Self-Training with Purpose Preserving Augmentation Improves Few-shot Generative Dialogue State Tracking
This work addresses labeling efficiency for dialogue systems, though it appears incremental as it builds on existing self-training and augmentation methods.
The paper tackles the problem of high labeling costs in dialogue state tracking by proposing a self-training framework with Purpose Preserving Augmentation for few-shot generative DST, achieving approximately 4% improvement in few-shot 10% performance on MultiWOZ 2.1 and 8.34% enhancement in slot-recall for unseen values compared to baseline.
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% performance by approximately 4% on MultiWOZ 2.1 and enhances the slot-recall 8.34% for unseen values compared to baseline.