Peer Revenue Impact factor: volatile

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Computation and Language

cs.CL Watch Computation and Language

Wednesday 12 November

Computation and Language

The Learning Dynamics of Subword Segmentation for Morphologically Diverse Languages

Francois Meyer and Jan Buys

Subword segmentation is typically applied in preprocessing and stays fixed during training. Alternatively, it can be learned during training to optimise the training objective. In this paper we study the learning dynamics of subword segmentation: if a language model can dynamically optimise tokenisation, how do its subwords evolve during pretraining and finetuning? To explore this, we extend the subword segmental language model (SSLM), a framework for learning subwords during training, to support pretraining and finetuning. We train models for three typologically diverse languages to study learning dynamics across the morphological spectrum: Isi-Xhosa is conjunctive (long word forms composed of many morphemes), Setswana is disjunctive (morphemes written as separate words), and English represents a typological middle ground. We analyse subword dynamics from a linguistic perspective, tracking morphology, productivity, and fertility. We identify four stages of subword learning, with the morphologically complex isi-Xhosa exhibiting greater instability. During finetuning, subword boundaries shift to become finer-grained. Lastly, we show that learnable subwords offers a promising approach to improve text generation and cross-lingual transfer for low-resource, morphologically complex languages.
Read the full abstract

Computation and Language

Context is Enough: Empirical Validation of $\textit{Sequentiality}$ on Essays

Amal Sunny, Advay Gupta and Vishnu Sreekumar

Recent work has proposed using Large Language Models (LLMs) to quantify narrative flow through a measure called sequentiality, which combines topic and contextual terms. A recent critique argued that the original results were confounded by how topics were selected for the topic-based component, and noted that the metric had not been validated against ground-truth measures of flow. That work proposed using only the contextual term as a more conceptually valid and interpretable alternative. In this paper, we empirically validate that proposal. Using two essay datasets with human-annotated trait scores, ASAP++ and ELLIPSE, we show that the contextual version of sequentiality aligns more closely with human assessments of discourse-level traits such as Organization and Cohesion. While zero-shot prompted LLMs predict trait scores more accurately than the contextual measure alone, the contextual measure adds more predictive value than both the topic-only and original sequentiality formulations when combined with standard linguistic features. Notably, this combination also outperforms the zero-shot LLM predictions, highlighting the value of explicitly modeling sentence-to-sentence flow. Our findings support the use of context-based sequentiality as a validated, interpretable, and complementary feature for automated essay scoring and related NLP tasks.
Read the full abstract

Computation and Language

A Hybrid Search for Complex Table Question Answering in Securities Report

Daiki Shirafuji, Koji Tanaka and Tatsuhiko Saito

Recently, Large Language Models (LLMs) are gaining increased attention in the domain of Table Question Answering (TQA), particularly for extracting information from tables in documents. However, directly entering entire tables as long text into LLMs often leads to incorrect answers because most LLMs cannot inherently capture complex table structures. In this paper, we propose a cell extraction method for TQA without manual identification, even for complex table headers. Our approach estimates table headers by computing similarities between a given question and individual cells via a hybrid retrieval mechanism that integrates a language model and TF-IDF. We then select as the answer the cells at the intersection of the most relevant row and column. Furthermore, the language model is trained using contrastive learning on a small dataset of question-header pairs to enhance performance. We evaluated our approach in the TQA dataset from the U4 shared task at NTCIR-18. The experimental results show that our pipeline achieves an accuracy of 74.6%, outperforming existing LLMs such as GPT-4o mini~(63.9%). In the future, although we used traditional encoder models for retrieval in this study, we plan to incorporate more efficient text-search models to improve performance and narrow the gap with human evaluation results.
Read the full abstract

Computation and Language

LoopTool: Closing the Data-Training Loop for Robust LLM Tool Calls

Kangning Zhang, Wenxiang Jiao, Kounianhua Du et al. (+4 more)

Augmenting Large Language Models (LLMs) with external tools enables them to execute complex, multi-step tasks. However, tool learning is hampered by the static synthetic data pipelines where data generation and model training are executed as two separate, non-interactive processes. This approach fails to adaptively focus on a model's specific weaknesses and allows noisy labels to persist, degrading training efficiency. We introduce LoopTool, a fully automated, model-aware data evolution framework that closes this loop by tightly integrating data synthesis and model training. LoopTool iteratively refines both the data and the model through three synergistic modules: (1) Greedy Capability Probing (GCP) diagnoses the model's mastered and failed capabilities; (2) Judgement-Guided Label Verification (JGLV) uses an open-source judge model to find and correct annotation errors, progressively purifying the dataset; and (3) Error-Driven Data Expansion (EDDE) generates new, challenging samples based on identified failures. This closed-loop process operates within a cost-effective, open-source ecosystem, eliminating dependence on expensive closed-source APIs. Experiments show that our 8B model trained with LoopTool significantly surpasses its 32B data generator and achieves new state-of-the-art results on the BFCL-v3 and ACEBench benchmarks for its scale. Our work demonstrates that closed-loop, self-refining data pipelines can dramatically enhance the tool-use capabilities of LLMs.
Read the full abstract

All authors Kangning Zhang, Wenxiang Jiao, Kounianhua Du, Yuan Lu, Weiwen Liu, Weinan Zhang, Yong Yu

Computation and Language

DoPE: Denoising Rotary Position Embedding

Jing Xiong, Liyang Fan, Hui Shen et al. (+4 more)

Positional encoding is essential for large language models (LLMs) to represent sequence order, yet recent studies show that Rotary Position Embedding (RoPE) can induce massive activation. We investigate the source of these instabilities via a spectral analysis of RoPE, and show that its low-frequency components concentrate structured energy, producing low-rank, over-aligned attention patterns. We theoretically reveal that this low-frequency alignment manifests as activation noise, degrading stability during long-context extrapolation. To mitigate this effect, we introduce Denoising Rotary Position Embedding (DoPE), a training-free method that identifies and suppresses noisy attention heads using truncated matrix entropy, then reparameterizes their attention maps with an isotropic Gaussian distribution. Across a range of settings, DoPE improves length extrapolation performance without fine-tuning, increases robustness to perturbations, and boosts both needle-in-a-haystack and many-shot in-context learning tasks. These results suggest that selective positional encoding is key to robust extrapolation. Our project page is Project: https://The-physical-picture-of-LLMs.github.io
Read the full abstract

All authors Jing Xiong, Liyang Fan, Hui Shen, Zunhai Su, Min Yang, Lingpeng Kong, Ngai Wong

Computation and Language

One-Topic-Doesn't-Fit-All: Transcreating Reading Comprehension Test for Personalized Learning

Jieun Han, Daniel Lee, Haneul Yoo et al. (+5 more)

Personalized learning has gained attention in English as a Foreign Language (EFL) education, where engagement and motivation play crucial roles in reading comprehension. We propose a novel approach to generating personalized English reading comprehension tests tailored to students' interests. We develop a structured content transcreation pipeline using OpenAI's gpt-4o, where we start with the RACE-C dataset, and generate new passages and multiple-choice reading comprehension questions that are linguistically similar to the original passages but semantically aligned with individual learners' interests. Our methodology integrates topic extraction, question classification based on Bloom's taxonomy, linguistic feature analysis, and content transcreation to enhance student engagement. We conduct a controlled experiment with EFL learners in South Korea to examine the impact of interest-aligned reading materials on comprehension and motivation. Our results show students learning with personalized reading passages demonstrate improved comprehension and motivation retention compared to those learning with non-personalized materials.
Read the full abstract

All authors Jieun Han, Daniel Lee, Haneul Yoo, Jinsung Yoon, Junyeong Park, Suin Kim, So-Yeon Ahn, Alice Oh

Computation and Language

Assessing the Capabilities of LLMs in Humor:A Multi-dimensional Analysis of Oogiri Generation and Evaluation

Ritsu Sakabe, Hwichan Kim, Tosho Hirasawa et al. (+1 more)

Computational humor is a frontier for creating advanced and engaging natural language processing (NLP) applications, such as sophisticated dialogue systems. While previous studies have benchmarked the humor capabilities of Large Language Models (LLMs), they have often relied on single-dimensional evaluations, such as judging whether something is simply “funny.” This paper argues that a multifaceted understanding of humor is necessary and addresses this gap by systematically evaluating LLMs through the lens of Oogiri, a form of Japanese improvisational comedy games. To achieve this, we expanded upon existing Oogiri datasets with data from new sources and then augmented the collection with Oogiri responses generated by LLMs. We then manually annotated this expanded collection with 5-point absolute ratings across six dimensions: Novelty, Clarity, Relevance, Intelligence, Empathy, and Overall Funniness. Using this dataset, we assessed the capabilities of state-of-the-art LLMs on two core tasks: their ability to generate creative Oogiri responses and their ability to evaluate the funniness of responses using a six-dimensional evaluation. Our results show that while LLMs can generate responses at a level between low- and mid-tier human performance, they exhibit a notable lack of Empathy. This deficit in Empathy helps explain their failure to replicate human humor assessment. Correlation analyses of human and model evaluation data further reveal a fundamental divergence in evaluation criteria: LLMs prioritize Novelty, whereas humans prioritize Empathy. We release our annotated corpus to the community to pave the way for the development of more emotionally intelligent and sophisticated conversational agents.
Read the full abstract

All authors Ritsu Sakabe, Hwichan Kim, Tosho Hirasawa, Mamoru Komachi

Computation and Language · Cross-listed from cs.AI

History-Aware Reasoning for GUI Agents

Ziwei Wang, Leyang Yang, Xiaoxuan Tang et al. (+4 more)

Advances in Multimodal Large Language Models have significantly enhanced Graphical User Interface (GUI) automation. Equipping GUI agents with reliable episodic reasoning capabilities is essential for bridging the gap between users' concise task descriptions and the complexities of real-world execution. Current methods integrate Reinforcement Learning (RL) with System-2 Chain-of-Thought, yielding notable gains in reasoning enhancement. For long-horizon GUI tasks, historical interactions connect each screen to the goal-oriented episode chain, and effectively leveraging these clues is crucial for the current decision. However, existing native GUI agents exhibit weak short-term memory in their explicit reasoning, interpreting the chained interactions as discrete screen understanding, i.e., unawareness of the historical interactions within the episode. This history-agnostic reasoning challenges their performance in GUI automation. To alleviate this weakness, we propose a History-Aware Reasoning (HAR) framework, which encourages an agent to reflect on its own errors and acquire episodic reasoning knowledge from them via tailored strategies that enhance short-term memory in long-horizon interaction. The framework mainly comprises constructing a reflective learning scenario, synthesizing tailored correction guidelines, and designing a hybrid RL reward function. Using the HAR framework, we develop a native end-to-end model, HAR-GUI-3B, which alters the inherent reasoning mode from history-agnostic to history-aware, equipping the GUI agent with stable short-term memory and reliable perception of screen details. Comprehensive evaluations across a range of GUI-related benchmarks demonstrate the effectiveness and generalization of our method.
Read the full abstract

All authors Ziwei Wang, Leyang Yang, Xiaoxuan Tang, Sheng Zhou, Dajun Chen, Wei Jiang, Yong Li

Computation and Language

Thinking Forward and Backward: Multi-Objective Reinforcement Learning for Retrieval-Augmented Reasoning

Wenda Wei, Yu-An Liu, Ruqing Zhang et al. (+6 more)

Retrieval-augmented generation (RAG) has proven to be effective in mitigating hallucinations in large language models, yet its effectiveness remains limited in complex, multi-step reasoning scenarios. Recent efforts have incorporated search-based interactions into RAG, enabling iterative reasoning with real-time retrieval. Most approaches rely on outcome-based supervision, offering no explicit guidance for intermediate steps. This often leads to reward hacking and degraded response quality. We propose Bi-RAR, a novel retrieval-augmented reasoning framework that evaluates each intermediate step jointly in both forward and backward directions. To assess the information completeness of each step, we introduce a bidirectional information distance grounded in Kolmogorov complexity, approximated via language model generation probabilities. This quantification measures both how far the current reasoning is from the answer and how well it addresses the question. To optimize reasoning under these bidirectional signals, we adopt a multi-objective reinforcement learning framework with a cascading reward structure that emphasizes early trajectory alignment. Empirical results on seven question answering benchmarks demonstrate that Bi-RAR surpasses previous methods and enables efficient interaction and reasoning with the search engine during training and inference.
Read the full abstract

All authors Wenda Wei, Yu-An Liu, Ruqing Zhang, Jiafeng Guo, Lixin Su, Shuaiqiang Wang, Dawei Yin, Maarten de Rijke, Xueqi Cheng

Computation and Language

Context-Aware Dynamic Chunking for Streaming Tibetan Speech Recognition

Chao Wang, Yuqing Cai, Renzeng Duojie et al. (+3 more)

In this work, we propose a streaming speech recognition framework for Amdo Tibetan, built upon a hybrid CTC/Atten-tion architecture with a context-aware dynamic chunking mechanism. The proposed strategy adaptively adjusts chunk widths based on encoding states, enabling flexible receptive fields, cross-chunk information exchange, and robust adaptation to varying speaking rates, thereby alleviating the context truncation problem of fixed-chunk methods. To further capture the linguistic characteristics of Tibetan, we construct a lexicon grounded in its orthographic principles, providing linguistically motivated modeling units. During decoding, an external language model is integrated to enhance semantic consistency and improve recognition of long sentences. Experimental results show that the proposed framework achieves a word error rate (WER) of 6.23% on the test set, yielding a 48.15% relative improvement over the fixed-chunk baseline, while significantly reducing recognition latency and maintaining performance close to global decoding.
Read the full abstract

All authors Chao Wang, Yuqing Cai, Renzeng Duojie, Jin Zhang, Yutong Liu, Nyima Tashi

Computation and Language

MM-CRITIC: A Holistic Evaluation of Large Multimodal Models as Multimodal Critique

Gailun Zeng, Ziyang Luo, Hongzhan Lin et al. (+5 more)

The ability of critique is vital for models to self-improve and serve as reliable AI assistants. While extensively studied in language-only settings, multimodal critique of Large Multimodal Models (LMMs) remains underexplored despite their growing capabilities in tasks like captioning and visual reasoning. In this work, we introduce MM-CRITIC, a holistic benchmark for evaluating the critique ability of LMMs across multiple dimensions: basic, correction, and comparison. Covering 8 main task types and over 500 tasks, MM-CRITIC collects responses from various LMMs with different model sizes and is composed of 4471 samples. To enhance the evaluation reliability, we integrate expert-informed ground answers into scoring rubrics that guide GPT-4o in annotating responses and generating reference critiques, which serve as anchors for trustworthy judgments. Extensive experiments validate the effectiveness of MM-CRITIC and provide a comprehensive assessment of leading LMMs' critique capabilities under multiple dimensions. Further analysis reveals some key insights, including the correlation between response quality and critique, and varying critique difficulty across evaluation dimensions. Our code is available at https://github.com/MichealZeng0420/MM-Critic.
Read the full abstract

All authors Gailun Zeng, Ziyang Luo, Hongzhan Lin, Yuchen Tian, Kaixin Li, Ziyang Gong, Jianxiong Guo, Jing Ma

Computation and Language · Cross-listed from cs.CV

PAN: A World Model for General, Actionable, and Long-Horizon World Simulation

PAN Team Institute of Foundation Models: Zihan Liu, Yi Gu, Mingkai Deng et al. (+11 more)

A world model is a cognitive simulator of the real-world environment allowing biological agents to reason about how the world evolves, whether spontaneously or in response to their actions, and accordingly to plan and strategize. In building Artificial Intelligence (AI) systems, world models represent the next frontier beyond large language models (LLMs) to enable physical and embodied intelligence in AI agents, allowing them to perform decision-making through simulative reasoning and reinforcement-learning through simulative trials. Recent advancements in world modeling have yielded impressive progress in video generation, 3-D scene evolution, robotic dynamics, and game simulation, but limitations persist in general, open-domain, action-driven prediction, long-horizon consistency, and abstract reasoning and planning. Moreover, fundamental architectural questions, whether it be state representation, information flow, or training objectives, remain unresolved. In this paper, we introduce PAN, a world model built on the Generative Latent Prediction (GLP) architecture. GLP combines stateful latent representations of world states; an encoder–decoder closed-loop information flow; an LLM/diffusion-based mixed reasoning backbone; and a non-degenerate generative reconstruction objective whose fidelity is “dampable” to balance fine-grained detail against semantic saliency. Compared to several existing systems, PAN demonstrates advantages beyond standard video generation in action-conditioned world simulation, long-horizon forecasting, and simulative reasoning and planning, capabilities we argue should serve as the primary criteria for evaluating world models.
Read the full abstract

All authors PAN Team Institute of Foundation Models: Zihan Liu, Yi Gu, Mingkai Deng, Guangyi Liu, Zeyu Feng, Qiyue Gao, Yiyan Hu, Benhao Huang, Yichi Yang, Kun Zhou, Jiannan Xiang, Zhiting Hu, Zhengzhong Liu, Eric P. Xing

Computation and Language · Cross-listed from cs.AI

Solving a Million-Step LLM Task with Zero Errors

Elliot Meyerson, Giuseppe Paolo, Roberto Dailey et al. (+6 more)

LLMs have achieved remarkable breakthroughs in reasoning, insights, and tool use, but chaining these abilities into extended processes at the scale of those routinely executed by humans, organizations, and societies has remained out of reach. The models have a persistent error rate that prevents scale-up: for instance, recent experiments in the Towers of Hanoi benchmark domain showed that the process inevitably becomes derailed after at most a few hundred steps. Thus, although LLM research is often still benchmarked on tasks with relatively few dependent logical steps, there is increasing attention on the ability (or inability) of LLMs to perform long range tasks. This paper describes MAKER, the first system that successfully solves a task with over one million LLM steps with zero errors, and, in principle, scales far beyond this level. The approach relies on an extreme decomposition of a task into subtasks, each of which can be tackled by focused microagents. The high level of modularity resulting from the decomposition allows error correction to be applied at each step through an efficient multi-agent voting scheme. This combination of extreme decomposition and error correction makes scaling possible. Thus, the results suggest that instead of relying on continual improvement of current LLMs, massively decomposed agentic processes (MDAPs) may provide a way to efficiently solve problems at the level of organizations and societies.
Read the full abstract

All authors Elliot Meyerson, Giuseppe Paolo, Roberto Dailey, Hormoz Shahrzad, Olivier Francon, Conor F. Hayes, Xin Qiu, Babak Hodjat, Risto Miikkulainen

Computation and Language

A Neurosymbolic Approach to Natural Language Formalization and Verification

Chenyang An, Sam Bayless, Stefano Buliani et al. (+27 more)

Large Language Models perform well at natural language interpretation and reasoning, but their lack of formal correctness guarantees limits their adoption in regulated industries like finance and health-care that operate under strict policies. To address this limitation, we launched Automated Reasoning checks (ARc): a public service that (1) uses LLMs with optional human guidance to formalize natural language policies, allowing fine-grained control of the formalization process, and (2) uses inference-time autoformalization to validate logical correctness of natural language statements against those policies. ARc performs multiple redundant formalization steps at inference time, checking the formalizations for semantic equivalence. Our benchmarks show that ARc exceeds 99% soundness and achieves a near-zero false positive rate in identifying logical validity. Our approach produces auditable artifacts that substantiate the verification outcomes and can be used to improve the original text. ARc is the first commercial offering from a major cloud provider to integrate automated reasoning into a generative AI guardrail.
Read the full abstract

All authors Chenyang An, Sam Bayless, Stefano Buliani, Darion Cassel, Byron Cook, Duncan Clough, Rémi Delmas, Nafi Diallo, Ferhat Erata, Nick Feng, Dimitra Giannakopoulou, Aman Goel, Aditya Gokhale, Joe Hendrix, Victor Heorhiadi, Marc Hudak, Dejan Jovanović, Andrew M. Kent, Benjamin Kiesl-Reiter, Jeffrey J. Kuna, Nadia Labai, Joseph Lilien, Divya Raghunathan, Zvonimir Rakamarić, Niloofar Razavi, Michael Tautschnig, Ali Torkamani, Nathaniel Weir, Michael W. Whalen, Jianan Yao

Computation and Language · Cross-listed from cs.AI

AI Founding Fathers: A Case Study of GIS Search in Multi-Agent Pipelines

Alvin Chauhan

Although Large Language Models (LLMs) show exceptional fluency, efforts persist to extract stronger reasoning capabilities from them. Drawing on search-based interpretations of LLM computation, this paper advances a systematic framework for understanding LLM reasoning and optimization. Namely, that enhancing reasoning is best achieved by structuring a multi-agent pipeline to ensure a traversal of the search space in a gradual, incremental, and sequential (GIS) manner. Stated succinctly, high-quality reasoning is a controlled, incremental search. To test this framework, we investigate the efficacy of recursive refinement (RR)–an iterative process of self-criticism, adversarial stress-testing, and integrating critical feedback–as a practical method for implementing GIS search. We designed an experiment comparing a simple, linear pipeline against a complex, explicitly structured pipeline leveraging a recursive refinement layer. The multi-agent models were constructed to reflect the historical personas of three US Founding Fathers (Hamilton, Jefferson, and Madison) using RAG-powered corpora and were prompted to generate responses to three contemporary political issues. Model performance was evaluated using a two-tiered approach: a quantitative score from an LLM arbiter agent and qualitative human judgment. Our results revealed that the complex model consistently outperformed the simple model across all nine test cases with an average arbiter-outputted score of 88.3 versus 71.7. The complex model's arguments were superior in analytical depth, structural nuance, and strategic framing. We conclude that recursive refinement is a robust architectural feature for enhancing LLM reasoning via GIS search.
Read the full abstract

Computation and Language

Detecting Emotional Dynamic Trajectories: An Evaluation Framework for Emotional Support in Language Models

Zhouxing Tan, Ruochong Xiong, Yulong Wan et al. (+8 more)

Emotional support is a core capability in human-AI interaction, with applications including psychological counseling, role play, and companionship. However, existing evaluations of large language models (LLMs) often rely on short, static dialogues and fail to capture the dynamic and long-term nature of emotional support. To overcome this limitation, we shift from snapshot-based evaluation to trajectory-based assessment, adopting a user-centered perspective that evaluates models based on their ability to improve and stabilize user emotional states over time. Our framework constructs a large-scale benchmark consisting of 328 emotional contexts and 1,152 disturbance events, simulating realistic emotional shifts under evolving dialogue scenarios. To encourage psychologically grounded responses, we constrain model outputs using validated emotion regulation strategies such as situation selection and cognitive reappraisal. User emotional trajectories are modeled as a first-order Markov process, and we apply causally-adjusted emotion estimation to obtain unbiased emotional state tracking. Based on this framework, we introduce three trajectory-level metrics: Baseline Emotional Level (BEL), Emotional Trajectory Volatility (ETV), and Emotional Centroid Position (ECP). These metrics collectively capture user emotional dynamics over time and support comprehensive evaluation of long-term emotional support performance of LLMs. Extensive evaluations across a diverse set of LLMs reveal significant disparities in emotional support capabilities and provide actionable insights for model development.
Read the full abstract

All authors Zhouxing Tan, Ruochong Xiong, Yulong Wan, Jinlong Ma, Hanlin Xue, Qichun Deng, Haifeng Jing, Zhengtong Zhang, Depei Liu, Shiyuan Luo, Junfei Liu

Computation and Language

SpiralThinker: Latent Reasoning through an Iterative Process with Text-Latent Interleaving

Shengmin Piao and Sanghyun Park

Recent advances in large reasoning models have been driven by reinforcement learning and test-time scaling, accompanied by growing interest in latent rather than purely textual reasoning. However, existing latent reasoning methods lack mechanisms to ensure stable reasoning dynamics in latent space and a systematic way to interleave implicit and explicit reasoning. We introduce SpiralThinker, a stabilized iterative latent reasoning framework that performs iterative updates over latent representations while interleaving latent and textual reasoning steps. At its core, it combines a progressive alignment objective that explicitly regulates latent representations across iterations with structured annotations for text-latent interleaving, thereby stabilizing latent updates and maintaining coherence with textual reasoning. Across mathematical, logical, and commonsense reasoning tasks, SpiralThinker achieves state-of-the-art performance among latent reasoning baselines. Further analysis shows that both iteration and alignment are essential, that the optimal numbers of latent tokens and iterations vary by dataset, and that proper alignment is crucial for effective iterative latent reasoning. Overall, SpiralThinker bridges iterative computation and latent reasoning, demonstrating that aligned iterative updates can reliably steer reasoning in the latent space.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

Bayesian Mixture of Experts For Large Language Models

Maryam Dialameh, Hossein Rajabzadeh, Weiwei Zhang et al. (+2 more)

We present Bayesian Mixture of Experts (Bayesian-MoE), a post-hoc uncertainty estimation framework for fine-tuned large language models (LLMs) based on Mixture-of-Experts architectures. Our method applies a structured Laplace approximation to the second linear layer of each expert, enabling calibrated uncertainty estimation without modifying the original training procedure or introducing new parameters. Unlike prior approaches, which apply Bayesian inference to added adapter modules, Bayesian-MoE directly targets the expert pathways already present in MoE models, leveraging their modular design for tractable block-wise posterior estimation. We use Kronecker-factored low-rank approximations to model curvature and derive scalable estimates of predictive uncertainty and marginal likelihood. Experiments on common-sense reasoning benchmarks with Qwen1.5-MoE and DeepSeek-MoE demonstrate that Bayesian-MoE improves both expected calibration error (ECE) and negative log-likelihood (NLL) over baselines, confirming its effectiveness for reliable downstream decision-making.
Read the full abstract

All authors Maryam Dialameh, Hossein Rajabzadeh, Weiwei Zhang, Walid Ahmed, Hyock Ju Kwon

Computation and Language

EVADE: LLM-Based Explanation Generation and Validation for Error Detection in NLI

Longfei Zuo, Barbara Plank and Siyao Peng

High-quality datasets are critical for training and evaluating reliable NLP models. In tasks like natural language inference (NLI), human label variation (HLV) arises when multiple labels are valid for the same instance, making it difficult to separate annotation errors from plausible variation. An earlier framework, VARIERR (Weber-Genzel et al., 2024), asks multiple annotators to explain their label decisions in the first round and flags errors through validity judgments in the second round. However, conducting two rounds of manual annotation is costly and may limit the coverage of plausible labels or explanations. Our study proposes a new framework, EVADE, for generating and validating explanations to detect errors using large language models (LLMs). We perform a comprehensive analysis comparing human- and LLM-detected errors for NLI across distribution comparison, validation overlap, and impact on model fine-tuning. Our experiments demonstrate that LLM validation refines generated explanation distributions to more closely align with human annotations, and that removing LLM-detected errors from training data yields improvements in fine-tuning performance than removing errors identified by human annotators. This highlights the potential to scale error detection, reducing human effort while improving dataset quality under label variation.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

TransactionGPT

Yingtong Dou, Zhimeng Jiang, Tianyi Zhang et al. (+26 more)

We present TransactionGPT (TGPT), a foundation model for consumer transaction data within one of the world's largest payment networks. TGPT is designed to understand and generate transaction trajectories while simultaneously supporting a variety of downstream prediction and classification tasks. We introduce a novel 3D-Transformer architecture specifically tailored for capturing the complex dynamics in payment transaction data. This architecture incorporates design innovations that enhance modality fusion and computational efficiency, while seamlessly enabling joint optimization with downstream objectives. Trained on billion-scale real-world transactions, TGPT significantly improves downstream anomaly transaction detection performance against a competitive production model and exhibits advantages over baselines in generating future transactions. We conduct extensive empirical evaluations utilizing a diverse collection of company transaction datasets spanning multiple downstream tasks, thereby enabling a thorough assessment of TGPT's effectiveness and efficiency in comparison to established methodologies. Furthermore, we examine the incorporation of LLM-derived embeddings within TGPT and benchmark its performance against fine-tuned LLMs, demonstrating that TGPT achieves superior predictive accuracy as well as faster training and inference. We anticipate that the architectural innovations and practical guidelines from this work will advance foundation models for transaction-like data and catalyze future research in this emerging field.
Read the full abstract

All authors Yingtong Dou, Zhimeng Jiang, Tianyi Zhang, Mingzhi Hu, Zhichao Xu, Shubham Jain, Uday Singh Saini, Xiran Fan, Jiarui Sun, Menghai Pan, Junpeng Wang, Xin Dai, Liang Wang, Chin-Chia Michael Yeh, Yujie Fan, Yan Zheng, Vineeth Rakesh, Huiyuan Chen, Guanchu Wang, Mangesh Bendre, Zhongfang Zhuang, Xiaoting Li, Prince Aboagye, Vivian Lai, Minghua Xu, Hao Yang, Yiwei Cai, Mahashweta Das, Yuzhong Chen

Computation and Language · Cross-listed from cs.AI

The Double Contingency Problem: AI Recursion and the Limits of Interspecies Understanding

Graham L. Bishop

Current bioacoustic AI systems achieve impressive cross-species performance by processing animal communication through transformer architectures, foundation model paradigms, and other computational approaches. However, these approaches overlook a fundamental question: what happens when one form of recursive cognition–AI systems with their attention mechanisms, iterative processing, and feedback loops–encounters the recursive communicative processes of other species? Drawing on philosopher Yuk Hui's work on recursivity and contingency, I argue that AI systems are not neutral pattern detectors but recursive cognitive agents whose own information processing may systematically obscure or distort other species' communicative structures. This creates a double contingency problem: each species' communication emerges through contingent ecological and evolutionary conditions, while AI systems process these signals through their own contingent architectural and training conditions. I propose that addressing this challenge requires reconceptualizing bioacoustic AI from universal pattern recognition toward diplomatic encounter between different forms of recursive cognition, with implications for model design, evaluation frameworks, and research methodologies.
Read the full abstract

Computation and Language

TiDAR: Think in Diffusion, Talk in Autoregression

Jingyu Liu, Xin Dong, Zhifan Ye et al. (+6 more)

Diffusion language models hold the promise of fast parallel generation, while autoregressive (AR) models typically excel in quality due to their causal structure aligning naturally with language modeling. This raises a fundamental question: can we achieve a synergy with high throughput, higher GPU utilization, and AR level quality? Existing methods fail to effectively balance these two aspects, either prioritizing AR using a weaker model for sequential drafting (speculative decoding), leading to lower drafting efficiency, or using some form of left-to-right (AR-like) decoding logic for diffusion, which still suffers from quality degradation and forfeits its potential parallelizability. We introduce TiDAR, a sequence-level hybrid architecture that drafts tokens (Thinking) in Diffusion and samples final outputs (Talking) AutoRegressively - all within a single forward pass using specially designed structured attention masks. This design exploits the free GPU compute density, achieving a strong balance between drafting and verification capacity. Moreover, TiDAR is designed to be serving-friendly (low overhead) as a standalone model. We extensively evaluate TiDAR against AR models, speculative decoding, and diffusion variants across generative and likelihood tasks at 1.5B and 8B scales. Thanks to the parallel drafting and sampling as well as exact KV cache support, TiDAR outperforms speculative decoding in measured throughput and surpasses diffusion models like Dream and Llada in both efficiency and quality. Most notably, TiDAR is the first architecture to close the quality gap with AR models while delivering 4.71x to 5.91x more tokens per second.
Read the full abstract

All authors Jingyu Liu, Xin Dong, Zhifan Ye, Rishabh Mehta, Yonggan Fu, Vartika Singh, Jan Kautz, Ce Zhang, Pavlo Molchanov

Computation and Language

HalluClean: A Unified Framework to Combat Hallucinations in LLMs

Yaxin Zhao and Yu Zhang

Large language models (LLMs) have achieved impressive performance across a wide range of natural language processing tasks, yet they often produce hallucinated content that undermines factual reliability. To address this challenge, we introduce HalluClean, a lightweight and task-agnostic framework for detecting and correcting hallucinations in LLM-generated text. HalluClean adopts a reasoning-enhanced paradigm, explicitly decomposing the process into planning, execution, and revision stages to identify and refine unsupported claims. It employs minimal task-routing prompts to enable zero-shot generalization across diverse domains, without relying on external knowledge sources or supervised detectors. We conduct extensive evaluations on five representative tasks-question answering, dialogue, summarization, math word problems, and contradiction detection. Experimental results show that HalluClean significantly improves factual consistency and outperforms competitive baselines, demonstrating its potential to enhance the trustworthiness of LLM outputs in real-world applications.
Read the full abstract

Computation and Language

Hierarchical Memorization in Large Language Models: Evidence from Citation Generation

Junichiro Niimi

Large language models (LLMs) generate fluent text across a wide range of tasks, but the fabrication of non-existent academic citations remains a critical and well-documented failure mode. Building on prior work that frames hallucination and verbatim memorization as outcomes of the same probabilistic process, this study uses citation count as a proxy for training data redundancy and asks how this redundancy is internally structured within a single bibliographic record. Using GPT-4.1, we generated and manually verified 100 citations across twenty computer-science domains, measuring factual fidelity via cosine similarity against authentic metadata. We find that (i) factual accuracy varies substantially across domains and scales log-linearly with citation count, (ii) the model crosses two empirically identifiable thresholds; an inflection around 90 citations and a saturation point near 1,200 citations beyond which records are reproduced nearly verbatim, (iii) memorization is hierarchical, with titles and first authors recalled earliest while venues and numeric fields require far greater redundancy and publication years remain essentially unlearned, and (iv) even highly cited records can be conflated when their titles and authors overlap, an effect interpretable as spurious-attractor interference. Memorization in LLMs is therefore not a binary on/off state but a graduated, hierarchically layered phenomenon shaped by the uneven distribution of knowledge in the pretraining corpus.
Read the full abstract

Computation and Language

BioVerge: A Comprehensive Benchmark and Study of Self-Evaluating Agents for Biomedical Hypothesis Generation

Fuyi Yang, Chenchen Ye, Mingyu Derek Ma et al. (+3 more)

Hypothesis generation in biomedical research has traditionally centered on uncovering hidden relationships within vast scientific literature, often using methods like Literature-Based Discovery (LBD). Despite progress, current approaches typically depend on single data types or predefined extraction patterns, which restricts the discovery of novel and complex connections. Recent advances in Large Language Model (LLM) agents show significant potential, with capabilities in information retrieval, reasoning, and generation. However, their application to biomedical hypothesis generation has been limited by the absence of standardized datasets and execution environments. To address this, we introduce BioVerge, a comprehensive benchmark, and BioVerge Agent, an LLM-based agent framework, to create a standardized environment for exploring biomedical hypothesis generation at the frontier of existing scientific knowledge. Our dataset includes structured and textual data derived from historical biomedical hypotheses and PubMed literature, organized to support exploration by LLM agents. BioVerge Agent utilizes a ReAct-based approach with distinct Generation and Evaluation modules that iteratively produce and self-assess hypothesis proposals. Through extensive experimentation, we uncover key insights: 1) different architectures of BioVerge Agent influence exploration diversity and reasoning strategies; 2) structured and textual information sources each provide unique, critical contexts that enhance hypothesis generation; and 3) self-evaluation significantly improves the novelty and relevance of proposed hypotheses.
Read the full abstract

All authors Fuyi Yang, Chenchen Ye, Mingyu Derek Ma, Yijia Xiao, Matthew Yang, Wei Wang

Tuesday 11 November

Computation and Language

DeCoRL: Decoupling Reasoning Chains via Parallel Sub-Step Generation and Cascaded Reinforcement for Interpretable and Scalable RLHF

Ziyuan Gao, Di Liang, Xianjie Wu et al. (+2 more)

Existing reinforcement learning methods for Chain-of-Thought reasoning suffer from two critical limitations. First, they operate as monolithic black boxes that provide undifferentiated reward signals, obscuring individual step contributions and hindering error diagnosis. Second, sequential decoding has O(n) time complexity. This makes real-time deployment impractical for complex reasoning tasks. We present DeCoRL (Decoupled Reasoning Chains via Coordinated Reinforcement Learning), a novel framework that transforms reasoning from sequential processing into collaborative modular orchestration. DeCoRL trains lightweight specialized models to generate reasoning sub-steps concurrently, eliminating sequential bottlenecks through parallel processing. To enable precise error attribution, the framework designs modular reward functions that score each sub-step independently. Cascaded DRPO optimization then coordinates these rewards while preserving inter-step dependencies. Comprehensive evaluation demonstrates state-of-the-art results across RM-Bench, RMB, and RewardBench, outperforming existing methods including large-scale models. DeCoRL delivers 3.8 times faster inference while maintaining superior solution quality and offers a 22.7% improvement in interpretability through explicit reward attribution. These advancements, combined with a 72.4% reduction in energy consumption and a 68% increase in throughput, make real-time deployment of complex reasoning systems a reality.
Read the full abstract

All authors Ziyuan Gao, Di Liang, Xianjie Wu, Philippe Morel, Minlong Peng

Computation and Language · Cross-listed from cs.LG

H-Model: Dynamic Neural Architectures for Adaptive Processing

Dmytro Hospodarchuk

This article explores the design and experimentation of a neural network architecture capable of dynamically adjusting its internal structure based on the input data. The proposed model introduces a routing mechanism that allows each layer to influence how its outputs are propagated through the network, enabling iterative and adaptive computation. This concept is loosely inspired by the idea of thought processes and dynamic reasoning, where information flow is conditioned not only on the data itself, but also on the internal state of the system. It is important to note that this work does not aim to compete with state-of-the-art language models in terms of performance. Instead, it presents a conceptual prototype-an architectural framework that opens up a new direction for exploring adaptable and potentially more interpretable networks. The goal is not optimization of existing benchmarks but rather the proposal of a system that can learn not only representations, but also the structure of computation itself. Due to practical constraints in computing resources and data, this study remains a preliminary investigation. Nevertheless, initial observations show promise, and the architecture's full potential can only be evaluated in future experiments under more favorable computational conditions.
Read the full abstract

Computation and Language · Cross-listed from cs.MA

Who Gets the Reward & Who Gets the Blame? Evaluation-Aligned Training Signals for Multi-LLM Agents

Chih-Hsuan Yang, Tanwi Mallick, Le Chen et al. (+4 more)

Large Language Models (LLMs) in multi-agent systems (MAS) have shown promise for complex tasks, yet current training methods lack principled ways to connect system-level evaluation with agent- and message-level learning. We propose a theoretical framework that unifies cooperative game-theoretic attribution with process reward modeling to transform system evaluation to agent credit to response-level signals. Unlike prior approaches that rely only on attribution (Shapley) or step-level labels (PRM), our method produces local, signed, and credit-conserving signals. In success cases, Shapley-based credit assignment fairly allocates outcomes across agents and is refined into per-message rewards that promote cooperation while discouraging redundancy or sabotage; in failure cases, first-error localization yields repair-aware preferences that penalize harmful steps while rewarding corrective attempts. The resulting signals are bounded, cooperative, and directly compatible with reinforcement- or preference-based post-training, providing a unified and auditable pathway from global evaluation to local supervision in LLM multi-agent training. Our contribution is conceptual: we present a theoretical foundation and training signals, leaving empirical validation for future work.
Read the full abstract

All authors Chih-Hsuan Yang, Tanwi Mallick, Le Chen, Krishnan Raghavan, Amal Gueroudji, Ian T. Foster, Rajeev Thakur

Computation and Language

A methodological analysis of prompt perturbations and their effect on attack success rates

Tiago Machado, Maysa Malfiza Garcia de Macedo, Rogerio Abreu de Paula et al. (+5 more)

This work aims to investigate how different Large Language Models (LLMs) alignment methods affect the models' responses to prompt attacks. We selected open source models based on the most common alignment methods, namely, Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Reinforcement Learning with Human Feedback (RLHF). We conducted a systematic analysis using statistical methods to verify how sensitive the Attack Success Rate (ASR) is when we apply variations to prompts designed to elicit inappropriate content from LLMs. Our results show that even small prompt modifications can significantly change the Attack Success Rate (ASR) according to the statistical tests we run, making the models more or less susceptible to types of attack. Critically, our results demonstrate that running existing 'attack benchmarks' alone may not be sufficient to elicit all possible vulnerabilities of both models and alignment methods. This paper thus contributes to ongoing efforts on model attack evaluation by means of systematic and statistically-based analyses of the different alignment methods and how sensitive their ASR is to prompt variation.
Read the full abstract

All authors Tiago Machado, Maysa Malfiza Garcia de Macedo, Rogerio Abreu de Paula, Marcelo Carpinette Grave, Aminat Adebiyi, Luan Soares de Souza, Enrico Santarelli, Claudio Pinhanez

Computation and Language

SpiderGen: Towards Procedure Generation For Carbon Life Cycle Assessments with Generative AI

Anupama Sitaraman, Bharathan Balaji and Yuvraj Agarwal

Investigating the effects of climate change and global warming caused by GHG emissions have been a key concern worldwide. These emissions are largely contributed to by the production, use and disposal of consumer products. Thus, it is important to build tools to estimate the environmental impact of consumer goods, an essential part of which is conducting Life Cycle Assessments (LCAs). LCAs specify and account for the appropriate processes involved with the production, use, and disposal of the products. We present SpiderGen, an LLM-based workflow which integrates the taxonomy and methodology of traditional LCA with the reasoning capabilities and world knowledge of LLMs to generate graphical representations of the key procedural information used for LCA, known as Product Category Rules Process Flow Graphs (PCR PFGs). We additionally evaluate the output of SpiderGen by comparing it with 65 real-world LCA documents. We find that SpiderGen provides accurate LCA process information that is either fully correct or has minor errors, achieving an F1-Score of 65% across 10 sample data points, as compared to 53% using a one-shot prompting method. We observe that the remaining errors occur primarily due to differences in detail between LCA documents, as well as differences in the "scope" of which auxiliary processes must also be included. We also demonstrate that SpiderGen performs better than several baselines techniques, such as chain-of-thought prompting and one-shot prompting. Finally, we highlight SpiderGen's potential to reduce the human effort and costs for estimating carbon impact, as it is able to produce LCA process information for less than \$1 USD in under 10 minutes as compared to the status quo LCA, which can cost over \$25000 USD and take up to 21-person days.
Read the full abstract