Research map / Architectures and efficiency / Pruning and efficient tuning
A Multi-Task Multi-Stage Transitional Training Framework for Neural Chat Translation
Chulun Zhou, Yunlong Liang, Fandong Meng, Jie Zhou, Jinan Xu, Hongji Wang, Min Zhang, Jinsong Su
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 MMT training instead of two-stage Flat-NCT+FT | Flat-NCT (Base) · BConTrasT En->De test | BLEU: 58.15 -> 59.33 | Improved |
| add MMT training instead of two-stage Flat-NCT+FT | Flat-NCT (Base) · BMELD En->Zh test | BLEU: 24.76 -> 27.43 | Improved |
| add MMT training instead of two-stage Flat-NCT+FT | Flat-NCT (Big) · BMELD En->Zh test | BLEU: 26.45 -> 28.62 | Improved |
| use context-aware models instead of sentence-level Transformer+FT | context-aware NMT with two-stage training · BConTrasT En->De test | BLEU: 58.43 vs 58.15 | No clear effect |
| add monolingual dialogues at stages 2 and 3 (ablation) | Flat-NCT+MMT · BConTrasT En->De valid | BLEU: 59.74 -> 60.86 | Improved |
| add source and target monolingual dialogues at stage 3 (ablation) | Flat-NCT+MMT · BConTrasT En->De valid | BLEU: 60.51 -> 60.86 | Improved |
| add both UD and SD tasks (ablation) | Flat-NCT+MMT · BConTrasT En->De valid | BLEU: 59.79 -> 60.86 | Improved |
| add UD task on monolingual dialogues (ablation) | Flat-NCT+MMT · BConTrasT En->De valid | BLEU: 59.96 -> 60.86 | Improved |
| add SD task on monolingual dialogues (ablation) | Flat-NCT+MMT · BConTrasT En->De valid | BLEU: 60.13 -> 60.86 | Improved |
| add UD task on monolingual dialogues (ablation) | Flat-NCT+MMT · BConTrasT De->En valid | BLEU: 60.41 vs 60.94 | Mixed |
| use authentic instead of pseudo Chinese monolingual dialogues (ablation) | Flat-NCT+MMT w/o SD (Base) · BMELD En->Zh test | BLEU: 27.35 -> 27.80 | Improved |
| use authentic instead of pseudo Chinese monolingual dialogues (ablation) | Flat-NCT+MMT w/o SD (Big) · BMELD Zh->En test | BLEU: 22.87 -> 23.36 | Improved |
Closest papers by meaning
- Attention-Informed Mixed-Language Training for Zero-Shot Cross-Lingual Task-Oriented Dialogue Systems (AAAI 2020 · similarity 0.88)
- Evaluating the Cross-Lingual Effectiveness of Massively Multilingual Neural Machine Translation (AAAI 2020 · similarity 0.87)
- Translating with Bilingual Topic Knowledge for Neural Machine Translation (AAAI 2019 · similarity 0.87)
- BiToD: A Bilingual Multi-Domain Dataset For Task-Oriented Dialogue Modeling (NeurIPS 2021 · similarity 0.87)
- Acquiring Knowledge from Pre-Trained Model to Neural Machine Translation (AAAI 2020 · similarity 0.87)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.85)
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.
