Remembering sentences not presented in lists

False memories are typically studied with lists of isolated words that are thematically related to unstudied critical items, a paradigm known as the Deese–Roediger–McDermott (DRM) task. Yet most real-world remembering concerns events expressed at a higher level of complexity. Here, we adapted the DRM to better describe such complexity by employing sentence materials. Leveraging sentence embeddings, we arranged recognition probes along a graded continuum of semantic similarity to each list’s theme. Participants incidentally encoded sentences and then performed a recognition task. Results showed that semantic similarity between new sentences and encoded one robustly predicted false recognition. Participants’ reliance on semantic information in this newly constructed task was then compared to the one observed in a classical word-based DRM task and showed a small-to-medium correlation (r = .39). Taken together, in showing that DRM-like false recognition reliably emerges for event-like sentence materials and that false alarms scale with graded semantic overlap, these findings advance a quantitative, ecologically oriented account of false memory.