應用字義相似度之測量方法於分析母語一致性對第二語言搭配詞的影響

全文

(1)國立臺灣師範大學英語學系博. 士. 論. 文. Doctoral Dissertation Department of English National Taiwan Normal University. 應用字義相似度之測量方法於分析母語一致性對第二語言搭配詞的影響. Application of Semantic Similarity Measures to Analyses of Congruency Effects on L2 Collocations. 指導教授：林至誠博士 Advisor: Dr. Chih-cheng Lin 研究生：李金瑛 Graduate: Ching-ying Lee. 中華民國一百零二年六月 June, 2013. 1.

(2) 2.

(3)

(4)

(5) 中文摘要應用搭配詞的能力是第二語言學習者發展近似外國人母語能力的重要顯著特徵。近年來，有關第二語言學習者所犯的搭配詞錯誤，和其受母語影響的問題均已受到關注。然而，母語與第二語言搭配詞之間錯綜複雜的關係仍是模糊不清，學習者對第二語言搭配詞的認知過程也欠缺充分的研究調查。因此，本研究目標為探討學習者在受母語一致性的影響之下，其第二語言搭配詞的表現為何，並分析學習者在組成搭配詞時有關詞彙選擇的思考過程。研究問題包括如何應用字義相似度計算方法來量測搭配詞與母語的一致性、搭配詞與母語一致性和學習者表現之間的關係、以及搭配詞促發歷程的跨語言型態。研究設計涵蓋質性與量化的四個分項實驗。研究工具包括字義相似度量測方法、搭配詞測驗、問卷、及有聲思考方法。受測者為345位台灣的醫學大學與護理專校學生，依其英語程度分為高、中、低能力區分的三組。第一個分項實驗是驗證字義相似度測量方法的使用性。第二個分項實驗為以字義相似度計算方法，對搭配詞測驗題組的母語一致性進行客觀評量和分類。第三個實驗為找出兩個獨立變數，(即搭配詞與母語一致性和學習者語言能力程度)，以及一個應變數，(即學習者的搭配詞表現)，這兩者之間的相互關係。第四個分項實驗則使用問卷調查及有聲思考方法，檢視受測者對搭配詞促發歷程的看法和思考過程。本研究的主要結果包括提出應用字義相似度計算方法對搭配詞與母語一致性的評估、對搭配詞與母語一致性和學習者語言程度兩項因素的分析結果、描述搭配詞學習型態的與母語一致性的表現模式、第二語言搭配詞促發歷程的影響因素及跨語言轉換型態。第一，本研究驗證了兩個字義相似度量測方法，能夠計算兩個字詞之間字義相似度的數值，作為可靠有效的指標，進而能作為搭配詞與母語一致性的客觀系統化的評估方法。第二，學習者的語言能力程度對其整體搭配詞表現有顯著影響，搭配詞與母語一致性於三組學習者的測驗中，呈現不同的結果，顯示其對學習過程的影響要大於對學習結果的影響。第三，本研究提出一個. i.

(6) 四區塊與母語一致性的學習表現模組，以描述搭配詞與母語一致性及學習者表現之間的互動型態。這個模組對第二語言搭配詞的階段性學習與搭配詞的認知錯誤，提供了一個整體的樣貌說明。第四，在搭配詞組成過程中,其主要影響因素包括中介搭配詞促發歷程中的字義配合、跨語言的字彙關聯性、以及母語和第二外語字彙的類比等因素。本研究結果可應用於搭配詞教學與自主學習方面。首先，搭配詞中母語一致性對學習者表現的模組，以及搭配詞使用與母語一致性之間更明確的關係，都有助於發展成可操作的知識。這些知識可延伸作為搭配詞有效學習策略，進而提升學習者的語言能力表現。再者，若能以引導式的學習過程，針對每一個搭配詞與母語一致性的影響，提供清楚的說明，這將能增進搭配詞的學習。要達成這個目標，接續發展出搭配詞診斷與諮詢服務的線上系統，將能幫助學習者找出搭配詞錯誤的原因、辨認所需的跨語言的字義關係、並建立適切的字義轉換或轉變。此系統可幫助教師了解受母語一致性影響的搭配詞學習方式。第三，編撰的常用搭配詞表，再附加情境文字與說明，可作為學習搭配詞之參考，且幫助學習者辨識適切的搭配字和從母語轉換後的錯誤用字之間的差異。瞭解搭配詞與母語一致的特性，可降低學習者的學習困難，並可幫助教師找出容易犯錯的搭配詞。此外，搭配詞在語言使用上的重要性，以及搭配詞字義關係的知識，均已受到重視。本研究主要成果為深入瞭解搭配詞與母語一致性對學習者表現的影響，並對第二語言搭配詞學習與教學提供明確指引。. 關鍵字：搭配詞、與母語一致性分類、與母語一致性的影響、跨語言促發歷程、字義學. ii.

(7) ABSTRACT Collocation use has been a significant and recommendable attribute to possess in developing native-like linguistic competence. Over decades, research attention has been attracted to the identification of second language (L2) learners’ miscollocation and the influence of first language (L1). However, the intricate yet influential interaction between L1 and L2 collocations remains ambiguous. Learners’ cognitive processing on L2 collocations also lacks adequate investigation. The current study aims to examine congruency effects on L2 learner’s collocation production and investigate their cognitive processing of collocate choice in collocation composition. Research questions concerning the applicability of lexical semantic similarity measures to congruency evaluation, the interaction of congruency effects and collocation performance, as well as cross-linguistic patterns of collocational priming, are formulated and addressed. The overall research design consisted of four sub-studies involving both quantitative and qualitative analyses. The research instruments included two semantic similarity measures, a set of collocation test, a set of questionnaire, and the think-aloud protocol. Three participant groups, formed by a total of 345 college students from a medical university and a nursing junior college in Taiwan, were recruited and classified into high, mid, and low proficiency levels. The first sub-study involved verifying the applicability of semantic similarity measures and using two semantic similarity measures to the experimental set of collocation so as to objectively evaluate the properties of congruency. In the second sub-study, the correlation between two independent variables, collocation congruency and proficiency level, and one dependent variable, L2 collocation performance, was investigated. In the third sub-study, a questionnaire survey was conducted to collect. iii.

(8) learners’ conceptions about collocational congruency processing. After the questionnaire survey, the think-aloud protocol was compiled to get insights into the learners’ dynamic thinking process within certain context of engagement. The major findings of the study included application of similarity measures for congruency evaluation, analysis of performance variation with congruency and learners’ proficiency level, a congruency-performance model for characterizing collocation learning, influential factors of L2 collocation processing gauged by questionnaire survey, and the cross-linguistic processing patterns extracted from think-aloud data. First, two selected semantic similarity measures were able to provide reliable and effective indication of semantic similarity between word pairs, and thus, can be considered as objective and systematic procedures for congruency evaluation. Second, learners’ language proficiency level had a considerable effect on their overall collocation performance. Learners’ performance on L2 collocations was somewhat divergent on congruency categories, which seemed to suggest that congruency affected more on L2 learners’ collocation learning process than on the learning outcome. Third, the congruency-performance four-quadrant model was derived to characterize the interaction patterns of collocation congruency and learners’ performance. This model provided an overall picture of how L2 collocations were learnt in progressive stages and why errors were made at the cognitive level. Fourth, the primary factors that influenced congruency processing included inter-lingual priming to make semantic matching, cross-linguistic word relations, and analogy drawing between L1 and L2. The implications of the research findings are discussed in both contexts of pedagogical and autonomy learning applications. First, the conceptualization of the congruency-performance model advances the field of L2 collocation learning by providing an operative view for clarifying the confusing relation between congruency iv.

(9) and collocation use. The knowledge can facilitate the development of effective learning strategies for learners’ L2 proficiency. Second, L2 learners may benefit greatly from a guided learning process in which both congruency effect and L1 influence are elucidated for individual collocation items. To this end, an online system can be developed to provide both diagnostic and consulting services so that learners can identify causes of collocation errors, recognize the required cross-linguistic semantic relations, and establish the felicitous lexical-semantic transfer or conversion. Teachers can also use this system to understand the learning trajectories with respect to congruency effects. Third, another application is to compile a useful listing of frequent collocations with a high probability of misled transfer and false conversion. The complied list can be further extended to a collocation learning reference by adding contexts. L2 users can thus identify the discrepancy between felicitous L2 collocates and erroneous L1 transferred words. The congruency characterization is facilitative for L2 learners to reduce their difficulty in collocation learning and for L2 teachers to identify error-prone collocations. In addition, the important role of collocations in language use and the compelling need to gain the knowledge about the semantic relations of collocation between lexical components have been widely recognized. The present study mainly contributes to gaining a deeper understanding of congruency effects on learners’ collocation production and providing a useful guidance for L2 collocational learning and teaching.. Keywords: collocation, congruency classification, congruency effect, cross-linguistic priming, lexical semantics. v.

(10) ACKNOWLEDGEMENT I owe plentiful debts of gratitude to my teachers and friends for their support and encouragement in completion of the dissertation. My hearty thanks are due to my advisor, Dr. Chih-cheng Lin (林至誠) whose rigorous academic coaching and valuable advice has been a critical resource. His encouragement and remarks yielded in-depth perspectives of research questions and inspired me to improve the study. It has also been a privilege to have a group of esteemed committees (in the alphabetic sequence of the last names), Dr. Pei-chin Chang (張佩青), Dr. Wu-chang Chang (張武昌), Dr. Hao-jan Chen (陳浩然), and Dr. Yow-yu Lin (林祐瑜).This study have been improved a great deal because of their insightful suggestions and guidance. I have been fortunate to have a highly supportive group of participant students at Taipei Medical University and Kang Ning Junior College, whose cooperation, time and patience have been very important at all stages of completing the study. During the years of 2011 to 2013, they gave me the chance to pilot my experiments, administrate the questionnaire and collect think-aloud data. I am grateful for their dedication and feedback. Also, I owe a special personal debt to the colleagues who have helped me throughout my data collection at Taipei Medical University and Kang Ning Junior College, thanks to Dr. Tsai-lin Hsu, Dr. Chien-lung Chen and Tsao-tsao Chien. Without their generous help, I would have been through a much harder and less rewarding experience. I am deeply indebted to my family. They are my strongest backup year-round. I would like thank them for their understanding and everlasting support. In particular, when the critical recurrence of Lupus had struck me for several times, they took good care of me and gave me courage to face my illness. After two years of recovery, I vi.

(11) gradually regained my strength and faith in moving on the study. With their unconditional love and help, my study and work could thus go a little easier on me. I would like to dedicate this dissertation to my advior and my family.. vii.

(12) viii.

(13) TABLE OF CONTENTS. CHAPTER ONE. INTRODUCTION .................................................................... 1. 1.1 Background of the Study .............................................................................. 1 1.2 Purpose of the Study ..................................................................................... 7 1.3 Significance of the Study .............................................................................. 8 CHAPTER TWO 2.1. LITERATURE REVIEW ....................................................... 10. Definition and Theory of Collocation ......................................................... 10. 2.2 Property and Classification ......................................................................... 16 2.3 Collocation Competence ............................................................................. 21 2.3.1 Development of Collocation Competence ....................................... 22 2.3.2 Collocational Problems and Errors .................................................. 23 2.4 Collocation Acquisition .............................................................................. 27 2.4.1 L1 Influence on Collocation Processing .......................................... 29 2.4.2 L2 Collocation Teaching .................................................................. 35 2.5 Measurement of Semantic Relation ............................................................ 40 2.5.1 Semantic Relatedness and Semantic Similarity ............................... 40 2.5.2 Measures of Semantic Distance ....................................................... 42 2.5.3 Uses of Semantic Similarity Measures ............................................ 44 2.6 Concluding Remarks ................................................................................... 47 CHAPTER THREE. RESEARCH METDHO ..................................................... 49. 3.1 Research Design........................................................................................... 49 3.2 Participants .................................................................................................. 53 3.3 Instruments .................................................................................................. 55 3.4 Data Collection Procedure .......................................................................... 62 3.5 Data Analysis .............................................................................................. 66 CHAPTER FOUR. RESULT ANALYSIS AND DISCUSSION ......................... 70. 4.1 Verification on the Lexical Similarity Measures ........................................ 70 4.2 Collocational Congruency Classification by Semantic Similarity Values .. 73 4.2.1 Applying Sematic Similarity Values and Examining Congruency . 77 4.2.2 Objective Congruency Classification .............................................. 89 4.2.3 A Four-Level Congruency Model .................................................... 96 4.3 Effects of Proficiency Level and Congruency on Learners’ Performance 102 4.3.1 Performance Variation by Participants’ Proficiency Level ............ 102 4.3.2 Performance Variation by Collocation Congruency ...................... 110 ix.

(14) 4.4 A Congruency-Performance Four-Quadrant Model ................................. 124 4.5 Learner Responses to Collocation Congruency ........................................ 142 4.5.1 Reliability and Validity of CUQ .................................................... 143 4.5.2 Interpretive Results of Collocational Use Questionnaire Survey .. 144 4.6 Think-aloud Protocol for Observation of Collocational Processing ......... 149 4.6.1 Learners’ Conceptions of L2 Collocation Processing ................... 149 4.6.2 Factors of Processing Collocational Congruency .......................... 157 4.6.3 Processing Strategies of Collocational Congruency ...................... 164 CHAPTER FIVE. CONCLUSION ...................................................................... 172. 5.1 Major Research Findings .......................................................................... 173 5.1.1 Semantic Similarity Measures for Congruency Evaluation ........... 173 5.1.2 Effects of Proficiency Level and Congruency on Collocation Performance ............................................................................................... 174 5.1.3 The Congruency-Performance Four-Quadrant Model ................... 176 5.1.4 Learners’ Cognitive Factors from Questionnaire Survey .............. 177 5.1.5 Cross-Linguistic Processing Patterns from Think-Aloud Data ..... 178 5.2 Implications of the Study .......................................................................... 179 5.2.1 Pedagogical Applications ............................................................... 179 5.2.2 Autonomy Learning Applications .................................................. 183 5.3 Limitations of the Study............................................................................ 184 5.4 Suggestions for Future Studies ................................................................. 186 REFERENCES .......................................................................................................... 189 Appendix A ................................................................................................................ 206 Appendix B ................................................................................................................ 209. x.

(15) LIST OF FIGURES Figure 3.1.1 Overall Research Design ……………………………………………... Figure 3.1.2 Operational Research Framework …………………………….............. 50 51. Figure 3.3.1 Semantic Search: Word Sense Selection ………...………………….... 56. Figure 3.3.2 Online Use of Semantic Similarity Measures ………..……….............. 56. Figure 3.4.1 Data Collection Procedure of Semantic Similarity Values …..……….. 64. Figure 4.2.1 Ideal Semantic Similarity Distribution and Congruency Classification. 75. Figure 4.2.2 Expectation of Semantic Similarity Distribution ……………………... 76. Figure 4.2.3 Disagreement between Objective Similarity Evaluation and Subjective Human Classification …………………………………….. Figure 4.2.4 Semantic Similarity Distribution of VN Collocations in Single Word Sense …………………………………………………………………. Figure 4.2.5 Semantic Similarity Distribution of VN Collocations in All Word Senses ………………………………………………………………… Figure 4.2.6 Semantic Similarity Distribution of AdjN Collocations in Single Word Sense …………………………………………………………... Figure 4.2.7 Semantic Similarity Distribution of AdjN Collocations in All Word Senses …………………………………………………………………. 76 81 81 84 84. Figure 4.4.1 The Congruency-Performance Four-Quadrant Model ……………….. 125 Figure 4.4.2 The Congruency-Performance Modeling of L2 Verb Noun Collocations in the Contexts of High-Proficiency Learners and Single Word Sense …………………………………………………………... 128 Figure 4.4.3 The Congruency-Performance Modeling of L2 Verb Noun Collocations in the Contexts of High-Level Learners and All Word Senses ………………………………………………………………… 128 Figure 4.4.4 The Congruency-Performance Modeling of L2 Adjective Noun Collocations in the Contexts of High-Level Learners and Single Word Sense …………………………………………………………………. 129 Figure 4.4.5 The Congruency-Performance Modeling of L2 Adjective Noun Collocations in the Contexts of High-Level Learners and All Word Senses ………………………………………………………………… 129 Figure 4.4.6 The Congruency-Performance Modeling of L2 Verb-Noun Collocations in the Contexts of Mid-Level Learners and Single Word Sense ….……………………………………………………………… 130 Figure 4.4.7 The Congruency-Performance Modeling of L2 Verb-Noun Collocations in The Contexts of Mid-Level Learners and All Word Senses …….…………………………………………………………... 130 Figure 4.4.8 The Congruency-Performance Modeling of L2 Adjective Noun Collocations in the Contexts of Mid-Level Learners and Single Word Sense …………………………………………………………………. 131. xi.

(16) Fiure 4.4.9. Figure 4.4.10. Figure 4.4.11. Figure 4.4.12. Figure 4.4.13. Figure 4.4.14. The Congruency-Performance Modeling of L2 Adjective Noun Collocations in the Contexts of Mid-Level Learners and All Word Senses ………………………………………………………………… 131 The Congruency-Performance Modeling of L2 Verb Noun Collocations in the Contexts of Low-Level Learners and Single Word Sense …………………………………………………………………. 132 The Congruency-Performance Modeling of L2 Verb Noun Collocations in the Contexts of Low-Level Learners and All Word Senses ………………………………………………………………… 132 The Congruency-Performance Modeling of L2 Adjective Noun Collocations in the Contexts of Low-Level Learners and Single Word Sense …………………………………………………………………. 133 The Congruency-Performance Modeling of L2 Adjective Noun Collocations in the Contexts of Low-Level Learners and All Word Senses ………………………………………………………………… 133 Space of Learning Behaviors in the Congruency-Performance Model ………………………………………………………………… 141. xii.

(17) LIST OF TABLES Table 2.2.1 Table 2.2.2. Types of Lexical Collocations by Benson et al. (1997) ……………… Types of Grammatical Collocations by Benson et al. (1997)……......... 18 19. Table 2.4.1. Synopsis of Collocation Studies on Cross-linguistic Influence ……..... 32. Table 3.2.1. Profile of Participant Groups and English Proficiency ……….............. 54. Table 3.3.1. Category Design and Allocation of Collocation Items ……………….. 58. Table 3.3.2. The Experimental Set of Collocation Items in Four Categories …..….. 59. Table 3.5.1. Think-Aloud Transcription Conventions ……………………………... 69. Table 4.1.1. Word Pairs Test Set for Semantic Similarity Measures …………..….. Statistical Descriptions of Semantic Similarity Values on Semantic Evaluation Test ………………………………………………………. Semantic Similarity Values of Verb-Noun L2/L1 Collocates in Subjective Congruency Classification ……………………..………… Semantic Similarity Values of Adjective-Noun L2/L1 Collocates in Subjective Congruency Classification ………………...……………... Descriptive Summary of Computational Semantic Similarity in Verb Noun Collocations ........................................…………………………. 71. Table 4.1.2 Table 4.2.1 Table 4.2.2 Table 4.2.3 Table 4.2.4 Table 4.2.5 Table 4.2.6 Table 4.2.7 Table 4.2.8 Table 4.2.9 Table 4.2.10. Table 4.2.11 Table 4.2.12 Table 4.2.13. One-Way ANOVA on Computational Semantic Similarity between Subjective Congruency Categories in Verb Noun Collocations ….….. Descriptive Summary of the Computational Semantic Similarity in Adjective Noun Collocations …………………….…………………... One-Way ANOVA on Computational Semantic Similarity between Subjective Congruency Categories in Adjective Noun Collocations ... Objective Congruency Classification on Verb Noun Collocations in the Context of Learners’ Single Word Sense …………...……………. Objective Congruency Classification on Verb Noun Collocations in the Context of Learners’ All Word Senses …………………………… Descriptive Summary of the Vector Measure for Objective Congruency Classification of Verb Noun Collocations………….…… One-Way ANOVA on the Vectors Measure for Objective Congruency Classification of Verb Noun Collocations …………..………………... Objective Congruency Classification on Adjective Noun Collocations in the Context of Learners’ Single Word Sense …...…………………. Objective Congruency Classification on Adjective Noun Collocations in the Context of Learners’ All Word Senses ………………………… Descriptive Summary of the Vector Measure for Objective Congruency Classification of Adjective Noun Collocations………….. xiii. 72 79 82 85 86 87 88 90 91 92 92. 93 94 95.

(18) Table 4.2.14 One-Way ANOVA on the Vector Measure for Objective Congruency Classification of Adjective Noun Collocations ……………………… Table 4.2.15 Four-Level Congruency Classification of Verb Noun Collocations ….. 95 100. Table 4.2.16 Four-Level Congruency Classification of Adjective Noun Collocations ………………………………………………………….. 101 Table 4.3.1 Learners’ Collocation Performance in Test Category …………...…… 103 Table 4.3.2 Descriptive Summary of Collocation Test Scores by Proficiency Level …………………………………………………………………. 103 Table 4.3.3 ANOVA on Participant Groups’ Overall Collocation Test Results …... 104 Table 4.3.4 Table 4.3.5 Table 4.3.6 Table 4.3.7 Table 4.3.8 Table 4.3.9 Table 4.3.10 Table 4.3.11 Table 4.3.12 Table 4.3.13 Table 4.3.14 Table 4.3.15 Table 4.3.16 Table 4.3.17 Table 4.3.18 Table 4.3.19. A Post-hoc Test on Participant Groups’ Collocation Test Results …… 104 Descriptive Summary of Participant Groups’ Test Scores in Syntactic Collocation Types ……………………………………………………. 105 ANOVA on Participant Groups’ Performance in Syntactic Collocation Types ……………………………………………………. 106 A Post-hoc Test on Participant Groups’ Performance in Syntactic Collocation Types…………………………………………………….. 107 One-Way ANOVA between Syntactic Collocation Types in Proficiency Groups…………………………………………………… 108 Descriptive Summary of All-Participant Group’s Test Scores between Syntactic Collocation Types…………………………………..……… 109 One-Way ANOVA between Syntactic Collocation Types on All-Participant Group………………………………………………… 109 Participant Groups’ Performance Rate Variation in VN Collocation by Objective Congruency in Learners’ Single Word Sense ………….. 111 Participant Groups’ Performance Variation in VN Collocation by Objective Congruency in Learners’ All Word Senses ………………... 112 Participant Groups’ Performance Variation in AdjN Collocation by Objective Congruency in Learners’ Single Word Sense ……………... 114 Participant Groups’ Performance Variation in AdjN Collocation by Objective Congruency in Learners’ All Word Senses ……….……….. 115 Participant Groups’ Performance Variation in VN Collocation by Subjective Congruency ………………………………………………. 118 Participant Groups’ Performance Variation in AdjN Collocation by Subjective Congruency ………………………………………………. 119 Descriptive Summary of Participants’ Performance in Subjective Congruency of Verb Noun Collocations ……………….…………….. 120 Descriptive Summary of Participants’ Performance in Subjective Congruency of Adjective Noun Collocations ………………………... 120 One-Way ANOVA on Subjective Congruency in Verb Noun Collocations ………………………………………………………….. 121. xiv.

(19) Table 4.3.20 One-Way ANOVA on Subjective Congruency in Adjective Noun Collocations ………………………………………………………….. Table 4.4.1 Correlation between Logical and Observed Learning Trajectory Patterns of Verb Noun Collocations ………………………………….. Table 4.4.2 Correlation between Logical and Observed Learning Trajectory Patterns of Adjective Noun Collocations …………………………….. Table 4.5.1 Demographic Information of Participants …………………………… Table 4.5.2 Table 4.5.3. 123 135 138 142. Item-Total Reliability Statistics ……………………………………… 143 KMO and Bartlett's Test ……………………………………………... 144. Table 4.5.5. Questionnaire Response Statistics……………………………………. 145 Comparison of Participants’ Responses Among Groups …….………. 147. Table 4.6.1. Congruency Processing Strategies from Think-aloud Data ……….…. 171. Table 4.5.4. xv.

(20) xvi.

(21) CHAPTER ONE. INTRODUCTION. Many second language (L2) learners strive for the ultimate goal of achieving native-like proficiency adequately demonstrated by linguistic delicacy and naturalsounding fluency. One of the sharp distinctions between L2 learners and native speakers is the fewer collocation use by L2 learners (Howarth, 1996; Laufer & Waldman, 2011). Thus, collocation use becomes a significant attribute to possess in developing native-like linguistic competence. Over decades, research attention has been attracted to the identification of L2 learners’ miscollocation and the influence of first language (L1), and has made much progress. However, the intricate yet influential interaction between L1 and L2 collocations remains ambiguous. Learners’ cognitive processing on L2 collocations also lacks adequate investigation. The current study aims to examine congruency effects on L2 learner’s collocation performance and investigate their cognitive processing of collocate choice in collocation composition. In addition, the research explores the semantic factors and the cognitive patterns of collocational processing in the impact circle of congruency effects. 1.1 Background of the Study Much linguistic research asserts that fixed forms of expression in every language are stored in the memories of native speakers as whole chunks, not just single words. Therefore, L2 vocabulary learning should not concern merely acquisition of individual words. It is also imperative to learn multiword expressions. Of all multiword expressions, collocation is one of the intricate and versatile aspects. L2 learners need to understand why a word is commonly found in company of one or more others, and how a particular word combination is formed. Over the years, many collocation studies have been conducted with either thesaurus-based or corpus-based approaches. In general, the thesaurus-based approaches probe into lexical meanings 1.

(22) and semantic relations. As a recent trend, the corpus-based approaches investigate L2 collocation processing and miscollocations and greatly contribute to the inquiry of word co-occurrence and relatedness (Jian, Chang & Chang, 2004; Nesselhauf, 2005; Nelson, 2006; Chen, 2007; Lin, 2010; Liu, 2010a, 2010b). Both approaches have its unique strength and suited purposes. To gain insights into L2 learners’ collocation use and cognitive processing, the present study adopts the computational notion of lexical semantic similarity and explores congruency effects on collocation processing. It has been an intriguing issue in collocation research to understand how L2 learners determine one felicitous lexical meaning to fit in a collocation. For instance, words with synonymous, polysemyous or L1 equivalent meanings have various influences on collocation processing. The word “cut” has fifty-one meanings in Merriam-Webster Dictionary (Sokolowski et al., 2006). L2 learners need to choose a felicitous meaning for collocation use such as “cut a deal”. If they fail to retrieve a felicitous meaning from their lexical network, arcane or awkward communication may occur. The multifaceted nature of lexical meaning makes collocation processing and formation more intricate. A number of researchers in the discipline of lexical semantics have long been studying how lexical semantic and conceptual links form (Firth, 1957; Sinclair, 1991; Bahns & Eldaw, 1993; Louw, 1993; Koya, 2002; Laufer, 2003; Nesselhauf, 2005; Brashi, 2006; Siyanova, 2008). It is crucial to understand how L2 learners tackle collocational meanings and to find out the factors which influence their conceptual links with each lexical component. In addition, a cross-linguistic conceptual link between a collocate and an L1 counterpart play its part in collocational composition. For instance, substratum L1-L2 transfer occurs constantly in the linguistic processing. Another reason is that theoretical concepts of collocation congruency remain vague and lack explicit criteria for subjectively dichotomous congruency classification (Koya, 2005; Webb & Kagimoto, 2009; 2.

(23) Yamashita & Jiang, 2010; Wolter & Gyllstad, 2011). One of the fundamental research issues in the current study is to elaborate congruency so as to better characterize cross-linguistic semantic relations and to better reveal L2 collocation processing behaviors. To modify the properties of collocational congruency, the present study first put forth the operational definition of collocation which is adapted as a recurring combination of lexemes (e.g., commit suicide), forming a particular syntactic unit (e.g., verb–object noun). A collocation is formulated by two elements, a “base” that keeps its usual meaning as auto-semantic words (Sinclair, 1991; Hausmann, 1999) and the other is a “collocate” that relies on syn-semantic words and has a less transparent meaning. Both of the elements have a certain internal relationship, determined by the collocational direction and attraction of one element towards the other. This oriented relation is multi-dimensional for both collocation elements in four aspects, leading to a positioning of each element on the respective dimensions – lexical, semantic, statistical, and structural. In such broad dimensions, the criteria on selection of collocation candidates are essential. In terms of the binary or ternary classification system proposed by Howarth (1998) and Nesselhauf (2003), the current study uses three criteria to retrieve collocation candidates. The criteria include verb-noun and adjective-noun combinations, semi-restricted collocations, as well as binary type at the span of two words. Congruency of binary collocations is the prime target for research observation of L2 learners’ collocational processing. Since the collocational processing involves collocations in context, not just collocation per se, it is necessary to take contextual information into account. As Louw (1993) indicated, lexical meaning is the result of a diachronic process, whereby meaning has been transferred from one word to another, as being a ‘‘consistent aura of meaning with which a form is imbued by its 3.

(24) collocates’’ (p. 157). For example, a CNN journalist reported, “Secretary of State Hillary Rodham Clinton ‘categorically denied’ Wednesday any US involvement in the car bombing of an Iranian nuclear…” For L2 learners, it is a daunting task to explain the meaning of “categorically denied” without the surrounding cotext. In general, the cotext could convey important lexical information which helped infer the meaning as “downright denied” because the event did not occur in partial possibilities. If the journalist used “completely denied”, it was easier for non-native speakers of English to grasp the meaning, especially for Chinese L2 learners because it was congruent with Chinese meaning, “完全否認 (wuan chuan fou ren, completely denied) ”. As indicated by Yamashita and Jiang (2010), collocations are largely cross-linguistic. It is thus believed that L2 collocations with L1 counterparts are more comprehensible to L2 learners. Also, collocations with similar L1-L2 culture-specific concepts can easily be inferred. For example, many languages have the expressions similar to “strong head”, “fierce argument” and “heavy traffic” (Carter & McCarthy, 1988; Schmitt & Carter, 2004), as in Chinese “意志堅強 (yi zhi jian qiang)”, “激烈爭論 (ji lie zheng lun)” and “交通繁忙 (jiao tong fan mang)”. Many empirical studies have focused on investigation of the cross-linguistic problems in L2 collocations (Zhang, 1993; Koya, 2002; Wolter, 2006; Xiao & McEnery, 2006; Siyanova, 2008; Webb & Kagimoto, 2009). As Wolter (2006) pointed out, L2 collocation processing and learning are more susceptible to the cross-linguistic factors and the particular components of word combination. The factors can be facilitative, but sometimes inhibitory. For example, Yamashita and Jiang (2010) claimed that L2 learners can grasp the collocational meaning immediately in the case of congruent collocations. On the other hand, incongruent collocations are more difficult for them to understand and produce. The characteristics of collocations cause more complex learning problems, even for the advanced learners. 4.

(25) Moreover, the cross-linguistic nature and flexibility of the component words link to form collocations carry significant effects on collocation variation. Various languages may use identical, similar or even totally different components to indicate the target meaning. Recent collocation studies have explored whether cross-linguistic lexical activation is valid for composing L2 collocations (Lesniewska & Witalisz, 2007; Durrant, 2008; Yamashita, & Jiang, 2010; Wolter & Gyllstad, 2011). Difficulties of collocation learning caused by the cross-linguistic factors are also evident in some empirical studies albeit evaluation of the cross-linguistic influence is not the focal point of their research. In this regard, such discussion can be found in the collocational teaching studies (Kellerman & Smith, 1986; Kennedy, 1990; Farghal & Obiedat, 1995; Gitsaki, 1999; Hoey, 2000; McCarthy & O’Dell, 2005; Hsieh, 2007; Laufer & Waldman, 2011). From a broader perspective, some collocation research evidences indicated the relationship between collocational problems and the L1 influence in collocational processing (Lennon, 1996; Nesselhauf, 2003; Murao, 2004; Martelli, 2006; Shehata, 2008; Walker, 2011). Overall, collocational congruency involves L2 learners’ conceptual network and cross-linguistic lexical processing. The rationale of the study is based on Kroll and Stewart’s (1994) Revised Hierarchical Model (RHM) and Dijkstrat and Van Heuven’s (1998, 2002) Bilingual Interactive Activation Model (BIA). Yamashita and Jiang’s (2010) adapted Jiang’s (2000) notions and the three memory representations in RHM, i.e., (1) concepts, (2) L1 lexicon and (3) L2 lexicon. These three representations are connected in terms of two levels: conceptual and lexical levels. The former refers to the link between concepts and L1 or L2 lexicons, and the latter refers to the link between L1 and L2 lexicons. Kroll and Tokowicz (2001) indicated that it is relatively easy to find an L1 lexicon corresponding to an L2 lexicon because the L1 lexical network normally has a larger number of lexical items. Based on this model, 5.

(26) Yamashita and Jiang (2010) postulated when learners try to understand an L2 collocation, they first translate it into L1, and retrieve L1 lexical links from their L1 lexical network. They further exerted the strong link from the L1 lexicon to its concepts to grasp the meaning of the L2 collocation. The role of L1 word meaning in L2 collocational processing has been extrinsically discussed in a few empirical studies. (Talamas, Kroll & Dufour, 1999; Jiang, 2001; Nesselhauf, 2003; Sunderman & Kroll, 2006). Nonetheless, the concepts that congruent collocations are more readily comprehensible and accessible to L2 learners have not been verified until more recent studies provided evidences (Koya, 2005; Nesselhauf, 2005; Ueno, 2006; Yamashita & Jiang, 2010; Wolter & Gyllstad, 2011). Ueno (2006) first posed an important concept that L2 processing involves not only L1 translation equivalents of single word but L1 lexical networks which are activated when L2 learners recognize an L2 word. Later, Yamashita and Jiang (2010), in view of the cross-linguistic relationship, made a distinction between congruent collocations and incongruent collocations. Wolter and Gyllstad (2011) further claimed that L1 has an effect on the development of L2 collocation knowledge. Their finding suggests that L2 learners recognize L1-L2 collocations more effectively than L2-only collocations. However, these related studies need a systematic and objective measure to determine the congruence or incongruence of a collocation. The current study employs two lexical semantic measures to evaluate the semantic differences between a collocate and a transferred word with L1 sense. L1-L2 (congruence) and L2-only (incongruence) collocations carry different numeric values of semantic similarity of the two words. According to Yamashita and Jiang’s model of three memory representations (2010), it is hypothesized that if the two words (collocate and transferred word) corresponding in a congruent collocation have closer meanings, it indicates it is easier to transfer from L1 word sense to L2 lexicon and then acquire the 6.

(27) collocation meaning. 1.2 Purpose of the Study The purpose of the study is to leverage existing computational methods on semantic similarity/distance as an assessment tool to refine examination of L1-L2 collocation congruency. Following the measurement, the present study uses the results to identify the problems of L2 collocation incongruence. It is expected that the semantic measures can be used for succinct and precise evaluation of lexical semantic distinction between L2 collocates and transferred words with L1 equivalents. For instance, Chinese collocation of “冷笑話 (leng xiao hua, cold joke)”corresponds to English collocation of “dry joke”, and “熱愛 (re ai, zealous love)” is expressed correctly as “big love” in English. Hypothetically, “dull” is semantically closer to “dry” than “hot” to “big”. This suggests that “big love” is somewhat more difficult than “dry joke”. Another example is the English collocation of “stand trial”, referring to “接受審判 (jie shou shen pan, stand trial)”. The primary sense of “stand” for L2 learners of Chinese native-speaking is “站立” (zhan li, stand) or “承擔…狀況(cheng dan…zhuan kuang, take on) ”. The value of semantic similarity between “站立 (zhan li, stand) ” and “接受 (jie shou, accept)” may be larger than that of“承擔 (cheng dan, take on)” and “接受 (jie shou, accept)”. It indicates that the similar lexical meanings of the L2 collocate and the transferred word from L1 counterpart is more conducive for L2 learners when the value of semantic similarity is smaller. The problem of collocational processing can thus be simplified and straightforward when they retrieve the similar or identical meanings from their lexical network. To fill the research gap emerging from inadequate evaluation on collocational congruency and rough understanding of congruency classification, the research questions are designed as follows:. 7.

(28) 1. Can the measures of semantic similarity be useed to evaluate collocational congruency? Can collocational classifications be defined on the basis of semantic similarity? 2. Can semantic similarity between the collocate and the transferred word with L1 sense be used to evaluate collocational congruency? What can be drawn to theorize collocational congruency? 3. Does the congruency between the collocate and the transferred word influence L2 learners’ collocational production? Are there any significant differences in the collocational production of the three groups? What relations can be verified between collocational congruency and L2 learners’ collocational performance? 4. What conceptions do L2 learners have towards collocational congruency processing? What are the cross-linguistic factors of processing congruent and incongruent collocations? What strategies do L2 learners use for collocational congruency production? 1.3 Significance of the Study The current study makes a novel hypothesis that congruency of collocations has a positive correlation with the semantic similarity of the two words, a collocate and a transferred word from L1 counterpart. The measures of collocational congruency can lead to a more salient characterization and distinction on different levels of congruency. For instance, the collocation of “relax the rules” is probably deemed as an incongruent collocation if the word “relax” is translated into Chinese as “放鬆 (fang sung, relax)”. However, after measuring the semantic similarity of the two words, “relax” and “loosen”, it is proved that “relax the rule” is a congruent collocation corresponding to its Chinese meaning. Another similar example is to. 8.

(29) determine the congruent collocation, “dog one’s career”. If an advanced L2 learner who has fuller recognition of the lexical network, he or she will consider the meaning of “dog” as “entangle”. As a result, it is postulated that the fuller the lexical recognition of advanced L2 learners is, the more decisive it is as a factor for the distinction between congruence and incongruence. Contrarily, the limited lexical recognition of the lower-level learners impedes the judgment of collocational congruency. The measures of semantic similarity provide an objective way to tackle the congruent problem and subdue the human subjective judgment. It is also surmised that congruency can serve as an effective indicator for collocational learning difficulty. The congruency characterization is facilitative for L2 learners to reduce their difficulty in collocation learning and for L2 teachers to identify error-prone collocations. From a broader perspective, collocation knowledge has its niche in a continuum of second language learning and can improve linguistic precision and natural-sounding language use. For L2 advanced learners, grammar is not a primary impediment to attain a higher level, from the intermediate plateau to the advanced level, or from the advanced to near native-like level. Yet, felicitous collocation use is a challenge to them. It has been widely recognized the important role of collocations in language use and the compelling need to gain the knowledge about the semantic and syntagmatic relations of collocation between lexical components (Gitsaki, 1999). The present study mainly contributes to gaining a deeper understanding of congruency effects on learners’ collocation production and providing a useful guidance for L2 collocational learning and teaching.. 9.

(30) CHAPTER TWO. LITERATURE REVIEW. In this chapter, an up-to-date research review is introduced to substantiate the nature of collocation in the context of second language learning and collocation pedagogy. In the first section, collocation is defined and explicated in terms of three linguistic perspectives—lexical, semantic and cognitive approaches. The second section depicts collocation properties and types. The third section examines collocation competence. Factors of difficulty in collocation learning are elicited on the account of the cross-linguistic influence in the fourth section. The fifth section evaluates semantic similarity and its application to collocational congruency, followed by a concluding remark in the last section. 2.1 Definition and Theory of Collocation Collocation is generally conceived as “the habitual co-occurrence of individual lexical items,” and a collocation refers to “two words or more adjacently juxtaposed or in close proximity to each other in text.” (Crystal, 2003, p. 82) The term “collocation” has been used interchangeably in literature with such terms as “word associations” (Murphy, 1983), “formulaic speech” (Kennedy, 1990), “multiword expressions” (Ellis, 1996) and “prefabricated patterns” (Granger, 1998). A common theme of study is the frequent association of a few words that defies systematic applications of linguistic rules. For semantic composition, semantically compatible word association does not always guarantee acceptable collocation. For instance, “foul language” and “dirty words” are acceptable collocations in English but “dirty language” and “foul words” are not. In the syntagmatic structure of a colloquial association, “care a damn” is well-composed but “damn good” is not. These fascinated yet peculiar linguistic phenomena of collocation have intrigued much research interest. 10.

(31) Past research on collocation had suggested a wide array of views to expound the intriguing linguistic phenomena of collocation. Definitions and theories of collocation developed by the lexical, semantic, and cognitive approaches were reviewed, respectively and chronologically. Firth (1957) initiated the notion of “meaning by collocation” at the syntagmatic level (p. 194), implying that word meaning was derived from its co-occurrence with other words in running text. He further gave a definition of collocation as “the company that words keep” (p.197). Firth’s theory on meaning by collocation was an abstraction at the syntagmatic level and was not downright related to the conceptual approach of lexical meaning. Firth explained the meaning of an abstraction as “a schematic construct for application especially to typical repetitive events in the social process”. The continuity of socially linguistic events ensured that “a text is attested as common usage in which the occasional, individual and idiosyncratic features are not in the focus of attention” (Firth, 1968, p. 176). From a theoretical perspective, it suggested that meaning of the composite structure was not constructed in an additive way, “It is more appropriate to say that the component structures motivate aspects of the composite structure and that the degree of motivation is variable (though typically quite substantial)” (Langacker 1987, p. 292). Firth’s notion of meaning by collocation was actually a theory of compositionality dwelling on the polysemous nature of words and their syntagmatic relations. Many Neo-Firthian linguists adopted Firth’s concepts and took the stance on the lexical approach. As a prominent representation, Sinclair (1966) extended Firth's theory and described the notion of word collocability in an integrated lexical theory. He maintained that grammar and lexis were two “interpenetrating ways” (p. 411) to study linguistic forms and that lexis was not independent of grammar. Instead, they interacted in a more dynamic way. Sinclair (1966) further described the structural 11.

(32) components of a collocation to include node, span, collocate and environment set, as in the following excerpt. We may use the term node to refer to an item whose collocations we are studying, and we may then define a span as the number of lexical items on each side of a node that we consider relevant to that node. Items in the environment set by the span we will call collocates (Sinclair, 1966, p. 415).. The proposed structure of collocation provided an important basis for objective analysis on the lexical behaviors of collocations. Early studies explored the patterns of collocation use which were not obvious when researchers relied on introspection alone (Sinclair, 1966, 1987, 1991; Kjellmer, 1982). For instance, “set in” as an intransitive prepositional verb was found to be collocated with negative subjects as “rot” or “decay” (Sinclair, 1987). Contraily, Lewis (2000) advocated the lexical approach and applied statistical measures in lexical collocation studies. Based on statistical measures and computational techniques, more objective measures such as descriptive statistics, co-occurrence frequency, and Mutual-Information (MI) score, could be used in collocation researuch. This development led to a set of quantifiable research to explore lexical behaviors. For instance, collocations could then be seen as units composed of co-occurring words within a distance of mutual expectancy and a distinction was often made between frequently and infrequently co-occurring words. As for lexical collocation, Halliday and Hasan (1976) indicated that collocation featured with the cohesive effect of word pairs such as ‘king...crown’. They further construed that the word pair “depends not so much on any systematic semantic relationship as on their tendency to share the same lexical environment, to occur in a collocation with one another” (p. 286). Collocational cohesion could be considered as a semantic interlace providing texts with their texture- or non-structural cohesion. 12.

(33) This indicated a vivid, variable and sophisticated expression of lexical co-occurrence in a general semantic description and text cohesion. It has been widely recognized that Halliday and Hasan’s collocation theory could be used in textual analysis, but it was restricted in lexically predictable collocational chains beyond the sentence level. Their theory did not construe unpredictable or idiosyncratic co-occurrences of words which were not syntagmatically or semantically combined with each other. The primary problem in the lexical approach lay in “the circularity of the definition of the basic unit of description, the lexical item” (Sinclair, 1966, p. 412). In other words, a lexical item was semantically construed by its environment which in turn was construed by the item. The lexical approach to the collocation research was found less precise than grammatical ones with a more explicit framework. One advantage of the lexical approach was that it shed lights on lexis, and indicated grammatical analysis could not thoroughly account for lexical patterns and collocational idiosyncrasies. The semantic approach of collocation studies described each lexical item defined by semantic markers and its meaning. Lehrer (1974) believed that the semantic approach could account for why certain collocated words were found conventionally acceptable. Contrarily, Cruise (1986) studied syntagmatic meaning relations between lexical associations and revealed collocational restrictions were arbitrarily established. Nonetheless, lexical field and semantic prosody have the potential to account for the non-arbitrary side of collocations. Based on the semantic theory, Lyons (1977) posited the notion of lexical fields embedded in “the relations of sense holding between pairs of syntagmatically connected lexemes” (p. 261). Lyons’ lexical field theory depicted that vocabulary of a language was a set of lexemes with each lexeme being a member of simply one field. Louw (1993) posed that “consistent aura of meaning with which a form is imbued by 13.

(34) its collocates”. The term, “semantic prosody” was also used by some writers (Nelson, 2006; Sinclair, 1995, 2004a, 2004b; Stubbs, 2001, 2009) in a wider sense to describe the way in which a lexical item could develop one of a range of different prosodies. From Lyon’s lexical field to Sinclair and Louw’s semantic prosody, linguistic researchers have gradually taken a broader view on collocations. They considered vocabulary of a language as an open system, and lexemes pertained to different fields because of their different meanings. Lyons suggested that descriptive semantics could work along well without syntagmatic relations (Lyons, 1977). As such, Lyons noted that the paradigmatic relations of word meaning were equally important in the study of semantics and syntagmatic relations. The semantic properties determined what lexical items were collocated. However, the semantic properties could not completely explain why there were numerous idiosyncratic co-occurrences that were arbitrarily restricted. As in Halliday's example, there was nothing in the meaning of “tea” to explain why it collocated with “strong” rather than “powerful”. The combination of “strong tea” could be categorized as an idiom and it might thus be overlooked in the study of lexical semantics. The proponents of the cognitive approach defined collocation as habitual composition of lexical units of which all surface structures were products of underlying deeper patterns. From the cognitive view, language was generated at an abstract level by logical, semantic and syntactic entities whose task was to conform a pre-existing idea with a linguistic string. The idea was somehow transformed into an abstract and organized set of entities which were in turn translated into language. This cognitive view had been adopted by a number of studies in cognitive processing of meaning of the L2 intralexical links (MacWhinny, 1997; Lucas, 2000; Wolter, 2001, 2006; Nesselhauf, 2003; Yamashita & Jiang, 2010). Conceptual links of a collocation were conjectured to provide a tendency for an activated word to 14.

(35) accelerate subsequent recognition of a collocate. Wolter (2006) evaluated the influence of conceptual knowledge and network structures in L2 lexical acquisition. He elaborated that conceptual knowledge led to the assumptions that word combination and lexical items mutually interacted to graft elements of one word with another. A modification of meaning was required to form a concept underlying one or both of the lexemes. That is, lexical interaction resulted in conceptual modification. In some cases lexical interaction did not demand any such modification and the concepts conceivably remained separate entities. Wolter postulated that transfer from L1 to L2 should be fairly straightforward, but when the formation of links did not exist in L1, it needed restructuring of the L2 lexical network. In view of Firth’s concept, the multilevel ‘polysystemic’ framework seemed compatible with the non-reductive approach found in cognitive linguistics, especially the notions of semantic frames and of dynamic schematic networks. Between them, these notions should be able to provide a cognitively based framework for explanation of the collocation phenomenon. The main problem with the cognitive view was that even ‘free combination’ appeared to be conventional. In fact, it was debatable that ‘free combination’ could turn out to be a phraseological unit. In Firth’s example, why was “powerful tea” and “strong car” unacceptable, when both “powerful argument” and “strong argument” were acceptable? Was it possible to argue that the meaning of “powerful” was somehow more ‘restricted’ in combination with car than with “argument”? Such an explanation seemed too ad hoc to have any explanatory value. It was acceptable in common sense that words were given by the linguistic community and overall people generally pointed to elements of existing common experiences. These approaches to collocation studies focused on various aspects of the collocational phenomenon. The advocators of the lexical approach regarded lexical 15.

(36) analysis as independent from grammar and considered a word as an autonomous entity, choosing its own collocates which could be enumerated and classified in lexical sets. Those of the semantic approach tried to find semantic features according to the meaning of lexical units that yielded prediction of their partner words. The advocators of the semantic approach maintained that lexical co-occurrences were arbitrarily restricted, but they did not provide ample evidence and explanation. Those of the cognitive approach emphasized more on the cognitive process of collocation meaning, including collocational priming and cross-linguistic transfer, less on collocational patterns. Overall, the three approaches indicated that vocabulary knowledge should be considered as a mapping of word association. However, only the cognitive approach highlighted language creativity, which could not explain the cases of restricted collocations. 2.2 Property and Classification The distinction among collocation and other forms of word association depends on various properties of collocation. Palmer (1981) characterized collocations with collocational restrictedness and cohesion. Palmer noted that some restrictedness came from the meaning of lexical units, while some restrictedness came from the range (p. 79). That is, lexeme was used with a whole set of lexical units that shared common semantic features. This explained why “pretty man” was conventionally unacceptable because “pretty” was used to describe females. Other restrictions were collocational in the most rigorous sense, such as “addled” with “brains”. Cruise (1986) indicated that word associations involving arbitrary collocational restrictions were semantically more cohesive than those with generalizable selectional restrictions because the former were more predictable. This notion of predictability was comparable to the Firthian notion of ‘mutual expectancy’. It was predictable that the constituents of. 16.

(37) collocations were mutually selective. Bahn and Eldaw (1993) suggested the primary property of collocation was that its meaning reflected the meaning of the constituent parts. This property was referred as semantic transparency of collocations and was used to determine the classification of free combinations, semi-collocations, collocations, and idioms. For instance, the meaning of a collocation, such as “observe the holiday”, is more semantically transparent than that of an idiom, such as “kick the bucket”. Moreover, collocations are semantically and psychologically salient in contrast with free combinations. Fernando (1996) characterized collocations as multiword expressions with subjective criteria, such as collocation restriction, syntactic structure and semantic opacity. The classical model of collocation categorization involved class membership depended on criteria attributes. Sinclair (1966) examined the relationship of collocational constituents based on co-occurrence frequency in large quantities of text and classified collocations into two categories of casual collocation and significant collocation. Casual collocation was interpreted as lexical combination which had least predictive power over the node and formed accidentally, while significant collocation had a strong tendency to occur near the node. Lewis (1997) extensively categorized collocations into strong, weak, frequent, and infrequent collocations based on fixedness and restrictedness. This meant shifting the focus of attention from the categories to the phraseological models underlying the categorization. Benson et al. (1997) distinguished between grammatical collocations and lexical collocations. Lexical collocations were multi-word combinations of nouns, adjectives, verbs, and adverbs with specific structures such as verb-noun, adjective-noun, noun-verb, noun-noun, adverb-adjective, adverb-verb. Table 2.2.1 summarized the seven basic types of lexical collocations. In contrast, grammatical. 17.

(38) collocations consisted of a major headword, e.g., noun, adjective/ participle, and verbs, accompanied by a preposition or a grammatical construction, such as to + infinitive or that-clause. Grammatical collocations were further divided into eight basic types of collocations (see Table 2.2.2). For the purpose of providing a more complete view on multiword expression, Howarth (1996) proposed a continuum model of multiword expressions with properties from syntactic frozenness or invariability to increasing semantic opacity or particularity. The properties included semantic transparency, degree of restrictedness, compositionality, analyzability. Multiword expressions were then classified by the extent to which they formed increasingly fixed units or chunks. There were four typical categories of collocations along the continuum: (1) free combinations, whose meanings were thoroughly compositional (blow a trumpet); (2) restricted collocations, which were less compositional (blow a fuse); (3) figurative idioms (e.g., blow your own trumpet); and (4) pure idioms (e.g., blow the gaff). Similar continuum models had also been adopted by other researchers (Palmer, 1981; Cowie, 1981, 1998; Benson et al. 1986; Nattinger & DeCarrico, 1992). Table 2.2.1 Types of Lexical Collocations by Benson et al. (1997) Type. Collocation Pattern. Adapted Example. L1. verb - noun / pronoun. stand trials. L2. verb - propositional phrase. come to an conclusion. L3. adj.- noun / noun - noun. burning ambition. L4. noun - verb. bombs explode. L5. noun1 - of - noun2. a cluster of admirers. L6. adv.- adj. / adv.- participle. sound asleep. L7. verb - adv.. argue heatedly. aptitude test. categorically denied. 18.

(39) Table 2.2.2 Types of Grammatical Collocations by Benson et al. (1997) Type G1. Pattern. Adapted Example. noun - preposition. apathy towards. (except noun - of ) G2. noun - to infinitive. It was a must to complete it.. (5 high frequent patterns). He had the foresight to do it. He felt an urge to do it. They made efforts to complete it. He was a fool to do it.. G3. noun - that - clause. He drew a conclusion that the boy is not guilty. He took an oath that he’ll tell the truth.. G4. prep. - noun. on demand, by chance. G5. prep. - adj.. He was angry at me.. G6. adj. - to - infinitive. It is necessary to finish it.. G7. adj. - that - clause. She was glad that she’s done with that.. G8. verb patterns. All verb patterns. Cowie (1998) suggested the notion of “a scale of idiomaticity” to represent the phraseological continuum defined by the properties of semantic transparency, restrictedness, compositionality, analyzability. Accordingly, free collocations were characterized as fully compositional, transparent and literal, whereas at the other end of the continuum, pure idioms like “break the ice” or “shoot the breeze” were viewed as non-compositional, opaque and figurative. In contrast to pure idioms, figurative idioms like “give attitude” and “make a U-turn” were considered as transparent since they could be interpreted by literal meaning. Restricted collocations tended to be. 19.

(40) analyzable like figurative idioms and free collocations, but fell in an intermediate position as being neither fully compositional nor unitary because of figurative or specialized meaning of one element and the literal meaning of the others. In phraseology, the verb in a verb + nominal object collocation has been characterized as typically having either a grammaticalized or delexical meaning as in “do the laundry”, a figurative meaning as in “run errands”, or a specialized meaning as in “rest the case”. This has led some researchers to explore a separate category of specialized or institutional lexical collocations (L’Homme and Bertrand, 2000).. In. this regard, there was still paucity of research to explore such questions as, how could the category be clearly defined? Could it be related to a technical register or a general domain like daily-life linguistic use? It was challenging for the continuum model how to treat institutionalized word combinations in which the constituents had a literal meaning; however, the model basically associated idiomaticity with non-literal meaning. All categories tended to be fuzzy at the edges and should have equal status. Geeraerts (1997) posed that the individual category including free and restricted collocation had a special status on the basis of good-example ratings of their centrality or prototypicality. Lakoff (1987) explained that prototype effects resulted from our judgment in terms of cognitive models, not that they represented inherent category structure. The prototype categories could “combine structural stability with flexibility” (Poulsen, 2005) and “new entities and new experiences can be readily associated, perhaps as peripheral members, to a prototype category, without necessarily causing any fundamental restructuring of the category system” (Taylor, 1995). The notions of the continumm model and prototype categories provided the essences to classify collocations, elaborate properties and serve as operational criteria for the present study. 20.