Papers with principled framework
The Data Frontier for Large Language Models: Selection, Synthesis, and Tools (2026.acl-tutorials)
Copied to clipboard
| Challenge: | acquiring and curating high-quality training data remains a significant bottleneck . acquiring such high-quality data is a key challenge for researchers and practitioners . |
| Approach: | This tutorial provides a comprehensive and practical guide to the state-of-the-art in data research directions for LLMs. |
| Outcome: | The tutorial covers methods for curating the most valuable information from vast, noisy datasets and the synthetic data revolution. |
Martingale Foresight Sampling: A Principled Approach to Inference-Time LLM Decoding (2026.eacl-long)
Copied to clipboard
| Challenge: | Standard autoregressive decoding in large language models is short-sighted, often failing to find globally optimal reasoning paths due to token-by-token generation process. |
| Approach: | They propose a principled framework that reformulates LLM decoding as a problem of identifying an optimal stochastic process. |
| Outcome: | The proposed framework surpasses state-of-the-art methods in accuracy while significantly improving computational efficiency. |
Do Audio LLMs Really LISTEN, or Just Transcribe? Measuring Lexical vs. Acoustic Emotion Cues Reliance (2026.eacl-long)
Copied to clipboard
| Challenge: | LISTEN is a controlled benchmark to disentangle lexical reliance from acoustic sensitivity in emotion understanding. |
| Approach: | They propose a benchmark to disentangle lexical reliance from acoustic sensitivity in emotion understanding. |
| Outcome: | LISTEN shows that current LALMs largely "transcribe" rather than "listen" authors note that models underutilize acoustic cues while relying on lexical semantics . |
OCP: Outlier-Centric Probing for Dynamic Structured Pruning of LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing structured pruning methods fail to identify outlier-triggering tokens and uniform layer-wise sparsity misaligns with heterogeneous outlier distributions. |
| Approach: | They propose a framework that prioritizes capturing outlier-triggering tokens rather than reconstructing full hidden distributions. |
| Outcome: | Experiments on LLaMA2, LLama3 and OPT show that the proposed framework outperforms state-of-the-art methods and achieves 25% perplexity reduction at 1.6 speedup. |
User-Assistant Bias in LLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Modern large language models are typically trained using structured role tags . asymmetries in training data associated with different role tags can potentially introduce inductive biases. |
| Approach: | They propose a task-agnostic benchmark to test user–assistant bias in large language models . they find human-preference alignment amplifies user bias, reasoning fine-tuning reduces it . |
| Outcome: | The proposed benchmark tests show that most instruction-tuned models exhibit strong user bias . human-preference alignment amplifies user bias, while reasoning fine-tuning reduces it. |
Challenges and Strategies in Cross-Cultural NLP (2022.acl-long)
Copied to clipboard
Daniel Hershcovich, Stella Frank, Heather Lent, Miryam de Lhoneux, Mostafa Abdou, Stephanie Brandl, Emanuele Bugliarello, Laura Cabello Piqueras, Ilias Chalkidis, Ruixiang Cui, Constanza Fierro, Katerina Margatina, Phillip Rust, Anders Søgaard
| Challenge: | Various efforts have been made to accommodate linguistic diversity and serve speakers of many different languages. |
| Approach: | They propose a framework to examine cultural differences in NLP to better serve users . they argue that cultural knowledge, preferences and values can affect NLP practices . |
| Outcome: | The proposed framework examines how cultural knowledge, preferences and values can affect NLP practices. |
Continuous Interpretive Steering for Scalar Diversity (2026.acl-long)
Copied to clipboard
| Challenge: | Existing studies on pragmatic inference in large language models rely on prompt-based manipulations to elicit a pragmatic interpretation. |
| Approach: | They propose a method that probes graded pragmatic interpretation by treating activation-level steering strength as a continuous experimental variable. |
| Outcome: | The proposed method increases pragmatic interpretations globally but collapses item-level variation whereas graded activation steering yields differentiated interpretive shifts aligned with scalar diversity grades. |
Powerful Training-Free Membership Inference Against Fine-Tuned Autoregressive Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for auditing fine-tuned language models have limited detection rates . membership inference attacks aim to determine if a specific record was in a model's training set . |
| Approach: | They propose a membership inference attack that exploits memorization at error positions . EZ-MIA achieves 3.8 higher detection than previous state-of-the-art . |
| Outcome: | The proposed attack achieves 3.8 higher detection than previous state-of-the-art models . EZ-MIA achieves 8 higher detectability than prior work, requiring no model training . |
SAVOIR: Learning Social Savoir-Faire via Shapley-based Reward Attribution (2026.findings-acl)
Copied to clipboard
Xiachong Feng, Yi Jiang, Xiaocheng Feng, Deyi Yin, Libo Qin, Yangfan Ye, Lei Huang, Weitao Ma, Yuxuan Gu, Chonghan Qin, Bing Qin, Lingpeng Kong
| Challenge: | Existing approaches to improve social intelligence of AI systems employ retrospective attributions and lack theoretical grounding. |
| Approach: | They propose a framework that uses Shapley values to ensure fair credit distribution with axiomatic guarantees of efficiency, symmetry, and marginality. |
| Outcome: | The proposed framework matches or exceeds proprietary models including GPT-4o and Claude-3.5-Sonnet. |
Uncertainty Quantification in LLM Agents: Foundations, Emerging Challenges, and Opportunities (2026.acl-long)
Copied to clipboard
Changdae Oh, Seongheon Park, To Eun Kim, Jiatong Li, Wendi Li, Samuel Yeh, Sean Du, Hamed Hassani, Paul Bogdan, Dawn Song, Sharon Li
| Challenge: | Uncertainty quantification (UQ) for large language models is a key building block for daily applications. |
| Approach: | They propose a general formulation of agent UQ that subsumes broad classes of existing UQ setups. |
| Outcome: | The proposed framework is based on the first general formulation of agent UQ that subsumes broad classes of existing setups. |
A Data-Centric Approach to Generalizable Speech Deepfake Detection (2026.acl-long)
Copied to clipboard
| Challenge: | Speech deepfake detection (SDD) is a critical research area as speech synthesis technologies become more sophisticated. |
| Approach: | They propose a data-centric approach to generalize SDD data from two perspectives . they propose naive aggregation strategies for mixing heterogeneous data and diversity-optimized sampling strategy for a single dataset and multiple datasets. |
| Outcome: | The proposed approach outperforms the naive aggregation baseline on a 12k-hour data pool while using only 3% of the total available data. |
Logical Phase Transitions: Understanding Collapse in LLM Logical Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Symbolic logical reasoning is a critical yet underexplored capability of large language models (LLMs). |
| Approach: | They propose a framework that aligns natural language with logical symbols to establish a shared representation and reshapes training dynamics around phase-transition boundaries to progressively strengthen reasoning at increasing logical depths. |
| Outcome: | The proposed framework mitigates logical reasoning collapse at high complexity while improving generalization to unseen logical compositions. |
Understanding the Thinking Process of Reasoning Models: A Perspective from Schoenfeld’s Episode Theory (2025.emnlp-main)
Copied to clipboard
Ming Li, Nan Zhang, Chenrui Fan, Hong Jiao, Yanbin Fu, Sydney Peters, Qingshu Xu, Robert Lissitz, Tianyi Zhou
| Challenge: | Large Reasoning Models (LRMs) generate extensive chain-of-thought reasoning, but we lack a principled framework for understanding how these thoughts are structured. |
| Approach: | They propose a method to analyze the reasoning traces of Large Reasoning Models using Schoenfeld’s Episode Theory. |
| Outcome: | The proposed framework provides a theoretically grounded methodology for interpreting LRM cognition and enables future work on more controllable and transparent reasoning systems. |
VCORE: Variance-Controlled Optimization-based Reweighting for Chain-of-Thought Supervision (2026.acl-long)
Copied to clipboard
| Challenge: | Empirical evaluations demonstrate that VCORE achieves the strongest overall average performance, with especially clear gains on lower-capacity models. |
| Approach: | They propose a framework that reformulates supervision as a constrained optimization problem. |
| Outcome: | Empirical evaluations show that VCORE achieves the strongest overall average performance, with especially clear gains on lower-capacity models. |
Controllable Contamination Detection for Reliable LLM Evaluation with Statistical Guarantees (2026.acl-long)
Copied to clipboard
Zheng Zhang, Qi Liu, Siyuan Liang, Ning Li, Zirui Hu, Weibo Gao, Rui Li, Zhenya Huang, Leszek Rutkowski, Baosheng Yu, Dacheng Tao
| Challenge: | Existing training data detectors fail to detect clean samples from contaminated test sets . existing methods fail to identify clean samples due to black-box nature of LLMs . |
| Approach: | They propose a framework that detects and filters contaminated evaluation data . they propose 'failure detection' to reduce the proportion of contaminated samples mistakenly retained . |
| Outcome: | The proposed framework reduces false discovery rate (FDR) under valid FDR control while maintaining evaluation consistency. |
The Strawberry Problem: Emergence of Character-level Understanding in Tokenized Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models fail at simple character-level tasks due to low mutual information, study finds . authors propose a lightweight architectural modification that improves character- level reasoning . |
| Approach: | They propose a lightweight architectural modification that improves character-level reasoning while preserving the inductive advantages of subword models. |
| Outcome: | The proposed model improves character-level reasoning while preserving the advantages of subword models. |
Magnet: Multi-turn Tool-use Data Synthesis and Distillation via Graph Translation (2025.acl-long)
Copied to clipboard
Fan Yin, Zifeng Wang, I-Hung Hsu, Jun Yan, Ke Jiang, Yanfei Chen, Jindong Gu, Long Le, Kai-Wei Chang, Chen-Yu Lee, Hamid Palangi, Tomas Pfister
| Challenge: | Large language models have been shown to be effective in multi-turn interactions . however, their performance may be limited in complex, multi-turned interactions involving users and multiple tools. |
| Approach: | They propose a framework for synthesizing high-quality training trajectories to enhance the function calling capability of large language model agents in multi-turn conversations with humans. |
| Outcome: | The proposed model outperforms the teacher model by 68.01 on BFCL-v3 and 73.30 on ToolQuery. |
ARCHITECT: Uncertainty-Aware Dynamic Tool Learning via Causal Intervention for Open-World Agents (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods treat all generated tools as equally trustworthy, a "blind trust" assumption that is untenable for reliable agent deployment. |
| Approach: | They propose a framework that moves beyond black-box reliability prediction to interpretable failure attribution. |
| Outcome: | The proposed framework achieves state-of-the-art on four benchmarks including StableToolBench, MINT, T-Eval, and SWE-bench Lite. |
Retracing the Past: LLMs Emit Training Data When They Get Lost (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for extracting training data from large language models exhibit limited success . existing methods offer limited insight into the fundamental drivers of memorization leakage . |
| Approach: | They propose a framework for extracting memorized data by maximizing model uncertainty . they propose mismatched fine-tuning to weaken alignment and induce confusion . |
| Outcome: | The proposed attacks outperform baselines on unaligned and aligned LLMs . the proposed attacks exploit the model uncertainty of the input snippets induced by the model entropy spike . |
Surprisal Minimisation over Goal-directed Alternatives Predicts Production Choice in Dialogue (2026.acl-long)
Copied to clipboard
| Challenge: | a method to model utterance production is based on information-theoretic notions of cost . a technique to generate alternative sets of utterables is proposed . |
| Approach: | They propose a procedure to generate both types of alternative sets using language models. |
| Outcome: | The proposed procedure allows for speaker- and listener-oriented interpretations of different cost measures. |
Structured Uncertainty guided Clarification for LLM Agents (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to clarifying tasks fail when user instructions are ambiguous or incomplete. |
| Approach: | They propose a principled formulation of structured uncertainty that operates directly over tool parameters and their domains. |
| Outcome: | The proposed framework improves when2call accuracy and training-time sample efficiency. |