Sequence-to-structure generation-based event extraction uniformly models all tasks in a single model and universally predicts different labels. From there, the model directly generates all arguments and their roles. An encoder-decoder structure is typically used.