Advanced large language models (LLMs) with long context windows can substantially reduce input truncation in document-level machine translation (DocMT). However, direct Doc2Doc translation remains prone to n-gram repetition and progressive quality degradation. A common remedy is to segment the document into finer-grained chunks. Nonetheless, conventional rule-based chunking approaches fail to handle the length distribution mismatch between training and inference. To address this, we introduce Fixed-Range Chunking (FRC), utilizing dynamic programming to partition documents into chunks within a predefined length interval. By consistently applying FRC during training and inference, the input documents of any length are mapped to the same length distribution, substantially reducing train-test length mismatch. Centered on FRC, we propose a lightweight dual-boundary matching algorithm for chunk alignment, alongside four distinct training strategies. Experimental results show that FRC-based fine-tuning substantially improves 7B LLMs over direct Doc2Doc fine-tuning and outperforms existing DocMT methods on IWSLT2017. We further construct GlobVDoc, a 10-language test set independent of mainstream DocMT training sources, and show that FRC improves out-of-distribution document translation.

LLM-based translation agents can plan, review, and revise translations, but most existing systems still rely on a single hand-crafted instruction set shared across inputs. This is suboptimal for translation, where texts vary widely in domain, context, style, and cultural grounding, leading to heterogeneous and input-specific errors. We propose STEINS, a self-evolving framework that treats the instruction set governing agent roles, translation strategies, and interaction constraints as an optimization target. During optimization, STEINS performs structured error diagnosis and uses the resulting evidence to refine the agent specification in a targeted and interpretable manner. The framework supports both offline corpus-level evolution and online instance-level adaptation at test time. Across WMT23 (general-domain, segment-level) and LITEVAL-CORPUS (literary, paragraph-level), and across four agentic translation architectures, offline evolution yields modest but generally positive corpus-level gains, while online instance-level adaptation yields larger and more consistent improvements, with macro-averaged XCOMET gains of +1.28 to +3.49 over 12 settings regardless of whether adaptation starts from the original or offline-evolved specification. Overall, our findings identify instance-level specification adaptation as the key ingredient in effective agentic machine translation.

As machine translation increasingly requires deeper contextual, linguistic, and cultural understanding, multi-agent collaboration has emerged as a promising approach. Multi-agent debate (MAD) frameworks, in which multiple agents deliberate to produce a final output, have shown strong performance on objective tasks, but remain underexplored in translation, where multiple valid renderings often exist. We adapt three MAD frameworks for English-Japanese translation and evaluate them against strong generative baselines, reasoning-capable LLMs, and a prompt-based self-reflection baseline. Across general-domain and culturally grounded datasets, the Society of Mind (SoM) variant yields the strongest results in the English-to-Japanese direction, showing that zero-shot translations leave substantial room for improvement through structured deliberation. Yet the gains of debate are front-loaded: later rounds do not reliably improve quality and often reintroduce translation errors. Diagnostic and error-span analyses show that hand-designed debate protocols tend to over-revise already strong translations, leading to semantic drift and process-induced degradation. These findings highlight both the promise and the limitations of agentic translation, and suggest that effective debate-based systems require mechanisms for preserving strong intermediate outputs.