以語料庫為本之半自動化英語母語者及學習者動名詞搭配詞比較

Abstract

在第二語言學習者的搭配詞研究中，動詞與名詞的搭配是最常見且重要的層面。然而，在近數十年的發現及研究方法上，較具效率且系統化的語料庫資料提取及分析技術尚未被充分討論及實行。本研究利用The Sketch Engine (SKE)此一線上平臺，期望能以半自動化之方式，檢視中文為母語之英語學習者於動詞－名詞搭配詞使用上之錯誤。透過對此線上平臺之實驗性測試，能達成更有效率及可靠之方法，並找出更具研究意義的結果。 本研究針對三個研究問題。他們分別為 (1) 在The Sketch Engine此線上平臺上實驗性測試Sketch Diff，一半自動化語料分析功能，來比較英語母語與學習者間之動詞－名詞搭配詞使用、(2) 探究中文為母語之英語學習者於動詞－名詞搭配詞使用上之錯誤類別、及 (3) 檢視這些錯誤背後之可能形成原因。 首先，利用SKE上的Corpus Creating功能，將英國國家語料庫 (BNC)，即英語母語者語料庫，及由四個中文為母語之英語學習者語料庫(CLEC、SWECCL、大考中心、臺灣英語學習者語料庫) 組合而成之一大型英語學習者語料庫，上傳至SKE平臺。接下來，在此一大型英語學習者語料庫中，提取出最常用的690 個英語名詞。最後，Sketch-Diff，一項原本在SKE平臺上用來比較同義字的功能，被實驗性地操作，用來比較前述690個中文母語之英語學習者最常用的英語名詞，他們分別於英國國家語料庫(BNC)及另一大型英語學習者語料庫中，常見的動詞搭配。 結果顯示，共計發現134項、2841筆的動詞－名詞搭配詞錯誤。關於錯誤類別，63項 (832筆) 為動詞錯誤、43項 (1502筆) 為介詞或動名詞性質之動詞錯誤、28項 (507筆) 為名詞錯誤。 錯誤成因部分，具影響力的來源因子排名為 a) 母語負面影響 (75項、1376筆)、b) 過度延伸 (23項、213筆)、c) 同義字錯誤 (21項、449筆)、d) 錯誤類比 (9項、728筆)、e) 拼字相似錯誤 (4項、64筆)、f) 輕動詞錯誤 (1項、6筆)、及g) 錯誤造義 (1項、5筆)。 依據上述結果，有下列幾項觀察。第一，中文，也就是英語學習者的母語，仍有壓倒性的負面影響力介入動詞－名詞搭配詞的學習歷程。第二，介詞或動名詞性質之動詞錯誤最為普遍，的確需要更多關注及應對策略。最後，可能由過度延伸及同義字錯誤引發的錯誤，指出了英語學習者在搭配詞上的學習盲點，也就是亟需更多的英語接受量及情境化式的練習。Among the studies of ESL learners' collocations, the aspect of Verb-Noun combinations has been the most popular and important one. Yet, with the findings discovered and research methods developed these decades, a more efficient and systematic manner of corpus data extraction and analysis has not been thoroughly discussed and practiced. This study, adopting the online platform The Sketch Engine (SKE), aims to examine Chinese ESL learners' Verb-Noun miscollocations with a semi-automated method. Through an experimental utilization of the online interface, a more streamlined and reliable approach is expected to be realized, and more interesting results are revealed through this way. Three research questions were targeted in this study. They are (1) to experiment Sketch Diff, a semi-automated corpus-based function of The Sketch Engine, an online platform, to compare the Verb-Noun collocations between native and non-native English speakers, (2) to explore the types of Chinese ESL learners' Vern-Noun miscollocations, and (3) to inspect the probable causes of learners' Verb-Noun miscollocations. First, with the Corpus Creating function on the SKE, a native speaker corpus, the British National Corpus (BNC), and a Chinese ESL learner corpus, merged by CLEC, SWECCL, JCEE Testees Corpus, and Taiwanese Learner Corpus, were uploaded unto the platform. Then, a 690-word-list of the most frequent nouns in the Chinese ESL learner corpus was generated. Finally, a Sketch-Diff function, originally for synonym comparison on the SKE, was alternatively manipulated in this study for the comparison of the 690 nouns' verb collocates respectively in the BNC and the Chinese ESL learner corpus. A total of 134 types of Chinese ESL learners' Verb-Noun miscollocations were found, with 2841 tokens overall. As for the general types of learners' V-N miscollocation types, 63 sorts (832 tokens) belonged to the deviant use of verbs, 43 types (1502 tokens) were categorized under the misuse of prepositional and phrasal verbs, and 28 types (507 tokens) were grouped by the misuse of nouns. In terms of contributing error sources, the influential factors are a) negative L1 transfer (75 types, 1376 tokens), b) overgeneralization (23 types, 213 tokens), c) erroneous use of synonyms (21 types, 449 tokens), d) false analogy (9 types, 728 tokens), e) approximation (4 types, 64 tokens), f) erroneous delexical verbs (1 type, 6 tokens), and g) erroneous coinage (1 type, 5 tokens). According to the results above, several observations were proposed. First, the overwhelming influence of Chinese, the target ESL learners' L1, still exerts a great impact on learners' acquisition of V-N collocations. Second, as the most found type of error, the misuse of prepositional and phrasal verbs indeed requires more attention and coping strategies. Finally, miscollocations possibly caused by overgeneralization and erroneous use of synonyms have pointed out the blind spots of learners' collocation learning, which is in serious need of more L2 input and contextualized exercises.