Search: “arxiv-cs-cl”

Monitoring the Internal Monologue: Probe Trajectories Reveal Reasoning Dynamics

arXiv:2605.18549v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) introduce new opportunities for safety monitoring through their Chain of Thought (CoT) reasoning. However, CoT is not alwa

Multilingual and Multimodal LLMs in the Wild: Building for Low-Resource Languages

DGX agent

arXiv:2605.17152v1 Announce Type: new Abstract: Multimodal LLMs are evolving from vision-language to tri-modality that see, hear, and read, yet pipelines and benchmarks remain English-centric and comp

safetyarxiv-cs-cl

19 May 2026

Multilingual OCR-Aware Fine-Tuning and Prompt-Guided Chain-of-Thought Reasoning for Multimodal Large Language Models

arXiv:2605.16409v1 Announce Type: cross Abstract: Optical character recognition (OCR) and multilingual text understanding remain major failure modes of multimodal large language models (MLLMs), partic

NewsLens: A Multi-Agent Framework for Adversarial News Bias Navigation

arXiv:2605.17364v1 Announce Type: new Abstract: Media bias detection has predominantly been framed as a classification task: assign a political label to an article or outlet. We argue this framing is

PaliBench: A Multi-Reference Blueprint for Classical Language Translation Benchmarks

arXiv:2605.16881v1 Announce Type: new Abstract: Digital humanities projects increasingly rely on machine translation and large language models to widen access to classical, religious, and otherwise un

PEGRL: Improving Machine Translation by Post-Editing Guided Reinforcement Learning

arXiv:2602.03352v2 Announce Type: replace Abstract: Reinforcement learning (RL) has shown strong promise for LLM-based machine translation, with recent methods such as GRPO demonstrating notable gains

Local Ai

PPAI: Enabling Personalized LLM Agent Interoperability for Collaborative Edge Intelligence

arXiv:2605.18067v1 Announce Type: new Abstract: Deploying large language model (LLM) on edge device enables personalized LLM agents for various users. The growing availability of diverse personalized

local-aiarxiv-cs-cl

PQR: A Framework to Generate Diverse and Realistic User Queries that Elicit QA Agent Failures

arXiv:2605.16551v1 Announce Type: new Abstract: Evaluating LLM-based agents remains challenging because identifying meaningful failure cases often requires substantial human effort to design realistic

Presupposition and Reasoning in Conditionals: A Theory-Based Study of Humans and LLMs

arXiv:2605.18352v1 Announce Type: new Abstract: Presupposition projection in conditionals is central to theories of meaning and pragmatics, yet it remains largely unevaluated in large language models.

Agents

Proof-Carrying Certificates for LLM Pipelines: A Trust-Boundary Architecture

arXiv:2605.16407v1 Announce Type: cross Abstract: We present a framework for verifying the deterministic structured computations surrounding a large language model rather than the model itself, extend

agentsarxiv-cs-cl

Protection Is (Nearly) All You Need: Structural Protection Dominates Scoring in Globally Capped KV Eviction

arXiv:2605.18053v1 Announce Type: cross Abstract: We study KV cache eviction under a shared globally capped decode-time harness. Seven policies (LRU, H2O, SnapKV, StreamingLLM, Ada-KV, QUEST, Random)

Provable Knowledge Acquisition and Extraction in One-Layer Transformers

arXiv:2508.00901v4 Announce Type: replace-cross Abstract: Large language models may encounter factual knowledge during pre-training yet fail to reliably use that knowledge after fine-tuning. Despite g

QuCo-RAG: Quantifying Uncertainty from the Pre-training Corpus for Dynamic Retrieval-Augmented Generation

arXiv:2512.19134v2 Announce Type: replace Abstract: Dynamic Retrieval-Augmented Generation adaptively determines when to retrieve during generation to mitigate hallucinations in large language models

Query-Aware Learnable Graph Pooling Tokens as Prompt for Large Language Models

arXiv:2501.17549v2 Announce Type: replace Abstract: Graph-structured data plays a vital role in numerous domains, such as social networks, citation networks, commonsense reasoning graphs and knowledge

Readers make targeted regressions to plausible errors in reanalysis of 'noisy-channel garden-path' sentences

arXiv:2605.18563v1 Announce Type: new Abstract: A key question in psycholinguistics is how inferences about the meaning of linguistic input unfold incrementally a comprehender's mind. In this work, we

Red-Bandit: Test-Time Adaptation for LLM Red-Teaming via Bandit-Guided LoRA Experts

arXiv:2510.07239v2 Announce Type: replace Abstract: Automated red-teaming has emerged as a scalable approach for auditing Large Language Models (LLMs) prior to deployment, yet existing approaches lack

Residual Semantic Decomposition of Word Embeddings

arXiv:2605.17482v1 Announce Type: new Abstract: We introduce Residual Semantic Decomposition (RSD), a neural additive decomposition of word embeddings that balances embedding reconstruction with relat

Responsible Federated LLMs via Safety Filtering and Constitutional AI

arXiv:2502.16691v2 Announce Type: replace Abstract: Recent research has increasingly focused on training large language models (LLMs) using federated learning, known as FedLLM. However, responsible AI

Rethinking 1-bit Optimization Leveraging Pre-trained Large Language Models

arXiv:2508.06974v2 Announce Type: replace Abstract: 1-bit LLM quantization offers significant advantages in reducing storage and computational costs. However, existing methods typically train 1-bit LL

Rethinking Table Pruning in TableQA: From Sequential Revisions to Gold Trajectory-Supervised Parallel Search

arXiv:2601.03851v2 Announce Type: replace Abstract: Table Question Answering (TableQA) benefits significantly from table pruning, which extracts compact sub-tables by eliminating redundant cells to st

Retrieval-Based Multi-Label Legal Annotation: Extensible, Data-Efficient and Hallucination-Free

arXiv:2605.16767v1 Announce Type: new Abstract: Multi-label legal annotation requires assigning multiple labels from large, evolving taxonomies to long, fact-intensive documents, often under limited s

Roll Out and Roll Back: Diffusion LLMs are Their Own Efficiency Teachers

arXiv:2605.16941v1 Announce Type: new Abstract: Diffusion Large Language Models (DLLMs) promise fast parallel generation, yet open-source DLLMs still face a severe quality-speed trade-off: acceleratin

RTI-Bench: A Structured Dataset for Indian Right-to-Information Decision Analysis

arXiv:2605.16843v1 Announce Type: new Abstract: India's Right to Information Act, 2005 gives every citizen the right to demand information from public authorities, yet in practice most people cannot m

SafeLens: Deliberate and Efficient Video Guardrails with Fast-and-Slow Screening

arXiv:2605.17610v1 Announce Type: cross Abstract: The rapid growth of online video platforms and AI-generated content has made reliable video guardrails a key challenge for safety and real-world deplo

Scale Determines Whether Language Models Organize Representation Geometry for Prediction

arXiv:2605.17084v1 Announce Type: cross Abstract: In language models, what a representation encodes is determined by the geometry of its representation space: distances, not activations, carry meaning

Scaling Accessible Mathematics on arXiv: HTML Conversion and MathML 4

arXiv:2605.16562v1 Announce Type: new Abstract: We report on the ongoing development of arXiv's HTML Papers offering, available on every new TeX/LaTeX submission since its initial release in 2023. The

Scaling Laws for Code: A More Data-Hungry Regime

arXiv:2510.08702v2 Announce Type: replace Abstract: Code Large Language Models (LLMs) are revolutionizing software engineering. However, scaling laws that guide the efficient training are predominantl

SEDD: Scalable and Efficient Dataset Deduplication with GPUs

arXiv:2501.01046v4 Announce Type: replace Abstract: Dataset deduplication is widely recognized as a crucial preprocessing step that enhances data quality and improves the performance of large language

Self-Improving CAD Generation Agents with Finite Element Analysis as Feedback

arXiv:2605.17448v1 Announce Type: cross Abstract: Computer-aided design (CAD) is the backbone of modern industrial design, yet learned CAD generators still fall short of real engineering pipelines: th

Applications

Semantic Reranking at Inference Time for Hard Examples in Rhetorical Role Labeling

arXiv:2605.18007v1 Announce Type: new Abstract: Rhetorical Role Labeling (RRL) assigns a functional role to each sentence in a document and is widely used in legal, medical, and scientific domains. Wh

applicationsarxiv-cs-cl

SIREM: Speech-Informed MRI Reconstruction with Learned Sampling

arXiv:2605.18221v1 Announce Type: cross Abstract: Real-time magnetic resonance imaging (rtMRI) of speech production enables non-invasive visualization of dynamic vocal-tract motion and is valuable for

Sometin Beta Pass Notin (SBPN): Improving Multilingual ASR for Nigerian Languages via Knowledge Distillation

arXiv:2605.17710v1 Announce Type: new Abstract: Although modern multilingual Automatic Speech Recognition (ASR) systems support several Nigerian languages, their performance consistently lags behind h

Sparse-to-Dense: A Free Lunch for Lossless Acceleration of Video Understanding in LLMs

arXiv:2505.19155v2 Announce Type: replace-cross Abstract: Due to the auto-regressive nature of current video large language models (Video-LLMs), the inference latency increases as the input sequence l

Speak Your Mind: The Speech Continuation Task as a Probe of Voice-Based Model Bias

arXiv:2509.22061v2 Announce Type: replace-cross Abstract: Speech Continuation (SC) is the task of generating a coherent extension of a spoken prompt while preserving both semantic context and speaker

Spherical Steering: Geometry-Aware Activation Rotation for Language Models

arXiv:2602.08169v2 Announce Type: replace-cross Abstract: Inference-time steering offers a promising way to control language models (LMs) without retraining. However, standard approaches typically rel

Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models

arXiv:2605.17672v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) achieve strong performance by generating long chains of thought (CoT), but often overthink, continuing to reason after a s

T-FIX: Text-Based Explanations with Features Interpretable to eXperts

arXiv:2511.04070v3 Announce Type: replace Abstract: As LLMs are deployed in knowledge-intensive settings (e.g., surgery, astronomy, therapy), users are often domain experts who expect not just answers

Taming 'Zombie'' Agents: A Markov State-Aware Framework for Resilient Multi-Agent Evolution

arXiv:2605.17348v1 Announce Type: new Abstract: Recent advancements in LLM-based multi-agent systems have demonstrated remarkable collaborative capabilities across complex tasks. To improve overall ef

Temporal Decay of Co-Citation Predictability: A 20-Year Statute Retrieval Benchmark from 396M Ukrainian Court Citations

arXiv:2605.17639v1 Announce Type: new Abstract: Co-citation structure is widely assumed to provide stable retrieval signal in legal information systems. We test this assumption longitudinally by const

The Expressive Power of Low Precision Softmax Transformers with (Summarized) Chain-of-Thought

arXiv:2605.18079v1 Announce Type: cross Abstract: Existing expressivity results for transformers typically rely on hardmax attention, high precision, and other architectural modifications that disconn

The Frequency Confound in Language-Model Surprisal and Metaphor Novelty

arXiv:2605.06506v2 Announce Type: replace Abstract: Language-model (LM) surprisal is widely used as a proxy for contextual predictability and has been reported to correlate with metaphor novelty judgm

The Unlearnability Phenomenon in RLVR for Language Models

arXiv:2605.16787v1 Announce Type: cross Abstract: Reinforcement Learning with Verifiable Reward (RLVR) has proven effective in improving Large Language Model's (LLM) reasoning ability. However, the le

To MRL or not to MRL: Text Embeddings are Robust to Truncation Without Matryoshka Embeddings, Except In Heavy Truncation Scenarios

arXiv:2605.16608v1 Announce Type: cross Abstract: Matryoshka Representation Learning (MRL) is a widely adopted approach for training text encoders so they provide useful text representations at variou

ToolMATH: A Diagnostic Benchmark for Long-Horizon Tool Use under Systematic Tool-Catalog Constraints

arXiv:2602.21265v2 Announce Type: replace Abstract: We introduce ToolMATH, a math-grounded diagnostic benchmark for evaluating long-horizon tool use under controllable tool-catalog conditions. ToolMAT

Applications

Traces of Social Competence in Large Language Models

arXiv:2603.04161v2 Announce Type: replace Abstract: The False Belief Test (FBT) has been the main method for assessing Theory of Mind (ToM) and related socio-cognitive competencies. For Large Language

applicationsarxiv-cs-cl

Trust No Tool: Evaluating and Defending LLM Agents under Untrusted Tool Feedback

arXiv:2605.17453v1 Announce Type: cross Abstract: Tool-using LLM agents increasingly rely on external tools to make consequential decisions, yet most existing agent-security benchmarks and defenses im

UbuntuGuard: A Culturally-Grounded Policy Benchmark for Equitable AI Safety in African Languages

arXiv:2601.12696v2 Announce Type: replace Abstract: Current guardian models are predominantly Western-centric and optimized for high-resource languages, leaving low-resource African languages vulnerab

Universal Adversarial Triggers

arXiv:2605.17936v1 Announce Type: new Abstract: Recent works have illustrated that modern NLP models trained for diverse tasks ranging from sentiment analysis to language generation succumb to univers