Abstract
INTRODUCTION: Despite growing global interest in the emotional dimensions of academic writing, Romanian academic discourse remains underexplored, particularly in multilingual contexts. This study addresses this gap by analyzing a bilingual corpus of texts written in Romanian (L1) and English (L2) across various disciplines and genres. It aims to uncover emotional dimensions conveyed through linguistic markers, exploring how language, culture, and academic context shape students' writing styles. Romania's historical and social emphasis on formality, hierarchy, and indirectness in communication serves as a backdrop for examining these dynamics. METHOD: A corpus-based approach was adopted, utilizing the Linguistic Inquiry and Word Count 2015 (LIWC2015) tool to analyze linguistic and emotional markers. The bilingual ROGER corpus, containing texts from nine Romanian universities spanning multiple disciplines and genres, served as the dataset. Advanced data analysis techniques included supervised machine learning for language classification, network analysis to explore interactions among linguistic features, and cluster analysis to detect discipline- and genre-specific linguistic patterns. RESULTS: The findings reveal distinct emotional patterns between Romanian and English academic writing. Romanian texts exhibit a higher degree of formality and indirectness, while English texts reflect greater assertiveness and personal engagement. Additionally, the Romanian corpus demonstrates less linguistic cohesion and a broader range of writing styles. Genre- and discipline-specific trends also emerge, with English coursework and analytical writing, predominantly from social sciences, displaying more personal and emotional expression than research-focused texts. In contrast, the Romanian corpus, characterized by a third cluster, presents less clear-cut patterns: humanities texts span both emotionally expressive and neutral tones, while research and academic papers frequently exhibit an achievement-oriented or entrepreneurial style, though a significant subset also reflects a highly disengaged profile. DISCUSSION: By integrating machine learning, network analysis, and automatic language analysis, this study offers a novel perspective on how language, genre, and discipline-specific conventions shape emotional expression in academic writing. The results suggest that the Romanian students' emotional personas in academic writing are influenced by all these factors, potentially shaped by the cultural norms of the second language, providing insights for teaching academic writing in multilingual settings.