跳到正文
研踪PaperTrail

YOUR RESEARCH, IN SIGHT

让下一篇重要论文，不再错过。

覆盖计算机全学科，以 AI 为快捷入口。关注方向、作者和关键词，把发现变成持续的研究积累。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+1032新增论文
+792其中 AI 相关
+128已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-12 06:59 · 全站最近入库：2026-11-12 06:37各信源进度 →

统计口径与数据新鲜度

统计 2026-11-11 09:44 — 2026-11-12 09:44（北京时间） · 点击数字查看对应简报；AI 相关包含在新增论文内。

最近完成检查：2026-11-12 06:59最近发现新增版本：2026-11-12 06:37库内论文最新更新时间：2026-11-11 08:00

每分钟刷新已保存的数据。空检查不代表采集异常；新收录可能包含历史论文。AI 导读按新增版本计数。

21,675 已收录论文18,323 AI 相关6,056 有修订112 HF 社区发现

浏览论文库

研究日报 / 周报 →按信源浏览 →

时间范围跟随排序：版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常，查看状态 →

找到 5,654 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-12 06:57（北京时间）
首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.64ACL Anthology · ACL 2026

Fusion Training for Mathematical Generalization in Large Language Models

Congfeng Cao, Pengyu Zhang, Jelke Bloem University of Amsterdam

原文摘要

Thinking Mode Fusion (TMF) enables large language models to support both concise responses and long-form reasoning by unifying a non-thinking mode and a thinking mode within a single model. However, its training dynamics, including the data ratio and training schedule between the two modes, remain underexplored. In this work, we present a systematic study of TMF by analyzing the effects of the training schedule and data ratio between thinking and non-thinking modes. Focusing on mathematical problem solving, we construct a benchmark with multiple thinking-to-non-thinking data ratios and three training schedules. Our results reveal an asymmetric interaction between the two modes: increasing the ratio of non-thinking supervision reduces the accuracy of the thinking mode. We further show that different training schedules modulate this trade-off and that the optimal schedule depends on the data ratio. Finally, we quantify a negative correlation between non-thinking and thinking mode supervision, highlighting an inherent tension between these two modes. These findings provide practical guidance for designing effective TMF training settings. All code and data are released to support further research at: Fusion Bench.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.65ACL Anthology · ACL 2026

Does Topic Sentiment Cause Perceived Ideology? Comparing Human and LLM Annotations in Political News Articles

Upasana Chatterjee

原文摘要

We ask whether topic sentiment has a causal effect on perceived political ideology, and whether the answer depends on who assigns the ideology label. Using articles from AllSides, paired with shared sentiment annotations from Llama-3.3-70b-versatile, we compare ideology labels from expert human annotators, GPT-4o-mini (baseline and finetuned), and Llama-3.3-70B. We apply Double Machine Learning (DML) and community-level mediation analysis across all four annotation paradigms. Human annotations yield no significant causal effects at the community level. Fine-tuned GPT-4o-mini achieves the highest classification accuracy (F1=72.48) and is the only annotator paradigm that produces significant community-level treatment effects and significant natural direct effects (NDEs) in mediation. We interpret this as evidence of shortcut learning: fine-tuning on ideology-labeled data causes the model to internalise a spurious sentiment–ideology coupling not operative in human judgment for this task. This coupling is structurally invisible to F1-based evaluation, with implications for the use of LLM annotations as silver labels and as proxies for human judgment in downstream causal analyses.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.66ACL Anthology · ACL 2026

Understanding Conversational Implicatures in Humans and LLMs

Daeun Kang

原文摘要

In conversational implicatures, speakers convey hidden intended meanings beyond the literal content of their utterances, and hearers are expected to infer what is implied. This study examines how Large Language Models (LLMs) interpret conversational implicatures, using human interpretation as a baseline and gold standard for comparison. The same experiments were conducted with two types of participants: humans and LLMs. Two metrics were adopted: a surprisal-based metric and a response-based metric. The results suggest that the response-based metric demonstrates higher accuracy, comparable to human responses, than the surprisal-based metric. In particular, humans and LLMs using the response-based metric performed better in the literal condition than in the implied condition. Additionally, they were more sensitive to capturing implied meanings for some-all trigger than for other triggers, whereas they showed lower performance on Manner implicatures. Overall, LLMs employing the response-based metric tend to exhibit human-like behavior, but still diverge from humans in their understanding of conversational implicatures.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.67ACL Anthology · ACL 2026

Thesis Proposal: Self-Adaptive and Epistemic Uncertainty-Guided ASR of Dense Intra-Sentential Code-Switched Speech for African Low-Resource Languages

Umar Baba Umar, Sulaimon Adebayo Bashir, Abdulmalik Danlami Mohammed Federal University of Technology, Minna, Amina Gogo Tafida

原文摘要

Automatic Speech Recognition (ASR) has achieved strong performance for high-resource languages, but dense intra-sentential code-switched speech in African low-resource settings remains underexplored. Existing multilingual and pretrained ASR systems improve general recognition accuracy, yet they remain weak at switch regions, are sensitive to language imbalance during adaptation, and are typically evaluated with metrics that obscure switching-specific errors. This thesis proposes a self-adaptive and epistemic uncertainty-guided framework for African low-resource code-switched ASR, using Hausa–English (Engausa) and Hausa–Yorùbá as case studies. The work investigates three linked questions: (1) how to design a linguistically informed code-switched corpus with explicit switch-region annotation and labeled/unlabeled partitions for adaptive learning, (2) whether epistemic uncertainty is systematically elevated around switch regions and can guide pseudo-label selection in semi-supervised training, and (3) whether switch-aware adaptation with auxiliary language identification and boundary supervision can reduce recognition errors without increasing catastrophic forgetting. The long-term goal is to develop scalable and data-efficient ASR systems that model code-switching as a structured linguistic phenomenon rather than as noise in multilingual African speech.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.68ACL Anthology · ACL 2026

RegTrack: A Fine-Grained Benchmark for Multi-Class Legal Change Detection

Joe Yu Technical University of Munich, Kevin Chenhao Li Technical University of Munich, Julian Ostarek Shuttle Technologies and Technical University of Munich

原文摘要

Organizations must continuously monitor evolving regulations to maintain compliance. While current tools are limited to surface-level text comparison, existing models lack the finegrained classification schemes to determine whether small changes impact legal obligations or merely update formatting. To address this gap, we introduce a novel benchmark for change detection in EU regulations. It comprises 4,772 manually annotated pairs of structurally distinct provisions, defined as Atomic Legal Units (ALUs), mapped to a six-class taxonomy of legal change types. We formalize three core tasks: structural alignment, change classification, and a combined task requiring simultaneous alignment and classification. Evaluating lexical algorithms, dense encoders, and Large Language Models (LLMs) as baselines, we find LLMs excel at isolated change classification, whereas domain-specific dense encoders offer the most robust combined performance. By providing fine-grained labeled data, this benchmark enables the development of AI systems that can help organizations analyze regulatory shifts and support version-aware retrieval in the legal domain.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.69ACL Anthology · ACL 2026

Validator-Guided Hard Negative Mining for Masked Language Modeling in Low-Resource Ancient Languages

Andrei Voinea

原文摘要

Masked language modeling for low-resource ancient languages remains challenging because pre-trained multilingual models lack exposure to these languages. We investigate rule-based linguistic constraints and hard negative mining for Sumerian, a language isolate not included in multilingual BERT’s training data. We build a hierarchical validator capturing subword, word, and part-of-speech patterns from 4,545 annotated sequences, using it to filter candidates and identify hard negatives for fine-tuning. Vanilla mBERT achieves 18.0% hit@10 accuracy. The validator alone improves this to 72.8%, while hard negative fine-tuning reaches 78.3%. Combining both yields 86.7%, a 68.7 percentage point improvement. Temporal generalization evaluation on tablets from 600 years earlier shows that both the hard negative mining and the validator alone improve performance, but the combined approach underperforms due to the validator’s period specific rules. These findings demonstrate that hard negative mining transfers across periods while explicit rule-based constraints provide strong in-domain improvements but limited cross-period generalization.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.70ACL Anthology · ACL 2026

Conformal LLM Routing with Distribution-Free Safety Guarantees

Iqtedar Uddin, André Bauer Illinois Institute of Technology

原文摘要

LLM routing directs queries to a cheaper model when it suffices and to an expensive model otherwise, reducing inference cost. Existing input-based routers optimize cost-performance trade-offs but provide no formal bound on how often the cheaper model fails among routed queries. We adapt a proactive conformal gate framework to LLM routing. A logistic regression gate trained on text embeddings predicts per-query safety, and Clopper-Pearson conformal calibration selects a routing threshold that guarantees the violation rate among routed queries stays below \alpha (the violation tolerance) with probability at least 1 - \delta (the confidence level). On two benchmarks covering math reasoning (GSM8K) and multi-domain knowledge (MMLU), routing between Mixtral-8x7B and GPT-4 (a 24.5\times cost difference), our method maintains the target \alpha within the \delta tolerance across a sweep from 0.05 to 0.50, while a validation-tuned baseline crosses the violation boundary on GSM8K. A feasibility analysis across all 10 RouterBench models reveals that routability is jointly model- and task-dependent. To our knowledge, this is the first input-based LLM router with distribution-free safety guarantees.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.71ACL Anthology · ACL 2026

Supervision versus Demonstration-Based In-Context Learning for Multiword Expression Classification

Sercan Karakaş University of Chicago, Yusuf Şimşek

原文摘要

Turkish idiomatic light verb constructions (LVCs) are challenging for multiword expression processing because they often share the same surface form as fully literal verb–object combinations while functioning as a single, partially idiomatic predicate. We frame Turkish LVC detection as a binary classification task (literal meaning vs. idiomatic meaning) and evaluate on a manually created controlled set (N=147) with matched negatives: out-of-domain random sentences and in-domain literal controls (NLVC), alongside LVC positives. We compare a supervised Turkish encoder baseline (BERTurk with a classifier head) to three instruction-tuned LLMs from different families under zero-shot, one-shot, and few-shot prompting, and analyze how demonstrations shift error profiles. In zero-shot, LLMs perform well on negatives but show very low LVC recall. One-shot prompting sharply improves LVC detection but can induce strong, model-specific biases (over- vs. under-predicting LVC). A richer few-shot prompt improves calibration and yields robust overall performance for GPT-OSS-20B and Qwen 2.5-14B. Overall, the results highlight substantial prompt sensitivity in Turkish metalinguistic classification: the supervised baseline remains competitive, while prompted LLMs can match or exceed it on LVCs with carefully constructed demonstrations. We release code, prompts, and evaluation materials to support reproducibility.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.72ACL Anthology · ACL 2026

When Models Hesitate: Answer Instability as a Label-Free Uncertainty Signal for LLMs

Jasper Meynard Arana Adamson University, Kristine Ann M. Carandang Professional Regulation Commission, Asian Institute of Management and Asian Institute of Management, Ethan Robert Casin, Christian Alis Asian Institute of Management, Christopher Monterola Asian Institute of Management

原文摘要

Large language models (LLMs) are increasingly deployed in high-stakes settings, yet reliably estimating when their outputs should be trusted remains an open challenge. Existing uncertainty estimation approaches—such as calibration, token-level probabilities, or semantic entropy—typically require access to model internals, additional supervision, or computationally intensive pipelines. We propose answer instability, defined as the variability of a model’s final answer across repeated stochastic generations of the same prompt, as a simple, label-free, and black-box uncertainty signal. Evaluated across three task types — reasoning, multiple-choice QA, and constraint-following — using four LLMs and 520 prompt-model pairs, our approach achieves performance competitive with semantic entropy while requiring no semantic similarity model. Our results show that instability strongly correlates with prediction errors and reliably discriminates correct from incorrect outputs. We further demonstrate its utility for selective prediction and targeted repair, improving reliability without access to internal probabilities or additional training.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.74ACL Anthology · ACL 2026

Thesis Proposal: Toward a Human-Centered and Perspective-Aware Framework for Reproducible ML Evaluation and AI Alignment

Deepak Pandita Rochester Institute of Technology, Christopher M. Homan Rochester Institute of Technology

原文摘要

Humans play a vital role at every stage of AI development, from data collection and curation to model development and evaluation. However, humans often disagree with each other and sometimes with themselves over time. It is essential to take disagreement into account when building human-centered AI systems, especially in domains where it is prevalent, such as AI safety, content moderation, or sentiment analysis. Disagreement often arises from subjective human opinion and can vary with one’s identity, beliefs, and social environment. Despite this, current LLM evaluation approaches frequently rely on aggregating labels (often via plurality voting) to represent consensus, thereby obscuring minority perspectives. By failing to account for human disagreement, these evaluation methods contribute to the reproducibility crisis in AI. Human feedback is also crucial for ensuring that AI systems align with human values. For these systems to be trustworthy, it is critical to ensure that they reflect diverse human values and perspectives. In this thesis proposal, we present a human-centered and perspective-aware framework for reproducible ML evaluation and AI alignment.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.75ACL Anthology · ACL 2026

One Panel Does Not Fit All: Case-Adaptive Multi-Agent Deliberation for Clinical Prediction

Yuxing Lu, Yushuhong Lin Dalian University of Technology, Jason Zhang Georgia Institute of Technology

原文摘要

Large language models applied to clinical prediction exhibit case-level heterogeneity: simple cases yield consistent outputs, while complex cases produce divergent predictions under minor prompt changes. Existing single-agent strategies sample from one role-conditioned distribution, and multi-agent frameworks use fixed roles with flat majority voting, discarding the diagnostic signal in disagreement. We propose CAMP (Case-Adaptive Multi-agent Panel), where an attending-physician agent dynamically assembles a specialist panel tailored to each case’s diagnostic uncertainty. Each specialist evaluates candidates via three-valued voting (KEEP/REFUSE/NEUTRAL), enabling principled abstention outside one’s expertise. A hybrid router directs each diagnosis through strong consensus, fallback to the attending physician’s judgment, or evidence-based arbitration that weighs argument quality over vote counts. On diagnostic prediction and brief hospital course generation from MIMIC-IV across four LLM backbones, CAMP consistently outperforms strong baselines while consuming fewer tokens than most competing multi-agent methods, with voting records and arbitration traces offering transparent decision audits.

cs.CL · cs.MA

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.77ACL Anthology · ACL 2026

LLMs for Now, Fine-Tuning for Later: An Ensemble Approach to Data Drift in Domain-Specific Tasks

Yuxuan Lu, Bingsheng Yao Northeastern University, Shao Zhang Shanghai Artificial Intelligence Laboratory, Yisi Sang Amazon, Yun Wang Microsoft Research, Hansu Gu Amazon, Peng Zhang, Tun Lu Fudan University 等 10 位作者

原文摘要

Deploying machine learning models in real-world domain-specific scenarios is challenged by the scarcity of expert annotations and by data drift, where the statistical properties of incoming data continuously evolve. Active Learning (AL) iteratively improves compact models with expert annotations but suffers from recurring cold-start degradation, while LLMs provide strong off-the-shelf performance yet cannot leverage newly accumulated labels, raising the question: how can we better leverage LLMs to assist the active learning process? Through an empirical study on five legal and biomedical datasets, we reveal a complementary temporal dynamic: LLMs excel during early and post-drift stages, while AL-assisted compact models eventually surpass them as annotations accumulate. Motivated by this finding, we propose an ensemble system that combines an LLM, an AL-assisted compact model, and an automatic switch module that routes predictions to the better-performing model in real time. Evaluated under simulated data drift on two mental health datasets, our system achieves 96–98% switch accuracy and consistently outperforms either model used alone.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.78ACL Anthology · ACL 2026

Thesis Proposal: When Does an Agent Know It Is Lost? Confidence Trajectory Analysis for Tool-Using LLMs

Zhenjiang Mao University of Florida

原文摘要

Large language model (LLM) agents that invoke external tools must make sequences of interdependent decisions, yet existing uncertainty quantification (UQ) methods treat each step in isolation, ignoring how confidence evolves and compounds across a full task trajectory.We propose a framework for trajectory-level confidence analysis in the tool-use agent setting. The thesis pursues three aims: (1) estimating action-level confidence by adapting step-wise UQ to the heterogeneous think-act-observe cycles of tool-using agents; (2) aggregating the diverse action space into semantically coherent action types to enable meaningful trajectory-level analysis; and (3) discovering temporal patterns in the resulting confidence trajectories that reliably predict task success or failure.We ground the work in standard tool-use benchmarks and expect the framework to expose early warning signals for agent failure and offer interpretable diagnostic tools for understanding when and why LLM agents lose confidence, with improved calibration of multi-step agentic pipelines as a secondary benefit.

cs.CL · cs.AI

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.79ACL Anthology · ACL 2026

Task Assignment meets Annotator Modeling: Human-LLM Collaborative Annotation with Constraints

Kei Moriyama The University of Tokyo, Tokyo Institute of Technology, Kouta Nakayama NII, Tokyo Institute of Technology, Yukino Baba University of Tokyo

原文摘要

Crowdsourced annotators and Large Language Models (LLMs) offer complementary, cost-effective ways to obtain labeled data, yet ensuring high label quality remains challenging.We observe that task features influence the accuracy of humans and LLMs, while real-world constraints, such as per-annotator assignment limits, further complicate allocation.Prior work typically addresses either task features or constraints, but not both.We present an integrated framework that (i) estimates per-task accuracy from task features using a learning from crowds model and (ii) incorporates these estimations into a linear programming formulation that assigns tasks under practical constraints. Experimental results demonstrate that the proposed method achieves accuracy comparable to that of baseline methods while satisfying given constraints.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.8ACL Anthology · ACL 2026

Reflection in the Dark: Exposing and Escaping the Black Box in Reflective Prompt Optimization

Shiyan Liu, Qifeng Xia, Qiyun Xia, Yisheng Liu, Xinyu Yu, Rui Qu

原文摘要

Automatic prompt optimization (APO) has emerged as a powerful paradigm for improving LLM performance without manual prompt engineering. Reflective APO methods such as GEPA iteratively refine prompts by diagnosing failure cases, but the optimization process remains black-box and label-free, leading to uninterpretable trajectories and systematic failure. We identify and empirically demonstrate four limitations: on GSM8K with a defective seed, GEPA degrades accuracy from 23.81% to 13.50%. We propose VISTA, a multi-agent APO framework that decouples hypothesis generation from prompt rewriting, enabling semantically labeled hypotheses, parallel minibatch verification, and interpretable optimization trace. A two-layer explore-exploit mechanism combining random restart and epsilon-greedy sampling further escapes local optima. VISTA recovers accuracy to 87.57% on the same defective seed and consistently outperforms baselines across all conditions on GSM8K and AIME2025.

cs.CL · cs.MA

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.80ACL Anthology · ACL 2026

Dynamic Meta-Metrics: Source-Sentence Conditioned Weighting for MT Evaluation

Luke Zhang, Justin Vasselli, Aditya Khan University of Toronto, York Hay Ng, En-Shiun Annie Lee

原文摘要

We propose Dynamic Meta-Metrics (DMM), a framework for machine translation evaluation that learns source-sentence conditioned combinations of existing metrics. Rather than relying on a single static ensemble or language-specific weighting, DMM adapts the metric combination based on properties of the source segment. We study hard conditioning, which fits an interpretable combiner per cluster, and an exploratory soft-conditioned extension whose weights vary continuously with source-cluster responsibilities. We evaluate DMM on the WMT Metrics Shared Task data across multiple language pairs using pairwise agreement measures at the system and segment levels. Across settings, MLP-based combinations outperform linear and Gaussian process-based ensembles, and introducing soft conditioning yields gains over linear models.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.81ACL Anthology · ACL 2026

Tonal Salience in Cognitive Decline: In-Context MCI Detection with Multimodal LLMs

Christopher Song, Abdullah P. Rashed Ahmed Brown University

原文摘要

Speech-based screening for mild cognitive impairment offers a highly accessible way to detect early cognitive decline. While most existing work focuses on English, cross-linguistic research is emerging to examine how cognitive decline manifests across languages. Studies on the Interspeech 2024 TAUKADIAL dataset, comprising English and Chinese speech recordings, consistently report higher classification performance on Chinese, yet the cause of this cross-lingual discrepancy remains unexplored. We examine this gap using Gemini 2.5 Pro, a multimodal large language model, using zero-shot and in-context-learning (ICL) paradigms. We hypothesize that this disparity is rooted in language typology: in tonal languages like Chinese, pitch encodes lexical meaning in every syllable (tone), whereas in non-tonal languages like English, pitch carries no lexical function. To test this, we pitch-flattened audio from TAUKADIAL and compared how classification performance changed across both languages. We found that Chinese classification degraded significantly under both zero-shot and ICL conditions (-4.78 and -5.92 UAR, respectively), while English performance increased (+0.11 and +2.98 UAR), implicating tonal pitch as the cross-lingual advantage. These findings suggest language typology should inform the design of audio-based cognitive screening tools, with raw audio preferred for tonal languages and text for non-tonal languages, a distinction critical for developing equitable cross-linguistic screening.

cs.CL · cs.CV

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.82ACL Anthology · ACL 2026

What Moves the Pareto Frontier in Tool-Using Agents? A Compute-Aware Study of ReAct Variants

Rishi N. Simhadri

原文摘要

Tool-using LLM agents are typically compared by accuracy alone, despite deployments being constrained by inference cost. We present a budgeted evaluation of common strategies for improving ReAct-style tool agents (multi-sample aggregation, iterative self-correction, and post-hoc answer revision) using Pareto analysis of cumulative accuracy versus token budget on three benchmarks (HotPotQA, FEVER, GSM8K) with Gemini 2.5 Flash. All experiments use three random seeds (N=500 per seed for HotPotQA/FEVER; N=1,015 for GSM8K); budgeted curves are computed post hoc from per-instance token logs. In our offline evaluation, Reflexion attains the highest accuracy on tool-heavy benchmarks (HotPotQA, FEVER), while CoT-SC leads on GSM8K. Reflexion’s reported token costs are optimistic lower bounds because retries are stopped using ground-truth feedback, and its accuracy is similarly optimistic: a deployment without access to ground-truth labels would not achieve the same accuracy because the gold-label stopping criterion would be unavailable; both costs and accuracy would differ in practice. Sampling-based approaches often spend 3-5x more tokens for comparatively small gains on tool-heavy tasks. GSM8K, a pure-math benchmark with minimal tool interaction, shows substantially larger gains for CoT-SC, TCAR, and Reflexion, larger than on tool-heavy benchmarks, though less sharply separated than headline accuracy alone would suggest, consistent with repeated tool trajectories being an important contributor to the observed efficiency gap in our tool-heavy settings. We provide a compute-aware evaluation protocol (frontier analysis and marginal-cost metrics) and practical guidance for choosing agent designs under different budget regimes.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.83ACL Anthology · ACL 2026

Filling the Long Tail: Structure-Aware Curriculum-Gap Completion for Medical Education with LLMs

Wenjie Lin Purdue University

原文摘要

Medical education resources are dense for common diseases but often sparse for under-covered conditions, atypical presentations, and fine-grained concept distinctions. This creates curriculum gaps that are difficult to repair manually, especially in long-tail domains where structured teaching materials are limited. We introduce Curriculum-Gap Completion (CGC), a new task for Large Language Model (LLM)-based medical education in which a model reconstructs missing educational units from a partially specified curriculum graph. Given topic nodes, pedagogical relations, and structured teaching slots, the model predicts omitted concepts, restores missing instructional links, and completes automatically verifiable teaching content. We instantiate this setting in a long-tail medical case study (hyperhidrosis) and evaluate five LLMs under three methods: direct prompting, retrieval-augmented prompting, and our proposed Structure-Aware Curriculum-Gap Completion (SACGC) framework. Across models, SACGC achieves the strongest overall performance, with the largest gains on structurally demanding masking settings. Ablation results show that explicit graph structure is the most important component, while schema constraints provide additional benefit. These findings suggest that LLMs are better suited for reconstructing an under-specified educational structure than for unrestricted medical tutoring, and they motivate CGC as a new natural language processing (NLP) problem for healthcare education.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.84ACL Anthology · ACL 2026

Mechanistic Analysis Of Universality: Numerical Comparison Circuits Across Transformer Architectures

Arya Bhardia, Julian Ramirez, Siddhanta Verma, Karen Mkrtchyan

原文摘要

Transformer language models reliably achieve high accuracy on many reasoning tasks; however, their internal mechanisms are not fully understood. Mechanistic interpretability seeks to remedy this gap by identifying task circuits within individual models, but it is unclear whether such circuits generalize across model families and scales. In this work, we study the universality of circuits through the lens of numerical comparisons, a simple and controlled task that enables clean and causal interventions. We conduct experiments on a set of transformer models spanning different families and sizes from 1.7b to 9b parameters. We find that models within the Qwen family exhibit a highly consistent circuit structure across architecture and scale, featuring localized attention heads that write a task relevant signal. In contrast, models from other families show qualitatively different implementations, where task relevant information emerges much earlier and is distributed across components as opposed to being concentrated within a small set of attention heads. These results serve as evidence that task behavior similarities do not imply mechanistic universality and highlight the necessity for cross model comparisons to claim generalization of internal circuits.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.85ACL Anthology · ACL 2026

How Hard is Math? Using Quantitative Metrics to Measure LLM Alignment to Human Intuitions of Difficulty

Micah Helzerman University of Michigan - Flint, Steven R Wilson University of Michigan - Flint, Cam McLeman University of Michigan - Flint

原文摘要

Modern LLMs have demonstrated advanced reasoning skills, including the ability to solve Olympiad-level mathematics problems. While solving more and more difficult problems is a hallmark of LLM progress, less attention has been placed on how “difficulty” is operationalized in the context of LLM problem solving tasks. This is particularly relevant in educational contexts where teachers or students may ask LLMs for “easy” or “hard” questions. In this paper, we explore various quantitative measurements from LLM-generated solutions and evaluate their inter-correlations, as well as their correlation to human-annotated difficulty scores. We find moderate correlations between metrics using log probabilities and output lengths, including some that are more strongly correlated to difficulty than LLM accuracy. We also train ModernBERT to predict difficulty scores, leading to reasonable accuracy within a given benchmark, but decreased performance when generalizing to other math benchmarks. Finally, to explore connections between difficulty scores and human performance, we collect problems, human solutions, and human performance data from the Putnam competition. We find poor alignment between LLM metrics and human-assigned difficulty scores, despite strong correlations between those scores and human performance on the problems.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.86ACL Anthology · ACL 2026

Fine-Grained Semantic Comparison of Legal Documents using LLMs

Elisei Rykov CompSem group, Nikolay Ivanov, Maria Bandulevich, Kseniia Petrushina, Valentin Malykh International IT University, Vasily Konovalov, Alexander Panchenko S-NLP group, Ilseyar Alimova Kazan Federal University

原文摘要

Frequent revisions of complex regulatory documents in large organizations often introduce inconsistencies and contradictions that are difficult for lawyers and auditors to detect manually. Existing tools rely on character-level diffs and therefore miss paraphrases and semantic shifts. We introduce LegDiff, a novel benchmark for evaluating span-aware semantic comparison of legal texts, and use it to investigate the ability of large language models to detect semantic changes beyond token- and character-level matching. LegDiff comprises manually annotated pairs of legal paragraphs drawn from different documents. In addition, we present a pipeline to generate synthetic training data that aligns with the manual annotations and mirrors the structure and label distribution of the manually curated benchmark, and a visualization tool for clearly displaying detected differences and inconsistencies. The dataset, code, and a visualization tool are publicly available to facilitate reproducibility and further research (https://github.com/s-nlp/SLeDoC).

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.87ACL Anthology · ACL 2026

From Fluent to Useful: Generative AI That Models Purpose, Audience, and Presenter for Scientific Communication

Ishani Mondal

原文摘要

Modern generative AI produces fluent text,polished slides, and clean diagrams — yetstill fails when an artifact must serve a specificpurpose for a specific reader, used by aspecific presenter. The missing piece is notfluency but a model of why content is beingproduced, for whom (presenter and audiencealike), and how it should adapt as goalsshift. My completed and published work developsfive systems across the scientific communicationpipeline: ADAPTIVE IE for intentdrivenextraction; Persona-Aware Slide Generationfor audience reframing rather than blanketsimplification; GPA for reconciling divergentgroup preferences; SciDoc2Diagrammer-MAF,whose multi-aspect critics distinguish purposefulabstraction from genuine omission or hallucination;and SMART-Editor, which modelscascading edits across multimodal layouts. Togetherthey show that aligning with intent, audience,and structure is necessary—but cannotanswer whether the resulting artifacts actuallycommunicate. I therefore propose three directionsin priority order: (RQ1) a goal-drivenframework that measures the educational utilityof document-to-video generation throughIRT-calibrated diagnostic questions, validatedagainst measured learning outcomes and accompaniedby inter-annotator agreement studieson human effectiveness judgments; (RQ2)presenter-side personalization that treats thepresenter—not just the audience—as a firstclassuser; and (RQ3) a unified SuperPersonalizationbenchmark for transferable user preferences.RQ3 is scoped to be deferrable topost-dissertation work if RQ1 expands. Thethesis shifts the target from generative AI thatproduces content that looks correct to systemswhose outputs demonstrably communicate

cs.CL · cs.CV

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.88ACL Anthology · ACL 2026

Beyond Discrete Search: Divergent Thinking as Intention Optimization in Latent Space

Mateusz Bystroński, Grzegorz Piotrowski, Tomasz Jan Kajdanowicz Wroclaw University of Science and Technology

原文摘要

We argue that LLM-based coding agents frequently fail to solve problems that lie within the model’s capacity and the bottleneck is often the conditioning context rather than the model itself. We formalize this for the full class of Turing-computable problems with verifiable specifications and introduce a framework that recasts coding as optimization overconditioning contexts that influence the generation of natural-languagesolution intentions. Guided by execution feedback, the method searches thiscontinuous context space to steer a coding agent toward correct solutions. The method operates as a plug-in layer that can wrap any coding agent without modifying its architecture or weights. On SWE-Bench Verified, our method raises the resolution rate of a weak, quantized 24B open-weight model to parity with frontier models +25× its size.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.89ACL Anthology · ACL 2026

Boosting Self-Consistency with Ranking

Maria Marina, Daniil Moskovskiy, Sergey Pletenev Applied AI Institute, Mikhail Salnikov Independent, Alexander Panchenko S-NLP group, Viktor Moskvoretskii EPFL - EPF Lausanne

原文摘要

Self-consistency improves large language models by sampling multiple reasoning paths and selecting the most frequent answer, but majority vote often fails to recover correct answers that are already present among samples. In this work, we reformulate answer selection in self-consistency as a ranking problem. Instead of relying on a single uncertainty or confidence signal, we train a lightweight reranker to score candidate answers using five carefully designed features that capture answer-level frequency, semantic centrality, and reasoning-trace consistency. We instantiate this approach with a LambdaRank model and evaluate it on three datasets under a range of test-time budgets. Across datasets, the proposed method consistently achieves a better accuracy-efficiency trade-off than standard self-consistency and strong baselines, with particularly large gains on question answering benchmarks. Further analysis shows that the proposed features are individually useful and, more importantly, complementary, highlighting the value of learning to combine multiple informative signals for test-time answer selection.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.9ACL Anthology · ACL 2026

Thesis Proposal: LLMs post-training for multilingual medical tasks. Instruction-Tuning, Continual-Pretraining or Reasoning?

Pietro Ferrazzi, Alberto Lavelli Fondazione Bruno Kessler, Bernardo Magnini Fondazione Bruno Kessler

原文摘要

Adapting Large Language Models to the medical domain remains an active area of research, with multiple strategies proposed to leverage annotated and unannotated data effectively. In this work, we propose a thesis outline to compare three common adaptation approaches—Instruction Tuning, Continual Pretraining, and Reasoning-oriented Training. We identify 5 dimensions to analyse: i) the interaction between the adaptation technique and the tasks; ii) the impact of the data size on the downstream performance; iii) the differences between datasets required by the three techniques; iv) the impact of the techniques given the model size; v) the impact of the techniques given the language.We construct an evaluation framework composed by 5 multilingual medical NLP tasks (named entity recognition, relation extraction, question answering, case report form filling, argument mining), spanning on 21 datasets in English, Italian, and Spanish, for a total of 61 combinations of language and sub-task.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.90ACL Anthology · ACL 2026

LLM-Based Zero-Shot Soft Labeling for Anticipating Disagreement in Negotiation Dialogues

Ken Watanabe Tokyo University of Agriculture and Technology, Tokyo Institute of Technology, Katsuhide Fujita Tokyo University of Agriculture and Technology

原文摘要

Negotiation involves complex emotional and strategic dynamics that pose challenges for AI agents in negotiation dialogues. This paper proposes a zero-shot soft-labeling method using an large language model-based embedding model and verifies its performance on negotiation dialogues. Furthermore, it examines the performance of predictive model training on rule-based annotated hard and soft labels obtained by the proposed method for the task of predicting whether agreement will be reached from partial dialogues, namely, final disagreement anticipation in negotiation mid-dialogues (FDANMD). Soft labeling obtained by the proposed method showed a maximum HIT@3 score of 0.87 against rule-based annotated hard labels, whereas failure cases also demonstrated the limitations of rule-based annotation. Furthermore, using ROC AUC, evaluations of FDANMD across three datasets (CB, DN, and JI) with negotiation progress rates of 0.25, 0.5, and 1.0 revealed that soft labeling is particularly effective at low negotiation progress rates and also offers superior performance on individual datasets and unseen datasets for models trained on multiple datasets. These results motivate the use of soft labeling to incorporate the complexity of negotiation dialogues into intermediate representations and support the generalizability of zero-shot soft labeling and generalizable predictors across a wide range of negotiations beyond known domains.

cs.CL · cs.AI

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.91ACL Anthology · ACL 2026

Analysis of the Neglect-Zero Effect in Large Language Models

Jin Tanaka, Daiki Matsuoka, Ryoma Kumon The University of Tokyo and RIKEN, Hitomi Yanaka RIKEN, Tohoku University, Tokyo Institute of Technology and the University of Tokyo

原文摘要

We investigate the extent to which the language processing of LLMs resembles human cognitive processes, focusing on a human cognitive bias called the neglect-zero effect. This effect refers to the human tendency to ignore zero-models, which are configurations that render a proposition vacuously true by virtue of an empty set. We focus on two types of inferences driven by the neglect-zero effect, and examine how LLMs process these inferences by comparing their behavior with that in an inference that does not involve the neglect-zero effect. For this purpose, we employ a paradigm based on structural priming, where recent exposure to a preceding sentence (the prime) facilitates the processing of a subsequent sentence (the target) due to their structural similarity. We prepare primes to force LLMs to consider the zero-model, and analyze whether they also consider it in the target. The results suggest that the neglect-zero effect may not occur in the LLMs analyzed in this study. Our code is available at https://github.com/ynklab/neglect_zero.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.92ACL Anthology · ACL 2026

Morphology-Aware Multi-Granularity Representation Learning for Agglutinative Languages

Zhonghao Zhang Inner Mongolia University of Technology, Na Liu Inner Mongolia University of Technology, Jiajia Ma Inner Mongolia University of Technology, Nier Wu Inner Mongolia University of Technology, Guiping Liu Inner Mongolia University of Technology

原文摘要

Low-resource agglutinative languages, characterized by rich morphological inflection and severe vocabulary sparsity in corpora, have long posed numerous challenges in the field of representation learning. Word-level representations preserve semantic integrity but struggle to handle sparse surface forms, whereas morpheme-level representations, though easier to learn, often lack holistic semantic information. Existing multi-granularity methods are typically modeled at the word and phrase levels, with very limited application to low-resource agglutinative languages. Focusing on the morphemes of agglutinative languages, this paper proposes MAGNet, a morphology-aware gated multi-granularity pre-training framework. At the morpheme granularity, this framework leverages morphological knowledge and integrates morpheme segmentation with morphological tagging to construct fine-grained representations. It further introduces a morphology-aware masked language modeling objective to facilitate the model in learning functional morphological regularities. Meanwhile, at the word granularity, a word-level encoder is employed to capture contextual semantics and maintain its semantic coherence.Finally, a gated fusion mechanism dynamically fuses representations of different granularities according to the context. Experiments conducted on two low-resource agglutinative languages, Mongolian and Turkish, for the tasks of dependency parsing and named entity recognition (NER) demonstrate that our method achieves consistent performance improvements over strong baseline models. Ablation studies further validate the complementary roles of morphological tagging and whole-word modeling in efficient representation learning.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-srw.94ACL Anthology · ACL 2026

Processing Inconsistency Predicts Language Competence: LLM Evaluation Without Answer Labels on Turkic Languages

Ilya Galyukshev ITMO University, Ilseyar Alimova Kazan Federal University

原文摘要

Most languages lack labeled evaluation benchmarks for large language models (LLMs). Creating such benchmarks requires native speakers, domain expertise, and answer annotation—resources unavailable for the vast majority of languages. We investigate whether a model’s internal processing signals—such as generation entropy and tokenizer statistics—correlate with its actual accuracy on a language, with the long-term goal of estimating language competence without labeled data. Our key observation is that for languages a model does not know, both tokenizer segmentation and generation entropy become highly variable across questions, whereas for known languages they remain consistent. We call this the inconsistency hypothesis and test it on 11 instruction-tuned LLMs (1B–70B parameters) across 14 language–script varieties (12 Turkic plus English and Russian controls). We extract over 25 processing features per model–language pair; individually, even the strongest correlate only moderately with accuracy (Pearson |r| up to 0.55). Yet combining just three complementary features—a tokenizer coverage ratio, entropy variability, and the model’s English/Russian benchmark score—explains 75% of accuracy variance in leave-one-language-out evaluation, nearly doubling the 44% explained by a model-mean baseline. The variability of processing signals (standard deviation) consistently outperforms mean values as a predictor across all five model families, but only for greedy-pass measures; sampling-based measures show no such pattern.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) · 2026 · 正式发表