State Value Generation with Prompt Learning and Self-Training for Low-Resource Dialogue State Tracking
This addresses the challenge of capturing context-dependent values in low-resource DST, offering a novel approach that is competitive with larger models, though it is incremental in improving existing decomposition methods.
The paper tackles the problem of generating state values in low-resource dialogue state tracking by proposing a framework that decomposes the task into value generation and slot generation, using self-training with an estimator for pseudo-label selection, achieving state-of-the-art performance on MultiWOZ 2.1 with under 1 billion parameters at data ratios of 5%, 10%, and 25%.
Recently, low-resource dialogue state tracking (DST) has received increasing attention. First obtaining state values then based on values to generate slot types has made great progress in this task. However, obtaining state values is still an under-studied problem. Existing extraction-based approaches cannot capture values that require the understanding of context and are not generalizable either. To address these issues, we propose a novel State VAlue Generation based framework (SVAG), decomposing DST into state value generation and domain slot generation. Specifically, we propose to generate state values and use self-training to further improve state value generation. Moreover, we design an estimator aiming at detecting incomplete generation and incorrect generation for pseudo-labeled data selection during self-training. Experimental results on the MultiWOZ 2.1 dataset show that our method which has only less than 1 billion parameters achieves state-of-the-art performance under the data ratio settings of 5%, 10%, and 25% when limited to models under 100 billion parameters. Compared to models with more than 100 billion parameters, SVAG still reaches competitive results.