The Incremental Named Entity Recognition (INER) task aims to update a model to extract entities from an expanding set of entity type candidates due to concerns related to data privacy and scarcity. However, conventional sequence labeling approaches to INER often suffer from the catastrophic forgetting problem, which leads to the degradation of the model’s performance on previously encountered entity types. In this paper, we formalize INER as a unified seq2seq generation task and propose a parameter-efficient dynamic prefix method. By employing the dynamic prefix as a task instructor to guide the generative model, our approach can preserve task-invariant knowledge while adapting to new entities with minimal parameter updates, making it particularly effective in low-resource scenarios. Additionally, we introduce a generative label augmentation strategy with dual optimization objectives including a self-entropy loss and a task-aware similarity loss to enable optimal balance between stability and plasticity. Empirical experiments on NER benchmarks demonstrate the effectiveness of our proposed method in addressing the challenges associated with INER.

Diffusion models have achieved state-of-the-art synthesis quality on both visual and audio tasks, and recent works further adapt them to textual data by diffusing on the embedding space. In this paper, we conduct systematic studies of the optimization challenges encountered with both the embedding space and the denoising model, which have not been carefully explored. Firstly, the data distribution is learnable for embeddings, which may lead to the collapse of the embedding space and unstable training. To alleviate this problem, we propose a new objective called the anchor loss which is more efficient than previous methods. Secondly, we find the noise levels of conventional schedules are insufficient for training a desirable denoising model while introducing varying degrees of degeneration in consequence. To address this challenge, we propose a novel framework called noise rescaling. Based on the above analysis, we propose Difformer, an embedding diffusion model based on Transformer. Experiments on varieties of seminal text generation tasks show the effectiveness of the proposed methods and the superiority of Difformer over previous state-of-the-art embedding diffusion baselines.