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

Thursday 11 December

Computation and Language · Cross-listed from cs.LG

Asynchronous Reasoning: Training-Free Interactive Thinking LLMs

George Yakushev, Nataliia Babina, Masoud Vahid Dastgerdi et al. (+4 more)

Many state-of-the-art LLMs are trained to think before giving their answer. Reasoning can greatly improve language model capabilities, but it also makes them less interactive: given a new input, a model must stop thinking before it can respond. Real-world use cases such as voice-based or embodied assistants require an LLM agent to respond and adapt to additional information in real time, which is incompatible with sequential interactions. In contrast, humans can listen, think, and act asynchronously: we begin thinking about the problem while reading it and continue thinking while formulating the answer. In this work, we augment LLMs capable of reasoning to operate in a similar way without additional training. Our method uses the properties of positional embeddings to enable LLMs built for sequential generation to simultaneously think, listen, and write outputs. We evaluate our approach on math, commonsense, and safety reasoning: it allows models to generate accurate thinking-augmented answers while reducing time to first non-thinking token from minutes to ${\le}$ 5s and the overall delays by up to $12{\times}$.
Read the full abstract

All authors George Yakushev, Nataliia Babina, Masoud Vahid Dastgerdi, Vyacheslav Zhdanovskiy, Denis Kuznedelev, Alina Shutova, Max Ryabinin

Computation and Language · Cross-listed from cs.HC

CompanionCast: Toward Social Collaboration with Multi-Agent Systems in Shared Experiences

Yiyang Wang, Chen Chen, Tica Lin et al. (+4 more)

Shared experiences are fundamental to social connection, yet media consumption is increasingly solitary. While AI companions offer real-time reactions and emotional regulation, existing systems either rely on single-agent designs or lack the social awareness and multi-party interaction required to replicate authentic group dynamics. We present CompanionCast, a general framework for orchestrating multiple specialized AI agents as social collaborators within a live shared context. CompanionCast integrates multimodal event detection, rolling context caching for improved grounding, and spatial audio to enhance co-presence. We validate CompanionCast through sports viewing, a domain with rich dynamics and strong social traditions. Pilot studies with soccer fans demonstrate that CompanionCast significantly improves perceived social presence and emotional sharing compared to solitary viewing. We conclude by discussing implications and open challenges for multi-agent systems as social collaborators in shared experiences.
Read the full abstract

All authors Yiyang Wang, Chen Chen, Tica Lin, Vishnu Raj, Josh Kimball, Alex Cabral, Josiah Hester

Computation and Language

Computational emotion analysis with multimodal LLMs: Current evidence on an emerging methodological opportunity

Hauke Licht

Research increasingly leverages audio-visual materials to analyze emotions in political communication. Multimodal large language models (mLLMs) promise to enable such analyses through in-context learning. However, we lack systematic evidence on whether current mLLMs can reliably measure emotions in real-world political settings. This paper closes this gap by evaluating open- and closed-weights mLLMs available as of early 2026 in video-based emotional arousal measurement using two complementary human-labeled datasets: speech actor recordings created under laboratory conditions and real-world parliamentary debates. I find a critical lab-vs-field performance gap. In videos created under laboratory conditions, the examined mLLMs arousal scores approach human-level reliability. However, in parliamentary debate recordings, all examined models' arousal scores correlate at best moderately with average human ratings. Moreover, in each dataset, all but one of the examined mLLMs exhibit systematic gender-differential bias, consistently underestimating arousal more for male than for female speakers, resulting in a net-positive intensity bias. These findings reveal important limitations of current mLLMs for real-world political video analysis and establish a rigorous evaluation framework for tracking future developments.
Read the full abstract

Computation and Language

Quantifying Emotional Tone in Tolkien's The Hobbit: Dialogue Sentiment Analysis with RegEx, NRC-VAD, and Python

Lilin Qiu

This study analyzes the emotional tone of dialogue in J. R. R. Tolkien's The Hobbit (1937) using computational text analysis. Dialogue was extracted with regular expressions, then preprocessed, and scored using the NRC-VAD lexicon to quantify emotional dimensions. The results show that the dialogue maintains a generally positive (high valence) and calm (low arousal) tone, with a gradually increasing sense of agency (dominance) as the story progresses. These patterns reflect the novel's emotional rhythm: moments of danger and excitement are regularly balanced by humor, camaraderie, and relief. Visualizations – including emotional trajectory graphs and word clouds – highlight how Tolkien's language cycles between tension and comfort. By combining computational tools with literary interpretation, this study demonstrates how digital methods can uncover subtle emotional structures in literature, revealing the steady rhythm and emotional modulation that shape the storytelling in The Hobbit.
Read the full abstract

Computation and Language

LabelFusion: Fusing Large Language Models with Transformer Encoders for Robust Financial News Classification

Michael Schlee, Christoph Weisser, Timo Kivimäki et al. (+2 more)

Financial news plays a central role in shaping investor sentiment and short-term dynamics in commodity markets. Many downstream financial applications, such as commodity price prediction or sentiment modeling, therefore rely on the ability to automatically identify news articles relevant to specific assets. However, obtaining large labeled corpora for financial text classification is costly, and transformer-based classifiers such as RoBERTa often degrade significantly in low-data regimes. Our results show that appropriately prompted out-of-the-box Large Language Models (LLMs) achieve strong performance even in such settings. Furthermore, we propose LabelFusion, a hybrid architecture that combines the output of a prompt-engineered LLM with contextual embeddings produced by a fine-tuned RoBERTa encoder through a lightweight Multilayer Perceptron (MLP) voting layer. Evaluated on a ten-class multi-label subset of the Reuters-21578 corpus, LabelFusion achieves a macro F1 score of 96.0% and an accuracy of 92.3% when trained on the full dataset, outperforming both standalone RoBERTa (F1 94.6%) and the standalone LLM (F1 93.9%). In low- to mid-data regimes, however, the LLM alone proves surprisingly competitive, achieving an F1 score of 75.9% even in a zero-shot setting and consistently outperforming LabelFusion until approximately 80% of the training data is available. These results suggest that LLM-only prompting is the preferred strategy under annotation constraints, whereas LabelFusion becomes the most effective solution once sufficient labeled data is available to train the encoder component. The code is available in an anonymized repository.
Read the full abstract

All authors Michael Schlee, Christoph Weisser, Timo Kivimäki, Melchizedek Mashiku, Benjamin Saefken

Computation and Language

The FACTS Leaderboard: A Comprehensive Benchmark for Large Language Model Factuality

Aileen Cheng, Alon Jacovi, Amir Globerson et al. (+62 more)

We introduce The FACTS Leaderboard, an online leaderboard suite and associated set of benchmarks that comprehensively evaluates the ability of language models to generate factually accurate text across diverse scenarios. The suite provides a holistic measure of factuality by aggregating the performance of models on four distinct sub-leaderboards: (1) FACTS Multimodal, which measures the factuality of responses to image-based questions; (2) FACTS Parametric, which assesses models' world knowledge by answering closed-book factoid questions from internal parameters; (3) FACTS Search, which evaluates factuality in information-seeking scenarios, where the model must use a search API; and (4) FACTS Grounding (v2), which evaluates whether long-form responses are grounded in provided documents, featuring significantly improved judge models. Each sub-leaderboard employs automated judge models to score model responses, and the final suite score is an average of the four components, designed to provide a robust and balanced assessment of a model's overall factuality. The FACTS Leaderboard Suite will be actively maintained, containing both public and private splits to allow for external participation while guarding its integrity. It can be found at https://www.kaggle.com/benchmarks/google/facts .
Read the full abstract

All authors Aileen Cheng, Alon Jacovi, Amir Globerson, Ben Golan, Charles Kwong, Chris Alberti, Connie Tao, Eyal Ben-David, Gaurav Singh Tomar, Lukas Haas, Yonatan Bitton, Adam Bloniarz, Aijun Bai, Andrew Wang, Anfal Siddiqui, Arturo Bajuelos Castillo, Aviel Atias, Chang Liu, Corey Fry, Daniel Balle, Deepanway Ghosal, Doron Kukliansky, Dror Marcus, Elena Gribovskaya, Eran Ofek, Honglei Zhuang, Itay Laish, Jan Ackermann, Lily Wang, Meg Risdal, Megan Barnes, Michael Fink, Mohamed Amin, Moran Ambar, Natan Potikha, Nikita Gupta, Nitzan Katz, Noam Velan, Ofir Roval, Ori Ram, Polina Zablotskaia, Prathamesh Bang, Priyanka Agrawal, Rakesh Ghiya, Sanjay Ganapathy, Simon Baumgartner, Sofia Erell, Sushant Prakash, Thibault Sellam, Vikram Rao, Xuanhui Wang, Yaroslav Akulov, Yulong Yang, Zhen Yang, Zhixin Lai, Zhongru Wu, Anca Dragan, Avinatan Hassidim, Fernando Pereira, Slav Petrov, Srinivasan Venkatachary, Tulsee Doshi, Yossi Matias, Sasha Goldshtein, Dipanjan Das

Computation and Language · Cross-listed from cs.AI

Replace, Don't Expand: Mitigating Context Dilution in Multi-Hop RAG via Fixed-Budget Evidence Assembly

Moshe Lahmy and Roi Yozevitch

Retrieval-Augmented Generation (RAG) systems often fail on multi-hop queries when the initial retrieval misses a bridge fact. Prior corrective approaches, such as Self-RAG, CRAG, and Adaptive-$k$, typically address this by adding more context or pruning existing lists. However, simply expanding the context window often leads to context dilution, where distractors crowd out relevant information. We propose SEAL-RAG, a training-free controller that adopts a “replace, don't expand” strategy to fight context dilution under a fixed retrieval depth $k$. SEAL executes a (Search $\rightarrow$ Extract $\rightarrow$ Assess $\rightarrow$ Loop) cycle: it performs on-the-fly, entity-anchored extraction to build a live gap specification (missing entities/relations), triggers targeted micro-queries, and uses entity-first ranking to actively swap out distractors for gap-closing evidence. We evaluate SEAL-RAG against faithful re-implementations of Basic RAG, CRAG, Self-RAG, and Adaptive-$k$ in a shared environment on HotpotQA and 2WikiMultiHopQA. On HotpotQA ($k=3$), SEAL improves answer correctness by +3–13 pp and evidence precision by +12–18 pp over Self-RAG. On 2WikiMultiHopQA ($k=5$), it outperforms Adaptive-$k$ by +8.0 pp in accuracy and maintains 96% evidence precision compared to 22% for CRAG. These gains are statistically significant ($p<0.001$). By enforcing fixed-$k$ replacement, SEAL yields a predictable cost profile while ensuring the top-$k$ slots are optimized for precision rather than mere breadth. We release our code and data at https://github.com/mosherino/SEAL-RAG.
Read the full abstract

Computation and Language

Script Gap: Evaluating LLM Triage on Indian Languages in Native vs Romanized Scripts in a Real World Setting

Manurag Khullar, Utkarsh Desai, Poorva Malviya et al. (+2 more)

Large Language Models (LLMs) are increasingly deployed in high-stakes clinical applications in India. Speakers of Indian languages frequently communicate using romanized text rather than native scripts, yet existing research rarely quantifies or evaluates this orthographic variation in real world applications. We investigate how romanization impacts the reliability of LLMs in a critical domain: maternal and newborn healthcare triage. We benchmark leading LLMs on a real world dataset of user-generated health queries spanning five Indian languages and Nepali. Our results reveal consistent degradation in performance for romanized messages, with gap reaching up to 24 points across languages and models. We propose and evaluate an Uncertainty-based Selective Routing method to close this script gap. At our partner maternal health organization alone, this gap could cause nearly 2 million excess errors in triage. Our findings highlight a critical safety blind spot in LLM-based health systems: models that appear to understand romanized input may still fail to act on it reliably.
Read the full abstract

All authors Manurag Khullar, Utkarsh Desai, Poorva Malviya, Aman Dalmia, Zheyuan Ryan Shi

Computation and Language

Outgrowing the Curse of Multilinguality: Scaling Strategies for Efficient Language Adaptation

Kevin Glocker, Kätriin Kukk, Romina Oji et al. (+3 more)

Training high-performing language models which include medium- and lower-resource languages remains a challenge. Massively multilingual models still underperform language-specific adaptations, especially at smaller scales, due to the curse of multilinguality. In this work, we investigate scaling as an efficient strategy for adapting pretrained models to new target languages. Through comprehensive ablations matched for data and training FLOPs, we test whether upscaling from English base models enables more effective and resource-efficient adaptation than standard continued pretraining. We find that upscaling is an efficient strategy for improving target-language performance by relieving capacity constraints, especially for higher-resourced languages. An initial upscaling step in English is beneficial for low-resource languages. However, in higher-resource scenarios it is more compute-efficient to directly scale up on the target language, albeit slightly compromising data efficiency for most languages. For multilingual continued pre-training, we identify directly scaling up using the target data as the optimal setup for both data efficiency and reduced English forgetting.
Read the full abstract

All authors Kevin Glocker, Kätriin Kukk, Romina Oji, Marcel Bollmann, Marco Kuhlmann, Jenny Kunz

Computation and Language

OPV: Outcome-based Process Verifier for Efficient Long Chain-of-Thought Verification

Zijian Wu, Lingkai Kong, Wenwei Zhang et al. (+12 more)

Large language models (LLMs) have achieved significant progress in solving complex reasoning tasks by Reinforcement Learning with Verifiable Rewards (RLVR). This advancement is also inseparable from the oversight automated by reliable verifiers. However, current outcome-based verifiers (OVs) are unable to inspect the unreliable intermediate steps in the long reasoning chains of thought (CoTs). Meanwhile, current process-based verifiers (PVs) have difficulties in reliably detecting errors in the complex long CoTs, limited by the scarcity of high-quality annotations due to the prohibitive costs of human annotations. Therefore, we propose the Outcome-based Process Verifier (OPV), which verifies the rationale process of summarized outcomes from long CoTs to achieve both accurate and efficient verification and enable large-scale annotation. To empower the proposed verifier, we adopt an iterative active learning framework with expert annotations to progressively improve the verification capability of OPV with fewer annotation costs. Specifically, in each iteration, the most uncertain cases of the current best OPV are annotated and then subsequently used to train a new OPV through Rejection Fine-Tuning (RFT) and RLVR for the next round. Extensive experiments demonstrate OPV's superior performance and broad applicability. It achieves new state-of-the-art results on our held-out OPV-Bench, outperforming much larger open-source models such as Qwen3-Max-Preview with an F1 score of 83.1 compared to 76.3. Furthermore, OPV effectively detects false positives within synthetic dataset, closely align with expert assessment. When collaborating with policy models, OPV consistently yields performance gains, e.g., raising the accuracy of DeepSeek-R1-Distill-Qwen-32B from 55.2% to 73.3% on AIME2025 as the compute budget scales.
Read the full abstract

All authors Zijian Wu, Lingkai Kong, Wenwei Zhang, Songyang Gao, Yuzhe Gu, Zhongrui Cai, Tianyou Ma, Yuhong Liu, Zhi Wang, Runyuan Ma, Guangyu Wang, Wei Li, Conghui He, Dahua Lin, Kai Chen

Computation and Language

TRIDENT: A Redundant Architecture for Caribbean-Accented Emergency Speech Triage

Elroy Galbraith, Chadwick Sutherland and Donahue Morgan

Emergency speech recognition systems exhibit systematic performance degradation on non-standard English varieties, creating a critical gap in services for Caribbean populations. We present TRIDENT (Transcription and Routing Intelligence for Dispatcher-Empowered National Triage), a three-layer dispatcher-support architecture designed to structure emergency call inputs for human application of established triage protocols (the ESI for routine operations and START for mass casualty events), even when automatic speech recognition fails. The system combines Caribbean-accent-tuned ASR, local entity extraction via large language models, and bio-acoustic distress detection to provide dispatchers with three complementary signals: transcription confidence, structured clinical entities, and vocal stress indicators. Our key insight is that low ASR confidence, rather than representing system failure, serves as a valuable queue prioritization signal – particularly when combined with elevated vocal distress markers indicating a caller in crisis whose speech may have shifted toward basilectal registers. A complementary insight drives the entity extraction layer: trained responders and composed bystanders may report life-threatening emergencies without elevated vocal stress, requiring semantic analysis to capture clinical indicators that paralinguistic features miss. We describe the architectural design, theoretical grounding in psycholinguistic research on stress-induced code-switching, and deployment considerations for offline operation during disaster scenarios. This work establishes a framework for accent-resilient emergency AI that ensures Caribbean voices receive equitable access to established national triage protocols. Empirical validation on Caribbean emergency calls remains future work.
Read the full abstract

Computation and Language

Intern-S1-MO: Long-horizon Reasoning Agent for Olympiad?Level Mathematical Problem Solving

Yuzhe Gu, Songyang Gao, Zijian Wu et al. (+18 more)

Large Reasoning Models (LRMs) have expanded the mathematical reasoning frontier through Chain-of-Thought (CoT) techniques and Reinforcement Learning with Verifiable Rewards (RLVR), capable of solving AIME-level problems. However, the performance of LRMs is heavily dependent on the extended reasoning context length. For solving ultra-hard problems like those in the International Mathematical Olympiad (IMO), the required reasoning complexity surpasses the space that an LRM can explore in a single round. Previous works attempt to extend the reasoning context of LRMs but remain prompt-based and built upon proprietary models, lacking systematic structures and training pipelines. Therefore, this paper introduces Intern-S1-MO, a long-horizon math agent that conducts multi-round hierarchical reasoning, composed of an LRM-based multi-agent system including reasoning, summary, and verification. By maintaining a compact memory in the form of lemmas, Intern-S1-MO can more freely explore the lemma-rich reasoning spaces in multiple reasoning stages, thereby breaking through the context constraints for IMO-level math problems. Furthermore, we propose OREAL-H, an RL framework for training the LRM using the online explored trajectories to simultaneously bootstrap the reasoning ability of LRM and elevate the overall performance of Intern-S1-MO. Experiments show that Intern-S1-MO can obtain 26 out of 35 points on the non-geometry problems of IMO2025, matching the performance of silver medalists. It also surpasses the current advanced LRMs on inference benchmarks such as HMMT2025, AIME2025, and CNMO2025. In addition, our agent officially participates in CMO2025 and achieves a score of 102/126 under the judgment of human experts, reaching the gold medal level.
Read the full abstract

All authors Yuzhe Gu, Songyang Gao, Zijian Wu, Lingkai Kong, Wenwei Zhang, Zhongrui Cai, Fan Zheng, Tianyou Ma, Junhao Shen, Haiteng Zhao, Duanyang Zhang, Huilun Zhang, Kuikun Liu, Chengqi Lyu, Yanhui Duan, Chiyu Chen, Ningsheng Ma, Jianfei Gao, Han Lyu, Dahua Lin, Kai Chen

Computation and Language

Textual Data Bias Detection and Mitigation – An Extensible Pipeline with Experimental Evaluation

Rebekka Görge, Sujan Sai Gannamaneni, Tabea Naeven et al. (+10 more)

Textual data used to train large language models (LLMs) exhibits multifaceted bias manifestations encompassing harmful language and skewed demographic distributions. Regulations such as the European AI Act require identifying and mitigating biases against protected groups in data, with the ultimate goal of preventing unfair model outputs. However, practical guidance and operationalization are lacking. We propose a comprehensive data bias detection and mitigation pipeline comprising four components that address two data bias types, namely representation bias and (explicit) stereotypes for a configurable sensitive attribute. First, we leverage LLM-generated word lists created based on quality criteria to detect relevant group labels. Second, representation bias is quantified using the Demographic Representation Score. Third, we detect and mitigate stereotypes using sociolinguistically informed filtering. Finally, we compensate representation bias through Grammar- and Context-Aware Counterfactual Data Augmentation. We conduct a two-fold evaluation using the examples of gender, religion and age. First, the effectiveness of each individual component on data debiasing is evaluated through human validation and baseline comparison. The findings demonstrate that we successfully reduce representation bias and (explicit) stereotypes in a text dataset. Second, the effect of data debiasing on model bias reduction is evaluated by bias benchmarking of several models (0.6B-8B parameters), fine-tuned on the debiased text dataset. This evaluation reveals that LLMs fine-tuned on debiased data do not consistently show improved performance on bias benchmarks, exposing critical gaps in current evaluation methodologies and highlighting the need for targeted data manipulation to address manifested model bias.
Read the full abstract

All authors Rebekka Görge, Sujan Sai Gannamaneni, Tabea Naeven, Hammam Abdelwahab, Héctor Allende-Cid, Armin B. Cremers, Lennard Helmer, Michael Mock, Anna Schmitz, Songkai Xue, Elif Yildirir, Maximilian Poretschkin, Stefan Wrobel

Computation and Language · Cross-listed from cs.AI

Remember Me, Refine Me: A Dynamic Procedural Memory Framework for Experience-Driven Agent Evolution

Zouying Cao, Jiaji Deng, Li Yu et al. (+4 more)

Procedural memory enables large language model (LLM) agents to internalize "how-to" knowledge, theoretically reducing redundant trial-and-error. However, existing frameworks predominantly suffer from a "passive accumulation" paradigm, treating memory as a static append-only archive. To bridge the gap between static storage and dynamic reasoning, we propose $\textbf{ReMe}$ ($\textit{Remember Me, Refine Me}$), a comprehensive framework for experience-driven agent evolution. ReMe innovates across the memory lifecycle via three mechanisms: 1) $\textit{multi-faceted distillation}$, which extracts fine-grained experiences by recognizing success patterns, analyzing failure triggers and generating comparative insights; 2) $\textit{context-adaptive reuse}$, which tailors historical insights to new contexts via scenario-aware indexing; and 3) $\textit{utility-based refinement}$, which autonomously adds valid memories and prunes outdated ones to maintain a compact, high-quality experience pool. Extensive experiments on BFCL-V3 and AppWorld demonstrate that ReMe establishes a new state-of-the-art in agent memory system. Crucially, we observe a significant memory-scaling effect: Qwen3-8B equipped with ReMe outperforms larger, memoryless Qwen3-14B, suggesting that self-evolving memory provides a computation-efficient pathway for lifelong learning. We release our code and the $\texttt{reme.library}$ dataset to facilitate further research.
Read the full abstract

All authors Zouying Cao, Jiaji Deng, Li Yu, Weikang Zhou, Zhaoyang Liu, Bolin Ding, Hai Zhao

Computation and Language

From Data Scarcity to Data Care: Reimagining Language Technologies for Serbian and other Low-Resource Languages

Smiljana Antonijevic Ubois

Large language models are commonly trained on dominant languages like English, and their representation of low resource languages typically reflects cultural and linguistic biases present in the source language materials. Using the Serbian language as a case, this study examines the structural, historical, and sociotechnical factors shaping language technology development for low resource languages in the AI age. Drawing on semi structured interviews with ten scholars and practitioners, including linguists, digital humanists, and AI developers, it traces challenges rooted in historical destruction of Serbian textual heritage, intensified by contemporary issues that drive reductive, engineering first approaches prioritizing functionality over linguistic nuance. These include superficial transliteration, reliance on English-trained models, data bias, and dataset curation lacking cultural specificity. To address these challenges, the study proposes Data Care, a framework grounded in CARE principles (Collective Benefit, Authority to Control, Responsibility, and Ethics), that reframes bias mitigation from a post hoc technical fix to an integral component of corpus design, annotation, and governance, and positions Data Care as a replicable model for building inclusive, sustainable, and culturally grounded language technologies in contexts where traditional LLM development reproduces existing power imbalances and cultural blind spots.
Read the full abstract

Computation and Language

AgriGPT-Omni: A Unified Speech-Vision-Text Framework for Multilingual Agricultural Intelligence

Bo Yang, Lanfei Feng, Yunkui Chen et al. (+5 more)

Despite rapid advances in multimodal large language models, agricultural applications remain constrained by the lack of multilingual speech data, unified multimodal architectures, and comprehensive evaluation benchmarks. To address these challenges, we present AgriGPT-Omni, an agricultural omni-framework that integrates speech, vision, and text in a unified framework. First, we construct a scalable data synthesis and collection pipeline that converts agricultural texts and images into training data, resulting in the largest agricultural speech dataset to date, including 492K synthetic and 1.4K real speech samples across six languages. Second, based on this, we train the first agricultural omni-model via a three-stage paradigm: textual knowledge injection, progressive multimodal alignment, and GRPO-based reinforcement learning, enabling unified reasoning across languages and modalities. Third, we propose AgriBench-Omni-2K, the first tri-modal benchmark for agriculture, covering diverse speech-vision-text tasks and multilingual slices, with standardized protocols and reproducible tools. Experiments show that AgriGPT-Omni significantly outperforms general-purpose baselines on multilingual and multimodal reasoning as well as real-world speech understanding. All models, data, benchmarks, and code will be released to promote reproducible research, inclusive agricultural intelligence, and sustainable AI development for low-resource regions.
Read the full abstract

All authors Bo Yang, Lanfei Feng, Yunkui Chen, Yu Zhang, Jianyu Zhang, Xiao Xu, Nueraili Aierken, Shijian Li

Computation and Language

RoleRMBench & RoleRM: Towards Reward Modeling for Profile-Based Role Play in Dialogue Systems

Hang Ding, Qiming Feng, Dongqi Liu et al. (+9 more)

Reward modeling has become a cornerstone of aligning large language models (LLMs) with human preferences. Yet, when extended to subjective and open-ended domains such as role play, existing reward models exhibit severe degradation, struggling to capture nuanced and persona-grounded human judgments. To address this gap, we introduce RoleRMBench, the first systematic benchmark for reward modeling in role-playing dialogue, covering seven fine-grained capabilities from narrative management to role consistency and engagement. Evaluation on RoleRMBench reveals large and consistent gaps between general-purpose reward models and human judgment, particularly in narrative and stylistic dimensions. We further propose RoleRM, a reward model trained with Continuous Implicit Preferences (CIP), which reformulates subjective evaluation as continuous consistent pairwise supervision under multiple structuring strategies. Comprehensive experiments show that RoleRM surpasses strong open- and closed-source reward models by over 24% on average, demonstrating substantial gains in narrative coherence and stylistic fidelity. Our findings highlight the importance of continuous preference representation and annotation consistency, establishing a foundation for subjective alignment in human-centered dialogue systems.
Read the full abstract

All authors Hang Ding, Qiming Feng, Dongqi Liu, Qi Zhao, Tao Yao, Shuo Wang, Dongsheng Chen, Jian Li, Zhenye Gan, Jiangning Zhang, Chengjie Wang, Yabiao Wang

Computation and Language

Causal Reasoning Favors Encoders: On The Limits of Decoder-Only Models

Amartya Roy, Elamparithy M, Kripabandhu Ghosh et al. (+2 more)

In context learning (ICL) underpins recent advances in large language models (LLMs), although its role and performance in causal reasoning remains unclear. Causal reasoning demands multihop composition and strict conjunctive control, and reliance on spurious lexical relations of the input could provide misleading results. We hypothesize that, due to their ability to project the input into a latent space, encoder and encoder decoder architectures are better suited for said multihop conjunctive reasoning versus decoder only models. To do this, we compare fine-tuned versions of all the aforementioned architectures with zero and few shot ICL in both natural language and non natural language scenarios. We find that ICL alone is insufficient for reliable causal reasoning, often overfocusing on irrelevant input features. In particular, decoder only models are noticeably brittle to distributional shifts, while finetuned encoder and encoder decoder models can generalize more robustly across our tests, including the non natural language split. Both architectures are only matched or surpassed by decoder only architectures at large scales. We conclude by noting that for cost effective, short horizon robust causal reasoning, encoder or encoder decoder architectures with targeted finetuning are preferable.
Read the full abstract

All authors Amartya Roy, Elamparithy M, Kripabandhu Ghosh, Ponnurangam Kumaraguru, Adrian de Wynter

Computation and Language

XDoGE: Multilingual Data Reweighting to Enhance Language Inclusivity in LLMs

Iñaki Lacunza, José Javier Saiz, Alexander Shvets et al. (+2 more)

Current large language models (LLMs) are trained on massive amounts of text data, primarily from a few dominant languages. Studies suggest that this over-reliance on high-resource languages, such as English, hampers LLM performance in mid- and low-resource languages. To mitigate this problem, we propose to (i) optimize the language distribution by training a small proxy model within a domain-reweighing DoGE algorithm that we extend to XDoGE for a multilingual setup, and (ii) rescale the data and train a full-size model with the established language weights either from scratch or within a continual pre-training phase (CPT). We target six languages possessing a variety of geographic and intra- and inter-language-family relations, namely, English and Spanish (high-resource), Portuguese and Catalan (mid-resource), Galician and Basque (low-resource). We experiment with Salamandra-2b, which is a promising model for these languages. We investigate the effects of substantial data repetition on minor languages and under-sampling on dominant languages using the IberoBench framework for quantitative evaluation. Finally, we release a new promising IberianLLM-7B-Instruct model centering on Iberian languages and English that we pretrained from scratch and further improved using CPT with the XDoGE weights.
Read the full abstract

All authors Iñaki Lacunza, José Javier Saiz, Alexander Shvets, Aitor Gonzalez-Agirre, Marta Villegas

Computation and Language

Grammaticality Judgments in Humans and Language Models: Revisiting Generative Grammar with LLMs

Lars G.B. Johnsen

What counts as evidence for syntactic structure? In traditional generative grammar, systematic contrasts in grammaticality such as subject-auxiliary inversion and the licensing of parasitic gaps are taken as evidence for an internal, hierarchical grammar. In this paper, we test whether large language models (LLMs), trained only on surface forms, reproduce these contrasts in ways that imply an underlying structural representation. We focus on two classic constructions: subject-auxiliary inversion (testing recognition of the subject boundary) and parasitic gap licensing (testing abstract dependency structure). We evaluate models including GPT-4 and LLaMA-3 using prompts eliciting acceptability ratings. Results show that LLMs reliably distinguish between grammatical and ungrammatical variants in both constructions, and as such support that they are sensitive to structure and not just linear order. Structural generalizations, distinct from cognitive knowledge, emerge from predictive training on surface forms, suggesting functional sensitivity to syntax without explicit encoding.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

When Reject Turns into Accept: Quantifying the Vulnerability of LLM-Based Scientific Reviewers to Indirect Prompt Injection

Devanshu Sahoo, Manish Prasad, Vasudev Majhi et al. (+5 more)

Driven by surging submission volumes, scientific peer review has catalyzed two parallel trends: individual over-reliance on LLMs and institutional AI-powered assessment systems. This study investigates the robustness of "LLM-as-a-Judge" systems to adversarial PDF manipulation via invisible text injections and layout aware encoding attacks. We specifically target the distinct incentive of flipping "Reject" decisions to "Accept," a vulnerability that fundamentally compromises scientific integrity. To measure this, we introduce the Weighted Adversarial Vulnerability Score (WAVS), a novel metric that quantifies susceptibility by weighting score inflation against the severity of decision shifts relative to ground truth. We adapt 15 domain-specific attack strategies, ranging from semantic persuasion to cognitive obfuscation, and evaluate them across 13 diverse language models (including GPT-5 and DeepSeek) using a curated dataset of 200 official and real-world accepted and rejected submissions (e.g., ICLR OpenReview). Our results demonstrate that obfuscation techniques like "Maximum Mark Magyk" and "Symbolic Masking & Context Redirection" successfully manipulate scores, achieving decision flip rates of up to 86.26% in open-source models, while exposing distinct "reasoning traps" in proprietary systems. We release our complete dataset and injection framework to facilitate further research on the topic (https://anonymous.4open.sciencer/llm-jailbreak-FC9E/).
Read the full abstract

All authors Devanshu Sahoo, Manish Prasad, Vasudev Majhi, Jahnvi Singh, Vinay Chamola, Yash Sinha, Murari Mandal, Dhruv Kumar

Computation and Language

Decoding Student Minds: Leveraging Conversational Agents for Psychological and Learning Analysis

Nour El Houda Ben Chaabene, Hamza Hammami and Laid Kahloul

This paper presents a psychologically-aware conversational agent designed to enhance both learning performance and emotional well-being in educational settings. The system combines Large Language Models (LLMs), a knowledge graph-enhanced BERT (KG-BERT), and a bidirectional Long Short-Term Memory (LSTM) network with attention to classify students' cognitive and affective states in real time. Unlike prior chatbots limited to either tutoring or affective support, our approach leverages multimodal data-including textual semantics, prosodic speech features, and temporal behavioral trends-to infer engagement, stress, and conceptual understanding. A pilot study with 45 university students demonstrated improved motivation, reduced stress, and moderate academic gains compared to unimodal baselines. We explicitly discuss the exploratory nature of this small-sample pilot, report effect sizes and inter-rater reliability alongside significance tests, and provide an ablation analysis isolating the contribution of the knowledge-graph component and of each modality. These results underline the promise-while acknowledging the current limits in scale and modality balance-of integrating semantic reasoning, multimodal fusion, and temporal modeling to support adaptive, student-centered educational interventions.
Read the full abstract

Computation and Language

Enhancing Next-Generation Language Models with Knowledge Graphs: Extending Claude, Mistral IA, and GPT-4 via KG-BERT

Nour El Houda Ben Chaabene and Hamza Hammami

Large language models (LLMs) like Claude, Mistral IA, and GPT-4 excel in NLP but lack structured knowledge, leading to factual inconsistencies. We address this by integrating Knowledge Graphs (KGs) via KG-BERT to enhance grounding and reasoning. Experiments show significant gains in knowledge-intensive tasks such as question answering and entity linking. This approach improves factual reliability and enables more context-aware next-generation LLMs.
Read the full abstract

2512.10440 · 11 Dec · This paper was accepted and scheduled for inclusion in the ICALT 2025 proceedings but was ultimately not published due to absence from the conference presentation. It appears in the official program booklet. Conference: 2025 IEEE International Conference on Advanced Learning Technologies (ICALT)

Abstract PDF $LANG ×2$RSNG ×1

Computation and Language

Semantic Reconstruction of Adversarial Plagiarism: A Context-Aware Framework for Detecting and Restoring "Tortured Phrases" in Scientific Literature

Agniva Maiti, Prajwal Panth and Suresh Chandra Satapathy

The integrity and reliability of scientific literature is facing a serious threat by adversarial text generation techniques, specifically from the use of automated paraphrasing tools to mask plagiarism. These tools generate "tortured phrases", statistically improbable synonyms (e.g. "counterfeit consciousness" for "artificial intelligence"), that preserve the local grammar while obscuring the original source. Most existing detection methods depend heavily on static blocklists or general-domain language models, which suffer from high false-negative rates for novel obfuscations and cannot determine the source of the plagiarized content. In this paper, we propose Semantic Reconstruction of Adversarial Plagiarism (SRAP), a framework designed not only to detect these anomalies but to mathematically recover the original terminology. We use a two-stage architecture: (1) statistical anomaly detection with a domain-specific masked language model (SciBERT) using token-level pseudo-perplexity, and (2) source-based semantic reconstruction using dense vector retrieval (FAISS) and sentence-level alignment (SBERT). Experiments on a parallel corpus of adversarial scientific text show that while zero-shot baselines fail completely (0.00 percent restoration accuracy), our retrieval-augmented approach achieves 23.67 percent restoration accuracy, significantly outperforming baseline methods. We also show that static decision boundaries are necessary for robust detection in jargon-heavy scientific text, since dynamic thresholding fails under high variance. SRAP enables forensic analysis by linking obfuscated expressions back to their most probable source documents.
Read the full abstract

Computation and Language

T-pro 2.0: An Efficient Russian Hybrid-Reasoning Model and Playground

Dmitrii Stoianov, Danil Taranets, Olga Tsymboi et al. (+12 more)

We introduce T-pro 2.0, an open-weight Russian LLM for hybrid reasoning and efficient inference. The model supports direct answering and reasoning-trace generation, using a Cyrillic-dense tokenizer and an adapted EAGLE speculative-decoding pipeline to reduce latency. To enable reproducible and extensible research, we release the model weights, the T-Wix 500k instruction corpus, the T-Math reasoning benchmark, and the EAGLE weights on Hugging Face. These resources allow users to study Russian-language reasoning and to extend or adapt both the model and the inference pipeline. A public web demo exposes reasoning and non-reasoning modes and illustrates the speedups achieved by our inference stack across domains. T-pro 2.0 thus serves as an accessible open system for building and evaluating efficient, practical Russian LLM applications.
Read the full abstract

All authors Dmitrii Stoianov, Danil Taranets, Olga Tsymboi, Ramil Latypov, Almaz Dautov, Vladislav Kruglikov, Nikita Surkov, German Abramov, Pavel Gein, Dmitry Abulkhanov, Mikhail Gashkov, Viktor Zelenkovskiy, Artem Batalov, Aleksandr Medvedev, Anatolii Potapov

Computation and Language

Cooperative Retrieval-Augmented Generation for Question Answering: Mutual Information Exchange and Ranking by Contrasting Layers

Youmin Ko, Sungjong Seo and Hyunjoon Kim

Since large language models (LLMs) have a tendency to generate factually inaccurate output, retrieval-augmented generation (RAG) has gained significant attention as a key means to mitigate this downside of harnessing only LLMs. However, existing RAG methods for simple and multi-hop question answering (QA) are still prone to incorrect retrievals and hallucinations. To address these limitations, we propose CoopRAG, a novel RAG framework for the question answering task in which a retriever and an LLM work cooperatively with each other by exchanging informative knowledge, and the earlier and later layers of the retriever model work cooperatively with each other to accurately rank the retrieved documents relevant to a given query. In this framework, we (i) unroll a question into sub-questions and a reasoning chain in which uncertain positions are masked, (ii) retrieve the documents relevant to the question augmented with the sub-questions and the reasoning chain, (iii) rerank the documents by contrasting layers of the retriever, and (iv) reconstruct the reasoning chain by filling the masked positions via the LLM. Our experiments demonstrate that CoopRAG consistently outperforms state-of-the-art QA methods on three multi-hop QA datasets as well as a simple QA dataset in terms of both the retrieval and QA performances. Our code is available.
Read the full abstract

Computation and Language

SWAA: Sliding Window Attention Adaptation for Efficient and Quality Preserving Long Context Processing

Yijiong Yu, Jiale Liu, Qingyun Wu et al. (+2 more)

The quadratic complexity of self attention in Transformer based LLMs renders long context inference prohibitively expensive. While Sliding Window Attention (SWA), the simplest sparse attention pattern, offers a linear complexity alternative, it suffers from catastrophic long context performance collapse, which stems from two fundamental factors: the training inference mismatch when naively applying SWA to models pretrained with Full Attention (FA), and the inherent structural inability to access distant information when applying SWA to every module at all times. To address these dual challenges, we propose Sliding Window Attention Adaptation (SWAA), a plug and play toolkit of recipes that adapts FA models to SWA without costly pretraining. SWAA systematically combines four core strategies to tackle these distinct issues: (1) Full Attention (FA) Decode and (2) Interleaving FA and SWA layers, which mitigate structural defects by selectively allowing access to distant information; alongside (3) preserving “sink” tokens and (4) lightweight fine tuning, which mitigate the training inference mismatch. Our experiments reveal that while isolated strategies are insufficient, specific synergistic combinations effectively recover long context performance. Despite varying computational overheads, our performance efficiency trade off analysis identifies optimal SWAA configurations for diverse scenarios, achieving 30% to 100% speedups for long context inference with acceptable quality retention. Our code, data and model weights are available at https://github.com/yuyijiong/sliding-window-attention-adaptation
Read the full abstract

All authors Yijiong Yu, Jiale Liu, Qingyun Wu, Huazheng Wang, Ji Pei

Computation and Language · Cross-listed from cs.SD

BRACE: A Benchmark for Robust Audio Caption Quality Evaluation

Tianyu Guo, Hongyu Chen, Hao Liang et al. (+5 more)

Automatic audio captioning is essential for audio understanding, enabling applications such as accessibility and content indexing. However, evaluating the quality of audio captions remains a major challenge, especially in reference-free settings where high-quality ground-truth captions are unavailable. While CLAPScore is currently the most widely used reference-free Audio Caption Evaluation Metric(ACEM), its robustness under diverse conditions has not been systematically validated. To address this gap, we introduce BRACE, a new benchmark designed to evaluate audio caption alignment quality in a reference-free setting. BRACE is primarily designed for assessing ACEMs, and can also be extended to measure the modality alignment abilities of Large Audio Language Model(LALM). BRACE consists of two sub-benchmarks: BRACE-Main for fine-grained caption comparison and BRACE-Hallucination for detecting subtle hallucinated content. We construct these datasets through high-quality filtering, LLM-based corruption, and human annotation. Given the widespread adoption of CLAPScore as a reference-free ACEM and the increasing application of LALMs in audio-language tasks, we evaluate both approaches using the BRACE benchmark, testing CLAPScore across various CLAP model variants and assessing multiple LALMs. Notably, even the best-performing CLAP-based ACEM achieves only a 70.01 F1-score on the BRACE-Main benchmark, while the best LALM reaches just 63.19. By revealing the limitations of CLAP models and LALMs, our BRACE benchmark offers valuable insights into the direction of future research.
Read the full abstract

All authors Tianyu Guo, Hongyu Chen, Hao Liang, Meiyi Qiang, Bohan Zeng, Linzhuang Sun, Bin Cui, Wentao Zhang

Computation and Language

Confucius Code Agent: Scalable Agent Scaffolding for Real-World Codebases

Sherman Wong, Zhenting Qi, Zhaodong Wang et al. (+8 more)

Real-world software engineering tasks require coding agents that can operate on massive repositories, sustain long-horizon sessions, and reliably coordinate complex toolchains at test time. Existing research-grade coding agents offer transparency but struggle when scaled to heavier, production-level workloads, while production-grade systems achieve strong practical performance but provide limited extensibility, interpretability, and controllability. We introduce the Confucius Code Agent (CCA), a software engineering agent that can operate at large-scale codebases. CCA is built on top of the Confucius SDK, an agent development platform structured around three complementary perspectives: Agent Experience (AX), User Experience (UX), and Developer Experience (DX). The SDK supports a unified orchestrator with advanced context management for long-context reasoning, a persistent note-taking system for cross-session continual learning, and a modular extension system for reliable tool use. In addition, we introduce a meta-agent that automates the construction, evaluation, and refinement of agents through a build-test-improve cycle, enabling rapid agent development on new tasks and tool stacks. Instantiated on the Confucius SDK using the meta-agent, CCA demonstrates strong performance on real-world software engineering tasks. On SWE-Bench-Pro, CCA achieves a Resolve@1 of 59%, exceeding prior research baselines as well as commercial results, under identical repositories, model backends, and tool access.
Read the full abstract

All authors Sherman Wong, Zhenting Qi, Zhaodong Wang, Nathan Hu, Samuel Lin, Jun Ge, Erwin Gao, Wenlin Chen, Yilun Du, Minlan Yu, Ying Zhang

Computation and Language · Cross-listed from cs.LG

GPG: Generalized Policy Gradient Theorem for Transformer-based Policies

Hangyu Mao, Guangting Dong and Zhicheng Dou

We present the Generalized Policy Gradient (GPG) Theorem, specifically designed for Transformer-based policies. Notably, we demonstrate that both standard Policy Gradient Theorem and GRPO emerge as special cases within our GPG framework. Furthermore, we explore its practical applications in training Large Language Models (LLMs), offering new insights into efficient policy optimization.
Read the full abstract