中文動詞之語意韻律處理:事件相關腦電位研究

全文

(1)國立台灣師範大學英語系碩士論文 Master’s Thesis Department of English National Taiwan Normal University. 中文動詞之語意韻律處理：事件相關腦電位研究 Semantic Prosody in the Processing of Mandarin-Chinese Verbs: An ERP Study. 指導教授：詹曉蕙博士 Advisor: Dr. Shiao-Hui Chan. 研究生：陳彥琳 Student: Yen-Lin Chen. 中華民國一百零五年一月 January 2016.

(2) 摘要字詞為語言之中傳遞訊息的最小核心單位。它的意義往往比字典裡所記錄的概括性解釋來的深廣，其中尤其重要的是說話者或作者透過使用該字詞來表達的思維、情緒、對於事件的感受以及立場等。這層語言使用上呈現出來的言外之意，產生自字詞在真實語境當中經常一同搭檔出現的「搭配詞」的一些語意特徵。因此，字詞與搭配詞之間的緊密關係，讓這些語意特徵逐漸感染到字詞的語意，而形成該字詞裡新增的語意。這部分的語意即是被字典忽略的部分。過去的語言學文獻稱這樣「搭配詞之語意特徵感染」的現象為 semantic prosody (語意韻律)，而它的存在驗證了 Sinclair (1991) 提出 the idiom principle，說明字詞的意義來自字詞在文章脈絡中和其他字詞的搭配關係，不是獨立且不受語言使用而改變的。本篇研究旨在以事件相關腦電位技術來探討中文受詞的顯著情緒特質和無情緒特質是否影響目標動詞(皆非情緒詞)的處理歷程。本實驗採用情緒判斷作業並操控目標動詞–中文雙字及物動詞–所誘發的感受(正面感受、中性、負面感受) 和目標動詞經常搭配之受詞之情緒高度 (高度情緒化、低度情緒化)。實驗結果顯示，目標動詞所誘發的感受會影響晚期的 N400 和 LPC 振幅特性。其中誘發正面感受的動詞所產生的 N400 振幅最小，說明當動詞帶有正面語意特徵時能提升語意提取之歷程，也意味人類對於正向性的偏好。另外，凡能誘發感受(尤其正面感受)的動詞所產生的 LPC 振幅都比中性動詞的振幅來的大。這說明帶有情緒語意特徵的動詞能在腦海中生成鮮明的畫面，並且激發大腦的認知分析作業。除此之外，這些情緒特徵的動詞所引誘右腦的 LPC 振幅又比左腦的大，可表示右腦是字詞情緒處理上不可或缺的功臣。本實驗結果似乎無法驗證語意韻律在大腦認知系統上扮演的角色。然而，我們發現這些動詞的情感判斷可能是受高頻率受詞之情緒特質感染而來。因此我們認為透過每日語言使用習慣，動詞和高頻受詞的搭配關係早已在腦海中深植，這些受詞的語意特徵也就被完全接納成為動詞語意的一部分了，故本實驗無法在大腦字詞處理的當下找到語言韻律的證據。. 關鍵詞：情緒、情緒詞處理、語意韻律、搭配、事件相關腦電位. i.

(3) Abstract A word tends to express more than its surface meaning does, of which is often denoted down as a firm explanation in a dictionary with no room for negotiation on the realistic use of the word. Since it is a general tendency for a word to co-occur with its habitual close surrounds in an authentic language context, the current study would approach the meaning of word through a phraseological analysis and conforms to the idiom principle proposed by Sinclair (1991), by which the core idea of it has been supported in several language studies nowadays with the adoption of computerized corpora into the investigation of word patterning in languages. The principle argues, as this study also leans towards, that meaning should arise from word patterns, not from word in insolation. An evidence for this tendency of language is semantic prosody, which states the transfer of semantic features from a word’s habitual collocates to the meaning of the word, yielding ultimately, due to recurrent collocational relation, an epistemic or pragmatic reading imposed on the word. The current study aimed to examine the nature of semantic prosody from the cognitive neuroscience perspective, with an expectation to capture its online computation. We anchored on the emotion aspect of semantic prosody, and employed a 3 x 2 factorial design based on two main factors: (1) Valence of the target two-character Mandarin Chinese verbs (Negative verbs, Neutral verbs, Positive verbs), and (2) Polarization of emotion of target verbs’ habitual nominal collocates (Highly emotion-polarized nominal collocates, Low emotion-polarized. ii.

(4) nominal collocates). Presenting only the target verbs (nominal collocates were invisible in the experiment for our manipulation on “Semantic Prosody” (High) vs. “Weaker or No Semantic Prosody” (Low) conditions), our experiment required participants to evaluate the emotion valence of the verbs. The results showed emotion effects in the N400 and the LPC time windows. The smaller N400 for the positive verbs compared to that of the neutral and negative ones demonstrated easier semantic processing of words with positive emotions. It also suggested the uniqueness of positive verbs, which may due to human’s general bias towards positivity. The enhanced LPC for the emotional (especially the positive) verbs may suggest vivid mental images activated by them and reflect more active cognitive analysis on words with distinctly negative and positive emotion. Furthermore, stronger LPC responses in the right hemisphere may indicate right-hemisphere involvement in processing the affective content of words. The current experiment failed to find an online computation of semantic prosody. However, we did observe that the valence of the verbs may come from the valence of the high-frequency collocates. Thus, we argue that semantic prosody might have been gradually formed during the process when a speaker learns and uses the combination of the verb and its collocates, and is directly associated with the verb once it is consolidated.. Keywords:. emotion,. emotional. words. processing,. event-related potential (ERPs). iii. semantic. prosody,. collocation,.

(5) Acknowledgements There were moments in life when I felt deeply moved. The moments as I looked back on the paths I have drifted through, joyfully or bitterly. And from those paths I finally reached where I was at that presence. Well, this is one of the moments, too, as I am writing this part of the thesis now. During this academic journey, I have been baptized by so many beautiful (yet not all well-comprehended within the time) theories and studies of language, and so brought up very special page of story in me. In the following, I would very much love to express my sincere gratitude to the people that have helped me become not only a better researcher, but also a better self along the road. First of all, my heartfelt thanks go to my advisor, Prof. Shiao-Hui Chan. She had the most amazing lectures on Neurolinguistics as well as Language and Brain. Ever since the days I enrolled in her Language and Brain class, she lighted up the pure curiosity in me so that I wished to know more about the mystery of our brain, and brain’s capability to decode and understand complicated information such as language. She never appears to get bored by my “shallow” or “not-good-or-interesting-enough” questions, which I think is the reason why I salute the passion of her as a researcher and the spirit in her of being a discoverer of this world. More importantly, she gives the most powerful encouragement whenever I found myself hindered by poor academic studying or struggles in life. Because of her, I could find my courage back to continue the research and be brave while facing unpredictable events. iv.

(6) during the time. I would like to give huge thanks to the committee members, Prof. Shu-Kai Hsieh and Prof. Yow-Yu Lin. I used to be afraid to take questions or comments in a regular Q&A session after presentation. But neither of you made me feel frightened! Thank you for accompanying me during this journey and being such memorable people in it. I could not better my study without each of your precious guidance and great insights for it. A must thank-you also goes to all the anonymous participants of the pilot test and the ERP experiment. Without them, this study would not be able to be accomplished. My gratitude also goes to the professors who have enlightened me with their wisdom during the days at NTNU. They are Prof Chiung-Huei Joan Chang, Prof. Chun-Yin Doris Chen, Prof. Gerardo Fernandez Salgueiro, Prof. Hsi-Nan Yeh, Prof. Hsi-Yao Su, Prof. Hsiao-Hung Iris Wu, Prof. Hui-Shan Lin, Prof. Jen-I Li, Prof. Jin-Lan Joy Wu, Prof. Miao-Hsia Chang, Prof. Miao-Ling Hsieh, Prof. Shau-Ju Chang, Prof. Wen-Ta Tseng, Prof. Yeu-Ting Liu, Prof. Yu-Chuan Shao, Prof. Yuh-Show Cheng, and Prof. Yung-O Biq (by the alphabetic order). My special gratitude goes to Prof. Yeu-Ting Liu for the consistent warm concerns and words of encouragement sent to me during the hard times, and for a living example of the indestructible perseverance a man can hold. To know you in this journey is truly a blessing! I would like to express my thanks to all my classmates at NTNU as well: Bill, Doris, Ivy, Jan, Jennifer, John, Karen, Ken, Kevin, Lucy and Vicky (by the alphabetic order). You guys. v.

(7) are great company! I kind of miss the days when we all studied hard together for the exams. (Except the exams part. Agree?) It would have been even harder for me if I did not have your support and some big laughs from you every now and then. To Ken and Lucy, you two are my best buddies! Thank you for sharing the ups and downs with me, adding so much life and energy into my research journey (, which might have turned dark and blank if you guys hadn’t come for me), and for always there for me whenever I was in troubles. I will keep all the tissues (with your words of encouragement) with me as long as I can. And they deserve so! To Kevin and Vicky, you are the source of joy and strength. Though I didn’t meet you guys often, but your occasional presence in room never ever failed to light me up! Moreover, I want to send my gratitude to all the lab members from the Neurolinguistics Lab of NTNU: Elvis, Gracie, Helen, Jeff, Julia, Ken, Lin, Matt, Ronald, Terry and Vivi (by the alphabetic order). It has been such an honor to know all you incredible people. To Elvis, thank you for the unplanned phone call we had several days ahead of the defense. Your words literally gave me a timely uplift. To Terry, thanks for your precious time spent on the ERP experiment and your insights for the study! To Helen, Matt, Ken and Lin, thank you guys for the huge support and great helpers for the ERP experiment. Special thanks go to Helen and Ken, for your patience to listen to some of my pessimistic thoughts about the chances for a successful experiment. Thank you for the experience sharing in data acquisition and follow-up analyses, and the peace and courage you brought to me when I felt lost or simply. vi.

(8) found no way out. Special thanks to Matt, for if it were not for your great assistance, I could not even move on to experiment recruitment at all. Thank you as well for the off hours you spent on me during that very week ahead of experiment. You always tried your best to explain details about the connection between the presentation computer and the recording computer, and figured out every possible ways to address troubleshooting and other technical problems. Special thanks go to Lin, for the positive energy and the optimistic view flowing over me during my stay in the lab. And to Gracie, you are my guidance angel. Thank you for growing up with me. Thank you, even now located oceans away, for still hearing my problems whenever you got time, sending inspiring songs and articles whenever I got overwhelmed, and for never turning down on my urgent requests for help. Lastly, I am deeply indebted to my family for their love and support along the way. To my younger sister, thank you for your patience and all the heartfelt talks with me. Sometimes I was astonished to learn how mature that you have grown. To Mom, thank you for your confidence in me and always telling me to look for God when I felt lost. To Dad in heaven, I miss you. Thank you for protecting my innocent dreams ever since I was little. Thank you for telling me not to look down on myself. You had taught me through your fights with illness that there must always be a way. So do not give in. Swim against the adversities. Thank you for these invaluable lessons, for giving me life, and for being our wonderful father. Thank You, God, for Your beautiful blueprint for this academic journey!. vii.

(9) TABLE OF CONTENTS 摘要 Abstract Acknowledgements Table of Contents List of Tables List of Figures Chapter 1 Introduction. i ii iv viii x xi 1. 1.1 Motivation. 1. 1.2 Research question. 3. Chapter 2 Literature Review. 5. 2.1 The idiomaticity of language: The open-choice principle and the idiom principle. 6. 2.2 Extended unit of meaning: Sinclair 2004a. 10. 2.2.1 Collocation. 12. 2.2.2 Semantic preference. 15. 2.2.3 Semantic prosody. 17. 2.2.3.1 Semantic prosody as a connotative meaning. 18. 2.2.3.2 Semantic prosody as a pragmatic meaning. 19. 2.2.3.3 Interim summary for semantic prosody. 21. 2.3. The ERPs evidence of emotional words processing. 22. 2.3.1 Early effect: N100. 23. 2.3.2 Early effect: P200. 26. 2.3.3 Early effect: EPN. 27. 2.3.4 Late effect: N400. 29. 2.3.5 Late effect: LPC. 30. Chapter 3 Methods. 32. 3.1 Participants. 32. 3.2 Materials. 32. 3.2.1 Emotion valence rating on verbs. 34. 3.2.2 Polarization of verbs’ habitual nominal collocates. 34. 3.2.3 Frequency. 37. 3.3 Procedure. 38. 3.4 Behavioral and EEG recordings. 41. 3.5 Data analysis. 41. viii.

(10) Chapter 4 Results. 44. 4.1 Behavioral data. 44. 4.2 ERP data. 46. 4.2.1 N400. 49. 4.2.2 LPC. 50. Chapter 5 Discussions. 54. Chapter 6 Conclusion. 60. References. 62. Appendix A: The rating questionnaire of the valence of stimuli Appendix B: A complete list of stimuli. 67 68. ix.

(11) List of Tables. Table 1.The six experimental conditions-------------------------------------------------------------33 Table 2. An illustration on the high and low emotion polarization of verb stimuli’s top 5 nominal collocates: The negative valence group----------------------------------------------------36 Table 3. An illustration on the high and low emotion polarization of verb stimuli’s top 5 nominal collocates: The neutral valence group------------------------------------------------------37 Table 4. An illustration on the high and low emotion polarization of verb stimuli’s top 5 nominal collocates: The positive valence group-----------------------------------------------------37 Table 5. Reaction times (ms) and accuracy rates for the experimental conditions--------------44. x.

(12) List of Figures. Figure 1.The procedure of a trial (followed by the first half the next trial)----------------------40 Figure 2.The grand average waveforms of negative, neutral, and positive stimuli-------------47 Figure 3.The topography of ERP differences between Negative stimuli vs. Neutral stimuli (Bin 1: Negative - Neutral) and between Positive stimuli vs. Neutral stimuli (Bin 2: Positive Neutral) during 400-800ms interval-------------------------------------------------------------------48 Figure 4.The grand average waveforms of highly and low emotion-polarized nominal collocates of stimuli-------------------------------------------------------------------------------------49. xi.

(13) Chapter 1 Introduction 1.1 Motivation. Mandarin-Chinese words usually “mean” more than their commonly recognized meanings denoted in dictionaries when co-occurring with other words. For example, consider the verb dao-zhi (導致). Although on the surface this verb is noted down as simple as “to bring about; to make something happen (引起, 促成)” (abstracted from The Revised Chinese Dictionary (教育部重編國語辭典修訂本), USL: http://dict.revised.moe.edu.tw), to most Mandarin-Chinese speakers it is very close to “to cause; to make the happening of something unpleasant or something you do not want” since it is usually followed by nouns describing consequences of the kind. In fact, this realistic side of word meaning is already borne in the mind of a proficient Mandarin-Chinese user. Our daily experience with how a certain word is commonly used in real contexts, such as a casual conversation with friends, an article from the magazine, a shopping program aired on the television, or a morning weather broadcast, offers implicit but useful instructions on the proper way of using it. Specifically, the collocational behavior of a word leads to an unconscious postulation on a word’s pragmatic usage; in other words, speaker’s intention of using a certain word and listener’s perception of what the word is made to mean. Xian & McEnery (2006) have once quoted a Chinese saying to illustrate this power of collocational influence, “he who stays near vermilion gets stained red, while he who stays. 1.

(14) near ink gets stained black”. The quote wisely articulates a transformation on word meaning, that the result of a word consistently keeping “bad company” is that “the use of the word alone may become enough to indicate something unfavorable (Xian & McEnery 2006:107)”. Now return to the example dao-zhi (導致). Since it frequently sticks around with object nouns of negativity, we unconsciously acquire important pragmatic knowledge initially not apparent from the denotation of this verb: it has a negative or unfavorable sense. This acquisition of knowledge can also explain why listeners are usually able to predict bad consequences coming after this verb even before speakers produce them. The collocational influence discussed above sums up the basic idea of the core of this study: semantic prosody. As will be reviewed in Chapter 2, the issue first aroused interests among corpus linguists, and then developed into a topic widely examined in many different angles. Although considerable attention has been devoted to semantic prosody, it awaits a cognitive neuroscience study to unveil the nature of it, especially when viewed as an embedded meaningful message that lies within words. It will be expected to help us answer some of the debated questions raised in the existing research; for example, is semantic prosody an element within the word level, or higher than the lexical level, an epistemic or pragmatic matter of issue? To put it in other words, we would like to know at what stages semantic prosody is decoded in our brain during the processing of a word’s content.. 2.

(15) 1.2 Research question. We expect to identify the brain responses to a collocational semantic relation, semantic prosody, as our focus of the study, between positive (or negative) verbs and their habitual object nouns which are mostly expressive with positive (or negative) emotions. This current study therefore anchors specifically on the emotion aspect of semantic prosody. The main research question comes hereafter: Do people process the semantic prosody of Mandarin-Chinese verbs, particularly the positive or negative emotive information induced by collocational relation with nouns? In order to better distinguish this “emotional prosody” effect on verbs, the other set of verbs are recruited in our experiment, whose emotional bond with their habitual object nouns are much weaker in comparison to the “emotional prosody” verbs. This set of verbs thus demonstrates the ones whose affective features actually arise within the semantic meaning of the verbs themselves, and thus are without the help of semantic prosody. With this experimental design applied, we are interested to see to what extent the emotion valence of the target verbs can be electrophysiologically modulated by the highly and low emotion-polarized habitual object nouns. For the verbs whose habitual object nouns are highly emotion-polarized (e.g. a positive verb whose object nouns are common in their affective content for positivity), they are the “emotional prosody” verbs; as for those whose habitual object nouns are low emotion-polarized (e.g. a positive verb whose object nouns are. 3.

(16) not preferred in their affective content for positivity), they are contrasting to “emotional prosody” verbs, with no help of semantic prosody.. 4.

(17) Chapter 2 Literature Review As we hear/read words or expressions that are totally new to us, one direct response of most people may be asking about the meaning of the new words or expressions. This direct reaction is consistent with the viewpoints expressed within some fields of linguistics, where the relationship between words and meanings are most often than not accounted for by a word-centric view of language (Philip 2011:35): that a word must carry a fixed meaning, and must be able to slot orderly into the position left open by the grammar. Later on, thanks to the creation of computational analysis in linguistic study, new branches of study that focus on lexical semantics in the context of applied linguistics, in particular corpus lexicography and computational linguistics, popped out and marked a brand new page on the development of word meaning study (Philip 2011:35). With sufficient language data provided by large corpora, it becomes possible for researchers to investigate how a word is used in actual text, and to identify the recurring patterns in the neighboring cotext of any word form of interest. This revolutionary step associated with computer technology leads to the emergence of a new view of language, which to some extent argues contradictorily to the position of the word-centric view─the idiom principle (Sinclair 1991): words make meanings by their combinations with collocates, and that meanings arise from language text. Since this study aims to answer whether native speakers of Mandarin Chinese have well. 5.

(18) developed an awareness of semantic prosody of Mandarin Chinese verbs─ the unsuspected patterns of form and use revealed by corpus data from the neighboring text where the verbs appear, my literature review will hence start from a general discussion on the idiom principle, an insightful model of language proposed by Sinclair that has truly held my curiosity and grown strong faith inside me along the road building up this study.. 2.1 The idiomaticity of language: The open-choice principle and the idiom principle Two principles are important in understanding how idiomaticity plays a crucial feature of natural language: the open-choice principle and the idiom principle. These two principles represent two conflicting models of interpretation on the issue of “a unit of meaning”: whether it is a word with a fixed meaning, or an idiomatic, phrasal structure with a conventionalized meaning that constitutes the basic unit of meaning in natural language. Between the two principles create a rich continuum (Sinclair 2004a:29). And it is within this continuum that each of the different language patterns finds which of the two extremes it favors more. In other words, there is by no means one all-time perfect model that can tackle all types of language patterns and that complete dependency of only one model is not enough. The open-choice principle is a way of seeing language text as a result of a very large number of complex choices (Sinclair 1991). It is also called a ‘slot-and-filler’ model, as it sees language text as a series of slots which have to be filled from a lexicon that satisfies. 6.

(19) local grammatical restraints. Literally speaking, any word (or a phrase, a clause) can occur in each slot; that is, any single word owns a free choice regarding to where it can occur. At the moment one slot is completed, a very complex pattern of choices operates simultaneously, and by then the only restraint would be the grammaticalness. To give a clearer demonstration, a tree structure may help illustrate the idea here. The slots or the choice points in this principle are similar to the nodes on a tree structure. As one slot (one node) is completed (by a word, phrase, or clause) and checked by its grammaticality, the following tree structure operates simultaneously and starts off the progressive choices of the following structure (Sinclair 1991). An analytic approach conforming to this principle can be categorized under terminological tendency, a tendency for a word to have a fixed meaning in reference to the world (Sinclair 2004a). It is also contended that words must be able to fill in orderly into the position left open by the grammar after they are chosen from the lexicon by the denoted meanings they carry (Sinclair 1991, 2004a). However, complete freedom of choice of a word, as suggested by the open-choice principle, is not unproblematic. There are still facts about language that cannot be captured by the principle. They are listed as below.. (a) It is clear that words do not occur at random in a text. The open-choice principle does not provide substantial restraints on how the consecutive choices are made. 7.

(20) (b) There are sets of linguistic choices which come under the heading of register, seen as large-scale conditioning choices. Once a register choice is made, this kind of social choices could affect and reduce in scope all the slot-by-slot choices. (Sinclair 1991). Before the involvement of computer technology, it is not easy for most linguists to detect word patterning in everyday language use. However, over the last 30 years, the adoption of computerized corpora has not only updated our vision on language investigation, but also suggested new ideas on the nature of language. The idiom principle proposed by Sinclair (1991) was one emerging from this trend, which tries to explain why some word combinations are more frequent, more meaningful, and more significant than others. Essentially, it affirms the phraseological, not compositional, tendency of language (Philip 2011). Clear (1993) also contended that phraseology exerts influence over meaning of each single word: meaning does not come from words in isolation; instead, it is the patterns of words that give rise to meaning. Leaning towards phraseology tendency, the principle of idiom argues that in a language user’s mind stores a large number of semi-preconstructed phrases that each of them constitutes single choices, or single units of meaning, even though these units might appear to be analyzable into segments (Sinclair 1991:110). Examples of such phrases are “a consequence of…” (e.g. a consequence of injury/late arrival) or “once in a …” (e.g. once in a decade/hundred years) (Philip 2011:54). Since these semi-preconstructed phrases are mostly memorized as single units in our lexicon just as the way we memorized true idioms of a 8.

(21) language, the principle is thereby named this way. Collocation and other features of idiomaticity make good evidence for this principle (See Section 2.2.1 for more detail). A semi-preconstructed phrase, in this sense, can be interpreted as a half-completed meaningful structure consisting of more than one word, with its well-formedness relying on what intervenes (Renouf & Sinclair 1991). In brief, the principle of idiom can be illustrated by any two words one can think of in no time. An example provided by Sinclair (1991:110-111) is of course. Although of and course have their own meaning, these two words must unit with each other in order to operate the distinct meaning, “(informal use) used to emphasize that what you are saying is true or correct” as noted in the Oxford Advanced Learner’s Dictionary (abstracted from Oxford Advanced Learner's Dictionary 8th edition CD-ROM). Other examples of this principle can be found in the specified use of synonymous adjectives in noun phrases, a phenomenon commonly referred to as collocation, like strong tea vs. powerful tea and powerful engine vs. strong engine (Halliday 1966). While strong and powerful share very similar cognitive or denotational meaning, on the collocational grounds they are not interchangeable (See also Sinclair 2004a:29; Xian & McEnery 2006:108). The idiom principle addresses problems left by the open-choice principle. In the account of open-choice principle, what follows of will be opened to any candidate as long as it belongs to the nominal group. In the case of of course, course is chosen. However, the. 9.

(22) meaning of course here is not the one mentioned in dictionaries. That is, its meaning is not pertained to the word form course, but only to be found in the phrase. If it were the meaning as noted in dictionaries, then it should be a countable noun in the singular form and thus would have to be preceded by a determiner to be grammatical (Sinclair 1991). But clearly this is not the case in of course. It will be suitable then if we categorize phrases like of course to compounds, where the independent meanings of phrases are not composed by adding up the meanings of its components. Likewise, idioms, proverbs, clichés, technical terms, jargon expressions, and phrasal verbs could also be treated in the same manner. The demonstration on the meaning of of course reiterates the strength of the idiom principle, especially the flexibility in defining the basic unit of meaning stored in our mind: one word, or some co-occurring words grouping together to engender a unit of meaningful structure.. 2.2. Extended unit of meaning: Sinclair 2004a In a sense, the existence of semi-preconstructed phrases indicates a natural tendency of humans to communicate idea via the economic routes to achieve efficient and sufficient communication (Sinclair 1991). Besides, their existence reflects the recurrence of similar situations in human affairs. Sinclair (1991) made one profound explanation on this phenomenon of words.. 10.

(23) To some extent, the nature of the world around us is reflected in the organization of language and contributes to the unrandomness. Things which occur physically together have a stronger chance of being mentioned together; also concepts in the same philosophical area, and the results of exercising a number of organizing features such as contrasts or series (Sinclair 1991:110).. Moreover, Sinclair suggested that the urgencies of on-line conversation may be part of the reason why these semi-preconstructed phrases were shaped in form and became highly accessible to a language user. No matter what motivates the existence of these phrases and how it arises, it has nevertheless been relegated to an inferior position in most current linguistics (Sinclair 1991:110, Philip 2011:35). Sinclair noted seven essential features of language from these semi-preconstructed phrases, which are captured by the idiom principle (Sinclair 1991):. (1) Many phrases have an indeterminate extent (e.g. set eyes on can attract a pronoun subject, never or the moment, the first time, as temporal conjunction, and has as an auxiliary to set). (2) Many phrases allow internal lexical variation (e.g. there seems to be not much difference between in some cases and in some instances; or between set x on fire and set fire on x). (3) Many phrases allow internal lexical syntactic variation (e.g. in “it’s not right for him to…”, is can change to was, not can be substituted by hardly or scarcely, and him can be replaced by any possessive pronoun like her, noun phrase like a reporter, or personal name like Henry). (4) Many phrases allow some variation in word order without affecting their meaning (e.g. it’s not right for him to discriminate can be rearranged to to discriminate is not right for him). (5) Many uses of words and phrases attract other words in collocation (e.g. pleasant journey, hard work, fast food). 11.

(24) (6) Many uses of words and phrases show a tendency to co-occur with certain grammatical choices (e.g. enjoy is followed by nouns or nominalized phrases). (7) Many uses of words and phrases show a tendency to occur in a certain semantic environment (e.g. happen is mostly associated with unpleasant things such in “A terrible car accident happened last night.”).. The features mentioned in (5)-(7) are further referred to as collocation, colligation, and semantic preference respectively. And, along with the fourth element, semantic prosody (which will be introduced in Section 2.2.3), these four features altogether attribute to the “extended unit of meaning” declared in the Sinclarian framework of words and phrases (Sinclair 2004a). Since the notion of colligation is not relevant to this study, it will not be addressed in the following subsections.. 2.2.1 Collocation As a phrase is recurrently used in a language, it becomes a conventionalized structure in mind as parallel to an idiom (Philip 2011). For example, once in a lifetime grounds a lexicogrammatical frame, “once in a…”, which in turn generates a rich list of expressions such as once in a while/long while/blue moon/decade/generation (Philip 2011:54). In this sense a primitive platform with a slot (alike to lexicogrammatical frame of an idiom) surfaces out of the conventionalized phrase later, which becomes prewired in speaker’s mind. Later, language users can put words that convey similar information into the slot of the platform to. 12.

(25) produce many newly-created expressions catering to communicative needs. For example, the right-hand collocate in “once in a…” must refer to a period of time in order to fit into this frame. Collocation was first introduced by J.R. Firth, of which he said “I propose to bring forward as a technical term, meaning by collocation, and to apply the test of collocability” (Firth 1951:194). Later, he described that “collocations of a given word are statements of the habitual or customary places of that word” (Xiao & McEnery 2006:105). This is in line with Sinclair (2004), where collocation was defined as “a frequent co-occurrence of words”. In respect to the frequent and habitual co-occurrence of words, Firth went on and emphasized that “the collocation of a word or a ‘piece’ is not to be regarded as mere juxtaposition. It is an order of mutual expectancy. The words are mutually expected and mutually prehended” (Firth 1957:181). So, Firth’s notion may imply that meaning in a language user’s mind is constructed by not only the individual meaning of word, but the combination of the word and its most expected collocates. It was also postulated that collocation concerns about word forms, not lemmas (Firth 1957). The combination of words can be viewed as a number of fossilized phrases and were hypothesized to be memorized as single units in a language user’s mind (Sinclair 1991). How frequent the co-occurring lexical items must be to be considered as a collocation? Hoey (1991:6-7) suggested that only when a lexical item appears with other items “with. 13.

(26) greater than random probability in its context” can the combination be counted a collocation. Such statistical approach to collocation was in line with Firth’s study, and was also adapted by many other corpus linguists, including Sinclair (1991), Stubbs (1995) and Partington (1998). Mutual information value (MI) is a statistical number frequently used to measure the collocational strength of a given node word in relation to its collocates (Xiao & McEnery 2006). It is determined by two numbers, Freq (y) and Freq (x, y). Freq (y) stands for the number of occurrence for the particular collocate within the database, while Freq (x, y) means the number of occurrence where the target word co-occurs with the particular collocate within the database. An M1 score above 0 implies the tendency for the target word to appear with the particular collocate, with the higher score meaning stronger collocational relationship. In contrast, an MI score below 0 means the target word tends not to co-occur with the particular collocate, and the lower it gets, the stronger the mutual exclusion. Previous research on collocation has devoted primary to English (e.g. Sinclair 1991; Renouf & Sinclair 1991; Stubbs 1995; Partington 1998; Sinclair 2004a), while only little attention has been paid to the phenomenon in other languages. In order to address this oversight, in the following, I shall begin by giving some Chinese examples of collocation. Xiao & McEnery (2006) was a contrastive study between Chinese and English, which compared the collocational behavior of near synonyms and the difference in the affective meanings they conveyed. It showed the significant left-hand collocates of the node, jie2guo3. 14.

(27) ‘result’ (結果) can be divided into two groups according to their lexical categories: Modifiers and Actions. For the Modifiers group, significant collocates are: da4xuan3 ‘general election’, bi4ran2 ‘inevitable’, shi4yan4 ‘experiment’, diao4cha2 ‘investigation’, ke3neng2 ‘possible’, jing1ji4 ‘economic’, and hao3 ‘good’. As for the Actions group, the action-verb collocates are: biao3ming2 ‘show’, zao4sheng2 ‘cause’, zeng1jia1 ‘increase’, chan3sheng1 ‘give rise to’, and you3 ‘have’. Viewing these collocates, a speaker can infer that jie2guo is mostly preceded by modifiers or action verbs in (textual) context. Besides, we can observe that this node word occurs in some specific genres: election, experiment, investigation and economy, and that jie2guo could be either a good result or a bad one, as evidenced by hao3 ‘good’, bi4ran2 ‘inevitable’, or zao4sheng2 ‘cause’. In sum, collocational behavior of words is a realization of the phraseological nature of language, and that some word combinations are outstandingly more recurrent than others in a language.. 2.2.2 Semantic preference As discussed in the previous section, the primitive platform involving jie2guo3 is prewired in speaker’s mind and is variable as different words fill in the slot of the platform. The words are said to carry similar information. With a further look into the left-hand collocates of jie2guo3, they do seem to be lexical items that derive from a few. 15.

(28) semantically-related fields. Consider the Modifiers group of jie2guo3. Among them we find da4xuan3 ‘general election’, shi4yan4 ‘experiment’, diao4cha2 ‘investigation’, and jing1ji4 ‘economic’, of which can fall under a semantic field labelled as “social event or process that usually involves a period of time”. The meaning of these four collocates thus designate respectively the contexts or environments under which the results come from. The rest in the Modifiers group are ke3neng2 ‘possible’, bi4ran2 ‘inevitable’, and hao3 ‘good’, whose meanings are semantically related to each other in that they are used for “indicating the expectancy/possibility or describing the quality” of the results. They therefore form another semantic field. The two semantic fields observed from the Modifier collocates of jie2guo3 illustrate another element of the extended unit of meaning: semantic preference. As its name puts it, this element reveals “a common preference” in semantic meaning or a shared semantic feature found in the collocates of a node word (Hunston 1995; Stubbs 2001; Sinclair 2004a; Sinclair 2004b; Partington 2004; Bednarek 2008; Philip 2011). So the lexical items that can pattern with the node are reduced within a range defined by semantic similarity, suggesting that only those conveying similar information are the most expected ones to co-occur with the node. As the result, they phrase together with the node more often and become conventionalized phrases in speaker’s mind. For some cases, Partington (2004) further specified a sub-category of semantic preference. 16.

(29) where the nodes show a strong tendency to co-occur with items that carry positive (favorable, pleasant) or negative (unfavorable, unpleasant) senses. This sheds light on the present study because it is of particular interest that the habitual co-occurrence can motivate a transfer of positive/negative evaluative meaning to the node word. We may make a temporal attempt to verify the point here with jie2guo3 again. A closer examination of its collocates reveals that they are mostly neutral lexical items, except for two: bi4ran2 ‘inevitable’ from the Modifiers group and zao4cheng2 ‘cause’ from the Actions group. They both indicate something negative (“bad”, “unfavorable” or “unpleasant”) about the results (jie2guo3): something about the result that is rather not expected to happen (bi4ran2+ jie2guo3), or that the result is a disfavored or disapproved one (zao4cheng2+ jie2guo3). So, as jie2guo3 co-occurs with the two lexical items, their negative senses spread over and a negative affective meaning may be transferred to jie2guo31. However, this “coloring” of meaning (Louw 1993) from the nodes’ close surrounds can only be fully discussed as we move a step further into the last abstraction of meaning within Sinclairian’s extended unit of meaning: semantic prosody. It will be detailed in the next section.. 2.2.3 Semantic prosody The notion of semantic prosody has been debated over the last two decades (Sinclair 1991; 1. However, it should be careful in the jie2guo3 case because only two collocates among all, bi4ran2 and zao4cheng2, are found to possess negative semantic senses. Two negative collocates may not be powerful enough to establish a negative affective meaning in jie2guo3. The point made here is purely to demonstrate the idea of the emergence of positive/negative affective meaning in the node word. 17.

(30) Louw 1993; Sinclair 2004a; Partington 2004; Hoey 2005; Bednarek 2008), and therefore the term itself has embraced different interpretations along the way. It started with Sinclair’s observation from the corpus he used as he studied the lexicogrammatical environment of the phrasal verb set in, whose grammatical subjects were found in general to be “unpleasant states of affairs”, including “rot, decay, malaise, despair, ill-will, decadence, impoverishment, infection, prejudice…” to name a few (Sinclair 1991:74-75). The author also noticed that the verb happen “...is associated with unpleasant things─accidents and the like” (Sinclair 1991:112). This interesting observation from corpora data surely brought up to a brand new chapter on the research field of lexical semantics.. 2.2.3.1 Semantic prosody as a connotative meaning Words are judged to be associated with positive or negative evaluation as they co-occur mostly with other words that belong to positive or negative semantic set. The term “semantic prosody” was publicly coined by Louw in his 1993 paper, where it described the “consistent aura of meaning with which a form is imbued by its collocates” (Louw 1993:157). It was specified as a kind of prosody since the phenomenon itself can be understood by analogy with that of “sound prosody” discussed by Firth (Stewart 2010:7)2.. 2. Firth’s discussion on phonological prosody aims to explain how phonemes transcend segmental boundaries to be phonetically realized. For instance, the realization of /k/ of kangaroo is not the same with that of keep because during the articulation of /k/, the mouth also prepares for the production of the sound that follows, which is /æ /. Thus, the actual realization of a phoneme will depend on the sounds adjacent to it, by a process called “phonological coloring” by Louw (1993:158). 18.

(31) Bublitz (1996) went along with Louw and concluded that semantic prosody is evidence for “meaning residing not in a single word but in several words”. He further pointed out that semantic prosody can vary according to the different basic meanings of the given words. This can be illustrated with the verb happen. Whereas Sinclair claimed an unfavorable semantic prosody for happen, this negative prosody stops applying to the verb in the situation when it is used to mean “by chance”, as in “I happened to meet him in person last week”.. 2.2.3.2 Semantic prosody as a pragmatic meaning Sinclair (2004a) investigated semantic prosody from beyond the scope of pure semantics. As mentioned in Section 2.2, Sinclair integrated four related aspects of unit of meaning: (1) collocation (lexical choices of the item), (2) colligation (grammatical choices of the item), (3) semantic preference (preferred semantic fields for the collocates of the item), and (4) semantic prosody. Semantic prosody, the fourth one, is the element that serves attitudinal and pragmatic function to the lexical item. Following Sinclair’s definition, semantic prosody is to explain speaker/writer’s purpose for making the utterance the way it is by using certain words or expressions. It is this definition that is followed in the remainder of the current study.. A semantic prosody …is attitudinal, and on the pragmatic side of the semantic/pragmatics continuum. It is thus capable of a wide range of realization, because in pragmatic expressions the normal semantic values of the words are not necessarily relevant. But once noticed among the variety of expression, it is immediately clear that the semantic 19.

(32) prosody has a leading role to play in the integration of an item with its surroundings. It expresses something close to the ‘function’ of an item─it shows how the rest of the item is to be interpreted functionally. Without it, the string of words just ‘means’ it is not put to use in a viable communication (Sinclair 2004a:34).. It should be notified that in Sinclair’s terminology, a “lexical item” or “item” denotes “a sequence of habitually co-occurring words” instead of a single word. Although this way of defining is not the most common one found in many linguistics studies, it builds on the contributions of previous work and reassures the fact of how a word’s consistent accompany can greatly influence our understanding of the word. Grounded on this argumentation, unit of meaning should arise from the juxtaposition of a word and its frequent collocating partners, and together to be considered as “one unit” or “one lexical item”. Additional importance of this definition is the concern of the pragmatic or epistemic meaning that can be communicated through the use of the whole pack of juxtaposition. We can use an English phrase, symptomatic of, to illustrate Sinclair’s point above. It is found to be habitually followed by lexical items describing unfavorableness, e.g. parental paralysis or management inadequacies (Louw 1993:170). This reoccurring collocational pattern makes the phrase inflected with a negative semantic color, and so when this phrase appears, the negative semantic prosody of it prepares the hearer/reader for the following, e.g. something that is in general negatively evaluated within the language community, or something that is going to be interpreted negatively from speaker/writer’s point of view. In. 20.

(33) other words, the negative semantic prosody has implicitly indicated to the hearer/reader about how the following information should be understood, in this case, “a sign for some bad situations or problems”. This is the very pragmatic function of semantic prosody reasoned about in Sinclair’s definition. Similar to Sinclair’s approach, Stubbs (2001:65) reiterated the pragmatic and discourse roles of semantic prosody by renaming it as “discourse prosody”. Stewart (2010:7) also assumed discourse function to be the primary, explaining that semantic prosody of a lexical item is a meaning established through the repeated use of the lexical item with other words. He remarked that some shared features existing in the regular collocates (e.g. undesirable things for the key verb cause) would be acquired by the verb over time, leading to an epistemic reading or readings imposed on the verb.. 2.2.3.3 Interim summary for semantic prosody From the previous discussions, there appears to be conflicting interpretations with regard to the essence of semantic prosody. Some authors have advocated that it is primarily evaluative or connotative meaning that is attached to the given words or expressions. This school of view has continued with Louw’s focus on the textual effects that can be achieved by producing collocative clash: manipulating words with positive or negative evaluation to collocate with words with the contrasting evaluation. For instance, writers can express irony. 21.

(34) by purposely making a typical negative verb, cause, followed by a positive word, as in “This song causes me happiness.” However, Louw objected to this “aura of meaning” definition of semantic prosody in his later work because it gave too much emphasis to the positive or negative evaluation. He clarified that semantic prosody is essentially a “collocational phenomenon” (Philip 2011:60, and see also Partington 2004; Hunston 2007; Bednarek 2008), and thus is different from connotation, which is more an issue of semantic associations made with that word, regardless of habitual co-occurrence factors. For other researchers, they hold a much more complex conception of semantic prosody. Clinging to Sinclair’s pragmatic approach on the subject, this school underscores the discourse function of semantic prosody: instead of being attached to a single word, semantic prosody is an implied meaning that articulates through a particular combination of words (which together are viewed as one unit of lexical item in Sinclair’s definition). In line with Sinclair, Louw also contended in his 2000 paper, Contextual prosodic theory: Bringing semantic prosodies to life, that semantic prosody is “the expression of the attitude of its speaker or writer towards some pragmatic situation” (Stewart 2010:13-14; Philip 2011:60).. 2.3 The ERPs evidence of emotional words processing The positivity or negativity of semantic prosody of lexical items may trigger listeners’/readers’ emotions. Emotions are of a crucial system in the human brain since it is in. 22.

(35) charge of our innate survival responses to critical situations, e.g., to signal and activate fight, attachment, and sexual behavior (Kissler et al. 2006). Though we express and receive emotional messages through various stimuli in daily life, e.g. face, sound, picture, word, an intriguing question that persists among literature is: how do the “emotional brain” interact with the “linguistic brain”, when written words with emotional connotations are encountered (Kissler et al. 2006:150)? Emotion theories have been converged with neurolingusitics and cognitive semantics, and posited that a linguistic expression can separately activate different subnetworks interconnected within a broader semantic network, with each of them specializing a function, i.e. linguistic features, pragmatic usages, and emotional connotations (Lang 1979; Bower 1981; Barsalou 1999; Kissler et al. 2006). ERP studies further revealed different processing stages by some indicative brain components, with an attempt to summarize how the human brain constructs the emotional content of word-stimuli and to capture how it does so in real time.. 2.3.1 Early effect: N100 Scott et al. (2009) examined whether emotionality of words will facilitate early lexical processes. One brain component looked for in their study was the N1, which had been demonstrated as an index of lexical access (See Hauk and Pulvermüller 2004 for review). Since they wondered how word frequency may play a role in the lexical access of emotional. 23.

(36) content, they also matched the frequency. A total of 80 sets of words were generated respectively for the positive, neutral, and negative word categories, with half of them being high frequency (HF) words and the other half low frequency (LF) words. Another three sets of 80 pseudowords were created as well for the lexical decision paradigm used in the experiment. Twenty-six participants (15 females, 11 males) were presented randomly with the word stimuli and pseudowords. An interaction in the N1 time window (135~180 ms after word onset) showed in their result suggested that frequency was modulated by the emotion valence of words. Within LF words, neutral words were found to generate larger N1 than either positive or negative words, which could be evidence for easier lexical access of words with emotionality. As for HF words, negative words yielded a larger N1 than either positive or neutral words. This was explained with two possible reasons. One was that early activation of high arousal words, such as those in the HF negative word set, may trigger an internal response that represents an enhancement or disruption in the N1 window. The other explained that HF arousal words tend to attract more attentional resources, which was reflected electrophysiologically as an early negative-going wave with larger amplitude for HF negative compared to HF positive words. Kissler et al. (2009) did not find consistent N1 effect for pleasant, neutral and unpleasant adjectives. The study examined whether the response to word’s emotionality depends on the availability of attentional resources. The experiment recruited 20 participants (10 females, 10. 24.

(37) males). A total of 198 words, 99 adjectives and 99 nouns, served as stimuli. Both two groups were further divided into three subgroups varying in emotional content: 33 of them were highly unpleasant, 33 neutral and 33 highly pleasant. The experiment consisted of three consecutive tasks: (1) to silently read the words, (2) to count the adjectives, and (3) to count the nouns. During the silent reading task, there were no significant main effects, only the N1 amplitude being slightly larger to pleasant words and tending to be larger over the right hemisphere. As for adjectives- and nouns-counting tasks, where subjects were selectively attentive to the word, still no clear main effects of task or emotional content of words emerged, except that N1 to pleasant words over the right hemisphere was reduced compared to neutral or unpleasant words. The authors concluded that these effects of N1 were too small for a solid interpretation. Taken together, Scott et al. (2009) showed that in a word-pseudoword lexical decision task, the N1 amplitude was reduced for words with emotionality, but was increased especially for HF arousal words such as HF negative words. Kissler et al. (2009) pointed out that when attention was directed to words as in adjectives- and nouns-counting tasks, the pleasant words evoked smaller N1 amplitude over the right hemisphere than both the neutral and unpleasant words. Otherwise, when the attention was relatively weak to words, as in the silent reading task, pleasant words induced slightly larger N1 compared to the other two word categories.. 25.

(38) 2.3.2 Early effect: P200 Kanske & Kotz (2007) examined how concreteness and emotionality can influence visual word processing, and expected to find early (P2) and late (N400, LPC) ERP effects elicited by emotionality. Two experiments were conducted. Thirty participants took part in Experiment 1, in which a visual hemifield lexical decision task was used. The task consisted of concrete and abstract words of negative (N=60), neutral (N=120), and positive (N=60) valence, as well as pseudowords (N =240). Participants responded to word or pseudoword by button press, with a left and right key representing word or pseudoword choices. The result showed a main effect of emotion in the positive words, reporting larger P2 amplitude elicited by positive words compared to neutral words. Since in Experiment 1 the behavioral response overlapped with the LPC time window, a second experiment was conducted in which participants only pressed button to pseudowords in a go/no-go lexical decision task. However, this time neither emotion nor concreteness modulation reached significance effects in the P2 time window. To explain why P2 effect was only present in the choice RT lexical decision task exployed in Experiment 1, the authors claimed that the difference in the level of word processing depth between the two tasks may be the vital factor. In the go/no-go task, participants only needed to detect pseudowords, whereas in the choice RT task each word-stimulus was evaluated in terms of its lexical status, during which the emotional content of words was assessed (Kanske & Kotz 2007:143). According to this proposal, a P2 emotion. 26.

(39) effect requires a certain level of processing depth, which may also account for the absence of P2 effect in letter-identification task employed in Begleiter et al. (1979).. 2.3.3 Early effect: EPN (The early posterior negativity) A posterior negative wave occurring between 200 and 300 ms after word onset has not only been reported to reflect early semantic processing (Sereno et al. 2003), but also found to be sensitive to a word’s emotionality. As revealed in Scott et al. (2009), a main emotion effect was observed with larger EPN amplitude elicited by both positive and negative words in comparison with neutral words. Kissler et al. (2009) obtained similar results, observing the EPN emotion effects in silent reading task and in word counting task as well. Planned comparisons revealed that the EPN of both unpleasant and pleasant words differed significantly from that of neutral ones, when participants simply read the stimuli presented on the screen and were free to allocate their attention as they wished. Subsequently, the task changed and the participants were instructed to attend to the word type (adjective or noun) to count the respective occurrences. This time, the emotional content of words still exerted an impact on the EPN, as planned comparison showed that the EPN to unpleasant and pleasant words was different from that to neutral words. In Schacht & Sommer (2009) the primary aim was to provide ERP evidence to the notion. 27.

(40) that emotion effects in word processing resemble that in face processing. Twenty-four participants (16 females, 8 males) took part in their experiments. They performed a lexical decision task on single verbs and face decision task, and were instructed to answer as fast as possible whether or not the presented letter string was a correct word or the displayed face was intact by button pressing. The target verbs set consisted of 40 verbs for each positive, neutral and negative word categories. All 120 target verbs were controlled for word frequency, number of orthographic neighbors, word length, and imageability. 120 pseudowords were created and matched to word stimuli with respect to word length. As for the face decision task, 240 portraits of different persons (half females, half males) were generated, containing 40 intact faces for angry, happy, or neutral facial expression. 120 distractor faces were produced by applying smearing tool to one of the facial features, i.e. left or right eye, month, or nose, 30 faces for each smeared facial feature. All portraits were edited to a unitary format. Significant early emotion effects were only observed from word and face stimuli of positive valence in the respective time windows: a posterior negativity to happy faces was triggered around 150 ms, and a bilateral occipito-temporal negativity and a fronto-central positivity to positive verbs were found around 380 ms. The posterior negativity elicited by positive words (e.g. at 380 ms) and happy faces (e.g. at 150 ms) were indistinguishable in scalp distribution and were typical for the EPN.. 28.

(41) 2.3.4 Late effect: N400 The N400 is widely known for its response to semantic incongruities (Kutas & Hillyard 1980, 1984; Kutas & Federmeier 2000). However, it is not simply an index of semantic anomaly. Kutas & Federmeier (2000) pointed out that it should be interpreted as a normal response to words and word-like stimuli (e.g. pseudowords). The N400 amplitude to a word has been discussed in different contexts, whether that context was a sentence, a discourse, or even a single word. In sum, N400 effects reflect the integration between a particular word and the sentential, discourse-level of context information, as well as the access into long-term semantic memory when a particular word is encountered (Kutas & Federmeier 2000). As for the electrophysiological response, the reduction of its amplitude often represents ease of processing or information retrieval (van Petten & Kutas 1990). Kanske & Kotz (2007) provided a new window into the N400 with their study. As reviewed earlier, their study examined how concreteness and emotionality could be two influential factors in word processing. Two experiments were conducted, and both of them observed N400 effects for concreteness and emotionality. In Experiment 1, within 390~590 ms after stimuli presentation, they found a main effect of concreteness, with concrete words eliciting a larger amplitude than abstract words. A main effect of emotion was also found in this time window, with both the two neutral conditions activating larger amplitudes than the corresponding positive and negative words.. 29.

(42) Although the N400 effect has not frequently reported in emotion processing, based on their findings, the authors reasoned that positive, negative and neutral words may not merely represent emotional categories, but they were also grouped as distinguished semantic categories in the semantic networks. As the N400 is an indication for semantic processing (Kutas & Federmeier 2000), it was proposed as a brain component modulated by emotional semantics (Kanske & Kotz 2007). A decreased N400 amplitude for emotional words compared to neutral words therefore implied facilitated processing when the content of a word was involved with emotion, which was in accord with the behavioral data shown in Experiment 1. As for the emotion effect in Experiment 2, although there was a numerical difference found between positive and neutral words, only negative words were reported to differ significantly from the corresponding neutral words. The authors explained that the decreased level of processing depth in the go/no-go task may reduce the difference between positive and neutral words.. 2.3.5 Late effect: LPC (The late positive complex) Prior studies of emotional word processing have consistently reported effects of word emotionality on the LPC around 500 ms after word onset, mostly yielding enhanced LPC amplitudes for both pleasant and unpleasant than for neutral words (Schapkin et al. 2000;. 30.

(43) Fischler & Bradley 2006; Herbert et al. 2006). For example, Schapkin et al. (2000) found larger P300 and LPC to pleasant compared to neutral and unpleasant words during emotion decision task. Herbert et al. (2006) also reported a larger LPC to differentiate pleasant words from unpleasant words. Schacht & Sommer (2009), as mentioned before, identified two stages of emotion effects in the processing of emotional words and faces with a decision-making paradigm. They found that the amplitude of the LPC was enhanced for both positive and negative verbs, starting around 540 ms. As for the face domain, only angry faces augmented the LPC at around 450 ms. The findings were reported to replicate previous studies on affective picture processing and emotional word processing (Fischler & Bradley 2006). Importantly, such LPC enhancements were recognized as an extension of the P300 component, which is augmented when the stimuli require active cognitive analysis. Thus the authors explained the LPC effects in emotional word processing should be attributed to the relevance of stimuli to the task and the arousal value of stimuli.. 31.

(44) Chapter 3 Methods 3.1 Participants Thirty-two native Mandarin-Chinese speakers (all undergraduate students of National Taiwan Normal University, 24 females), aged between 20 to 32 years (Mean age = 22.91, SD = 2.87) took part in the ERP experiment. They were all strongly right-handed, had not previously been diagnosed with neurological or psychiatric disorders, and had normal or corrected-to-normal version. In accordance with the guidelines approved by the Research Ethics Office of National Taiwan University, written informed consent was obtained from these participants prior to the experiment. They were paid for their participation after the experiment.. 3.2 Materials It is a general tendency that a word is in consistent company with a particular semantic set, whose semantic features will gradually attach onto the word, yielding a semantic prosody of the word. The current study anchored on the emotion aspect of semantic prosody. To this end, we used Mandarin Chinese two-character transitive verbs as the target stimuli. And in order to better distinguish the brain responses to the “emotional prosody” effect of verbs, the other set of verbs were recruited in our experiment, whose emotional bond with their habitual object nouns were much weaker in comparison to the “emotional prosody” verbs. This set of. 32.

(45) verbs thus demonstrated the ones whose affective contents actually arose within the semantic meaning of the verbs themselves, and thus were without the help of semantic prosody. Taken together, the experimental design applied in this study was to examine whether the emotion valence of the target verb can be modulated by the emotional polarization of target verb’s habitual nominal collocates (i.e., the high or low emotion-arousing degree of the following high-frequency object nouns). A 3 x 2 factorial design was used in this study, yielding six experimental conditions on the basis of two dimensions: (1) Valence: negative, neutral, or positive valence of the target verb, and (2) Polarization: high or low emotion-arousing degree of the verb’s top five habitual nominal collocates (See Table 1 for the six conditions). Each condition contained 35 two-character transitive verbs, and thus there were 210 verbs in total for the experiment.. Table 1.The six experimental conditions Valence. negative neutral positive. Polarization. Condition. Semantic prosody in emotion. Example. High. NH. strong. niang-cheng (釀成). Low. NL. weak. po-huai (破壞). High. NUH. strong. jian-qing (減輕). Low. NUL. weak. pei-dai (配戴). High. PH. strong. chuang-zao (創造). Low. PL. weak. an-fu (安撫). For the verb-stimuli, a pool of 719 Mandarin Chinese two-character verbs was firstly obtained from a Chinese Dictionary (國語活用辭典) and Chinese Gigaword Corpus from. 33.

(46) Chinese Word Sketch Engine (URL: http://wordsketch.ling.sinica.edu.tw/). These verbs were then filtered through the following steps to finalize the 210 verb-stimuli. Each process will be discussed in order in the sections below.. 3.2.1 Emotion valence rating on verbs To ensure the emotion valence of verbs, an emotion-rating questionnaire was conducted online in which all the 719 Mandarin Chinese two-character transitive verbs was evaluated by a normative group of 30 participants with the 7-point-scale (1 = very negative, 4 = neutral, 7 = very positive) (See Appendix A for the instruction of the rating questionnaire). The following criteria for word selection were employed. Each valence group had 70 verbs. Valence values were less than 3.8 for negative verbs, greater than 4.9 for positive verbs, and between those values for neutral verbs. The valence rating differed significantly among all the three valence groups (F(2,68) = 735.569, p < .001), with the positive valence rated higher than the neutral valence, and the neutral valence higher than the negative valence (Positive vs. Neutral: t(69) = -20.011, p < .001; Positive vs. Negative: t(69) = -37.563, p < .001; Neutral vs. Negative: t(69) = -20.79, p < .001).. 3.2.2 Polarization of verbs’ habitual nominal collocates Among each valence group, we further manipulated emotional polarization (i.e. the high. 34.

(47) or low emotion-arousing degree of the verb’s top five following object nouns) so that there were 35 highly emotion-polarized verbs and 35 low emotion-polarized verbs under each valence group. This manipulation was done by consulting the statistical data generated from Chinese GigaWord Corpus (where the gigaword2all corpus was chosen), which was operated through the platform of Chinese Word Sketch Engine. After the target verb was set in the corpus, firstly, we applied the frequency function to the first candidates on the right side of the target verb (named 1R by the corpus), and selected to display their grammatical class. Secondly, these 1R candidates were filtered by the collocation function to generate only those that collocated with the target verb. Lastly, these 1R collocating candidates were sorted by the frequency function again, this time displaying the word. By so doing, the final corpora result showed not only the 1R collocates, ranking from the most frequent one to the less, but also the grammatical class of these 1R collocates. The main criterion for the operational definition of high or low degree of emotion polarization relied on the first five frequent nominal collocates. For High conditions, the five nominal collocates of each verb were biased for a specific emotion valence. Take the NH condition for example (i.e. the High condition under the negative valence). The Positive: Neutral: Negative distributional patterns of the five nominal collocates were 0:0:5, 1:0:4, or 0:1:4, showing a preference for negative valence. Other patterns of distribution (with the value of Negative being less than 4, including 0:5:0, 0:4:1, 1:4:0, 0:2:3, 0:3:2, 2:3:0) were. 35.

(48) categorized for the NL condition (i.e. the Low condition under the negative valence). Likewise, the Positive value was 4 or 5 in the distribution for the PH condition (including 5:0:0, 4:0:1, 4:1:0), leaving the rest (with the value of Positive being less than 4, including 0:5:0, 1:4:0, 2:3:0, 2:2:1) the PL condition. As for the NUH condition, either the Positive or the Negative value reached 4 or 5 in the distribution (including 5:0:0, 4:0:1, 4:1:0, 0:0:5, 1:0:4, 0:1:4), leaving other distributional patterns (i.e. the value of Positive or Negative being less than 4, including 0:5:0, 1:4:0, 0:4:1, 0:3:2, 1:3:1, 2:2:1) categorized under the NUL condition. See Tables 2-4 for the illustration on the high and low emotion polarization of verb-stimuli’s top five nominal collocates, respectively, for the negative, neutral, and positive valence groups.. Table 2. An illustration on the high and low emotion polarization of verb stimuli’s top 5 nominal collocates: The negative valence group Negative Valence (of verb-stimuli) 釀成破壞 Verb-stimulus Top 5 nominal collocates Collocate 1 災害 2 大禍 3 悲劇 4 意外 5 慘劇 Positive:Neutral:Negative Polarization (of the 5 nominal collocates) Condition. Emotion negative negative. Collocate 社會秩序國家稅制. Emotion neutral neutral. negative negative negative. 環境生態生態環境. neutral neutral neutral. 0:0:5. 0:5:0. High. Low. (highly polarized for negativity). (no explicit emotion). NH. NL. 36.

(49) Table 3. An illustration on the high and low emotion polarization of verb stimuli’s top 5 nominal collocates: The neutral valence group Neutral Valence (of verb-stimuli) 減輕配戴 Verb-stimulus Top 5 nominal collocates 1 2 3 4 5. Collocate 農民負擔企業個體負擔政府財政負擔學生負擔體重. Positive:Neutral:Negative Polarization (of the 5 nominal collocates). Emotion negative negative negative negative neutral. Collocate 隱形眼鏡安全帽眼鏡識別證口罩. Emotion neutral neutral neutral neutral neutral. 0:1:4. 0:5:0. High. Low. (highly polarized for negativity). (no explicit emotion). NUH. NUL. Condition. Table 4. An illustration on the high and low emotion polarization of verb stimuli’s top 5 nominal collocates: The positive valence group Positive Valence (of verb-stimuli) 創造安撫 Verb-stimulus Top 5 nominal collocates 1 2 3 4 5 Positive:Neutral:Negative Polarization (of the 5 nominal collocates) Condition. Collocate 條件經濟奇蹟歷史價值財富機會. Emotion neutral positive positive positive positive. Collocate 民心人心民眾情緒群眾. Emotion neutral neutral neutral neutral neutral. 4:1:0. 0:5:0. High. Low. (highly polarized for positivity). (no explicit emotion). PH. PL. 3.2.3 Frequency Lastly, in order to avoid potential modulation on emotion valence from word frequency (Scott et al. 2009), verb frequencies were controlled by checking the frequency information. 37.

(50) in the Chinese Gigaword Corpus. A 3 (Valence: Negative, Neutral, Positive) x 2 (Polarization: High, Low) ANOVA was conducted to reassure that verb frequencies did not differ among the six experimental conditions (Valence x Polarization: F(1,34) = 1.15, p = .291).. 3.3 Procedure The experiment was conducted in the Neurolinguistics Lab at National Taiwan Normal University. Prior to the experiment, the participant read and signed a consent form, and was then interviewed with a demographic questionnaire, which included his/her health and medical status, linguistic background, handedness, and language use. Hereafter, the participant was put on a cap with electrodes to measure his/her brainwaves. Since there were six electrodes that would be attached onto the skin around the eyes and behind the ears, we used an alcohol pad to clean the skin first, and then applied some abrasive gel to remove the dead skin. In order to better measure the brain signals, some gel was inserted into the electrodes on the cap to improve the electrical conductance. After the experiment setup, the participant was guided to sit in front of a computer screen at a distance of 90100 cm and to put his/her right/left (which hand to use was counterbalanced across participants) index finger, middle finger and ring finger on the respective three central buttons of the response box. Then, the participant was instructed to evaluate the target verbs (which appeared at the center of the black screen, one verb at a time). 38.

(51) with their negativity, neutrality, or positivity. The participant was told to give response until a response cue showed up, and then respond via button pressing. The participant was encouraged to judge deliberately on the emotion valence of each verb before giving response, and reminded that he/she got to answer each trial at his/her own pace. After the response, there would be a 2-second interval for eye blinking. Each trial began with a central white cross at the center of the black screen. It was presented for 700 ms. Then a verb-stimulus was displayed in white, with a duration of 1.5 seconds. As stimulus presentation ended, a response cue in yellow (“ ? ”) appeared and stayed on the screen until the participant answered by pressing one of the three buttons (3 seconds as the upper limit). The screen then turned blank for 1 second after button pressing. Lastly, at the end of each trial, a symbol of eye blinking showed up signaling that blinking was possible during a 2-second interval. After a 500 ms blank interval, a white cross showed up again to signal the start of a new trial (See Figure 1 for a demonstration on the procedure of a trial). Experimental trials were randomly presented for each participant. Except for the 2-second interval for eye blinking use, the participant was reminded to keep the eyes fixated on the center of the screen, and was told to minimize body movements (e.g., blinking, eye movements, and muscle movements) since any movement could contaminate the electrical signals. A practice block of 10 trials was first provided to make the participant accustomed to the task and the experiment procedure. The button pressing was counterbalanced across. 39.