s use as other lexical categories in Chinese Gigaword Corpus2 is below 8*10-9) were kept. Finally, a manual inspection of homographs was made to delete the word candidates with multiple senses which are conceptually unrelated, resulting in 750 words. To control the word stimuli, all the remaining 750 word candidates were rated online by 91, 99, and 89 college students with the 7-point Likert Type scale on three variables respectively: emotion valence (-3 = negative to 3 = positive), arousal (1 = low arousing to 7 = high arousing), and imageability (1 = hardly imageable to 7 = very imageable) (see Appendix C, D, and E for the instruction of the rating questionnaire). We selected 108 positive, 108 negative, and 108 neutral words in total, and each word is a meaningful two-character compound in Chinese. The valence ratings differed significantly among all the three valence conditions, F(2,321)=2843.71, p<.001. Among each valence category, we further manipulated the lexical categories so that there were 36 stative verbs, event verbs, and nouns respectively for each valence category. Arousal levels of positive or negative stimuli were 1. Chinese Word Sketch Engine is a Corpus Query System incorporating concordance, word sketches, grammatical relations, and a distributional thesaurus based on Chinese Gigaword Corpus and Sinica Corpus 5.0 and maintained by Lexical Computing Ltd. in Academia Sinica, see http://wordsketch.ling.sinica.edu.tw/. 2 Chinese Gigaword Corpus contains about 1.1 billion Chinese characters, including more than 700 million characters from Taiwan’s Central News Agency, and nearly 400 million characters from China’s Xinhua News Agency. 31.

(43) also controlled. Statistical comparison of the arousal ratings revealed larger arousal values for positive and negative than neutral words, respectively, t’s(107)>.68, p’s<.01, and no significant difference of arousal level between positive and negative words, t(107)=.09, p=.269. A two-way repeated-measures analysis of variance (ANOVA) applied to word frequency,. arousal. ratings,. and. imageability. with. the. factors. of. valence. (positive/neutral/negative) and lexical category (stative verb/event verb/noun) indicated that there was no interaction between valence and lexical category regarding frequency, arousal, and imageability, F’s(4,140)<2.62, p’s>.05, but there was a main effect of lexical category on imageability, F(2,70)=8.1, p<.001. A Bonferroni-corrected post-hoc comparison showed that the imageability rating of eventive verbs was greater than that of stative verbs. All 9 word categories (3 valence conditions * 3 lexical categories) were further matched for frequency of use, the first- and second-character strokes, and numbers of first-character orthographic neighbors, F’s(8,315)<1.76, p’s>.08, except for the second-character stokes, F(8,315)=2.08, p=.038. A post-hoc comparison using Dunnett's T3 test due to its robustness in analyses with unequal variances showed no significant difference between the categories. The numbers of each character’s strokes, and the orthographic neighbors were computed automatically by using python-based Load and Analysis Chinese Corpus-Natural Language Toolkit (LACC-NLTK) based on Academia Sinica Balanced Corpus with tagged texts. In addition, 324 pseudowords were created by re-pairing the characters randomly in the word condition to 32.

(44) form meaningless two-character compounds. Stimulus characteristics are summarized in Table 1 (see Appendix F for a complete list of stimulus words).. 33.

(45) Table 1. Descriptive statistics (mean (M) values with standard deviation (SD)) for controlled variables and rating results of the stimulus set.. Positive. Valence (range -3 to +3). Arousal (range 1 to 7). Imageability (range 1 to 7). Word Frequency (1/109, Chinese GigaWordCorpus). Character 1 Stroke. Character 2 Stroke. Character 1 Character 2 No. of No. of Orthographic Orthographic Neighbors Neighbors. Stative Verbs. 1.58(0.31). 3.94(0.36). 4.05(0.67). 6076.28(8337.40). 11.25(4.47). 11.56(4.10). 43.46(48.22). 29.36(43.11). Eventive Verbs. 1.47(0.27). 3.98(0.46). 4.55(0.89). 23896.19(62346.91). 11.22(4.30). 11.81(3.98). 61.00(84.85). 32.53(31.38). Nouns. 1.58(0.21). 3.87(0.42). 4.02(1.18). 23440.56(35719.89). 10.08(3.51). 10.11(5.74). 49.58(42.53). 98.28(86.01). -0.09(0.49). 3.53(0.53). 3.87(0.91). 11350.78(43184.26). 9.03(3.93). 10.81(4.40). 64.75(61.02). 39.96(46.73). Eventive Verbs. 0.03(0.30). 3.24(0.56). 4.71(0.92). 14944.22(28520.01). 9.86(3.70). 11.14(3.30). 46.42(41.48). 57.39(86.83). Nouns. -0.02(0.35). 2.99(0.69). 4.57(1.25). 33944.81(98639.44). 8.83(3.68). 10.28(3.44). 53.47(44.07). 65.78(49.77). Stative Verbs. -1.72(0.20). 4.10(0.27). 4.24(0.60). 4753.25(17979.24). 11.17(5.68). 11.97(4.90). 41.86(82.68). 30.03(44.54). -1.80(0.28). 4.09(0.34). 4.59(0.82). 6751.39(14932.07). 10.22(3.91). 12.42(5.65). 37.61(40.76). 28.44(36.25). -1.62(0.32). 3.86(0.33). 4.35(1.03). 11113.08(26984.22). 11.00(4.50). 9.00(4.15). 44.47(52.75). 77.03(98.98). Stative Verbs Neutral. Negative Eventive Verbs Nouns. 34.

(46) 3.3 Procedure The experiment was carried out at the Neurolinguistics Lab in the English Department at National Taiwan Normal University. All participants were seated in front of a computer monitor at a distance of 80~100 cm when they underwent EEG recording in a sound-proof room. The participants were instructed to carefully and silently read the visually presented words in bold white characters (Font: Courier New; Point size: 18; Visual angle: 2.5~3.2°) on a black background designed by E-prime software (Psychology Software Tools, Inc.) and decide whether they were Chinese words or not as accurately and as quickly as possible by button press. A response box with two buttons corresponding to “yes/no” answers was provided and the button configuration was counterbalanced across participants. Before the experiment, participants were given ten practice trials to familiarize themselves with the procedure. The experiment started with a fixation cross in the center of the screen for 500 ms. After that, the screen would blank out for 200 ms and a word immediately appeared and remained on the screen for 1 second, followed by a question mark at the center of the screen. Once the question mark showed up, the participants needed to make a self-paced lexical decision by pressing the YES or NO button for the response. The screen would then blank with a jitter of 1500~2000 ms, during which time participants were allowed to blink (see Figure 1 for illustration). Three hundred and twenty-four words and 324 pseudowords were mixed together and randomly separated into four lists. The four lists were further rotated for 35.

(47) each participant to counterbalance practice effects across the conditions of the experiment. The whole session of the experiment was divided into eight blocks. The number of word and pseudoword stimuli was balanced in each block, together with the valence and lexical categories of word stimuli. The word order within blocks was randomized. Each block would take approximately 6-7 minutes to complete, and a short break would follow each block, during which participants could decide when to resume the experiment. The entire EEG recording lasted for about 45 minutes. After completing the task, participants had to leave the EEG recording room and were presented again with the emotion stimuli in a newly randomized order on the computer. They needed to rate the physiological activation level of each word stimulus using the 7-point Likert Type scale so that individual variability in the subjective perception of these emotion words could be evaluated.. Figure 1. Sequence of events under lexical decision task in this experiment. 36.

(48) 3.4 Data acquisition and analysis The electroencephalogram (EEG) was recorded from 30 electrodes placed in an electrode cap (FP1, FP2, F7, F3, FZ, F4, F8, FT7, FC3, FCZ, FC4, FT8, T3, C3, CZ, C4, T4, TP7, CP3, CPZ, CP4, TP8, T5, P3, PZ, P4, T6, O1, OZ, O2) according to the international 10-20 system (Pivik et al., 1993). Two external electrodes were placed at the left (A1) and right (A2) mastoid for reference points. The vertical and horizontal electrooculogram would be recorded through two electrodes placed laterally to the left and right eye (HEOL and HEOR) and two electrodes placed at the upper and the lower position of the left eye (VEOU and VEOL). Electrode impedance was kept below 5 kΩ. Data was recorded with a sampling rate of 1000 Hz and ampliﬁed by NuAmps system (NeuroScan Inc.) with an amplifier rate (gain) of 19, corresponding to an input range of ±131.5mV. The online low pass filter was set to 100Hz and the high pass filter was set as DC recording. Reaction times (RTs) and accuracy rates in the lexical decision task were calculated by E-prime software (Psychology Software Tools, Inc.) and analyzed with a two-way repeated-measures ANOVA, including the factors of emotional valence (positive, negative, neutral), and lexical category (stative verb, eventive verb, noun). To understand individual differences, two three-way repeated-measures ANOVAs with the previous factors and additional factors, extraversion/neuroticism (high, low), were also carried out. EEG recordings were further processed offline with EDIT 4.5 software (NeuroScan Inc.). 37.

(49) A linear derivation of channels converted the four mono-polar channels (HEOL, HEOR, VEOU, and VEOL) into two bipolar channels (HEOG and VEOG). The raw EEG data then went through ocular artifact correction by setting a trigger on VEOG. The ocular artifacts were constructed with a minimum of 20 sweeps and 400 ms duration on VEOG signal, and the computed average waves were subtracted from the original raw EEG data. The continuous EEG files were then epoched by setting the interval as 200 ms before and 800 ms after the stimulus onset. The baseline correction was applied with the pre-stimulus interval of -200 to 0 ms. Trials in which the voltage in any channel(s) exceeded ±100uV during the epoch interval were rejected first. Then, trials were visually inspected and rejected if they were contaminated by blinks, eye movements, absurd voltage shifts, alpha waves, and other body movements. Those participants whose remaining trials were below 50% were excluded from further analysis. Afterwards, sorting and averaging were computed separately for each condition. Grand averaging was also performed by averaging all subjects' averaged ERP data within the same experimental condition. Finally, the ERP data was bandpass filtered with frequency values set as 0.1Hz to 30Hz. With the ERP average amplitudes, five-way repeated-measures ANOVAs with the same factors as for the behavioral data (emotional valence, lexical category, extraversion/neuroticism) and the additional factors of laterality (left, midline, right sites) and anteriority (anterior, central, and posterior sites) were conducted on time windows of 200~250 ms, 300~450 ms, and 450~650 ms post stimulus. 38.

(50) The factor of anteriority divided region of interests into anterior sites (F3, FZ, F4, FC3, FCZ, and FC4), central sites (C3, CZ, C4, CP3, CPZ, and CP4), and posterior sites (P3, PZ, P4, O1, OZ, and O2), whereas the factor of laterality, left sites (F3, FC3, C3, CP3, P3, and O1), midline (FZ, FCZ, CZ, CPZ, PZ, and OZ), and right sites (F4, FC4, C4, CP4, P4, and O2). To investigate the lexical complexity modulation on the temporal aspect of emotion word processing, a three-way repeated ANOVA on the factor of emotion valence, lexical category, and neuroticism was conducted to examine the differences of 50% fractional area latency across conditions on time window of 300~400 ms (N400) with CZ, one of the typical electrodes showing N400 in literature. The 50% fractional area latency was computed by using functions from the EEGLAB toolbox (Delorme and Makeig, 2004). To adjust the degrees of freedom (DF) of the F-values for violations of the sphericity assumption, Greenhouse-Geisser correction was applied. Corrected p-values and uncorrected DF values were reported. The alpha levels in all post-hoc comparisons were Bonferroni-corrected.. 39.

(51) Chapter 4 Results The present study recruited 35 participants. Due to artifact contamination, only 18 participants (7 males; Mean=26.7, SD=4.94) were included for statistical analysis. The participants were then categorized into high/low extraversion and high/low neuroticism according to the median scores of extraversion (4.5, Mean=5.5, SD=4.26) and neuroticism (8, Mean=6.83, SD=3.57) levels on EPQ-R. The EPQ-R scores of the participants were provided in Table 2.. Table 2. The EPQ-R Scores of the Participants Participant No.. 1. 2. 5. 6. 7. 8. Extraversion Scale. 6. 2 11 10 3. 3. 3. 8 10 3. Neuroticism Scale. 2. 3. 3. 3. 4. 3. 9 10 11 12 13 14 15 16 17 18. 8 11 2 10 10 9. 1 11 0 7. 0 11 0 10 7. 9 12 10 11 6. 4. 3. 4.1 Behavioral Data The participants’ average reaction times (RTs) and accuracy rates for word and pseudoword stimuli are illustrated in Table 3. Two one-way repeated ANOVAs with the factors of lexicality (word, pseudoword) were conducted on the RTs and accuracy rate, respectively. The results showed that there was a main lexicality effect on RTs, F(1,17)=5.83, p<.03, with the participants responding significantly faster to word than pseudoword stimuli. 40.