Although large language models (LLMs) have demonstrated remarkable reasoning capabilities, they still face challenges in knowledge-intensive multi-hop reasoning. Recent work explores iterative retrieval to address complex problems. However, the lack of intermediate guidance often results in inaccurate retrieval and flawed intermediate reasoning, leading to incorrect reasoning. To address these, we propose Self-Critique Guided Iterative Reasoning (SiGIR), which uses self-critique feedback to guide the iterative reasoning process. Specifically, through end-to-end training, we enable the model to iteratively address complex problems via question decomposition. Additionally, the model is able to self-evaluate its intermediate reasoning steps. During iterative reasoning, the model engages in branching exploration and employs self-evaluation to guide the selection of promising reasoning trajectories. Extensive experiments on three multi-hop reasoning datasets demonstrate the effectiveness of our proposed method, surpassing the previous SOTA by $8.6\%$. Furthermore, our thorough analysis offers insights for future research. Our code, data, and models are available at Github: https://github.com/zchuz/SiGIR-MHQA.
Investigating hallucination issues in large language models (LLMs) within cross-lingual and cross-modal scenarios can greatly advance the large-scale deployment in real-world applications. Nevertheless, the current studies are limited to a single scenario, either cross-lingual or cross-modal, leaving a gap in the exploration of hallucinations in the joint cross-lingual and cross-modal scenarios. Motivated by this, we introduce a novel joint Cross-lingual and Cross-modal Hallucinations benchmark (CCHall) to fill this gap. Specifically, CCHall simultaneously incorporates both cross-lingual and cross-modal hallucination scenarios, which can be used to assess the cross-lingual and cross-modal capabilities of LLMs. Furthermore, we conduct a comprehensive evaluation on CCHall, exploring both mainstream open-source and closed-source LLMs. The experimental results highlight that current LLMs still struggle with CCHall. We hope CCHall can serve as a valuable resource to assess LLMs in joint cross-lingual and cross-modal scenarios.
Selective state-space models have achieved great success in long-sequence modeling. However, their capacity for language representation, especially in complex hierarchical reasoning tasks, remains underexplored. Most large language models rely on flat Euclidean embeddings, limiting their ability to capture latent hierarchies. To address this limitation, we propose Hierarchical Mamba (HiM), integrating efficient Mamba2 with exponential growth and curved nature of hyperbolic geometry to learn hierarchy-aware language embeddings for deeper linguistic understanding. Mamba2-processed sequences are projected to the Poincare ball (via tangent-based mapping) or Lorentzian manifold (via cosine and sine-based mapping) with "learnable" curvature, optimized with a combined hyperbolic loss. Our HiM model facilitates the capture of relational distances across varying hierarchical levels, enabling effective long-range reasoning. This makes it well-suited for tasks like mixed-hop prediction and multi-hop inference in hierarchical classification. We evaluated our HiM with four linguistic and medical datasets for mixed-hop prediction and multi-hop inference tasks. Experimental results demonstrated that: 1) Both HiM models effectively capture hierarchical relationships for four ontological datasets, surpassing Euclidean baselines. 2) HiM-Poincare captures fine-grained semantic distinctions with higher h-norms, while HiM-Lorentz provides more stable, compact, and hierarchy-preserving embeddings favoring robustness over detail.
Manga, or Japanese comics, is a richly multimodal narrative form that blends images and text in complex ways. Teaching large multimodal models (LMMs) to understand such narratives at a human-like level could help manga creators reflect on and refine their stories. To this end, we introduce two benchmarks for multimodal manga understanding: MangaOCR, which targets in-page text recognition, and MangaVQA, a novel benchmark designed to evaluate contextual understanding through visual question answering. MangaVQA consists of 526 high-quality, manually constructed question-answer pairs, enabling reliable evaluation across diverse narrative and visual scenarios. Building on these benchmarks, we develop MangaLMM, a manga-specialized model finetuned from the open-source LMM Qwen2.5-VL to jointly handle both tasks. Through extensive experiments, including comparisons with proprietary models such as GPT-4o and Gemini 2.5, we assess how well LMMs understand manga. Our benchmark and model provide a comprehensive foundation for evaluating and advancing LMMs in the richly narrative domain of manga.
Role-Playing Agents (RPAs), benefiting from large language models, is an emerging interactive AI system that simulates roles or characters with diverse personalities. However, existing methods primarily focus on mimicking dialogues among roles in textual form, neglecting the role's voice traits (e.g., voice style and emotions) as playing a crucial effect in interaction, which tends to be more immersive experiences in realistic scenarios. Towards this goal, we propose OmniCharacter, a first seamless speech-language personality interaction model to achieve immersive RPAs with low latency. Specifically, OmniCharacter enables agents to consistently exhibit role-specific personality traits and vocal traits throughout the interaction, enabling a mixture of speech and language responses. To align the model with speech-language scenarios, we construct a dataset named OmniCharacter-10K, which involves more distinctive characters (20), richly contextualized multi-round dialogue (10K), and dynamic speech response (135K). Experimental results showcase that our method yields better responses in terms of both content and style compared to existing RPAs and mainstream speech-language models, with a response latency as low as 289ms. Code and dataset are available at https://github.com/AlibabaResearch/DAMO-ConvAI/tree/main/OmniCharacter.
Using massive datasets, foundation models are large-scale, pre-trained models that perform a wide range of tasks. These models have shown consistently improved results with the introduction of new methods. It is crucial to analyze how these trends impact the medical field and determine whether these advancements can drive meaningful change. This study investigates the application of recent state-of-the-art foundation models, DINOv2, MAE, VMamba, CoCa, SAM2, and AIMv2, for medical image classification. We explore their effectiveness on datasets including CBIS-DDSM for mammography, ISIC2019 for skin lesions, APTOS2019 for diabetic retinopathy, and CHEXPERT for chest radiographs. By fine-tuning these models and evaluating their configurations, we aim to understand the potential of these advancements in medical image classification. The results indicate that these advanced models significantly enhance classification outcomes, demonstrating robust performance despite limited labeled data. Based on our results, AIMv2, DINOv2, and SAM2 models outperformed others, demonstrating that progress in natural domain training has positively impacted the medical domain and improved classification outcomes. Our code is publicly available at: https://github.com/sajjad-sh33/Medical-Transfer-Learning.
The Massive Multitask Language Understanding (MMLU) benchmark has been widely used to evaluate language models across various domains. However, existing MMLU datasets primarily focus on high-resource languages such as English, which leaves low-resource languages like Bengali underrepresented. In this paper, we introduce BnMMLU, a benchmark to evaluate the multitask language understanding capabilities of Bengali in language models. The dataset spans 23 domains, including science, humanities, mathematics and general knowledge and is structured in a multiple-choice format to assess factual knowledge, application-based problem-solving and reasoning abilities of language models. It consists of 138,949 question-option pairs. We benchmark several proprietary and open-source large language models (LLMs) on the BnMMLU test set. Additionally, we annotate the test set with three cognitive categories-factual knowledge, procedural application and reasoning-to gain deeper insights into model strengths and weaknesses across various cognitive tasks. The results reveal significant performance gaps, highlighting the need for improved pre-training and fine-tuning strategies tailored to Bengali data. We release the dataset and benchmark results to facilitate further research in this area.
Instruction-tuned large language models (LLMs) employ structured templates, such as role markers and special tokens, to enforce format consistency during inference. However, we identify a critical limitation of such formatting: it induces a phenomenon we term diversity collapse, where the model generates semantically similar outputs for open-ended inputs, undermining creativity and variability. We systematically evaluate this effect across tasks like story completion and free-form generation, finding that (1) diversity collapse persists even under high-temperature sampling, and (2) structural tokens in templates significantly constrain the model's output space. To contextualize these findings, we fine-tune the same model using a range of structured prompts and then evaluate them across three axes: downstream task performance, alignment behavior, and output diversity. Our analysis shows that format consistency between fine-tuning and inference is crucial for structure-sensitive tasks (e.g., GSM8K, IFEval), but has marginal influence on knowledge-heavy tasks (e.g., MMLU, WebQuestions). In contrast, output diversity is primarily governed by the presence or absence of structural tokens, with minimal formatting yielding the most diverse outputs. These findings reveal that current prompting conventions, while beneficial for alignment, may inadvertently suppress output diversity, underscoring the need for diversity-aware prompt design and instruction tuning.
Human social interactions depend on the ability to infer others' unspoken intentions, emotions, and beliefs-a cognitive skill grounded in the psychological concept of Theory of Mind (ToM). While large language models (LLMs) excel in semantic understanding tasks, they struggle with the ambiguity and contextual nuance inherent in human communication. To bridge this gap, we introduce MetaMind, a multi-agent framework inspired by psychological theories of metacognition, designed to emulate human-like social reasoning. MetaMind decomposes social understanding into three collaborative stages: (1) a Theory-of-Mind Agent generates hypotheses user mental states (e.g., intent, emotion), (2) a Domain Agent refines these hypotheses using cultural norms and ethical constraints, and (3) a Response Agent generates contextually appropriate responses while validating alignment with inferred intent. Our framework achieves state-of-the-art performance across three challenging benchmarks, with 35.7% improvement in real-world social scenarios and 6.2% gain in ToM reasoning. Notably, it enables LLMs to match human-level performance on key ToM tasks for the first time. Ablation studies confirm the necessity of all components, which showcase the framework's ability to balance contextual plausibility, social appropriateness, and user adaptation. This work advances AI systems toward human-like social intelligence, with applications in empathetic dialogue and culturally sensitive interactions. Code is available at https://github.com/XMZhangAI/MetaMind.
While AI agents hold transformative potential in business, effective performance benchmarking is hindered by the scarcity of public, realistic business data on widely used platforms. Existing benchmarks often lack fidelity in their environments, data, and agent-user interactions, with limited coverage of diverse business scenarios and industries. To address these gaps, we introduce CRMArena-Pro, a novel benchmark for holistic, realistic assessment of LLM agents in diverse professional settings. CRMArena-Pro expands on CRMArena with nineteen expert-validated tasks across sales, service, and 'configure, price, and quote' processes, for both Business-to-Business and Business-to-Customer scenarios. It distinctively incorporates multi-turn interactions guided by diverse personas and robust confidentiality awareness assessments. Experiments reveal leading LLM agents achieve only around 58% single-turn success on CRMArena-Pro, with performance dropping significantly to approximately 35% in multi-turn settings. While Workflow Execution proves more tractable for top agents (over 83% single-turn success), other evaluated business skills present greater challenges. Furthermore, agents exhibit near-zero inherent confidentiality awareness; though targeted prompting can improve this, it often compromises task performance. These findings highlight a substantial gap between current LLM capabilities and enterprise demands, underscoring the need for advancements in multi-turn reasoning, confidentiality adherence, and versatile skill acquisition.
Lay paraphrasing aims to make scientific information accessible to audiences without technical backgrounds. However, most existing studies focus on a single domain, such as biomedicine. With the rise of interdisciplinary research, it is increasingly necessary to comprehend knowledge spanning multiple technical fields. To address this, we propose Sci-LoRA, a model that leverages a mixture of LoRAs fine-tuned on multiple scientific domains. In particular, Sci-LoRA dynamically generates and applies weights for each LoRA, enabling it to adjust the impact of different domains based on the input text, without requiring explicit domain labels. To balance domain-specific knowledge and generalization across various domains, Sci-LoRA integrates information at both the data and model levels. This dynamic fusion enhances the adaptability and performance across various domains. Experimental results across twelve domains on five public datasets show that Sci-LoRA significantly outperforms state-of-the-art large language models and demonstrates flexible generalization and adaptability in cross-domain lay paraphrasing.
Recent advances in Multimodal Large Language Models (MLLMs) have significantly enhanced the naturalness and flexibility of human computer interaction by enabling seamless understanding across text, vision, and audio modalities. Among these, voice enabled models such as SpeechGPT have demonstrated considerable improvements in usability, offering expressive, and emotionally responsive interactions that foster deeper connections in real world communication scenarios. However, the use of voice introduces new security risks, as attackers can exploit the unique characteristics of spoken language, such as timing, pronunciation variability, and speech to text translation, to craft inputs that bypass defenses in ways not seen in text-based systems. Despite substantial research on text based jailbreaks, the voice modality remains largely underexplored in terms of both attack strategies and defense mechanisms. In this work, we present an adversarial attack targeting the speech input of aligned MLLMs in a white box scenario. Specifically, we introduce a novel token level attack that leverages access to the model's speech tokenization to generate adversarial token sequences. These sequences are then synthesized into audio prompts, which effectively bypass alignment safeguards and to induce prohibited outputs. Evaluated on SpeechGPT, our approach achieves up to 89 percent attack success rate across multiple restricted tasks, significantly outperforming existing voice based jailbreak methods. Our findings shed light on the vulnerabilities of voice-enabled multimodal systems and to help guide the development of more robust next-generation MLLMs.
We introduce the Exemplar-Based Expository Text Generation task, aiming to generate an expository text on a new topic using an exemplar on a similar topic. Current methods fall short due to their reliance on extensive exemplar data, difficulty in adapting topic-specific content, and issues with long-text coherence. To address these challenges, we propose the concept of Adaptive Imitation and present a novel Recurrent Plan-then-Adapt (RePA) framework. RePA leverages large language models (LLMs) for effective adaptive imitation through a fine-grained plan-then-adapt process. RePA also enables recurrent segment-by-segment imitation, supported by two memory structures that enhance input clarity and output coherence. We also develop task-specific evaluation metrics--imitativeness, adaptiveness, and adaptive-imitativeness--using LLMs as evaluators. Experimental results across our collected three diverse datasets demonstrate that RePA surpasses existing baselines in producing factual, consistent, and relevant texts for this task.
Diffusion models have achieved state-of-the-art performance in generating images, audio, and video, but their adaptation to text remains challenging due to its discrete nature. Prior approaches either apply Gaussian diffusion in continuous latent spaces, which inherits semantic structure but struggles with token decoding, or operate in categorical simplex space, which respect discreteness but disregard semantic relation between tokens. In this paper, we propose Smoothing Diffusion on Token Embeddings (Smoothie), a novel diffusion method that combines the strengths of both approaches by progressively smoothing token embeddings based on semantic similarity. This technique enables gradual information removal while maintaining a natural decoding process. Experimental results on several sequence-to-sequence generation tasks demonstrate that Smoothie outperforms existing diffusion-based models in generation quality. Furthermore, ablation studies show that our proposed diffusion space yields better performance than both the standard embedding space and the categorical simplex. Our code is available at https://github.com/ashaba1in/smoothie.
We introduce Grade School Math with Distracting Context (GSM-DC), a synthetic benchmark to evaluate Large Language Models' (LLMs) reasoning robustness against systematically controlled irrelevant context (IC). GSM-DC constructs symbolic reasoning graphs with precise distractor injections, enabling rigorous, reproducible evaluation. Our experiments demonstrate that LLMs are significantly sensitive to IC, affecting both reasoning path selection and arithmetic accuracy. Additionally, training models with strong distractors improves performance in both in-distribution and out-of-distribution scenarios. We further propose a stepwise tree search guided by a process reward model, which notably enhances robustness in out-of-distribution conditions.
Direct Preference Optimization (DPO) has emerged as a promising framework for aligning Large Language Models (LLMs) with human preferences by directly optimizing the log-likelihood difference between chosen and rejected responses. However, existing methods assign equal importance to all tokens in the response, while humans focus on more meaningful parts. This leads to suboptimal preference optimization, as irrelevant or noisy tokens disproportionately influence DPO loss. To address this limitation, we propose \textbf{O}ptimal \textbf{T}ransport-based token weighting scheme for enhancing direct \textbf{P}reference \textbf{O}ptimization (OTPO). By emphasizing semantically meaningful token pairs and de-emphasizing less relevant ones, our method introduces a context-aware token weighting scheme that yields a more contrastive reward difference estimate. This adaptive weighting enhances reward stability, improves interpretability, and ensures that preference optimization focuses on meaningful differences between responses. Extensive experiments have validated OTPO's effectiveness in improving instruction-following ability across various settings\footnote{Code is available at https://github.com/Mimasss2/OTPO.}.
ICD Coding aims to assign a wide range of medical codes to a medical text document, which is a popular and challenging task in the healthcare domain. To alleviate the problems of long-tail distribution and the lack of annotations of code-specific evidence, many previous works have proposed incorporating code knowledge to improve coding performance. However, existing methods often focus on a single type of knowledge and design specialized modules that are complex and incompatible with each other, thereby limiting their scalability and effectiveness. To address this issue, we propose GKI-ICD, a novel, general knowledge injection framework that integrates three key types of knowledge, namely ICD Description, ICD Synonym, and ICD Hierarchy, without specialized design of additional modules. The comprehensive utilization of the above knowledge, which exhibits both differences and complementarity, can effectively enhance the ICD coding performance. Extensive experiments on existing popular ICD coding benchmarks demonstrate the effectiveness of GKI-ICD, which achieves the state-of-the-art performance on most evaluation metrics. Code is available at https://github.com/xuzhang0112/GKI-ICD.
Knowledge editing aims to update the embedded knowledge within Large Language Models (LLMs). However, existing approaches, whether through parameter modification or external memory integration, often suffer from inconsistent evaluation objectives and experimental setups. To address this gap, we conduct a comprehensive benchmarking study. In addition to fact-level datasets, we introduce more complex event-based datasets and general-purpose datasets drawn from other tasks. Our evaluation covers both instruction-tuned and reasoning-oriented LLMs, under a realistic autoregressive inference setting rather than teacher-forced decoding. Beyond single-edit assessments, we also evaluate multi-edit scenarios to better reflect practical demands. We employ four evaluation dimensions, including portability, and compare all recent methods against a simple and straightforward baseline named Selective Contextual Reasoning (SCR). Empirical results reveal that parameter-based editing methods perform poorly under realistic conditions. In contrast, SCR consistently outperforms them across all settings. This study offers new insights into the limitations of current knowledge editing methods and highlights the potential of context-based reasoning as a more robust alternative.
Machine translation (MT) systems that support low-resource languages often struggle on specialized domains. While researchers have proposed various techniques for domain adaptation, these approaches typically require model fine-tuning, making them impractical for non-technical users and small organizations. To address this gap, we propose Tulun, a versatile solution for terminology-aware translation, combining neural MT with large language model (LLM)-based post-editing guided by existing glossaries and translation memories. Our open-source web-based platform enables users to easily create, edit, and leverage terminology resources, fostering a collaborative human-machine translation process that respects and incorporates domain expertise while increasing MT accuracy. Evaluations show effectiveness in both real-world and benchmark scenarios: on medical and disaster relief translation tasks for Tetun and Bislama, our system achieves improvements of 16.90-22.41 ChrF++ points over baseline MT systems. Across six low-resource languages on the FLORES dataset, Tulun outperforms both standalone MT and LLM approaches, achieving an average improvement of 2.8 ChrF points over NLLB-54B.
Large Language Models (LLMs) have achieved remarkable success in Natural Language Processing (NLP), yet their cross-lingual performance consistency remains a significant challenge. This paper introduces a novel methodology for efficiently identifying inherent cross-lingual weaknesses in LLMs. Our approach leverages beam search and LLM-based simulation to generate bilingual question pairs that expose performance discrepancies between English and target languages. We construct a new dataset of over 6,000 bilingual pairs across 16 languages using this methodology, demonstrating its effectiveness in revealing weaknesses even in state-of-the-art models. The extensive experiments demonstrate that our method precisely and cost-effectively pinpoints cross-lingual weaknesses, consistently revealing over 50\% accuracy drops in target languages across a wide range of models. Moreover, further experiments investigate the relationship between linguistic similarity and cross-lingual weaknesses, revealing that linguistically related languages share similar performance patterns and benefit from targeted post-training. Code is available at https://github.com/xzx34/Cross-Lingual-Pitfalls.