This dataset contains synthetic story quintuplets designed to support research on disentangling core narrative aspects in a latent embedding space (Higgins et al., 2017).Each record includes an anchor story, a stylistic variant that preserves the same core aspects, and three aspect-perturbed variants where exactly one core aspect differs: theme, course, or outcome.
The dataset was created for training… See the full description on the dataset page: https://huggingface.co/datasets/ufal/NarrativeAspect-EntangledSynthetic.
