4arXiv cs.CL (Computation and Language)·
RLVR reasoning traces improve neural machine translation quality at increased inference cost
A new arXiv preprint investigates whether quality gains from reinforcement learning with verifiable rewards (RLVR) in neural machine translation stem from the reasoning trace itself or from the training paradigm more broadly. Experiments show that including reasoning during inference specifically improves translation quality, while omitting it during training has less impact. The paper quantifies the cost-quality tradeoff introduced by longer reasoning outputs, finding that better translations come at meaningful computational expense.
Related guides (3)
Related events (8)
Reinforcement learning with verifiable rewards improves small-model legal machine translation
Researchers evaluate multiple training paradigms for legal machine translation, comparing supervised fine-tuning and reinforcement learning with verifiable rewards (RLVR) on small models (Qwen3.5 4B/9B, Gemma 3 12B) against frontier reasoning models. Using the Swiss multilingual legal system as a testbed, they find RLVR outperforms SFT for legal NMT and brings small models close to frontier performance, though a gap remains. The study also observes diminishing returns from re-training as model size increases. Code and models are publicly released.
Synthetic linguistic reasoning traces improve low-resource machine translation via in-context learning
Researchers propose a pipeline that generates step-by-step linguistic reasoning traces from Universal Dependencies treebanks, dictionaries, and grammar-rule banks to assist LLMs in translating extremely low-resource languages. Evaluated on Xibe and Chintang across ICL, SFT, and RFT settings, the traces prove most effective as inference-time guidance rather than training data. Models can leverage reliable grammatical analyses at inference time but struggle to learn to generate accurate traces themselves, identifying trace generation quality as the key bottleneck.
LongTraceRL: Reinforcement Learning for Long-Context Reasoning via Search Agent Trajectories and Rubric Rewards
LongTraceRL is a new RL training framework for improving long-context reasoning in LLMs, addressing limitations of existing RLVR methods. It constructs challenging training data using multi-hop questions from knowledge graph random walks and tiered distractors derived from search agent trajectories (high-confusability: read but uncited; low-confusability: seen but unopened). A rubric reward provides entity-level process supervision along reasoning chains, applied only to correct responses to prevent reward hacking. Experiments across three LLMs (4B–30B parameters) on five long-context benchmarks show consistent improvements over strong baselines.
Instruction tuning plus model merging improves reasoning LMs in verifiable and unverifiable domains
A new arXiv preprint proposes a two-step method for adapting reasoning language models (RLMs) to domains lacking reliable verifiers: first apply standard supervised fine-tuning (without reasoning traces) on the target domain, then merge the resulting model with the original RLM to recover reasoning behavior. The technique is evaluated on both verifiable domains (coding) and hard-to-verify domains (text summarization), showing improvements while preserving general RLM capabilities. Notably, the authors report the full adaptation can be achieved for under $3, making it highly accessible.
ExpRL: RL-based mid-training using human QA data as reward scaffolds for LLM reasoning
ExpRL proposes an automated approach to LLM mid-training that replaces manually curated reasoning traces with large corpora of human-written QA data used as reward scaffolds rather than imitation targets. Reference solutions are hidden from the policy and used only to construct problem-specific grading rubrics, enabling dense process-level rewards that reinforce partial progress and intermediate reasoning steps. On challenging math reasoning benchmarks, ExpRL outperforms SFT, sparse-reward GRPO, and self-distillation as an RL initialization strategy, with additional mixed-domain experiments suggesting broader applicability.
Luar: Selective Translation via Reinforcement Learning for Multilingual Reasoning
Luar is a reinforcement learning framework that trains reasoning language models to selectively invoke English translation only when direct understanding of a non-English input is deemed unreliable. The approach, built on top of GRPO, outperforms standard multilingual baselines across reasoning benchmarks, with especially large gains on low-resource languages. Analysis confirms the model learns to avoid unnecessary translation when direct reasoning suffices, and generalizes the translation-call behavior to unseen low-resource languages.
Context-aware translation cascades improve multilingual reasoning across 285 languages
A new arXiv preprint identifies a structural flaw in standard translation cascades for multilingual reasoning—each stage discards context needed by later stages—and proposes a training-free fix: providing the original question, English translation, and reasoning trace to the final translation module. The intervention is evaluated on nine multilingual benchmarks across three backbone models and 285 languages, showing strong gains for open-ended generation. The key finding is that preserving the original-language question until the end of the pipeline captures most of the benefit.
CORA: Consistency-Oriented Reasoning Alignment addresses thinking-answer gap in multimodal RLVR
Researchers identify and analyze a systematic inconsistency between reasoning traces and final answers in RLVR-trained large vision-language models, showing the problem persists throughout GRPO training and inference. They propose CORA, which introduces a lightweight plug-and-play consistency reward model and a Hybrid Reward Advantage Splitting (HRAS) mechanism to coordinate task and consistency optimization. Experiments across multimodal reasoning benchmarks show CORA improves both task performance and reasoning faithfulness.