We developed Litteraturmaskinen, a graphical annotation and exploration interface that enables students to collaborate on labeling sentiment in literary passages, comparing their decisions with model predictions, and justifying their interpretations. We deployed the system in two educational settings: A university module on computational literary studies and regular teaching by two first-language high school teachers. Based on observations, collected teaching plans, and interviews, we find that tensions between epistemic and academic traditions are both a barrier for integration and a productive entry point for literary reflection and argumentation. We conclude with recommendations for integrating NLP into literature and first-language curricula.