Research map / Image generation / Editing and style transfer
Stylized Dialogue Generation with Multi-Pass Dual Learning
Jinpeng Li, Yingce Xia, Rui Yan, Hongda Sun, Dongyan Zhao, Tie-Yan Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add quality discriminators that weight pseudo data (ablation) | MPDL · TCFC avg of S0 and S1 responses | BLEU-1: 10.9 -> 12.0 | Improved |
| add quality discriminators that weight pseudo data (ablation) | MPDL · TCFC avg of S0 and S1 responses | Distinct-2: 52.8 -> 50.1 | Hurt |
| add quality discriminators that weight pseudo data (ablation) | MPDL · TCFC avg of S0 and S1 responses | BERT style accuracy: 83.1 -> 84.4 | Improved |
| add multi-pass dual loss L_tra closed loop (ablation) | MPDL · TCFC avg of S0 and S1 responses | BLEU-1: 9.89 -> 12.0 | Improved |
| add multi-pass dual loss L_tra closed loop (ablation) | MPDL · TCFC avg of S0 and S1 responses | Distinct-2: 45.1 -> 50.1 | Improved |
| initialize from pretrained DialoGPT (ablation) | MPDL · TCFC avg of S0 and S1 responses | BLEU-2: 0.87 -> 1.38 | Improved |
| initialize from pretrained DialoGPT (ablation) | MPDL · TCFC avg of S0 and S1 responses | Distinct-2: 33.6 -> 50.1 | Improved |
| initialize from pretrained DialoGPT (ablation) | MPDL · TCFC avg of S0 and S1 responses | BERT style accuracy: 84.0 -> 84.4 | Mixed |
| MPDL vs SRJT on formal response generation | SRJT · TCFC formal response (S1) | BLEU-1: 15.1 -> 16.5 | Improved |
| MPDL vs SRJT on Shakespearean response generation | SRJT · SDGC Shakespearean response (S1) | BLEU-1: 12.2 -> 13.9 | Improved |
| MPDL vs SRJT on Shakespearean response generation | SRJT · SDGC Shakespearean response (S1) | BERT style accuracy: 65.0 -> 64.3 | Hurt |
Closest papers by meaning
- Stylized Dialogue Response Generation Using Stylized Unpaired Texts (AAAI 2021 · similarity 0.91)
- Mitigating Negative Style Transfer in Hybrid Dialogue System (AAAI 2023 · similarity 0.87)
- From Natural Alignment to Conditional Controllability in Multimodal Dialogue (ICLR 2026 · similarity 0.87)
- A Dataset for Low-Resource Stylized Sequence-to-Sequence Generation (AAAI 2020 · similarity 0.86)
- NTPP: Generative Speech Language Modeling for Dual-Channel Spoken Dialogue via Next-Token-Pair Prediction (ICML 2025 · similarity 0.84)
- CoVoMix2: Advancing Zero-Shot Dialogue Generation with Fully Non-Autoregressive Flow Matching (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
