I asked my Muse agent to count how many papers were awarded at #EMNLP2026. There are 34(!) awarded papers compared to 12 last year. My guess is this accounts for the almost 2x increase in submissions and acceptances, and excludes the SAC highlights (35 last year). Maybe best paper per track? I guess we’ll find out tomorrow! How it was counted (from Muse): filtered the Underline program to oral sessions only (70 sessions), opened each one, and counted the trophy icon above a paper entry. Here's the full list: - In-Place Feedback: Reliable Refinement for Multi-Turn Expert-LLM Collaboration - Conformal Certification of Reasoning Trace Prefixes - Counterfactual Constraint-Conditioned On-Policy Distillation for Multi-Constraint Instruction Following - Format as Nuisance: Intervention-Inspired Cross-Format Distillation for Knowledge Conflicts - Do Benchmarks Measure the Same Thing? Evaluation-Free Benchmark Redundancy Analysis with SAE - UReason: Benchmarking Reasoning-to-Generation Alignment in Unified Multimodal Models - InterPol: De-anonymizing LM Arena via Interpolated Preference Learning - Race in the Record: Measuring Racial Signal and Model Inference in Police Arrest Narratives - The Cost of Sounding Different: Accent Bias in Audio Language Models - SWE-MEM: Learning Adaptive Memory Management for Long-Horizon Coding Agents - Gold vs. Translation: Are We Measuring Language, Knowledge, or Locality in Translated Benchmarks? - Rethinking Clustering as a Conversational Supervision Problem - Who Counts as "the People"? Measuring Contested Political Concepts from Political Speech using Masked Language Models - Accelerating Constrained Decoding with Token Space Compression - OptiMer: Optimal Distribution Vector Merging Is Better than Data Mixing for Continual Pre-Training - Clinically Grounded Privacy Evaluation of Medical LMs - Can MLLMs Perceive and Describe Medical Image Quality Beyond Scores? - When Must Faithful Chain-of-Thought Fail? An Information-Theoretic Bound and Its Empirical Limits - From Debate to Decision: Conformal Social Choice for Safe Multi-Agent Deliberation - Language Identification for Non-Standard Varieties: Language Varieties as Geolects on a Geodesic Grid - Aligned but Not Partner-Specific: How Multimodal LLM Agents Succeed in Reference Games Without Forming Conceptual Pacts - SkillHarm: Lifecycle-Aware Skill-Based Attacks via Automated Construction - Omnilingual MT: Machine Translation for 1,600 Languages - Cognitively Decoupled Meta-Induction for Out-of-Distribution Generalization - Text as a Universal Interface for Transferable Personalization - From Profiles to Steering Vectors: Global Sparse Priors and Local Semantic Calibration for Personalized Text Generation - When Hate Meets Facts: LLMs-in-the-Loop for Check-worthiness Detection in Hate Speech - Tampered ML: Evaluating LLM Judgment of Machine Learning Research - Diffusion Large Language Models for Visual Speech Recognition - Repetition over Diversity: High-Signal Data Filtering for Sample-Efficient German Language Modeling - HarnessForge: Joint Harness and Policy Evolution for Adaptive Agent Systems - Beyond "To whom it may concern": Tailoring Machine Translation to Audience and Intent - Rethinking the Evaluation of Efficiency Methods for Multi-Agent Systems - Why and When to Use Statistical Significance in Explanation Evaluation: A Systematic Survey and Guidelines for Explainable NLP