Peer Revenue Past citations are no guarantee of future returns

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Wednesday 28 January

Computation and Language

Beyond Divergent Creativity: A Human-Based Evaluation of Creativity in Large Language Models

Kumiko Nakajima, Jan Zuiderveld and Sandro Pezzelle

Large language models (LLMs) are increasingly used in verbal creative tasks. However, previous assessments of the creative capabilities of LLMs remain weakly grounded in human creativity theory and are thus hard to interpret. The widely used Divergent Association Task (DAT) focuses on novelty, ignoring appropriateness, a core component of creativity. We evaluate a range of state-of-the-art LLMs on DAT and show that their scores on the task are lower than those of two baselines that do not possess any creative abilities, undermining its validity for model evaluation. Grounded in human creativity theory, which defines creativity as the combination of novelty and appropriateness, we introduce Conditional Divergent Association Task (CDAT). CDAT evaluates novelty conditional on contextual appropriateness, separating noise from creativity better than DAT, while remaining simple and objective. Under CDAT, smaller model families often show the most creativity, whereas advanced families favor appropriateness at lower novelty. We hypothesize that training and alignment likely shift models along this frontier, making outputs more appropriate but less creative. We release the dataset and code.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

PathWise: Planning through World Model for Automated Heuristic Design via Self-Evolving LLMs

Oguzhan Gungordu, Siheng Xiong and Faramarz Fekri

Large Language Models (LLMs) have enabled automated heuristic design (AHD) for combinatorial optimization problems (COPs), but existing frameworks' reliance on fixed evolutionary rules and static prompt templates often leads to myopic heuristic generation, redundant evaluations, and limited reasoning about how new heuristics should be derived. We propose a novel multi-agent reasoning framework, referred to as Planning through World Model for Automated Heuristic Design via Self-Evolving LLMs (PathWise), which formulates heuristic generation as a sequential decision process over an entailment graph serving as a compact, stateful memory of the search trajectory. This approach allows the system to carry forward past decisions and reuse or avoid derivation information across generations. A policy agent plans evolutionary actions, a world model agent generates heuristic rollouts conditioned on those actions, and critic agents provide routed reflections summarizing lessons from prior steps, shifting LLM-based AHD from trial-and-error evolution toward state-aware planning through reasoning. Experiments across diverse COPs show that PathWise converges faster to better heuristics, generalizes across different LLM backbones, and scales to larger problem sizes.
Read the full abstract

Computation and Language

In-Context Learning for Diagram Code Generation: What is the Effect on Faithfulness and Quality?

Evanfiya Logacheva, Arto Hellas, Tsvetomila Mihaylova et al. (+3 more)

We investigate how provision of in-context examples based on discourse similarity affects quality and hallucination in diagram code generation from educational materials. Using a sample of texts as sources for diagram generation, we test two methods using a conventional text-to-diagram task format, instruction-based zero-shot generation and ICL with text-to-diagram examples, and an unconventional ICL method with a discourse analysis-to-diagram example, which requires a model to generate a diagram from a discourse analysis. We evaluate AI-generated diagrams for faithfulness to provided context and their quality outcomes using both human and automated evaluation. Bayesian data analysis of the human diagram evaluation shows that there is significant evidence that the unconventional task format improves context faithfulness in diagram generation compared to the zero-shot instruction with reducing the odds of context faithfulness hallucination by nearly $70\%$; however, we do not find the ICL having a significant effect on the diagram aesthetic presentation and superficial coherence. Evaluation with the LLM-as-a-judge G-Eval metric shows nearly 4.5-fold increased odds of the perfect faithfulness outcome for the novel ICL task. Our study provides a novel insight into ICL format effect on superficial fluency and faithfulness in context-grounded structured diagram code generation.
Read the full abstract

All authors Evanfiya Logacheva, Arto Hellas, Tsvetomila Mihaylova, Juha Sorva, Ava Heinonen, Juho Leinonen

Computation and Language · Cross-listed from cs.AI

CtrlCoT: Dual-Granularity Chain-of-Thought Compression for Controllable Reasoning

Zhenxuan Fan, Jie Cao, Yang Dai et al. (+5 more)

Chain-of-thought (CoT) prompting improves LLM reasoning but incurs high latency and memory cost due to verbose traces, motivating CoT compression with preserved correctness. Existing methods either shorten CoTs at the semantic level, which is often conservative, or prune tokens aggressively, which can miss task-critical cues and degrade accuracy. Moreover, combining the two is non-trivial due to sequential dependency, task-agnostic pruning, and distribution mismatch. We propose CtrlCoT, a dual-granularity CoT compression framework that harmonizes semantic abstraction and token-level pruning through three components: Hierarchical Reasoning Abstraction produces CoTs at multiple semantic granularities; Logic-Preserving Distillation trains a logic-aware pruner to retain indispensable reasoning cues (e.g., numbers and operators) across pruning ratios; and Distribution-Alignment Generation aligns compressed traces with fluent inference-time reasoning styles to avoid fragmentation. On MATH-500 with Qwen2.5-7B-Instruct, CtrlCoT uses 30.7% fewer tokens while achieving 7.6 percentage points higher than the strongest baseline, demonstrating more efficient and reliable reasoning. Our code will be publicly available at https://github.com/fanzhenxuan/Ctrl-CoT.
Read the full abstract

All authors Zhenxuan Fan, Jie Cao, Yang Dai, Zheqi Lv, Wenqiao Zhang, Zhongle Xie, Peng LU, Beng Chin Ooi

Computation and Language

BMAM: Brain-inspired Multi-Agent Memory Framework

Yang Li, Jiaxiang Liu, Yusong Wang et al. (+2 more)

Language-model-based agents operating over extended interaction horizons face persistent challenges in preserving temporally grounded information and maintaining behavioral consistency across sessions, a failure mode we term soul erosion. We present BMAM (Brain-inspired Multi-Agent Memory), a general-purpose memory architecture that models agent memory as a set of functionally specialized subsystems rather than a single unstructured store. Inspired by cognitive memory systems, BMAM decomposes memory into episodic, semantic, salience-aware, and control-oriented components that operate at complementary time scales. To support long-horizon reasoning, BMAM organizes episodic memories along explicit timelines and retrieves evidence by fusing multiple complementary signals. Experiments on the LoCoMo benchmark show that BMAM achieves 78.45 percent accuracy under the standard long-horizon evaluation setting, and ablation analyses confirm that the hippocampus-inspired episodic memory subsystem plays a critical role in temporal reasoning.
Read the full abstract

All authors Yang Li, Jiaxiang Liu, Yusong Wang, Yujie Wu, Mingkun Xu

Computation and Language

MuVaC: A Variational Causal Framework for Multimodal Sarcasm Understanding in Dialogues

Diandian Guo, Fangfang Yuan, Cong Cao et al. (+5 more)

The prevalence of sarcasm in multimodal dialogues on the social platforms presents a crucial yet challenging task for understanding the true intent behind online content. Comprehensive sarcasm analysis requires two key aspects: Multimodal Sarcasm Detection (MSD) and Multimodal Sarcasm Explanation (MuSE). Intuitively, the act of detection is the result of the reasoning process that explains the sarcasm. Current research predominantly focuses on addressing either MSD or MuSE as a single task. Even though some recent work has attempted to integrate these tasks, their inherent causal dependency is often overlooked. To bridge this gap, we propose MuVaC, a variational causal inference framework that mimics human cognitive mechanisms for understanding sarcasm, enabling robust multimodal feature learning to jointly optimize MSD and MuSE. Specifically, we first model MSD and MuSE from the perspective of structural causal models, establishing variational causal pathways to define the objectives for joint optimization. Next, we design an alignment-then-fusion approach to integrate multimodal features, providing robust fusion representations for sarcasm detection and explanation generation. Finally, we enhance the reasoning trustworthiness by ensuring consistency between detection results and explanations. Experimental results demonstrate the superiority of MuVaC in public datasets, offering a new perspective for understanding multimodal sarcasm.
Read the full abstract

All authors Diandian Guo, Fangfang Yuan, Cong Cao, Xixun Lin, Chuan Zhou, Hao Peng, Yanan Cao, Yanbing Liu

Computation and Language

PEARL: Plan Exploration and Adaptive Reinforcement Learning for Multihop Tool Use

Qihao Wang, Mingzhe Lu, Jiayue Wu et al. (+2 more)

Large Language Models show great potential with external tools, but face significant challenges in complex, multi-turn tool invocation. They often exhibit weak planning, tool hallucination, erroneous parameter generation, and struggle with robust interaction. To tackle these issues, we present PEARL, a novel framework to enhance LLM planning and execution for sophisticated tool use. PEARL adopts a two-stage approach: an offline phase where the agent explores tools to learn valid usage patterns and failure conditions, and an online reinforcement learning phase. In the online phase, a dedicated Planner is trained via group Relative Policy Optimization (GRPO) with a carefully designed reward function that provides distinct signals for planning quality. Experiments on the ToolHop and T-Eval benchmarks show PEARL significantly outperforms existing methods, achieving a new state-of-the-art success rate of 56.5% on ToolHop while maintaining a low invocation error rate. Our work marks a key advance in addressing the complex planning challenges of tool use, contributing to the development of more robust and reliable LLM-based agents.
Read the full abstract

All authors Qihao Wang, Mingzhe Lu, Jiayue Wu, Yue Hu, Yanbing Liu

Computation and Language

Hopes and Fears – Emotion Distribution in the Topic Landscape of Finnish Parliamentary Speech 2000-2020

Anna Ristilä, Otto Tarkka, Veronika Laippala et al. (+1 more)

Existing research often treats parliamentary discourse as a homogeneous whole, overlooking topic-specific patterns. Parliamentary speeches address a wide range of topics, some of which evoke stronger emotions than others. While everyone has intuitive assumptions about what the most emotive topics in a parliament may be, there has been little research into the emotions typically linked to different topics. This paper strives to fill this gap by examining emotion expression among the topics of parliamentary speeches delivered in Eduskunta, the Finnish Parliament, between 2000 and 2020. An emotion analysis model is used to investigate emotion expression in topics, from both synchronic and diachronic perspectives. The results strengthen evidence of increasing positivity in parliamentary speech and provide further insights into topic-specific emotion expression within parliamentary debate.
Read the full abstract

All authors Anna Ristilä, Otto Tarkka, Veronika Laippala, Kimmo Elo

Computation and Language

SpeechMapper: Speech-to-text Embedding Projector for LLMs

Biswesh Mohapatra, Marcely Zanon Boito and Ioan Calapodescu

Current speech LLMs bridge speech foundation models to LLMs using projection layers, training all of these components on speech instruction data. This strategy is computationally intensive and susceptible to task and prompt overfitting. We present SpeechMapper, a cost-efficient speech-to-LLM-embedding training approach that mitigates overfitting, enabling more robust and generalizable models. Our model is first pretrained without the LLM on inexpensive hardware, and then efficiently attached to the target LLM via a brief 1K-step instruction tuning (IT) stage. Through experiments on speech translation and spoken question answering, we demonstrate the versatility of SpeechMapper's pretrained block, presenting results for both task-agnostic IT, an ASR-based adaptation strategy that does not train in the target task, and task-specific IT. In task-agnostic settings, Speechmapper rivals the best instruction-following speech LLM from IWSLT25, despite never being trained on these tasks, while in task-specific settings, it outperforms this model across many datasets, despite requiring less data and compute. Overall, SpeechMapper offers a practical and scalable approach for efficient, generalizable speech-LLM integration without large-scale IT.
Read the full abstract

Computation and Language

Beyond Accuracy: A Cognitive Load Framework for Mapping the Capability Boundaries of Tool-use Agents

Qihao Wang, Yue Hu, Mingzhe Lu et al. (+3 more)

The ability of Large Language Models (LLMs) to use external tools unlocks powerful real-world interactions, making rigorous evaluation essential. However, current benchmarks primarily report final accuracy, revealing what models can do but obscuring the cognitive bottlenecks that define their true capability boundaries. To move from simple performance scoring to a diagnostic tool, we introduce a framework grounded in Cognitive Load Theory. Our framework deconstructs task complexity into two quantifiable components: Intrinsic Load, the inherent structural complexity of the solution path, formalized with a novel Tool Interaction Graph; and Extraneous Load, the difficulty arising from ambiguous task presentation. To enable controlled experiments, we construct ToolLoad-Bench, the first benchmark with parametrically adjustable cognitive load. Our evaluation reveals distinct performance cliffs as cognitive load increases, allowing us to precisely map each model's capability boundary. We validate that our framework's predictions are highly calibrated with empirical results, establishing a principled methodology for understanding an agent's limits and a practical foundation for building more efficient systems.
Read the full abstract

All authors Qihao Wang, Yue Hu, Mingzhe Lu, Jiayue Wu, Yanbing Liu, Yuanmin Tang

Computation and Language · Cross-listed from cs.LG

LLM-AutoDP: Automatic Data Processing via LLM Agents for Model Fine-tuning

Wei Huang, Anda Cheng, Yinggui Wang et al. (+2 more)

Large Language Models (LLMs) can be fine-tuned on domain-specific data to enhance their performance in specialized fields. However, such data often contains numerous low-quality samples, necessitating effective data processing (DP). In practice, DP strategies are typically developed through iterative manual analysis and trial-and-error adjustment. These processes inevitably incur high labor costs and may lead to privacy issues in high-privacy domains like healthcare due to direct human access to sensitive data. Thus, achieving automated data processing without exposing the raw data has become a critical challenge. To address this challenge, we propose LLM-AutoDP, a novel framework that leverages LLMs as agents to automatically generate and optimize data processing strategies. Our method generates multiple candidate strategies and iteratively refines them using feedback signals and comparative evaluations. This iterative in-context learning mechanism enables the agent to converge toward high-quality processing pipelines without requiring direct human intervention or access to the underlying data. To further accelerate strategy search, we introduce three key techniques: Distribution Preserving Sampling, which reduces data volume while maintaining distributional integrity; Processing Target Selection, which uses a binary classifier to identify low-quality samples for focused processing; Cache-and-Reuse Mechanism}, which minimizes redundant computations by reusing prior processing results. Results show that models trained on data processed by our framework achieve over 80% win rates against models trained on unprocessed data. Compared to AutoML baselines based on LLM agents, LLM-AutoDP achieves approximately a 65% win rate. Moreover, our acceleration techniques reduce the total searching time by up to 10 times, demonstrating both effectiveness and efficiency.
Read the full abstract

All authors Wei Huang, Anda Cheng, Yinggui Wang, Lei Wang, Tao Wei

Computation and Language · Cross-listed from cs.LG

TABED: Test-Time Adaptive Ensemble Drafting for Robust Speculative Decoding in LVLMs

Minjae Lee, Wonjun Kang, Byeongkeun Ahn et al. (+6 more)

Speculative decoding (SD) has proven effective for accelerating LLM inference by quickly generating draft tokens and verifying them in parallel. However, SD remains largely unexplored for Large Vision-Language Models (LVLMs), which extend LLMs to process both image and text prompts. To address this gap, we benchmark existing inference methods with small draft models on 11 datasets across diverse input scenarios and observe scenario-specific performance fluctuations. Motivated by these findings, we propose Test-time Adaptive Batched Ensemble Drafting (TABED), which dynamically ensembles multiple drafts obtained via batch inference by leveraging deviations from past ground truths available in the SD setting. The dynamic ensemble method achieves an average robust walltime speedup of 1.74x over autoregressive decoding and a 5% improvement over single drafting methods, while remaining training-free and keeping ensembling costs negligible through parameter sharing. With its plug-and-play compatibility, we further enhance TABED by integrating advanced verification and alternative drafting methods. Code and custom-trained models are available at https://github.com/furiosa-ai/TABED.
Read the full abstract

All authors Minjae Lee, Wonjun Kang, Byeongkeun Ahn, Christian Classen, Kevin Galim, Seunghyuk Oh, Minghao Yan, Hyung Il Koo, Kangwook Lee

Computation and Language

Improving Diffusion Language Model Decoding through Joint Search in Generation Order and Token Space

Yangyi Shen, Tianjian Feng, Jiaqi Han et al. (+5 more)

Diffusion Language Models (DLMs) offer order-agnostic generation that can explore many possible decoding trajectories. However, current decoding methods commit to a single trajectory, limiting exploration in trajectory space. We introduce Order-Token Search to explore this space through jointly searching over generation order and token values. Its core is a likelihood estimator that scores denoising actions, enabling stable pruning and efficient exploration of diverse trajectories. Across mathematical reasoning and coding benchmarks, Order-Token Search consistently outperforms baselines on GSM8K, MATH500, Countdown, and HumanEval (3.1%, 3.8%, 7.9%, and 6.8% absolute over backbone), matching or surpassing diffu-GRPO post-trained d1-LLaDA. Our work establishes joint search as a key component for advancing decoding in DLMs.
Read the full abstract

All authors Yangyi Shen, Tianjian Feng, Jiaqi Han, Wen Wang, Tianlang Chen, Chunhua Shen, Jure Leskovec, Stefano Ermon

Computation and Language

MobileBench-OL: A Comprehensive Chinese Benchmark for Evaluating Mobile GUI Agents in Real-World Environment

Qinzhuo Wu, Zhizhuo Yang, Hanhao Li et al. (+3 more)

Recent advances in mobile Graphical User Interface (GUI) agents highlight the growing need for comprehensive evaluation benchmarks. While new online benchmarks offer more realistic testing than offline ones, they tend to focus on the agents' task instruction-following ability while neglecting their reasoning and exploration ability. Moreover, these benchmarks do not consider the random noise in real-world mobile environments. This leads to a gap between benchmarks and real-world environments. To addressing these limitations, we propose MobileBench-OL, an online benchmark with 1080 tasks from 80 Chinese apps. It measures task execution, complex reasoning, and noise robustness of agents by including 5 subsets, which set multiple evaluation dimensions. We also provide an auto-eval framework with a reset mechanism, enabling stable and repeatable real-world benchmarking. Evaluating 12 leading GUI agents on MobileBench-OL shows significant room for improvement to meet real-world requirements. Human evaluation further confirms that MobileBench-OL can reliably measure the performance of leading GUI agents in real environments. Our data and code will be released upon acceptance.
Read the full abstract

All authors Qinzhuo Wu, Zhizhuo Yang, Hanhao Li, Pengzhi Gao, Wei Liu, Jian Luan

Computation and Language

PsychePass: Calibrating LLM Therapeutic Competence via Trajectory-Anchored Tournaments

Zhuang Chen, Dazhen Wan, Zhangkai Zheng et al. (+4 more)

While large language models show promise in mental healthcare, evaluating their therapeutic competence remains challenging due to the unstructured and longitudinal nature of counseling. We argue that current evaluation paradigms suffer from an unanchored defect, leading to two forms of instability: process drift, where unsteered client simulation wanders away from specific counseling goals, and standard drift, where static pointwise scoring lacks the stability for reliable judgment. To address this, we introduce Ps, a unified framework that calibrates the therapeutic competence of LLMs via trajectory-anchored tournaments. We first anchor the interaction trajectory in simulation, where clients precisely control the fluid consultation process to probe multifaceted capabilities. We then anchor the battle trajectory in judgments through an efficient Swiss-system tournament, utilizing dynamic pairwise battles to yield robust Elo ratings. Beyond ranking, we demonstrate that tournament trajectories can be transformed into credible reward signals, enabling on-policy reinforcement learning to enhance LLMs' performance. Extensive experiments validate the effectiveness of PsychePass and its strong consistency with human expert judgments.
Read the full abstract

All authors Zhuang Chen, Dazhen Wan, Zhangkai Zheng, Guanqun Bi, Xiyao Xiao, Binghang Li, Minlie Huang

Computation and Language

CE-RM: A Pointwise Generative Reward Model Optimized via Two-Stage Rollout and Unified Criteria

Xinyu Hu, Yancheng He, Weixun Wang et al. (+6 more)

Automatic evaluation is crucial yet challenging for open-ended natural language generation, especially when rule-based metrics are infeasible. Compared with traditional methods, the recent LLM-as-a-Judge paradigms enable better and more flexible evaluation, and show promise as generative reward models for reinforcement learning. However, prior work has revealed a notable gap between their seemingly impressive benchmark performance and actual effectiveness in RL practice. We attribute this issue to some limitations in existing studies, including the dominance of pairwise evaluation and inadequate optimization of evaluation criteria. Therefore, we propose CE-RM-4B, a pointwise generative reward model trained with a dedicated two-stage rollout method, and adopting unified query-based criteria. Using only about 5.7K high-quality data curated from the open-source preference dataset, our CE-RM-4B achieves superior performance on diverse reward model benchmarks, especially in Best-of-N scenarios, and delivers more effective improvements in downstream RL practice.
Read the full abstract

All authors Xinyu Hu, Yancheng He, Weixun Wang, Tao Feng, Li Lin, Jiashun Liu, Wenbo Su, Bo Zheng, Xiaojun Wan

Computation and Language

Beyond Speedup – Utilizing KV Cache for Sampling and Reasoning

Zeyu Xing, Xing Li, Hui-Ling Zhen et al. (+2 more)

KV caches, typically used only to speed up autoregressive decoding, encode contextual information that can be reused for downstream tasks at no extra cost. We propose treating the KV cache as a lightweight representation, eliminating the need to recompute or store full hidden states. Despite being weaker than dedicated embeddings, KV-derived representations are shown to be sufficient for two key applications: (i) Chain-of-Embedding, where they achieve competitive or superior performance on Llama-3.1-8B-Instruct and Qwen2-7B-Instruct; and (ii) Fast/Slow Thinking Switching, where they enable adaptive reasoning on Qwen3-8B and DeepSeek-R1-Distil-Qwen-14B, reducing token generation by up to $5.7\times$ with minimal accuracy loss. Our findings establish KV caches as a free, effective substrate for sampling and reasoning, opening new directions for representation reuse in LLM inference. Code: https://github.com/cmd2001/ICLR2026_KV-Embedding.
Read the full abstract

All authors Zeyu Xing, Xing Li, Hui-Ling Zhen, Mingxuan Yuan, Sinno Jialin Pan

Computation and Language

SAPO: Self-Adaptive Process Optimization Makes Small Reasoners Stronger

Kaiyuan Chen, Guangmin Zheng, Jin Wang et al. (+2 more)

Existing self-evolution methods overlook the influence of fine-grained reasoning steps, which leads to the reasoner-verifier gap. The computational inefficiency of Monte Carlo (MC) process supervision further exacerbates the difficulty in mitigating the gap. Motivated by the Error-Related Negativity (ERN), which the reasoner can localize error following incorrect decisions, guiding rapid adjustments, we propose a Self-Adaptive Process Optimization (SAPO) method for self-improvement in Small Language Models (SLMs). SAPO adaptively and efficiently introduces process supervision signals by actively minimizing the reasoner-verifier gap rather than relying on inefficient MC estimations. Extensive experiments demonstrate that the proposed method outperforms most existing self-evolution methods on two challenging task types: mathematics and code. Additionally, to further investigate SAPO's impact on verifier performance, this work introduces two new benchmarks for process reward models in both mathematical and coding tasks.
Read the full abstract

All authors Kaiyuan Chen, Guangmin Zheng, Jin Wang, Xiaobing Zhou, Xuejie Zhang

Computation and Language

MiLorE-SSL: Scaling Multilingual Capabilities in Self-Supervised Models without Forgetting

Jing Xu, Minglin Wu, Xueyuan Chen et al. (+2 more)

Self-supervised learning (SSL) has greatly advanced speech representation learning, but multilingual SSL models remain constrained to languages encountered during pretraining. Retraining from scratch to incorporate new languages is computationally expensive, while sequential training without migitation strategies often leads to catastrophic forgetting. To address this, we propose MiLorE-SSL, a lightweight framework that combines LoRA modules with a soft mixture-of-experts (MoE) mechanism for efficient continual multilingual training. LoRA provides efficient low-rank adaptation, while soft MoE promotes flexible expert sharing across languages, reducing cross-lingual interference. To further mitigate forgetting, we introduce limited replay data from existing languages, avoiding reliance on large historical corpora. Experiments on ML-SUPERB demonstrate that MiLorE-SSL achieves strong performance in new languages and improves the ability in existing ones with only 2.14% trainable parameters.
Read the full abstract

All authors Jing Xu, Minglin Wu, Xueyuan Chen, Xixin Wu, Helen Meng

Computation and Language · Cross-listed from cs.LG

Truthfulness Despite Weak Supervision: Evaluating and Training LLMs Using Peer Prediction

Tianyi Alex Qiu, Micah Carroll and Cameron Allen

The evaluation and post-training of large language models (LLMs) rely on supervision, but strong supervision for difficult tasks is often unavailable, especially when evaluating frontier models. In such cases, models are demonstrated to exploit evaluations built on such imperfect supervision, leading to deceptive results. However, underutilized in LLM research, a wealth of mechanism design research focuses on game-theoretic incentive compatibility, i.e., eliciting honest and informative answers with weak supervision. Drawing from this literature, we introduce the peer prediction method for model evaluation and post-training. It rewards honest and informative answers over deceptive and uninformative ones, using a metric based on mutual predictability and without requiring ground truth labels. We demonstrate the method's effectiveness and resistance to deception, with both theoretical guarantees and empirical validation on models with up to 405B parameters. We show that training an 8B model with peer prediction-based reward recovers most of the drop in truthfulness due to prior malicious finetuning, even when the reward is produced by a 0.135B language model with no finetuning. On the evaluation front, in contrast to LLM-as-a-Judge which requires strong and trusted judges, we discover an inverse scaling property in peer prediction, where, surprisingly, resistance to deception is strengthened as the capability gap between the experts and participants widens, enabling reliable evaluation of strong models with weak supervision. In particular, LLM-as-a-Judge become worse than random guess when facing deceptive models 5-20x the judge's size, while peer prediction thrives when such gaps are large, including in cases with over 100x size difference.
Read the full abstract

Computation and Language · Cross-listed from cs.IR

One Word is Enough: Minimal Adversarial Perturbations for Neural Text Ranking

Tanmay Karmakar, Sourav Saha, Debapriyo Majumdar et al. (+1 more)

Neural ranking models (NRMs) achieve strong retrieval effectiveness, yet prior work has shown they are vulnerable to adversarial perturbations. We revisit this robustness question with a minimal, query-aware attack that promotes a target document by inserting or substituting a single, semantically aligned word - the query center. We study heuristic and gradient-guided variants, including a white-box method that identifies influential insertion points. On TREC-DL 2019/2020 with BERT and monoT5 re-rankers, our single-word attacks achieve up to 91% success while modifying fewer than two tokens per document on average, achieving competitive rank and score boosts with far fewer edits under a comparable white-box setup to ensure fair evaluation against PRADA. We also introduce new diagnostic metrics to analyze attack sensitivity beyond aggregate success rates. Our analysis reveals a Goldilocks zone in which mid-ranked documents are most vulnerable. These findings demonstrate practical risks and motivate future defenses for robust neural ranking.
Read the full abstract

All authors Tanmay Karmakar, Sourav Saha, Debapriyo Majumdar, Surjyanee Halder

Computation and Language

Beyond the Needle's Illusion: Decoupled Evaluation of Evidence Access and Use under Semantic Interference at 326M-Token Scale

Tianwei Lin, Zuyi Zhou, Xinda Zhao et al. (+6 more)

Long-context LLM agents must access the right evidence from large environments and use it faithfully. However, the popular Needle-in-a-Haystack (NIAH) evaluation mostly measures benign span localization. The needle is near-unique, and the haystack is largely irrelevant. We introduce EverMemBench-S (EMB-S), an adversarial NIAH-style benchmark built on a 326M-token MemoryBank. While the full MemoryBank spans 326M tokens for retrieval-based (RAG) evaluation, we evaluate native long-context models only at scales that fit within each model's context window (up to 1M tokens in this work) to ensure a fair comparison. EMB-S pairs queries with collision-tested near-miss hard negatives and gold evidence sets spanning one or more documents, validated via human screening and LLM verification. We also propose a decoupled diagnostic protocol that reports evidence access (document-ID localization) separately from end-to-end QA quality under full-context prompting. This enables consistent diagnosis for both native long-context prompting and retrieval pipelines. Across a reference-corpus ladder from domain-isolated 64K contexts to a globally shared 326M-token environment, we observe a clear reality gap. Systems that saturate benign NIAH degrade sharply in evidence access under semantic interference. These results indicate that semantic discrimination, not context length alone, is the dominant bottleneck for long-context memory at scale.
Read the full abstract

All authors Tianwei Lin, Zuyi Zhou, Xinda Zhao, Chenke Wang, Xiaohong Li, Yu Chen, Chuanrui Hu, Jian Pei, Yafeng Deng

Computation and Language

RusLICA: A Russian-Language Platform for Automated Linguistic Inquiry and Category Analysis

Elina Sigdel and Anastasia Panfilova

Defining psycholinguistic characteristics in written texts is a task gaining increasing attention from researchers. One of the most widely used tools in the current field is Linguistic Inquiry and Word Count (LIWC) that originally was developed to analyze English texts and translated into multiple languages. Our approach offers the adaptation of LIWC methodology for the Russian language, considering its grammatical and cultural specificities. The suggested approach comprises 96 categories, integrating syntactic, morphological, lexical, general statistical features, and results of predictions obtained using pre-trained language models (LMs) for text analysis. Rather than applying direct translation to existing thesauri, we built the dictionary specifically for the Russian language based on the content from several lexicographic resources, semantic dictionaries and corpora. The paper describes the process of mapping lemmas to 42 psycholinguistic categories and the implementation of the analyzer as part of RusLICA web service.
Read the full abstract

Computation and Language

SoftHateBench: Evaluating Moderation Models Against Reasoning-Driven, Policy-Compliant Hostility

Xuanyu Su, Diana Inkpen and Nathalie Japkowicz

Online hate on social media ranges from overt slurs and threats (hard hate speech) to soft hate speech: discourse that appears reasonable on the surface but uses framing and value-based arguments to steer audiences toward blaming or excluding a target group. We hypothesize that current moderation systems, largely optimized for surface toxicity cues, are not robust to this reasoning-driven hostility, yet existing benchmarks do not measure this gap systematically. We introduce \textbf{SoftHateBench}, a generative benchmark that produces soft-hate variants while preserving the underlying hostile standpoint. To generate soft hate, we integrate the Argumentum Model of Topics (AMT) and Relevance Theory (RT) in a unified framework: AMT provides the backbone argument structure for rewriting an explicit hateful standpoint into a seemingly neutral discussion while preserving the stance, and RT guides generation to keep the AMT chain logically coherent. The benchmark spans 7 sociocultural domains and 28 target groups, comprising 4,745 soft-hate instances. Evaluations across encoder-based detectors, general-purpose LLMs, and safety models show a consistent drop from hard to soft tiers: systems that detect explicit hostility often fail when the same stance is conveyed through subtle, reasoning-based language. \textcolor{red}{Disclaimer. Contains offensive examples used solely for research.}
Read the full abstract

Computation and Language · Cross-listed from cs.LG

HE-SNR: Uncovering Latent Logic via Entropy for Guiding Mid-Training on SWE-bench

Yueyang Wang, Jiawei Fu, Baolong Bi et al. (+2 more)

SWE-bench has emerged as the premier benchmark for evaluating Large Language Models on complex software engineering tasks. While these capabilities are fundamentally acquired during the mid-training phase and subsequently elicited during Supervised Fine-Tuning (SFT), there remains a critical deficit in metrics capable of guiding mid-training effectively. Standard metrics such as Perplexity (PPL) are compromised by the "Long-Context Tax" and exhibit weak correlation with downstream SWE performance. In this paper, we bridge this gap by first introducing a rigorous data filtering strategy. Crucially, we propose the Entropy Compression Hypothesis, redefining intelligence not by scalar Top-1 compression, but by the capacity to structure uncertainty into Entropy-Compressed States of low orders ("reasonable hesitation"). Grounded in this fine-grained entropy analysis, we formulate a novel metric, HE-SNR (High-Entropy Signal-to-Noise Ratio). We validate our approach on models with up to 560B parameters across different context windows (32K/128K). This work provides both the theoretical foundation and practical tools for optimizing the latent potential of LLMs in complex engineering domains.
Read the full abstract

All authors Yueyang Wang, Jiawei Fu, Baolong Bi, Xili Wang, Xiaoqing Liu

Computation and Language

Automated Benchmark Generation from Domain Guidelines Informed by Bloom's Taxonomy

Si Chen, Le Huy Khiem, Annalisa Szymanski et al. (+3 more)

Open-ended question answering (QA) evaluates a model's ability to perform contextualized reasoning beyond factual recall. This challenge is especially acute in practice-based domains, where knowledge is procedural and grounded in professional judgment, while most existing LLM benchmarks depend on pre-existing human exam datasets that are often unavailable in such settings. We introduce a framework for automated benchmark generation from expert-authored guidelines informed by Bloom's Taxonomy. It converts expert practices into implicit violation-based scenarios and expands them into auto-graded multiple-choice questions (MCQs) and multi-turn dialogues across four cognitive levels, enabling deterministic, reproducible, and scalable evaluation. Applied to three applied domains: teaching, dietetics, and caregiving, we find differences between model and human-like reasoning: LLMs sometimes perform relatively better on higher-order reasoning (Analyze) but fail more frequently on lower-level items (Remember). We produce large-scale, psychometrically informed benchmarks that surface these non-intuitive model behaviors and enable evaluation of contextualized reasoning in real-world settings.
Read the full abstract

All authors Si Chen, Le Huy Khiem, Annalisa Szymanski, Ronald Metoyer, Ting Hua, Nitesh V. Chawla

Computation and Language

Unit-Based Agent for Semi-Cascaded Full-Duplex Dialogue Systems

Haoyuan Yu, Yuxuan Chen and Minjie Cai

Full-duplex voice interaction is crucial for natural human computer interaction. We present a framework that decomposes complex dialogue into minimal conversational units, enabling the system to process each unit independently and predict when to transit to the next. This framework is instantiated as a semi-cascaded full-duplex dialogue system built around a multimodal large language model, supported by auxiliary modules such as voice activity detection (VAD) and text-to-speech (TTS) synthesis. The resulting system operates in a train-free, plug-and-play manner. Experiments on the HumDial dataset demonstrate the effectiveness of our framework, which ranks second among all teams on the test set of the Human-like Spoken Dialogue Systems Challenge (Track 2: Full-Duplex Interaction). Code is available at the GitHub repository https://github.com/yu-haoyuan/fd-badcat.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

Scaling Medical Reasoning Verification via Tool-Integrated Reinforcement Learning

Hang Zhang, Ruheng Wang, Yuelyu Ji et al. (+7 more)

Large language models have achieved strong performance on medical reasoning benchmarks, yet their deployment in clinical settings demands rigorous verification to ensure factual accuracy. While reward models offer a scalable approach for reasoning trace verification, existing methods face two limitations: they produce only scalar reward values without explicit justification, and they rely on single-pass retrieval that precludes adaptive knowledge access as verification unfolds. We introduce $\method$, an agentic framework that addresses these limitations by training medical reasoning verifiers to iteratively query external medical corpora during evaluation. Our approach combines tool-augmented verification with an iterative reinforcement learning paradigm that requires only trace-level supervision, alongside an adaptive curriculum mechanism that dynamically adjusts training data distribution. Across four medical reasoning benchmarks, $\method$ achieves substantial gains over existing methods, improving MedQA accuracy by 23.5% and MedXpertQA by 32.0% relative to the base generator in particular. Crucially, $\method$ demonstrates an $\mathbf{8\times}$ reduction in sampling budget requirement compared to prior reward model baselines. These findings establish that grounding verification in dynamically retrieved evidence offers a principled path toward more reliable medical reasoning systems.
Read the full abstract

All authors Hang Zhang, Ruheng Wang, Yuelyu Ji, Mingu Kwak, Xizhi Wu, Chenyu Li, Li Zhang, Wenqi Shi, Yifan Peng, Yanshan Wang

Computation and Language · Cross-listed from cs.LG

Spark: Strategic Policy-Aware Exploration via Dynamic Branching for Long-Horizon Agentic Learning

Jinyang Wu, Shuo Yang, Changpeng Yang et al. (+4 more)

Reinforcement learning has empowered large language models to act as intelligent agents, yet training them for long-horizon tasks remains challenging due to the scarcity of high-quality trajectories, especially under limited resources. Existing methods typically scale up rollout sizes and indiscriminately allocate computational resources among intermediate steps. Such attempts inherently waste substantial computation budget on trivial steps while failing to guarantee sample quality. To address this, we propose Spark (Strategic Policy-Aware exploRation via Key-state dynamic branching), a novel framework that selectively branches at critical decision states for resource-efficient exploration. Our key insight is to activate adaptive branching exploration at critical decision points to probe promising trajectories, thereby achieving precise resource allocation that prioritizes sampling quality over blind coverage. This design leverages the agent's intrinsic decision-making signals to reduce dependence on human priors, enabling the agent to autonomously expand exploration and achieve stronger generalization. Experiments across diverse tasks (e.g., embodied planning), demonstrate that Spark achieves superior success rates with significantly fewer training samples, exhibiting robust generalization even in unseen scenarios. Our code and checkpoints are available at https://github.com/jinyangwu/SPARK.
Read the full abstract

All authors Jinyang Wu, Shuo Yang, Changpeng Yang, Yuhao Shen, Shuai Zhang, Zhengqi Wen, Jianhua Tao

Computation and Language

Improving X-Codec-2.0 for Multi-Lingual Speech: 25 Hz Latent Rate and 24 kHz Sampling

Husein Zolkepli

X-Codec-2.0 has shown strong performance in neural audio compression and multilingual speech modeling, operating at a 50 Hz latent rate and a 16 kHz sampling rate using frozen HuBERT features. While effective, this configuration limits temporal efficiency and audio fidelity. In this work, we explore a simple and effective modification by introducing additional pooling and increasing the decoder hop size. This reduces the latent rate from 50 Hz to 25 Hz and simultaneously raises the output sampling rate from 16 kHz to 24 kHz, improving efficiency and perceptual quality without altering the core architecture. Evaluated on the multilingual Common Voice 17 test set, the proposed configuration achieves a 0.29 MOS improvement over the original X-Codec-2.0 baseline based on UTMOSv2, and attains the best reported performance among all codecs operating at 25 Hz. The source code, checkpoints, and generation comparisons are released at \href{https://huggingface.co/Scicom-intl/xcodec2-25TPS-24k}{https://huggingface.co/Scicom-intl/xcodec2-25TPS-24k}.
Read the full abstract