Research map / Large language models / Language, safety and interpretability
Unsupervised Multi-Modal Neural Machine Translation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 image input during pretraining fine-tuning (ablation) | UMNMT · Multi30K 2016 test, En→Fr | BLEU: 37.20 -> 39.79 | Helped |
| Add image input during pretraining fine-tuning (ablation) | UMNMT · Multi30K 2016 test, Fr→En | BLEU: 38.51 -> 40.53 | Helped |
| Add image input during pretraining fine-tuning (ablation) | UMNMT · Multi30K 2016 test, En→De | BLEU: 20.97 -> 23.52 | Helped |
| Add image input during pretraining fine-tuning (ablation) | UMNMT · Multi30K 2016 test, De→En | BLEU: 25.00 -> 26.39 | Helped |
| Add image input when training from scratch (ablation) | UMNMT · Multi30K 2016 test, En→Fr | BLEU: 6.01 -> 9.40 | Helped |
| Add image input when training from scratch (ablation) | UMNMT · Multi30K 2016 test, Fr→En | BLEU: 6.75 -> 10.04 | Helped |
| Add image input when training from scratch (ablation) | UMNMT · Multi30K 2016 test, En→De | BLEU: 6.27 -> 8.85 | Helped |
| Add image input when training from scratch (ablation) | UMNMT · Multi30K 2016 test, De→En | BLEU: 6.81 -> 9.97 | Helped |
| Set the joint text-image attention weight λ₂ to 0 (ablation) | UMNMT · Multi30K | BLEU | No clear effect |
| Train with image input, then evaluate using text only (ablation) | UMNMT · Multi30K 2016 test, text-only inference | BLEU: 37.20 -> 39.44; 38.51 -> 40.30; 20.97 -> 23.18; 25.00 -> 25.47 | Helped |
Closest papers by meaning
- Unsupervised Neural Machine Translation (ICLR 2018 · similarity 0.90)
- UC2: Universal Cross-Lingual Cross-Modal Vision-and-Language Pre-Training (CVPR 2021 · similarity 0.87)
- CL2CM: Improving Cross-Lingual Cross-Modal Retrieval via Cross-Lingual Knowledge Transfer (AAAI 2024 · similarity 0.86)
- UniGen: Enhanced Training & Test-Time Strategies for Unified Multimodal Understanding and Generation (NeurIPS 2025 · similarity 0.85)
- Split to Merge: Unifying Separated Modalities for Unsupervised Domain Adaptation (CVPR 2024 · similarity 0.85)
- Towards Identifiable Unsupervised Domain Translation: A Diversified Distribution Matching Approach (ICLR 2024 · 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.