TOD-BERT: Pre-trained Natural Language Understanding for Task-Oriented Dialogue
This work addresses the data scarcity and performance gap for task-oriented dialogue systems, which is crucial for improving conversational AI in practical domains.
The authors tackled the problem of pre-trained language models being less effective for task-oriented dialogue due to linguistic differences, by developing TOD-BERT, which outperforms BERT on four downstream applications and shows stronger few-shot ability to mitigate data scarcity.
The underlying difference of linguistic patterns between general text and task-oriented dialogue makes existing pre-trained language models less useful in practice. In this work, we unify nine human-human and multi-turn task-oriented dialogue datasets for language modeling. To better model dialogue behavior during pre-training, we incorporate user and system tokens into the masked language modeling. We propose a contrastive objective function to simulate the response selection task. Our pre-trained task-oriented dialogue BERT (TOD-BERT) outperforms strong baselines like BERT on four downstream task-oriented dialogue applications, including intention recognition, dialogue state tracking, dialogue act prediction, and response selection. We also show that TOD-BERT has a stronger few-shot ability that can mitigate the data scarcity problem for task-oriented dialogue.