DialogBERT: Discourse-Aware Response Generation via Learning to Recover and Rank Utterances
This work provides an incremental improvement in response generation coherence for dialogue systems by better modeling discourse structure.
This paper introduces DialogBERT, a conversational response generation model that addresses the limitation of existing pre-trained language models by incorporating discourse-level coherence. It achieves this through a hierarchical Transformer and two novel training objectives, outperforming baselines like BART and DialoGPT on three multi-turn conversation datasets.
Recent advances in pre-trained language models have significantly improved neural response generation. However, existing methods usually view the dialogue context as a linear sequence of tokens and learn to generate the next word through token-level self-attention. Such token-level encoding hinders the exploration of discourse-level coherence among utterances. This paper presents DialogBERT, a novel conversational response generation model that enhances previous PLM-based dialogue models. DialogBERT employs a hierarchical Transformer architecture. To efficiently capture the discourse-level coherence among utterances, we propose two training objectives, including masked utterance regression and distributed utterance order ranking in analogy to the original BERT training. Experiments on three multi-turn conversation datasets show that our approach remarkably outperforms the baselines, such as BART and DialoGPT, in terms of quantitative evaluation. The human evaluation suggests that DialogBERT generates more coherent, informative, and human-like responses than the baselines with significant margins.