Peer Revenue Fully funded

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Saturday 13 December

Computation and Language

SCIR: A Self-Correcting Iterative Refinement Framework for Enhanced Information Extraction Based on Schema

Yushen Fang, Jianjun Li, Mingqian Ding et al. (+3 more)

Although Large language Model (LLM)-powered information extraction (IE) systems have shown impressive capabilities, current fine-tuning paradigms face two major limitations: high training costs and difficulties in aligning with LLM preferences. To address these issues, we propose a novel universal IE paradigm, the Self-Correcting Iterative Refinement (SCIR) framework, along with a Multi-task Bilingual (Chinese-English) Self-Correcting (MBSC) dataset containing over 100,000 entries. The SCIR framework achieves plug-and-play compatibility with existing LLMs and IE systems through its Dual-Path Self-Correcting module and feedback-driven optimization, thereby significantly reducing training costs. Concurrently, the MBSC dataset tackles the challenge of preference alignment by indirectly distilling GPT-4's capabilities into IE result detection models. Experimental results demonstrate that SCIR outperforms state-of-the-art IE methods across three key tasks: named entity recognition, relation extraction, and event extraction, achieving a 5.27 percent average improvement in span-based Micro-F1 while reducing training costs by 87 percent compared to baseline approaches. These advancements not only enhance the flexibility and accuracy of IE systems but also pave the way for lightweight and efficient IE paradigms.
Read the full abstract

All authors Yushen Fang, Jianjun Li, Mingqian Ding, Chang Liu, Xinchi Zou, Wenqi Yang

Computation and Language · Cross-listed from cs.CV

From Human Intention to Action Prediction: Intention-Driven End-to-End Autonomous Driving

Huan Zheng, Yucheng Zhou, Tianyi Yan et al. (+9 more)

While end-to-end autonomous driving has achieved remarkable progress in geometric control, current systems remain constrained by a command-following paradigm that relies on simple navigational instructions. Transitioning to genuinely intelligent agents requires the capability to interpret and fulfill high-level, abstract human intentions. However, this advancement is hindered by the lack of dedicated benchmarks and semantic-aware evaluation metrics. In this paper, we formally define the task of Intention-Driven End-to-End Autonomous Driving and present Intention-Drive, a comprehensive benchmark designed to bridge this gap. We construct a large-scale dataset featuring complex natural language intentions paired with high-fidelity sensor data. To overcome the limitations of conventional trajectory-based metrics, we introduce the Imagined Future Alignment (IFA), a novel evaluation protocol leveraging generative world models to assess the semantic fulfillment of human goals beyond mere geometric accuracy. Furthermore, we explore the solution space by proposing two distinct paradigms: an end-to-end vision-language planner and a hierarchical agent-based framework. The experiments reveal a critical dichotomy where existing models exhibit satisfactory driving stability but struggle significantly with intention fulfillment. Notably, the proposed frameworks demonstrate superior alignment with human intentions.
Read the full abstract

All authors Huan Zheng, Yucheng Zhou, Tianyi Yan, Jiayi Su, Hongjun Chen, Dubing Chen, Xingtai Gui, Wencheng Han, Runzhou Tao, Zhongying Qiu, Jianfei Yang, Jianbing Shen

Computation and Language

F5-TTS-RO: Extending F5-TTS to Romanian TTS via Lightweight Input Adaptation

Radu-Gabriel Chivereanu and Tiberiu Boros

This work introduces a lightweight input-level adapter for the F5-TTS model that enables Romanian Language support. To preserve the existing capabilities of the model (voice cloning, English and Chinese support), we keep the original weights frozen, append a sub-network to the model and train it as an extension for the textual embedding matrix of the text encoder. For simplicity, we rely on ConvNeXt module implemented in F5-TTS to also model the co-dependencies between the new character-level embeddings. The module serves as a “soft“ letter-to-sound layer, converting Romanian text into a continuous representation that the F5-TTS model uses to produce naturally sounding Romanian utterances. We evaluate the model with a pool of 20 human listeners across three tasks: (a) audio similarity between reference and generated speech, (b) pronunciation and naturalness and (c) Romanian-English code-switching. The results indicate that our approach maintains voice cloning capabilities and enables, to a certain extent, code-switching within the same utterance; however, residual English accent characteristics remain. We open-source our code and provide example audio samples at https://github.com/racai-ro/Ro-F5TTS.
Read the full abstract

Computation and Language

Market-Bench: Evaluating Large Language Models on Introductory Quantitative Trading and Market Dynamics

Abhay Srivastava, Sam Jung and Spencer Mateega

We introduce MARKET-BENCH, a benchmark that evaluates large language models (LLMs) on introductory quantitative trading tasks by asking them to construct executable backtesters from natural language strategy descriptions and market assumptions. Each instance specifies one of three canonical strategies: scheduled trading on Microsoft (NASDAQ: MSFT), pairs trading on Coca-Cola (NASDAQ: KO) and Pepsi (NASDAQ: PEP), or delta hedging on MSFT. Models must produce code whose profit and loss (P and L), drawdown, and position paths match a verifiable reference implementation. We assess thirteen state-of-the-art models using a multi-round evaluation that separates structural reliability (whether the backtest runs) from numerical accuracy (mean absolute error of the backtest metrics), assigning failed outputs a duplicated-metrics baseline MAE. While most models reliably execute the simplest strategy (average executable passes of 4.08 out of 5 rounds), errors vary by orders of magnitude across models and tasks. Gemini 3 Pro and Claude 4.5 Sonnet combine strong reliability with low error on simpler strategies. GPT-5.2 achieves strong overall performance with perfect executability. GPT-5.1 Codex-Max achieves the lowest best-run error on the easiest task. Qwen3 Max attains perfect executability yet sometimes produces inaccurate profit and loss paths. These results show that current LLMs can scaffold basic trading infrastructure but still struggle to reason robustly about prices, inventory, and risk. We release MARKET-BENCH and a public leaderboard at https://marketbench.ai.
Read the full abstract

Computation and Language

Adversarially Probing Cross-Family Sound Symbolism in 27 Languages

Anika Sharma, Tianyi Niu, Emma Wrenn et al. (+1 more)

The phenomenon of sound symbolism, the non-arbitrary mapping between word sounds and meanings, has long been demonstrated through anecdotal experiments like Bouba Kiki, but rarely tested at scale. We present the first computational cross-linguistic analysis of sound symbolism in the semantic domain of size. We compile a typologically broad dataset of 810 adjectives (27 languages, 30 words each), each phonemically transcribed and validated with native-speaker audio. Using interpretable classifiers over bag-of-segment features, we find that phonological form predicts size semantics above chance even across unrelated languages, with both vowels and consonants contributing. To probe universality beyond genealogy, we train an adversarial scrubber that suppresses language identity while preserving size signal (also at family granularity). Language prediction averaged across languages and settings falls below chance while size prediction remains significantly above chance, indicating cross-family sound-symbolic bias. We release data, code, and diagnostic tools for future large-scale studies of iconicity.
Read the full abstract

All authors Anika Sharma, Tianyi Niu, Emma Wrenn, Shashank Srivastava

Computation and Language

Semantic Distance Measurement based on Multi-Kernel Gaussian Processes

Yinzhu Cheng, Haihua Xie, Yaqing Wang et al. (+2 more)

Semantic distance measurement is a fundamental problem in computational linguistics, providing a quantitative characterization of similarity or relatedness between text segments, and underpinning tasks such as text retrieval and text classification. From a mathematical perspective, a semantic distance can be viewed as a metric defined on a space of texts or on a representation space derived from them. However, most classical semantic distance methods are essentially fixed, making them difficult to adapt to specific data distributions and task requirements. In this paper, a semantic distance measure based on multi-kernel Gaussian processes (MK-GP) was proposed. The latent semantic function associated with texts was modeled as a Gaussian process, with its covariance function given by a combined kernel combining Matérn and polynomial components. The kernel parameters were learned automatically from data under supervision, rather than being hand-crafted. This semantic distance was instantiated and evaluated in the context of fine-grained sentiment classification with large language models under an in-context learning (ICL) setup. The experimental results demonstrated the effectiveness of the proposed measure.
Read the full abstract

All authors Yinzhu Cheng, Haihua Xie, Yaqing Wang, Miao He, Mingming Sun

Computation and Language · Cross-listed from cs.CV

Journey Before Destination: On the importance of Visual Faithfulness in Slow Thinking

Rheeya Uppaal, Phu Mon Htut, Min Bai et al. (+3 more)

Reasoning-augmented vision language models (VLMs) generate explicit chains of thought that promise greater capability and transparency but also introduce new failure modes: models may reach correct answers via visually unfaithful intermediate steps, or reason faithfully yet fail on the final prediction. Standard evaluations that only measure final-answer accuracy cannot distinguish these behaviors. We introduce the visual faithfulness of reasoning chains as a distinct evaluation dimension, focusing on whether the perception steps of a reasoning chain are grounded in the image. We propose a training- and reference-free framework that decomposes chains into perception versus reasoning steps and uses off-the-shelf VLM judges for step-level faithfulness, additionally verifying this approach through a human meta-evaluation. Building on this metric, we present a lightweight self-reflection procedure that detects and locally regenerates unfaithful perception steps without any training. Across multiple reasoning-trained VLMs and perception-heavy benchmarks, our method reduces Unfaithful Perception Rate while preserving final-answer accuracy, improving the reliability of multimodal reasoning.
Read the full abstract

All authors Rheeya Uppaal, Phu Mon Htut, Min Bai, Nikolaos Pappas, Zheng Qi, Sandesh Swamy

Computation and Language · Cross-listed from cs.SE

Training Versatile Coding Agents in Synthetic Environments

Yiqi Zhu, Apurva Gandhi and Graham Neubig

Prior works on training software engineering agents have explored utilizing existing resources such as issues on GitHub repositories to construct software engineering tasks and corresponding test suites. These approaches face two key limitations: (1) their reliance on pre-existing GitHub repositories offers limited flexibility, and (2) their primary focus on issue resolution tasks restricts their applicability to the much wider variety of tasks a software engineer must handle. To overcome these challenges, we introduce SWE-Playground, a novel pipeline for generating environments and trajectories which supports the training of versatile coding agents. Unlike prior efforts, SWE-Playground synthetically generates projects and tasks from scratch with strong language models and agents, eliminating reliance on external data sources. This allows us to tackle a much wider variety of coding tasks, such as reproducing issues by generating unit tests and implementing libraries from scratch. We demonstrate the effectiveness of this approach on three distinct benchmarks, and results indicate that SWE-Playground produces trajectories with dense training signal, enabling agents to reach comparable performance with significantly fewer trajectories than previous works.
Read the full abstract

Computation and Language

Diffusion Language Model Inference with Monte Carlo Tree Search

Zheng Huang, Kiran Ramnath, Yueyan Chen et al. (+8 more)

Diffusion language models (DLMs) have recently emerged as a compelling alternative to autoregressive generation, offering parallel generation and improved global coherence. During inference, DLMs generate text by iteratively denoising masked sequences in parallel; however, determining which positions to unmask and which tokens to commit forms a large combinatorial search problem. Existing inference methods approximate this search using heuristics, which often yield suboptimal decoding paths; other approaches instead rely on additional training to guide token selection. To introduce a principled search mechanism for DLMs inference, we introduce MEDAL, an inference-time scaling framework that integrates Monte Carlo Tree SEarch initialization for Diffusion LAnguage Model inference. We employ Monte Carlo Tree Search at the initialization stage to explore promising unmasking trajectories, providing a robust starting point for subsequent refinement. This design enables efficient inference-time scaling, allowing generation quality to improve as the search budget increases, without additional training. Across multiple benchmarks, MEDAL achieves up to 22.0% improvement over existing inference strategies, establishing a new paradigm for search-based inference in DLMs.
Read the full abstract

All authors Zheng Huang, Kiran Ramnath, Yueyan Chen, Aosong Feng, Sangmin Woo, Balasubramaniam Srinivasan, Zhichao Xu, Kang Zhou, Shuai Wang, Haibo Ding, Lin Lee Cheong

Computation and Language

Extending the Context of Pretrained LLMs by Dropping Their Positional Embeddings

Yoav Gelberg, Koshi Eguchi, Takuya Akiba et al. (+1 more)

So far, expensive finetuning beyond the pretraining sequence length has been a requirement for effectively extending the context of language models (LM). In this work, we break this key bottleneck by Dropping the Positional Embeddings of LMs after training (DroPE). Our simple method is motivated by three key theoretical and empirical observations. First, positional embeddings (PEs) serve a crucial role during pretraining, providing an important inductive bias that significantly facilitates convergence. Second, over-reliance on this explicit positional information is also precisely what prevents test-time generalization to sequences of unseen length, even when using popular PE-scaling methods. Third, positional embeddings are not an inherent requirement of effective language modeling and can be safely removed after pretraining, following a short recalibration phase. Empirically, DroPE yields seamless zero-shot context extension without any long-context finetuning, quickly adapting pretrained LMs without compromising their capabilities in the original training context. Our findings hold across different models and dataset sizes, far outperforming previous specialized architectures and established rotary positional embedding scaling methods.
Read the full abstract

All authors Yoav Gelberg, Koshi Eguchi, Takuya Akiba, Edoardo Cetin

Friday 12 December

Computation and Language

Operation Veja: Fixing Fundamental Concepts Missing from Modern Roleplaying Training Paradigms

Yueze Liu, Ajay Nagi Reddy Kumdam, Ronit Kanjilal et al. (+2 more)

Modern roleplaying models are increasingly sophisticated, yet they consistently struggle to capture the essence of believable, engaging characters. We argue this failure stems from training paradigms that overlook the dynamic interplay of a character's internal world. Current approaches, including Retrieval-Augmented Generation (RAG), fact-based priming, literature-based learning, and synthetic data generation, exhibit recurring limitations in modeling the deliberative, value-conflicted reasoning that defines human interaction. In this paper, we identify four core concepts essential for character authenticity: Values, Experiences, Judgments, and Abilities (VEJA). We propose the VEJA framework as a new paradigm for data curation that addresses these systemic limitations. To illustrate the qualitative ceiling enabled by our framework, we present a pilot study comparing a manually curated, VEJA-grounded dataset against a state-of-the-art synthetic baseline. Using an LLM-as-judge evaluation, our findings demonstrate a significant quality gap, suggesting that a shift toward conceptually grounded data curation, as embodied by VEJA, is necessary for creating roleplaying agents with genuine depth and narrative continuity. The full dataset is available at https://github.com/HyouinKyoumaIRL/Operation-Veja
Read the full abstract

All authors Yueze Liu, Ajay Nagi Reddy Kumdam, Ronit Kanjilal, Hao Yang, Yichi Zhang

Computation and Language

TeleMem: Building Long-Term and Multimodal Memory for Agentic AI

Chunliang Chen, Ming Guan, Xiao Lin et al. (+8 more)

Large language models (LLMs) excel at many NLP tasks but struggle to sustain long-term interactions due to limited attention over extended dialogue histories. Retrieval-augmented generation (RAG) mitigates this issue but lacks reliable mechanisms for updating or refining stored memories, leading to schema-driven hallucinations, inefficient write operations, and minimal support for multimodal reasoning.To address these challenges, we propose TeleMem, a unified long-term and multimodal memory system that maintains coherent user profiles through narrative dynamic extraction, ensuring that only dialogue-grounded information is preserved. TeleMem further introduces a structured writing pipeline that batches, retrieves, clusters, and consolidates memory entries, substantially improving storage efficiency, reducing token usage, and accelerating memory operations. Additionally, a multimodal memory module combined with ReAct-style reasoning equips the system with a closed-loop observe, think, and act process that enables accurate understanding of complex video content in long-term contexts. Experimental results show that TeleMem surpasses the state-of-the-art Mem0 baseline with 19% higher accuracy, 43% fewer tokens, and a 2.1x speedup on the ZH-4O long-term role-play gaming benchmark.
Read the full abstract

All authors Chunliang Chen, Ming Guan, Xiao Lin, Jiaxu Li, Luxi Lin, Qiyi Wang, Xiangyu Chen, Jixiang Luo, Changzhi Sun, Dell Zhang, Xuelong Li

Computation and Language

TrueBrief: Faithful Summarization through Small Language Models

Kumud Lakara, Ruibo Shi and Fran Silavong

Large language models (LLMs) have exhibited remarkable proficiency in generating high-quality text; however, their propensity for producing hallucinations poses a significant challenge for their deployment in security-critical domains. In this work, we present TrueBrief, an end-to-end framework specifically designed to enhance the faithfulness of small LLMs (SLMs) primarily for the task of text summarization through a preference-optimization paradigm. Central to our framework is a data generation module that facilitates controlled hallucination injection to generate synthetic preference data. Our work provides insights into the impact of data quality and model size on preference-based optimization, highlighting the conditions under which these methods are most effective.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

NoveltyRank: A Retrieval-Augmented Framework for Conceptual Novelty Estimation in AI Research

Zhengxu Yan, Han Li and Yuming Feng

The accelerating pace of scientific publication makes it difficult to identify truly original research among incremental work. We propose a framework for estimating the conceptual novelty of research papers by combining semantic representation learning with retrieval-based comparison against prior literature. We model novelty as both a binary classification task (novel vs. non-novel) and a pairwise ranking task (comparative novelty), enabling absolute and relative assessments. Experiments benchmark three model scales, ranging from compact domain-specific encoders to a zero-shot frontier model. Results show that fine-tuned lightweight models outperform larger zero-shot models despite their smaller parameter count, indicating that task-specific supervision matters more than scale for conceptual novelty estimation. We further deploy the best-performing model as an online system for public interaction and real-time novelty scoring.
Read the full abstract

Computation and Language · Cross-listed from cs.CV

VEGAS: Mitigating Hallucinations in Large Vision-Language Models via Vision-Encoder Attention Guided Adaptive Steering

Zihu Wang, Boxun Xu, Yuxuan Xia et al. (+1 more)

Large vision-language models (LVLMs) exhibit impressive ability to jointly reason over visual and textual inputs. However, they often produce outputs that are linguistically fluent but factually inconsistent with the visual evidence, i.e., they hallucinate. Despite growing efforts to mitigate such hallucinations, a key question remains: what form of visual attention can effectively suppress hallucinations during decoding? In this work, we provide a simple answer: the vision encoder's own attention map. We show that LVLMs tend to hallucinate when their final visual-attention maps fail to concentrate on key image objects, whereas the vision encoder's more concentrated attention maps substantially reduce hallucinations. To further investigate the cause, we analyze vision-text conflicts during decoding and find that these conflicts peak in the language model's middle layers. Injecting the vision encoder's attention maps into these layers effectively suppresses hallucinations. Building on these insights, we introduce VEGAS, a simple yet effective inference-time method that integrates the vision encoder's attention maps into the language model's mid-layers and adaptively steers tokens which fail to concentrate on key image objects. Extensive experiments across multiple benchmarks demonstrate that VEGAS consistently achieves state-of-the-art performance in reducing hallucinations.
Read the full abstract

All authors Zihu Wang, Boxun Xu, Yuxuan Xia, Peng Li

Computation and Language

BLASST: Dynamic BLocked Attention Sparsity via Softmax Thresholding

Jiayi Yuan, Cameron Shinn, Kai Xu et al. (+19 more)

The growing demand for long-context inference capabilities in Large Language Models (LLMs) has intensified the computational and memory bottlenecks inherent to the self-attention mechanism. To address this challenge, we introduce BLASST, a drop-in, dynamic sparse attention mechanism that accelerates inference by using only a fixed scalar threshold to skip attention blocks. Our method targets practical inference deployment by removing the barriers to adoption present in existing works. As such, BLASST eliminates training requirements, avoids expensive pre-computation passes, accelerates both prefill and decode across all major attention variants (MHA, GQA, MQA, and MLA), provides optimized support for modern hardware, and easily integrates into existing frameworks. This is achieved by reusing online softmax statistics to identify negligible attention scores, skipping softmax, value block loads, and the subsequent matrix multiplication. We demonstrate the BLASST algorithm by delivering optimized kernels with negligible latency overhead. Our automated threshold calibration procedure reveals a simple inverse relationship between optimal threshold and context length, meaning we require only a single threshold each for prefill and decode per model. Preserving benchmark accuracy, we demonstrate a 1.52x speedup for prefill at 71.9% sparsity and a 1.48x speedup for decode at 73.2% sparsity on modern GPUs.
Read the full abstract

All authors Jiayi Yuan, Cameron Shinn, Kai Xu, Jingze Cui, George Klimiashvili, Guangxuan Xiao, Perkz Zheng, Bo Li, Yuxin Zhou, Zhouhai Ye, Weijie You, Tian Zheng, Dominic Brown, Pengbo Wang, Markus Hoehnerbach, Richard Cai, Julien Demouth, John D. Owens, Xia Hu, Song Han, Timmy Liu, Huizi Mao

Computation and Language

VOYAGER: A Training Free Approach for Generating Diverse Datasets using LLMs

Avinash Amballa, Yashas Malur Saidutta, Chi-Heng Lin et al. (+2 more)

Large language models (LLMs) are increasingly being used to generate synthetic datasets for the evaluation and training of downstream models. However, prior work has noted that such generated data lacks diversity. In this paper, we propose Voyager, a novel principled approach to generate diverse datasets. Our approach is iterative and directly optimizes a mathematical quantity that optimizes the diversity of the dataset using the machinery of determinantal point processes. Furthermore, our approach is training-free, applicable to closed-source models, and scalable. In addition to providing theoretical justification for the working of our method, we also demonstrate through comprehensive experiments that Voyager significantly outperforms popular baseline approaches by providing a 1.5-3 times improvement in diversity.
Read the full abstract

All authors Avinash Amballa, Yashas Malur Saidutta, Chi-Heng Lin, Vivek Kulkarni, Srinivas Chappidi

Computation and Language · Cross-listed from cs.CR

Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoring

Peichun Hua, Hao Li, Shanghao Shi et al. (+2 more)

Large Vision-Language Models (LVLMs) are vulnerable to a growing array of multimodal jailbreak attacks, necessitating defenses that are both generalizable to novel threats and efficient for practical deployment. Many current strategies fall short, either targeting specific attack patterns, which limits generalization, or imposing high computational overhead. While lightweight anomaly-detection methods offer a promising direction, we find that their common one-class design tends to confuse unseen benign inputs with malicious ones, leading to unreliable over-rejection. To address this, we propose Representational Contrastive Scoring (RCS), a framework built on a key insight: the most potent safety signals reside within the LVLM's own internal representations. Our approach inspects the internal geometry of these representations, learning a lightweight projection to maximally separate benign and malicious inputs in safety-critical layers. This enables a simple yet powerful contrastive score that differentiates true malicious intent from mere distribution shift. Our instantiations, MCD (Mahalanobis Contrastive Detection) and KCD (K-nearest Contrastive Detection), achieve state-of-the-art performance on a challenging evaluation protocol designed to test generalization to unseen attack types. This work demonstrates that effective jailbreak detection can be achieved by applying simple, interpretable statistical methods to the internal representations, offering a practical path towards safer LVLM deployment.
Read the full abstract

All authors Peichun Hua, Hao Li, Shanghao Shi, Zhiyuan Yu, Ning Zhang

Computation and Language · Cross-listed from cs.LG

The Instability of Safety: How Random Seeds and Temperature Expose Inconsistent LLM Refusal Behavior

Erik Larsen

Current safety evaluations of large language models rely on single-shot testing, implicitly assuming that model responses are deterministic and representative of the model's safety alignment. We challenge this assumption by investigating the stability of safety refusal decisions across random seeds and temperature settings. Testing four instruction-tuned models from three families (Llama 3.1 8B, Qwen 2.5 7B, Qwen 3 8B, Gemma 3 12B) on 876 harmful prompts across 20 sampling configurations (4 temperatures x 5 seeds), we find that 18-28% of prompts exhibit decision flips (the REFUSE/PARTIAL/COMPLY label changes across configurations), depending on the model. Our Safety Stability Index (SSI) reveals that higher temperatures significantly reduce decision stability (Friedman chi-squared = 396.81, p < 0.001), with mean within-temperature SSI dropping from 0.977 at temperature 0.0 to 0.942 at temperature 1.0. We validate findings across all model families using Claude 3.5 Haiku as a unified external judge, achieving 89.1% inter-judge agreement with the Llama 70B judge on the two models both judges labeled (Cohen's kappa = 0.62). Within each model, prompts with higher compliance rates exhibit lower stability (Spearman rho = -0.47 to -0.70, all p < 0.001), indicating that models "waver" more on borderline requests. These findings demonstrate that single-shot safety evaluations are insufficient for reliable safety assessment and that evaluation protocols must account for stochastic variation in model behavior. For Llama 3.1 8B, single-shot evaluation agrees with multi-sample ground truth only 92.5% of the time when pooling across temperatures (98.7% at greedy to 90.3% at temperature 1.0), and we recommend scaling samples with temperature–one at greedy, three at low temperature, more at higher temperatures (where three reach only ~95%), and ten when pooling–rather than a single flat threshold.
Read the full abstract

2512.12066 · 12 Dec · 17 pages, 7 figures, 9 tables. Code and data: https://github.com/erikl2/safety-refusal-stability . v4: no reported numbers change. The Claude 3.5 Haiku judge ran at temperature 1.0, not 0.0 (Table 9); at t=0.0, at least 9 of Qwen 2.5's 81 flips are judge noise, so t=0.0 flip rates include judge noise (Secs. 4.3.1, 5.2); a flip is now defined as any REFUSE/PARTIAL/COMPLY label change

Abstract PDF $PRMT ×3$ALIGN ×1$LANG ×1$STOC ×1

Computation and Language

Benchmarking Contextual Understanding for In-Car Conversational Systems

Philipp Habicht, Lev Sorokin, Abdullah Saydemir et al. (+2 more)

In-Car Conversational Question Answering (ConvQA) systems significantly enhance user experience by enabling seamless voice interactions. However, assessing their accuracy and reliability remains a challenge. This paper explores the use of Large Language Models (LLMs) alongside advanced prompting techniques and agent-based methods to evaluate the extent to which ConvQA system responses adhere to user utterances. The focus lies on contextual understanding and the ability to provide accurate venue recommendations considering user constraints and situational context. To evaluate utterance-response coherence using an LLM, we synthetically generate user utterances accompanied by correct and modified failure-containing system responses. We use input-output, chain-of-thought, self-consistency prompting, and multi-agent prompting techniques with 13 reasoning and non-reasoning LLMs of varying sizes and providers, including OpenAI, DeepSeek, Mistral AI, and Meta. We evaluate our approach on a case study involving restaurant recommendations. The most substantial improvements occur for small non-reasoning models when applying advanced prompting techniques, particularly multi-agent prompting. However, reasoning models consistently outperform non-reasoning models, with the best performance achieved using single-agent prompting with self-consistency. Notably, DeepSeek-R1 reaches an F1-score of 0.99 at a cost of 0.002 USD per request. Overall, the best balance between effectiveness and cost-time efficiency is reached with the non-reasoning model DeepSeek-V3. Our findings show that LLM-based evaluation offers a scalable and accurate alternative to traditional human evaluation for benchmarking contextual understanding in ConvQA systems.
Read the full abstract

All authors Philipp Habicht, Lev Sorokin, Abdullah Saydemir, Ken E. Friedl, Andrea Stocco

Computation and Language · Cross-listed from cs.CV

Semantic-Drive: Democratizing Long-Tail Data Curation via Open-Vocabulary Grounding and Neuro-Symbolic VLM Consensus

Antonio Guillen-Perez

The development of robust Autonomous Vehicles (AVs) is bottlenecked by the scarcity of "Long-Tail" training data. While fleets collect petabytes of video logs, identifying rare safety-critical events (e.g., erratic jaywalking, construction diversions) remains a manual, cost-prohibitive process. Existing solutions rely on coarse metadata search, which lacks precision, or cloud-based VLMs, which are privacy-invasive and expensive. We introduce Semantic-Drive, a local-first, neuro-symbolic framework for semantic data mining. Our approach decouples perception into two stages: (1) Symbolic Grounding via a real-time open-vocabulary detector (YOLOE) to anchor attention, and (2) Cognitive Analysis via a Reasoning VLM that performs forensic scene analysis. To mitigate hallucination, we implement a "System 2" inference-time alignment strategy, utilizing a multi-model "Judge-Scout" consensus mechanism. Benchmarked on the nuScenes dataset against the Waymo Open Dataset (WOD-E2E) taxonomy, Semantic-Drive achieves a Recall of 0.966 (vs. 0.475 for CLIP) and reduces Risk Assessment Error by 40% ccompared to the best single scout models. The system runs entirely on consumer hardware (NVIDIA RTX 3090), offering a privacy-preserving alternative to the cloud.
Read the full abstract

Computation and Language

Hold Onto That Thought: Assessing KV Cache Compression On Reasoning

Minghui Liu, Aadi Palnitkar, Tahseen Rabbani et al. (+9 more)

Large language models (LLMs) have demonstrated remarkable performance on long-context tasks, but are often bottlenecked by memory constraints. Namely, the KV cache, which is used to significantly speed up attention computations, grows linearly with context length. A suite of compression algorithms has been introduced to alleviate cache growth by evicting unimportant tokens. However, several popular strategies are targeted towards the prefill phase, i.e., processing long prompt context, and their performance is rarely assessed on reasoning tasks requiring long decoding. In particular, short but complex prompts, such as those in benchmarks like GSM8K and MATH500, often benefit from multi-step reasoning and self-reflection, resulting in thinking sequences thousands of tokens long. In this work, we benchmark the performance of several popular compression strategies on long-reasoning tasks. For the non-reasoning Llama-3.1-8B-Instruct, we determine that no singular strategy fits all, and that performance is heavily influenced by dataset type. However, we discover that H2O and our decoding-enabled variant of SnapKV are dominant strategies for reasoning models, indicating the utility of heavy-hitter tracking for reasoning traces. We also find that eviction strategies at low budgets can produce longer reasoning traces, revealing a tradeoff between cache size and inference costs.
Read the full abstract

All authors Minghui Liu, Aadi Palnitkar, Tahseen Rabbani, Hyunwoo Jae, Kyle Rui Sang, Dixi Yao, Shayan Shabihi, Fuheng Zhao, Tian Li, Ce Zhang, Furong Huang, Kunpeng Zhang

Computation and Language

Direct Confidence Alignment: Aligning Verbalized Confidence with Internal Confidence In Large Language Models

Glenn Zhang, Treasure Mayowa, Jason Fan et al. (+4 more)

Producing trustworthy and reliable Large Language Models (LLMs) has become increasingly important as their usage becomes more widespread. Calibration seeks to achieve this by improving the alignment between the model's confidence and the actual likelihood of its responses being correct or desirable. However, it has been observed that the internal confidence of a model, derived from token probabilities, is not well aligned with its verbalized confidence, leading to misleading results with different calibration methods. In this paper, we propose Direct Confidence Alignment (DCA), a method using Direct Preference Optimization to align an LLM's verbalized confidence with its internal confidence rather than ground-truth accuracy, enhancing model transparency and reliability by ensuring closer alignment between the two confidence measures. We evaluate DCA across multiple open-weight LLMs on a wide range of datasets. To further assess this alignment, we also introduce three new calibration error-based metrics. Our results show that DCA improves alignment metrics on certain model architectures, reducing inconsistencies in a model's confidence expression. However, we also show that it can be ineffective on others, highlighting the need for more model-aware approaches in the pursuit of more interpretable and trustworthy LLMs.
Read the full abstract

All authors Glenn Zhang, Treasure Mayowa, Jason Fan, Yicheng Fu, Aaron Sandoval, Sean O'Brien, Kevin Zhu

Computation and Language

SUMFORU: An LLM-Based Review Summarization Framework for Personalized Purchase Decision Support

Yuming Feng and Xinrui Jiang

Online product reviews contain rich but noisy signals that overwhelm users and hinder effective decision-making. Existing LLM-based summarizers remain generic and fail to account for individual preferences, limiting their practical utility. We propose SUMFORU, a steerable review summarization framework that aligns outputs with explicit user personas to support personalized purchase decisions. Our approach integrates a high-quality data pipeline built from the Amazon 2023 Review Dataset with a two-stage alignment procedure: (1) persona-aware Supervised Fine-Tuning (SFT) via asymmetric knowledge distillation, and (2) Reinforcement Learning with AI Feedback (RLAIF) using a preference estimator to capture fine-grained, persona-relevant signals. We evaluate the model across rule-based, LLM-based, and human-centered metrics, demonstrating consistent improvements in consistency, grounding, and preference alignment. Our framework achieves the highest performance across all evaluation settings and generalizes effectively to unseen product categories. Our results highlight the promise of steerable pluralistic alignment for building next-generation personalized decision-support systems.
Read the full abstract

Computation and Language · Cross-listed from cs.HC

From Signal to Turn: Interactional Friction in Modular Speech-to-Speech Pipelines

Tittaya Mairittha, Tanakon Sawanglok, Panuwit Raden et al. (+4 more)

While voice-based AI systems have achieved remarkable generative capabilities, their interactions often feel conversationally broken. This paper examines the interactional friction that emerges in modular Speech-to-Speech Retrieval-Augmented Generation (S2S-RAG) pipelines. By analyzing a representative production system, we move beyond simple latency metrics to identify three recurring patterns of conversational breakdown: (1) Temporal Misalignment, where system delays violate user expectations of conversational rhythm; (2) Expressive Flattening, where the loss of paralinguistic cues leads to literal, inappropriate responses; and (3) Repair Rigidity, where architectural gating prevents users from correcting errors in real-time. Through system-level analysis, we demonstrate that these friction points should not be understood as defects or failures, but as structural consequences of a modular design that prioritizes control over fluidity. We conclude that building natural spoken AI is an infrastructure design challenge, requiring a shift from optimizing isolated components to carefully choreographing the seams between them.
Read the full abstract

All authors Tittaya Mairittha, Tanakon Sawanglok, Panuwit Raden, Jirapast Buntub, Thanapat Warunee, Napat Asawachaisuvikrom, Thanaphum Saiwongin

Computation and Language

Speculative Decoding Speed-of-Light: Optimal Lower Bounds via Branching Random Walks

Sergey Pankratov and Dan Alistarh

Speculative generation has emerged as a promising technique to accelerate inference in large language models (LLMs) by leveraging parallelism to verify multiple draft tokens simultaneously. However, the fundamental limits on the achievable speedup remain poorly understood. In this work, we establish the first “tight” lower bounds on the runtime of any deterministic speculative generation algorithm. This is achieved by drawing a parallel between the token generation process and branching random walks, which allows us to analyze the optimal draft tree selection problem. We prove, under basic assumptions, that the expected number of tokens successfully predicted per speculative iteration is bounded as $\mathbb{E}[X] \leq (\mu + \mu_{(2)})\log(P )/\mu^2 + O(1)$, where $P$ is the verifier's capacity, $\mu$ is the expected entropy of the verifier's output distribution, and $\mu_{(2)}$ is the expected second log-moment. This result provides new insights into the limits of parallel token generation, and could guide the design of future speculative decoding systems. Empirical evaluations on Llama models validate our theoretical predictions, confirming the tightness of our bounds in practical settings.
Read the full abstract

Computation and Language

Automating Historical Insight Extraction from Large-Scale Newspaper Archives via Neural Topic Modeling

Keerthana Murugaraj, Salima Lamsiyah, Marten During et al. (+1 more)

Extracting coherent and human-understandable themes from large collections of unstructured historical newspaper archives presents significant challenges due to topic evolution, Optical Character Recognition (OCR) noise, and the sheer volume of text. Traditional topic-modeling methods, such as Latent Dirichlet Allocation (LDA), often fall short in capturing the complexity and dynamic nature of discourse in historical texts. To address these limitations, we employ BERTopic. This neural topic-modeling approach leverages transformerbased embeddings to extract and classify topics, which, despite its growing popularity, still remains underused in historical research. Our study focuses on articles published between 1955 and 2018, specifically examining discourse on nuclear power and nuclear safety. We analyze various topic distributions across the corpus and trace their temporal evolution to uncover long-term trends and shifts in public discourse. This enables us to more accurately explore patterns in public discourse, including the co-occurrence of themes related to nuclear power and nuclear weapons and their shifts in topic importance over time. Our study demonstrates the scalability and contextual sensitivity of BERTopic as an alternative to traditional approaches, offering richer insights into historical discourses extracted from newspaper archives. These findings contribute to historical, nuclear, and social-science research while reflecting on current limitations and proposing potential directions for future work.
Read the full abstract

All authors Keerthana Murugaraj, Salima Lamsiyah, Marten During, Martin Theobald

Computation and Language

Bounding Hallucinations: Merlin-Arthur Protocols for Mutual-Information Bounds in Language Models

Björn Deiseroth, Max Henning Höth, Kristian Kersting et al. (+1 more)

Retrieval-augmented generation (RAG) relies on retrieved context to guide large language models (LLM), yet treats the retrieval as a heuristic rather than verifiable evidence – leading to unsupported answers, hallucinations, and reliance on spurious context. We introduce a novel evaluation, data augmentation, and training framework that treats the RAG pipeline as an interactive proof system by adapting the Merlin-Arthur (M/A) protocol: Arthur (the generator LLM) receives context of unknown provenance and Merlin gives helpful evidence, while Morgana injects adversarial, misleading context. We implement them both with an XAI method to self-assess and modify evidence most influential to Arthur. Based on this framework we propose the Explained Information Fraction (EIF) score, that disentangles explanation fidelity from model predictive errors and imperfect benchmarks, and normalizes M/A mutual-information lower bounds to realistic empirical settings. When trained with those contexts, Arthur learns to answer when evidence supports the answer and abstain when evidence is insufficient. Across five QA benchmarks and four LLMs (1B to 32B parameters), M/A reduces incorrect answers under insufficient context by up to 35pp during training and by 18-20pp over vanilla finetuning. Abstention emerges even without any manually annotated unanswerable example or preference pair. We improve EIF-cond by 0.1-0.4 during M/A training and by 0.33-0.38 over vanilla finetuning. Reusing Merlin/Morgana contexts as automatic hard positives and negatives also raises retriever Recall@1 by 2pp. While high accuracy does not guarantee entropy flow from context to answer, our EIF scores – to our knowledge, the first context-to-answer information bound for RAG generators – show that autonomous interactive-proof-style supervision enables RAG systems that treat retrieved documents as verifiable evidence.
Read the full abstract

All authors Björn Deiseroth, Max Henning Höth, Kristian Kersting, Letitia Parcalabescu

Computation and Language

Visualizing token importance for black-box language models

Paulius Rauba, Qiyao Wei and Mihaela van der Schaar

We consider the problem of auditing black-box large language models (LLMs) to ensure they behave reliably when deployed in production settings, particularly in high-stakes domains such as legal, medical, and regulatory compliance. Existing approaches for LLM auditing often focus on isolated aspects of model behavior, such as detecting specific biases or evaluating fairness. We are interested in a more general question – can we understand how the outputs of black-box LLMs depend on each input token? There is a critical need to have such tools in real-world applications that rely on inaccessible API endpoints to language models. However, this is a highly non-trivial problem, as LLMs are stochastic functions (i.e. two outputs will be different by chance), while computing prompt-level gradients to approximate input sensitivity is infeasible. To address this, we propose Distribution-Based Sensitivity Analysis (DBSA), a lightweight model-agnostic procedure to evaluate the sensitivity of the output of a language model for each input token, without making any distributional assumptions about the LLM. DBSA is developed as a practical tool for practitioners, enabling quick, plug-and-play visual exploration of LLMs reliance on specific input tokens. Through illustrative examples, we demonstrate how DBSA can enable users to inspect LLM inputs and find sensitivities that may be overlooked by existing LLM interpretability methods.
Read the full abstract

Computation and Language

Extending a Parliamentary Corpus with MPs' Tweets: Automatic Annotation and Evaluation Using MultiParTweet

Mevlüt Bagci, Ali Abusaleh, Daniel Baumartz et al. (+3 more)

Social media serves as a critical medium in modern politics because it both reflects politicians' ideologies and facilitates communication with younger generations. We present MultiParTweet, a multilingual tweet corpus from X that connects politicians' social media discourse with German political corpus GerParCor, thereby enabling comparative analyses between online communication and parliamentary debates. MultiParTweet contains 39 546 tweets, including 19 056 media items. Furthermore, we enriched the annotation with nine text-based models and one vision-language model (VLM) to annotate MultiParTweet with emotion, sentiment, and topic annotations. Moreover, the automated annotations are evaluated against a manually annotated subset. MultiParTweet can be reconstructed using our tool, TTLABTweetCrawler, which provides a framework for collecting data from X. To demonstrate a methodological demonstration, we examine whether the models can predict each other using the outputs of the remaining models. In summary, we provide MultiParTweet, a resource integrating automatic text and media-based annotations validated with human annotations, and TTLABTweetCrawler, a general-purpose X data collection tool. Our analysis shows that the models are mutually predictable. In addition, VLM-based annotation were preferred by human annotators, suggesting that multimodal representations align more with human interpretation.
Read the full abstract

All authors Mevlüt Bagci, Ali Abusaleh, Daniel Baumartz, Giueseppe Abrami, Maxim Konca, Alexander Mehler