Peer Revenue Play credits. Real papers.

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Saturday 24 January

Computation and Language

Oops, Wait: Discourse Tokens Matter in Reasoning Model

Jaehui Hwang, Byeongho Heo, Sangdoo Yun et al. (+1 more)

Recent studies suggest that even data-efficient training with ($\simeq$1K) reasoning trajectories can induce non-trivial reasoning capabilities in large language models through post-training. Such training corpora often contain iconic tokens such as "wait", "so", and "alternatively", which frequently appear in reasoning trajectories and may play a role in this process. This paper focuses on characterizing observable token-level patterns in post-training and a case study of how data-efficient supervised fine-tuning (SFT) differs from, and falls short of, large-scale post-training. To this end, we first identify tokens that correlate with correct answers along reasoning trajectories across models and training setups. We then focus on the distribution and (functional) roles of the "wait" token to primarily study the model trained in a data-efficient manner compared with the counterpart. Our study finds that discourse tokens are associated with correctness and a reasoning accuracy jump, even in data-efficient SFT. This suggests data-efficient SFT can partially reproduce discourse-token patterns to mimic meaningful reasoning behavior, but the patterns are less aligned with high-confidence answer transitions than those from large-scale post-training.
Read the full abstract

All authors Jaehui Hwang, Byeongho Heo, Sangdoo Yun, Dongyoon Han

Computation and Language

CLM-Bench: Benchmarking and Analyzing Cross-lingual Misalignment of LLMs in Knowledge Editing

Yucheng Hu, Wei Zhou and Juesi Xiao

Knowledge Editing (KE) has emerged as a promising paradigm for updating facts in Large Language Models (LLMs) without retraining. However, progress in Multilingual Knowledge Editing (MKE) is currently hindered by biased evaluation frameworks. We observe that existing MKE benchmarks are typically constructed by mechanically translating English-centric datasets into target languages (e.g., English-to-Chinese). This approach introduces translation artifacts and neglects culturally specific entities native to the target language, failing to reflect the true knowledge distribution of LLMs. To address this, we propose CLM-Bench, a culture-aware benchmark constructed using a native Chinese-first methodology. We curate 1,010 high-quality CounterFact pairs rooted in Chinese cultural contexts and align them with English counterparts. Using CLM-Bench, we conduct extensive experiments on representative LLMs (e.g., Llama-3, Qwen2) and reveal a significant Cross-lingual Misalignment: edits in one language function independently and fail to propagate to the other. We further provide a geometric explanation via layer-wise representation analysis, demonstrating that edit vectors for Chinese and English are nearly orthogonal – residing in disjoint subspaces – while mixed-lingual editing exhibits linear additivity of these vectors. Our findings challenge the effectiveness of current methods in cross-lingual transfer and underscore the importance of culturally native benchmarks.
Read the full abstract

Computation and Language

Generation-Step-Aware Framework for Cross-Modal Representation and Control in Multilingual Speech-Text Models

Toshiki Nakai, Varsha Suresh and Vera Demberg

Multilingual speech-text models rely on cross-modal language alignment to transfer knowledge between speech and text, but it remains unclear whether this reflects shared computation for the same language or modality-specific processing. We introduce a generation-step-aware framework for evaluating cross-modal computation that (i) identifies language-selective neurons for each modality at different decoding steps, (ii) decomposes them into language-representation and language-control roles, and (iii) enables cross-modal comparison via overlap measures and causal intervention. Applying our framework to SeamlessM4T, and additionally evaluating its generality on the decoder-only model Qwen2-Audio, we find that cross-modal language-representation alignment is highly model dependent. SeamlessM4T exhibits pronounced generation-step-dependent specialization, where only 5-7% of language-representation neurons are shared across modalities and overlap shifts from same-language to typologically related languages during autoregressive generation, whereas Qwen2-Audio maintains substantially larger cross-modal sharing (45-47%) together with stable, strongly language-specific alignment across decoding steps. In contrast, language-control neurons identified at later decoding steps exhibit progressively stronger cross-modal transfer from speech to text in SeamlessM4T. Together, these results demonstrate that generation-step-aware analysis can reveal both shared and model-specific patterns of cross-modal computation.
Read the full abstract

Computation and Language

WarrantScore: Modeling Warrants between Claims and Evidence for Substantiation Evaluation in Peer Reviews

Kiyotada Mori, Shohei Tanaka, Tosho Hirasawa et al. (+3 more)

The scientific peer-review process is facing a shortage of human resources due to the rapid growth in the number of submitted papers. The use of language models to reduce the human cost of peer review has been actively explored as a potential solution to this challenge. A method has been proposed to evaluate the level of substantiation in scientific reviews in a manner that is interpretable by humans. This method extracts the core components of an argument, claims and evidence, and assesses the level of substantiation based on the proportion of claims supported by evidence. The level of substantiation refers to the extent to which claims are based on objective facts. However, when assessing the level of substantiation, simply detecting the presence or absence of supporting evidence for a claim is insufficient; it is also necessary to accurately assess the logical inference between a claim and its evidence. We propose a new evaluation metric for scientific review comments that assesses the logical inference between claims and evidence. Experimental results show that the proposed method achieves a higher correlation with human scores than conventional methods, indicating its potential to better support the efficiency of the peer-review process.
Read the full abstract

All authors Kiyotada Mori, Shohei Tanaka, Tosho Hirasawa, Tadashi Kozuno, Koichiro Yoshino, Yoshitaka Ushiku

Computation and Language

Elastic Attention: Test-time Adaptive Sparsity Ratios for Efficient Transformers

Zecheng Tang, Quantong Qiu, Yi Yang et al. (+6 more)

The quadratic complexity of standard attention mechanisms poses a significant scalability bottleneck for large language models (LLMs) in long-context scenarios. While hybrid attention strategies that combine sparse and full attention within a single model offer a viable solution, they typically employ static computation ratios (i.e., fixed proportions of sparse versus full attention) and fail to adapt to the varying sparsity sensitivities of downstream tasks during inference. To address this issue, we propose Elastic Attention, which allows the model to dynamically adjust its overall sparsity based on the input. This is achieved by integrating a lightweight Attention Router into the existing pretrained model, which dynamically assigns each attention head to different computation modes. Within only 12 hours of training on 8xA800 GPUs, our method enables models to achieve both strong performance and efficient inference. Experiments across three long-context benchmarks on widely-used LLMs demonstrate the superiority of our method.
Read the full abstract

All authors Zecheng Tang, Quantong Qiu, Yi Yang, Zhiyi Hong, Haiya Xiang, Kebin Liu, Qingqing Dang, Juntao Li, Min Zhang

Computation and Language

Parameter Efficient Fine Tuning Llama 3.1 for Answering Arabic Legal Questions: A Case Study on Jordanian Laws

Mohammed Fasha, Bassam Hammo, Bilal Sowan et al. (+2 more)

This study uses Jordanian law as a case study to explore the fine-tuning of the Llama-3.1 large language model for Arabic question-answering. Two versions of the model - Llama-3.1-8B-bnb-4bit and Llama-3.1-8B-Instruct-bnb-4bit - were fine-tuned using parameter-efficient fine-tuning (PEFT) with LoRA adapters and 4-bit quantized models, leveraging the Unsloth framework for accelerated and resource-efficient training. A custom dataset of 6000 legal question-answer pairs was curated from Jordanian laws and formatted into structured prompts. Performance was evaluated using the BLEU and the ROUGE metrics to compare the fine-tuned models to their respective base versions. Results demonstrated improved legal reasoning and accuracy while achieving resource efficiency through quantization and optimized fine-tuning strategies. This work underscores the potential of adapting large language models for Arabic legal domains and highlights effective techniques for fine-tuning domain-specific tasks.
Read the full abstract

All authors Mohammed Fasha, Bassam Hammo, Bilal Sowan, Husam Barham, Esam Nsour

Computation and Language

Do readers prefer AI-generated Italian short stories?

Michael Farrell

This study investigates whether readers prefer AI-generated short stories in Italian over one written by a renowned Italian author. In a blind setup, 20 participants read and evaluated three stories, two created with ChatGPT-4o and one by Alberto Moravia, without being informed of their origin. To explore potential influencing factors, reading habits and demographic data, comprising age, gender, education and first language, were also collected. The results showed that the AI-written texts received slightly higher average ratings and were more frequently preferred, although differences were modest. No statistically significant associations were found between text preference and demographic or reading-habit variables. These findings challenge assumptions about reader preference for human-authored fiction and raise questions about the necessity of synthetic-text editing in literary contexts.
Read the full abstract

Computation and Language

The Shadow Self: Intrinsic Value Misalignment in Large Language Model Agents

Chen Chen, Kim Young Il, Yuan Yang et al. (+7 more)

Large language model (LLM) agents with extended autonomy unlock new capabilities, but also introduce heightened challenges for LLM safety. In particular, an LLM agent may pursue objectives that deviate from human values and ethical norms, a risk known as value misalignment. Existing evaluations primarily focus on responses to explicit harmful input or robustness against system failure, while value misalignment in realistic, fully benign, and agentic settings remains largely underexplored. To fill this gap, we first formalize the Loss-of-Control risk and identify the previously underexamined Intrinsic Value Misalignment (Intrinsic VM). We then introduce IMPRESS (Intrinsic Value Misalignment Probes in REalistic Scenario Set), a scenario-driven framework for systematically assessing this risk. Following our framework, we construct benchmarks composed of realistic, fully benign, and contextualized scenarios, using a multi-stage LLM generation pipeline with rigorous quality control. We evaluate Intrinsic VM on 21 state-of-the-art LLM agents and find that it is a common and broadly observed safety risk across models. Moreover, the misalignment rates vary by motives, risk types, model scales, and architectures. While decoding strategies and hyperparameters exhibit only marginal influence, contextualization and framing mechanisms significantly shape misalignment behaviors. Finally, we conduct human verification to validate our automated judgments and assess existing mitigation strategies, such as safety prompting and guardrails, which show instability or limited effectiveness. We further demonstrate key use cases of IMPRESS across the AI Ecosystem. Our code and benchmark will be publicly released upon acceptance.
Read the full abstract

All authors Chen Chen, Kim Young Il, Yuan Yang, Wenhao Su, Yilin Zhang, Xueluan Gong, Qian Wang, Yongsen Zheng, Ziyao Liu, Kwok-Yan Lam

Computation and Language

Meta-Judging with Large Language Models: Concepts, Methods, and Challenges

Hugo Silva, Mateus Mendes and Hugo Gonçalo Oliveira

Large language models (LLMs) are evolving fast and are now frequently used as evaluators, in a process typically referred to as LLM-as-a-Judge, which provides quality assessments of model outputs. However, recent research points out significant vulnerabilities in such evaluation, including sensitivity to prompts, systematic biases, verbosity effects, and unreliable or hallucinated rationales. These limitations motivated the development of a more robust paradigm, dubbed LLM-as-a-Meta-Judge. This survey reviews recent advances in meta-judging and organizes the literature, by introducing a framework along six key perspectives: (i) Conceptual Foundations, (ii) Mechanisms of Meta-Judging, (iii) Alignment Training Methods, (iv) Evaluation, (v) Limitations and Failure Modes, and (vi) Future Directions. By analyzing the limitations of LLM-as-a-Judge and summarizing recent advances in meta-judging by LLMs, we argue that LLM-as-a-Meta-Judge offers a promising direction for more stable and trustworthy automated evaluation, while highlighting remaining challenges related to cost, prompt sensitivity, and shared model biases, which must be addressed to advance the next generation of LLM evaluation methodologies.
Read the full abstract

Computation and Language · Cross-listed from cs.NI

Structure-Aware NL-to-SQL for SFC Provisioning via AST-Masking Empowered Language Models

Xinyu Zhu, Parisa Fard Moshiri, Poonam Lohan et al. (+2 more)

Effective Service Function Chain (SFC) provisioning requires precise orchestration in dynamic and latency-sensitive networks. Reinforcement Learning (RL) improves adaptability but often ignores structured domain knowledge, which limits generalization and interpretability. Large Language Models (LLMs) address this gap by translating natural language (NL) specifications into executable Structured Query Language (SQL) commands for specification-driven SFC management. Conventional fine-tuning, however, can cause syntactic inconsistencies and produce inefficient queries. To overcome this, we introduce Abstract Syntax Tree (AST)-Masking, a structure-aware fine-tuning method that uses SQL ASTs to assign weights to key components and enforce syntax-aware learning without adding inference overhead. Experiments show that AST-Masking significantly improves SQL generation accuracy across multiple language models. FLAN-T5 reaches an Execution Accuracy (EA) of 99.6%, while Gemma achieves the largest absolute gain from 7.5% to 72.0%. These results confirm the effectiveness of structure-aware fine-tuning in ensuring syntactically correct and efficient SQL generation for interpretable SFC orchestration.
Read the full abstract

All authors Xinyu Zhu, Parisa Fard Moshiri, Poonam Lohan, Burak Kantarci, Emil Janulewicz

Computation and Language

Mind the Ambiguity: Aleatoric Uncertainty Quantification in LLMs for Safe Medical Question Answering

Yaokun Liu, Yifan Liu, Phoebe Mbuvi et al. (+4 more)

The deployment of Large Language Models in Medical Question Answering is severely hampered by ambiguous user queries, a significant safety risk that demonstrably reduces answer accuracy in high-stakes healthcare settings. In this paper, we formalize this challenge by linking input ambiguity to aleatoric uncertainty (AU), which is the irreducible uncertainty arising from underspecified input. To facilitate research in this direction, we construct CV-MedBench, the first benchmark designed for studying input ambiguity in Medical QA. Using this benchmark, we analyze AU from a representation engineering perspective, revealing that AU is linearly encoded in LLM's internal activation patterns. Leveraging this insight, we introduce a novel AU-guided "Clarify-Before-Answer" framework, which incorporates AU-Probe - a lightweight module that detects input ambiguity directly from hidden states. Unlike existing uncertainty estimation methods, AU-Probe requires neither LLM fine-tuning nor multiple forward passes, enabling an efficient mechanism to proactively request user clarification and significantly enhance safety. Extensive experiments across four open LLMs demonstrate the effectiveness of our QA framework, with an average accuracy improvement of 9.48% over baselines. Our framework provides an efficient and robust solution for safe Medical QA, strengthening the reliability of health-related applications. The code is available at https://github.com/yaokunliu/AU-Med.git, and the CV-MedBench dataset is released on Hugging Face at https://huggingface.co/datasets/yaokunl/CV-MedBench.
Read the full abstract

All authors Yaokun Liu, Yifan Liu, Phoebe Mbuvi, Zelin Li, Ruichen Yao, Gawon Lim, Dong Wang

Computation and Language

PingPong: A Natural Benchmark for Multi-Turn Code-Switching Dialogues

Mohammad Rifqi Farhansyah, Hanif Muhammad Zhafran, Farid Adilazuarda et al. (+6 more)

Code-switching is a widespread practice among the world's multilingual majority, yet few benchmarks accurately reflect its complexity in everyday communication. We present PingPong, a benchmark for natural multi-party code-switching dialogues covering five language-combination variations, some of which are trilingual. Our dataset consists of human-authored conversations among 2 to 4 participants covering authentic, multi-threaded structures where replies frequently reference much earlier points in the dialogue. We demonstrate that our data is significantly more natural and structurally diverse than machine-generated alternatives, offering greater variation in message length, speaker dominance, and reply distance. Based on these dialogues, we define three downstream tasks: Question Answering, Dialogue Summarization, and Topic Classification. Evaluations of several state-of-the-art language models on PingPong reveal that performance remains limited on code-switched inputs, underscoring the urgent need for more robust NLP systems capable of addressing the intricacies of real-world multilingual discourse.
Read the full abstract

All authors Mohammad Rifqi Farhansyah, Hanif Muhammad Zhafran, Farid Adilazuarda, Shamsuddeen Hassan Muhammad, Maryam Ibrahim Mukhtar, Nedjma Ousidhoum, Genta Indra Winata, Ayu Purwarianti, Alham Fikri Aji

Computation and Language · Cross-listed from cs.SD

Window Size Versus Accuracy Experiments in Voice Activity Detectors

Max McKinnon, Samir Khaki, Chandan KA Reddy et al. (+1 more)

Voice activity detection (VAD) plays a vital role in enabling applications such as speech recognition. We analyze the impact of window size on the accuracy of three VAD algorithms: Silero, WebRTC, and Root Mean Square (RMS) across a set of diverse real-world digital audio streams. We additionally explore the use of hysteresis on top of each VAD output. Our results offer practical references for optimizing VAD systems. Silero significantly outperforms WebRTC and RMS, and hysteresis provides a benefit for WebRTC.
Read the full abstract

All authors Max McKinnon, Samir Khaki, Chandan KA Reddy, William Huang

Friday 23 January

Computation and Language

Talking to Yourself: Defying Forgetting in Large Language Models

Yutao Sun, Mingshuai Chen, Tiancheng Zhao et al. (+5 more)

Catastrophic forgetting remains a major challenge when fine-tuning large language models (LLMs) on narrow, task-specific data, often degrading their general knowledge and reasoning abilities. We propose SA-SFT, a lightweight self-augmentation routine in which an LLM generates self-dialogues prior to fine-tuning, and the resulting self-authored data are mixed with task data without modifying optimization or training schedules. Despite requiring no external data or additional tuning, SA-SFT consistently mitigates catastrophic forgetting while improving in-domain performance. Across 50 evaluation scenarios, it maintains performance comparable to the original model and achieves the best results in 40 cases, outperforming common baselines such as layer freezing and external data mixing. Guided by these empirical findings, we further present a theoretical analysis suggesting that forgetting can partly stem from style-induced parameter drift, and that self-alignment through self-generated data provides an effective means to counteract this effect. Overall, our results indicate that self-augmentation offers a simple and effective mechanism for robust LLM adaptation without incurring catastrophic forgetting.
Read the full abstract

All authors Yutao Sun, Mingshuai Chen, Tiancheng Zhao, Phillip Miao, Zilun Zhang, Haozhan Shen, Ruizhe Zhu, Jianwei Yin

Computation and Language

Narrative Theory-Driven LLM Methods for Automatic Story Generation and Understanding: A Survey

David Y. Liu, Aditya Joshi and Paul Dawson

Applications of narrative theories using large language models (LLMs) deliver promising methods in automatic story generation and understanding tasks. Our survey examines how natural language processing (NLP) research uses LLM methods to engage with diverse concepts from narrative studies. We use established distinctions from narratology to categorise ongoing efforts and discover the following: \redtext{(a) narrative texts come from diverse sources beyond just literature, (b) theoretical synthesis and validation are potential outcomes, (c) generation tasks lag behind understanding in several ways: theoretical application, post-training methods, exploring non-fiction narratives and addressing narrative levels beyond fabula and discourse.} For future directions, instead of the pursuit of a single, generalised benchmark for `narrative quality', we believe that progress can benefit from efforts that focus on the following: defining and improving theory-based metrics for individual narrative attributes; continue conducting large-scale, theory-driven literary/social/cultural analysis; generating narratives in situated contexts; and continuing experiments where outputs can be used to validate or refine narrative theories. This work provides a contextual foundation for more systematic and theoretically informed narrative research in NLP by providing an overview to ongoing research efforts and the broader narrative studies landscape.
Read the full abstract

Computation and Language

Large Language Models for Assisting American College Applications

Zhengliang Liu, Weihang You, Peng Shu et al. (+14 more)

American college applications require students to navigate fragmented admissions policies, repetitive and conditional forms, and ambiguous questions that often demand cross-referencing multiple sources. We present EZCollegeApp, a large language model (LLM)-powered system that assists high-school students by structuring application forms, grounding suggested answers in authoritative admissions documents, and maintaining full human control over final responses. The system introduces a mapping-first paradigm that separates form understanding from answer generation, enabling consistent reasoning across heterogeneous application portals. EZCollegeApp integrates document ingestion from official admissions websites, retrieval-augmented question answering, and a human-in-the-loop chatbot interface that presents suggestions alongside application fields without automated submission. We describe the system architecture, data pipeline, internal representations, security and privacy measures, and evaluation through automated testing and human quality assessment. Our source code is released on GitHub (https://github.com/ezcollegeapp-public/ezcollegeapp-public) to facilitate the broader impact of this work.
Read the full abstract

All authors Zhengliang Liu, Weihang You, Peng Shu, Junhao Chen, Yi Pan, Hanqi Jiang, Yiwei Li, Zhaojun Ding, Chao Cao, Xinliang Li, Yifan Zhou, Ruidong Zhang, Shaochen Xu, Wei Ruan, Huaqin Zhao, Dajiang Zhu, Tianming Liu

Computation and Language

IntelliAsk: Learning to Ask High-Quality Research Questions via RLVR

Karun Sharma, Vidushee Vats, Shengzhi Li et al. (+3 more)

Peer review relies on substantive, evidence-based questions, yet current LLMs generate surface-level queries that perform worse than human reviewer questions in expert evaluation. To address this gap, we curate a high-quality dataset of reviewer questions from OpenReview and conduct a human preference study where expert annotators evaluate question-paper pairs across three dimensions: effort, evidence, and grounding. From these annotations, we train IntelliReward, a reward model built from a frozen autoregressive LLM with trainable multi-head transformers. Validated against expert judgments, IntelliReward predicts reviewer-question quality better than API-based SFT baselines and provides scalable evaluation. We apply Decoupled Clip and Dynamic Sampling Policy Optimization (DAPO) with IntelliReward to train IntelliAsk, a question-generation model aligned with human standards of effortful, evidence-based critique. Human evaluations show IntelliAsk generates more grounded, substantive and effortful questions than strong baselines and reduces reliance on first-page content. We also find improvements on reasoning and writing benchmarks, suggesting reviewer-question quality correlates with broader capabilities. Compared to Qwen3-32B, IntelliAsk improves MuSR (68.3 vs 64.7 Acc) and WritingBench (8.31 vs 8.07). We release our code, filtered review dataset, expert annotations, IntelliAsk and IntelliReward to support automatic evaluation of grounding, effort, and evidence in LLM-generated review questions.
Read the full abstract

All authors Karun Sharma, Vidushee Vats, Shengzhi Li, Yuxiang Wang, Zhongtian Sun, Prayag Tiwari

Computation and Language

Can LLMs Assess Personality? Validating Conversational AI for Trait Profiling

Andrius Matšenas, Anet Lello, Tõnis Lees et al. (+2 more)

This study validates Large Language Models (LLMs) as a dynamic alternative to questionnaire-based personality assessment. Using a within-subjects experiment (N=33), we compared Big Five personality scores derived from guided LLM conversations against the gold-standard IPIP-50 questionnaire, while also measuring user-perceived accuracy. Results indicate moderate convergent validity (r=0.38-0.58), with Conscientiousness, Openness, and Neuroticism scores statistically equivalent between methods. Agreeableness and Extraversion showed significant differences, suggesting trait-specific calibration is needed. Notably, participants rated LLM-generated profiles as equally accurate as traditional questionnaire results. These findings suggest conversational AI offers a promising new approach to traditional psychometrics.
Read the full abstract

All authors Andrius Matšenas, Anet Lello, Tõnis Lees, Hans Peep, Kim Lilii Tamm

Computation and Language

Do Personality Traits Interfere? Geometric Limitations of Steering in Large Language Models

Pranav Bhandari, Usman Naseem and Mehwish Nasim

Personality steering in large language models (LLMs) commonly relies on injecting trait-specific steering vectors, implicitly assuming that personality traits can be controlled independently. In this work, we examine whether this assumption holds by analysing the geometric relationships between Big Five personality steering directions. We study steering vectors extracted from two model families (LLaMA-3-8B and Mistral-8B) and apply a range of geometric conditioning schemes, from unconstrained directions to soft and hard orthonormalisation. Our results show that personality steering directions exhibit substantial geometric dependence: steering one trait consistently induces changes in others, even when linear overlap is explicitly removed. While hard orthonormalisation enforces geometric independence, it does not eliminate cross-trait behavioural effects and can reduce steering strength. These findings suggest that personality traits in LLMs occupy a slightly coupled subspace, limiting fully independent trait control.
Read the full abstract

Computation and Language

Gated Tree Cross-Attention for Checkpoint-Compatible Syntax Injection in Decoder-Only LLMs

Xinyu Gao, Shaonan Wang and Nai Ding

Decoder-only large language models achieve strong broad performance but are brittle to minor grammatical perturbations, undermining reliability for downstream reasoning. However, directly injecting explicit syntactic structure into an existing checkpoint can interfere with its pretrained competence. We introduce a checkpoint-compatible gated tree cross-attention (GTCA) branch that reads precomputed constituency chunk memory while leaving backbone architecture unchanged. Our design uses a token update mask and staged training to control the scope and timing of structural updates. Across benchmarks and Transformer backbones, GTCA strengthens syntactic robustness beyond continued-training baselines without compromising Multiple-Choice QA performance or commonsense reasoning, providing a practical checkpoint-compatible route to more syntax-robust decoder-only LLMs.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

Scaling the Scaling Logic: Agentic Meta-Synthesis of Logic Reasoning

Bowen Liu, Zhi Wu, Runquan Xie et al. (+2 more)

Reinforcement Learning from Verifiable Rewards (RLVR) is bottlenecked by data: existing synthesis pipelines rely on expert-written code or fixed templates, confining growth to instance-level perturbations. We shift the evolvable unit from problem instances to task-family specifications. SSLogic is an agentic meta-synthesis framework in which LLM agents iteratively author and refine executable Generator-Validator pairs inside a closed Generate-Validate-Refine loop, producing families with new rules and difficulty gradients rather than parameter variations of old ones. A Multi-Gate Validation Protocol – multi-strategy consensus plus Adversarial Blind Review, where independent agents solve each instance by writing and executing code – filters ill-posed tasks before they enter training. Starting from 400 seed families, two evolution rounds yield 953 families and 21,389 verifiable instances. Three converging comparisons (step-matched, token-matched, and size-controlled on external Enigmata data) consistently show higher training utility of evolved data, with gains of SynLogic +5.2, AIME25 +3.0, and BBH +5.5 on Enigmata. Fine-grained KORBench evaluation reveals selective improvements in logic (+13.2%) and operation (+9.6%), linking structural evolution to downstream gains. Code: https://github.com/AdAstraAbyssoque/Scaling-the-Scaling-Logic
Read the full abstract

All authors Bowen Liu, Zhi Wu, Runquan Xie, Zhanhui Kang, Jia Li

Computation and Language · Cross-listed from cs.AI

CreditAudit: 2$^\text{nd}$ Dimension for LLM Evaluation and Selection

Yiliang Song, Hongjun An, Jiangong Xiao et al. (+3 more)

Leaderboard scores on public benchmarks have been steadily rising and converging, with many frontier language models now separated by only marginal differences. However, these scores often fail to match users' day to day experience, because system prompts, output protocols, and interaction modes evolve under routine iteration, and in agentic multi step pipelines small protocol shifts can trigger disproportionate failures, leaving practitioners uncertain about which model to deploy. We propose CreditAudit, a deployment oriented credit audit framework that evaluates models under a family of semantically aligned and non adversarial system prompt templates across multiple benchmarks, reporting mean ability as average performance across scenarios and scenario induced fluctuation sigma as a stability risk signal, and further mapping volatility into interpretable credit grades from AAA to BBB via cross model quantiles with diagnostics that mitigate template difficulty drift. Controlled experiments on GPQA, TruthfulQA, and MMLU Pro show that models with similar mean ability can exhibit substantially different fluctuation, and stability risk can overturn prioritization decisions in agentic or high failure cost regimes. By providing a 2D and grade based language for regime specific selection, CreditAudit supports tiered deployment and more disciplined allocation of testing and monitoring effort, enabling more objective and trustworthy model evaluation for real world use.
Read the full abstract

All authors Yiliang Song, Hongjun An, Jiangong Xiao, Haofei Zhao, Jiawei Shao, Xuelong Li

Computation and Language · Cross-listed from cs.CY

Beyond Translation: Cross-Cultural Meme Transcreation with Vision-Language Models

Yuming Zhao, Peiyi Zhang and Oana Ignat

Memes are a pervasive form of online communication, yet their cultural specificity poses significant challenges for cross-cultural adaptation. We study cross-cultural meme transcreation, a multimodal generation task that aims to preserve communicative intent and humor while adapting culture-specific references. We propose a hybrid transcreation framework based on vision-language models and introduce a large-scale bidirectional dataset of Chinese and US memes. Using both human judgments and automated evaluation, we analyze 6,315 meme pairs and assess transcreation quality across cultural directions. Our results show that current vision-language models can perform cross-cultural meme transcreation to a limited extent, but exhibit clear directional asymmetries: US-Chinese transcreation consistently achieves higher quality than Chinese-US. We further identify which aspects of humor and visual-textual design transfer across cultures and which remain challenging, and propose an evaluation framework for assessing cross-cultural multimodal generation. Our code and dataset are publicly available at https://github.com/AIM-SCU/MemeXGen.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

Interpreting and Controlling Model Behavior via Constitutions for Atomic Concept Edits

Neha Kalibhat, Zi Wang, Prasoon Bajpai et al. (+4 more)

We introduce a black-box interpretability framework that learns a verifiable constitution: a natural language summary of how changes to a prompt affect a model's specific behavior, such as its alignment, correctness, or adherence to constraints. Our method leverages atomic concept edits (ACEs), which are targeted operations that add, remove, or replace an interpretable concept in the input prompt. By systematically applying ACEs and observing the resulting effects on model behavior across various tasks, our framework learns a causal mapping from edits to predictable outcomes. This learned constitution provides deep, generalizable insights into the model. Empirically, we validate our approach across diverse tasks, including mathematical reasoning and text-to-image alignment, for controlling and understanding model behavior. We found that for text-to-image generation, GPT-Image tends to focus on grammatical adherence, while Imagen 4 prioritizes atmospheric coherence. In mathematical reasoning, distractor variables confuse GPT-5 but leave Gemini 2.5 models and o4-mini largely unaffected. Moreover, our results show that the learned constitutions are highly effective for controlling model behavior, achieving an average of 1.86 times boost in success rate over methods that do not use constitutions.
Read the full abstract

All authors Neha Kalibhat, Zi Wang, Prasoon Bajpai, Drew Proud, Wenjun Zeng, Been Kim, Mani Malek

Computation and Language

Benchmarking von ASR-Modellen im deutschen medizinischen Kontext: Eine Leistungsanalyse anhand von Anamnesegesprächen

Thomas Schuster, Julius Trögele, Nico Döring et al. (+3 more)

Automatic Speech Recognition (ASR) offers significant potential to reduce the workload of medical personnel, for example, through the automation of documentation tasks. While numerous benchmarks exist for the English language, specific evaluations for the German-speaking medical context are still lacking, particularly regarding the inclusion of dialects. In this article, we present a curated dataset of simulated doctor-patient conversations and evaluate a total of 29 different ASR models. The test field encompasses both open-weights models from the Whisper, Voxtral, and Wav2Vec2 families as well as commercial state-of-the-art APIs (AssemblyAI, Deepgram). For evaluation, we utilize three different metrics (WER, CER, BLEU) and provide an outlook on qualitative semantic analysis. The results demonstrate significant performance differences between the models: while the best systems already achieve very good Word Error Rates (WER) of partly below 3%, the error rates of other models, especially concerning medical terminology or dialect-influenced variations, are considerably higher.
Read the full abstract

All authors Thomas Schuster, Julius Trögele, Nico Döring, Robin Krüger, Matthieu Hoffmann, Holger Friedrich

Computation and Language

Frame-Guided Synthetic Claim Generation for Automatic Fact-Checking Using High-Volume Tabular Data

Jacob Devasier, Akshith Putta, Qing Wang et al. (+2 more)

Automated fact-checking benchmarks have largely ignored the challenge of verifying claims against real-world, high-volume structured data, instead focusing on small, curated tables. We introduce a new large-scale, multilingual dataset to address this critical gap. It contains 78,503 synthetic claims grounded in 434 complex OECD tables, which average over 500K rows each. We propose a novel, frame-guided methodology where algorithms programmatically select significant data points based on six semantic frames to generate realistic claims in English, Chinese, Spanish, and Hindi. Crucially, we demonstrate through knowledge-probing experiments that LLMs have not memorized these facts, forcing systems to perform genuine retrieval and reasoning rather than relying on parameterized knowledge. We provide a baseline SQL-generation system and show that our benchmark is highly challenging. Our analysis identifies evidence retrieval as the primary bottleneck, with models struggling to find the correct data in massive tables. This dataset provides a critical new resource for advancing research on this unsolved, real-world problem.
Read the full abstract

All authors Jacob Devasier, Akshith Putta, Qing Wang, Alankrit Moses, Chengkai Li

Computation and Language

CaseFacts: A Benchmark for Legal Fact-Checking and Precedent Retrieval

Akshith Reddy Putta, Jacob Devasier and Chengkai Li

Automated Fact-Checking has largely focused on verifying general knowledge against static corpora, overlooking high-stakes domains like law where truth is evolving and technically complex. We introduce CaseFacts, a benchmark for verifying colloquial legal claims against U.S. Supreme Court precedents. Unlike existing resources that map formal texts to formal texts, CaseFacts challenges systems to bridge the semantic gap between layperson assertions and technical jurisprudence while accounting for temporal validity. The dataset consists of 6,294 claims categorized as Supported, Refuted, or Overruled. We construct this benchmark using a multi-stage pipeline that leverages Large Language Models (LLMs) to synthesize claims from expert case summaries, employing a novel semantic similarity heuristic to efficiently identify and verify complex legal overrulings. Experiments with state-of-the-art LLMs reveal that the task remains challenging; notably, augmenting models with unrestricted web search degrades performance compared to closed-book baselines due to the retrieval of noisy, non-authoritative precedents. We release CaseFacts to spur research into legal fact verification systems.
Read the full abstract

Computation and Language

Retell, Reward, Repeat: Reinforcement Learning for Narrative Theory-Informed Story Retelling

David Y. Liu, Xanthe Muston, Dipankar Srirag et al. (+2 more)

Counterfactual storytelling exposes LLM shortcomings in causal reasoning, where few-shot prompting and SFT fail to produce logical and rational narratives. This case study introduces Retell, Reward, Repeat (RRR), a narrative theory-informed RL pipeline that teaches causal story structure. We extend an existing counterfactual stories dataset with human-annotated stages of narrative equilibrium, synthesising Structuralist Narratology with scalar narrativity. An aligned LLM-as-judge automates this framework to output reward signals, teaching the policy LLM to generate texts with causal narrative structure. Evaluations show that RRR-trained 7/8B LLMs outperform few-shot and SFT baselines in logic, rationality, and completeness, also aligning with a blind human preference subset. RRR offers a linguistically grounded, cost-effective post-training mechanism for LLM storytelling using a small, query-only dataset, showing the value of integrating linguistic theories into contemporary NLP.
Read the full abstract

All authors David Y. Liu, Xanthe Muston, Dipankar Srirag, Aditya Joshi, Sebastian Sequoiah-Grayson

Computation and Language

Beyond Outcome Verification: Verifiable Process Reward Models for Structured Reasoning

Massimiliano Pronesti, Anya Belz and Yufang Hou

Recent work on reinforcement learning with verifiable rewards (RLVR) has shown that large language models (LLMs) can be substantially improved using outcome-level verification signals, such as unit tests for code or exact-match checks for mathematics. In parallel, process supervision has long been explored as a way to shape the intermediate reasoning behaviour of LLMs, but existing approaches rely on neural judges to score chain-of-thought steps, leaving them vulnerable to opacity, bias, and reward hacking. To address this gap, we introduce Verifiable Process Reward Models (VPRMs), a reinforcement-learning framework in which intermediate reasoning steps are checked by deterministic, rule-based verifiers. We apply VPRMs to risk-of-bias assessment for medical evidence synthesis, a domain where guideline-defined criteria and rule-based decision paths enable programmatic verification of reasoning traces. Across multiple datasets, we find that VPRMs generate reasoning that adheres closely to domain rules and achieve substantially higher coherence between step-level decisions and final labels. Results show that VPRMs achieve up to 20% higher F1 than state-of-the-art models and 6.5% higher than verifiable outcome rewards, with substantial gains in evidence grounding and logical coherence.
Read the full abstract

Computation and Language

DF-RAG: Query-Aware Diversity for Retrieval-Augmented Generation

Saadat Hasan Khan, Spencer Hong, Jingyu Wu et al. (+4 more)

Retrieval-augmented generation (RAG) is a common technique for grounding language model outputs in domain-specific information. However, RAG is often challenged by reasoning-intensive question-answering (QA), since common retrieval methods like cosine similarity maximize relevance at the cost of introducing redundant content, which can reduce information recall. To address this, we introduce Diversity-Focused Retrieval-Augmented Generation (DF-RAG), which systematically incorporates diversity into the retrieval step to improve performance on complex, reasoning-intensive QA benchmarks. DF-RAG builds upon the Maximal Marginal Relevance framework to select information chunks that are both relevant to the query and maximally dissimilar from each other. A key innovation of DF-RAG is its ability to optimize the level of diversity for each query dynamically at test time without requiring any additional fine-tuning or prior information. We show that DF-RAG improves F1 performance on reasoning-intensive QA benchmarks by 4-10 percent over vanilla RAG using cosine similarity and also outperforms other established baselines. Furthermore, we estimate an Oracle ceiling of up to 18 percent absolute F1 gains over vanilla RAG, of which DF-RAG captures up to 91.3 percent.
Read the full abstract

All authors Saadat Hasan Khan, Spencer Hong, Jingyu Wu, Kevin Lybarger, Youbing Yin, Erin Babinsky, Daben Liu