prefixTuning Dialogue State Tracking Module

Dialogue State Tracking plays an important role in improving the overall performance of Task-Oriented Dialogue System.

In this task, (slot, value) pairs as dialogue states need to be predicted at each turn of the dialogue by the model. To address this problem, we proposed a sequence-to-sequence model based on conditional generation of the dialogue states. To steer the language model to this task efficiently, we apply continuous prompt tuning to guide the generation process and during the encoding process, only parameters relevant to prompt prefix will be optimized.