PAMT is proposed, a process-aligned training framework that combines cold-start domain-aware Long-CoT supervision with reinforcement learning that improves over base models, outperforms MT-specialized baselines on average, and remains competitive with strong LLMs/LRMs across in-domain, OOD, and multilingual settings.
Abstract
Multi-domain machine translation (MDMT) requires more than fluent generation: it demands domain-sensitive translation decisions such as domain disambiguation, terminology control, and stylistic adaptation. Large reasoning models (LRMs) make such decisions explicit through intermediate translation steps, but our analysis across 15 domains and four translation directions shows that this explicit reasoning is double-edged: it improves long-form and high-difficulty translation, yet often drifts in terminology-intensive and stylistically constrained settings. We trace this failure to a credit-assignment bottleneck: existing methods optimize final outputs or coarse trajectories, but cannot identify which translation steps actually help the final translation. To address this, we propose PAMT, a process-aligned training framework that combines cold-start domain-aware Long-CoT supervision with reinforcement learning. PAMT uses sequence-level format and outcome rewards for the final translation, together with a step-level process reward that measures how much each explicit translation step increases the likelihood of the reference translation. Across two backbones, PAMT improves over base models, outperforms MT-specialized baselines on average, and remains competitive with strong LLMs/LRMs across in-domain, OOD, and multilingual settings.
TwT (Translation with Thought), a resource-rational framework that learns to modulate inference between intuitive and deliberate reasoning, confirms that aligning translation behavior with cognitive principles enables robust generalization, high translation quality, and efficient reasoning in MDMT.
Yongshi Ye, Biao Fu, Chongxuan Huang et al.· Annual Meeting of the Associ...· 0 citations
Data-limited multi-domain machine translation remains challenging because parallel corpora are scarce in specialized domains, domain terminology is highly constrained, and large language models may generate fluent but unfaithful translations. Direct prompting or ordinary fine-tuning is therefore insufficient for domain-sensitive translation scenarios. To address these issues, this paper proposes DKA-LLM-MT, a large-language-model-enhanced adaptive optimization framework for data-limited multi-domain machine translation. The framework follows a data–model–reliability design. First, a domain-knowledge-constrained data augmentation strategy generates pseudo-parallel corpora under terminology, semantic consistency, and domain-style constraints. Second, a retrieval-augmented parameter-efficient adaptation mechanism integrates domain memory retrieval, lightweight LoRA adapters, and dynamic domain routing. Third, a reliability-aware optimization mechanism incorporates semantic fidelity, terminology consistency, and hallucination risk into both training-time data selection and inference-time candidate reranking. Experiments are conducted on five public data-limited domain translation benchmarks covering medical, legal, technical, news, and spoken-style texts. The proposed method achieves an average BLEU of 36.18, chrF of 62.14, COMET of 0.816, and TER of 40.62, consistently outperforming strong neural, multilingual, and LLM-based baselines. Additional matched-backbone and same-language-pair analyses are included to separate the effect of domain adaptation from language-pair variation. Reliability evaluation further shows that DKA-LLM-MT improves terminology accuracy to 89.6% and reduces hallucination rate to 3.2%. The proposed framework provides an effective and reliable solution for domain-sensitive machine translation under limited bilingual supervision and offers practical support for bilingual reading, specialized translation assistance, and domain-oriented language learning.
LatentMT is introduced, the first systematic study of latent-reasoning LoopLMs for machine translation that adapts a small 2.6B-parameter backbone model with lightweight training and shows that hidden-representation differences shrink along the recurrent reasoning-step axis, supporting the observed saturation in performance.
Wei-Rui Chen, Samar M. Magdy, Chiyu Zhang et al.· 0 citations
Machine translation for low-resource languages suffers from domain-imbalanced corpora, causing quality degradation on technical text. However, in-context learning opens the possibility to rely on limited in-domain corpora to inform translation. We present lessons learned from Tulun, a retrieval-augmented system combining neural MT with LLM post-editing, guided by user-configurable translation memories and glossaries. Deployed for medical translation in Timor-Leste (Tetun) and disaster relief translation in Vanuatu (Bislama), the system achieves accuracy improvements over baseline MT by 16.90-22.41 ChrF++ points, while offering rapid adaptability and transparency to end-users. Key recommendations include: domain granularity matters more than broad categories; translation target audience should inform retrieval; and RAG-augmented MT is most effective for languages that lack domain corpora but remain within LLM pretraining distributions.
Raphael Merx, Ekaterina Vylomova· Annual International ACM SIG...· 0 citations
The results show that the quality of small ``base'' models can be greatly enhanced, and that reinforcement learning with verifiable rewards can be applied to NMT in the legal domain and surpasses the translation quality of supervised fine-tuning.
Aixiu An, Michael Jungo, Eloi Eynard et al.· 0 citations
This work proposes augmenting existing benchmarks to increase translation difficulty by combining adversarial optimization with a differentiable translation difficulty estimator, and uses gradients from a combined difficulty and fluency objective to iteratively replace tokens in Adversarial Translation Optimization (ATO).
William Kalikman, Šimon Sukup, Michal Tesnar et al.· 0 citations