We present Lumen, our submission to the WMT2026 General Machine Translation Shared Task. We focus on instruction-specialized translation, where systems must produce accurate translations while satisfying fine-grained requirements on style, terminology, and structured output. Lumen follows a four-stage pipeline. First, continued pre-training and supervised fine-tuning jointly strengthen multilingual translation competence and the ability to follow translation-specific instructions, such as terminology, style, and format constraints. Second, we extend the originally offline M{2}PO preference-optimization framework into an online preference-optimization stage: the current policy samples candidate translations, an LLM judge assigns continuous direct-assessment scores, and the resulting multi-pair preferences are used to update the model. Third, on-policy distillation transfers stronger translation-instruction-following behavior from a larger teacher model to the Qwen3-14B student, improving translation-instruction adherence while retaining a 14B primary translation backbone. Finally, structure-aware inference and postprocessing preserve required formats, while selective repair addresses clear structural or decoding failures before submission. We participate in 23 translation directions. Under an absolute 1–5 LLM-judge rubric that evaluates translation-instruction following and translation quality separately, Lumen reaches an overall mean score of 4.60. Despite using a 14B translation backbone, it achieves performance comparable to leading proprietary systems in both translation quality and translation-instruction following.

This paper presents Wayfinder, our submission system for the WMT26 General Machine Translation (GenMT) shared task. GenMT evaluates document-level translation under natural-language instructions, where structural preservation, formatting constraints, and output cleanliness are part of translation quality. Rather than training a new translation model, Wayfinder treats four API-based system outputs as a document-level candidate pool and produces one final hypothesis through a four-stage agentic pipeline. It aggregates source, instruction, metadata, candidates, and a DeepSeek-V4-Pro ranking-then-scoring signal; applies an anonymized pick-or-rewrite agent; and runs a conservative postcheck for severe objective defects. During development, a judge-guided loop compares outputs with a GPT-5.5 baseline, analyzes loss cases, updates prompts and repair constraints, and selectively reprocesses difficult documents. In automatic pairwise evaluation against GPT-5.5, Wayfinder obtains a pooled document-level weighted win rate of 0.6537 and remains above parity in every evaluated direction. The results suggest that candidate-pool translation, constrained agentic selection, conservative repair, and selective judge-guided iteration are practical components for instruction-conditioned shared-task MT.

Large Language Models (LLMs) have enabled a shift from sentence-level to document-to-document (Doc2Doc) machine translation, promising improved global coherence. However, document-to-document generation in a single pass frequently suffers from structural misalignment, manifesting as sentence omissions or hallucinations that violate the core requirement of source-target correspondence. To address this, we introduce **S**entence **T**ranslation **A**lignment **R**ate (**STAR**), an auxiliary metric that explicitly quantifies sentence-level structural fidelity. Building on this, we propose **STAR**-masked **P**reference **O**ptimization (**StarPO**), a framework that ranks document-level hypotheses by structural quality and utilizes a dynamic alignment mask to focus optimization on misaligned segments. Experimental results across news and literary domains demonstrate that StarPO significantly enhances translation quality and structural integrity. Notably, StarPO allows compact models to surpass the performance of massive proprietary systems like GPT-4o while maintaining superior token efficiency. We will release our code and datasets.

Multimodal language models achieve near-ceiling scores on food recognition benchmarks, yet it remains unclear whether this success reflects genuine cultural understanding or mere visual matching. To probe this distinction, we introduce CulturalMenuBench, a benchmark of 4,870 items in 10 languages across 18 regions; its 10 tasks pair final-dish and step-by-step cooking images with ingredients, procedural text, and regional labels, spanning basic recognition to process-grounded cultural attribution. Evaluating 12 models exposes a substantial knowledge-application gap: models exceeding 94% on standard multiple-choice tasks drop to at most 56% when attributing dishes to Chinese regional cuisines, despite an identical four-way format. Diagnostic analyses explain why: error patterns are consistent with random guessing, accuracy tracks visual distinctiveness rather than cultural structure, and models classify cuisines more accurately from dish names alone than from images (+7–18 points). The knowledge is thus present but cannot be activated through visual input. An ablation confirms these tasks genuinely require procedural evidence: removing sequential cooking images selectively degrades process-grounded tasks while others remain stable. Overall, CulturalMenuBench shows that near-perfect recognition can conceal an inability to apply cultural knowledge, motivating training that explicitly connects perception, procedure, and cultural context. Code and data are publicly available.

LLMs are transforming customer service, yet evaluating service agents across diverse business scenarios remains costly and fragmented, as each new domain requires building a dedicated benchmark from scratch. Existing benchmarks rely on static, scenario-specific scripts with two fundamental limitations: ***Scenario Silos***, where each domain demands costly independent annotation with no shared evaluation structure, and ***Dialogue Dynamics Gap***, where fixed scripts fail to capture the full range of real user behaviors. We propose **SAGE** (**S**ervice **A**gent **G**raph-guided **E**valuation), to our knowledge the first scenario-extensible benchmark to jointly assess logical SOP compliance and conversational quality, abstracting any SOP as a directed graph and bridging Scenario Silos through a lightweight two-step extension mechanism. SAGE further enables dynamic multi-turn simulation via adversarial user agents to overcome the Dynamics Gap. Across six industrial scenarios representative of real-world customer service deployment, we evaluate 27 mainstream LLMs spanning both closed-source systems and open-source families, offering a fine-grained capability map to guide real-world model selection. Our experiments further reveal two systematic failure patterns shared across all tested models: while models generally succeed at recognizing user intent, they frequently struggle to follow complex business logic and execute the correct subsequent action; and they sustain a warm and empathetic conversational tone even when their underlying procedural reasoning has already failed. **Code and resources are available at: https://github.com/lingshi6565/SAGE-Bench.**

Retrieval-Augmented Generation (RAG) systems often fail to maintain contextual faithfulness, generating responses that conflict with the provided context. Existing methods attempt to improve faithfulness through external interventions, such as specialized prompting, decoding-based calibration, or preference optimization. However, since these approaches treat the LLM as a black box, they lack a reliable mechanism to assess how these conflicts occur. Consequently, they tend to be brittle, data-intensive, and agnostic to the model’s internal reasoning process. In this paper, we move beyond black-box interventions to analyze the model’s internal reasoning process. We discover that conflicting and aligned knowledge states are linearly separable in the model’s latent space, and contextual noise systematically increases the entropy of these representations. Based on these findings, we propose ProbeRAG, a novel framework for faithful RAG that operates in three stages: (i) fine-grained knowledge pruning to filter irrelevant context, (ii) latent conflict probing to identify hard conflicts in the model’s latent space, and (iii) conflict-aware attention to modulate attention heads toward faithful context integration. Extensive experiments demonstrate that ProbeRAG substantially improves both accuracy and contextual faithfulness. The related resources are available at https://github.com/XMUDeepLIT/ProbeRAG.

Sparse Mixture-of-Experts (MoE) architectures have emerged as an increasingly influential paradigm as they offer a strategic balance between parameter scalability and computational efficiency. However, low-resource language tokens are often routed to different experts than those predominantly activated by high-resource inputs, which limits cross-lingual expert sharing. This cross-lingual routing divergence consequently hinders their efficacy in multilingual contexts. To address this issue, we propose SARA (Semantically Anchored Routing Alignment), a framework designed to transfer specialized capabilities from high-resource languages as anchors to low-resource languages. SARA explicitly aligns the routing distribution of multilingual inputs with high-resource semantic anchors using a symmetric Jensen-Shannon (JS) divergence constraint. Unlike traditional distillation methods that operate on output logits, SARA directly aligns the internal routing distributions of MoE layers, encouraging mechanistic consistency in expert selection across languages. We conduct experiments on 2 LLMs across 5 low-resource languages and 3 benchmarks. Experiment results demonstrate that SARA outperforms standard instruction tuning (e.g., +0.8% on Qwen3-30B-A3B and +1.2% on Phi-3.5-MoE-instruct on Global-MMLU benchmark). Further analyses show that SARA effectively addresses performance bottlenecks in low-resource languages, providing a scalable pathway to enhance multilingual capabilities in sparse architectures.

Large Language Models (LLMs) have advanced Machine Translation (MT), yet they remain vulnerable by hallucinations. Most existing benchmarks, originally designed for neural MT models, do not effectively elicit hallucinations from multilingual LLMs, resulting in a low density of samples for LLM translation hallucination analysis. To address this, we create HalloMTBench, a human-verified hallucination benchmark derived from tens of millions of translations, for studying signals that trigger LLM translation hallucinations across 11 English-to-X directions. We curate 6,908 authentic hallucination instances through a rigorous process: frontier LLMs first translate a large source corpus, and their outputs are then validated by both an LLM-judge ensemble and human experts. Based on our analysis of these hallucination instances, we introduce a taxonomy for translation hallucination that distinguishes Instruction Detachment from Source Detachment. Our evaluation of 21 model variants on HalloMTBench reveals several observable vulnerability patterns, including model Characteristics, source length, language proximity, hallucination reproducibility, and model-dependent effects of reasoning-style decoding. HalloMTBench offers a forward-looking testbed for diagnosing LLM translation hallucination. All resources will be released.

Aligning Large Language Models (LLMs) to human preferences is pivotal for Machine Translation (MT), yet current approaches are often hindered by misleading reward signals. Our analysis reveals that prevailing Quality Estimation (QE) models exhibit a systematic blind spot towards partial errors—specifically partial hallucinations and omissions—often favoring superficially fluent but unfaithful translations. To address this, we propose M²PO (Multi-Perspective Multi-Pair Preference Optimization), a data-centric framework for preference optimization in machine translation. First, to correct the bias towards fluency, M²PO uses a multi-perspective alignment mechanism that decouples semantic fidelity from fluency, prioritizing faithfulness via a curriculum strategy. Second, with the bias corrected, partial errors fall between perfect and severely incorrect translations, making them inefficient to learn via standard best-versus-worst comparisons. We thus introduce a multi-pair objective that leverages the full candidate list to capture these fine-grained error signals. Experiments on WMT23, WMT24, and FLORES-200 show that M²PO enables a 9B model to outperform leading open-source baselines and achieve parity with proprietary models like GPT-4o and Gemini-2.0-Flash, demonstrating significant potential for efficient, high-fidelity LLM-based translation.

While mechanistic interpretability tools like Sparse Autoencoders (SAEs) can uncover meaningful features within Large Language Models (LLMs), a critical gap remains in transforming these insights into practical actions for model optimization. We bridge this gap with the hypothesis that data selection guided by a model’s internal task features is a effective training strategy. Inspired by this, we propose Interpretability-Guided Data Selection (IGDS), a framework that first identifies these causal task features through frequency recall and interventional filtering, then selects “Feature-Resonant Data” that maximally activates task features for fine-tuning. We validate IGDS on mathematical reasoning, summarization, and translation tasks within Gemma-2, LLaMA-3.1, and Qwen3 models. Our experiments demonstrate exceptional data efficiency: on the Math task, IGDS surpasses full-dataset fine-tuning by a remarkable 17.4% on Gemma-2-2B while using only 50% of the data, and outperforms established baselines focused on data quality and diversity. Analysis confirms a strong positive correlation between feature amplification and task performance improvement. IGDS thus provides a direct and effective framework to enhance LLMs by leveraging their internal mechanisms, validating our core hypothesis.

Cross-cultural entity translation remains challenging for large language models (LLMs) as literal or phonetic renderings are usually yielded instead of culturally appropriate translations in context. However, relevant knowledge may already be encoded in model parameters during large-scale pre-training. To incentivize the effective use of parametric knowledge, we propose EA-RLVR (Entity-Anchored Reinforcement Learning with Verifiable Rewards), a training framework that optimizes cross-cultural entity translation without relying on external knowledge bases. EA-RLVR anchors supervision on a verifiable, entity-level reward signal and incorporates lightweight structural gates to stabilize optimization. This design steers the model toward learning a robust reasoning process rather than merely imitating reference translations. We evaluate EA-RLVR on XC-Translate and observe consistent improvements in both entity translation accuracy and out-of-domain generalization. Specifically, training on merely 7k samples boosts Qwen3-14B’s entity translation accuracy from 23.66% to 31.87% on a 50k test set comprising entirely unseen entities. The learned entity translation ability also transfers to general translation, yielding +1.35 XCOMET on WMT24pp, which scales to +1.59 with extended optimization. Extensive analyses of pass@k dynamics and reward formulations attribute these gains to superior sampling efficiency and a stable optimization landscape.

This paper presents the Marco-MT-Algharb system, our submission to the WMT2025 General Machine Translation Shared Task from Alibaba International Digital Commerce (AIDC). Built on a large language model (LLM) foundation, the system’s strong performance stems from novel quality-aware training and decoding techniques: (1) a two-step supervised fine-tuning (SFT) process incorporating data distillation, (2) a two-step reinforcement learning (RL) framework for preference alignment, and (3) a hybrid decoding strategy that integrates word alignment with Minimum Bayes Risk (MBR) re-ranking to improve faithfulness. These approaches jointly ensure high accuracy and robustness across diverse languages and domains. In the official human evaluation, our system secured five first‐place finishes, one second, and four third‐place results in the constrained category across the 13 directions we participated in. Notably, for the English-Chinese, our results surpassed all open/closed‐source systems.