China

Step-by-step reasoning has become a standard approach for large language models (LLMs) to tackle complex tasks. While this paradigm has proven effective, it raises a fundamental question: How can we verify that an LLM&#39;s reasoning is accurately grounded in knowledge? To address this question, we introduce a novel evaluation suite that systematically assesses the knowledge grounding of intermediate reasoning. Our framework comprises three key components. (1) Principal Knowledge Collection, a large-scale repository of atomic knowledge essential for reasoning. Based on the collection, we propose (2) knowledge-grounded evaluation metrics designed to measure how well models recall and apply prerequisite knowledge in reasoning. These metrics are computed by our (3) evaluator LLM, a lightweight model optimized for cost-effective and reliable metric computation. Our evaluation suite demonstrates remarkable effectiveness in identifying missing or misapplied knowledge elements, providing crucial insights for uncovering fundamental reasoning deficiencies in LLMs. Beyond evaluation, we demonstrate how these metrics can be integrated into preference optimization, showcasing further applications of knowledge-grounded evaluation.

EMNLP 2025

Assessing LLM Reasoning Steps via Principal Knowledge Grounding

llm knowledge usage evaluation

llm reasoning evaluation

evaluation methodology

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

AAAI 2026

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

Step-by-step reasoning has become a standard approach for large language models (LLMs) to tackle complex tasks. While this paradigm has proven effective, it raises a fundamental question: How can we verify that an LLM's reasoning is accurately grounded in knowledge? To address this question, we introduce a novel evaluation suite that systematically assesses the knowledge grounding of intermediate reasoning. Our framework comprises three key components. (1) Principal Knowledge Collection, a large-scale repository of atomic knowledge essential for reasoning. Based on the collection, we propose (2) knowledge-grounded evaluation metrics designed to measure how well models recall and apply prerequisite knowledge in reasoning. These metrics are computed by our (3) evaluator LLM, a lightweight model optimized for cost-effective and reliable metric computation. Our evaluation suite demonstrates remarkable effectiveness in identifying missing or misapplied knowledge elements, providing crucial insights for uncovering fundamental reasoning deficiencies in LLMs. Beyond evaluation, we demonstrate how these metrics can be integrated into preference optimization, showcasing further applications of knowledge-grounded evaluation.

poster

## 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." 

*Message from the General Chair, Dirk Hovy*

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

Recent work shows that post-training datasets for LLMs can be substantially downsampled without noticeably deteriorating performance. However, data selection often incurs high computational costs or is limited to narrow domains. In this paper, we demonstrate that data selection can be both---efficient and universal---by using a multi-step pipeline in which we efficiently bin data points into groups, estimate quality using specialized models, and score difficulty with a robust, lightweight method. Task-based categorization allows us to control the composition of our final data---crucial for finetuning multi-purpose models. To guarantee diversity, we improve upon previous work using embedding models and a clustering algorithm. This integrated strategy enables high-performance fine-tuning with minimal overhead.

Stratified Selective Sampling for Instruction Tuning with Dedicated Scoring Strategy

Chain-of-Thought (CoT) distillation has emerged as a promising paradigm to enhance the reasoning ability of small language models by imitating the reasoning and outputs of larger teacher models. However, existing approaches suffer from a critical limitation: a distribution mismatch between teacher-generated training trajectories and the student model's own generative distribution. This mismatch leads to exposure bias during inference and often induces mode collapse or mode averaging, thereby degrading the student model's generative diversity and robustness. To address these issues, we propose CoTD-PO (Chain-of-Thought Distillation with Preference Optimization), a reinforcement learning framework that shifts the training paradigm from passive imitation to active trajectory exploration. Instead of forcing the student to imitate exact teacher traces, our method enables the student to sample its own reasoning paths. To support training with non-open-source teacher models, we approximate the teacher’s output distribution through preference-based scoring. Furthermore, we adopt an offline iterative training procedure that enables stable and efficient optimization. Experiments on diverse open-ended generation tasks demonstrate that CoTD-PO significantly outperforms standard CoT distillation baselines, achieving higher output quality while mitigating mode collapse and preserving semantic diversity.

CoTD-PO: Chain-of-Thought Distillation with Preference Optimization

We examine three evaluation paradigms: standard benchmarks (e.g., MMLU and BBH), interactive games (e.g., Signalling Games or Taboo), and cognitive tests (e.g., for working memory or theory of mind). First, we investigate which of the former two—benchmarks or games—is most effective at discriminating LLMs of varying quality. Then, inspired by human cognitive assessments, we compile a suite of targeted tests that measure cognitive abilities deemed essential for effective language use, and we investigate their correlation with model performance in benchmarks and games. Our analyses reveal that interactive games are superior to standard benchmarks in discriminating models. Causal and logical reasoning correlate with both static and interactive tests, while differences emerge regarding core executive functions and social/emotional skills, which correlate more with games. We advocate for the development of new interactive benchmarks and targeted cognitive tasks inspired by assessing human abilities but designed specifically for LLMs.

Triangulating LLM Progress through Benchmarks, Games, and Cognitive Tests

Large Language Models (LLMs) demonstrate remarkable ability to comprehend instructions and generate human-like text, enabling sophisticated agent simulation beyond basic behavior replication. However, the potential for creating freely customisable characters remains underexplored. We introduce the Customisable Conversation Agent Framework, which employs LLMs to simulate real-world characters through personalised characteristic feature injection, enabling diverse character creation according to user preferences. We propose the SimsConv dataset, comprising 68 customised characters and 13,971 multi-turn role-playing dialogues across 1,360 real-world scenes. Characters are initially customised using pre-defined elements (career, aspiration, traits, skills), then expanded through personal and social profiles. Building on this, we present SimsChat, a freely customisable role-playing agent incorporating various realistic settings and topic-specified character interactions. Experimental results on both SimsConv and WikiRoleEval datasets demonstrate SimsChat's superior performance in maintaining character consistency, knowledge accuracy, and appropriate question rejection compared to existing models. Our framework provides valuable insights for developing more accurate and customisable human simulacra. Our data and code are publicly available https://anonymous.4open.science/r/SimsChat-60D0/.

Crafting Customisable Characters with LLMs: A Persona-Driven Role-Playing Agent Framework

Large Language Models (LLMs) are highly sensitive to subtle, non-semantic variations in prompt phrasing and formatting. In this work, we present the first systematic evaluation of 4 methods for improving prompt robustness within a unified experimental framework. We benchmark these techniques on 8 models from Llama, Qwen and Gemma families across 52 tasks from Natural Instructions dataset. Our evaluation covers robustness methods from both fine-tuned and in-context learning paradigms, and tests their generalization against multiple types of distribution shifts. Finally, we extend our analysis to GPT-4.1 and DeepSeek V3 to assess frontier models' current robustness to format perturbations. Our findings offer actionable insights into the relative effectiveness of these robustness methods, enabling practitioners to make informed decisions when aiming for stable and reliable LLM performance in real-world applications. Code: \urlstyle{tt}\url{https://anonymous.4open.science/r/prompt-instability-B6CD/}.

When Punctuation Matters: A Large-Scale Comparison of Prompt Robustness Methods for LLMs

We present NarratEX, a dataset designed for the task of explaining the choice of the Dominant Narrative in a news article, and intended to support the research community in addressing challenges such as discourse polarization and propaganda detection. Our dataset comprises 1,056 news articles in four languages, Bulgarian, English, Portuguese, and Russian, covering two globally significant topics: the Ukraine-Russia War (URW) and Climate Change (CC). Each article is manually annotated with a dominant narrative and sub-narrative labels, and an explanation justifying the chosen labels. We describe the dataset, the process of its creation, and its characteristics. We present experiments with two new proposed tasks: Explaining Dominant Narrative based on Text, which involves writing a concise paragraph to justify the choice of the dominant narrative and sub-narrative of a given text, and Inferring Dominant Narrative from Explanation, which involves predicting the appropriate dominant narrative category based on an explanatory text. The proposed dataset is a valuable resource for advancing research on detecting and mitigating manipulative content, while promoting a deeper understanding of how narratives influence public discourse.

NarratEX Dataset: Explaining the Dominant Narratives in News Texts

Inference-time alignment enhances the performance of large language models without requiring additional training or fine-tuning but presents challenges due to balancing computational efficiency with high-quality output. Best-of-N (BoN) sampling, as a simple yet powerful approach, generates multiple responses and selects the best one, achieving improved performance but with a high computational cost. We propose TreeBoN, a novel framework that integrates a speculative tree-search strategy into Best-of-N (BoN) Sampling. TreeBoN maintains a set of parent nodes, iteratively branching and pruning low-quality responses, thereby reducing computational overhead while maintaining high output quality. Our approach also leverages token-level rewards from Direct Preference Optimization (DPO) to guide tree expansion and prune low-quality paths. We evaluate TreeBoN using AlpacaFarm, UltraFeedback, GSM8K, HH-RLHF, and TutorEval datasets, demonstrating consistent improvements. Specifically, TreeBoN achieves a 65% win rate at maximum lengths of 192 and 384 tokens, outperforming standard BoN with the same computational cost. Furthermore, TreeBoN achieves around a 60% win rate across longer responses, showcasing its scalability and alignment efficacy.

TreeBoN: Enhancing Inference-Time Alignment with Speculative Tree-Search and Best-of-N Sampling

Long-context extension seeks to expand the contextual window in pre-trained large language models (LLMs), allowing them to handle several multiples of their original training context lengths. The primary method for extending the window length involves expanding the initial positional encodings, such as interpolating and extrapolation new positions based on Rotary Position Embedding (RoPE). This expansion inevitably disrupts the positional encodings learned during pre-training, thereby affecting the attention allotment and introducing unseen positional encoding distributions. To address this issue, we propose a new extension strategy based on RoPE, namely Periodic Extrapolation Positional Encodings (PEPE). This strategy expands pre-trained high dimensional components of positional encodings by replicating them in a periodic manner, thereby neither altering the learned positional encoding spaces nor introducing new positional encoding distributions. Experiments demonstrate that PEPE-based approaches can significantly improve long-context extension capabilities using just one-fourth the fine-tuning steps required by state-of-the-art methods. In addition, we analyze the characteristics of PEPE based methods and the key parameters that contribute to their effectiveness. The code will be available publicly.

PEPE: Long-context Extension for Large Language Models via Periodic Extrapolation Positional Encodings

Jailbreak attacks, where harmful prompts bypass generative models' built-in safety, raise serious concerns about model vulnerability. While many defense methods have been proposed, the trade-offs between safety and helpfulness, and their application to Large Vision-Language Models (LVLMs), are not well understood. This paper systematically examines jailbreak defenses by reframing the standard generation task as a binary classification problem to assess model refusal tendencies for both harmful and benign queries. We identify two key defense mechanisms: \textit{safety shift}, which increases refusal rates across all queries, and \textit{harmfulness discrimination}, which improves the model's ability to differentiate between harmful and benign inputs. Using these mechanisms, we develop two ensemble defense strategies—inter-mechanism and intra-mechanism ensembles—to balance safety and helpfulness. Experiments on the MM-SafetyBench and MOSSBench datasets with LLaVA-1.5 models show that these strategies effectively improve model safety or optimize the trade-off between safety and helpfulness.

How Jailbreak Defenses Work and Ensemble? A Mechanistic Investigation

Large language models have demonstrated considerable capabilities in various mathematical tasks, yet they often fall short in rigorous, proof-based reasoning essential for research-level mathematics. Retrieval-augmented generation presents a promising direction for enhancing these capabilities. This paper systematically explores RAG for natural language theorem proving, revealing that LLMs, when augmented with retrieved proofs rather than just theorems, can function as potent mimetic theorem provers: these models can effectively generalize proof techniques found in unstructured retrieved contexts to construct correct proofs for novel theorems. Building upon this finding, we introduce Dual RAG, a simple yet effective RAG framework. Dual RAG employs LLMs to identify underlying reasoning challenges within theorems, augmenting both queries and document contexts to improve retrieval performance. Our experiments show that Dual RAG achieves substantial improvements in retrieval performance, with gains of up to 34.19%. Expert evaluations further confirm that these retrieval enhancements directly translate into higher quality proof generation. Notably, when integrated with the arXiv API, Dual RAG demonstrates the ability to prove research-level theorems in theoretical machine learning, highlighting its strong potential as a foundational element for a practical mathematical copilot.

Downloads

Next from EMNLP 2025

Stratified Selective Sampling for Instruction Tuning with Dedicated Scoring Strategy

Similar lecture

Decoupling Knowledge and Reasoning in LLMs: An Exploration Using Cognitive Dual-System Theory