解碼情緒： 以語料庫為本方法探索華語流行歌詞中的情緒概念

Abstract

本研究旨在分析華語流行音樂歌詞中，與情緒有關之概念，以及這些概念如何以概念隱喻和概念轉喻之方式在歌詞文本中將抽象的情緒概念具體化。我們建立了從2013年至2023年共73,358首歌曲的華語流行歌詞語料庫，並利用語料庫語言學中搭配詞概念、詞向量模型和網絡分析等技術，將和七大基本情緒 (快樂、悲傷、恐懼、驚訝、憤怒、厭惡、以及輕視) 有關的名詞搭配詞取出，建立每個情緒的搭配詞網絡。網絡中的節點(nodes)為各情緒的顯著名詞搭配詞，邊(edges)則為搭配詞之間的語意相似值。我們透過詞向量計算每個搭配詞之間的語意相似值，並以網絡分析中社群偵測 (community detection) 彙整出與情緒相關之語義場，進而匯聚為更廣義的語意類別，分析與情緒相關之概念。最後，我們也隨機挑選出每個語義場中的搭配詞索引，進一步觀察、分析其在歌詞文本中可能出現的概念隱喻、轉喻，以及相關概念。根據社群偵測的結果顯示，每種情緒都和不同的媒介產生關聯。和快樂有關的概念數量最多，其次依序為悲傷、恐懼、驚訝、憤怒、厭惡、以及輕視。其中，我們發現快樂（JOY）和悲傷（SADNESS）在不同具體程度的語義類別中有共同的相關概念：抽象概念傾向於和快樂及悲傷有關，但具體概念則顯現出不同情緒偏好。儘管和恐懼、驚訝、憤怒等情緒相關的概念數量較少，本研究分析的七大情緒皆與「人」（HUMAN）、時間（TIME）相關的概念有關聯，突顯了情緒表達短暫且人本中心的性質。而透過分析隨機選取的搭配詞索引，我們發現華語流行歌詞運用了概念隱喻和概念轉喻來理解情緒，同時，單純相關的概念則反映出歌詞文本中經常出現的主題。透過語料庫為本的方法，我們發現許多過往文獻未提及的隱喻和轉喻。總的來說，本研究結果顯示，華語流行歌詞中七大基礎情緒和不同概念有關，且歌詞文本常以概念隱喻和轉喻的方式將抽象情緒概念具體化以助理解。而利用搭配詞和網絡分析，本研究採用之語料庫為本、資料導向的研究方法能有效幫助我們找出華語流行歌詞背後和情緒相關的語意概念和群體。This current study aims to investigate emotion-associated concepts and the conceptual mappings (e.g., conceptual metaphor and conceptual metonymy) used to understand emotion in Mandarin pop lyrics. We established a corpus of 73,358 song lyrics in Mandarin from 2013 to 2023 and examined seven basic emotions in Ekman (1992a, 1992b)—JOY, SADNESS, ANGER, FEAR, SURPRISE, DISGUST, and CONTEMPT—via collocation extraction and network science. We extracted significant noun collocates of each emotion and constructed lexical networks with collocates as nodes and pair-wise cosine similarity as edges. For each emotion network, we performed community detection to identify semantic coherent groups, categorized them into emerging semantic fields and further collapsed them into broader semantic categories. Finally, we analyzed a random sample of concordance lines in each major community to explore potential conceptual mappings (i.e., conceptual metaphor and conceptual metonymy) and pure associated concepts.The results revealed that each emotion was associated with different mediums. JOY accounted for the largest number of associated concepts, followed by SADNESS, FEAR, SURPRISE, and ANGER/DISGUST/CONTEMPT. We found that JOY and SADNESS shared semantic categories in terms of different extent of concreteness: abstract concepts tended to be shared by JOY and SADNESS, whereas concrete concepts showed distinct preferences. Although FEAR, SURPRISE, and ANGER/DISGUST/CONTEMPT were associated with fewerconcepts, they shared HUMAN and TIME related concepts with JOY and SADNESS, highlighting the human centered and transient characteristics of emotion expressions. The analysis of randomly sampled concordance lines suggested that Mandarin pop lyrics employed both conceptual metaphors and conceptual metonymies for the comprehension of emotions, along with purely associated concepts that reflected topics frequently discussed with emotions. Notably, we identified several novel metaphors and metonymies compared to prior studies. We conclude that pop song lyrics featured distinct concepts associated with basic emotions, leveraging metaphor and metonymy for conceptualization. Using collocation and network analysis, corpus-based and data-driven methods help uncover the hidden grouping and association with emotions in pop song lyrics.