Self-Training with Purpose Preserving Augmentation Improves Few-shot Generative Dialogue State Tracking
In dialogue state tracking (DST), labeling the dataset involves considerable human labor. We propose a new self-training framework for few-shot generative DST that utilize unlabeled data. Our self-training method iteratively improves the model by pseudo labeling and employs Purpose Preserving Augmentation (PPAug) to prevent overfitting. We increaese the few-shot 10 approximately 4 values compared to baseline.
READ FULL TEXT