以ChatGPT的搭配詞詞典與《麥克米倫英語搭配詞詞典》之比較研究

Abstract

隨著人工智慧在教育領域中的應用日益普及，ChatGPT等工具逐漸被用來輔助語言學習。本研究旨在探討ChatGPT作為搭配詞學習資源的可行性，並與廣泛使用的學習型詞典《麥克米倫英語搭配詞典》（Macmillan English Collocations Dictionary,MCD）進行比較。研究聚焦於兩種語法結構：形容詞＋名詞（Adj+N）與動詞＋名詞（V+N），並探討三個研究問題：(1)ChatGPT所產生的搭配詞與MCD所列出的搭配詞之間有多少重疊？(2)ChatGPT所產生而MCD未收錄的搭配詞，是否能在外部參考資源中（如OZDIC與COCA）找到佐證？(3)ChatGPT與MCD在語意分類上有何差異？本研究選取MCD中的四十個名詞作為搭配詞詞頭，依據其詞典頁面長度將詞頭分類為四個搭配豐富度層級。將詞頭放進ChatGPT使用標準化提示詞（prompt）從中產出形容詞+名詞（Adj+N）與動詞+名詞（V+N）搭配詞，並與MCD所列內容進行比較。研究結果顯示，ChatGPT產生的搭配詞有超過五成與MCD重疊，其中形容詞＋名詞類型的重疊率特別高。另外，在ChatGPT產生而MCD未收錄的搭配詞中，有79%的形容詞＋名詞搭配詞與41.26%的動詞＋名詞搭配詞可在OZDIC或COCA中找到，顯示許多未被MCD收錄的搭配詞仍具實際使用性。最後，語意分類的分析顯示，ChatGPT傾向以「情感語氣」、「程度強弱」、「類型」等語意概念分類，而MCD則依據語用功能進行分類。結論，雖然ChatGPT尚無法完全取代傳統學習型詞典，但其在產出常見搭配詞與語意分類上的靈活性，展現了作為輔助教學工具的潛力。搭配傳統詞典使用時，ChatGPT可提升英語學習者對搭配詞的理解與應用能力，特別適用於英語為外語（EFL）之學習環境。As artificial intelligence continues to evolve, tools like ChatGPT are increasingly used in educational settings. This study investigates the potential of ChatGPT as a collocation dictionary by comparing its output with the Macmillan English Collocations Dictionary (MCD), a well-established learner resource. Focusing on adjective + noun (Adj + N) and verb + noun (V + N) patterns, the study addressed three research questions: (1) To what extent do ChatGPT-generated collocations overlap with those listed in the MCD? (2) To what extent are collocations generated exclusively by ChatGPT supported by external reference sources such as the Oxford Collocations Dictionary (OZDIC) and the Corpus of Contemporary American English (COCA)? (3) How do the ChatGPT-based collocation dictionary and the MCD differ in semantic groupings of collocates? Forty head nouns were selected from the MCD and classified into four levels of collocational richness. Using the same prompts for each noun, collocations were generated from ChatGPT and compared with those listed in the MCD. For RQ1, results showed that ChatGPT covered over 50% of the MCD-listed collocations across both patterns, with especially high overlap in the Adj + N pattern. For RQ2, 79% of Adj + N and 41.26% of V + N collocations generated exclusively by ChatGPT were found in OZDIC or COCA, indicating that many of these collocations are used in real English even if they aren’t listed in the MCD. Lastly, for RQ3, a qualitative analysis revealed that ChatGPT grouped collocates based on broad semantic themes (e.g., “emotional tone,” “intensity,” or “type”), whereas the MCD provided more specific, function-based groupings tailored for learner use. Overall, while ChatGPT cannot fully replace traditional learner dictionaries, it shows strong potential as a supplementary tool for learning collocations. It can generate widely used word combinations and organize them in meaningful ways, making it a useful complement to traditional learner dictionaries in EFL contexts.