基於語料庫分析之國中生社會情緒學習語意主題與性別差異研究

A Corpus-Based Analysis of Semantic Themes and Gender Differences in Junior High Students’ Social and Emotional Learning

在全球教育日益重視社會情緒學習（Social-Emotional Learning,SEL）與臺灣十二年國民基本教育強調核心素養的政策脈絡下，學生在真實課程情境中如何透過語言表達情緒、協商關係並展現社會情緒能力，仍缺乏系統性的語言證據。本研究以CASEL五大核心能力架構及「語言作為心理狀態外顯線索」之觀點為理論基礎，結合語料庫語言學，探討國中生在SDGs跨領域統整課程後之反思語料所呈現的語意主題、情緒傾向與性別差異。研究對象為臺灣北部某國中147位九年級學生。研究蒐集課程結束後之書面反思294筆，經篩選後納入239筆有效語料。分析程序結合CKIP斷詞、TF-IDF向量化、LDA主題建模、自編中文情緒詞庫與多項邏輯斯迴歸分析，以檢驗主題分布、情緒傾向及其與性別之關聯。主題數驗證結果顯示K=4具有較佳語意一致性與模型適配性。研究結果顯示，學生反思語料可歸納為四個語意主題：「挑戰經驗與舞臺焦慮」（46.86%）、「團隊合作與溝通互動」（28.45%）、「負向情緒與情緒調節歷程」（3.35%）與「正向情緒與學習動機提升」（21.34%）。整體情緒傾向略偏負向，但不同主題之情緒表現具有明顯差異。女生在情緒詞彙使用頻率與多樣性上均高於男生，語言表達較為細緻；男生則較傾向採取簡潔且行動導向的表述。進一步分析顯示，情緒傾向可顯著預測主題二、主題三與主題四相對於主題一之出現傾向，而性別僅在團隊合作主題上呈現接近顯著的差異。本研究的原創性，在於將SEL理論、語料庫語言學、主題建模與情緒傾向分析整合為一套可觀察學生社會情緒歷程的語言分析框架，補足過去研究偏重問卷量表與整體課程成效、較少直接處理學生自然語言資料之不足。研究結果顯示，學生反思語言可作為觀察其情緒覺察、人際協商與自我調節歷程的重要指標。實務上，教師可透過結構化反思寫作、情緒轉折語句辨識與語料分析工具，及早掌握學生在焦慮、合作與動機轉化中的需求，並據以調整SEL教學設計與差異化情緒支持策略。

Against the growing policy emphasis on social and emotional learning (SEL) in global education and Taiwan’s 12-year basic education curriculum, relatively little research has provided systematic linguistic evidence on how students use language to express emotions, negotiate relationships, and demonstrate social and emotional competencies in authentic classroom contexts. Grounded in CASEL’s five-competency framework and the perspective of “language as an explicit cue of psychological states,” this study investigated the semantic themes, sentiment orientations, and gender differences embedded in junior high school students’ written reflections following an SDGs-themed interdisciplinary unit. Participants were 147 ninth-grade students from a junior high school in northern Taiwan. A total of 294 written reflections were collected after the course, of which 239 valid texts were retained for analysis after screening. The analytical procedures combined CKIP word segmentation, TF-IDF vectorization, Latent Dirichlet Allocation (LDA) topic modeling, a researcher-developed Chinese emotion lexicon, and multinomial logistic regression to examine the associations among topic distribution, sentiment orientation, and gender. Validation of the number of topics indicated that a four-topic model (K=4) yielded optimal semantic coherence and model fit. Four themes were identified in the student corpus: “challenge experiences and performance anxiety” (46.86%), “teamwork and interactive communication” (28.45%), “negative emotions and emotion regulation” (3.35%), and“positive emotions and enhanced learning motivation” (21.34%). Overall sentiment in the corpus was slightly negative, although meaningful differences emerged across themes. Female students used emotional vocabulary more frequently and diversely than male students, demonstrating more nuanced linguistic expressions; male students, by contrast, tended to adopt more concise and action-oriented expressions. Further analyses revealed that sentiment orientation significantly predicted the occurrence of themes two, three, and four relative to theme one, whereas gender only demonstrated a marginally significant effect on the teamwork theme. The originality of this study lies in integrating SEL theory, corpus linguistics, topic modeling, and sentiment analysis into a linguistic framework for observing students’ SEL processes, thereby addressing the lack of direct engagement with students’ natural language data in previous research, which heavily relied on self-report scales and holistic course evaluations. The findings demonstrate that students’ reflective language serves as a crucial indicator for observing their emotional awareness, interpersonal negotiation, and self-regulation processes. Practically, teachers can utilize structured reflective writing, identification of emotional transitions, and corpus analysis tools to promptly grasp students’ needs amidst anxiety, collaboration, and motivational shifts, thereby adjusting SEL instructional designs and differentiating emotional support strategies.