Computation and Language
Linhao Zhong, Linyu Wu, Bozhen Fang et al. (+6 more)
Diffusion Language Models (DLMs) offer a promising alternative for
language modeling by enabling parallel decoding through iterative refinement. However, most DLMs rely on hard binary masking and discrete
token assignments, which hinder the revision of early decisions and underutilize intermediate probabilistic representations. In this paper, we propose EvoToken-DLM, a novel
diffusion-based
language modeling approach that replaces hard binary masks with evolving soft
token distributions. EvoToken-DLM enables a progressive transition from masked states to discrete outputs, supporting revisable decoding. To effectively support this
evolution, we introduce continuous trajectory supervision, which aligns training objectives with iterative probabilistic updates. Extensive experiments across multiple
benchmarks show that EvoToken-DLM consistently achieves superior performance, outperforming strong
diffusion-based and masked DLM baselines. Project webpage: https://aim-uofa.github.io/EvoTokenDLM.
Read the full abstract ▾
All authors Linhao Zhong, Linyu Wu, Bozhen Fang, Tianjian Feng, Chenchen Jing, Wen Wang, Jiaheng Zhang, Hao Chen, Chunhua Shen
Computation and Language
Reward Modeling from Natural Language Human Feedback
Zongqi Wang, Rui Wang, Yuchuan Wu et al. (+5 more)
Reinforcement Learning with Verifiable reward (RLVR) on preference data has become the mainstream approach for training Generative Reward Models (GRMs). Typically in pairwise rewarding tasks, GRMs generate
reasoning chains ending with critiques and preference labels, and RLVR then relies on the correctness of the preference labels as the training reward. However, in this paper, we demonstrate that such binary classification tasks make GRMs susceptible to guessing correct outcomes without sound critiques. Consequently, these spurious successes introduce substantial noise into the reward signal, thereby impairing the effectiveness of reinforcement learning. To address this issue, we propose Reward Modeling from Natural
Language Human Feedback (RM-NLHF), which leverages natural
language feedback to obtain process reward signals, thereby mitigating the problem of limited solution space inherent in binary tasks. Specifically, we compute the similarity between GRM-generated and human critiques as the training reward, which provides more accurate reward signals than outcome-only supervision. Additionally, considering that human critiques are difficult to scale up, we introduce Meta Reward Model (MetaRM) which learns to predict process reward from
datasets with human critiques and then generalizes to data without human critiques. Experiments on multiple
benchmarks demonstrate that our method consistently outperforms state-of-the-art GRMs trained with outcome-only reward, confirming the superiority of integrating natural
language over binary human feedback as supervision.
Read the full abstract ▾
All authors Zongqi Wang, Rui Wang, Yuchuan Wu, Yiyao Yu, Pinyi Zhang, Shaoning Sun, Yujiu Yang, Yongbin Li
Computation and Language
Controlled Self-Evolution for Algorithmic Code Optimization
Tu Hu, Ronghao Chen, Shuo Zhang et al. (+9 more)
Self-
evolution methods enhance code generation through iterative "generate-verify-refine" cycles, yet existing approaches suffer from low exploration efficiency, failing to discover solutions with superior complexity within limited budgets. This inefficiency stems from initialization bias trapping
evolution in poor solution regions, uncontrolled
stochastic operations lacking feedback guidance, and insufficient experience utilization across tasks. To address these bottlenecks, we propose Controlled Self-
Evolution (CSE), which consists of three key components. Diversified Planning Initialization generates structurally distinct algorithmic strategies for broad solution space coverage. Genetic
Evolution replaces
stochastic operations with feedback-guided mechanisms, enabling targeted mutation and compositional crossover. Hierarchical
Evolution Memory captures both successful and failed experiences at inter-task and intra-task levels. Experiments on EffiBench-X demonstrate that CSE consistently outperforms all baselines across various LLM backbones. Furthermore, CSE achieves higher efficiency from early generations and maintains continuous improvement throughout
evolution. Our code is publicly available at https://github.com/QuantaAlpha/EvoControl.
Read the full abstract ▾
All authors Tu Hu, Ronghao Chen, Shuo Zhang, Jianghao Yin, Mou Xiao Feng, Jingping Liu, Shaolei Zhang, Wenqi Jiang, Yuqi Fang, Sen Hu, Huacan Wang, Yi Xu
Computation and Language
DiffER: Diffusion Entity-Relation Modeling for Reversal Curse in Diffusion Large Language Models
Shaokai He, Kaiwen Wei, Xinyi Zeng et al. (+5 more)
The "reversal curse" refers to the phenomenon where large
language models (LLMs) exhibit predominantly unidirectional behavior when processing logically bidirectional relationships. Prior work attributed this to autoregressive training – predicting the next
token inherently favors left-to-right information flow over genuine bidirectional knowledge associations. However, we observe that
Diffusion LLMs (DLLMs), despite being trained bidirectionally, also suffer from the reversal curse. To investigate the root causes, we conduct systematic experiments on DLLMs and identify three key reasons: 1) entity fragmentation during training, 2) data asymmetry, and 3) missing entity relations. Motivated by the analysis of these reasons, we propose
Diffusion Entity-Relation Modeling (DiffER), which addresses the reversal curse through entity-aware training and balanced data construction. Specifically, DiffER introduces whole-entity masking, which mitigates entity fragmentation by predicting complete entities in a single step. DiffER further employs distribution-symmetric and relation-enhanced data construction strategies to alleviate data asymmetry and missing relations. Extensive experiments demonstrate that DiffER effectively alleviates the reversal curse in
Diffusion LLMs, offering new perspectives for future research.
Read the full abstract ▾
All authors Shaokai He, Kaiwen Wei, Xinyi Zeng, Xiang Chen, Xue Yang, Zhenyang Li, Jiang Zhong, Yu Tian
Computation and Language
Beyond Literal Mapping: Benchmarking and Improving Non-Literal Translation Evaluation
Yanzhi Tian, Cunxiang Wang, Zeming Liu et al. (+5 more)
Large
Language Models (LLMs) have significantly advanced Machine Translation (MT), applying them to linguistically complex domains-such as Social Network Services, literature etc. In these scenarios, translations often require handling non-literal expressions, leading to the inaccuracy of MT metrics. To systematically investigate the reliability of MT metrics, we first curate a meta-evaluation
dataset focused on non-literal translations, namely MENT. MENT encompasses four non-literal translation domains and features source sentences paired with translations from diverse MT systems, with 7,530 human-annotated scores on translation quality. Experimental results reveal the inaccuracies of traditional MT metrics and the limitations of LLM-as-a-Judge, particularly the knowledge cutoff and score inconsistency problem. To mitigate these limitations, we propose RATE, a novel agentic translation evaluation framework, centered by a reflective Core
Agent that dynamically invokes specialized sub-
agents. Experimental results indicate the efficacy of RATE, achieving an improvement of at least 3.2 points in combined system- and segment-level correlation with human judgments compared with current methods. Further experiments demonstrate the robustness of RATE to general-domain MT evaluation. Code and
dataset are available at: https://github.com/BITHLP/RATE.
Read the full abstract ▾
All authors Yanzhi Tian, Cunxiang Wang, Zeming Liu, Heyan Huang, Wenbo Yu, Dawei Song, Jie Tang, Yuhang Guo
Computation and Language
BayesRAG: Probabilistic Mutual Evidence Corroboration for Multimodal Retrieval-Augmented Generation
Xuan Li, Yining Wang, Haocai Luo et al. (+6 more)
Retrieval-Augmented Generation (RAG) has become a pivotal paradigm for Large
Language Models (LLMs), yet current approaches struggle with visually rich documents by treating text and images as isolated
retrieval targets. Existing methods relying solely on cosine similarity often fail to capture the semantic reinforcement provided by cross-modal
alignment and layout-induced coherence. To address these limitations, we propose BayesRAG, a novel
multimodal retrieval framework grounded in
Bayesian inference and Dempster-Shafer evidence theory. Unlike traditional approaches that rank candidates strictly by similarity, BayesRAG models the intrinsic consistency of retrieved candidates across modalities as probabilistic evidence to refine
retrieval confidence. Specifically, our method computes the posterior association probability for combinations of
multimodal retrieval results, prioritizing text-image pairs that mutually corroborate each other in terms of both semantics and layout. Extensive experiments demonstrate that BayesRAG significantly outperforms state-of-the-art (SOTA) methods on challenging
multimodal benchmarks. This study establishes a new paradigm for
multimodal retrieval fusion that effectively resolves the isolation of heterogeneous modalities through an evidence fusion mechanism and enhances the robustness of
retrieval outcomes. Our code is available at https://github.com/TioeAre/BayesRAG.
Read the full abstract ▾
All authors Xuan Li, Yining Wang, Haocai Luo, Shengping Liu, Jerry Liang, Ying Fu, Weihuang, Jun Yu, Junnan Zhu
Computation and Language
How to predict creativity ratings from written narratives: A comparison of co-occurrence and textual forma mentis networks
Roberto Passaro, Edith Haim and Massimo Stella
This tutorial paper provides a step-by-step workflow for building and analysing semantic networks from short creative texts. We introduce and compare two widely used text-to-network approaches: word co-occurrence networks and textual forma mentis networks (TFMNs). We also demonstrate how they can be used in machine learning to predict human creativity ratings. Using a corpus of 1029 short stories, we guide readers through text preprocessing, network construction, feature extraction (structural measures, spreading-activation indices, and
emotion scores), and application of regression models. We evaluate how network-construction choices influence both network
topology and predictive performance. Across all modelling settings, TFMNs consistently outperformed co-occurrence networks through lower prediction errors (best MAE = 0.581 for TFMN, vs 0.592 for co-occurrence with window size 3). Network-structural features dominated predictive performance (MAE = 0.591 for TFMN), whereas
emotion features performed worse (MAE = 0.711 for TFMN) and spreading-activation measures contributed little (MAE = 0.788 for TFMN). This paper offers practical guidance for researchers interested in applying network-based methods for cognitive fields like creativity research. we show when syntactic networks are preferable to surface co-occurrence models, and provide an open, reproducible workflow accessible to newcomers in the field, while also offering deeper methodological insight for experienced researchers.
Read the full abstract ▾
Computation and Language
Mitrasamgraha: A Comprehensive Classical Sanskrit Machine Translation Dataset
Sebastian Nehrdich, David Allport, Sven Sellmer et al. (+5 more)
While machine translation is regarded as a "solved problem" for many high-resource
languages, close analysis quickly reveals that this is not the case for content that shows challenges such as poetic
language, philosophical concepts, multi-layered metaphorical expressions, and more. Sanskrit literature is a
prime example of this, as it combines a large number of such challenges in addition to inherent linguistic features like sandhi, compounding, and heavy morphology, which further complicate NLP downstream tasks. It spans multiple millennia of text production time as well as a large breadth of different domains, ranging from ritual formulas via epic narratives, philosophical treatises, poetic verses up to scientific material. As of now, there is a strong lack of publicly available resources that cover these different domains and temporal layers of Sanskrit. We therefore introduce Mitrasamgraha, a high-quality Sanskrit-to-English machine translation
dataset consisting of 391,548 bitext pairs, more than four times larger than the largest previously available Sanskrit
dataset Itih=asa. It covers a time period of more than three millennia and a broad range of historical Sanskrit domains. In contrast to web-crawled
datasets, the temporal and domain annotation of this
dataset enables fine-grained study of domain and time period effects on MT performance. We also release a validation set consisting of 5,587 and a test set consisting of 5,552 post-corrected bitext pairs. We conduct experiments benchmarking commercial and open models on this
dataset and fine-tune NLLB and Gemma models on the
dataset, showing significant improvements, while still recognizing significant challenges in the translation of complex compounds, philosophical concepts, and multi-layered metaphors. We also analyze how in-context learning on this
dataset impacts the performance of commercial models
Read the full abstract ▾
All authors Sebastian Nehrdich, David Allport, Sven Sellmer, Jivnesh Sandhan, Manoj Balaji Jagadeeshan, Pawan Goyal, Sujeet Kumar, Kurt Keutzer
Computation and Language
PsyCLIENT: Client Simulation via Conversational Trajectory Modeling for Trainee Practice and Model Evaluation in Mental Health Counseling
Huachuan Qiu, Zhaoming Chen, Yuqian Chen et al. (+3 more)
LLM-based client simulation provides a scalable approach to novice counselor training, counseling-dialogue synthesis, and interactive evaluation of automated counseling systems. However, existing approaches are limited by insufficient profile diversity, weak behavioral grounding, and the lack of open Chinese-
language resources for simulated counseling clients. We propose PsyCLIENT, a framework that conditions simulated-client responses on client profiles, dialogue histories, and conversational trajectories specifying target behaviors and content constraints at each client turn. We also construct PsyCLIENT-CP, a
dataset of 120 Chinese client profiles spanning 60 counseling topics. Evaluations involving 24 professional counselors show that PsyCLIENT receives higher ratings of perceived authenticity and training utility than the comparison methods. In a separate source-identification study, PsyCLIENT dialogues were more frequently misclassified as human-client interactions than those generated by the baselines. These results suggest that conversational trajectory modeling can transform static client profiles into behavior-guided, dynamically unfolding counseling interactions.
Read the full abstract ▾
All authors Huachuan Qiu, Zhaoming Chen, Yuqian Chen, Yuan Xie, Yu Lu, Zhenzhong Lan
Computation and Language · Cross-listed from cs.AI
LRAS: Advanced Legal Reasoning with Agentic Search
Yujin Zhou, Chuxue Cao, Jinluan Yang et al. (+4 more)
While Large
Reasoning Models (LRMs) have demonstrated exceptional logical capabilities in mathematical domains, their application to the legal field remains hindered by the strict requirements for procedural rigor and adherence to legal logic. Existing legal LLMs, which rely on "closed-loop
reasoning" derived solely from internal parametric knowledge, frequently suffer from lack of self-awareness regarding their knowledge boundaries, leading to confident yet incorrect conclusions. To address this challenge, we present Legal
Reasoning with Agentic Search (LRAS), the first framework designed to transition legal LLMs from static and parametric "closed-loop thinking" to dynamic and interactive "Active Inquiry". By integrating Introspective Imitation Learning and Difficulty-aware Reinforcement Learning, LRAS enables LRMs to identify knowledge boundaries and handle legal
reasoning complexity. Empirical results demonstrate that LRAS outperforms state-of-the-art baselines by 8.2-32%, with the most substantial gains observed in tasks requiring deep
reasoning with reliable knowledge. We will release our data and models for further exploration soon.
Read the full abstract ▾
All authors Yujin Zhou, Chuxue Cao, Jinluan Yang, Lijun Wu, Conghui He, Sirui Han, Yike Guo
Computation and Language
ReasonTabQA: A Comprehensive Benchmark for Table Question Answering from Real World Industrial Scenarios
Changzai Pan, Jie Zhang, Kaiwen Wei et al. (+15 more)
Recent advancements in Large
Language Models (LLMs) have significantly catalyzed table-based question answering (TableQA). However, existing TableQA
benchmarks often overlook the intricacies of industrial scenarios, which are characterized by multi-table structures, nested headers, and massive scales. These environments demand robust table
reasoning through deep structured inference, presenting a significant challenge that remains inadequately addressed by current methodologies. To bridge this gap, we present ReasonTabQA, a large-scale bilingual
benchmark encompassing 1,932 tables across 30 industry domains such as energy and automotive. ReasonTabQA provides high-quality annotations for both final answers and explicit
reasoning chains, supporting both thinking and no-thinking paradigms. Furthermore, we introduce TabCodeRL, a reinforcement learning method that leverages table-aware verifiable rewards to guide the generation of logical
reasoning paths. Extensive experiments on ReasonTabQA and 4 TableQA
datasets demonstrate that while TabCodeRL yields substantial performance gains on open-source LLMs, the persistent performance gap on ReasonTabQA underscores the inherent complexity of real-world industrial TableQA.
Read the full abstract ▾
All authors Changzai Pan, Jie Zhang, Kaiwen Wei, Chenshuo Pan, Yu Zhao, Jingwang Huang, Jian Yang, Zhenhe Wu, Haoyang Zeng, Xiaoyan Gu, Weichao Sun, Yanbo Zhai, Yujie Mao, Zhuoru Jiang, Jiang Zhong, Shuangyong Song, Yongxiang Li, Zhongjiang He
Computation and Language
Towards Comprehensive Semantic Speech Embeddings for Chinese Dialects
Kalvin Chang, Yiwen Shao, Jiahong Li et al. (+1 more)
Despite having hundreds of millions of speakers, Chinese dialects lag behind Mandarin in
speech and
language technologies. Most varieties are primarily spoken, making dialect-to-Mandarin
speech-LLMs (large
language models) more practical than dialect LLMs. Building dialect-to-Mandarin
speech-LLMs requires
speech representations with cross-dialect semantic
alignment between Chinese dialects and Mandarin. In this paper, we achieve such a cross-dialect semantic
alignment by training a
speech encoder with ASR (automatic
speech recognition)-only data, as demonstrated by
speech-to-
speech retrieval on a new
benchmark of spoken Chinese varieties that we contribute. Our
speech encoder further demonstrates state-of-the-art ASR performance on Chinese dialects. Together, our Chinese dialect
benchmark, semantically aligned
speech representations, and
speech-to-
speech retrieval evaluation lay the groundwork for future Chinese dialect
speech-LLMs. We release the
benchmark at https://github.com/kalvinchang/yubao.
Read the full abstract ▾
All authors Kalvin Chang, Yiwen Shao, Jiahong Li, Dong Yu
Computation and Language
Document-Level Zero-Shot Relation Extraction with Entity Side Information
Mohan Raj Chanthran, Soon Lay Ki, Ong Huey Fang et al. (+1 more)
Document-Level Zero-Shot Relation Extraction (DocZSRE) aims to predict unseen relation labels in text documents without prior training on specific relations. Existing approaches rely on Large
Language Models (LLMs) to generate synthetic data for unseen labels, which poses challenges for low-resource
languages like Malaysian English. These challenges include the incorporation of local linguistic nuances and the risk of factual inaccuracies in LLM-generated data. This paper introduces Document-Level Zero-Shot Relation Extraction with Entity Side Information (DocZSRE-SI) to address limitations in the existing DocZSRE approach. The DocZSRE-SI framework leverages Entity Side Information, such as Entity Mention Descriptions and Entity Mention Hypernyms, to perform ZSRE without depending on LLM-generated synthetic data. The proposed low-complexity model achieves an average improvement of 11.6% in the macro F1-Score compared to baseline models and existing
benchmarks. By utilizing Entity Side Information, DocZSRE-SI offers a robust and efficient alternative to error-prone, LLM-based methods, demonstrating significant advancements in handling low-resource
languages and linguistic diversity in relation extraction tasks. This research provides a scalable and reliable solution for ZSRE, particularly in contexts like Malaysian English news articles, where traditional LLM-based approaches fall short.
Read the full abstract ▾
All authors Mohan Raj Chanthran, Soon Lay Ki, Ong Huey Fang, Bhawani Selvaretnam
Computation and Language
The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agents
Weihao Xuan, Qingcheng Zeng, Heli Qi et al. (+3 more)
Autonomous
agents based on large
language models (LLMs) are rapidly evolving to handle multi-turn tasks, but ensuring their trustworthiness remains a critical challenge. A fundamental pillar of this trustworthiness is calibration, which refers to an
agent's ability to express confidence that reliably reflects its actual performance. While calibration is well-established for static models, its dynamics in tool-integrated agentic workflows remain underexplored. In this work, we systematically investigate verbalized calibration in tool-use
agents, revealing a fundamental confidence dichotomy driven by tool type. Specifically, our pilot study identifies that evidence tools (e.g., web search) systematically induce severe overconfidence due to inherent noise in retrieved information, while verification tools (e.g., code interpreters) can ground
reasoning through deterministic feedback and mitigate miscalibration. To robustly improve calibration across tool types, we propose a reinforcement learning (RL) fine-tuning framework that jointly optimizes task accuracy and calibration, supported by a holistic
benchmark of reward designs. We demonstrate that our trained
agents not only achieve superior calibration but also exhibit robust generalization from local training environments to noisy web settings and to distinct domains such as mathematical
reasoning. Our results highlight the necessity of domain-specific calibration strategies for tool-use
agents. More broadly, this work establishes a foundation for building self-aware
agents that can reliably communicate uncertainty in high-stakes, real-world deployments.
Read the full abstract ▾
All authors Weihao Xuan, Qingcheng Zeng, Heli Qi, Yunze Xiao, Junjue Wang, Naoto Yokoya
Computation and Language
ActiShade: Activating Overshadowed Knowledge to Guide Multi-Hop Reasoning in Large Language Models
Huipeng Ma, Luan Zhang, Dandan Song et al. (+10 more)
In multi-hop
reasoning, multi-round
retrieval-augmented generation (RAG) methods typically rely on LLM-generated content as the
retrieval query. However, these approaches are inherently vulnerable to knowledge overshadowing - a phenomenon where critical information is overshadowed during generation. As a result, the LLM-generated content may be incomplete or inaccurate, leading to irrelevant
retrieval and causing error accumulation during the iteration process. To address this challenge, we propose ActiShade, which detects and activates overshadowed knowledge to guide large
language models (LLMs) in multi-hop
reasoning. Specifically, ActiShade iteratively detects the overshadowed keyphrase in the given query, retrieves documents relevant to both the query and the overshadowed keyphrase, and generates a new query based on the retrieved documents to guide the next-round iteration. By supplementing the overshadowed knowledge during the formulation of next-round queries while minimizing the introduction of irrelevant noise, ActiShade reduces the error accumulation caused by knowledge overshadowing. Extensive experiments show that ActiShade outperforms existing methods across multiple
datasets and LLMs.
Read the full abstract ▾
All authors Huipeng Ma, Luan Zhang, Dandan Song, Linmei Hu, Yuhang Tian, Jun Yang, Changzhi Zhou, Chenhao Li, Yizhou Jin, Xudong Li, Meng Lin, Mingxing Zhang, Shuhao Zhang
Computation and Language · Cross-listed from cs.AI
Learning to Trust the Crowd: A Multi-Model Consensus Reasoning Engine for Large Language Models
Pranav Kallem
Large
language models (LLMs) achieve strong average performance yet remain unreliable at the instance level, with frequent
hallucinations, brittle failures, and poorly calibrated confidence. We study reliability through the lens of multi-model consensus: given responses from several heterogeneous LLMs, can we learn which answer is most likely correct for a given query? We introduce a Multi-Model Consensus
Reasoning Engine that treats the set of LLM outputs as input to a supervised meta-learner. The system maps natural
language responses into structured features using semantic
embeddings, pairwise similarity and clustering statistics, lexical and structural cues,
reasoning-quality scores, confidence estimates, and model-specific priors, and then applies gradient-boosted trees, listwise ranking, and graph neural networks over similarity graphs of answers. Using three open-weight LLMs evaluated on compact, resource-constrained subsets of GSM8K, ARC-Challenge, HellaSwag, and TruthfulQA, our best graph-
attention-based consensus model improves macro-average accuracy by 4.6 percentage points over the strongest single LLM and by 8.1 points over majority vote, while also yielding lower Brier scores and fewer TruthfulQA
hallucinations. Ablation and feature-importance analyses show that semantic agreement and clustering features are most influential, with
reasoning-quality and model-prior features providing complementary gains, suggesting supervised multi-model consensus is a practical route toward more reliable LLM behavior, even in a modest single-machine setup.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
Lost in the Noise: How Reasoning Models Fail with Contextual Distractors
Seongyun Lee, Yongrae Jo, Minju Seo et al. (+2 more)
Recent advances in
reasoning models and agentic AI systems have led to an increased reliance on diverse external information. However, this shift introduces input contexts that are inherently noisy, a reality that current sanitized
benchmarks fail to capture. We introduce NoisyBench, a comprehensive
benchmark that systematically evaluates model robustness across 11
datasets in RAG,
reasoning,
alignment, and tool-use tasks against diverse noise types, including random documents, irrelevant chat histories, and hard negative distractors. Our evaluation reveals a catastrophic performance drop of up to 80% in state-of-the-art models when faced with contextual distractors. Crucially, we find that agentic workflows often amplify these errors by over-trusting noisy tool outputs, and distractors can trigger emergent misalignment even without
adversarial intent. We find that prompting, context engineering, SFT, and outcome-reward only RL fail to ensure robustness; in contrast, our proposed Rationale-Aware Reward (RARE) significantly strengthens resilience by incentivizing the identification of helpful information within noise. Finally, we uncover an inverse scaling trend where increased test-time computation leads to worse performance in noisy settings and demonstrate via
attention visualization that models disproportionately focus on distractor
tokens, providing vital insights for building the next generation of robust,
reasoning-capable
agents.
Read the full abstract ▾
All authors Seongyun Lee, Yongrae Jo, Minju Seo, Moontae Lee, Minjoon Seo
Computation and Language
The Roots of Performance Disparity in Multilingual Language Models: Intrinsic Modeling Difficulty or Design Choices?
Chen Shani, Yuval Reif, Nathan Roll et al. (+2 more)
Multilingual language models (LMs) promise broader NLP access, yet current systems deliver uneven performance across the world's
languages. This survey examines why these gaps persist and whether they reflect intrinsic linguistic difficulty or modeling artifacts. We organize the literature around two questions: do linguistic disparities arise from representation and allocation choices (e.g., tokenization, encoding, data exposure, parameter sharing) rather than inherent complexity; and which design choices mitigate inequities across typologically diverse
languages. We review linguistic features, such as orthography, morphology, lexical diversity, syntax, information density, and typological distance, linking each to concrete modeling mechanisms. Gaps often shrink when
segmentation, encoding, and data exposure are normalized, suggesting much apparent difficulty stems from current modeling choices. We synthesize these insights into design recommendations for tokenization, sampling, architectures, and evaluation to support more balanced
multilingual LMs.
Read the full abstract ▾
All authors Chen Shani, Yuval Reif, Nathan Roll, Dan Jurafsky, Ekaterina Shutova
Computation and Language
MI-PRUN: Optimize Large Language Model Pruning via Mutual Information
Hao Zhang, Zhibin Zhang, Guangxin Wu et al. (+3 more)
Large
Language Models (LLMs) have become indispensable across various domains, but this comes at the cost of substantial computational and
memory resources. Model
pruning addresses this by removing redundant components from models. In particular, block
pruning can achieve significant compression and inference acceleration. However, existing block
pruning methods are often unstable and struggle to attain globally optimal solutions. In this paper, we propose a mutual information based
pruning method MI-PRUN for LLMs. Specifically, we leverages mutual information to identify redundant blocks by evaluating transitions in hidden states. Additionally, we incorporate the Data Processing Inequality (DPI) to reveal the relationship between the importance of entire contiguous blocks and that of individual blocks. Moreover, we develop the Fast-Block-Select
algorithm, which iteratively updates block combinations to achieve a globally optimal solution while significantly improving the efficiency. Extensive experiments across various models and
datasets demonstrate the stability and effectiveness of our method.
Read the full abstract ▾
All authors Hao Zhang, Zhibin Zhang, Guangxin Wu, He Chen, Jiafeng Guo, Xueqi Cheng
Computation and Language · Cross-listed from cs.LG
MAESTRO: Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization
Yang Zhao, Hepeng Wang, Xiao Ding et al. (+8 more)
Group-Relative Policy Optimization (GRPO) has emerged as an efficient paradigm for aligning Large
Language Models (LLMs), yet its efficacy is primarily confined to domains with verifiable ground truths. Extending GRPO to open-domain settings remains a critical challenge, as unconstrained generation entails multi-faceted and often conflicting objectives - such as creativity versus factuality - where rigid, static reward scalarization is inherently suboptimal. To address this, we propose MAESTRO (Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization), which introduces a meta-cognitive orchestration layer that treats reward scalarization as a dynamic latent policy, leveraging the model's terminal hidden states as a semantic bottleneck to perceive task-specific priorities. We formulate this as a contextual bandit problem within a bi-level optimization framework, where a lightweight Conductor network co-evolves with the policy by utilizing group-relative advantages as a meta-reward signal. Across seven
benchmarks, MAESTRO consistently outperforms single-reward and static multi-objective baselines, while preserving the efficiency advantages of GRPO, and in some settings even reducing redundant generation.
Read the full abstract ▾
All authors Yang Zhao, Hepeng Wang, Xiao Ding, Yangou Ouyang, Bibo Cai, Kai Xiong, Jinglong Gao, Zhouhao Sun, Li Du, Bing Qin, Ting Liu
Computation and Language
Relink: Constructing Query-Driven Evidence Graph On-the-Fly for GraphRAG
Manzong Huang, Chenyang Bu, Yi He et al. (+2 more)
Graph-based
Retrieval-Augmented Generation (GraphRAG) mitigates
hallucinations in Large
Language Models (LLMs) by grounding them in structured knowledge. However, current GraphRAG methods are constrained by a prevailing build-then-reason paradigm, which relies on a static, pre-constructed Knowledge Graph (KG). This paradigm faces two critical challenges. First, the KG's inherent incompleteness often breaks
reasoning paths. Second, the graph's low signal-to-noise ratio introduces distractor facts, presenting query-relevant but misleading knowledge that disrupts the
reasoning process. To address these challenges, we argue for a reason-and-construct paradigm and propose Relink, a framework that dynamically builds a query-specific evidence graph. To tackle incompleteness, Relink instantiates required facts from a latent relation pool derived from the original text corpus, repairing broken paths on the fly. To handle misleading or distractor facts, Relink employs a unified, query-aware evaluation strategy that jointly considers candidates from both the KG and latent relations, selecting those most useful for answering the query rather than relying on their pre-existence. This empowers Relink to actively discard distractor facts and construct the most faithful and precise evidence path for each query. Extensive experiments on five Open-Domain Question Answering
benchmarks show that Relink achieves significant average improvements of 5.4% in EM and 5.2% in F1 over leading GraphRAG baselines, demonstrating the superiority of our proposed framework.
Read the full abstract ▾
All authors Manzong Huang, Chenyang Bu, Yi He, Xingrui Zhuo, Xindong Wu
Computation and Language
Jinyi Han, Zixiang Di, Zishang Jiang et al. (+4 more)
Large
language models (LLMs) achieve strong performance by generating long chains of thought, but longer traces always introduce redundant or ineffective
reasoning steps. One typical behavior is that they often perform unnecessary verification and revisions even if they have reached the correct answers. This limitation stems from the unstructured nature of
reasoning trajectories and the lack of targeted supervision for critical
reasoning abilities. To address this, we propose Structured
Reasoning (SCR), a framework that decouples
reasoning trajectories into explicit, evaluable, and trainable components. We mainly implement SCR using a Generate-Verify-Revise paradigm. Specifically, we construct structured training data and apply Dynamic Termination Supervision to guide the model in deciding when to terminate
reasoning. To avoid interference between learning signals for different
reasoning abilities, we adopt a progressive two-stage reinforcement learning strategy: the first stage targets initial generation and self-verification, and the second stage focuses on revision. Extensive experiments on three backbone models show that SCR substantially improves
reasoning efficiency and self-verification. Besides, compared with existing
reasoning paradigms, it reduces output
token length by up to 50%.
Read the full abstract ▾
All authors Jinyi Han, Zixiang Di, Zishang Jiang, Ying Liao, Jiaqing Liang, Yongqi Wang, Yanghua Xiao
Computation and Language
Can Large Language Models Understand, Reason About, and Generate Code-Switched Text?
Genta Indra Winata, David Anugraha, Patrick Amadeus Irawan et al. (+15 more)
Code-switching is a pervasive phenomenon in
multilingual communication, yet the robustness of large
language models (LLMs) in mixed-
language settings remains insufficiently understood. In this work, we present a comprehensive evaluation of LLM capabilities in understanding,
reasoning over, and generating code-switched text. We introduce CodeMixQA a novel
benchmark with high-quality human annotations, comprising 16 diverse parallel code-switched
language-pair variants that span multiple geographic regions and code-switching patterns, and include both original scripts and their transliterated forms. Using this
benchmark, we analyze the
reasoning behavior of LLMs on code-switched question-answering tasks, shedding light on how models process and reason over mixed-
language inputs. We further conduct a systematic evaluation of LLM-generated synthetic code-switched text, focusing on both naturalness and semantic fidelity, and uncover key limitations in current generation capabilities. Our findings reveal persistent challenges in both
reasoning and generation under code-switching conditions and provide actionable insights for building more robust
multilingual LLMs. We release the
dataset and code as open source.
Read the full abstract ▾
All authors Genta Indra Winata, David Anugraha, Patrick Amadeus Irawan, Anirban Das, Haneul Yoo, Paresh Dashore, Shreyas Kulkarni, Ruochen Zhang, Haruki Sakajo, Frederikus Hudi, Anaelia Ovalle, Syrielle Montariol, Felix Gaschi, Michael Anugraha, Rutuj Ravindra Puranik, Zawad Hayat Ahmed, Adril Putra Merin, Emmanuele Chersoni
Computation and Language · Cross-listed from cs.AI
Rewarding Creativity: A Human-Aligned Generative Reward Model for Reinforcement Learning in Storytelling
Zhaoyan Li, Hang Lei, Yujia Wang et al. (+3 more)
While Large
Language Models (LLMs) can generate fluent text, producing high-quality creative stories remains challenging. Reinforcement Learning (RL) offers a promising solution but faces two critical obstacles: designing reliable reward signals for subjective storytelling quality and mitigating training instability. This paper introduces the Reinforcement Learning for Creative Storytelling (RLCS) framework to systematically address both challenges. First, we develop a Generative Reward Model (GenRM) that provides multi-dimensional analysis and explicit
reasoning about story preferences, trained through supervised fine-tuning on demonstrations with
reasoning chains distilled from strong teacher models, followed by GRPO-based refinement on expanded preference data. Second, we introduce an
entropy-based reward shaping strategy that dynamically prioritizes learning on confident errors and uncertain correct predictions, preventing overfitting on already-mastered patterns. Experiments demonstrate that GenRM achieves 68%
alignment with human creativity judgments, and RLCS significantly outperforms strong baselines including Gemini-2.5-Pro in overall story quality. This work provides a practical pipeline for applying RL to creative domains, effectively navigating the dual challenges of reward modeling and training stability.
Read the full abstract ▾
All authors Zhaoyan Li, Hang Lei, Yujia Wang, Lanbo Liu, Hao Liu, Liang Yu
Computation and Language
Measuring Iterative Temporal Reasoning with Time Puzzles
Zhengxiang Wang and Zeyu Dong
Tool use, such as web search, has become a standard capability even in freely available large
language models (LLMs). However, existing
benchmarks evaluate temporal
reasoning mainly in static, non-tool-using settings, which poorly reflect how LLMs perform temporal
reasoning in practice. We introduce Time Puzzles, a constraint-based date inference task for evaluating iterative temporal
reasoning with tools. Each puzzle combines factual temporal anchors with (cross-cultural) calendar relations and may admit one or multiple valid dates. The puzzles are algorithmically generated, enabling controlled and continual evaluation. Across 13 LLMs, even the best model (GPT-5) achieves only 55.3% accuracy without tools, despite using easily searchable facts. While web search improves performance, models perform substantially better when constraints are rewritten with explicit dates, removing the need for factual lookup. These results reveal a gap in reliable tool use for iterative temporal
reasoning.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.IR
ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System
Sungguk Cha, DongWook Kim, Mintae Kim et al. (+3 more)
Multi-vector
embedding models have emerged as a powerful paradigm for document
retrieval, preserving fine-grained visual and textual details through
token-level representations. However, this expressiveness comes at a staggering cost: storing
embeddings for every
token inflates index sizes by over $1000\times$ compared to single-vector approaches, severely limiting scalability. We introduce ReinPool, a reinforcement learning framework that learns to dynamically filter and pool multi-vector
embeddings into compact,
retrieval-optimized representations. By training with an inverse
retrieval objective and NDCG-based rewards, ReinPool identifies and retains only the most discriminative vectors without requiring manual importance annotations. On the Vidore V2
benchmark across three
vision-
language embedding models, ReinPool compresses multi-vector representations by $746$–$1249\times$ into single vectors while recovering 76–81% of full multi-vector
retrieval performance. Compared to static mean pooling baselines, ReinPool achieves 22–33% absolute NDCG@3 improvement, demonstrating that learned selection significantly outperforms heuristic aggregation.
Read the full abstract ▾
All authors Sungguk Cha, DongWook Kim, Mintae Kim, Youngsub Han, Byoung-Ki Jeon, Sangyeob Lee
Computation and Language
ReMIND: Orchestrating Modular Large Language Models for Controllable Serendipity A REM-Inspired System Design for Emergent Creative Ideation
Makoto Sato
Large
language models (LLMs) are increasingly used not only for problem solving but also for creative ideation; however, generating ideas that are both novel and coherent remains challenging. While high-temperature sampling can promote originality, it often compromises consistency and usefulness. Here, we propose ReMIND, a four-stage framework comprising wake, which establishes a stable semantic baseline through low-temperature generation; dream, which performs high-temperature exploratory generation; judge, which evaluates candidate outputs for consistency and extracts salient novel ideas; and rewake, which consolidates selected ideas into coherent final outputs. By assigning these functions to independent LLM modules, ReMIND explicitly separates exploration from stabilization. We systematically evaluated diverse model configurations using multiple creative ideation tasks. External evaluations showed that novelty enhancement could emerge through two separable stages: first from the wake to the dream phase through high-temperature exploration, and subsequently from the dream to the rewake phase through judge-mediated selection and consolidation. Notably, different LLM families exhibited distinct functional tendencies. These findings suggest that serendipitous ideation in LLMs is not solely a property of individual models but emerges from interactions among models assigned to specialized cognitive roles. ReMIND provides a general framework for investigating computational creativity and demonstrates how modular LLM orchestration can bridge exploratory generation and coherent idea formation.
Read the full abstract ▾
Computation and Language
The Need for a Socially-Grounded Persona Framework for User Simulation
Pranav Narayanan Venkit, Yu Li, Yada Pruksachatkun et al. (+1 more)
Synthetic personas are widely used to condition large
language models (LLMs) for social simulation, yet most personas are still constructed from coarse sociodemographic attributes or summaries. We revisit persona creation by introducing SCOPE, a socially grounded framework for persona construction and evaluation, built from a 141-item, two-hour sociopsychological protocol collected from 124 U.S.-based participants. Across seven models, we find that demographic-only personas are a structural bottleneck: demographics explain only ~1.5% of variance in human response similarity. Adding sociopsychological facets improves behavioral prediction and reduces over-accentuation, and non-demographic personas based on values and identity achieve strong
alignment with substantially lower bias. These trends generalize to SimBench (441 aligned questions), where SCOPE personas outperform default prompting and NVIDIA Nemotron personas, and SCOPE augmentation improves Nemotron-based personas. Our results indicate that persona quality depends on sociopsychological structure rather than demographic templates or summaries.
Read the full abstract ▾
All authors Pranav Narayanan Venkit, Yu Li, Yada Pruksachatkun, Chien-Sheng Wu