Categorizing Crossmodal Interaction Between Web Search and GenAI Chat in Translation: Evidence from Student Users

This study examines crossmodal interaction between web search and GenAI chat during translation. Thirty-six participants completed a translation task with access to both modalities. Queries and prompts were coded using the Community of Inquiry framework and analyzed using lag sequential analysis and epistemic network analysis; translation quality was assessed with COMET and human ratings. Based on interaction-time shares, participants were classified as Web-dominant, GenAI-dominant, or Balanced. Web-dominant users followed an augmentative pattern, verifying via search with minimal social presence; GenAI-dominant users displayed an assimilative pattern, converging on GenAI with active feedback; while Balanced users demonstrated an integrative dual-hub workflow, circulating between web verification and GenAI synthesis. Overall quality and task time did not differ significantly across categories, though Balanced users showed the highest cultural-adequacy ratings. Exploratory PCA indicated category-specific quality-related behaviors, suggesting that performance depends more on how modalities are coordinated than on modality dominance itself.