一個以語料庫為本對現代漢語中反義詞共現現象之研究

全文

(1)國立臺灣師範大學英語學系碩士論. 文. Master Thesis Graduate Institute of English National Taiwan Normal University. 一個以語料庫為本對現代漢語中反義詞共現現象之研究. A Corpus-based Approach to Antonym Co-occurrences in Mandarin Chinese. 指導教授：畢永峨博士 Advisor: Dr. Yung-O Biq 研究生：許展嘉 Student：Chan-Chia Hsu. 中華民國九十七年七月 July, 2008.

(2) 摘要. 本論文採用以語料庫為本的研究方法，探討現代漢語中反義詞共現之現象。本研究一共選擇了二十五組反義詞，從中文詞彙特性素描系統(Chinese Word Sketch Engine)中選取三千六百零五個句子作為檢測語料。本研究的分析採用 Jones (2002)針對反義詞共現之篇章功能所提出的理論架構。在中文裡，我們發現十三個反義詞共現的篇章功能。其中，對等反義詞 (Coordinated Antonymy)與輔助反義詞(Ancillary Antonymy)是最常見的功能。除此之外，本研究發現中文反義詞的篇章功能分佈會受到以下因素的影響：傳統上反義詞分類的類型、詞類、現代化程度、詞素音節結構。本研究更進一步發現，反義詞共現篇章結構之跨語言差異，可歸因於各個語言獨特的語法特徵。本論文也發現，中文裡每種反義詞共現的篇章功能，皆有其典型的詞彙語法框架。這些不斷出現的構式可能會增強反義詞之間的連結。我們又發現當中文反義詞共現時，會偏向特定的詞序。在中文裡，影響反義詞共現詞序的因素包括正面性與詞頻。當篇章中的觀點轉變時，另一種共現詞序可能會出現。本論文探討現代漢語中反義詞共現之現象，為 Jones (2002)針對反義詞篇章功能所提出的理論架構提供跨語言的支持，也觸及中文反義詞共現現象所反映出的認知機制。本篇是一針對傳統上屬於語義學的反義詞現象，以語料庫及篇章分析的角度所做的研究，彰顯出反義詞之間的橫列(syntagmatic)關係。我們建議未來關於反義詞的研究可採用跨語料庫、跨語言的研究方法，探討反義詞此一語內現象，與各語言獨特的語法特徵及各項語外因素之互動。. i.

(3) ABSTRACT. The thesis adopts a corpus-based approach to investigating how antonym co-occurrences in Mandarin Chinese function in text. In total, 25 antonymous pairs are selected for analysis, and 3,605 sentences are sampled from the Chinese Word Sketch Engine. The analysis is mainly based on Jones’ (2002) functional framework. In Mandarin Chinese, thirteen textual functions of antonym co-occurrences are identified. Coordinated Antonymy and Ancillary Antonymy are the most dominant functions. In addition, it is found that the functional distribution of antonyms in Mandarin Chinese may be influenced by their traditional category of semantic oppositeness, grammatical category, extent of modernization, and morpho-syllabic structure. Moreover, cross-linguistic discrepancies in how antonyms function in text are attributed to language-specific structural properties. Furthermore, the thesis identifies the typical lexico-syntactic frames that are associated with the textual functions of antonymy in Mandarin Chinese. The recurrent linguistic formulas may enhance the pairing between antonyms. In addition, antonyms in Mandarin Chinese are also found to prefer a particular sequencing in text. In Mandarin Chinese, factors that affect antonym sequencing include positivity and frequency. However, the reverse sequencing order shows up when perspective shifts take place in text. The. thesis. on. antonym. co-occurrences. in. Mandarin. Chinese. lends. cross-linguistic support to Jones’ (2002) functional framework and touches on the cognitive facet of antonymy in Mandarin Chinese. The thesis illustrates an attempt to treat antonymy, a traditionally semantic issue, from the corpus and discourse analysis perspective, focusing on the syntagmatic aspect of antonymy. It is suggested that further research on antonymy should take a multi-corpus, cross-linguistic approach to ii.

(4) examine how such a language-internal phenomenon interacts with language-specific structural properties and language-external factors.. iii.

(5) ACKNOWLEDGEMENTS 撰寫論文是一段漫長的旅程。這段旅程充滿了挑戰，常常讓我傷透腦筋。然而，在解決問題的過程中，卻也讓我發現到語言學這塊領域令人驚奇之處，不枉先前的努力。在這段旅程中，有許多人幫助過我，在此要向他們致上由衷的感謝。首先，我要向我的指導教授，畢永峨老師，致上最深的謝意。我與畢老師的緣分在我正式進入師大英語系之前就開始了。當時參加入學口試時，我緊張萬分，是畢老師的微笑讓我的心情放鬆下來。進了研究所之後，我當了畢老師兩年的研究助理。在替畢老師工作時，除了學習到語言學的知識，更重要的是，我學習到做研究時認真的態度。雖然沒有修過畢老師的課，但畢老師相信我可以將論文完成，願意擔任我的指導教授，給了我很大的信心。從一開始到現在，畢老師花了很多時間與我討論語料，替我訂出計劃，給了我許多建議，忍耐我錯誤百出的初稿。若是沒有畢老師的引導與鼓勵，我絕對不可能完成這篇論文。我也要感謝兩位口試委員，林雪娥老師與安可思(Kathleen Ahrens)老師。我一年級時修了林老師的語用學，林老師當時耐心地引導我們的期末報告，最後好多位同學的期末報告都可以在國際級的研討會發表，而我參加了第十八屆北美漢語語言學會議。這是我第一次參加研討會，都要歸功於林老師。我很感激林老師願意來擔任這篇論文的口試委員，繼續給予我建議。我三年級時在臺大語言所旁聽安可思老師的詞彙語意學。雖然我只是旁聽生，安老師在課堂上對我的關心一點也不減，而且對於我的每份作業都給了評語。我很感謝安老師對我的照顧與鼓勵，並且願意來擔任這篇論文的口試委員，給了我很多有用的建議。接著，我想感謝師大英語系的所有老師。我從修過的課中學到語言學的知識，點點滴滴累積到可以開始寫論文的程度。更重要的是，我在課堂上的討論中得到許多靈感與啟發，讓我找到繼續往前的動力。除此之外，我還要感謝我一大群好朋友提供我精神上的支持。我有一群很特別的同學：何信昌、黃郁欣、謝孟璇、張建斌、施姵如、沈正嵐、蔡貴如、孫安霖。因為有了他們，我的研究所生活充滿了歡樂，很多老師都說我們是最吵鬧的一屆。此外，在課堂討論之中，大家的想法激盪出許多有趣的火花，讓我看到我從來沒有思考過的面向。其中，我要特別感謝何信昌、張建斌與沈正嵐：感謝何信昌常常替我修改我的文章，感謝張建斌願意在深夜來搶救我的電腦，感謝沈正嵐願意陪我聊些垃圾話。另外，我還要特別感謝兩位高中同學：王炳勻與林柏仲。有趣的是，我們三個都選擇唸語言學。他們在我沮喪的時候聽我抱怨、陪我逛街，王炳勻還常常給我一些研究上的建議。對於這段將近十年的友誼，我心存感激。最後，我由衷地感謝我最愛也最愛我的家人，特別是我的媽媽。雖然他們似乎不清楚我究竟在唸什麼，我的家人總是以我為榮，並且毫無保留地提供我物質上與精神上的支持。正因為他們的愛，讓我有勇氣面對前方的重重挑戰。我要將這篇論文獻給我在這段漫長的旅程中所有的貴人，希望我沒有讓大家失望。. iv.

(6) TABLE OF CONTENTS Chinese Abstract English Abstract Acknowledgements Table of Contents List of Tables. i ii iv v viii. Chapter 1 Introduction 1.1 General Background 1.2 Research Questions 1.3 Organization of the Thesis. 1 1 3 4. Chapter 2 Literature Review 2.1 Intuition-based Approaches to Antonymy 2.1.1 Complementary Antonymy 2.1.2 Gradable Antonymy 2.1.3 Relational Antonymy 2.1.4 Directional Antonymy 2.1.5 Interim Summary. 6 6 7 9 11 12 13. 2.2 2.2.1 2.2.2. 2.2.3. Data-based Approaches to Antonymy 14 Antonymy: Substitutability versus Co-occurrence 14 Textual Functions of Antonym Co-occurrences 16 2.2.2.1 Fellbaum (1995) 17 2.2.2.2 Jones (2002) 18 2.2.2.3 Follow-up Studies of Jones’ (2002) Functional Framework of Antonymy 25 Interim Summary 27. 2.3. Summary. 28. Chapter 3 Methodology 3.1 Corpus 3.2 Antonymous Pairs for Analysis 3.3 Sampling Method 3.4 Summary. v. 30 30 32 39 41.

(7) Chapter 4 Functions of Antonymy in Mandarin Chinese 4.1 Functions of Antonymy in Mandarin Chinese: Definitions and Examples 4.1.1 Ancillary Antonymy 4.1.2 Coordinated Antonymy 4.1.3 Comparative Antonymy 4.1.4 Distinguished Antonymy 4.1.5 Transitional Antonymy 4.1.6 Negated Antonymy 4.1.7 Extreme Antonymy 4.1.8 Idiomatic Antonymy 4.1.9 Specified Antonymy 4.1.10 Associative Antonymy 4.1.11 Simultaneous/Equivalent Antonymy. 43 44 46 47 49 51 52 53 54 55 56 58. 4.1.12 4.1.13 4.1.14. Transitive Antonymy Negated Ancillary Antonymy Interim Summary. 59 61 62. 4.2 4.2.1 4.2.2. Database Distribution Overall Distribution Factors that Affects the Database Distribution 4.2.2.1 Traditional Category of Oppositeness 4.2.2.2 Grammatical Category 4.2.2.3 Extent of Modernization 4.2.2.4 Morpho-syllabic Structure 4.2.2.5 Morphological Structure 4.2.2.6 Near Synonyms Interim Summary. 63 63 70 71 75 77 79 81 82 85. 4.3.1 4.3.2 4.3.3. A Cross-linguistic Comparison: Antonym Co-occurrences in Mandarin Chinese and English Overall Distributions of Antonymy in Mandarin Chinese and English Pair-by-pair Comparison Interim Summary. 87 87 89 96. 4.4. Summary. 97. 4.2.3 4.3. Chapter 5 5.1 5.2. Constructions of Antonym Co-occurrences in Mandarin Chinese Constructions of Coordinated Antonymy Constructions of Ancillary Antonymy. 43. vi. 98 98 103.

(8) 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11. Constructions of Transitive Antonymy Constructions of Transitional Antonymy Constructions of Specified Antonymy Constructions of Negated Antonymy Constructions of Comparative Antonymy Constructions of Associative Antonymy Constructions of Idiomatic Antonymy Constructions of Distinguished Antonymy Summary. Chapter 6 Antonym Sequences in Mandarin Chinese 6.1 Sequence Statistics 6.2 Sequence Principles 6.2.1 Positivity. 106 108 112 115 117 120 121 123 125 130 130 133 133. 6.2.2 6.2.3 6.2.4. Frequency 137 Why do antonyms sometimes co-occur in an uncanonical sequence? 138 Interim Summary 139. 6.3 6.4. Sequencing Preferences Summary. 139 143. Chapter 7 Concluding Remarks 7.1 Findings of the Thesis 7.2 Implications of the Thesis 7.3 Suggestions for Further Research. 145 145 149 150. References. 153. vii.

(9) LIST OF TABLES Table 1 Table 2 Table 3 Table 4 Table 5 Table 6 Table 7 Table 8 Table 9 Table 10 Table 11 Table 12 Table 13 Table 14. Cruse’s (1986) framework of verbal complementary antonymy Cruse’s (1986) classification of adjective gradable antonymy Antonymous pairs selected in Jones (2002) Functions of antonymy identified in Jones (2002) Antonymous pairs selected for analysis Distribution of pairs across textual functions Functional distribution of antonymy in Mandarin Chinese Distribution of antonymy in Mandarin Chinese by traditional category of oppositeness Distribution of antonymy in Mandarin Chinese by grammatical category Functional distribution of maifang/maifang, fangdong/fangke, laoshi/xuesheng, and fumu/haizi Functional distributions of antonymy in Mandarin Chinese and English Comparison of the results in the present study and Jones (2002) Constructions typical of each textual function of antonym co-occurrences in Mandarin Chinese Antonym sequences in Mandarin Chinese. viii. 8 10 20 22 37 64 69 71 75 82 88 90 126 132.

(10) Chapter 1 Introduction. 1.1. General Background Antonymy is commonly understood as a concept which refers to a pair of words. with opposite meaning. It is also commonly believed that antonyms help people to organize their perception of various phenomena in the world. For example, it is fairly natural and fundamental to divide the human race into men and women (Trudgill 2000:61). This division is taken for granted, encoded in antonym pairs such as man/woman, male/female, and masculine/feminine. In the field of linguistics, antonymy is an intriguing topic with a host of issues discussed in the literature. For instance, how is semantic oppositeness defined? How are words associated with each other as antonyms? No consensus has been reached so far. Traditional approaches to the study of antonymy are intuition-based, with a focus on making logical distinctions between members in an antonymous pair. There are several well-established categories of meaning oppositeness, such as complementary antonymy, gradable antonymy, relational antonymy, and directional antonymy (Cruse 1986; Leech 1974; Saeed 1997). With the rapid development of corpus linguistics, a significant finding of many corpus-based studies on antonymy is that antonyms co-occur far more frequently than expected by chance (e.g., Fellbaum 1995; Jones 2002; Justeson and Katz 1991). Therefore, more and more attention has been directed to the syntagmatic aspect of antonymy, i.e., antonym co-occurrences in language use. One of the most important studies of antonym co-occurrences is Jones (2002). With the aid of a 280-million-word corpus, Jones (2002) proposes eight textual 1.

(11) functions of antonymy in English. In addition, each of these textual functions of antonymy is associated with some particular lexico-syntactic frames, which are argued to contribute to antonym associations in the mental lexicon. Furthermore, antonyms in English are found to prefer a particular sequence in text. Finally, the same functional framework has also been used to account for antonyms in child-produced speech, child-directed speech (Jones and Murphy 2005), and spoken language (Jones 2006). To our knowledge, the only study on antonym co-occurrences in Mandarin Chinese is Lien (1989), in which four functions are identified. For example, antonym co-occurrences in Mandarin Chinese might signal “totality” (Lien 1989:283), as exemplified in the following:. (1). 東拉西扯 dong-la-xi-che east-drag-west-pull ‘talk at random’ (taken from Lien 1989:299). According to Lien (1989), the antonyms dong ‘east’ and xi ‘west’ in (1) function as polar affixes (cf. Zhu 1982:36), signaling the infinity of the event la-che ‘to drag-pull’. While Lien (1989) has paid attention to the functions of antonym co-occurrences in Mandarin Chinese, he limits the research scope to four-character phrases with antonyms as affixes and does not consider authentic production data. Presently, there is no study that takes a broader scope nor a corpus-based perspective to explore the textual functions of antonymy in Mandarin Chinese. To fill the gap, this thesis adopts a corpus-based approach to investigating how antonyms in 2.

(12) Mandarin Chinese function in text. Relevant issues such as lexico-syntactic frames associated with antonyms and antonym sequence preferences are also addressed. In the following section, research questions to be investigated in this thesis are summarized.. 1.2. Research Questions This thesis poses the following research question. Jones (2002), having. established a list of 56 antonymous pairs, comprehensively investigates how antonyms in English function in text. With a data-based approach, Jones (2002) identifies eight textual functions of antonym co-occurrences, touching on the syntagmatic aspect of antonymy and complementing traditional approaches to antonymy. To our knowledge, Jones’ (2002) functional framework has not been applied to languages other than English. From a cross-linguistic perspective, we would like to explore the extent to which Jones’ (2002) functional framework of antonymy is able to account for data in Mandarin Chinese. The purposes of this research question are twofold, i.e., to investigate how antonyms in Mandarin Chinese function in text, and to examine the generality of Jones’ (2002) functional framework of antonymy. After arriving at a holistic picture of how antonyms in Mandarin Chinese function in text, the present study will zoom in to examine to what extent the functional distribution of antonymy in Mandarin Chinese is affected by various linguistic variables. Some variables, such as an antonymous pair’s traditional category of oppositeness, are also Jones’ (2002) concerns; others, such as the morpho-syllabic structure of the antonym words, reflect structural properties specific to Mandarin Chinese and are explored further.. 3.

(13) From a constructional perspective, what constructions 1 in Mandarin Chinese can serve as lexico-syntactic frames for antonym co-occurrences? In constructional linguistics, recurrent linguistic formulas are multi-word units stored in our memory which fluent speakers rely on in communication (e.g., Biq 2004; Fillmore, Kay and O’Connor 1988; Goldberg 1995, 2006; Sinclair 1996; Stubbs 2001; Tao and McCarthy 2001; Thompson 2002). When antonyms co-occur in recurrent linguistic patterns, antonym associations might be enhanced (Charles and Miller 1989; Fellbaum 1995; Jones 2002; Justeson and Katz 1991). With the above implications, the present study aims to establish a collocational profile for each textual function of antonymy in Mandarin Chinese. Another research question that seems to have constructional implications is whether antonym pairs in Mandarin Chinese prefer a particular occurrence sequence in text (e.g., good and bad versus bad and good). If yes, what factors determine the preferred sequence? Is the preference over occurrence sequencing the same across all the antonym pairs? These language-internal phenomena might be conditioned by language-external factors, such as cognitive operations, text types, and cultural features. The present study attempts to address these issues. All the research questions posed in this thesis have their theoretical and empirical implications. They will be answered in the following chapters. The organization of the thesis is presented in the next section.. 1.3. Organization of the Thesis This thesis is organized as follows. Chapter 2 reviews two main types of. approaches to antonymy, i.e., the intuition-based approach and the data-based. 1. In general terms, the term “construction” (Fillmore, Kay and O’Connor 1988; Goldberg 1995, 2006) might be interchangeable with “lexical items” and “extended units of meaning” (Sinclair 1996, 1998). 4.

(14) approach. Chapter 3 introduces the corpus used for this study, the antonymous pairs selected for analysis, and the sampling method adopted to retrieve sentences for analysis. Chapter 4 explores how antonyms in Mandarin Chinese function in written text and the factors that affect the functional distribution of antonymy in Mandarin Chinese. Chapter 5 identifies constructions that are associated with each textual function of antonymy in Mandarin Chinese. Chapter 6 discusses factors that determine antonym occurrence sequences in Mandarin Chinese. Chapter 7 offers the conclusion and some suggestions for further research.. 5.

(15) Chapter 2 Literature Review. The present study aims to explore how antonym co-occurrences in Mandarin Chinese function in text. In this chapter, major studies of antonymy are reviewed. Generally speaking, there are two types of approaches to antonymy, i.e., the intuition-based approach and the data-based approach. Section 2.1 reviews the intuition-based approach, which focuses on traditional categories of antonymy. Section 2.2, which consists of two subsections, reviews the data-based approach to antonymy. Section 2.2.1 introduces two competing hypotheses on how two words are associated with each other as antonyms: the Substitutability Hypothesis and the Co-occurrence Hypothesis. The latter reflects how antonymy is considered from a data-based perspective. Section 2.2.2 presents Jones’ (2002) functional framework of antonym co-occurrences, which is also the framework the present study adopts. Finally, Section 2.3 is the summary.. 2.1. Intuition-based Approaches to Antonymy Generally speaking, an antonym is defined as “a word which is opposite in. meaning to another word” (Richards, Platt and Weber 1985:14). Another definition is that antonymy “refers collectively to all types of semantic oppositeness” (Crystal 1985:18). In the following subsections, four major traditional categories of semantic oppositeness, i.e., complementary antonymy (Section 2.1.1), gradable antonymy (Section 2.1.2), relational antonymy (Section 2.1.3), and directional antonymy (Section 2.1.4) (Cruse 2001; Saeed 1997), are reviewed. This classification is regarded as intuition-based because it depends mainly on linguists’ intuitive judgments. 6.

(16) 2.1.1. Complementary Antonymy Of all the types of semantic oppositeness, complementary antonymy might be. conceptually the simplest (Cruse 1986:198). This class is also referred to as “binary taxonomy” (Leech 1974:106). By definition, complementary antonyms “divide some conceptual domain into two mutually exclusive compartments, so that what does not fall into one of the compartments must necessarily fall into the other” (Cruse 1986:198). 2 For example, one cannot be alive and dead at the same time, so alive and dead form a complementary antonymous pair. Due to the dichotomous nature of complementary antonymy, the following sentences featuring complementary antonyms are semantically anomalous:. (2). ? John is neither alive nor dead. (taken from Leech 1974:106). (3). ? The statement that John has blue eyes is neither true nor false. (taken from Cruse 1986:199). From a componential perspective, complementary antonyms can be represented by semantic features, as the following examples illustrate (Leech 1974:96):. (4). man:. [＋human]. [＋adult]. [＋male]. woman:. [＋human]. [＋adult]. [－male]. boy:. [＋human]. [－adult]. [＋male]. 2. In fact, an absolute boundary between complementary antonyms does not seem to exist, as the following example shows: (i). In a technical sense he was alive, but for all practical purposes he was dead (taken from Leech 1974:106). Still, complementary antonymy is a verbal strategy that helps human beings organize the world (Leech 1974). 7.

(17) girl:. [＋human]. [－adult]. [－male]. Cruse (1986:201-202) proposes a framework of verbal complementary antonymy, as summarized in Table 1. Note that verbal antonymous triplets are formed for each category, with the first verb denoting a given condition and the complementary antonymous pair denoting the consequences.. Table 1. Cruse’s (1986) framework of verbal complementary antonymy. Categories. Definitions. Antonymous Triplets. Interactive. The verb expressing the precondition for complementarity denotes an action which has as its goal the elicitation of the response denoted by its interactive opposite, which, in turn, is one of the terms of the complementary pair.. request: grant/refuse learn: remember/forget. Satisfactive. The verb denotes an attempt to do something, and the complementary antonymous pair denotes the performance.. try: succeed/fail compete: win/lose. Counteractive The verb denotes an aggressive action, attack: defend/submit and the complementary antonymous shoot: save/let pair represents an active response and a passive response to the action.. Jones (2002) criticizes that Cruse’s (1986) categories of verbal complementary antonyms are highly restrictive so that it is difficult to find antonymous pairs that fit the above framework and at the same time are recognized as good opposites by most native speakers.. 8.

(18) 2.1.2. Gradable Antonymy According to Cruse (1986), gradable antonyms have the following properties: (i) They are fully gradable (most are adjectives; a few are verbs). (ii) Members of a pair denote degrees of some variable property such as length, speed, weight, accuracy, etc. (iii) When more strongly intensified, the members of a pair move, as it were, in opposite directions along the scale representing degrees of the relevant variable property. Thus, very heavy and very light, for instance, are more widely separated on the scale of weight than fairly heavy and fairly light. (iv) The terms of a pair do not strictly bisect a domain: there is a range of values of the variable property, lying between those covered by the opposed terms, which cannot be properly referred to by either term. As a result, a statement containing one member of an antonym pair stands in a relation of contrariety with the parallel statement containing the other term. Thus, It’s long and It’s short are contrary, not contradictory, statements. (taken from Cruse 1986:204) This class is also referred to as “polar opposition” (Leech 1974:108), for a. polarity can accommodate a region on a given scale which does not belong to either end of the scale, as the following sentence exemplifies:. (5). This man is neither rich nor poor. (taken from Leech 1974:108). In (5), there is a middle ground on the scale of wealth that fits the polarity phrase neither rich nor poor. Cruse (1986:206-214) proposes a framework of adjective gradable antonyms, as summarized in Table 2. This framework is based on impartiality and committedness. 9.

(19) For example, for a given entity to be hotter than something else, the entity is supposed to be hot. However, for a given entity to be longer than something else, the entity is not necessarily long. According to Cruse (1986), words such as hotter are regarded as committed, while words such as longer are regarded as impartial.. Table 2. Cruse’s (1986) classification of adjective gradable antonymy. Categories. Definitions. Examples. Polar Antonyms. Both members of a gradable antonymous pair are impartial in the comparative.. long/short, high/low, wide/narrow, thick/thin. Overlapping Antonyms. One member of a gradable antonymous pair is committed in the comparative, and the other is impartial.. good/bad, kind/cruel, clean/dirty, polite/rude. Equipollent Antonyms. Both members of a gradable antonymous pair are committed in the comparative.. hot/cold, happy/sad, proud of/ashamed of. Now consider the following sentence:. (6). I’m sad, but I’m happier than yesterday.. Because the sentence in (6) is considered semantically unacceptable (Cruse 1986:207), the pair happy/sad is classified as an equipollent antonymous pair. However, Cruse’s (1986) analysis is not compatible with every native speaker’s intuitive judgment. The main problem with Cruse’s (1986) classification of adjective gradable antonyms lies in its “subjective nature of the criteria” (Jones 2002:16).. 10.

(20) 2.1.3. Relational Antonymy In general terms, relational antonyms describe the same relationship, with the. noun phrase arguments interchanged (Cruse 1986:231), 3 as shown in the following pair of sentences:. (7). a. A is above B. b. B is below A.. Here are several typical examples of relational antonyms: parent/offspring, ancestor/descendant, in front of/behind, buy/sell, and husband/wife. This category is also referred to as “converses” (Leech 1974:111). According to how many arguments are involved, relational antonyms might denote 2-place, 3-place, or 4-place relationships (Cruse 2001:255). Consider the following examples:. (8). a. John gave a book to Jenny. b. Jenny received a book from John.. (9). a. John bought the book for one thousand dollars from Jenny. b. Jenny sold the book to John for one thousand dollars.. In (8a) and (8b), three arguments are involved, i.e., John, a book, and Jenny, so the pair give/receive is regarded as 3-place. In (9a) and (9b), four arguments are involved,. 3. To prevent symmetrical synonyms, such as resemble/be similar to, from being classified as relational antonyms, Cruse (1986:233, emphasis added) poses additional conditions to the permutation test: X and Y are converses if any sentence in which X expresses a relation between two noun phrases N1 and N2 is logically equivalent to the sentence which results when (i) N1 and N2 are interchanged and (ii) X is replaced by Y, but is not equivalent to the sentences which result when operations (i) and (ii) are carried out singly. 11.

(21) i.e., John, a book, one thousand dollars, and Jenny, so the pair buy/sell is regarded as 4-place. Cruse (1986:233-234) further divides relational antonymous pairs into two subcategories. Consider the following sentences:. (10) a. John gave a book to Jenny. b. Jenny received a book from John. (11) a. Monday precedes Tuesday. b. Tuesday follows Monday.. In (10), the interchanged arguments are John and Jenny. The word John serves as the subject in the double object construction, while the word Jenny serves as the indirect object in the double object construction, occupying a syntactically more peripheral position than John. On the other hand, both Monday and Tuesday in (11) occupy a syntactically essential position. According to Cruse (1986:233-234), relational antonymous pairs in which the interchanged arguments both occupy essential positions (e.g., precede/follow) are considered “direct”, while those in which an essential argument and a peripheral argument are interchanged are considered “indirect”.. 2.1.4. Directional Antonymy There are two main categories of directional antonymy. The first is “static”. directional antonyms (Cruse 2001:254), which denote a relationship between two spatial terms on the ends of a given axis. Here are a few typical examples of static directional antonyms: up/down, north/south, east/west, and forwards/backwards. The other type of directional antonymy is “dynamic” directional antonyms 12.

(22) (Cruse 2001:254). This class is also referred to as “reversives” (Cruse 1986:226). There are two subtypes of reversives, i.e., restitutives and independent reversives. In the case of restitutives, one of the members “necessarily denotes the restitution of a former state” (Cruse 1986:228). Take the pair remove/replace, for example. One can remove something to a certain place where it has never been to, but one cannot replace something in a place where it has never been to. That is, remove is an independent member, while replace is logically dependent on a former state. Here are some more examples of restitutives: stop/resume, kill/resurrect, and damage/repair. As for independent reversives, there is “no necessity for the final state of either verb to be a recurrence of a former state” (Cruse 1986:228). Take the pair lengthen/shorten, for example. A line which lengthens does not need to return to a former state of being long, after it is shortened. Here are two more examples of independent reversives: increase/decrease and widen/narrow.. 2.1.5. Interim Summary In Section 2.1, four traditional categories of antonymy are reviewed. A. complementary antonymous pair (e.g., alive/dead) is a pair of opposites which are mutually exclusive categories. A gradable antonymous pair (e.g., easy/difficult) is a pair of opposites that represent the opposite ends of a semantic scale. A relational antonymous pair (e.g., teacher/student) is a pair of opposites that represent a symmetrical relationship. A directional antonymous pair might be either static or dynamic. The former (e.g., up/down) represents a symmetrical spatial relationship, while the latter (e.g., lengthen/shorten) represents a change of state. The above traditional classification of antonymy is useful in making logical distinctions between antonymous pairs. However, the intuition-based approach to antonymy has a few limitations. First, 13.

(23) some words are semantically opposite, but they are not regarded as antonyms by most native speakers. For example, although affluent and broke operate along the same semantic scale, i.e., the scale of wealth, most native speakers of English do not regard them as antonyms (Jones 2002:10). Second, while big and large are synonymous and little and small are synonymous, most native speakers of English might intuitively feel that big/little and large/small achieve greater antonymous status than big/small and large/little (Charles and Miller 1989:373-374). As the above examples show, the traditional definition of antonymy, i.e., semantic oppositeness, is apparently insufficient.. 2.2. Data-based Approaches to Antonymy As Muehleisen (1997:113) has suggested, typical antonyms “not only occupy. opposite ends of a shared semantic dimension, but also describe the same kind of things” (cited in Jones 2002:11). That is, the antonymous status of a pair depends on its members’ individual collocational profiles. If two words (e.g., affluent and broke) have very different collocational profiles, they might not be considered to be an antonymous pair. This additional criterion, i.e., collocational profiles, complements traditional definitions of antonymy. Since a pair’s antonymous status has a lot to do with its individual members’ collocational patterns, more and more linguistic studies exploring antonymy take a data-based perspective. In the following subsections, major studies adopting data-based approaches to antonymy are reviewed.. 2.2.1. Antonymy: Substitutability versus Co-occurrence One of the central issues in previous studies of antonymy is how two words are. associated with each other as antonyms. There are at least two hypotheses concerning this issue, i.e., the Substitutability Hypothesis and the Co-occurrence Hypothesis. 14.

(24) The Substitutability Hypothesis (Deese 1962, 1964, 1965) states that two words are learned and associated with each other as antonyms because they are syntactically and/or semantically interchangeable in most contexts. In other words, paradigmatic substitutability contributes to the antonymic association between a pair of words. To test the hypothesis, Charles and Miller (1989) design three context-sorting tasks. Here are two experimental sentences in Charles and Miller (1989):. (12) Third, there were those notably Patrice Lumumba who favored a Unified Congo with a very _____ central government. (13) It is no harder to raise big, healthy, blooming plants than _____, sickly, little things; in fact it is easier. (taken from Charles and Miller 1989:362). The participants are asked to sort the sentences, according to which antonym, say strong or weak for the above sentences, can be used to fill in the blanks. Based on the participants’ intuition, each sentence in the Charles and Miller’s (1989) experiment is biased for a given antonym rather than neutral between both members. That is, antonyms are seldom interchangeable in the context. This finding does not lend support to the Substitutability Hypothesis. On the other hand, the Co-occurrence Hypothesis suggests that two words are associated with each other as antonyms because they co-occur at a rate higher than expected by chance (Charles and Miller 1989:360). The hypothesis has been empirically supported by many corpus-based studies on antonymy. For example, Justeson and Katz (1991) select the Brown corpus, a database “containing 1,000,000 words of English text balanced across 15 general categories, divided into 500 text extracts of about 2,000 words each” (Justeson and Katz 1991:3), finding that 15.

(25) antonyms co-occur far more frequently than expected by chance. Their findings are further confirmed in the APHB corpus, “a much larger but grammatically untagged corpus of 25,000,000 words, obtained from the American Publishing House for the Blind and archived at IBM’s Watson Research Center” (Justeson and Katz 1991:4). Following Justeson and Katz (1991), Fellbaum (1995) makes the Co-occurrence Hypothesis even more general, suggesting that “regardless of their syntactic category, words expressing semantically opposed concepts tend to be used together in the same sentences” (Fellbaum 1995:289). 4 In effect, the Substitutability Hypothesis and the Co-occurrence Hypothesis are complementary, as Stubbs (2001:39, italics added) suggests: Antonymy has traditionally been regarded as a paradigmatic opposition permanently available in the lexicon of the language. However, it is better seen in addition as a syntagmatic relation, which is realized in co-text. When it comes to syntagmatic relations, a key concept is collocation, which is defined as “a lexical relation between two or more words which have a tendency to co-occur within a few words of each other in running text” (Stubbs 2001:24). From a collocational perspective, it is of great interest how antonym co-occurrences function in text and interact with their co-text. These issues need to be addressed by analyzing a large amount of production data, and the findings might complement traditional approaches to antonymy.. 2.2.2. Textual Functions of Antonym Co-occurrences As for how antonyms function in text, a preliminary study from a data-based. perspective is Fellbaum (1995) (Section 2.2.2.1). By analyzing data from a 280-million-word corpus, Jones (2002) revisits the textual functions of antonym 4. The corpus which Fellbaum (1995) uses is the Brown Corpus. 16.

(26) co-occurrences in English and proposes a functional framework of antonymy (Section 2.2.2.2). This framework has also been used to account for different types of English data (Section 2.2.2.3).. 2.2.2.1. Fellbaum (1995). Fellbaum (1995) is a preliminary study on how antonyms function in text. Fellbaum. (1995:294-297). suggests. some. possible. reasons. for. antonym. co-occurrences. First, there are a large number of lexico-syntactic frames with two variables, which might allow antonyms. Typical lexico-syntactic frames available for antonym co-occurrences include X or Y, X and Y alike, and between X and Y:. (14) A matter of life or death (15) Rich and poor alike (16) I asked about the battle between life and death in his plays. (taken from Fellbaum 1995:295). Second, antonyms are sometimes redundantly used for emphasis, as illustrated by the following sentence:. (17) Just to remind the communists that the bombs dropped on Japan were to end a war not start one. (taken from Fellbaum 1995:296). In (17), since end and start are obviously antonymous, the phrase end a war implies the negation of start one (= a war). The redundancy is to put emphasis on the real intention of dropping two nuclear bombs on Japan, i.e. to end a war. 17.

(27) Third, antonyms are sometimes used for rhetoric reasons, such as humor and irony, as shown in the following:. (18) How easily he could hate the lovable Irish. (taken from Fellbaum 1995:296). The sentence in (18) has a tinge of irony due to the incompatibility between the antonymous words hate and loveable. Fourth, antonyms sometimes co-occur to refer to a change of state, as the following example illustrates:. (19) After being closed for seven months, the Garden of the Gods Club will have its gala summer opening Saturday, June 3. (taken from Fellbaum 1995:297). In the above instance, a transition is involved in that the Garden of the Gods Club was closed and then reopened. Fellbaum (1995) has touched on the textual functions of antonym co-occurrences in English. However, as Fellbaum (1995:295) acknowledges, the above textual functions of antonymy are merely crude suggestions for further research. In addition, the corpus selected for use in Fellbaum (1995), i.e., the Brown Corpus, is relatively small in size. Comprehensive studies on how antonyms function in text do not appear until Jones (2002).. 2.2.2.2. Jones (2002). Jones (2002) takes a data-based perspective to explore how antonyms function in 18.

(28) text. The corpus selected for use in Jones (2002) consists of 280 million words of newspaper data from The Independent, including all the articles printed from October 1, 1988 to December 31, 1996. Jones’ (2002) first step towards establishing a fresh list of antonymous pairs is to consider Deese’s (1964) list, and the following criteria are taken into consideration (Jones 2002:30). First, more non-gradable antonymous pairs should be added. Second, a representative set of antonymous pairs should not be restricted. to. adjectives.. Third,. morphologically. related. antonyms. (e.g.,. correct/incorrect) should be included. Finally, 56 antonymous pairs, which are presented in Table 3 (Jones 2002:33), are selected for analysis.. 19.

(29) Table 3. Antonymous pairs selected in Jones (2002). new/old private/public bad/good hate/love poor/rich active/passive failure/success female/male directly/indirectly heavy/light old/young feminine/masculine fail/succeed false/true right/wrong cold/hot lose/win alive/dead badly/well begin/end large/small agree/disagree optimistic/pessimistic private/public happy/sad guilt/innocence rightly/wrongly advantage/disadvantage. 254 134 117 104 102 96 88 87 79 77 69 68 63 62 60 59 58 54 53 51 50 49 47 47 45 44 44 36. long/short fact/fiction strength/weakness confirm/deny gay/straight hard/soft high/low illegal/legal married/unmarried dry/wet explicitly/implicitly. 36 36 35 34 33 32 32 31 31 31 30. attack/defend discourage/encourage fast/slow quickly/slowly permanent/temporary difficult/easy major/minor officially/unofficially rural/urban boom/recession optimism/pessimism punishment/reward correct/incorrect drunk/sober peace/war disprove/prove dishonest/honest. 30 28 28 28 28 27 27 25 24 24 21 19 18 18 15 14 12. 20.

(30) The next step is to sample sentences from the corpus. In total, 3,000 sentences are needed for analysis. Jones (2002) suggests that a representative set of sentences for analysis should not be biased towards adjective gradable antonymous pairs, which have drawn a lot of attention in the literature. Therefore, the following guidelines are observed (Jones 2002:32): (i) no more than 60 percent of database sentences should feature adjectival antonyms; at least 10 percent should feature antonymous nouns, at least 10 percent should feature antonymous verbs, and at least 10 percent should feature antonymous adverbs; (ii) at least 250 database sentences should feature non-gradable antonyms; (iii) at least 250 database sentences should feature morphologically related antonyms; (iv) where possible, while still meeting the criteria above, sample size should reflect co-occurrence frequency. With the above guidelines taken into consideration, the numbers of sentences retrieved for each antonymous pair are manually determined. The figures in Table 3 indicate the numbers of sentences randomly sampled for each antonymous pair. In fact, there are only 2,844 sentences recorded in Table 3. The shortfall, i.e., 156 sentences, is left for sentences featuring un- antonyms (e.g., known/unknown), for the morpheme of opposition un- is fairly productive in English. In total, 3,000 sentences, with 156 sentences featuring un- antonyms included, are randomly sampled for analysis. Note that when a sentence is somehow not interpretable, or antonyms in a sentence are not in any contextual relationship, another sentence is randomly sampled to replace the sentence in question. Jones (2002) identifies eight textual functions of antonym co-occurrences in English, as summarized in Table 4. Note that the percentage in parentheses indicates 21.

(31) the proportion a given textual function accounts for in the database. As Table 4 shows, Ancillary Antonymy (38.7%) and Coordinated Antonymy (38.4%) dominate in English, accounting for 77.1% of the database sentences.. Table 4. Functions of antonymy identified in Jones (2002). Ancillary Antonymy (38.7%) The antonymous pair signals a contrastive interpretation of a non-contrastive pair. (e.g., Stamps are popular, but collecting is unpopular.) Coordinated Antonymy (38.4%) Antonyms are used in a unified, coordinated way to show inclusiveness or exhaustiveness. (e.g., He had a particular gift for metaphor, mixed and unmixed, which he deployed to greater advantage.) Typical Constructions: both…and…, either…or…, neither…nor…, …and…alike, …as well as…, whether…or… Comparative Antonymy (6.8%) Antonyms are used to establish some form of comparison. (e.g., A study of institutionalized mental patients at the turn of the century found that castrated men lived longer than uncastrated ones.) Typical Constructions: more…than…, … be more [adjective] than…, …rather than… Distinguished Antonymy (5.4%) Antonyms are used in frames like the gap/difference between X and Y. (e.g., But the gap between rich and poor has widened and there’s a dwindling middle class.) Typical Constructions: the difference between…and…, separating…and…, a gap between…and… Transitional Antonymy (3.0%) The context signals a movement or change from one antonymous state to the other. (e.g., The younger generation is leading the way from passive to active forms of entertainment and information-gathering.) Typical Constructions: from…to…, turning…to…, …give way to…. 22.

(32) Table 4. Functions of antonymy identified in Jones (2002) (cont.). Negated Antonymy (2.1%) An antonym is negated, modifying the other member in the pair. (e.g., If you look at employment, not unemployment, that too fell in the first quarter of the year.) Typical Constructions: …not…, …instead of…, …as opposed to… Extreme Antonymy (1.3%) Extremity-signaling adverbs are used to modify either antonym. (e.g., The meticulous lawnsman will aerate it every fortnight throughout the year except when the soil is too wet or too dry.) Typical Constructions: the very…and the very…, either too…or too…, deeply…and deeply… Idiomatic Antonymy (0.8%) An antonymous pair is embedded in an idiomatic frame. (e.g., But old dog that he is, Clough has little interest in new tricks.). In addition, it has been found in Jones (2002) that each textual function of antonymy has its typical lexico-syntactic frames. Typical lexico-syntactic frames associated with each textual function of antonymy are recorded in Table 4. According to Jones (2002:154), it is important to identify typical lexico-syntactic frames in which antonyms co-occur, for words which often occupy the same lexico-syntactic frames but have not been regarded as antonyms might evolve into new antonyms. Take …instead of…, for example. While this lexico-syntactic frame is typical of Negated Antonymy, it is open to words which are not prototypical antonyms, signaling a contrastive interpretation. By placing a given word in the contrast-signaling construction …instead of… and searching for words which co-occur with the target word in the above lexico-syntactic frame, it is possible to establish the antonymous profile of the target word and identify new opposites which arise from constant textual contrasts. 23.

(33) Furthermore, Jones (2002:122-130) has found that antonyms in English tend to favor a particular sequence in text. A root word tends to precede its morphologically related antonym (e.g., advantage/disadvantage); a positive word tends to precede its negative antonymous partner (e.g., good/bad); a word of greater magnitude tends to precede its antonymous partner (e.g., large/small); a word which tends to precede its antonym in the real world tends to precede its antonym in text (e.g., begin/end); a phonologically shorter word tends to precede its phonologically longer antonymous partner (e.g., easy/difficult). These tendencies might remain dormant when the reverse sequence is contextually more significant, as illustrated by the following two examples:. (20) In my seventeen years with the group, I served indirectly and directly, no fewer than twenty chairmen and managing directors. (taken from Jones 2002:134) (21) The stricter approach coincides with growing demands for curbs from Congress, where bills have been tabled aimed at reducing both illegal and legal immigrants. (taken from Jones 2002:134). The tendency that a root word usually precedes its morphologically related antonym is not observed in the above sentences. In (20), it is likely that the twenty chairmen were mostly served indirectly, so the antonym indirectly is more significant and takes precedence in this context. Similarly, illegal is more readily accessible than legal in (21) because the phrase illegal immigrants occurs more frequently than its antonymous counterpart legal immigrants in text.. 24.

(34) 2.2.2.3. Follow-up Studies of Jones’ (2002) Functional Framework of Antonymy. Jones’ (2002) functional framework of antonymy is derived from the examination of written English. This framework is adopted to examine how antonyms function in other text types such as child-produced speech, child-directed speech (Jones and Murphy 2005), and spoken language (Jones 2006). A comparison of antonym co-occurrences across the above four domains is established (Jones 2007). The corpus used to investigate antonym co-occurrences in child-produced and child-directed speech is the Child Language Data Exchange System (CHILDS), 5 from which the spoken texts of three children observed longitudinally are collected, with 281,736 words of child-produced speech and 247,076 words of child-directed speech. The corpus selected for examining antonym co-occurrences in spoken language consists of 855 spoken texts from the National British Corpus (BNC), 6 containing 9,980,799 words in total. Since adults’ and children’ languages are greatly different, two separate lists of antonymous pairs are established. For the two lists of antonymous pairs, see Table 2 in Jones (2007). The findings in Jones and Murphy (2005) and Jones (2006, 2007) suggest that Jones’ (2002) functional framework of antonymy is highly applicable to examining antonym co-occurrences in different domains. In addition, Ancillary Antonymy and Coordinated Antonymy dominate in all of the four domains. A longitudinal analysis of child-produced speech shows that while the proportional use of Ancillary Antonymy and Coordinated Antonymy remains constant during childhood, changes in the use of Negated Antonymy and Transitional. 5. The Child Language Data Exchange System (CHILDS) is a database for studying interactions between adults and children. The database is available on-line at http://childs.psy.cmu.edu. 6 The British National Corpus (BNC) consists of 89.74 million words of written language and 10.37 million words of spoken language. Not all the spoken texts in the British National Corpus are included in Jones (2006). The excluded spoken texts (55 in total) are “either demographically-governed conversations recorded by respondents under the age of 16 or else data collected in an educational context” (Jones 2006:1106-1107). 25.

(35) Antonymy are apparent (Jones and Murphy 2005). On the one hand, the frequency of Negated Antonymy decreases with age, perhaps because children, as their language competence improves, become more capable of using antonyms in complicated contexts so that they use antonyms less often in such an explicitly contrastive context as Negated Antonymy (Jones and Murphy 2005:411). On the other hand, the frequency of Transitional Antonymy increases with age, perhaps because children are more aware of changes around them and become able to describe them as they grow older (Jones and Murphy 2005:412). Furthermore, it is found that antonym co-occurrences in child-produced speech do not completely reflect those in child-directed speech (Jones and Murphy 2005:413). In the comparison between antonym co-occurrences in written language and spoken language, the antonyms investigated are found to co-occur 1.36 times more often in written language than in spoken language (Jones 2006:191). The two modes are different in at least two other aspects of antonym use (Jones 2006:198): (i) While spoken language favors pairs that are casual and conversational in nature and often concerned with judgments and feelings (e.g., badly/well, hate/love), written language favors pairs that are more conventional in nature and more likely to be technical (e.g., directly/indirectly, confirm/deny). (ii) When two antonymous pairs operate along the same semantic scale, one tends to occur more commonly in spoken language (e.g., right/wrong, big/little) and the other more commonly in written language (e.g., correct/incorrect, large/small). To summarize, the three follow-up studies on the textual functions of antonym co-occurrences, i.e., Jones and Murphy (2005) and Jones (2006, 2007), all suggest that the functional framework proposed in Jones (2002) has a strong descriptive power in the discussion of antonym co-occurrences across various domains. The 26.

(36) results show that antonym co-occurrences in the four domains (i.e., adult-produced writing, adult-produced speech, child-produced speech, and child-directed speech) are generally similar in how they function in language use, with Ancillary Antonymy and Coordinated Antonymy being the most dominant functions of antonymy in English. Nevertheless, minor differences do exist, and they might be attributed to different communicative needs in different language modes.. 2.2.3. Interim Summary Fellbaum (1995), from a corpus-based perspective, sketches how antonyms. function in text. Structurally speaking, there are many lexico-syntactic frames allowing antonym co-occurrences. Fellbaum (1995) suggests that antonyms co-occur for rhetorical reasons, such as emphasis, humor, and irony. Moreover, antonyms sometimes co-occur to signal a change of state. Jones (2002) is a larger-scale study of antonym co-occurrences in text. Based on a 280-million-word corpus of newspaper data, Jones (2002) proposes eight textual functions of antonym co-occurrences. The most dominant functions are Coordinated Antonymy and Ancillary Antonymy. Jones’ (2002) functional framework of antonymy in English has been used to describe how antonyms function in child-produced speech, child-directed speech (Jones and Murphy 2005), and spoken language (Jones 2006). The framework is found to be useful in accounting for the textual functions of antonym co-occurrences across various domains. Jones’ (2002) functional framework of antonymy should be further tested in other languages, say, Mandarin Chinese. A cross-linguistic comparison is further called for to see whether typological properties in different languages affect how antonyms function in text. It is hoped that multi-corpus and cross-linguistic 27.

(37) approaches to antonymy can enhance our understanding of how antonym co-occurrences function in language use.. 2.3. Summary In Chapter 2, two types of approaches to antonymy are reviewed, i.e., the. intuition-based approach (Section 2.1) and the data-based approach (Section 2.2). Generally speaking, the intuition-based approach to antonymy focuses on logical distinctions between members in an antonymous pair, ignoring how antonyms actually function in text. Another problem with the intuition-based approach is that the definition of antonymy, i.e., semantic oppositeness, cannot explain why words that are opposite on a given scale (e.g., affluent/broke) do not always form an antonymous pair. Despite the inadequacies of the intuition-based approach to antonymy, the traditional categories of oppositeness identified with such an approach are still deemed valid and useful for the analysis conducted in the present study. The rapid development of corpus linguistics has made it possible to explore the textual functions of antonymy from a data-based perspective. The findings might complement the intuition-based approach to antonymy. Jones (2002) analyzes a considerable amount of corpus data to investigate how antonyms in English function in text, and proposes eight textual functions of antonym co-occurrences. Jones’ (2002) functional framework of antonymy has been successfully applied to various language modes (Jones 2006, 2007; Jones and Murphy 2005), such as child-produced speech, child-directed speech, and spoken language. So far, Jones’ (2002) functional framework of antonymy has not been adopted to investigate antonym co-occurrences in languages other than English. The present study is intended to examine whether corpus data in Mandarin Chinese can lend cross-linguistic support to Jones’ (2002) functional framework of antonymy. The 28.

(38) following chapter will introduce the corpus, the antonymous pairs, and the sampling method for this study.. 29.

(39) Chapter 3 Methodology. Chapter 2 reviews two types of approaches to antonymy: the intuition-based approach (e.g., Cruse 1986; Leech 1974; Lyons 1977; Saeed 1997) and the data-based approach (e.g., Charles and Miller 1989; Fellbaum 1995; Jones 2002, 2006, 2007; Muehleisen 1997). While the intuition-based approach is useful in making logical distinctions between different types of antonymous pairs, it is inadequate when the question comes to how antonyms function in text, for they seldom consider real production data (Jones 2002:25). To complement previous accounts of antonymy, Jones (2002) resorts to corpus data and proposes eight textual functions of antonymy in English. The present study adopts the data-based approach to examine to what extent Jones’ (2002) functional framework of antonymy is applicable to Mandarin Chinese. This chapter introduces the corpus (Section 3.1), the antonymous pairs (Section 3.2), and the sampling method (Section 3.3) for this study.. 3.1. Corpus A well-known corpus in Mandarin Chinese is the Academia Sinica Balanced. Corpus of Modern Chinese, Sinica Corpus hereafter. 7 In the Sinica Corpus, texts have been collected and classified according to the following five variables: genre, style, mode, topic, and source. The corpus is segmented and tagged. The current version was completed in 1997, containing roughly 5 million words. For the purpose of studying antonym co-occurrences in Mandarin Chinese, the. 7. The Academia Sinica Balanced Corpus of Modern Chinese is open to the research community online. It is available at http://www.sinica.edu.tw/SinicaCorpus/. 30.

(40) Sinica Corpus is relatively small in size and does not offer adequate data for analysis. For instance, there are only 21 lines featuring the pair ying/shu ‘to win/to lose’ in the database. Any patterns and generalizations based on such a small sampling pool can be flawed with biases due to insufficient data. Another Chinese corpus currently available for research is the Chinese Word Sketch Engine. 8 This corpus takes as its input the Chinese Gigaword Corpus (the second edition), which is an archive of newswire text data in Mandarin Chinese. The second edition of the Chinese Gigaword Corpus includes three international sources of Chinese newswire: Central News Agency of Taiwan (1991-December 2004), Xinhua News Agency of Beijing (1990-December 2004), and Zaobao Newspaper of Singapore (October 2000-September 2003). The Chinese Word Sketch Engine is selected for use in this study. The reasons are as follows. First, texts in the Chinese Word Sketch Engine are natural and authentic. Second, the size of the Chinese Word Sketch Engine is large enough for the present study. For example, in the Chinese Word Sketch Engine there are 899 lines featuring the antonymous pair ying/shu, roughly 40 times more than in the Sinica Corpus. Third, the present study is intended to examine the applicability of Jones’ (2002) functional framework of antonymy to Mandarin Chinese, and the Chinese Word Sketch Engine makes the present study comparable with Jones (2002), for both studies are based on newswire texts. Last but not least, an essential feature of corpus linguistics is that data for analysis should be publicly accessible (Stubbs 2001:123), and the Chinese Word Sketch Engine is open to the research community.. 8. The Chinese Word Sketch Engine is officially open to the research community after the analysis of this study is almost done. The current version of the Chinese Word Sketch Engine is available at http://wordsketch.ling.sinica.edu.tw/. An earlier version, on which the present study is based, is still accessible at http://corpora.fi.muni.cz/chinese_all/. 31.

(41) 3.2. Antonymous Pairs for Analysis Since there is still no consensus on an adequate definition of antonymy, “any list. of antonyms is immediately and inherently flawed” (Jones 2002:29). In other words, different definitions of antonymy may identify different sets of antonymous pairs, so it is difficult to select a representative set of antonymous pairs. The best one could do is to examine a wide range of pairs which most native speakers recognize as good opposites. In this section, how a list of antonymous pairs is established in the present study is explained. Jones’ (2002) list (see Table 3 in Section 2.2.2.2) includes gradable (e.g., old/young) and non-gradable (e.g., right/wrong) pairs, lexical (e.g., right/wrong) and morphological (e.g., correct/incorrect) pairs, and pairs of different grammatical categories (e.g., good/bad, adjectival; well/badly, adverbial; confirm/deny, verbal; advantage/disadvantage, nominal). The antonymous pairs in this list are translated into Mandarin Chinese as the first step to establish our Chinese list. When these antonyms are translated into Mandarin Chinese, several problems emerge. First, it is impossible to find an equivalent in Mandarin Chinese for certain pairs, such as explicitly and implicitly. Second, even when it is possible to obtain an equivalent in Mandarin Chinese, the Chinese counterparts do not always form good opposites. For example, while encourage and discourage are good opposites in English, their Chinese equivalents, i.e., guli and quanzu respectively, are hardly considered anonymous by native speakers. Another example is the pair fact/fiction. The Chinese equivalents, i.e., shishi/xiaoshuo, are not considered antonymous by most native speakers, either. If fiction is translated as xugou, the pair shishi/xugou might be even lower on the scale of antonymity, for shishi is a noun while xugou is usually regarded as a verb. Third, there might be no lexical equivalent in Mandarin Chinese for the negative member of a morphological antonymous pair in English. For 32.

(42) example, the word dishonest might be naturally translated into bu-chengshi ‘not-honest’, which is a phrase rather than a word in Mandarin Chinese. In view of the above problems, some antonymous pairs in Jones’ (2002) list, such as explicitly/implicitly, encourage/discourage, fact/fiction, and honest/dishonest, are excluded in the present study. In selecting a representative set of antonymous pairs, another consideration is co-occurrence frequency. Some antonyms in Jones’ (2002) list can be easily translated, but their equivalents in Mandarin Chinese do not co-occur very often. For example, in the Chinese Word Sketch Engine there are only 11 hits featuring the antonymous pair gan/shi, i.e., ‘dry/wet’. In the present study, antonyms that co-occur fewer than 150 times in the Chinese Word Sketch Engine, such as gan/shi ‘dry/wet’ and kuaile/beishang ‘happy/sad’, are not considered. With the above limitations, only seven antonymous pairs are selected from Jones’ (2002) list, i.e., ai/hen ‘to love/to hate’, lao/nianqing ‘old/young’, yi/nan ‘easy/difficult’, chengren/fouren ‘to confirm/to deny’, dui/cuo ‘right/wrong’, nan/nu ‘male/female’, and ying/shu ‘to win/to lose’. The seven antonymous pairs in Mandarin Chinese are semantically close to their English equivalents in Jones’ (2002) list, and they do achieve great antonymous status. In addition, the seven antonymous pairs all pass the co-occurrence frequency threshold, i.e., each pair has at least 150 hits in the Chinese Word Sketch Engine. Seven pairs are apparently not adequate for a study that aims to test whether Jones’ (2002) framework works for Chinese data. More pairs are selected, therefore, through considering other linguistic variables and structural properties specific to Mandarin Chinese. First, Jones’ (2002) classification of antonymy is divided into gradable antonymous pairs and non-gradable ones, but the present study adopts a fine-grained 33.

(43) categorization of antonymy, including gradable, complementary, relational, and directional antonyms (Cruse 2001; Saeed 1997). Among the seven pairs selected from Jones’ (2002) list, three (i.e., ai/hen, lao/nianqing, yi/nan) are gradable, three (i.e., chengren/fouren, dui/cuo, nan/nu) are complementary, and one (i.e., ying/shu) is relational. 9 To balance the distribution across different categories of oppositeness, four pairs are added in the present list. For the complementary category, the pair jia/zhen ‘fake/real’ is added. For the relational category, mai/mai ‘to buy/to sell’ is added. For the directional category, jin/tui ‘to move forwards/to move backwards’ and jin/chu ‘to enter/to go out’ are selected. In the Chinese Word Sketch Engine, the members of the above four pairs are found to co-occur frequently. Second, the present study aims to cover different grammatical categories of antonymous pairs, just as Jones (2002) does. So far, there are six verbal pairs and three adjectival pairs. Only nan/nu is nominal. To balance the distribution across grammatical categories, the following three pairs are added to the nominal category: qian/hou ‘front; before/back; after’, 10 fumu/haizi ‘parents/child’, and laoshi/xuesheng ‘teacher/student’. In the Chinese Word Sketch Engine, the members of the above three pairs are found to co-occur frequently. In addition, the three pairs are recognized to be good opposites for most native speakers of Mandarin Chinese. Third, some antonymous pairs in Mandarin Chinese have a counterpart in Classical Chinese that is still in use. We want to investigate whether an antonymous pair’s extent of modernization has an effect on its functional distribution. Therefore,. 9. In fact, some other antonymous pairs in Jones’ (2002) list, such as punishment/reward, are highly translatable into Mandarin Chinese, and their Chinese equivalents do achieve great antonymous status and pass the co-occurrence frequency threshold. However, they are not considered in the present study because they cannot fit the present four-category framework of oppositeness well. 10 In Mandarin Chinese, qian ‘front; before’ and hou ‘back; after’ are locative markers (Chao 1968; Li and Thompson 1981). While locative markers in Mandarin Chinese are not regarded as nouns, they function as nouns in Classical Chinese and still retain a number of nominal properties in Modern Chinese (Huang, Li and Li, to appear). Therefore, the present study includes qian and hou even though they are not prototypical nouns. 34.

(44) to contrast with the pair lao/nianqing, lao/shao is selected as its Classical Chinese counterpart. In like fashion, shi/fei is chosen as the Classical Chinese counterpart to contrast with dui/cuo. Both lao/shao and shi/fei pass the co-occurrence frequency threshold in the present study. It is noted that shi and fei are versatile in terms of their grammatical properties. They can function as adjectives, nouns, and even copulas. Despite its multiple grammatical functions, the pair is chosen to contrast with its vernacular counterpart, i.e., dui/cuo, to focus on the functional variation arising from the classical/vernacular parallel. Therefore, in the present study, the grammatical functions of shi and fei are not distinguished. Fourth, Jones’ (2002) list includes a number of morphological antonymous pairs, such as correct/incorrect and honest/dishonest. The present study also takes an antonymous pair’s morphological structure into consideration. As widely documented in the literature, a morpho-lexical fact related to the Classical Chinese issue is the monomorphemic versus double-morphemic, and the resulting monosyllabic versus disyllabic, parallels for some lexical items in Modern Chinese (Chao 1968; Li and Thompson 1981). For example, while the two monomorphemic and monosyllabic words dui ‘right’ and cuo ‘wrong’ form an antonymous pair, the two double-morphemic and disyllabic words zhengque ‘right’ and cuowu ‘wrong’ form another pair with the same semantic oppositeness. The co-existing, synonymous sets (of antonymous pairs) may differ from each other in their distribution and function across text types. To verify whether the single versus double morphemic/syllabic structure has any impact. on the textual. functions. of. antonyms,. three. double-morphemic, disyllabic antonymous pairs are selected to contrast with their monomorphemic, monosyllabic counterparts that are already in the present list, i.e., rongyi/kunnan ‘easy/difficult’ versus yi/nan, qianjin/houtui ‘to move forwards/to move backwards’ versus jin/tui, and zhengque/cuowu ‘right/wrong’ versus dui/cuo. 35.

(45) Fifth, in Mandarin Chinese, there exist some antonyms that share a morpheme with each other. For example, maifang ‘buyer’ and maifang ‘seller’, a nominal antonymous pair, share the second morpheme fang. This pair is selected not only for this purpose but also because its functional distribution can be further compared with that of its verbal counterpart mai/mai ‘to buy/to sell’. The other case in point is the pair fangdong/fangke ‘landlord/tenant’, which is selected to contrast with the pair maifang/maifang, since the members of the former share the first morpheme while the members of the latter share the second morpheme. Sixth, the antonymous pair nan/nu ‘male/female’ is particularly noteworthy, for it has at least four synonymous relatives, including nanren/nuren, nansheng/nusheng, nanxing/nuxing, and nanzi/nuzi. To investigate whether synonymous pairs show any variation as they are used in text, the present study includes all of the above five antonymous pairs related to gender distinction. With the above linguistic variables and Chinese structural features taken into consideration, 18 antonymous pairs are added. Along with the seven pairs from Jones’ (2002) list, 25 antonymous pairs in total are selected for analysis. Table 5 presents all the antonymous pairs selected in the present study.. 36.

(46) Table 5. Antonymous pairs selected for analysis*. Gradable. Complementary. Verbs. Adjectives. ai ‘to love’/ hen ‘to hate’ (247). lao ‘old’/ nianqing ‘young’ (256) lao ‘old’/ shao ‘young’ (651) yi ‘easy’/ nan ‘difficult’ (458) chengren ‘to confirm’/ fouren ‘to deny’ (387). Nouns. dui ‘right’/ cuo ‘wrong’ (470) shi ‘right’/ fei ‘wrong’ (9406) jia ‘fake’/ zhen ‘real’ (1238). rongyi ‘easy’/ kunnan ‘difficult’ (212) zhengque ‘right’/ cuowu ‘wrong’ (164). * The figure following each antonymous pair indicates the number of lines featuring the pair.. 37. nan ‘male’/ nu ‘female’ (14869). nanren ‘male’/ nuren ‘female’ (872) nansheng/nusheng (1603) nanxing/nuxing (5579) nanzi/nuzi (8273).

(47) Table 5. Antonymous pairs selected for analysis (continued). Verbs Relational. Adjectives. Nouns. ying ‘to win’/ shu ‘to lose’ (899) mai ‘to buy’/ mai ‘to sell’ (1062). fangdong ‘landlord’/ fangke ‘tenant’ (127) maifang ‘seller’/ maifang ‘buyer’ (462) fumu ‘parents’/ haizi ‘child’ (1514) laoshi ‘teacher’/ xuesheng ‘student’ (4027). Directional. jin ‘to move forwards’/ tui ‘to move backwords’ (917) jin ‘to enter’/ chu ‘to go out’ (1273). qianjin ‘to move forwards’/ houtui ‘to move backwards’ (121). qian ‘front; before’/ hou ‘back; after’ (13532). 38.

(48) Regarding the frequency of these 25 pairs, as Table 5 shows, there are two cases not meeting our frequency criterion of having 150 hits or more in the database. There are only 121 lines featuring the pair qianjin/houtui and 127 lines featuring the pair fangdong/fangke. The former is selected to contrast with its monomorphemic, monosyllabic counterpart jin/tui. In the latter pair, the members share the first morpheme, i.e., fang. This pair is selected to contrast with maifang/maifang ‘seller/buyer’, in which the members share the second morpheme. To select a representative set of antonymous pairs for analysis, the present study adopts the English list (Jones 2002) as a point of departure and then considers various linguistic variables and structural features specific to Mandarin Chinese, such as categorization of oppositeness, grammatical category, modern versus classical contrast, morpho-syllabic structure, and synonymous alternates. In total, 25 antonymous pairs in Mandarin Chinese are selected for analysis. The following step is to sample sentences from the database.. 3.3. Sampling Method After the corpus and the set of antonymous pairs for analysis are selected, the. next step is to sample sentences from the Chinese Word Sketch Engine. In the present study, a sentence is loosely regarded as a concordance line, for antonyms may co-occur in simple sentences and complex sentences. As Table 5 shows, the antonymous pairs selected for analysis are greatly different in how frequently their members co-occur. However, for each antonymous pair, an equal number of database sentences, i.e., 150, is randomly sampled with the rationale that this 150 frequency threshold can exclude antonyms which do not. 39.

(49) co-occur often and thus cannot be studied from a corpus-based perspective. 11 In the process of sampling sentences from the Chinese Word Sketch Engine, we use the set default values for search, such as window size (i.e., 21 words) and search span (i.e., -5~5). Not all the sentences retrieved from the database meet our needs. Some examples are of accidental antonym co-occurrences.. (22) 考生有老有少，年輕的體能較佔優勢， Kaosheng you lao you shao, nianqing de tineng jiao zhan youshi. examinee have old have young, young. NOMINALIZER. physical strength. more take advantage ‘There were old and young examinees, and younger ones were at an advantage in their physical strength.’ (23) 他在比賽中戰術正確，不犯低級錯誤， Ta zai bisai zhong zhanshu zhengque, bu fan diji cuowu. he in game middle strategy right, not commit low-level mistake ‘He adopted right strategies in the game and did not make stupid mistakes.’. Note that the sentence in (22) is retrieved for the pair lao/nianqing ‘old/young’, not for the pair lao/shao ‘old/young’. In (22), although lao and nianqing co-occur in close lexical proximity, they seem to have little to do with each other. In effect, lao is coordinated with shao in the lexico-syntactic frame you…you… ‘have…have…’. Such sentences as the above are not included for analysis. As for the sentence in (23), 11. The only two exceptions to this 150 threshold are qianjin/houtui ‘to move forwards/to move backwards’ (121 lines) and fangdong/fangke ‘landlord/tenant’ (127 lines). The reasons for keeping the two pairs in the list are given in Section 3.2. 40.