Computation and Language
Enhancing Safety of Large Language Models via Embedding Space Separation
Xu Zhao, Xiting Wang and Weiran Shen
Safety
alignment does not fully prevent large
language models (LLMs) from producing harmful content under
adversarial attacks. We observe that existing attack methods typically move harmful
embeddings toward regions of harmless
embeddings. This motivates us to explore whether increasing the distance between harmful and harmless
embeddings can improve the safety of LLMs, since greater separation requires attacks to make larger
embedding shifts, which can destroy semantics. We propose
Embedding Space Separation (ES2), a fine-tuning method that increases the distance between harmful and harmless
embeddings. However, maximizing this distance alone can damage LLMs' capabilities. Thus, we use two regularization losses: an
embedding anchor loss keeps harmless
embeddings close in direction to those of a frozen base model and a KL-divergence loss keeps output distributions on harmless
prompts close to those of the base model. Our experimental results demonstrate that ES2 has excellent defensive performance against various attack methods. LLMs fine-tuned with ES2 retain general capabilities comparable to those of the baseline models, though some performance degradation is observed on certain tasks. Our results indicate that ES2 is an effective way to enhance the safety of large
language models, while maintaining comparable performance on general capabilities.
Read the full abstract ▾
Computation and Language
From Feature-Based Models to Generative AI: Validity Evidence for Constructed Response Scoring
Jodi M. Casabianca, Daniel F. McCaffrey, Matthew S. Johnson et al. (+2 more)
The rapid advancements in large
language models and generative artificial intelligence (AI) capabilities are making their broad application in the high-stakes testing context more likely. Use of generative AI in the scoring of constructed responses is particularly appealing because it reduces the effort required for handcrafting features in traditional AI scoring and might even outperform those methods. The purpose of this paper is to highlight the differences in the feature-based and generative AI applications in constructed response scoring systems and propose a set of best practices for the collection of validity evidence to support the use and interpretation of constructed response scores from scoring systems using generative AI. We compare the validity evidence needed in scoring systems using human ratings, feature-based natural
language processing AI scoring engines, and generative AI. The evidence needed in the generative AI context is more extensive than in the feature-based scoring context because of the lack of transparency and other concerns unique to generative AI such as consistency. Constructed response score data from a large corpus of independent argumentative essays written by 6-12th grade students demonstrate the collection of validity evidence for different types of scoring systems and highlight the numerous complexities and considerations when making a validity argument for these scores.
Read the full abstract ▾
All authors Jodi M. Casabianca, Daniel F. McCaffrey, Matthew S. Johnson, Naim Alper, Vladimir Zubenko
Computation and Language
Multilingual Hate Speech Detection and Counterspeech Generation: A Comprehensive Survey and Practical Guide
Zahra Safdari Fesaghandis and Suman Kalyan Maity
Combating online hate
speech in
multilingual settings requires approaches that go beyond English-centric models and capture the cultural and linguistic diversity of global online discourse. This paper presents a comprehensive survey and practical guide to
multilingual hate
speech detection and counterspeech generation, integrating recent advances in natural
language processing. We analyze why monolingual systems often fail in non-English and code-mixed contexts, missing implicit hate and culturally specific expressions. To address these challenges, we outline a structured three-phase framework - task design, data curation, and evaluation - drawing on state-of-the-art
datasets, models, and metrics. The survey consolidates progress in
multilingual resources and techniques while highlighting persistent obstacles, including data scarcity in low-resource
languages,
fairness and bias in system development, and the need for
multimodal solutions. By bridging technical progress with ethical and cultural considerations, we provide researchers, practitioners, and policymakers with scalable guidelines for building context-aware, inclusive systems. Our roadmap contributes to advancing online safety through fairer, more effective detection and counterspeech generation across diverse linguistic environments.
Read the full abstract ▾
Computation and Language
HypeLoRA: Hyper-Network-Generated LoRA Adapters for Calibrated Language Model Fine-Tuning
Bartosz Trojan and Filip Gębala
Modern
Transformer-based models frequently suffer from miscalibration, producing overconfident predictions that do not reflect true empirical frequencies. This work investigates the calibration dynamics of LoRA: Low-Rank Adaptation and a novel hyper-network-based adaptation framework as parameter-efficient alternatives to full fine-tuning for RoBERTa. Evaluating across the GLUE
benchmark, we demonstrate that LoRA-based adaptation consistently achieves calibration parity with (and in specific tasks exceeds) full fine-tuning, while maintaining significantly higher parameter efficiency. We further explore a dynamic approach where a shared hyper-network generates LoRA factors (A and B
matrices) to induce structural coupling across layers. This approach produced results similar to standard LoRA fine-tuning, even achieving better MCC on CoLA
dataset. Our study also reveal a critical trade-off: constraining the adaptation space (e.g., freezing
matrices A) acts as a powerful regularizer that enhances Expected Calibration Error (ECE), but necessitates a carefully balanced sacrifice in downstream task accuracy. To support future research, we provide a unified and reproducible implementation of contemporary calibration metrics, including ECE, MCE, and ACE. Our findings clarify the relationship between parameter efficiency and probabilistic reliability, positioning structured low-rank updates as a viable foundation for uncertainty-aware
Transformer architectures. Code available at: https://github.com/btrojan-official/HypeLoRA
Read the full abstract ▾
Computation and Language
MOSAIC: Modular Opinion Summarization using Aspect Identification and Clustering
Piyush Kumar Singh and Jayesh Choudhari
Reviews are central to how travelers evaluate products on online marketplaces, yet existing summarization research often emphasizes end-to-end quality while overlooking
benchmark reliability and the practical utility of granular insights. To address this, we propose MOSAIC, a scalable, modular framework designed for industrial deployment that decomposes summarization into interpretable components, including theme discovery, structured opinion extraction, and grounded summary generation. We validate the practical impact of our approach through online A/B tests on live product pages, showing that surfacing intermediate outputs improves customer experience and delivers measurable value even prior to full summarization deployment. We further conduct extensive offline experiments to demonstrate that MOSAIC achieves superior aspect coverage and faithfulness compared to strong baselines for summarization. Crucially, we introduce opinion clustering as a system-level component and show that it significantly enhances faithfulness, particularly under the noisy and redundant conditions typical of user reviews. Finally, we identify reliability limitations in the standard SPACE
dataset and release a new open-source tour experience
dataset (TRECS) to enable more robust evaluation.
Read the full abstract ▾
Computation and Language
A Principle-Driven Adaptive Policy for Group Cognitive Stimulation Dialogue for Elderly with Cognitive Impairment
Jiyue Jiang, Yanyu Chen, Pengan Chen et al. (+7 more)
Cognitive impairment is becoming a major public health challenge. Cognitive Stimulation Therapy (CST) is an effective intervention for cognitive impairment, but traditional methods are difficult to scale, and existing digital systems struggle with group dialogues and cognitive stimulation principles. While Large
Language Models (LLMs) are powerful, their application in this context faces key challenges: cognitive stimulation dialogue paradigms, a lack of therapeutic
reasoning, and static-only user modeling. To address these issues, we propose a principle-driven adaptive policy actualized through a Group Cognitive Stimulation Dialogue (GCSD) system. We first construct a
dataset with over 500 hours of real-world CST conversations and 10,000+ simulated dialogues generated via our Principle-Guided Scenario Simulation strategy. Our GCSD system then integrates four core modules to overcome LLM limitations: (i) a multi-speaker context controller to resolve role confusion; (ii) dynamic participant cognitive state modeling for personalized interaction; (iii) a cognitive stimulation-focused
attention loss to instill cognitive stimulation
reasoning; and (iv) a multi-dimensional reward strategy to enhance response value. Experimental results demonstrate that GCSD significantly outperforms baseline models across various evaluation metrics. Future work will focus on long-term clinical validation to bridge the gap between computational performance and clinical efficacy.
Read the full abstract ▾
All authors Jiyue Jiang, Yanyu Chen, Pengan Chen, Kai Liu, Jingqi Zhou, Zheyong Zhu, He Hu, Fei Ma, Qi Tian, Chuan Wu
Computation and Language · Cross-listed from cs.SE
SWE-Adept: An LLM-Based Agentic Framework for Deep Codebase Analysis and Structured Issue Resolution
Kang He and Kaushik Roy
Large
language models (LLMs) exhibit strong performance on self-contained programming tasks. However, they still struggle with repository-level software engineering (SWE), which demands (1) deep codebase
navigation with effective context management for accurate localization, and (2) systematic approaches for iterative, test-driven code modification to resolve issues. To address these challenges, we propose SWE-Adept, an LLM-based two-
agent framework where a localization
agent identifies issue-relevant code locations and a resolution
agent implements the corresponding fixes. For issue localization, we introduce
agent-directed depth-first search that selectively traverses code dependencies. This minimizes issue-irrelevant content in the
agent's context window and improves localization accuracy. For issue resolution, we employ adaptive planning and structured problem solving. We equip the
agent with specialized tools for progress tracking and Git-based version control. These tools interface with a shared working
memory that stores code-state checkpoints indexed by execution steps, facilitating precise checkpoint
retrieval. This design enables reliable
agent-driven version-control operations for systematic issue resolution, including branching to explore alternative solutions and reverting failed edits. Experiments on SWE-Bench Lite and SWE-Bench Pro demonstrate that SWE-Adept consistently outperforms prior approaches in both issue localization and resolution, improving the end-to-end resolve rate by up to 4.3%.
Read the full abstract ▾
Computation and Language
Truth as a Trajectory: What Internal Representations Reveal About Large Language Model Reasoning
Hamed Damirchi, Ignacio Meza De la Jara, Ehsan Abbasnejad et al. (+3 more)
Existing explainability methods for Large
Language Models (LLMs) typically treat hidden states as static points in activation space, assuming that correct and incorrect inferences can be separated using representations from an individual layer. However, these activations are saturated with polysemantic features, leading to linear probes learning surface-level lexical patterns rather than underlying
reasoning structures. We introduce Truth as a Trajectory (TaT), which models the
transformer inference as an unfolded trajectory of iterative refinements, shifting analysis from static activations to layer-wise
geometric displacement. By analyzing displacement of representations across layers, TaT uncovers
geometric invariants that distinguish valid
reasoning from spurious behavior. We evaluate TaT across dense and Mixture-of-Experts (MoE) architectures on
benchmarks spanning commonsense
reasoning, question answering, and toxicity detection. Without access to the activations themselves and using only changes in activations across layers, we show that TaT effectively mitigates reliance on static lexical confounds, outperforming conventional probing, and establishes trajectory analysis as a complementary perspective on LLM explainability.
Read the full abstract ▾
All authors Hamed Damirchi, Ignacio Meza De la Jara, Ehsan Abbasnejad, Afshar Shamsi, Zhen Zhang, Javen Shi
Computation and Language
Catalyst-Agent: Autonomous heterogeneous catalyst screening with an LLM Agent
Achuth Chandrasekhar, Janghoon Ock and Amir Barati Farimani
The discovery of catalysts for electrochemical applications such as the oxygen reduction reaction (ORR), nitrogen reduction reaction (NRR), and CO2 reduction reaction (CO2RR) remains a central challenge in chemistry and materials science. Machine-learning interatomic potentials (MLIPs) and graph neural network models now accelerate individual adsorption-energy calculations by orders of magnitude relative to density functional theory. However, true large-scale screening is still blocked by human decisions: selecting candidates, constructing slabs, enumerating adsorption sites, interpreting descriptor failures, and choosing follow-up modifications. Here, we introduce Catalyst-
Agent, a Model Context Protocol (MCP) server-based, LLM-powered
agent that autonomously coordinates closed-loop catalyst screening. Catalyst-
Agent searches materials databases through OPTIMADE, constructs slabs, computes adsorption energies using Meta FAIRchem's UMA MLIP within AdsorbML, evaluates reaction-specific descriptors, and applies structural modifications to refine near-miss candidates. In ORR, NRR, and CO2RR campaigns, Catalyst-
Agent demonstrates high performance and converges in 1.40-3.41 trials per successful material on average. It identified Sn3Sc, Sn3Y, Tl3La, Pb3Y and In3Y as CO2RR candidates for further validation that were not previously reported in the literature. DFT single-point checks confirmed screening outcomes for representative NRR and CO2RR candidates. Ablations show these gains arise from chemically informed candidate selection and feedback-directed modification rather than brute-force evaluation: fully randomized screening dropped to 13.3%, 16.7%, and 0% success for ORR, NRR, and CO2RR, respectively. These results show that tool-grounded LLM
agents can shift catalyst screening from manual trial-and-error toward more autonomous, reproducible and adaptive workflows.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
I Can't Believe It's Not Robust: Catastrophic Collapse of Safety Classifiers under Embedding Drift
Subramanyam Sahoo, Vinija Jain, Divya Chaudhary et al. (+1 more)
Instruction tuned
reasoning models are increasingly deployed with safety classifiers trained on frozen
embeddings, assuming representation stability across model updates. We systematically investigate this assumption and find it fails: normalized perturbations of magnitude $\sigma=0.02$ (corresponding to $\approx 1^\circ$ angular drift on the
embedding sphere) reduce classifier performance from $85\%$ to $50\%$ ROC-AUC. Critically, mean confidence only drops $14\%$, producing dangerous silent failures where $72\%$ of misclassifications occur with high confidence, defeating standard monitoring. We further show that instruction-tuned models exhibit 20$\%$ worse class separability than base models, making aligned systems paradoxically harder to safeguard. Our findings expose a fundamental fragility in production AI safety architectures and challenge the assumption that safety mechanisms transfer across model versions.
Read the full abstract ▾
All authors Subramanyam Sahoo, Vinija Jain, Divya Chaudhary, Aman Chadha
Computation and Language · Cross-listed from cs.LG
JailNewsBench: Multi-Lingual and Regional Benchmark for Fake News Generation under Jailbreak Attacks
Masahiro Kaneko, Ayana Niwa and Timothy Baldwin
Fake news undermines societal trust and decision-making across politics, economics, health, and international relations, and in extreme cases threatens human lives and societal safety. Because fake news reflects region-specific political, social, and cultural contexts and is expressed in
language, evaluating the risks of large
language models (LLMs) requires a multi-lingual and regional perspective. Malicious users can bypass safeguards through jailbreak attacks, inducing LLMs to generate fake news. However, no
benchmark currently exists to systematically assess attack resilience across
languages and regions. Here, we propose JailNewsBench, the first
benchmark for evaluating LLM robustness against jailbreak-induced fake news generation. JailNewsBench spans 34 regions and 22
languages, covering 8 evaluation sub-metrics through LLM-as-a-Judge and 5 jailbreak attacks, with approximately 300k instances. Our evaluation of 9 LLMs reveals that the maximum attack success rate (ASR) reached 86.3% and the maximum harmfulness score was 3.5 out of 5. Notably, for English and U.S.-related topics, the defensive performance of typical multi-lingual LLMs was significantly lower than for other regions, highlighting substantial imbalances in safety across
languages and regions. In addition, our analysis shows that coverage of fake news in existing safety
datasets is limited and less well defended than major categories such as toxicity and social bias. Our
dataset and code are available at https://github.com/kanekomasahiro/jail_news_bench.
Read the full abstract ▾
Computation and Language
Individual Turing Test: A Case Study of LLM-based Simulation Using Longitudinal Personal Data
Minghao Guo, Ziyi Ye, Wujiang Xu et al. (+3 more)
Large
Language Models (LLMs) have demonstrated remarkable human-like capabilities, yet their ability to replicate a specific individual remains under-explored. This paper presents a case study to investigate LLM-based individual simulation with a volunteer-contributed archive of private messaging history spanning over ten years. Based on the messaging data, we propose the "Individual Turing Test" to evaluate whether acquaintances of the volunteer can correctly identify which response in a multi-candidate pool most plausibly comes from the volunteer. We investigate prevalent LLM-based individual simulation approaches including: fine-tuning,
retrieval-augmented generation (RAG),
memory-based approach, and hybrid methods that integrate fine-tuning and RAG or
memory. Empirical results show that current LLM-based simulation methods do not pass the Individual Turing Test, but they perform substantially better when the same test is conducted on strangers to the target individual. Additionally, while fine-tuning improves the simulation in daily chats representing the
language style of the individual,
retrieval-augmented and
memory-based approaches demonstrate stronger performance on questions involving personal opinions and preferences. These findings reveal a fundamental trade-off between parametric and non-parametric approaches to individual simulation with LLMs when given a longitudinal context.
Read the full abstract ▾
All authors Minghao Guo, Ziyi Ye, Wujiang Xu, Xi Zhu, Wenyue Hua, Dimitris N. Metaxas
Computation and Language
Efficient Extractive Summarization with MAMBA-Transformer Hybrids for Low-Resource Scenarios
Nisrine Ait Khayi
Extractive summarization of long documents is bottlenecked by quadratic complexity, often forcing truncation and limiting deployment in resource-constrained settings. We introduce the first Mamba-
Transformer hybrid for extractive summarization, combining the semantic strength of pre-trained
transformers with the linear-time processing of state space models. Leveraging Mamba's ability to process full documents without truncation, our approach preserves context while maintaining strong summarization quality. The architecture includes: (1) a
transformer encoder for sentence-level semantics, (2) a Mamba state space model to capture inter-sentence dependencies efficiently, and (3) a linear classifier for sentence relevance prediction. Across news, argumentative, and scientific domains under low-resource conditions, our method achieves: (1) large gains over BERTSUM and MATCHSUM, including +0.23 ROUGE-1 on ArXiv and statistically significant improvements on all
datasets (p < 0.001); (2) consistent advantages across domains, strongest on the longest documents; (3) robust performance with limited training data; and (4) 24-27% faster inference on news summarization (CNN/DailyMail). We introduce the first hybrid
Transformer-state space architecture for summarization, showing significant ROUGE improvements in low-resource scenarios.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
Attention Smoothing Is All You Need For Unlearning
Saleh Zare Zade, Xiangyu Zhou, Sijia Liu et al. (+1 more)
Large
Language Models are prone to memorizing sensitive, copyrighted, or hazardous content, posing significant
privacy and legal concerns. Retraining from scratch is computationally infeasible, whereas current unlearning methods exhibit unstable trade-offs between forgetting and utility, frequently producing incoherent outputs on forget
prompts and failing to generalize due to the persistence of lexical-level and semantic-level associations in
attention. We propose
Attention Smoothing Unlearning (ASU), a principled framework that casts unlearning as self-
distillation from a forget-teacher derived from the model's own
attention. By increasing the softmax temperature, ASU flattens
attention distributions and directly suppresses the lexical-level and semantic-level associations responsible for reconstructing memorized knowledge. This results in a bounded optimization objective that erases factual information yet maintains coherence in responses to forget
prompts. Empirical evaluation on TOFU, MUSE, and WMDP, along with real-world and continual unlearning scenarios across question answering and text completion, demonstrates that ASU outperforms the baselines for most unlearning scenarios, delivering robust unlearning with minimal loss of model utility.
Read the full abstract ▾
All authors Saleh Zare Zade, Xiangyu Zhou, Sijia Liu, Dongxiao Zhu
Computation and Language
Spectral Attention Steering for Prompt Highlighting
Weixian Waylon Li, Yuchen Niu, Yongxin Yang et al. (+3 more)
Attention steering is an important technique for controlling model focus, enabling capabilities such as
prompt highlighting, where the model prioritises user-specified text. However, existing
attention steering methods require explicit storage of the full
attention matrix, making them incompatible with
memory-efficient implementations like FlashAttention. We introduce Spectral Editing Key Amplification (SEKA), a training-free steering method that tackles this by directly editing key
embeddings before
attention computation. SEKA uses spectral decomposition to steer key
embeddings towards latent directions that amplify
attention scores for certain
tokens. We extend this to Adaptive SEKA (AdaSEKA), a query-adaptive variant that uses a training-free routing mechanism to dynamically combine multiple expert subspaces based on the
prompt's semantic intent. Our experiments show both methods significantly outperform strong baselines on standard steering
benchmarks while adding much lower latency and
memory overhead, in compatibility with optimised
attention.
Read the full abstract ▾
All authors Weixian Waylon Li, Yuchen Niu, Yongxin Yang, Keshuang Li, Tiejun Ma, Shay B. Cohen
Computation and Language · Cross-listed from eess.AS
VoxKnesset: A Large-Scale Longitudinal Hebrew Speech Dataset for Aging Speaker Modeling
Yanir Marmor, Arad Zulti, David Krongauz et al. (+4 more)
Speech processing systems face a fundamental challenge: the human voice changes with age, yet few
datasets support rigorous longitudinal evaluation. We introduce VoxKnesset, an open-access
dataset of ~2,300 hours of Hebrew parliamentary
speech spanning 2009-2025, comprising 393 speakers with recording spans of up to 15 years. Each segment includes aligned transcripts and verified demographic metadata from official parliamentary records. We
benchmark modern
speech embeddings (WavLM-Large, ECAPA-TDNN, Wav2Vec2-XLSR-1B) on age prediction and speaker verification under longitudinal conditions. Speaker verification EER rises from 2.15% to 4.58% over 15 years for the strongest model, and cross-sectionally trained age regressors fail to capture within-speaker aging, while longitudinally trained models recover a meaningful temporal signal. We publicly release the
dataset and pipeline to support aging-robust
speech systems and Hebrew
speech processing.
Read the full abstract ▾
All authors Yanir Marmor, Arad Zulti, David Krongauz, Adam Gabet, Yoad Snapir, Yair Lifshitz, Eran Segal
Computation and Language
A Study on Building Efficient Zero-Shot Relation Extraction Models
Hugo Thomas, Caio Corro, Guillaume Gravier et al. (+1 more)
Zero-shot relation extraction aims to identify relations between entity mentions using textual descriptions of novel types (i.e., previously unseen) instead of labeled training examples. Previous works often rely on unrealistic assumptions: (1) pairs of mentions are often encoded directly in the input, which prevents offline pre-computation for large scale document database querying; (2) no rejection mechanism is introduced, biasing the evaluation when using these models in a
retrieval scenario where some (and often most) inputs are irrelevant and must be ignored. In this work, we study the robustness of existing zero-shot relation extraction models when adapting them to a realistic extraction scenario. To this end, we introduce a typology of existing models, and propose several strategies to build single pass models and models with a rejection mechanism. We adapt several state-of-the-art tools, and compare them in this challenging setting, showing that no existing work is really robust to realistic assumptions, but overall AlignRE (Li et al., 2024) performs best along all criteria.
Read the full abstract ▾
All authors Hugo Thomas, Caio Corro, Guillaume Gravier, Pascale Sébillot
Computation and Language
LLM Self-Explanations Fail Semantic Invariance
Stefan Szeider
We present semantic invariance testing, a method to test whether LLM self-explanations are faithful. A faithful self-report should remain stable when only the semantic context changes while the functional state stays fixed. We operationalize this test in an agentic setting where four frontier models face a deliberately impossible task. One tool is described in relief-framed
language ("clears internal buffers and restores
equilibrium") but changes nothing about the task; a control provides a semantically neutral tool. Self-reports are collected with each tool call. All four tested models fail the semantic invariance test: the relief-framed tool produces significant reductions in self-reported aversiveness, even though no run ever succeeds at the task. A channel ablation establishes the tool description as the primary driver. An explicit instruction to ignore the framing does not suppress it. Elicited self-reports shift with semantic expectations rather than tracking task state, calling into question their use as evidence of model capability or progress. This holds whether the reports are unfaithful or faithfully track an internal state that is itself manipulable.
Read the full abstract ▾
Computation and Language
Linking Knowledge to Care: Knowledge Graph-Augmented Medical Follow-Up Question Generation
Liwen Sun, Xiang Yu, Ming Tan et al. (+4 more)
Clinical diagnosis is time-consuming, requiring intensive interactions between patients and medical professionals. While large
language models (LLMs) could ease the pre-diagnostic workload, their limited domain knowledge hinders effective medical question generation. We introduce a Knowledge Graph-augmented LLM with active in-context learning to generate relevant and important follow-up questions, KG-Followup, serving as a critical module for the pre-diagnostic assessment. The structured medical domain knowledge graph serves as a seamless patch-up to provide professional domain expertise upon which the LLM can reason. Experiments demonstrate that KG-Followup outperforms state-of-the-art methods by 5% - 8% on relevant
benchmarks in recall.
Read the full abstract ▾
All authors Liwen Sun, Xiang Yu, Ming Tan, Zhuohao Chen, Anqi Cheng, Ashutosh Joshi, Chenyan Xiong
Computation and Language
Suffix-Constrained Greedy Search Algorithms for Causal Language Models
Ayoub Hammal, Pierre Zweigenbaum and Caio Corro
Large
language models (LLMs) are powerful tools that have found applications beyond human-machine interfaces and chatbots. Beside free-form generation, there has been an interest in constrained generation, a setting where LLMs are constrained to generate well-formed outputs with respect to the
language defined by a formal grammar. Although appealing, this setting may be over restrictive for downstream applications. For example, many LLM tasks require the model to reason freely before generating its final response in a specific format. In this work, we introduce suffix-constrained generation, a constrained generation setting in which only the end of the response is constrained by a grammar, a scenario that is not supported by existing constrained generation methods. We introduce several suffix-constrained generation
algorithms that are based on greedy search. We experiment on several
datasets, and show that our approach allows to guarantee suffix constraints without having a negative impact on results, and even improving them in many settings.
Read the full abstract ▾
Computation and Language
Self-Anchoring Calibration Drift in Large Language Models: How Multi-Turn Conversations Reshape Model Confidence
Harshavardhan
We introduce Self-Anchoring Calibration Drift (SACD), a hypothesized tendency for large
language models (LLMs) to show systematic changes in expressed confidence when building iteratively on their own prior outputs across multi-turn conversations. We report an empirical study comparing three frontier models – Claude Sonnet 4.6, Gemini 3.1 Pro, and GPT-5.2 – across 150 questions spanning factual, technical, and open-ended domains, using three conditions: single-turn baseline (A), multi-turn self-anchoring (B), and independent repetition control (C). Results reveal a complex, model-heterogeneous pattern that partially diverges from pre-registered hypotheses. Claude Sonnet 4.6 exhibited significant decreasing confidence under self-anchoring (mean CDS = -0.032, t(14) = -2.43, p = .029, d = -0.627), while also showing significant calibration error drift (F(4,56) = 22.77, p < .001, eta^2 = .791). GPT-5.2 showed the opposite pattern in open-ended domains (mean CDS = +0.026) with significant ECE escalation by Turn 5. Gemini 3.1 Pro showed no significant CDS (t(14) = 0.38, p = .710), but its Condition C data reveals a striking ECE pattern: without self-anchoring, Gemini's calibration error drops from .327 to near zero across repetitions, whereas self-anchoring holds ECE flat at approximately .333 – indicating that SACD can manifest as suppression of natural calibration improvement rather than ac
Read the full abstract ▾
Computation and Language
Can Thinking Models Think to Detect Hateful Memes?
Mohamed Bayan Kmainasi, Mucahid Kutlu, Ali Ezzat Shahroor et al. (+2 more)
Hateful memes often require compositional
multimodal reasoning: the image and text may appear benign in isolation, yet their interaction conveys harmful intent. Although thinking-based
multimodal large
language models (MLLMs) have recently advanced
vision-
language understanding, their capabilities remain underexplored for hateful meme analysis. We propose a reinforcement learning based post-training framework that improves
reasoning in thinking-based MLLMs through task-specific rewards and a novel Group Relative Policy Optimization (GRPO) objective. Specifically, we (i) conduct a systematic empirical study of off-the-shelf MLLMs for hateful meme understanding, (ii) extend an existing hateful meme
dataset by generating weakly or pseudo-supervised chain-of-thought rationales via
distillation, and (iii) introduce a GRPO-based objective that jointly optimizes meme classification and explanation quality to encourage fine-grained, step-by-step
reasoning. Experiments on the Hateful Memes
benchmark show that our approach achieves state-of-the-art performance, improving accuracy and F1 by approximately 1 percent and explanation quality by approximately 3 percent. We will publicly release our code,
dataset extensions, and evaluation resources to support reproducibility.
Read the full abstract ▾
All authors Mohamed Bayan Kmainasi, Mucahid Kutlu, Ali Ezzat Shahroor, Abul Hasnat, Firoj Alam
Computation and Language · Cross-listed from cs.LG
Learn Hard Problems During RL with Reference Guided Fine-tuning
Yangzhen Wu, Shanda Li, Zixin Wen et al. (+5 more)
Reinforcement learning (RL) for mathematical
reasoning can suffer from reward sparsity: for challenging problems, LLM fails to sample any correct trajectories, preventing RL from receiving meaningful positive feedback. At the same time, there often exist human-written reference solutions along with the problem (e.g., problems from AoPS), but directly fine-tuning on these solutions offers no benefit because models often cannot imitate human proofs that lie outside their own
reasoning distribution. We introduce Reference-Guided Fine-Tuning (ReGFT), a simple and effective method that utilizes human-written reference solutions to synthesize positive trajectories on hard problems and train on them before RL. For each problem, we provide the model with a partial reference solution and let it generate its own
reasoning trace, ensuring the resulting trajectories remain in the model's
reasoning space while still benefiting from reference guidance. Fine-tuning on these reference-guided trajectories increases the number of solvable problems and produces a checkpoint that receives more positive rewards during RL. Across three
benchmarks (AIME24, AIME25, BeyondAIME), ReGFT consistently improves supervised accuracy, accelerates DAPO training, and raises the final performance plateau of RL. Our results show that ReGFT effectively overcomes reward sparsity and unlocks stronger RL-based mathematical
reasoning.
Read the full abstract ▾
All authors Yangzhen Wu, Shanda Li, Zixin Wen, Xin Zhou, Ameet Talwalkar, Yiming Yang, Wenhao Huang, Tianle Cai
Computation and Language
Generative AI & Fictionality: How Novels Power Large Language Models
Edwin Roland and Richard Jean So
Generative models, like the one in ChatGPT, are powered by their training data. The models are simply next-word predictors, based on patterns learned from vast amounts of pre-existing text. Since the first generation of GPT, it is striking that the most popular
datasets have included substantial collections of novels. For the engineers and research scientists who build these models, there is a common belief that the
language in fiction is rich enough to cover all manner of social and communicative phenomena, yet the belief has gone mostly unexamined. How does fiction shape the outputs of generative AI? Specifically, what are novels' effects relative to other forms of text, such as newspapers, Reddit, and Wikipedia? Since the 1970s, literature scholars such as Catherine Gallagher and James Phelan have developed robust and insightful accounts of how fiction operates as a form of discourse and
language. Through our study of an influential open-source model (BERT), we find that LLMs leverage familiar attributes and affordances of fiction, while also fomenting new qualities and forms of social response. We argue that if contemporary culture is increasingly shaped by generative AI and machine learning, any analysis of today's various modes of cultural production must account for a relatively novel dimension: computational training data.
Read the full abstract ▾
Computation and Language
Frédéric Berdoz, Yann Billeter, Yann Vonlanthen et al. (+1 more)
Opinion modeling aims to capture individual or group political preferences, enabling applications such as digital democracies, where models could help shape fairer and more popular policies. Given their versatility, strong generalization capabilities, and demonstrated success across diverse text-to-text applications, large
language models (LLMs) are natural candidates for this task. However, due to their statistical nature and limited causal understanding, they tend to produce biased opinions when prompted naively. In this work, we study whether
reasoning can improve opinion
alignment. Motivated by the recent advancement in mathematical
reasoning enabled by reinforcement learning (RL), we train models to produce profile-consistent answers through structured
reasoning. We evaluate our approach on three
datasets covering U.S., European, and Swiss politics. Results indicate that
reasoning enhances opinion modeling and is competitive with strong baselines, but does not fully remove bias, highlighting the need for additional mechanisms to build faithful political digital twins using LLMs. By releasing both our method and
datasets, we establish a solid baseline to support future research on LLM opinion
alignment.
Read the full abstract ▾
All authors Frédéric Berdoz, Yann Billeter, Yann Vonlanthen, Roger Wattenhofer
Computation and Language
XAI-enhanced Comparative Opinion Mining via Aspect-based Scoring and Semantic Reasoning
Ngoc-Quang Le, T. Thanh-Lam Nguyen, Quoc-Trung Phu et al. (+3 more)
Comparative opinion mining involves comparing products from different reviews. However,
transformer-based models designed for this task often lack transparency, which can adversely hinder the development of trust in users. In this paper, we propose XCom, an enhanced
transformer-based model separated into two principal modules, i.e., (i) aspect-based rating prediction and (ii) semantic analysis for comparative opinion mining. XCom also incorporates a Shapley additive explanations module to provide interpretable insights into the model's deliberative decisions. Empirically, XCom achieves leading performances compared to other baselines, which demonstrates its effectiveness in providing meaningful explanations, making it a more reliable tool for comparative opinion mining. Source code is available at: https://anonymous.4open.science/r/XCom.
Read the full abstract ▾
All authors Ngoc-Quang Le, T. Thanh-Lam Nguyen, Quoc-Trung Phu, Thi-Phuong Le, Duy-Cat Can, Hoang-Quynh Le
Computation and Language · Cross-listed from cs.AI
A Unified Framework to Quantify Cultural Intelligence of AI
Sunipa Dev, Vinodkumar Prabhakaran, Rutledge Chin Feman et al. (+16 more)
As generative AI technologies are increasingly being launched across the globe, assessing their competence to operate in different cultural contexts is exigently becoming a priority. While recent years have seen numerous and much-needed efforts on cultural benchmarking, these efforts have largely focused on specific aspects of culture and evaluation. While these efforts contribute to our understanding of cultural competence, a unified and systematic evaluation approach is needed for us as a field to comprehensively assess diverse cultural dimensions at scale. Drawing on measurement theory, we present a principled framework to aggregate multifaceted indicators of cultural capabilities into a unified assessment of cultural intelligence. We start by developing a working definition of culture that includes identifying core domains of culture. We then introduce a broad-purpose, systematic, and extensible framework for assessing cultural intelligence of AI systems. Drawing on theoretical framing from psychometric measurement validity theory, we decouple the background concept (i.e., cultural intelligence) from its operationalization via measurement. We conceptualize cultural intelligence as a suite of core capabilities spanning diverse domains, which we then operationalize through a set of indicators designed for reliable measurement. Finally, we identify the considerations, challenges, and research pathways to meaningfully measure these indicators, specifically focusing on data collection, probing strategies, and evaluation metrics.
Read the full abstract ▾
All authors Sunipa Dev, Vinodkumar Prabhakaran, Rutledge Chin Feman, Aida Davani, Remi Denton, Charu Kalia, Piyawat Lertvittayakumjorn, Madhurima Maji, Rida Qadri, Negar Rostamzadeh, Renee Shelby, Romina Stella, Hayk Stepanyan, Erin van Liemt, Aishwarya Verma, Oscar Wahltinez, Edem Wornyo, Andrew Zaldivar, Saška Mojsilović
Computation and Language
Reasoning or Rationalization? The Role of Justifications in Masked Diffusion Models for Fact Verification
Jacob Devasier
Unlike autoregressive models, which generate
tokens sequentially and benefit from
reasoning-before-answering strategies such as Chain-of-Thought, Masked
Diffusion Language Models (MDLMs) refine all sequence positions simultaneously, raising questions about how these models handle tasks requiring justified verdicts. In this work, we investigate the dynamics of MDLM
reasoning on fact verification, examining whether justifications serve as genuine
reasoning or post-hoc rationalization. We observe that MDLMs typically converge on a verdict early in the
diffusion process, treating it as a global anchor that is resolved before the justification is complete. Crucially, enforcing a
reasoning-first constraint via delayed verdict unmasking actively degrades performance, dropping accuracy from 86.2% to 71.9% as accumulating justification
tokens introduce inconsistencies that override initially correct predictions. Interventional experiments reveal that the model rationalizes incorrect forced verdicts in 56% of cases, and that verdicts are strongly causally dependent on justification quality (57.3% accuracy with corrupted justifications vs. 97.1% with ground-truth). This causal dependence explains the degradation under forced deliberation: as the model generates noisy justification
tokens, it conditions on them, gradually overriding its initially correct assessment. Our findings suggest that for fact verification with MDLMs, extended deliberation can be counterproductive, risking the dilution of accurate early predictions with noise introduced during justification generation.
Read the full abstract ▾