China

Mamba&#39;s theoretical infinite-context potential is limited in practice when sequences far exceed training lengths. This work explores unlocking Mamba&#39;s long-context memory ability by a simple-yet-effective method, Recall with Reasoning (RwR), by distilling chain-of-thought (CoT) summarization from a teacher model. Specifically, RwR prepends these summarization as CoT prompts during fine-tuning, teaching Mamba to actively recall and reason over long contexts. Experiments on LONGMEMEVAL and HELMET show that RwR outperforms existing long-term memory methods on the Mamba model. Furthermore, under similar pre-training conditions, RwR improves the long-context performance of Mamba relative to comparable Transformer/hybrid baselines while preserving short-context capabilities, all without changing the architecture.

EMNLP 2025

Recall with Reasoning: Chain-of-Thought Distillation for Mamba’s Long-Context Memory and Extrapolation

mamba

memory

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community."<br>

*Message from the General Chair, Dirk Hovy*





<br>

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

Mamba's theoretical infinite-context potential is limited in practice when sequences far exceed training lengths. This work explores unlocking Mamba's long-context memory ability by a simple-yet-effective method, Recall with Reasoning (RwR), by distilling chain-of-thought (CoT) summarization from a teacher model. Specifically, RwR prepends these summarization as CoT prompts during fine-tuning, teaching Mamba to actively recall and reason over long contexts. Experiments on LONGMEMEVAL and HELMET show that RwR outperforms existing long-term memory methods on the Mamba model. Furthermore, under similar pre-training conditions, RwR improves the long-context performance of Mamba relative to comparable Transformer/hybrid baselines while preserving short-context capabilities, all without changing the architecture.

poster

As large language models (LLMs) often generate plausible but incorrect content, error detection has become increasingly critical to ensure truthfulness. However, existing detection methods often overlook a critical problem we term as self-consistent error, where LLMs repeatedly generate the same incorrect response across multiple stochastic samples. This work formally defines **self-consistent errors** and evaluates mainstream detection methods on them. Our investigation reveals two key findings: (1) Unlike inconsistent errors, whose frequency diminishes significantly as LLM scale increases, the frequency of self-consistent errors remains stable or even increases. (2) All four types of detection methods significantly struggle to detect self-consistent errors. These findings reveal critical limitations in current detection methods and underscore the need for improved methods. Motivated by the observation that self-consistent errors often differ across LLMs, we propose a simple but effective cross‑model probe method that fuses hidden state evidence from an external verifier LLM. Our method significantly enhances performance on self-consistent errors across three LLM families.

Too Consistent to Detect: A Study of Self-Consistent Errors in LLMs

Existing studies have optimized retrieval-augmented generation (RAG) across various sub-tasks, such as query understanding and retrieval refinement, but integrating these optimizations into a unified framework remains challenging. To tackle this problem, this work proposes RoleRAG, a unified RAG framework that achieves efficient multi-task processing through role-specific token optimization. RoleRAG comprises six modules, each handling a specific sub-task within the RAG process. Additionally, we introduce a query graph to represent the decomposition of the query, which can be dynamically resolved according to the decomposing state. All modules are driven by the same underlying LLM, distinguished by task-specific role tokens that are individually optimized. This design allows RoleRAG to dynamically activate different modules within a single LLM instance, thereby streamlining deployment and reducing resource consumption. Experimental results on five open-domain question-answering datasets demonstrate the effectiveness, generalizability, and flexibility of our framework.

Single LLM, Multiple Roles: A Unified Retrieval-Augmented Generation Framework Using Role-Specific Token Optimization

The proliferation of Machine Learning as a Service (MLaaS) has enabled widespread deployment of large language models (LLMs) via cloud APIs, but also raises critical concerns about model integrity and security. Existing black-box tamper detection methods, such as watermarking and fingerprinting, rely on the stability of model outputs—a property that does not hold for inherently stochastic LLMs. In this paper, we propose a novel fingerprinting approach tailored for LLM tamper detection in the black-box setting. By formulating the problem as hypothesis testing, we introduce a Regularized Entropy-Sensitive Fingerprinting (RESF) objective that leverages a first-order surrogate for KL divergence to maximize sensitivity while controlling false positives. To robustly distinguish genuine tampering from benign temperature changes, we develop a lightweight two-tier sequential test combining support-based and distributional checks, with rigorous control of the global false-alarm rate. Our method is supported by comprehensive theoretical analysis and explicit performance guarantees. Extensive experiments across multiple LLMs and tampering scenarios demonstrate that RESF achieves up to 98.80\% detection accuracy, even under minimal LoRA fine-tuning modifications, establishing a new standard for practical and reliable LLM integrity verification.

RESF: Regularized-Entropy-Sensitive Fingerprinting for Black-Box Tamper Detection of Large Language Models

In this paper, we introduce MIO, a novel foundation model built on multimodal tokens, capable of understanding and generating speech, text, images, and videos in an end-to-end, autoregressive manner. While the emergence of large language models (LLMs) and multimodal large language models (MM-LLMs) propels advancements in artificial general intelligence through their versatile capabilities, they still lack true any-to-any understanding and generation. Recently, the release of GPT-4o has showcased the remarkable potential of any-to-any LLMs for complex real-world tasks, enabling omnidirectional input and output across images, speech, and text. However, it is closed-source and does not support the generation of multimodal interleaved sequences. To address this gap, we present MIO, which is trained on a mixture of discrete tokens across four modalities using causal multimodal modeling. MIO undergoes a four-stage training process: (1) alignment pre-training, (2) interleaved pre-training, (3) speech-enhanced pre-training, and (4) comprehensive supervised fine-tuning on diverse textual, visual, and speech tasks. Our experimental results indicate that MIO exhibits competitive, and in some cases superior, performance compared to previous dual-modal baselines, any-to-any model baselines, and even modality-specific baselines. Moreover, MIO demonstrates advanced capabilities inherent to its any-to-any feature, such as interleaved video-text generation, chain-of-visual-thought reasoning, visual guideline generation, instructional image editing, etc.

MIO: A Foundation Model on Multimodal Tokens

Current research on long-form context in Large Language Models (LLMs) primarily focuses on the understanding of long-contexts, the **Open-ended Long Text Generation** (Open-LTG) remains insufficiently explored. Training a long text generation model requires curation of gold-standard reference data, which is typically nonexistent for informative Open-LTG tasks. However, previous methods only utilize general assessments as reward signals, which limits accuracy. To bridge this gap, we introduce **ProxyReward**, an innovative reinforcement learning (RL) based framework, which includes a data synthesis method and a novel reward signal. Firstly, **ProxyReward Dataset** synthesis is accomplished through simple prompts that enables the model to create automatically, obviating extensive labeled data or significant manual effort. Secondly, **ProxyReward Signal** offers a targeted evaluation of information comprehensiveness and accuracy for specific questions. The experimental results indicate that our method ProxyReward **surpasses even GPT-4-Turbo**. It can significantly enhance performance by 20% on the Open-LTG task when training widely used open-source models, while also surpassing the LLM-as-a-Judge approach. Our work presents effective methods to enhance the ability of LLMs to address complex open-ended questions posed by humans.

From General Reward to Targeted Reward: Improving Open-ended Long-context Generation Models

Selecting high-quality and diverse training samples from extensive datasets plays a crucial role in reducing training overhead and enhancing the performance of Large Language Models (LLMs). However, existing studies fall short in assessing the overall value of selected data, focusing primarily on individual quality, and struggle to strike an effective balance between ensuring diversity and minimizing data point traversals. Therefore, this paper introduces a novel choice-based sample selection framework that shifts the focus from evaluating individual sample quality to comparing the contribution value of different samples when incorporated into the subset. Thanks to the advanced language understanding capabilities of LLMs, we utilize LLMs to evaluate the value of each option during the selection process. Furthermore, we design a greedy sampling process where samples are incrementally added to the subset, thereby improving efficiency by eliminating the need for exhaustive traversal of the entire dataset with the limited budget. Extensive experiments demonstrate that selected data from our method not only surpasses the performance of the full dataset but also achieves competitive results with state-of-the-art (SOTA) studies, while requiring fewer selections. Moreover, we validate our approach on a larger medical dataset, highlighting its practical applicability in real-world applications.

Add-One-In: Incremental Sample Selection for Large Language Models via a Choice-Based Greedy Paradigm

The reward model (RM) plays a crucial role in aligning Large Language Models (LLMs) with human preferences through Reinforcement Learning, where the Bradley-Terry (BT) objective has been recognized as simple yet powerful, specifically for pairwise preference comparison. However, BT-based RMs often struggle to effectively distinguish between similar preference responses, leading to insufficient separation between preferred and non-preferred outputs. Consequently, they may easily overfit easy samples and cannot generalize well to Out-Of-Distribution (OOD) samples, resulting in suboptimal performance. To address these challenges, this paper introduces an effective enhancement to BT-based RMs through an adaptive margin mechanism. Specifically, we design to dynamically adjust the RM focus on more challenging samples through margins, based on both semantic similarity and model-predicted reward differences that is approached from a distributional perspective solvable with Optimal Transport (OT). By incorporating these factors into a principled OT cost matrix design, our adaptive margin enables the RM to better capture distributional differences between chosen and rejected responses, yielding significant improvements in performance, convergence speed, and generalization capabilities. Experimental results across multiple benchmarks demonstrate that our method significantly outperforms several existing RM techniques, showcasing enhanced performance in both In-Distribution (ID) and OOD settings.

APLOT: Robust Reward Modeling via Adaptive Preference Learning with Optimal Transport

Large Language Models (LLMs) have demonstrated remarkable capabilities in various tasks, yet code generation remains a major challenge. Despite the abundant sources of code data, constructing high-quality training datasets at scale poses a significant challenge. Pre-training code data typically suffers from inconsistent data quality issues. Conversely, instruction-based methods which use a high-quality subset as seed samples suffer from limited task diversity. In this paper, we introduce \textbf{UnitCoder}, which directly supervises pre-training data quality through automatically generated unit tests, while ensuring the correctness via an iterative fix and refine flow. Code synthesized by UnitCoder benefits from both the diversity of pre-training corpora and the high quality ensured by unit test supervision. Our experiments demonstrate that models fine-tuned on our synthetic dataset exhibit consistent performance improvements. Our work presents a scalable approach that leverages model-generated unit tests to guide the synthesis of high-quality code data from pre-training corpora, demonstrating the potential for producing diverse and high-quality post-training data at scale. All code and data will be released.

UnitCoder: Scalable Code Synthesis from Pre-training Corpora

Current LLM evaluations often rely on single instruction templates, overlooking models' sensitivity to instruction style—a critical aspect for real-world deployments. We present RCScore, a multi-dimensional framework quantifying how instruction formulation affects model responses. By systematically transforming benchmark problems into multiple instruction styles, RCScore reveals performance variations undetected by conventional metrics. Our experiments across ten LLMs on four reasoning benchmarks demonstrate that instruction style can shift accuracy by up to 16.7% points. We introduce Cross-Response Similarity (CRS), a method applying RCScore metrics to measure stylistic self-consistency, and establish its strong correlation with task accuracy, suggesting consistency as a valuable proxy for model reliability. Additional findings show that deterministic decoding produces more stylistically stable outputs, and model scale correlates positively with cross-style consistency. RCScore offers a principled approach to assess instruction robustness, enabling more comprehensive model evaluation beyond traditional accuracy-focused metrics.

RCScore: Quantifying Response Consistency in Large Language Models

This paper introduces ThoughtProbe, a novel inference-time framework that leverages the hidden reasoning features of Large Language Models (LLMs) to improve their reasoning performance. Unlike previous works that manipulate the hidden representations to steer LLM generation, we harness them as discriminative signals to guide the tree-structured response space exploration. In each node expansion, a classifier serves as a scoring and ranking mechanism that efficiently allocates computational resources by prioritizing higher score candidates for continuation. After completing the tree expansion, we collect answers from all branches to form a candidate answer pool. We then propose a branch-aggregation method that marginalizes over all supporting branches by aggregating their CoT scores, thereby identifying the optimal answer from the pool. Experimental results show that our framework's comprehensive exploration not only covers valid reasoning chains but also effectively identifies them, achieving significant improvements across multiple arithmetic reasoning benchmarks.

Downloads

Next from EMNLP 2025

Too Consistent to Detect: A Study of Self-Consistent Errors in LLMs

Similar lecture

Exploring the Limitations of Mamba in COPY and CoT Reasoning

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES