We describe the HITSZ system submitted to the Sorbian track of the WMT 2026 Shared Task on Multitask LLMs with Limited Resources. The track requires a single Qwen3.5-2B model to jointly perform machine translation (MT), question answering (QA), spell checking (SC), grammar checking (GC), and mathematical reasoning (MR) for Upper and Lower Sorbian. To strengthen the model’s limited Sorbian linguistic knowledge while retaining previously acquired capabilities, we first perform mixed continued pre-training (CPT) on Sorbian monolingual and parallel data together with multilingual and task-oriented replay. We then apply full-parameter multitask supervised fine-tuning (SFT) to align the adapted model with the five heterogeneous task formats. Compared with the official Qwen3.5-2B baseline, our primary submission improves MT by 41.44 chrF++ points and improves QA, SC, GC, and MR by 13.94, 66.47, 60.77, and 13.20 points, respectively, ranking third overall in the Sorbian track. Additional analysis shows that removing CPT reduces our five-task development score by 5.24 points, with the largest losses on SC, MT, and GC. We also find that heavily increasing MT supervision does not further improve translation and can degrade other jointly trained tasks.

Document-level translation remains one of the most challenging tasks for large language models, which are constrained by limited context windows that impede global cohesion, while simultaneously suffering from redundant contextual information that degrades translation quality. To address this, we propose a human-like long document translation agent called Loong, which leverages a 3E memory module (Essence-Exemplar-Entity) to store summaries, sentence pairs, and entity records as historical context. Instead of passively attending to all history, Loong performs deep reasoning to adaptively identify the optimal context for translation guidance. Loong optimizes its context policy through reinforcement learning, utilizing preference data derived from its own sampled observe-and-act reasoning trajectories. Empirical evaluations demonstrate that Loong achieves substantial translation quality improvements in English ⇔ Chinese, German, and French directions, with average gains of up to 13.0 points across the three evaluation metrics. Furthermore, Loong exhibits strong generalization across domains and robustness against contextual noise, while maintaining remarkable stability in ultra-long document translation. Our code is freely available at https://github.com/YutongWang1216/LoongDocMT.

Large Language Reasoning Models have demonstrated remarkable success on static tasks, yet their application to multi-round agentic planning in interactive environments faces two fundamental challenges. First, the intractable credit assignment problem renders conventional reinforcement learning ineffective in sparse-reward settings. Second, the computational overhead of verbose, step-by-step reasoning histories is prohibitive. To address these challenges, we propose BPO, a three-stage framework (bootstrapping, extrapolation, and refinement) that establishes a self-improving data flywheel to develop robust reasoning models for long-horizon, sparse-reward environments. Our framework first bootstraps efficient reasoning using the proposed planning quaternions with long-short chain-of-thought fusion. It then extrapolates to out-of-distribution tasks through complexity-stratified curriculum learning. Finally, the model iteratively refines itself by learning exclusively on experiences selected via reward-gated rejection sampling. Experiments on ALFWorld, ScienceWorld, and WebShop demonstrate that our approach achieves state-of-the-art with significant token efficiency, providing a new recipe for reasoning models in agentic planning.

Idiom translation remains a formidable challenge for Large Language Models (LLMs), as the constraints of static parametric memory and the noise in sentence-level retrieval often lead to literal misinterpretations. To address this, we propose DeReA, a detect-retrieve-arbitrate framework. The system employs a preference-aligned detector to identify idiomatic spans by reasoning over semantic conflicts between literal and contextual meanings. Subsequently, an idiom-centric translator invokes a fine-tuned embedding model to efficiently retrieve canonical definitions from an external knowledge base. The translator then utilizes a dual-path arbitration mechanism to select the optimal rendering by weighing the retrieval-augmented translations against direct translation. To evaluate our framework, we introduce LoMI, a high-difficulty benchmark with low data contamination. Experimental results demonstrate that DeReA significantly enhances performance across various model scales, improving GPT-5-mini by over 5.2 points in both idiomatic quality and consistency according to LLM-based metrics. Furthermore, evaluations on an emerging slang dataset from Urban Dictionary validate the potential of our approach in handling novel and evolving linguistic data. Our code is available at https://github.com/jrongqing/DeReA.

In this paper, we propose a BERT-based hierarchical text classification framework to address the challenges of training multi-level classification tasks. As part of the SemEval-2025 Task 10 challenge (Subtask 2), the framework performs fine-grained text classification by training dedicated sub-category classifiers for each top-level category. Experimental results demonstrate the feasibility of the proposed approach in multi-class text classification tasks.

Nous présentons une étude comparative de deux paradigmes de classification hiérarchique multi-labels de texte dans le contexte de l’extraction des narratifs d’articles de presse. La première approche utilise un cadre séquentiel basé sur BERT qui identifie les narratifs et leurs sous-narratifs correspondants. La seconde utilise des agents LLM spécialisés, chacun effectuant une classification binaire pour des catégories narratives spécifiques. En évaluant les deux approches sur l’ensemble de données SemEval-2025 Task 10 dans cinq langues, nous constatons que l’approche basée sur BERT offre une efficacité de calcul et des performances interlinguistiques cohérentes (moyenne F1 macro : 0,475), tandis que la méthode basée sur les agents démontre une meilleure gestion des narratifs nuancés et de meilleures performances sur les données en anglais (F1 macro : 0,513). Notre analyse révèle des forces complémentaires entre ces paradigmes. Nous discutons des implications pratiques et proposons des orientations pour des systèmes hybrides potentiels.

Proper initialization is crucial for any system, particularly in multi-agent systems (MAS), where it plays a pivotal role in determining both the system’s efficiency and effectiveness. However, existing MAS initialization methods do not fully account for the collaborative needs of the generated agents in subsequent stages. Inspired by the principles of effective team composition, we propose , which aims to optimize the structure of agent teams. Specifically, in addition to multi-round interactions and reflections between agents during agent generation, AgentInit incorporates a Natural Language to Format mechanism to ensure consistency and standardization. Balanced team selection strategies using Pareto principles are subsequently applied to jointly consider agent team diversity and task relevance to promote effective and efficient collaboration and enhance overall system performance. Experiments show that AgentInit consistently outperforms state-of-the-art initialization methods and pre-defined strategies across various frameworks and tasks, achieving an overall performance improvement of up to 1.2 and 1.7, respectively, while also significantly reducing token consumption. Further analysis confirms its strong transferability to similar tasks and verifies the effectiveness of its key components, demonstrating its capability and adaptability as a reliable MAS initialization method. Source code and models are available at https://github.com/1737423697/AgentInit.

This study investigates the automatic detection and classification of persuasion techniques across five Slavic languages (Bulgarian, Croatian, Polish, Russian, and Slovenian), addressing two subtasks: binary detection of persuasion techniques in text fragments (Subtask 1) and multi-label classification of specific technique types (Subtask 2). To overcome limited training resources, we implemented a multi-level cross-lingual augmentation strategy utilizing GPT-4o for non-Slavic to Slavic conversion and intra-Slavic language migration. We employ XLM-RoBERTa architecture with two LLM-enhanced variants that use explanations to improve classification performance. The experimental results demonstrate varied performance across languages and tasks, with our approach achieving first place in the Russian subtask 1 and second place in Bulgarian subtask 2, confirming that larger parameter models excel in complex classification tasks. These findings highlight the significant potential of LLMs for enhancing multilingual classification and the persistent difficulties in ensuring consistent cross-linguistic performance.

Multi-agent systems (MAS) based on large language models (LLMs) have demonstrated significant potential in collaborative problem-solving. However, they still face substantial challenges of low communication efficiency and suboptimal task performance, making the careful design of the agents’ communication topologies particularly important. Inspired by the management theory that roles in an efficient team are often dynamically adjusted, we propose AgentDropout, which identifies redundant agents and communication across different communication rounds by optimizing the adjacency matrices of the communication graphs and eliminates them to enhance both token efficiency and task performance. Compared to state-of-the-art methods, AgentDropout achieves an average reduction of 21.6% in prompt token consumption and 18.4% in completion token consumption, along with a performance improvement of 1.14 on the tasks. Furthermore, the extended experiments demonstrate that AgentDropout achieves notable domain transferability and structure robustness, revealing its reliability and effectiveness. We release our code at https://github.com/wangzx1219/AgentDropout.

Large language models (LLMs) have exhibited remarkable performance in various natural language processing tasks. Techniques like instruction tuning have effectively enhanced the proficiency of LLMs in the downstream task of machine translation. However, the existing approaches fail to yield satisfactory translation outputs that match the quality of supervised neural machine translation (NMT) systems. One plausible explanation for this discrepancy is that the straightforward prompts employed in these methodologies are unable to fully exploit the acquired instruction-following capabilities. To this end, we propose the TasTe framework, which stands for translating through self-reflection. The self-reflection process includes two stages of inference. In the first stage, LLMs are instructed to generate preliminary translations and conduct self-assessments on these translations simultaneously. In the second stage, LLMs are tasked to refine these preliminary translations according to the evaluation results. The evaluation results in four language directions on the WMT22 benchmark reveal the effectiveness of our approach compared to existing methods. Our work presents a promising approach to unleash the potential of LLMs and enhance their capabilities in MT. The codes and datasets are open-sourced at https://github.com/YutongWang1216/ReflectionLLMMT.

The ability of commonsense reasoning (CR) decides whether a neural machine translation (NMT) model can move beyond pattern recognition. Despite the rapid advancement of NMT and the use of pretraining to enhance NMT models, research on CR in NMT is still in its infancy, leaving much to be explored in terms of effectively training NMT models with high CR abilities and devising accurate automatic evaluation metrics. This paper presents a comprehensive study aimed at expanding the understanding of CR in NMT.For the training, we confirm the effectiveness of incorporating pretrained knowledge into NMT models and subsequently utilizing these models as robust testbeds for investigating CR in NMT. For the evaluation, we propose a novel entity-aware evaluation method that takes into account both the NMT candidate and important entities in the candidate, which is more aligned with human judgement. Based on the strong testbed and evaluation methods, we identify challenges in training NMT models with high CR abilities and suggest directions for further unlabeled data utilization and model design. We hope that our methods and findings will contribute to advancing the research of CR in NMT. Source data, code and scripts are freely available at https://github.com/YutongWang1216/CR-NMT.