Research map / Reinforcement learning / Visual and representation RL
On the Weaknesses of Reinforcement Learning for Neural Machine Translation
Read the paperCheck 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 |
|---|---|---|---|
| Fine-tune with expected BLEU reward | Pretrained NMT system · news2014 test | BLEU: 30.31 -> 30.73 | Helped |
| Replace expected BLEU reward with a constant reward of 1 | Reinforce fine-tuning of the pretrained NMT system · news2014 test | BLEU: 30.73 vs. 30.72 | No clear effect |
| Apply Reinforce with constant reward | Pretrained NMT conditional distributions · NMT conditional distributions | entropy: 3.45 -> 2.82 | Helped |
| Apply a single Reinforce step with constant reward | Pretrained token distribution · Controlled simulation, Constant Reward | entropy: 2.9 -> 2.85 | Helped |
| Train with simulated reward for 100K steps | Pretrained token distribution · Controlled simulation, Simulated Reward | entropy: 3 -> about 0.001 | Helped |
| Use Contrastive MRT instead of Reinforce | Controlled token-distribution simulations · Simulated Reward, target initially second-ranked | target-token rank | Helped |
| Use Contrastive MRT when the target token is initially third-ranked or lower | Controlled token-distribution simulations · Simulated Reward | target-token rank | Hurt |
Closest papers by meaning
- Reducing the Probability of Undesirable Outputs in Language Models Using Probabilistic Inference (NeurIPS 2025 · similarity 0.88)
- Towards Cost-Effective Reward Guided Text Generation (ICML 2025 · similarity 0.87)
- Reinforcement Learning Finetunes Small Subnetworks in Large Language Models (NeurIPS 2025 · similarity 0.86)
- Degeneration-free Policy Optimization: RL Fine-Tuning for Language Models without Degeneration (ICML 2024 · similarity 0.86)
- Teaching Models to Improve on Tape (AAAI 2025 · similarity 0.85)
- Reward-Guided Prompt Evolving in Reinforcement Learning for LLMs (ICML 2025 · 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.