论文检索

Hao Sun, Yingyan Hou, Jiayan Guo, Bo Wang, Chunyu Yang, Jinsong Ni, Yan Zhang

Document retrieval in real-world scenarios faces significant challenges due to diverse document formats and modalities. Traditional text-based approaches rely on tailored parsing techniques that disregard layout information and are prone to errors, while recent parsing-free visual methods often struggle to capture fine-grained textual semantics in text-rich scenarios. To address these limitations, we propose Unveil, a novel visual-textual embedding framework that effectively integrates textual and visual features for robust document representation. Through knowledge distillation, we transfer the semantic understanding capabilities from the visual-textual embedding model to a purely visual model, enabling efficient parsing-free retrieval while preserving semantic fidelity. Experimental results demonstrate that our visual-textual embedding method surpasses existing approaches, while knowledge distillation successfully bridges the performance gap between visual-textual and visual-only methods, improving both retrieval accuracy and efficiency.

Haoxiang Sun, Ruize Gao, Pei Zhang, Baosong Yang, Rui Wang

Model alignment methods like Direct Preference Optimization and Contrastive Preference Optimization have enhanced machine translation performance by leveraging preference data to enable models to reject suboptimal outputs. During preference data construction, previous approaches primarily rely on humans, strong models like GPT4 or model self-sampling. In this study, we first explain the shortcomings of this practice. Then, we propose Self-Supervised Preference Optimization (SSPO), a novel framework which efficiently constructs translation preference data for iterative DPO training. Applying SSPO to 14B parameters large language models (LLMs) achieves comparable or better performance than GPT-4o on FLORES and multi-domain test datasets. We release an augmented MQM dataset in https://github.com/sunny-sjtu/MQM-aug.

Marc Feger, Katarina Boland, Stefan Dietze

Identifying arguments is a necessary prerequisite for various tasks in automated discourse analysis, particularly within contexts such as political debates, online discussions, and scientific reasoning. In addition to theoretical advances in understanding the constitution of arguments, a significant body of research has emerged around practical argument mining, supported by a growing number of publicly available datasets. On these benchmarks, BERT-like transformers have consistently performed best, reinforcing the belief that such models are broadly applicable across diverse contexts of debate. This study offers the first large-scale re-evaluation of such state-of-the-art models, with a specific focus on their ability to generalize in identifying arguments. We evaluate four transformers, three standard and one enhanced with contrastive pre-training for better generalization, on 17 English sentence-level datasets as most relevant to the task. Our findings show that, to varying degrees, these models tend to rely on lexical shortcuts tied to content words, suggesting that apparent progress may often be driven by dataset-specific cues rather than true task alignment. While the models achieve strong results on familiar benchmarks, their performance drops markedly when applied to unseen datasets. Nonetheless, incorporating both task-specific pre-training and joint benchmark training proves effective in enhancing both robustness and generalization.

Zhengyu Chen, Siqi Wang, Teng Xiao, Yudong Wang, Shiqi Chen, Xunliang Cai, Junxian He, Jingang Wang

Traditional scaling laws in natural language processing suggest that increasing model size and training data enhances performance. However, recent studies reveal deviations, particularly in large language models, where performance improvements decelerate—a phenomenon known as sub-scaling. This paper revisits these scaling laws by examining the impact of data quality and training strategies on model performance. Through extensive empirical analysis of over 400 models, we identify high data density and non-optimal resource allocation as key factors contributing to sub-scaling. High data density leads to diminishing returns due to redundant information, while optimal resource allocation is crucial for sustained performance improvements. We propose a sub-optimal scaling law that better predicts performance in sub-scaling regimes, highlighting the importance of data quality and diversity.

Junjie Wu, Gefei Gu, Yanan Zheng, Dit-Yan Yeung, Arman Cohan

Long-context language models (LCLMs) have exhibited impressive capabilities in long-context understanding tasks. Among these, long-context referencing—a crucial task that requires LCLMs to attribute items of interest to specific parts of long-context data—remains underexplored. To bridge this gap, this paper proposes Referencing Evaluation for Long-context Language Models (Ref-Long), a novel benchmark designed to assess the long-context referencing capability of LCLMs. Specifically, Ref-Long requires LCLMs to identify the indexes of documents that reference a specific key, emphasizing contextual relationships between the key and the documents over simple retrieval. Based on the task design, we construct three subsets ranging from synthetic to realistic scenarios to form the Ref-Long benchmark. Experimental results of 13 LCLMs reveal significant shortcomings in long-context referencing, even among advanced models like GPT-4o. To further investigate these challenges, we conduct comprehensive analyses, including human evaluations, task format adjustments, fine-tuning experiments, and error analyses, leading to several key insights. Our data and code will be publicly released, and the data is also attached in the submission.

Yutong Liu, Lida Shi, Rui Song, Hao Xu

Negotiation agents need to influence the attitudes or intentions of users to reach a consensus. Strategy planning and expressive optimization are crucial aspects of effective negotiations. However, previous studies have typically focused on only one of these aspects, neglecting the fact that their combined synergistic effect can lead to better performance. Inspired by the dual-process theory in human cognition, we propose a Dual-Mind Negotiation Agent (DMNA) framework. This framework integrates an intuitive module for rapid, experience-based response and a deliberative module for slow, expression optimization. The intuitive module is trained using Monte Carlo Tree Search (MCTS) and Direct Preference Optimization (DPO), enabling it to make suitable strategic planning and expression. The deliberative module employs a multifaceted reflexion mechanism to enhance the quality of expression. Experiments conducted on negotiation datasets confirm that DMNA achieves state-of-the-art results, demonstrating an enhancement in the negotiation ability of agents.

Diana Galvan-Sosa, Gabrielle Gaudeau, Pride Kavumba, Yunmeng Li, Hongyi Gu, Zheng Yuan, Keisuke Sakaguchi, Paula Buttery

The performance and usability of Large-Language Models (LLMs) are driving their use in explanation generation tasks. However, despite their widespread adoption, LLM explanations have been found to be unreliable, making it difficult for users to distinguish good from bad explanations. To address this issue, we present Rubrik’s CUBE–an education-inspired rubric and a dataset of 26k explanations, written and later quality-annotated using the rubric by both humans and six open- and closed-source LLMs. The CUBE dataset focuses on two reasoning and two language tasks, providing the necessary diversity for us to effectively test our proposed rubric. Using Rubrik, we find that explanations are influenced by both task and perceived difficulty. Low quality stems primarily from a lack of conciseness in LLM-generated explanations, rather than cohesion and word choice. The full dataset, rubric, and code are available at https://github.com/RubriksCube/rubriks_cube.

Haozhen Zhang, Tao Feng, Jiaxuan You

Retrieval-augmented generation (RAG) has revitalized Large Language Models (LLMs) by injecting non-parametric factual knowledge. Compared with long-context LLMs, RAG is considered an effective summarization tool in a more concise and lightweight manner, which can interact with LLMs multiple times using diverse queries to get comprehensive responses. However, the LLM-generated historical responses, which contain potentially insightful information, are largely neglected and discarded by existing approaches, leading to suboptimal results. In this paper, we propose \textit{graph of records} (\textbf{GoR}), which leverages historical responses generated by LLMs to enhance RAG for long-context global summarization. Inspired by the \textit{retrieve-then-generate} paradigm of RAG, we construct a graph by establishing an edge between the retrieved text chunks and the corresponding LLM-generated response. To further uncover the intricate correlations between them, GoR features a \textit{graph neural network} and an elaborately designed \textit{BERTScore}-based objective for self-supervised model training, enabling seamless supervision signal backpropagation between reference summaries and node embeddings. We comprehensively compare GoR with 12 baselines across four long-context summarization datasets, and the results indicate that our proposed method reaches the best performance (\textit{e.g.}, 15%, 8%, and 19% improvement over retrievers w.r.t. Rouge-L, Rouge-1, and Rouge-2 on the WCEP dataset). Extensive experiments further demonstrate the effectiveness of GoR.

Ruiyang Xu, Jialun Cao, Yaojie Lu, Ming Wen, Hongyu Lin, Xianpei Han, Ben He, Shing-Chi Cheung, Le Sun

Code benchmarks such as HumanEval are widely adopted to evaluate Large Language Models’ (LLMs) coding capabilities. However, there is an unignorable programming language bias in existing code benchmarks – over 95% code generation benchmarks are dominated by Python, leaving the LLMs’ capabilities in other programming languages such as Java and C/C++ unknown. Moreover, coding task bias is also crucial. Most benchmarks focus on code generation capability, while benchmarks for code reasoning (given input, reasoning output; and given output, reasoning input), an essential coding capability, are insufficient. Yet, constructing multi-lingual benchmarks can be expensive and labor-intensive, and codes in contest websites such as Leetcode suffer from data contamination during training. To fill this gap, we propose CRUXEVAL-X, a multi-lingual code reasoning benchmark that contains 19 programming languages. It comprises at least 600 subjects for each language, along with 19K content-consistent tests in total. In particular, the construction pipeline of CRUXEVAL-X works in a fully automated and test-guided manner, which iteratively generates and repairs based on execution feedback. Also, to cross language barriers (e.g., dynamic/static type systems in Python/C++), we formulated various transition rules between language pairs to facilitate translation. Our intensive evaluation of 24 representative LLMs reveals the correlation between language pairs. For example, TypeScript and JavaScript show a significant positive correlation, while Racket has less correlation with other languages. More interestingly, even a model trained solely on Python can achieve at most 34.4% Pass@1 in other languages, revealing the cross-language generalization of LLMs.

Qiming Ge, Shuhao Xing, Songyang Gao, Yunhua Zhou, Yicheng Zou, Songyang Zhang, Zhi Chen, Hang Yan, Qi Zhang, Qipeng Guo 等

Scaling law builds the relationship between training computation and validation loss, enabling researchers to effectively predict the loss trending of models across different levels of computation. However, a gap still remains between validation loss and the model’s downstream capabilities, making it untrivial to apply scaling law to direct performance prediction for downstream tasks. The loss typically represents a cumulative penalty for predicted tokens, which are implicitly considered to have equal importance. Nevertheless, our studies have shown evidence that when considering different training data distributions, we cannot directly model the relationship between downstream capability and computation or token loss. To bridge the gap between validation loss and downstream task capabilities, in this work, we introduce Capability Salience Vector, which decomposes the overall loss and assigns different importance weights to tokens to assess a specific meta-capability, aligning the validation loss with downstream task performance in terms of the model’s capabilities. Experiments on various popular benchmarks demonstrate that our proposed Capability Salience Vector could significantly improve the predictability of language model performance on downstream tasks.

Beatrix Miranda Ginn Nielsen, Iuri Macocco, Marco Baroni

Hubness, the tendency for a few points to be among the nearest neighbours of a disproportionate number of other points, commonly arises when applying standard distance measures to high-dimensional data, often negatively impacting distance-based analysis. As autoregressive large language models (LLMs) operate on high-dimensional representations, we ask whether they are also affected by hubness. We first prove that the only large-scale representation comparison operation performed by LLMs, namely that between context and unembedding vectors to determine continuation probabilities, is not characterized by the concentration of distances phenomenon that typically causes the appearance of nuisance hubness. We then empirically show that this comparison still leads to a high degree of hubness, but the hubs in this case do not constitute a disturbance. They are rather the result of context-modulated frequent tokens often appearing in the pool of likely candidates for next token prediction. However, when other distances are used to compare LLM representations, we do not have the same theoretical guarantees, and, indeed, we see nuisance hubs appear. There are two main takeaways. First, hubness, while omnipresent in high-dimensional spaces, is not a negative property that needs to be mitigated when LLMs are being used for next token prediction. Second, when comparing representations from LLMs using Euclidean or cosine distance, there is a high risk of nuisance hubs and practitioners should use mitigation techniques if relevant.

Ukyo Honda, Tatsushi Oka

In-context learning (ICL) has emerged as a successful paradigm for leveraging large language models (LLMs). However, it often struggles to generalize beyond the distribution of the provided demonstrations. A recent advancement in enhancing robustness is ICL with explanations (X-ICL), which improves prediction reliability by guiding LLMs to understand and articulate the reasoning behind correct labels. Building on this approach, we introduce an advanced framework that extends X-ICL by systematically exploring explanations for all possible labels (X^2-ICL), thereby enabling more comprehensive and robust decision-making. Experimental results on multiple natural language understanding datasets validate the effectiveness of X^2-ICL, demonstrating significantly improved robustness to out-of-distribution data compared to the existing ICL approaches.

Yooseop Lee, Suin Kim, Yohan Jo

In designing multiple-choice questions (MCQs) in education, creating plausible distractors is crucial for identifying students’ misconceptions and gaps in knowledge and accurately assessing their understanding. However, prior studies on distractor generation have not paid sufficient attention to enhancing the difficulty of distractors, resulting in reduced effectiveness of MCQs. This study presents a pipeline for training a model to generate distractors that are more likely to be selected by students. First, we train a pairwise ranker to reason about students’ misconceptions and assess the relative plausibility of two distractors. Using this model, we create a dataset of pairwise distractor ranks and then train a distractor generator via Direct Preference Optimization (DPO) to generate more plausible distractors. Experiments on computer science subjects (Python, DB, MLDL) demonstrate that our pairwise ranker effectively identifies students’ potential misunderstandings and achieves ranking accuracy comparable to human experts. Furthermore, our distractor generator outperforms several baselines in generating plausible distractors and produces questions with a higher item discrimination index (DI).

Yongquan He, Wenyuan Zhang, Xuancheng Huang, Peng Zhang, Lingxun Meng, Xiang Zhou, Ke Zeng, Xunliang Cai

Instruction tuning for large language models (LLMs) can drive them to produce results consistent with human goals in specific downstream tasks. However, the process of continual instruction tuning (CIT) for LLMs may bring about the catastrophic forgetting (CF) problem, where previously learned abilities are degraded. Recent methods try to alleviate the CF problem by modifying models or replaying data, which may only remember the surface-level pattern of instructions and get confused on held-out tasks. In this paper, we propose a novel continual instruction tuning method based on Key-part Information Gain (KPIG). Our method computes the information gain on masked parts to dynamically replay data and refine the training objective, which enables LLMs to capture task-aware information relevant to the correct response and alleviate overfitting to general descriptions in instructions. In addition, we propose two metrics, P-score and V-score, to measure the generalization and instruction-following abilities of LLMs. Experiments demonstrate our method achieves superior performance on both seen and held-out tasks.

Parth Thakkar, Ankush Agarwal, Prasad Kasu, Pulkit Bansal, Chaitanya Devaguptapu

While Multi-modal Large Language Models (MLLMs) have shown impressive capabilities in document understanding tasks, their ability to locate and reason about fine-grained details within complex documents remains understudied. Consider searching a restaurant menu for a specific nutritional detail or identifying a disclaimer in a lengthy newspaper article — tasks that demand careful attention to small but significant details within a broader narrative, akin to Finding Needles in Images (NiM). To address this gap, we introduce NiM-Benchmark, a carefully curated benchmark spanning diverse real-world documents including newspapers, menus, and lecture images, specifically designed to evaluate MLLMs’ capability in these intricate tasks. Building on this, we further propose Spot-IT, a simple yet effective approach that enhances MLLMs capability through intelligent patch selection and Gaussian attention, motivated from how humans zoom and focus when searching documents. Our extensive experiments reveal both the capabilities and limitations of current MLLMs in handling fine-grained document understanding tasks, while demonstrating the effectiveness of our approach. Spot-IT achieves significant improvements over baseline methods, particularly in scenarios requiring precise detail extraction from complex layouts.

Deniz Ekin Yavas, Timothée Bernard, Benoît Crabbé, Laura Kallmeyer

Topological properties of embeddings, such as isotropy and uniformity, are closely linked to their expressiveness, and improving these properties enhances the embeddings’ ability to capture nuanced semantic distinctions. However, fine-tuning can reduce the expressiveness of the embeddings of language models. This study investigates the relation between fine-tuning, topology of the embedding space, and task performance in the context of sense knowledge enhancement, focusing on identifying the topological properties that contribute to the success of sense-enhanced embeddings. We experiment with two fine-tuning methods: Supervised Contrastive Learning (SCL) and Supervised Predictive Learning (SPL). Our results show that SPL, the most standard approach, exhibits varying effectiveness depending on the language model and is inconsistent in producing successful sense-enhanced embeddings. In contrast, SCL achieves this consistently. Furthermore, while the embeddings with only increased sense-alignment show reduced task performance, those that also exhibit high isotropy and balance uniformity with sense-alignment achieve the best results. Additionally, our findings indicate that supervised and unsupervised tasks benefit from these topological properties to varying degrees.

Siyin Wang, Wenyi Yu, Xianzhao Chen, Xiaohai Tian, Jun Zhang, Lu Lu, Yu Tsao, Junichi Yamagishi, Yuxuan Wang, Chao Zhang

This paper explores a novel perspective to speech quality assessment by leveraging natural language descriptions, offering richer, more nuanced insights than traditional numerical scoring methods. Natural language feedback provides instructive recommendations and detailed evaluations, yet existing datasets lack the comprehensive annotations needed for this approach. To bridge this gap, we introduce QualiSpeech, a comprehensive low-level speech quality assessment dataset encompassing 11 key aspects and detailed natural language comments that include reasoning and contextual insights. Additionally, we propose the QualiSpeech Benchmark to evaluate the low-level speech understanding capabilities of auditory large language models (LLMs). Experimental results demonstrate that finetuned auditory LLMs can reliably generate detailed descriptions of noise and distortion, effectively identifying their types and temporal characteristics. The results further highlight the potential for incorporating reasoning to enhance the accuracy and reliability of quality assessments. The dataset can be found at https://huggingface.co/datasets/tsinghua-ee/QualiSpeech.

Weixiang Zhao, Yulin Hu, Yang Deng, Tongtong Wu, Wenxuan Zhang, Jiahe Guo, An Zhang, Yanyan Zhao, Bing Qin, Tat-Seng Chua 等

Large language models (LLMs) have become increasingly central to AI applications worldwide, necessitating robust multilingual safety alignment to ensure secure deployment across diverse linguistic contexts. Existing preference learning methods for safety alignment, such as RLHF and DPO, are primarily monolingual and struggle with noisy multilingual data. To address these limitations, we introduce Multilingual reward gaP Optimization (MPO), a novel approach that leverages the well-aligned safety capabilities of the dominant language (e.g., English) to improve safety alignment across multiple languages. MPO directly minimizes the reward gap difference between the dominant language and target languages, effectively transferring safety capabilities while preserving the original strengths of the dominant language. Extensive experiments on three LLMs, LLaMA-3.1, Gemma-2 and Qwen2.5, validate MPO’s efficacy in multilingual safety alignment without degrading general multilingual utility.

Zhiyu Shen, Yunhe Pang, Yanghui Rao, Jianxing Yu

Emotion Recognition in Conversations (ERC) is crucial for machines to understand dynamic human emotions. While Large Language Models (LLMs) show promise, their performance is often limited by challenges in interpreting complex conversational streams. We introduce a Clue of Emotion (CoE) framework, which progressively integrates key conversational clues to enhance the ERC task. Building on CoE, we implement a multi-stage auxiliary learning strategy that incorporates role-playing, speaker identification, and emotion reasoning tasks, each targeting different aspects of conversational emotion understanding and enhancing the model’s ability to interpret emotional contexts. Our experiments on EmoryNLP, MELD, and IEMOCAP demonstrate that CoE consistently outperforms state-of-the-art methods, achieving a 2.92% improvement on EmoryNLP. These results underscore the effectiveness of clues and multi-stage auxiliary learning for ERC, offering valuable insights for future research.

Zehan Wang, Ke Lei, Chen Zhu, Jiawei Huang, Sashuai Zhou, Luping Liu, Xize Cheng, Shengpeng Ji, Zhenhui Ye, Tao Jin 等

Text-to-audio (T2A) generation has achieved remarkable progress in generating a variety of audio outputs from language prompts. However, current state-of-the-art T2A models still struggle to satisfy human preferences for prompt-following and acoustic quality when generating complex multi-event audio. To improve the performance of the model in these high-level applications, we propose to enhance the basic capabilities of the model with AI feedback learning. First, we introduce fine-grained AI audio scoring pipelines to: 1) verify whether each event in the text prompt is present in the audio (Event Occurrence Score), 2) detect deviations in event sequences from the language description (Event Sequence Score), and 3) assess the overall acoustic and harmonic quality of the generated audio (Acoustic&Harmonic Quality). We evaluate these three automatic scoring pipelines and find that they correlate significantly better with human preferences than other evaluation metrics. This highlights their value as both feedback signals and evaluation metrics. Utilizing our robust scoring pipelines, we construct a large audio preference dataset, T2A-FeedBack, which contains 41k prompts and 249k audios, each accompanied by detailed scores. Moreover, we introduce T2A-EpicBench, a benchmark that focuses on long captions, multi-events, and story-telling scenarios, aiming to evaluate the advanced capabilities of T2A models. Finally, we demonstrate how T2A-FeedBack can enhance current state-of-the-art audio model. With simple preference tuning, the audio generation model exhibits significant improvements in both simple (AudioCaps test set) and complex (T2A-EpicBench) scenarios.