臺灣華語詞彙結構：詞彙聯想的實證 - 政大學術集成

全文

(1)國立政治大學語言學研究所碩士論文 National Chengchi University Graduate Institute of Linguistics Master Thesis. 指導教授：萬依萍博士 Advisor: Dr. I-Ping Wan. 立. 政治大. ‧ 國. 學 ‧. 臺灣華語詞彙結構：詞彙聯想的實證. sit. y. Nat. Lexical Organization in Taiwan Mandarin:. io. n. al. er. Evidence from Word Association Tasks. Ch. engchi. i n U. v. 研究生：曾子伊撰 Student: Tseng, Tzu-Yi 中華民國一百零九年一月 January, 2020. DOI:10.6814/NCCU202000123.

(2) Lexical Organization in Taiwan Mandarin: Evidence from Word Association Tasks. 立. 政治大 By. ‧. ‧ 國. 學 Tseng, Tzu-Yi. er. io. sit. y. Nat. n. al. C h Submitted toUthen i A Thesis engchi. v. Graduate Institute of Linguistics in Partial Fulfillment of the Requirements for the Degree of Master of Arts. January, 2020 DOI:10.6814/NCCU202000123.

(3) 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Copyright © 2020 Tseng, Tzu-Yi All Rights Reserved. DOI:10.6814/NCCU202000123.

(4) ACKNOWLEDGEMENTS It is such a blessing to have professor I-Ping Wan as the advisor of my thesis. Since we met in her Psycholinguistics class during my first semester in National Chengchi University (NCCU), I have admired her professional and the keen development on her transdisciplinary research. With the training at her Phonetics and Psycholinguistic Lab for two years, I am capable for the considerable experiment design as well as the analysis of Mandarin phonology in a psycholinguistic viewpoint. I would not come up with the idea. 政治大. of this thesis without her instruction; I would not dare to explore this topic across various. 立. fields in linguistics without her encouragements. I am very grateful for her instruction and. ‧ 國. 學. supports to me, especially the time she spent to reading my thesis word by word, which complete my master degree. Of course, professor Wan is also the crucial reason for me to. ‧. have opportunities for the invaluable advice from the committee members, professor Shu-. Nat. sit. y. Kai Hsieh and professor One-Soon Her. It is such an honor to have the chair of Graduate. er. io. Institute of Linguistics at National Taiwan University, professor Shu-Kai Hsieh, as one of. al. n. iv n C h experimental possible research limitations on both i U and data analysis, but also e n g c h materials. my committee members. I really appreciate that professor Hsieh not only pointed out the. gave me the detailed suggestions to my statistic calculation. With professor Hsieh’s evaluation, my thesis is permitted with a well-knit statistic support. Thanks to professor One-Soon Her’s instruction in Syntax during my second semester in NCCU as well, I was allowed the related knowledge to arrange the key element of my material design. Since my thesis proposal, professor Her has given me the great help on the coherence and cohesion throughout the thesis. Both of them have treated me as their own student by having my. i. DOI:10.6814/NCCU202000123.

(5) thesis carefully examined with high criterion as well as lots of encouragements and suggestions for the future and derived research. 將我最誠摯的感謝獻給我親愛的家人，二十幾年來的心力與支持成就我的學業，以及我現在能夠擁有的知識與能力都歸功於我的家人。(I would like to give my biggest thanks to my family who supports me studying for more than two decades. The knowledge and abilities I possess now are all from their love, time and efforts on me.) Special thanks go to my best friends, Cece, 靖瑜, and Valentin, who has always been there. 政治大. as the strongest mental supports for me whenever I was down that you reminded me to. 立. keep writing my thesis and moving on again and again. Thanks to Kayla, Alyssa, and. ‧ 國. 學. Katrina for your sweet supports since I have met you babies in Mississippi as well as your time and patience to check words for me again and again even after I came back to Taiwan.. ‧. Thanks to professor Rik De Busser’s instruction in Semantics during my first semester in. y. Nat. sit. NCCU, which has enhanced my motivation in linguistic research; the knowledge I have. n. al. er. io. learnt from his course also facilitated me a lot with the semantic analysis in my study.. i n U. v. Thanks to 惠鈴助教學姊 who helps me a lot during my study in NCCU with the. Ch. engchi. suggestions and encouragements cheering me up when I was lost. Finally, I would like to appreciate the angelic friends around me, that is, my roommates Michelle, who introduced some participants for my study, and made my days with your thoughtful sweetness; and 筱翎, who would always calm me down with cheerful encouragements and positive energy. It is really my pleasure to have these angels surrounding me; I would always remember and be grateful for your presence in my life.. ii. DOI:10.6814/NCCU202000123.

(6) CONTENT. ACKNOWLEDGEMENTS .............................................................................................. I CONTENT ....................................................................................................................... III INDEX OF FIGURES .................................................................................................... VI INDEX OF TABLES .................................................................................................... VII. 政治大 CHAPTER ONE. INTRODUCTION ............................................................................. 1 立 ABSTRACT .................................................................................................................. VIII. CHAPTER TWO. LITERATURE REVIEW ................................................................ 5. ‧ 國. 學. 2.1 MENTAL LEXICON .................................................................................................... 6 2.1.1 Models of Meaning: Semantic Representation .............................................. 8. ‧. 2.1.2 Models of Speech Production Planning ........................................................ 11. y. Nat. 2.1.3 Organization between Semantics and Phonology: Connectionism ............ 14. io. sit. 2.2 INFLUENCES OF WORD ACCESS AND ORGANIZATION ........................................... 17. er. 2.2.1 Frequency ........................................................................................................ 18. al. n. iv n C 2.2.3 Lexical Bias ..................................................................................................... 20 hengchi U 2.2.2 Imageability ..................................................................................................... 19. 2.2.4 Syntactic Category .......................................................................................... 20 2.2.5 Phonological Similarity .................................................................................. 21 2.3 WORD ASSOCIATION TASK .................................................................................... 23 2.3.1 Free versus Controlled Association............................................................... 24 2.3.2 Continued versus Single Response ................................................................ 25 2.3.3 Semantic vs. Phonological Relation .............................................................. 25 2.4 MANDARIN MENTAL LEXICON AND WORD ASSOCIATION .................................... 27 2.4.1 Mandarin Mental Lexicon ............................................................................. 28 2.4.2 Mandarin Word Association ......................................................................... 31 2.5 RESEARCH QUESTIONS AND HYPOTHESIS ............................................................. 35 iii. DOI:10.6814/NCCU202000123.

(7) CHAPTER THREE. METHODOLOGY ..................................................................... 40 3.1 DATA COLLECTION AND ELICITATION .................................................................. 40 3.1.1 Participants ..................................................................................................... 40 3.1.2 Materials .......................................................................................................... 41 3.1.3 Procedures ....................................................................................................... 48 3.1.4 Instruments ..................................................................................................... 50 3.2 DATA ANALYSIS ...................................................................................................... 50 3.2.1 Frequency ........................................................................................................ 50 3.2.2 Syntactic Category .......................................................................................... 52. 政治大. 3.2.3 Semantic Relations ......................................................................................... 52 3.2.4 Syllable Structure and Tone Structure ......................................................... 53. 立. 3.2.5 Interaction between Semantics and Phonology ........................................... 54. ‧ 國. 學. CHAPTER FOUR. RESULTS AND FINDINGS ........................................................ 56 4.1 FREQUENCY ............................................................................................................ 57. ‧. 4.2 SYNTACTIC CATEGORY .......................................................................................... 59 4.3 SEMANTIC RELATION ............................................................................................. 61. y. Nat. sit. 4.4 PHONOLOGICAL RELATION .................................................................................... 64. al. er. io. 4.4.1 Phonological Similarity .................................................................................. 65. n. 4.4.2 Syllable Structure and Tone Structure ......................................................... 67. Ch. i n U. v. 4.4.3 Syllable Type ................................................................................................... 69. engchi. 4.5 INTERACTION BETWEEN SEMANTICS AND PHONOLOGY ....................................... 70 CHAPTER FIVE. DISCUSSION .................................................................................. 74 5.1 TENDENCY OF LEXICAL ORGANIZATION IN TAIWAN MANDARIN ........................ 74 5.2 LINGUISTIC INFLUENCES IN LEXICAL ORGANIZATION ......................................... 76 5.2.1 Frequency Effect & Mental Networks .......................................................... 76 5.2.2 Semantic Networks & Syntactic Networks .................................................. 77 5.2.3 Phonological Networks ................................................................................... 78 5.2.4 Networks between Linguistic Influences ...................................................... 80 CHAPTER SIX. CONCLUSION .................................................................................. 82. iv. DOI:10.6814/NCCU202000123.

(8) REFERENCES ................................................................................................................ 85 APPENDIX I. EXPERIMENT MATERIALS ............................................................. 92 APPENDIX II. EXPERIMENT PROCEDURES ........................................................ 93 APPENDIX III. DATA ANALYSES ............................................................................ 96 APPENDIX IV. PROGRAMMING CODES IN R .................................................... 111. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. v. DOI:10.6814/NCCU202000123.

(9) INDEX OF FIGURES. Figure 2-1-1. An example of a hierarchically organized memory structure in Collins and Quillian (1969) ............................................................................................................ 8 Figure 2-1-2. An example of a spreading activation model in Collins and Loftus (1975) ........................................................................................................................................... 10 Figure 2-1-3. Garrett’s (1975) speech production processing model ............................ 12 Figure 2-1-4. An overview of language production processes reviewed in Bock and Levelt (1994) ..................................................................................................................... 13. 政治大 Figure 2-1-6. The lexical and word-shape networks in the model by Dell (1988) ........ 16 立 Figure 2-2-1. A sample of words from the English phonological network analyzed in Figure 2-1-5. A sample of a network for phonological encoding in Dell (1986) .......... 15. ‧ 國. 學. Vitevitch (2008) ................................................................................................................ 22 Figure 2-4-1. The pyramid structure model in Hsieh (2006, 2016). .............................. 29. ‧. Figure 3-2-1. The boxplots of the distribution among the frequency of the nouns and verbs .................................................................................................................................. 43. y. Nat. sit. Figure 4-1-1. The boxplots of the frequency distributions among the responses between. al. er. io. nouns and verbs ............................................................................................................... 57. n. Figure 4-1-2. The boxplots of the frequency distributions between stimuli and. Ch. i n U. v. responses........................................................................................................................... 58. engchi. Figure 4-2-1. Bar plot (left) and mosaic plot (right) of the observed frequencies and the corresponding Pearson residuals of the syntactic categories of the responses ............. 59 Figure 4-3-1. Bar plot and mosaic plot of the observed frequencies and the corresponding Pearson residuals of the semantic relations of the responses ............... 63. vi. DOI:10.6814/NCCU202000123.

(10) INDEX OF TABLES. Table 2-4-1. Main types of speech errors identified in Wan’s corpus ........................... 31 Table 3-2-1. The distributions of the word frequency in the task. ................................. 42 Table 3-2-2. The proportion of onsets with tones and syllable structures of target verbs ........................................................................................................................................... 44 Table 3-2-3. The proportion of the onsets, tones, syllable structures, and imageability of target nouns ...................................................................................................................... 45 Table 4-2-1. The values of the frequencies and residuals of the responses based on the. 政治大 Table 4-3-1. The values of the frequencies and residuals of the responses based on the 立 semantic relations by chi-square test............................................................................... 64 syntactic categories by chi-square test............................................................................. 60. ‧ 國. 學. Table 4-4-1. The distribution of the phonological similarity of the responses in syllable level and word level .......................................................................................................... 67. ‧. Table 4-4-2. The distribution of syllable structure and tone structure of the responses69 Table 4-4-3. The statistics of the distributions of the syllable types ............................... 70. y. Nat. sit. Table 4-5-1. The statistics of the distributions of the semantic and phonological. al. er. io. relations ............................................................................................................................ 73. n. Table 5-2-1. The distributions of each linguistic features between word pairs of. Ch. i n U. v. associations regardless syllable types .............................................................................. 80. engchi. vii. DOI:10.6814/NCCU202000123.

(11) 國. 立. 政. 治. 大. 學. 碩. 士. 論. 文. 提. 要. 研究所別：語言學研究所論文名稱：臺灣華語詞彙結構：詞彙聯想的實證指導教授：萬依萍博士研究生：曾子伊論文內容：(共一冊，20,574 字，分六章；以下扼要說明內容，共 666 字) 至今心理詞彙結構的相關理論和模型多數支持概念間的觸發連結是透過激發. 政治大. 作用(activation)傳遞的。而影響該連結與激發的因素又涵蓋詞頻、具體抽象度、真. 立. 詞作用(lexical effect)、詞性與音韻關聯。詞彙聯想測試被廣泛應用於心裡詞彙與記. ‧ 國. 學. 憶的研究，透過反應詞與刺激詞之間的關係，來追究詞彙聯想與擷取過程中各階段間的作用。然而目前多數與臺灣華語心理詞彙相關的研究並未以重要的語言學影響. ‧. 因素為實驗設計基礎，因此本研究目標以涵蓋上述影響特徵的自由雙字詞彙聯想測試，為臺灣華語心理詞彙結構提出一個概括性的探討。主要的研究問題分為兩大方. y. Nat. sit. 向：（一）語意與音韻何者對臺灣華語詞彙結構的影響較大？（二）各種語言學特. n. al. er. io. 徵對臺灣華語詞彙結構的影響為何？. i n U. v. 本研究提出的新興實驗方法，以平衡臺灣華語所有 21 個音節子音首，在雙. Ch. engchi. 字刺激詞的兩個子音音節首的分布，自 10 位（5 男 5 女；平均年齡 23.8 歲，標準差 1.9 歲）臺灣華語母語語者收集共 406 個華語雙字詞彙。詞彙聯想結果顯示：（一）臺灣華語詞彙偏向以語意組構；（二）語意關聯以聚合關係(paradigmatic relation)為主；語意、韻尾、名詞刺激詞與反應詞間的一致性等在臺灣華語詞彙結構中有較強的連結。鑒於本研究尚有幾許研究限制，建議後續較大規模的跨聲調語言研究；而本研究探討人類大腦詞彙的組構能對於未來在人類與人工智慧的神經網路聯結上提供概括性參考與貢獻。. 關鍵詞：詞彙結構、詞彙聯想、心理詞彙、臺灣華語、語意與音韻關聯 viii. DOI:10.6814/NCCU202000123.

(12) ABSTRACT. Current theories and models of lexical organization assume that the stimulation of an association results in an activation spreading to the related concepts within mental networks. Several factors have also been suggested to affect those connections to meet different purposes of lexical access such as frequency, imageability, lexical effect, lexical categories, and phonological similarity. Word association tasks have been widely implemented to psychological research of mental lexicon and memory; through the relations between stimuli and responses, different levels of processing are able to be traced.. 政治大 without concerning linguistic立 features in Mandarin as the basis. The present study therefore. However, most research related to Mandarin lexicon investigated the associated responses. aims to provide an outline of the lexical organization in Taiwan Mandarin by free word. ‧ 國. 學. associations. Topics to be explored involve 1) the tendency of lexical organization, and 2) the influences of linguistic features in Taiwan Mandarin lexicon.. ‧. With the new methodology conducted, in which the 21 possible onsets for. sit. y. Nat. disyllabic words and linguistic influences in Taiwan Mandarin are considered, total of 406 responses are collected from ten Taiwan Mandarin native speakers (age mean= 23.8 years. io. er. old; SD= 1.9; 5 males and 5 females). The results of free word associations in the present. al. n. iv n C U h e n grelation, Mandarin; 2) the tendency of paradigmatic instead of phonological c h i semantic study suggest the following: 1) a semantic tendency of lexical organization in Taiwan. relation, rhyming relation, and the consistency of syntactic categories in nouns are presented during associations in Taiwan Mandarin lexicon. However, regarding limited literature and the small scale of the present study, further research on phonological organization in mental lexicon across languages is suggested. The understanding of lexical organization in human brains may contribute to the further research on the functions of associations and networks between human and artificial intelligence.. Keywords: Lexical networks, Word associations, Semantic-phonological organization, Mental lexicon, Taiwan Mandarin ix. DOI:10.6814/NCCU202000123.

(13) CHAPTER ONE. INTRODUCTION Mental lexicon represents the structure of linguistic environment built in human minds and brains. It functions as a dictionary where we store the known words, and the organization of those words is different from people to people, i.e., each unique dictionary records how words are arranged after entering into one’s mind. Word formation in different languages with the linguistic constraints may be one of the factors that morphologically categorizes words together in the mental lexicon. Words with other similar linguistic. 政治大. features such as phonological, semantic, and syntactic ones are also possibly put in the. 立. same page of the mental dictionary, e.g., ‘cake’ and ‘cookie’ in English have the same part-. ‧ 國. 學. of-speech and initial sound; 開心 [kaj55ɕin55] and 愉快 [y35kʰwaj51] in Taiwan Mandarin share the similar meaning as a state of feeling happiness, and can be the same. ‧. part-of-speech. Frequency plays an integral role in lexical organization as well, so some. Nat. sit. y. words may be bookmarked because of habitual usages or for different purposes in certain. n. al. er. io. linguistic environments. Language usages make different demands upon the lexicon, which. i n U. v. regards the mapping and decoding process between concepts and meanings with forms. Ch. engchi. during perception and production. Therefore, understanding lexical organization in human minds may help with the exploration of the retrieval, comprehension, acquisition and usage of mental lexicon. Previous theories have been developed two major types of models detailing the possible lexical access as the sequential search models and parallel access models (e.g., Forster’s (1976) autonomous search model and Morton’s (1969) logogen model, respectively). On the other hand, numerous studies have put the focus on models of conceptual and lexical organization in mental lexicon. For example, Collins and Quillain’s 1. DOI:10.6814/NCCU202000123.

(14) (1969) hierarchical network model; Collins and Loftus’ (1975) spreading activation network model of semantic representation; and Fromkin (1971), Garrett (1975) and Levelt’s (1983, 1989) models of speech production planning based on speech errors. To further explore the organization and interaction between semantics and phonology during lexical access and selection, the connectionist models were also developed with linguistic features linked through the activation spreading (e.g., McClelland & Rumelhart, 1981; Dell, 1985, 1986, 1988).. 政治大 stimulation of an association results in an activation spreading to the related concepts 立. A common assumption shared by current theories of lexical organization is that the. ‧ 國. 學. within the mental networks. Several factors have been suggested to influence those connections that the spread of activation follows to meet different purposes of word. ‧. retrieval and selection including frequency (e.g., Dell, 1990; Kittredge et al., 2008; Nozari. sit. y. Nat. et al., 2010), imageability (e.g., De Groot, 1989; De Deyne & Storms, 2008; Malhi, 2018),. io. er. lexical effect (e.g., Dell, 1985; Dell & O’Seaghdha, 1991, 1992; Oppenheim & Dell, 2008),. al. iv n C 2016), and phonological similarity (e.g., 1986, 1988; Dell et al., 2000; Vitevitch, h eDell, n g1985, chi U n. lexical categories (e.g., Dell, 1990; Nissen & Henriksen, 2006; De Simone & Collina,. 2008; Vitevitch & Goldstein, 2014; Beckage & Colunga, 2016; Rebei et al., 2019). With the influences concerned, psychological research of mental lexicon and memory have traditionally applied experiments such as the Stroop, lexical decision, word categorization, recognition memory, and word association tasks; or supported by the evidence from speech errors (e.g., Warren (1974) for the interference to memory between colors and words by a Stroop task; and De Groot (1989) for the relations of lexical frequency by the lexical decision and word association tasks). Word association tasks have been widely. 2. DOI:10.6814/NCCU202000123.

(15) implemented to develop semantic networks (e.g., Entwisle, 1966; Dell, Oppenheim & Kittredge, 2008; De Deyne et al., 2013) and to establish free-association norms of mental lexicon (e.g., Nelson et al. (2004) for English; Hu et al. (2017) for Mandarin Chinese). The relations between stimuli and responses have been accepted being able to trace different levels of word processing, ranging from phonology (e.g., Vitevitch, 2008; Vitevitch & Goldstein, 2014; Beckage & Colunga, 2016; Rebei, Anderson & Dell, 2019) to semantics (e.g., Entwisle, 1966; Nissen & Henriksen, 2006; Wan & Ting, 2019). Moreover, the. 治政 words in the mental lexicon (e.g., Dell, 1990; Kittredge et大 al., 2008; Nozari et al., 2010). 立. frequency of responses is typically regarded as the strength of the connection between two. ‧ 國. 學. On the other hand, as mentioned above, different linguistic rules shaping languages can affect the lexical organization; therefore, to investigate the organization of mental. ‧. lexicon in Taiwan Mandarin, the linguistic features of this language should be taken into. sit. y. Nat. account as the foundation first and supported by the experimental evidence later. However,. io. er. most research related to Mandarin lexicon focuses on bilingual comparison (e.g., Lee, 2012;. al. iv n C in Mandarin concerned as the basis. Therefore, h e n gthec possible h i U influences on Mandarin lexicon n. Zhao, 2013), or analyzes the responses to association tasks without the linguistic features. such as word frequency, syllable type, syllable structure, tone structure, imageability and syntactic categories are all considered as the experimental foundation in the present study to fill this gap1. To further explore the lexical organization in the aspects of the relationship. 1. Though several works related to word associations implemented experiments with monosyllabic. words (e.g., Chen & Li, 2011), regarding the common tendency of disyllabification in the Mandarin lexicon (see Wang & Sun, 2015 for more information), only disyllabic words are involved in the word association task of the present study. 3. DOI:10.6814/NCCU202000123.

(16) between phonology and semantics with parts-of-speech in Taiwan Mandarin, a free word association task will be conducted with disyllabic nouns and verbs in the present study2. During the word association task, participants are required to produce the first word emerging in their mind upon a lexical item as the stimulus presentation every trail. The analyses and the categorizations of responses to the free word association tasks are later concentrating the relations between phonology and semantic in the Mandarin lexicon with the effects of parts-of-speech and lexicography explored as well.. 政治大 proposed by the previous scholars with the methodology of the present study explained in 立. The following chapter will review the basic and classic concepts of mental lexicon. ‧ 國. 學. the third chapter. The results of the free word association task will be presented in the fourth chapter with the findings discussed in the fifth chapter. Finally, the research. ‧. limitations and suggestions will be stated with a conclusion remarked in the last chapter.. n. er. io. sit. y. Nat. al. 2. Ch. engchi. i n U. v. Even though the word association task is commonly applied to assess semantic memory in. psycholinguistics, the aspect of the phonology will be investigated as well in the present study; therefore, the International Phonetic Alphabet (IPA) symbols will be used to identify and analyze the phonological relations among Taiwan Mandarin words. As for the material involved in the tasks for the participants, the words will be presented as Mandarin Phonetic Symbols (MPS; Zhuyin Symbols) to get rid of the influences from lexicography (for more information please see Chapter Three). 4. DOI:10.6814/NCCU202000123.

(17) CHAPTER TWO. LITERATURE REVIEW Mental lexicon functions as a word storage in human mind, in which concepts are mapping to words, and words are connected to each other by shared features relevant to phonological, morpho-syntactic, and semantic functions. Numerous studies have already proposed the conceptual and lexical organization by various models of meaning and sound to investigate the structure and access in the mental lexicon. For example, Collins and Quillain’s (1969) hierarchical network model; Smith, Shoben and Rips’ (1974) feature. 政治大. comparison model; and Collins and Loftus’ (1975) spreading activation network model of. 立. semantic representation. Fromkin (1971), Garrett (1975) and Levelt’s (1983, 1989) models. ‧ 國. 學. of speech production planning based on speech errors have contributed as well. With the concept of the activation spreading mentioned, the connectionist models were also. ‧. developed to explain the organization and interaction between semantics and phonology. Nat. sit. y. during lexical access and selection (e.g., McClelland & Rumelhart, 1981; Dell, 1985, 1986,. er. io. 1988). Several key factors influencing word access and organization were suggested by. al. n. iv n C h esimilarity phonological i UDell (1990), n g c h(e.g.,. various empirical findings mainly including frequency effect, imageability, lexical bias, lexical categories, and. Kittredge, Dell,. Verkuilen and Schwartz (2008), and Nozari, Kittredge, Dell and Schwartz (2010) for frequency effect; De Groot (1989), De Deyne and Storms (2008), and Malhi (2018) for imageability; Dell (1985), Dell and O’Seaghdha (1991, 1992), and Oppenheim and Dell (2008) for lexical bias; Dell (1990), Nissen and Henriksen (2006), and De Simone and Collina (2016) for lexical categories; Dell (1985, 1986, 1988), Dell, Reed, Adams and Meyer (2000), Vitevitch (2008), Vitevitch and Goldstein (2014), Beckage and Colunga (2016), and Rebei, Anderson and Dell (2019) for phonological similarity). 5. DOI:10.6814/NCCU202000123.

(18) Through the above models of conceptual and lexical organization in the mental lexicon, the psychological networks between sounds and meanings in the Mandarin lexicon are intended to be outlined in the present study. To introduce the basic and classic concepts of the mental lexicon proposed by the previous scholars, in this chapter, the English lexical access and organization will be investigated first, followed by some previous studies in the Mandarin mental lexicon. The models of English mental lexicon built by previous researchers with their reviews on the development of psycholinguistic theories will be. 治政大 in 2.1.1, 2.1.2 and 2.1.3 semantic and phonological organizations will be demonstrated 立 presented in the section 2.1 along with the models of meanings and sounds, and those of. respectively. The influences of the English word access and organization will be discussed. ‧ 國. 學. in the section 2.2. Evidence on different research methodologies of English word. ‧. association tasks in Mandarin will be reviewed in the section 2.3. Mandarin mental lexicon and studies of word association tasks will be researched with the language differences. y. Nat. hypotheses will be proposed in the last section.. n. al. 2.1 Mental Lexicon. Ch. engchi. er. io. sit. against English considered in the section 2.4. Finally, the research questions and. i n U. v. Not only as a knowledge storage, mental lexicon is also assumed to reflect the linguistic environment due to the acquisition, retrieval, and usage of mental lexicon as a network derived from languages. A number of researchers have put much effort to explore the English mental lexicon for accessing approximations of the structure of language and knowledge stored in human minds and brains. Levelt (1993) described mental lexicon as a repository of knowledge about the words of the language, which is a listing of four features, i.e., meaning, syntactic properties, morphological specification, and phonological 6. DOI:10.6814/NCCU202000123.

(19) segments. Jarema and Libben (2007) defined the mental lexicon as the cognitive system that constitutes the capacity for lexical activity. Aitchison (2012) mentioned that the mental lexicon should be regarded as a metaphor with links, cores, and frameworks but not locations or fixed details, where language and cognition overlap and interact for the purpose of easy production and comprehension, i.e., communication. Vitevitch and Goldstein (2014) found that keywords occupy a unique position in the English lexicon, in which the overall distance of keywords is minimized and held together as smaller. 治政大organize previous viewpoints and microscopic properties of mental lexicon to conclude and 立 components. De Deyne, Verheyen and Storms (2016) further discussed the macro-, meso-,. from broad aspects to specific ones. The macroscopic level covers all the connections. ‧ 國. 學. between the concept representations; the mesoscopic level presents a subset in the network,. ‧. i.e., the meaning of words, achieved by computing the distance through a set of direct and indirect paths connecting them; the microscopic level explains the function of the network. y. Nat. er. io. al. n. and semantic properties.. sit. in terms of the effects of variables to the connections such as word frequency, imageability,. Having the concept. iv n C and structure U lexicon h e n gof ctheh imental. relatively concrete,. numerous models presented by previous researchers symbolizing mental lexicon as a network consist of nodes representing lexicalized concepts, and lines between these nodes linking lexico-semantic relationships in the lexicon. For example, the models of meaning such as Collins and Quillain’s (1969) hierarchical network model, and Smith, Shoben and Rips’ (1974) feature comparison model, along with Collins and Loftus’ (1975) spreading activation network model; the models of speech production planning such as Fromkin. 7. DOI:10.6814/NCCU202000123.

(20) (1971), Garrett (1975) and Levelt’s (1983 and 1989) models; and the models about the interaction between semantics and phonology during lexical access by connectionism.. 2.1.1 Models of Meaning: Semantic Representation Semantic networks are often used to represent the organization of the connection between meaning of words in mental lexicon regarding cognitive processes of language. Therefore, words should be connected to those with the related meanings, and the strength of the connection should represent the semantic similarity between two words. However,. 政治大. semantic networks function more complicated than shared meanings because the semantic. 立. features can be connected dependent on context; semantic relations capture more than. ‧ 國. 學. shared meanings, but also can be refer to hierarchical, inclusional and other complex relationships.. ‧. To investigate such semantic organization, Collins and Quillian (1969) presented. sit. y. Nat. io. n. al. er. the first cognitive model of semantic representation as Figure 2-1-1.. Ch. engchi. i n U. v. Figure 2-1-1. An example of a hierarchically organized memory structure in Collins and Quillian (1969) 8. DOI:10.6814/NCCU202000123.

(21) This hierarchical network model demonstrates that lexical meanings are arranged by increasing features with the relation to those at the highest relevant level. Words are stored with pointers to those at the same level, and with semantic connections between associated concepts. The model groups and labels the associated semantic features which can be subdivided into a subordinate category, and the superordinate layer is similarly to be included in a higher category. For example, the associated semantic representation of ‘Bird’ are pointed with the connections as ‘has wings’, ‘can fly’ and ‘has feathers’, and it shares the increasing feature of ‘Animal’, in the higher level—level 2, with the semantic. 治政 representation of ‘Fish’, in the same level—level 1. The 大 cognitive economy of the mental 立 lexicon thus seems able to be achieved via this hierarchical structure.. ‧ 國. 學. Smith, Shoben and Rips (1974) therefore based on the feature similarity and created. ‧. the feature comparison model that the meaning of a word can be represented as a list of. sit. y. Nat. semantic features. Some of these features indicate essential aspects to the meaning of a. io. er. word, i.e., defining features, while others show accidental aspects i.e., characteristic. al. iv n C characteristic features refer to the common unnecessary features that members of the h e nbut gchi U n. features. Therefore, defining features concern the inclusional relationships in a category;. category have. The distinction between semantic representations thus takes its place by listing, ordering, and comparing their overlapping features. For example, the comparison and distinction of the features between ‘bird’ and ‘canary’ are that both of them have the defining features of ‘can fly’ and ‘has wings’, but ‘is yellow’ can only be the defining features for ‘canary’ as a characteristic feature. In other words, since ‘canary’ contains more defining features than ‘bird’, this also illustrates another idea of Smith et al. that the more the number of defining features, the more abstract the lexical item is. However, even. 9. DOI:10.6814/NCCU202000123.

(22) the feature comparison model classifies the semantic distances between shared features by comparison, and solved some problems from the hierarchical network model, the distinction between defining and characteristic features is unclear3. Collins and Loftus (1975) further represented another type of semantic networks on semantic relations as the activation spreading and accumulating from the given initial priming, as shown in Figure 2-1-2.. 立. 政治大. ‧. ‧ 國. 學 er. io. sit. y. Nat. al. n. iv n C h e activation Figure 2-1-2. An example of a spreading i U in Collins and Loftus (1975) n g c hmodel Unlike the strict hierarchy and redundant processing as former models have, the spreading activation model presents as a complex web to express the more complicated relations. 3. Smith, Shoben and Rips (1974) argued that more typical members of a category should be. verified more quickly when processing the semantic representation, i.e., frequency effects. However, this process was not dealt with in the hierarchical network model (Collins & Quillian, 1969), in which all members of a category are treated as equal members of that category (see Smith, Shoben & Rips, 1974; also Reeves, Hirsh-Pasek & Golinkoff, 1998 for more information). 10. DOI:10.6814/NCCU202000123.

(23) between semantic representations. Nodes are the semantic representations connected by lines where the length of each line shows the degree of association between two concepts, i.e., semantic distance. With the activation spread between several nodes simultaneously, the nodes can accumulate in various states of activation that can be further spread to other nodes connected. As one can see in the figure, if the initial priming was given as ‘Red’, some of the activations may spread to a subordinate category that seemingly can be named as ‘fruit’ (in the right side of the ‘Red’ node), in which the direct connections only shown from ‘Red’ to ‘Apples’ and ‘Cherries’, but not ‘Pears’, which also shares the connection. 治政大 connection between ‘Red’ with both ‘Apples’ and ‘Cherries’. Therefore, there is an indirect 立. and ‘Pears’, but somehow they simultaneously share the direct relations with ‘Apples’ and. ‧ 國. 學. ‘Cherries’, respectively.. ‧. 2.1.2 Models of Speech Production Planning. y. Nat. sit. At the same time, several speech production planning models were also proposed. n. al. er. io. since the attention on the processing of phonological encoding for serial-order speech error. i n U. v. patterns during the lexical retrieval and organization (e.g., Fromkin, 1971; Garrett, 1975;. Ch. engchi. Levelt, 1983, 1989). Following Fromkin’s (1971) model that first specified the stages and representations in speech production based on speech errors, Garrett (1975) proposed a more detailed speech production planning model starting from a message level that conveys information by words encoding with concepts and other linguistic features shown as Figure 2-1-3.. 11. DOI:10.6814/NCCU202000123.

(24) 立. 政治大. ‧ 國. 學. Figure 2-1-3. Garrett’s (1975) speech production processing model. According to this model, to generate a production, messages with concepts embedding in. ‧. words are selected and attached with grammatical encodings in the function level, and. Nat. sit. y. therefore phonological encodings, for instance, syllabification and prosody, are placed into. n. al. er. io. the slots of selected phonetic segments in the position level, for preparing pronounceable. i n U. v. patterns in the sound level before introduced to the articulatory gestures. It is the position. Ch. engchi. level where sounds, morphemes, or words exchange in Garrett’s model that makes it different from Fromkin’s model, in which the phonological rules determining the fully specified phonetic segments are applied. With the observation that speakers tend to self-monitor their intended productions, Levelt (1983, 1989) followed Garrett’s outline to create a model including four levels of processing, the message level, the functional level, the positional level, and the phonological level with the interaction processing between grammatical and phonological encodings as Figure 2-1-4. 12. DOI:10.6814/NCCU202000123.

(25) 政治大 Figure 2-1-4. An overview of立 language production processes reviewed in Bock and Levelt. ‧ 國. 學. (1994). The processes of the phonological encoding comprising both sound and intonation forms. ‧. interact with that of the grammatical encoding comprising both selection of lexical. y. Nat. sit. concepts and syntactic properties in the functional level and the positional level. Therefore,. n. al. er. io. this model seems to more specifically provide a further explanation of the stages where. i n U. v. different patterns of speech errors occur according to which specific linguistic features. Ch. engchi. being dealt with. For example, word exchanges accounting for lexical selection and syntactic feature assignment will happen during the functional process before the occurrences of sounds errors in the last stage, phonological encoding, which deals with both the phonological segments of word forms and the prosody of larger units (e.g., Bock & Levelt, 1994). Through the models of semantic representation and spoken word production discussed in this section, the organization of the semantic relations and phonological processes in the mental lexicon are included. In the following section, the models of lexical 13. DOI:10.6814/NCCU202000123.

(26) access regarding the interaction between semantics and phonology will be introduced to further propose the activation between perceptual stimuli about a lexical item and other linguistic units, and gradually develop the bidirectional connections of access routes to lexical production.. 2.1.3 Organization between Semantics and Phonology: Connectionism Regarding the speech production planning models based on serial-order speech errors (Fromkin, 1971; Garrett, 1975) and the previous network models (e.g., Collins &. 政治大. Loftus, 1975; McClelland & Rumelhart, 1981), Dell (1985) further proposed the. 立. bidirectional connections between nodes we discussed above to explain that not only do. ‧ 國. 學. activated letter units send activation to all words that contain them, but the words also send activation back to the letter units in language production. The nodes in the network stand. ‧. for linguistic units such as concepts, words, morphemes, phonemes, and phonemic features,. y. Nat. sit. as well as syllables and syllabic constituents, and these conceptual nodes thereby define. n. al. er. io. the words by connecting to other nodes. Dell (1986) reviewed his former model to construct. i n U. v. the model on the basis of spreading activation network from the semantic to the. Ch. engchi. phonological level, and supported by the experimental evidence from phonological encoding as Figure 2-1-5.. 14. DOI:10.6814/NCCU202000123.

(27) 立. 政治大. Figure 2-1-5. A sample of a network for phonological encoding in Dell (1986). ‧ 國. 學. This model contributes the explanation of the linguistic causes to speech errors with the. ‧. internal representations of utterances-to-be-spoken by both top-down and bottom-up connections. The demonstration of the various linguistic units fitting into nodes includes. y. Nat. er. io. sit. syllables, consonant clusters, rimes, morphemes, and phonological features. Each of the phonemic nodes is marked as onset, nucleus or coda. A linguistic representation that. n. al. Ch. i n U. v. involves in various units must be represented by matched nodes such as /k/, one of which. engchi. is marked as a potential onset and another one as a potential coda. Dell (1988) further assumed that the only levels of lexical representation are a word and a phoneme level, i.e., two aspects of the lexical network are words and sounds. Illustrating the higher hierarchical aspect of the model we saw above as Figure 2-1-5, another advanced model followed presents the distinction between the lexical and wordshape networks with the demonstration of the connections between two networks by the dotted lines, and the sequence of activation by the arrows between phoneme nodes in the. 15. DOI:10.6814/NCCU202000123.

(28) word-shape network as Figure 2-1-6.. 立. 政治大. ‧ 國. 學. Figure 2-1-6. The lexical and word-shape networks in the model by Dell (1988). ‧. In this case, the intended phrase to illustrate in the model is ‘deal back’ labeled by numbered flags on the word nodes through the process of activation and spread of. y. Nat. er. io. sit. activation from the word ‘deal’ to the selection of phonemes before the sounds of ‘back’ are retrieved and selected in proper order.. n. al. According to Dell et. iv n C al. (1999), the complexity of h e nsince gchi U. the mapping from. perception to production during the process of word recognition and retrieval, the lexical access is divided to two steps—lemma selection and phonological encoding. To process the language acquisition, comprehension and production, a concept is mapped onto a lemma, a nonphonological representation with the grammatical properties of a word, and the lemma will be transformed into an organized sequence of speech sounds. Therefore, words are accessed and ordered also by their grammatical similarities, and their sounds are retrieved and organized for articulations.. 16. DOI:10.6814/NCCU202000123.

(29) Suggested by various findings of the models of lexical access that word recognition and retrieval are influenced by several key characteristics of words, in the next section, the influences of word access and organization will be investigated based on the viewpoints of the structure of the mental lexicon models we discussed before.. 2.2 Influences of Word Access and Organization Word access and organization are sorted and influenced by the following factors: (a) frequency effect, (b) imageability, (c) lexical bias, (d) lexical categories, and (e). 政治大. phonological similarity4. For instance, evidence of frequency effect was found by Dell. 立. (1990), Kittredge, Dell, Verkuilen and Schwartz (2008), and Nozari, Kittredge, Dell and. ‧ 國. 學. Schwartz (2010); evidence of imageability was found by De Groot (1989), De Deyne and Storms (2008), and Malhi (2018); evidence of lexical bias was found by Dell (1985), Dell. ‧. and O’Seaghdha (1991, 1992), and Oppenheim and Dell (2008); evidence of lexical. y. Nat. sit. categories was found by Dell (1990), Nissen and Henriksen (2006), and De Simone and. n. al. er. io. Collina (2016); evidence of phonological similarity was found by Dell (1985, 1986, 1988),. i n U. v. Dell, Reed, Adams and Meyer (2000), Vitevitch (2008), Vitevitch and Goldstein (2014),. Ch. engchi. Beckage and Colunga (2016), and Rebei, Anderson and Dell (2019). To understand and figure out the possible variabilities which may affect the result of the word association tasks in the present study, the influences of lexical access is mainly. 4. Several studies also concerned the effect from the age of acquisition in lexical retrieval;. however, this influence does not fit the research purpose in the present study, so it will not be discussed thoroughly. 17. DOI:10.6814/NCCU202000123.

(30) according to Reeves et al.’s (1998) organization regarding the evidence from other researchers mentioned above, and will be carefully discussed in the following sections5.. 2.2.1 Frequency Frequency effect is caused by words being experienced in diverse contexts rather just an individual’s language experience. Evidence from a large number of research have been consistently shown that lexical frequency has an effect on lexical processing. Dell (1990) applied a spreading-activation model of lexical retrieval in both production and. 政治大. recognition, and found that it was frequency but not the vocabulary types (function and. 立. content morphemes) that put influences on phonological errors. Kittredge et al. (2008). ‧ 國. 學. found evidence from aphasic picture-naming error that lexical frequency had significant influences on both phonologically and semantically related errors. Nozari, Kittredge, Dell. ‧. and Schwartz (2010) found that results were sensitive to frequency effect in both aphasic. y. Nat. sit. picture naming and auditory word repetition tasks with both of the semantic and. n. al. er. io. phonological steps involved.. Ch. i n U. v. Further pieces of the evidence about the frequency effects were located such as at. engchi. the retrieval of a word’s phonological and semantic characteristics. Therefore, frequency effect influences the way a word being chosen to produce for a presentation of the meaning that a speaker wants to convey. The principle of frequency in lexical access also interacts with concreteness which will be introduced in the following.. 5. Reeves, Hirsh-Pasek and Golinkoff (1998) proposed five factors influencing word access and. organization including frequency, imageability and concreteness and abstractness, semantics, grammatical class, and phonology. 18. DOI:10.6814/NCCU202000123.

(31) 2.2.2 Imageability Imageability deals with the issue between concreteness and abstractness of a concept, in which the former concept is easier to be come up with internal pictures such as ‘tree’; on the contrary, the latter one is more difficult to generate an image of the word meaning, for instance, ‘freedom’. As mentioned in the last section, imageability also interacts with frequency; therefore, words with high frequency and high concreteness (e.g., ‘apple’) are best accessed and recalled, and vice versa. That is, in comparison, abstract. 政治大. words with low frequency are later and/or more difficult to be retrieved from the mental. 立. lexicon. De Groot (1989) found that word imageability strongly determines responses in. ‧ 國. 學. word association, and this effect also interacts with the frequency variable in lexical decision task. However, De Deyne and Storms’ (2008) experimental results somehow. ‧. contradicted De Groot’s finding that most core nodes in the mental lexicon have high. Nat. sit. y. frequency due to being early acquired, i.e., the nodes are therefore retrieved more times,. er. io. but not necessarily associated with mental imagery. On the other hand, Caplan and Madan. al. n. iv n C U research, words with higher h organization. the process of the lexical access and e n g c h Ini their (2016) also proposed that imageability is one of the most important factors that involves in. imageability were better recalled than those with lower imageability during the associative processing; furthermore, that was the imageability instead of the lexical effect, a preference on a real word or not, that enhanced the strength between association and memory. Therefore, concrete words with rich semantic representations are attested to be accessed easier and be stored deeper in the brain, i.e., in the core of the mental lexicon.. 19. DOI:10.6814/NCCU202000123.

(32) 2.2.3 Lexical Bias Most evidence of lexical bias effect come from speech errors, and this effect refers to the tendency for phonological speech errors to result in real words rather than nonwords. One of the general effects on phoneme-morpheme feedback in the model proposed by Dell (1985) is lexical bias, i.e., a speaker would have the tendency to create morphemes for an existent word. The causes of lexical bias were described with his explanations of the semantic influences on different levels of the language perception including the effects of. 政治大. the lexical context on the perception of letters. Also, the semantic stimuli on the perception. 立. of words, and that on the production of a string of phonemes lead to the lexical bias effect. ‧ 國. 學. as well. This effect has been supported by the latter studies about lexical priming and the stages of lexical access (e.g., Dell & O’Seaghdha, 1991, 1992). Oppenheim and Dell (2008). ‧. further examined both lexical bias and phonemic similarity effect between inner and overt. Nat. sit. y. speech, in which the inner speech indicates the voices produced in mind, and the latter one. n. al. er. io. means the speech spoken out loud. The results contributed that lexical bias presented in. i n U. v. both inner and overt speech errors, which demonstrate the great effect of the lexical bias. Ch for the lexical access and organization.. engchi. 2.2.4 Syntactic Category Different syntactic categories serve different functions in language. Words from different syntactic categories operate differently in terms of the manner and degree of integration into the word web, along with the way to establish vertical and horizontal semantic and syntactic relations to other words. Therefore, it is important to distinguish. 20. DOI:10.6814/NCCU202000123.

(33) their respective networks and understand its influences on the lexical access and organization. In Dell’s experiment (1990) about the effects of frequency and vocabulary type on phonological speech errors, the evidence showed a distinctive frequency effect between two types of vocabulary, i.e., function and content words, when they were presented as a whole syntactic unit, even if the function and content morphemes were not distinguished in the lexical access and recognition. Nissen and Henriksen (2006) found a clear evidence. 政治大 elicit a higher proportion of paradigmatic responses than verbs and adjectives. Words are 立 of lexical category affecting the responses in word association test, especially that nouns. ‧ 國. 學. classified as paradigmatic relation by being distinguished between alternative semantic connections, but locating within the same syntactic slot; words are classified as syntagmatic. ‧. relation by connecting syntactic relations and combining the semantic connections.. Nat. sit. y. De Simone and Collina (2016) proposed an evidence of the syntactic category. n. al. er. io. effects from the activation of lexical categories. This type of activation conveys between. i n U. v. the representation of the lexical categories during lexical access and production, and before. Ch. engchi. the target word is selected, all the candidate word with strong connections, i.e., the words sharing the same grammatical category, spreads the activation to each other. Furthermore, the syntactic category effects were considered as independent from semantic in their experiments because they still emerged though one of the influences was controlled.. 2.2.5 Phonological Similarity Phonology constructs the integral basis of language network representations. A large amount of studies have investigated phonological networks mostly by the analysis of. 21. DOI:10.6814/NCCU202000123.

(34) speech production errors (e.g., Dell, 1985, 1986, 1988; Dell, Reed, Adams & Meyer, 2000; Vitevitch, 2008; Vitevitch & Goldstein, 2014; Beckage & Colunga, 2016; Rebei, Anderson & Dell, 2019). Dell, Reed, Adams and Meyer (2000) proposed that the presence of nonwords from phonological speech errors happens often, but the chance of the violation of phonotactic constraints shows relatively small because the preferences that certain phonemes appear in the position of the word-initial or word-final are ruled by the language. Take English consonant [ŋ] for example, if [n] is mispronounced as [ŋ], the [ŋ] will always. 治政大 network by 20,000 English (2008) presented a small portion of the English phonological 立. appear in the word-final. Phonological similarity was illustrated as well when Vitevitch. word-forms for word speech and the neighbors of speech, in which nodes represent words. ‧ 國. 學. in the lexicon; links connect words with phonological similarity as Figure 2-2-1 below.. ‧. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 2-2-1. A sample of words from the English phonological network analyzed in Vitevitch (2008) The figure illustrates the word speech, the neighbors of speech, and the neighbors of those neighbors. For example, the node for the word ‘peach’ /pitʃ/ would have a link connecting. 22. DOI:10.6814/NCCU202000123.

(35) it to the nodes representing the words ‘preach’ /pɹitʃ/, ‘peat’ /pit/, ‘patch’ /pæ tʃ/, ‘each’ /itʃ/, etc. These examples explain the phonological similarity of the words with /p/ as the onset, their minimal pairs (e.g., ‘preach’ /pɹitʃ/ and ‘each’ /itʃ/), or words with similar phonological conditions (e.g., ‘speak’ /spik/ and ‘speed’ /spid/). The first two ones connect to each other directly; however, the latter ones have at least one node between them in the phonological network of the mental lexicon. Beckage and Colunga (2016) mentioned the same phonological similarity as. 政治大 deletion) of one word into another word by a series of phoneme changes. For example, the 立 Vitevitch based on ‘edit distance’, i.e., transformations (e.g., substitution, insertion, and. ‧ 國. 學. word ‘kit’ can transfer to ‘hit’ with the substitution of the onset from /k/ to /h/; to the word ‘skit’ via inserting another consonant /s/ in the word-initial to constitute the consonant. ‧. cluster /sk/; to the word ‘it’ by deleting the onset /k/ and remaining the rhyme.. y. Nat. er. io. sit. 2.3 Word Association Task. Word association is assumed able to encode mental representations and reflect. n. al. Ch. i n U. v. fundamental characteristics of the relations between words in the mental lexicon. Since it. engchi. allows the concepts to be free from the use of pragmatics for extended purpose, or break out of the frame of syntax as a sentence, word association is believed to be simply the reflection from mental constraints of what prominently comes to mind. In this section, numerous word association tasks will be reviewed as a serious of examples to investigate the appropriateness of the research methodology in this study. Types and purposes of tasks are described by the way that cued words are perceived for association, i.e., how those words can be associated (e.g., the word ‘leaf’ may be associated. 23. DOI:10.6814/NCCU202000123.

(36) by the cued word ‘tree’), and the relations between responses as well as between stimuli and responses during associations. Free and controlled association tasks (e.g., Nelson et al., 2004; Sailor et al., 2009) will be introduced first, followed by the tasks with single and continuous word responses (e.g., De Deyne et al., 2013; Aldridge et al., 2018), and the responses of stimuli with syntagmatic or paradigmatic relations (e.g., Entwisle, 1966; Nissen & Henriksen, 2006; Dell, Oppenheim & Kittredge, 2008).. 2.3.1 Free versus Controlled Association. 政治大. According to the access that cued words are perceived, two types of word. 立. association tasks can be classified. Free association allows an examinee to give the. ‧ 國. 學. response associated directly from the cued word; while the controlled association provides the fixed responses to the targets for a participant to choose the preferences which relate to. ‧. the cues more. Examples of each type of tasks are presented in the following. One of the. y. Nat. sit. most famous free association tasks were implemented by Nelson, McEvoy and Schreiber. n. al. er. io. (2004) in which the examinees were allowed to produce only a single word in response to. i n U. v. another word given as input. The researchers also proposed that this type of association can. Ch. engchi. provide a relative response, rather than an absolute one, which agrees with the underlying networks with general constraints. On the other hand, De Young, Lavender, Washington, Looby and Anderson (2010) conducted a controlled association task in which the participants were required to sort words into categories when the categories are paired with one another. Therefore, the participants may need to decide a cued word (e.g., spider) belongs in, for instance, the. 24. DOI:10.6814/NCCU202000123.

(37) ‘good’ or ‘flower’ category, or the ‘bad’ or ‘insect’ category, and the result suggested decreasing discomfort from pre- to post-test for words that were targeted in the task.. 2.3.2 Continued versus Single Response Regarding the process of the perceived cues are associated, two types of word association tasks can be classified. Continued word association allows a sequence of associated responses which is naturally seen as dynamic process for a stream of ideas, while single word association limits the response only one directly association from each cued. 政治大. word to prevent the possibility of other indirect or unrelated associations.. 立. In the research of De Deyne, Navarro, and Storms (2013), each participant was. ‧ 國. 學. permitted to generate free associations for three different responses to each cue. The result. ‧. reported that this type of format enables a better approximation of weak connections in the network, and also provides better predictions of lexical access and semantic relatedness.. sit. y. Nat. io. er. On the other hand, since more responses given to the same cued word, the associative connections may become less predictable and less well researched, Aldridge, Fontaine,. n. al. Ch. i n U. v. Bowen, and Smith (2018) suggested to focus on the tasks with single response to explore. engchi. the access of the mental lexicon. Their result also illustrated that the more the responses are elicited, the weaker the connections they have.. 2.3.3 Semantic vs. Phonological Relation The lexical organization in the mental lexicon reveals word development and knowledge; the syntagmatic and paradigmatic relations therefore suggest the function of language exposure and the knowledge of an individual word. Research by word association tasks usually analyzes the result of the responses to the prompt words regarding the relation 25. DOI:10.6814/NCCU202000123.

(38) between paradigmatic and syntagmatic shift (e.g., Entwisle, 1966; Nissen & Henriksen, 2006; Sheng, McGregor & Marian, 2006; Dell, Oppenheim & Kittredge, 2008). Words are classified as paradigmatic relation by distinguished between alternative semantic connections, in terms of one of the semantic relations of synonymy, antonymy, hyponymy or meronymy, but locating within the same syntactic slot, i.e., the same word class. In addition, the words which could be understood as a conventional and commonly accepted icon or symbol of the target word (e.g., ‘love’ and ‘heart’) can also be categorized. 政治大 syntactic relations and the semantic combination, i.e., with a different word class that the 立 as this relation. By contrast, words are classified into syntagmatic relation by connecting. ‧ 國. 學. target word has (e.g., ‘dog’ and ‘bark’). Also, words in the same syntactic category could still be classified as syntagmatic relation when a clear sequential connection happens (e.g.,. sit. y. Nat. and ‘car’).. ‧. compound relations as ‘accident’ and ‘car accident’, or argument relations as ‘accident’. n. al. er. io. Numerous studies targeting on paradigmatic and syntagmatic relations from the. i n U. v. responses of word association tasks are discussed. In Entwisle’s (1966) word association. Ch. engchi. tasks, the unequal proportion of responses between syntagmatic and paradigmatic relation was reported that a higher amount of responses with the paradigmatic relation to noun targets was found, while a negative correlation between verb and adverb associates showed for the syntactic relation. Nissen and Henriksen (2006) have demonstrated that target words with different syntactic categories tend to trigger a distinguishing proportion of paradigmatic and syntagmatic responses that nouns prime a higher proportion of paradigmatic responses than verbs and adjectives. Dell, Oppenheim and Kittredge (2008) investigated the function between the syntagmatic and paradigmatic interference in speech 26. DOI:10.6814/NCCU202000123.

(39) production. They found that during speech production, semantic information is dealt by paradigmatic relation while syntagmatic relation retrieves and predicts the words. On another hand, as we mentioned in the section 2.2.5, many studies have investigated phonological relation in term of syllable structure within word association by the analysis of speech production errors (e.g., Dell, 1988; Meyer, 1991; Adams & Meyer, 2000; Vitevitch, 2008; Rebei, Anderson & Dell, 2019). Most of them all found that syllable-initial consonants work in a distinct way when participating in phonological errors.. 政治大 segments within a syllable retrieved in order was reported that the entire first syllable of 立 For example, in Meyer’s (1991) speech production tasks with disyllable cued words,. ‧ 國. 學. the responses showed a stronger connection to the targets than just the onset. In Vitevitch’s (2008) demonstration of English phonological network, words with the same initial. ‧. consonant (e.g., ‘peach’ and ‘peat’), and minimal pairs including the same onset, rhyme,. sit. y. Nat. or coda (e.g., ‘peach’, ‘preach’ and ‘each’) showed direct connections to the target word. io. al. n. indirect connections.. er. while others with loose phonological similarity (e.g., ‘peach’ and ‘speak’) have only. i n U. Ch. v. i e nWord 2.4 Mandarin Mental Lexicon and g c hAssociation In this section, previous works on the Mandarin mental lexicon and word association tasks will be reviewed and investigated for approaching to the research target of the present study. Theories of Mandarin lexicon will be introduced (e.g., Huang & Hsieh, 2015; Hsieh, 2006, 2016), followed by some studies about the influences on the lexical access and organization in the mental lexicon (e.g., Zhang, 2010; Zhao, 2012; Yu, 2014; Wan & Ting, 2019). As for the reviews of the word association tasks, those in Beijing. 27. DOI:10.6814/NCCU202000123.

(40) Mandarin (e.g., Da & Lee, 2000; Zhao, 2013; Zhang & Chen, 2018) and in Taiwan Mandarin (e.g., Chen, Chen & Dell, 2002; Huang et al, 2009; Lee & Lee, 2011; Hu et al, 2017) will be discussed separately considering the language differences between two dialects of Mandarin, which affect the result of association6.. 2.4.1 Mandarin Mental Lexicon Since lexicon is viewed as the repository of conceptual structures of a language, understanding the lexicon can be an access to explore the mapping between concept and. 政治大. lexicon to represent meaning in the mental lexicon. Considering the heavy dependence of. 立. Mandarin lexicon on its morpho-semantics system resulting from its relationship between. ‧ 國. 學. orthography and language, few models have been proposed to explain how Mandarin lexicography reflects its conceptual organization in the mental lexicon (e.g., Huang &. ‧. Hsieh, 2015; Hsieh, 2006, 2016)7.. sit. y. Nat. io. er. Based on the idea of character-centered lexicon in Mandarin minds, Huang and Hsieh (2015) reviewed previous studies and proposed that Mandarin lexicon is encoded. n. al. Ch. i n U. v. the conventionalized semantic information by the orthographic system due to poverty of. engchi. morpho-syntactic markings contained inside. A pyramid structure model schematizes their. 6. Chinese language relates to the terms such as ‘Mandarin’, ‘Standard Chinese’, ‘Modern. Chinese’, and ‘Putonghua’, which all refer to the language based on Beijing speech (see Wang & Sun, 2015 for more). In the present study, only broad Beijing Mandarin and Taiwan Mandarin are involved, other specific dialects of Mandarin will be not identified thoroughly. 7. In Mandarin writing system, radicals can compose various characters, and characters can. constitute various words. Sometimes a single radical can be a character, and a single character can be a mono-syllabic word. 28. DOI:10.6814/NCCU202000123.

(41) viewpoints by a three-layered representation to present how characters can contain the core conceptual information in Mandarin as Figure 2-4-1.. 立. 政治大. ‧. ‧ 國. 學 sit. y. Nat. Figure 2-4-1. The pyramid structure model in Hsieh (2006, 2016).. n. al. er. io. The interactions between characters and morphemes in the character level constitute the. i n U. v. interface of concept and word, i.e., the character level reflects the conceptual organization. Ch. engchi. in the mental lexicon. Characters with similar conceptual information will group as a cluster together in the middle level, and construct the connections down to the word level, where the words are connected to each other through different semantic relations. This model offers the demonstration of the connections and representations from the semantic system to the orthographic system in Mandarin. Therefore, rich linguistic information encoded by its character system in Mandarin also offers its semantic networking some clues how a word relates to other words with semantic similarity. Besides the syntagmatic relations exist between words that occur 29. DOI:10.6814/NCCU202000123.

(42) together in syntactic structures, Huang and Hsieh (2015) further concluded several subcategories of the semantic paradigmatic relations in Mandarin that are also supported by his another project, Chinese WordNet 8 . For example, synonymy shares the same meaning (e.g., [cong2bu4]/[cong2wei4] ‘never’); near-synonymy has similar meanings (e.g., [bao1rong2]/[rong2ren3], ‘tolerate/endure’); antonymy contrasts meanings (e.g., [sheng1]/[si3] ‘life/death’); hypernymy/hyponymy defines a ‘type of’ relation (e.g., [ge1]/[min2yao2] ‘song/folk song’); holonymy/meronymy indicates a ‘part of’ relation of meaning (e.g., [shou3]/[shou3zhang3] ‘hand/palm’), and so on. A similar categorization is. 治政大lexical-semantic relationships also applied by Wan and Ting (2019) when investigating the 立 in Mandarin lexicon by lexical substitution speech errors. They classified performance. ‧ 國. 學. errors into the syntagmatic and paradigmatic relations, in which the latter one contains five. ‧. major subcategories involving coordinates, subsumatives, contrastive, near-synonymy, and association (e.g., [dongci]/[mingci] ‘verb/noun’; [xiaohai]/[ren] ‘child/person’; [leng]/[re]. y. Nat. er. io. sit. ‘cold/hot’; [wanxiao]/[xiaohua] ‘joke’; [gui]/[hao] ‘expensive/good’ respectively).. al. iv n C semantic h lexical errors with e n g c h i Uthe interaction. n. Another corpus study of lexical speech errors in Mandarin by Tang and Wan (2019) further categorized the. between both. phonological and sematic relations considered as Table 2-4-1.. 8. Starting from 2003, this project by Huang and Hsieh has been gathering research results for. been reconsidering and revising how senses should be disambiguated and represented in Mandarin. 30. DOI:10.6814/NCCU202000123.

(43) Table 2-4-1. Main types of speech errors identified in Wan’s corpus9 Error type. Semantic. Phonological. Semantic errors. +. -. Phonological errors. -. +. Mixed-type errors. +. +. Environmental errors. -. -. Since semantic activation spreads to the target lemma that can have either phonological or semantic correlates, errors should not be marked just by binary system as a semantic or. 治政 phonological error because both semantics and phonology 大 can have an effect on the 立 occurrence of speech errors (e.g., Wan and Tang, manuscript). Therefore, the connection ‧ 國. 學. between lexicon could be classify as either a semantic [+sem, -phon] (e.g.,. ‘OT’. ‘stairs/optimality. theory’). relationship,. or. ‧. [ming2cheng1]/[shu4zi4] ‘name/number’) or phonological [-sem, +phon] (e.g., ([lou2ti1]/ both. +phon]. (e.g.,. Nat. y. [+sem,. io. sit. [tong3ji4]/[zong3ji4] ‘statistics/ sum’) or neither [-sem, -phon], which generally relates to. er. context-induced relationship.. n. al. i n C 2.4.2 Mandarin Word Association hengchi U. v. Words have been supposed to be the core linguistic units for the investigation of languages, with the particular properties of Mandarin lexicon we discussed above, a number of studies are related to lexical organization in Mandarin by using word association tasks in Beijing Mandarin (e.g., Da & Lee, 2000; Zhao, 2013; Zhang & Chen, 2018) and. 9. The speech errors investigated in Tang and Wan (2019) were obtained from a corpus with over. 8,000 errors produced by over 100 native speakers of Taiwan Mandarin collected by Wan in a naturalistic setting (see Wan, 2007, and Tang & Wan, 2019, for more information). 31. DOI:10.6814/NCCU202000123.