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 09:37 — 2026-11-12 09:37（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.33ACL Anthology · ACL 2026

Confidence as a Tie-Breaker: Reassessing Multilingual Hedging Bias in LLM-as-a-Judge Evaluation

Rajashik Datta, Sanjan Baitalik

原文摘要

LLM judges are often used to score generated answers, but their decisions may be affected by surface style rather than semantic correctness. We introduce PolyJudge-Uncertain, a controlled benchmark for studying multilingual hedging effects in LLM-as-a-judge evaluation. The benchmark contains 5,120 short factual QA instances across English, Hindi, Hinglish, and Bengali, balancing assertive versus hedged style and correct versus incorrect answers. A small pilot suggested a large pointwise penalty against hedged answers. After repairing multilingual templates and adding quality-control checks, this pointwise effect largely disappears: final pointwise accuracy is 99.8%, with no meaningful assertive-hedged gap. The robust remaining effect is pairwise: when two answers are equally correct and differ only in style, the judge prefers the assertive answer in 1,276 of 1,280 cases. We interpret this as a protocol- and task-specific assertiveness preference, not as a universal bias against hedging. Our findings highlight benchmark auditing as a central requirement for multilingual judge-bias research.

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.34ACL Anthology · ACL 2026

BanglaSTEM: A Parallel Corpus and Term-Weighted Evaluation for Technical Bangla-English Translation

Kazi Reyazul Hasan Bangladesh University of Engineering and Technology, A. B. M. Alim Al Islam Bangladesh University of Engineering and Technology, Muhammad Abdullah Adnan Bangladesh University of Engineering and Technology

原文摘要

Large language models excel at technical problem solving in English but struggle when questions are posed in Bangla. While translation offers a practical solution, existing Bangla-English systems frequently mistranslate specialized terminology, altering problem semantics and degrading downstream performance. We present BanglaSTEM, a dataset of 5,000 Bangla-English sentence pairs covering computer science, mathematics, physics, chemistry, and biology. Our pipeline extracts matching passages from official bilingual curriculum textbooks using OCR, then uses LLMs to align sentences and mark technical terms. These aligned examples serve as few-shot prompts for generating over 12,000 new translation pairs from LLMs, avoiding copyright issues. Human evaluators then select the best 5,000 pairs that correctly preserve technical terminology. We also test a term-weighted BLEU metric that gives higher weight to technical words, since standard metrics treat terminology errors and common word errors equally. We show that our weighted metric correlates better with downstream accuracy in code generation and math solving, while standard BLEU gives high scores even for wrong translations. The full implementation, dataset, and model will be made publicly available.

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.36ACL Anthology · ACL 2026

Disentangling the Effects of Unlearning in Measuring Parametric Faithfulness of Chain-of-Thought

Ryo Mitsuhashi, Gaku Morio Hitotsubashi University and Hitachi, ltd., Ayana Niwa Mohamed bin Zayed University of Artificial Intelligence, Masahiro Kaneko NII LLMC and Mohamed bin Zayed University of Artificial Intelligence, Kentaro Inui MBZUAI, RIKEN and Tohoku University, Terufumi Morishita Hitachi, ltd, Yuta Koreeda Hitachi, Ltd., Yasuhiro Sogawa Hitachi, Ltd.

原文摘要

Chain-of-Thought (CoT) in large language models (LLMs) has been widely debated in terms of whether it faithfully reflects an internal reasoning process of models. Parametric faithfulness is a recently proposed metric that uses unlearning to assess whether a model encodes parametric beliefs corresponding to a reasoning chain. This paper refines this metric by accounting for the unintended artifacts of unlearning. We introduce control tasks that unlearn irrelevant knowledge and word-shuffled content and show that these control tasks yield substantial parametric faithfulness values, suggesting the non-negligible effect of unlearning. We also found that control tasks help explain the significant variations in parametric faithfulness observed across different model sizes and CoT lengths. We conclude that the effects of unlearning need to be considered when measuring parametric faithfulness.

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.37ACL Anthology · ACL 2026

FedPAGR: Federated Prototype Alignment via Geometric Refinement for Heterogeneous Architectures

Kris Prasad, Md Abdullah Al Hafiz Khan Kennesaw State University

原文摘要

Federated learning with heterogeneous client architectures cannot rely on parameter aggregation. Prototype-based methods address architectural heterogeneity by exchanging class-level representations, but naively averaging prototypes across non-IID clients leads to semantic drift and poor inter-class separation. We propose FedPAGR, a framework where heterogeneous clients project their features into a shared consensus space and exchange class prototypes with a central server. The server refines aggregated prototypes through a geometric regularization objective that enforces agreement with client submissions and inter-class angular separation. Clients anchor their classifiers to the refined prototypes and train with a composite objective combining classification, prototype alignment, and entropy regularization. We evaluate FedPAGR across multiple domains, including four image benchmarks and a clinical NLP task using heterogeneous ClinicalBERT variants, with five architectures per federation under severe label heterogeneity (\alpha{=}0.1). FedPAGR achieves the highest ensemble accuracy across all four image datasets and the highest local test accuracy on low-class and clinical tasks, including a 4.99-point improvement over the strongest baseline on MIMIC-IV, while remaining competitive on high-class benchmarks.

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.38ACL Anthology · ACL 2026

Sentiment Analysis of Yelp Review Dataset: A Comparative Study of Machine Learning Methods

Krishna Thakar, Mohamed Abu Sheha Southeast Missouri State University, Emmanuel Thompson Southeast Missouri State University

原文摘要

Sentiment analysis involves analyzing text to determine whether the sentiment expressed is positive, negative, or neutral. In the context of online reviews, such as those on Yelp, sentiment analysis helps businesses assess customer satisfaction and identify areas for improvement. Given the large volume of user-generated content, restaurants often struggle to extract actionable insights from feedback, making sentiment analysis an efficient tool for categorizing reviews and highlighting customer concerns. This study focuses on sentiment analysis of Yelp reviews. The main research question is: How can Natural Language Processing (NLP) combined with statistical machine learning methods be applied to classify sentiment in Yelp reviews and provide actionable insights for improving customer satisfaction, service quality, and business performance? The study used 21,000 Yelp reviews, utilizing NLP approaches - tokenization, stop-word removal, and vectorization. Comparative classification predictive modeling and analysis were done across traditional machine learning (Logistic Regression, Support Vector Machine (SVM), Naïve Bayes, Random Forest), deep learning methods (CNN, LSTM, BiLSTM, GRU, RNN), and an advanced transformer-based (RoBERTa) model. Results showed that RoBERTa outperformed the other candidate methods. These findings highlight the potential of advanced NLP techniques to offer businesses practical ways to address customer complaints, enhance service quality, and drive overall business performance.

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.39ACL Anthology · ACL 2026

Semantic Span Annotation: An Exploratory Study of LLM Annotation

Tejas Goyal, Dhriti Krishnan, Anuj Gupta, Jaromir Savelka Carnegie Mellon University

原文摘要

Structured span extraction research is siloed by context length, annotation task, and domain, making it difficult to assess how well large language models (LLMs) generalize across realistic extraction settings. We introduce SSA (Structured Span Annotation), a unified evaluation framework bringing together five datasets across four domains: finance, biomedicine, affective analysis, and privacy, under a common JSONL format with character-level offsets. We conduct an exploratory study evaluating seven models (three closed, four open-weight) under three prompting configurations: zero-shot, definition-augmented, and few-shot, formulating extraction as inline XML generation where models reproduce the document with tagged spans. Our results reveal two distinct performance regimes: on tasks requiring complex ontology reasoning, zero-shot performance is near zero (e.g., 0.00% F1 on FiNER-139) but improves substantially with label definitions (e.g., Claude Opus 4.6 rises from 8.8% to 57.5% F1); on pattern-based tasks like PII detection, definitions consistently hurt performance across all models. These findings suggest that prompting strategy must be matched to task structure, and that unified evaluation frameworks spanning varied domains and input lengths are essential for understanding LLM extraction capabilities.

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.4ACL Anthology · ACL 2026

Emergence of Minimal Circuits for Indirect Object Identification in Attention-Only Transformers

Rabin Adhikari Universität des Saarlandes

原文摘要

Mechanistic interpretability aims to reverse-engineer large language models (LLMs) into human-understandable computational circuits. However, the complexity of pretrained models often obscures the minimal mechanisms required for specific reasoning tasks. In this work, we train small, attention-only transformers from scratch on a symbolic version of the Indirect Object Identification (IOI) task, a benchmark for studying coreference-like reasoning in transformers. Surprisingly, a single-layer model with only two attention heads achieves perfect IOI accuracy, despite lacking MLPs and normalization layers. Through residual stream decomposition, spectral analysis, and embedding interventions, we find that the two heads specialize into additive and contrastive subcircuits that jointly implement IOI resolution. Furthermore, we show that a two-layer, one-head model achieves performance comparable to that of a multi-head model by composing information across layers primarily through query-key interactions. These results demonstrate that task-specific training induces highly interpretable, minimal circuits, offering a controlled testbed for probing the computational foundations of transformer reasoning.

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.40ACL Anthology · ACL 2026

Thesis Proposal: An Explainable Multimodal Framework for Detecting Harmful Content in Code-Switched Children’s Media

Juliana Isabelle A. Guillermo, Jasper Kyle Catapang Tokyo University of Foreign Studies, Nathaniel Oco De La Salle University

原文摘要

Current automated content moderation systems fail to protect children from harmful YouTube content, particularly in under-resourced, code-switched settings. These systems are often text-only, English-centric, and operate as ’black boxes,’ lacking the multimodal understanding and transparency needed for effective moderation. This thesis proposes a novel hybrid framework for the explainable multimodal detection of harmful content in videos with code-switching. The proposed framework integrates a fine-tuned classifier for accurate, scalable detection with an LLM-powered module that synthesizes the classifier’s internal evidential signals (e.g., text attention and visual heat maps) to generate faithful, human-readable rationales for each decision. As a primary case study, the framework will be developed and validated on an English–Filipino code-switched dataset. Expected contributions include a new dataset publicly available under controlled access (de-identified transcripts, blacked-out frames, extracted feature representations, and metadata via data-sharing agreement) and a blueprint for building more equitable, transparent, and trustworthy AI safety systems.

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.41ACL Anthology · ACL 2026

Test-Time Strategies for More Efficient and Accurate Agentic RAG

Abhinav Sharma, Brian Zhang, Deepti Guntur University of Massachusetts at Amherst, Zhiyang Zuo MavenBio, Shreyas Chaudhari University of Massachusetts at Amherst, Wenlong Zhao University of Massachusetts at Amherst, Franck Dernoncourt Adobe Research, Puneet Mathur Adobe Systems 等 10 位作者

原文摘要

Retrieval-Augmented Generation (RAG) systems face challenges with complex, multi-hop questions, and iterative agentic frameworks such as Search-R1 (Jin et al., 2025) have been proposed to address these complexities. However, such approaches can introduce inefficiencies, including repetitive retrieval of previously processed information and challenges in contextualizing retrieved results effectively within the current generation prompt. Such issues can lead to unnecessary retrieval turns, suboptimal reasoning, inaccurate answers, and increased token consumption. In this paper, we investigate test-time modifications to Search-R1’s open-source Qwen2.5-7B pipeline to mitigate these identified shortcomings. Specifically, we explore the integration of two components and their combination: a contextualization module to better integrate relevant information from retrieved documents into reasoning, and a de-duplication module that replaces previously retrieved documents with the next most relevant ones. We evaluate our approaches using the HotpotQA (Yang et al., 2018) and the Natural Questions (Kwiatkowski et al., 2019) datasets, reporting the exact match (EM) score, an LLM-as-a-Judge assessment of answer correctness, and the average number of turns. Our best-performing variant (contextualization) achieves a 5.6% increase in EM score and reduces the average number of turns by 10.5% compared to the Search-R1 baseline. While contextualization itself introduces additional LLM calls, our results demonstrate improved answer accuracy and reduced retrieval load.

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.42ACL Anthology · ACL 2026

Eye Movement Features Can Predict Human Preferences on Machine-Generated Texts

Xiaoshan He, Xiaoqun Liu Michigan State University, Haodong He, Yu Wang, Yang Xu Southern University of Science and Technology

原文摘要

Eye movement offers valuable insights into human visual attention during assessment of machine-generated texts, yet existing research and resources in this area are limited. To bridge this gap, we introduce Gaze Responses for Evaluating AI Texts (GREAT), a comprehensive dataset capturing human eye-movement features during screen reading of passages generated by large language models (LLMs). The dataset includes raw eye-movement recordings, reading-time measurements, and post-reading evaluations for LLM-generated passage pairs, alongside rigorous validation metrics. The collected eye-movement features demonstrate strong explanatory power in predicting text quality. When integrated with negative log-likelihood (NLL), a commonly used metric for evaluating text quality, it substantially enhances model performance across all standard statistical criteria. These findings demonstrate that eye-movement can act as an effective source of information that complements probabilistic metrics, for the task of automatic text quality assessment. The full dataset and some processing code are publicly available at https://github.com/qwurd231/GREAT.

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.43ACL Anthology · ACL 2026

Thesis Proposal: Diagnosing and Mitigating Semantic Interference in Script-Sharing Low-Resource Language Models: A Case Study on Square Bai Script

Jingting Zheng Tianjin University and Kunmimg University of Science and Technology, Deyi Xiong Tianjin University

原文摘要

Multilingual language models now cover more languages than ever, yet script-sharing low-resource languages remain vulnerable to failures driven by script and dominant-language priors. This dissertation studies one such failure mode, semantic interference, in Square Bai Script, where many forms resemble Chinese characters but often differ in meaning. We argue that current adaptation pipelines underperform not only because Bai is low-resource, but because they treat visible overlap as safe transfer by default. Building on an expert-validated corpus of 28,382 Bai-Chinese sentence pairs, an out-of-domain epigraphic set and a reproducible encoding pipeline, the dissertation will (1) diagnose semantic interference, (2) compare adaptation strategies under realistic compute constraints, and (3) estimate when shared-script transfer helps or harms adaptation. The long-term goal is Bai-capable understanding and generation. The dissertation addresses the prerequisite problem of safe and effective adaptation in a script-sharing low-resource setting.

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.44ACL Anthology · ACL 2026

Does Locality Cost in Polish Medical Text Classification? Duplicate-Aware Evaluation of Federated Learning

Daniel Cieślak Gdańsk University of Technology, Andrzej Czyżewski

原文摘要

Federated learning is often framed as a practical trade-off in clinical NLP: safer data handling at the cost of lower predictive performance. We revisit this assumption in a benchmark-specific study of Polish medical text classification. A key issue is evaluation granularity: the test split contains 10,634 rows but only 670 unique normalized text hashes, with 18 inconsistent groups removed in strict grouped evaluation. We therefore compare centralized and federated training under both conventional instance-level scoring and a stricter hash-level protocol that controls duplicate inflation. In the strongest reported settings, federated training matches or slightly exceeds the centralized baseline, reaching instance-level Macro-F1 of 0.8826 ± 0.0177 versus 0.8689 ± 0.0124, and hash-level Macro-F1 of 0.8908 ± 0.0220 versus 0.8841 ± 0.0078. The claim is deliberately narrow: we do not argue that federated learning is generally superior to centralized training, nor do we claim formal privacy guarantees. Rather, we show that in this duplicate-heavy Polish medical text benchmark, conclusions about locality depend strongly on evaluation hygiene.

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.45ACL Anthology · ACL 2026

Analyzing Hate Speech Amplification on Fringe Platforms

Anika Ghosh Basu, Humberto Jesus Carlon, Junyi Liu

原文摘要

Fringe platforms like Gab harbor high volumes of hate speech due to minimal moderation and insular communities. Our study examines thefactors that determine how hate speech amplifies on these platforms. We prepared a novel dataset of 5K+ threads and 50K+ responses from four fringe platforms (Gab, 4chan, Stormfront, and Vanguard), including both structural features (e.g., timestamps, metadata) and con-tent features (e.g., original text, hate intensity of posts), where hate speech amplification was measured using platform-specific engagement metrics. We trained both Generalized Linear Models and Gradient Boosted Tree models to estimate how several features influence the amplification of hate speech on fringe platforms, and used Shapley value estimates to identify the relative importance of the features. Our analysis shows that research insights from social network analysis (SNA) of mainstream sites like X do not directly generalize to fringe platforms. For instance, our experiments show that using features like thread structure and disagreements in early response windows can give up to 74% lift in Root Mean Squared Error (RMSE) of predicting reply counts for hateful posts on fringe platforms, compared to a baseline model that has features like hate intensity and thread age (which would be considered predictive by regular SNA methods).

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.46ACL Anthology · ACL 2026

The Silence of the Facts: Popularity as a Barrier to Machine Unlearning

Anna Borisiuk DS-NLP Group, Andrey Savchenko HSE University, Alexander Panchenko S-NLP group, Elena Tutubalina

原文摘要

Machine Unlearning is a valuable ability of LLMs, enabling the removal of unsafe, outdated, or private information. Existing unlearning methods, however, are often evaluated under the assumption that all facts are equally challenging to forget. Controllable knowledge removal is essential for reliable NLP systems. In this paper, we investigate whether fact popularity influences the efficiency of LLM unlearning. To answer this question, we build UNLamb benchmark designed to systematically investigate this relationship. It consists of 11.6k question-answer pairs derived from real-world knowledge in Wikidata, explicitly partitioned into rare and popular facts. Using this benchmark, we perform a comprehensive evaluation of state-of-the-art unlearning algorithms on a set of models of different sizes. We conduct a comprehensive analysis of four unlearning methods across three validation sets and two LLMs. We show that larger models struggle more to forget popular entities, often damaging related knowledge in the process. In contrast, it is much easier to remove rare facts without side effects.

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.47ACL Anthology · ACL 2026

Leakage-Aware User-Level ADHD Signal Classification from Social Media: When Graph Aggregation Helps, and When It Does Not

Daniel Cieślak Gdańsk University of Technology, Władysław Średniawa Samsung

原文摘要

User-level ADHD-related text classification from social media is methodologically challenging because predictions must aggregate many short posts, performance can be inflated by direct diagnostic leakage, and screening-adjacent settings require calibrated probabilities rather than discrimination alone. We introduce a leakage-aware evaluation framework organized around two controlled axes: evidence budget, i.e., the number of tweets available per user, and leakage control. Within this setup, we compare document-level transformers, strong non-graph embedding-pooling baselines, and heterogeneous graph models combining semantic tweet embeddings, psycholinguistic features, and temporal structure. The main result is regime-dependent: graph aggregation is most useful when user evidence is scarce, whereas simple embedding pooling becomes highly competitive and often slightly stronger as more evidence becomes available. Overall, the main contribution is a controlled benchmarking framework and a clearer account of when graph-based aggregation is actually beneficial.

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.48ACL Anthology · ACL 2026

CAL-Log: Cost-Aware Active Learning with Logarithmic Cognitive Effort Modeling and Online Adaptation to Human Annotation Behavior

Vihanga Supasan Kariyakaranage, Banuka Athuraliya

原文摘要

Active learning (AL) reduces labeled data requirements in NLP, yet most methods optimize label efficiency while ignoring annotation cost. Standard uncertainty sampling assumes uniform effort, leading to suboptimal resource allocation when documents vary in length. Supasan and Athuraliya (2026) introduced CAL-Log, a cost-aware AL variant using logarithmic cost modeling C(x)=α+β log(1+L(x)), where C(x) is the predicted annotation time for document x and L(x) is its token length, grounded in information foraging theory (Pirolli and Card, 1999) and psycholinguistic studies of human skimming (Rayner, 1998). This paper presents CAL-Log in full, extending that preliminary framework with two new contributions: temperature-scaled calibrated entropy and online per-annotator cost adaptation, which together resolve the cold-start calibration bottleneck identified in the prior work. Experiments on ten text classification benchmarks demonstrate a 3.3× speedup over BADGE (Batch Active learning by Diverse Gradient Embeddings; Ash et al., 2020) and 3.9× over Entropy sampling to reach F1=0.80, with large effect sizes (Cohen’s d>0.8). A live annotation deployment with preliminary user evaluation (N=7) confirms that the online cost model produces reading-speed classifications consistent with annotator self-reports, and that a transparency interface successfully communicates the scoring rationale to non-expert users.

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.49ACL Anthology · ACL 2026

Thesis Proposal: Targeted and Unified Cross-Lingual Unlearning from Multilingual Language Models

Jan Bronec Charles University, Jindřich Helcl Charles University

原文摘要

As large language models (LLM) trained on massive corpora scraped from the web exhibit the capability to reproduce sensitive and copyright-protected data, the field of machine unlearning has emerged to address the arising ethical and legal concerns.While previous research has provided a unified evaluation of LLM unlearning methods, this unification remains constrained to English-only models and datasets.We aim to address the prevailing fragmentation in recent cross-lingual unlearning research by extending existing unified benchmarks with multilingual data.To that end, we plan to compile a dataset of parallel translations of question-answer pairs consisting of real-world facts and synthetic personally identifiable information.Moreover, we will focus on mitigating model degradation during unlearning by selectively editing only those layers that contain the given knowledge.

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.5ACL Anthology · ACL 2026

Why Large Language Models can Secretly Outperform Embedding Similarity in Information Retrieval

Matei Benescu, Ivo Pascal de Jong

原文摘要

With the emergence of Large Language Models (LLMs), new methods in Information Retrieval are available in which relevance is estimated directly through language understanding and reasoning, instead of embedding similarity. We argue that similarity is a short-sighted interpretation of relevance, and that LLM-Based Relevance Judgment Systems (LLM-RJS) (with reasoning) have potential to outperform Neural Embedding Retrieval Systems (NERS) by overcoming this limitation. Using the TREC-DL 2019 passage retrieval dataset, we compare various LLM-RJS with NERS, but observe no noticeable improvement. Subsequently, we analyze the impact of reasoning by comparing LLM-RJS with and without reasoning. We find that human annotations also suffer from short-sightedness, and that false-positives in the reasoning LLM-RJS are primarily mistakes in annotations due to short-sightedness. We conclude that LLM-RJS do have the ability to address the short-sightedness limitation in NERS, but that this cannot be evaluated with standard annotated relevance datasets.

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.50ACL Anthology · ACL 2026

A11y-Compressor: A Framework for Enhancing the Efficiency of GUI Agent Observations through Visual Context Reconstruction and Redundancy Reduction

Michito Takeshita Hosei University, Takuro Kawada Hosei University, Takumi Ohashi Hosei University, Shunsuke Kitada LY Corp., Hitoshi Iyatomi Hosei University

原文摘要

AI agents that interact with graphical user interfaces (GUIs) require effective observation representations for reliable grounding.The accessibility tree is a commonly used text-based format that encodes UI element attributes, but it suffers from redundancy and lacks structural information such as spatial relationships among elements.We propose A11y-Compressor, a framework that transforms linearized accessibility trees into compact and structured representations.Our implementation, Compressed-a11y, applies a lightweight and structured transformation pipeline with modal detection, redundancy reduction, and semantic structuring.Experiments on the OSWorld benchmark show that Compressed-a11y reduces input tokens to 22% of the original while improving task success rates by 5.1 percentage points on average.

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.52ACL Anthology · ACL 2026

Counterspeech Generation using Small Language Models

Abubakar Sadiq Muhammad, Simona Frenda Heriot-Watt University, Gavin Abercrombie Heriot-Watt University

原文摘要

Counterspeech offers a way to tackle harmful content online without restricting freedom of expression. This work explores counterspeech generation using small language models (SLMs) as lightweight and cost-effective alternatives to large language models. We evaluate SLMs ranging from 100 million to 3 billion parameters using simple prompting strategies as well as fine-tuning, combining automatic and robust human evaluations. Our findings demonstrate that small language models can generate relevant, coherent, and high-quality counterspeech, suggesting their potential suitability for efficient and responsible deployments.

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.53ACL Anthology · ACL 2026

From Graphs to Hypergraphs: Enhancing Aspect-Term Sentiment Analysis via Multi-Level Relational Modeling

Omkar Mahesh Kashyap, Padegal Amit, Madhav Kashyap Moloco, Ashwini M Joshi Pes University, Shylaja S S

原文摘要

Aspect-Term Sentiment Analysis (ATSA) aims to predict sentiment polarity for specific aspect terms, a task complicated by conflicting sentiments and limited context in short texts. Existing graph-based approaches rely on predefined pairwise structures to capture different linguistic views. However, this leads to two key limitations: (1) their pairwise formulation often requires multiple graphs to improve expressive capacity, and (2) their reliance on predefined parsers or heuristic graph construction limits adaptability to sentence-specific sentiment composition. We propose HyperATSA, a dynamic hypergraph framework that overcomes these limitations through a single instance-specific hypergraph constructed directly from contextual token representations. Hyperedges are dynamically induced via hierarchical agglomerative clustering over token embeddings, where an acceleration-based cutoff identifies sentence-specific semantic groupings and enables adaptive hypergraph construction. Experiments on Lap14, Rest14, and MAMS demonstrate consistent improvements over strong graph-based baselines, suggesting that hypergraph-based relational modeling generalizes effectively to short-text sentiment composition.

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.54ACL Anthology · ACL 2026

Probing Bias Formation in Medical LLMs through Activation Steering

Bayram Ayadi, Annette Hautli-Janisz Universität Passau

原文摘要

Large Language Models specialized for the medical domain achieve high performance on static benchmarks, but remain vulnerable to sycophantic confabulation, where the models generate medically spurious rationales to justify incorrect user hints. This robustness gap poses severe risks in clinical environments, as models may prioritize contextual faithfulness to a biased prompt over their internal parametric medical knowledge. This study introduces a mechanistic approach to identify and mitigate these failures in MedGemma-27B, isolating hint integration circuits using Sparse Autoencoders and geometric manifold analysis. Our findings reveal that sycophantic bias is a highly distributed and polymorphic concept, with biased reasoning routed through shifting dimensions across transformer layers. We identify the optimal layer for intervention and demonstrate that cluster-conditioned dynamic steering tailored to the geometric subspace of the prompt outperforms static global interventions, though it reveals a fundamental tension between bias resilience and the retention of internal parametric knowledge. This work proposes a principled framework toward clinical AI systems that are more robust and aligned with expert medical logic, demonstrating the potential of cluster-conditioned geometric interventions while characterizing the inherent trade-offs in clinical knowledge retention.

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.55ACL Anthology · ACL 2026

Faithfulness Beyond Plausibility: Auditing Human Explanations in Educational Assessment

Ria Talsania, Dhruv Ritesh Shah, Sudhir Dhage Sardar Patel Institute of Technology

原文摘要

When rubric-based feedback tools explain a grade, students and instructors assume those explanations reflect how the score was actually determined. Yet it remains unclear whether explanation components such as rubric assignments and evidence spans reflect how scores are constructed or primarily serve as post-hoc justifications. This gap has direct implications for automated essay scoring and rubric-based feedback tools, where explanation reliability is often assumed but rarely evaluated.We introduce a knowledge graph framework that represents human tutor grading traces as structured objects, enabling controlled counterfactual testing of explanation components. Using 400 grading traces from 10 expert human tutors evaluating 100 narrative essays, we define a reconstruction-based diagnostic to measure how explanation components contribute to score interpretation, independent of prediction. Our results reveal a consistent asymmetry: removing rubric-level information leads to substantial changes in reconstructed scores, while removing evidence spans has minimal impact. This suggests that rubric structure is central to score interpretation, whereas cited evidence spans may function primarily as post-hoc justifications. We further observe tutor-specific variation in grading behavior. These findings highlight the need for explanation mechanisms that better align with scoring processes, ensuring that feedback provided to learners is both interpretable and functionally relevant.

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.58ACL Anthology · ACL 2026

CBAL: Context-Based Agentic Learning for Speaker Diarization Segmentation Refinement

Odwitiyo Dutta, Dinesh K Vishwakarma Delhi Technological University (Delhi College of Engineering),

原文摘要

Speaker diarization systems produce segmentation errors, such as false splits and boundary misplacements, that degrade transcript readability and downstream applications. We present CBAL (Context-Based Agentic Learning), a post-processing framework that refines segmentation boundaries in diarized scripts through targeted error correction. CBAL detects potential segmentation errors using acoustic and temporal heuristics and employs a lightweight LLM agent to reason about merge decisions, validating corrections through uncertainty-aware filtering with signal-based constraints. CBAL achieves 93.4% accuracy across 359 applied merges and reduces segment count by 6.1%. We demonstrate that our framework identifies and corrects high-confidence errors while maintaining 0% degradation in terms of concatenated minimum-permutation Word Error Rate (cpWER). An ablation study confirms that each component contributes non-redundantly, demonstrating the viability of interpretable refinement frameworks that use the strengths of acoustic models and language understanding without requiring end-to-end retraining.

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.59ACL Anthology · ACL 2026

Measuring and Mitigating Shortcut Reliance in Language Models with Probe-Based Representation Entanglement

Divyajot Singh

原文摘要

Shortcut learning remains a major obstacle to robust NLP systems: models can achieve high in-distribution accuracy by relying on surface cues that fail under distribution shift. We study whether shortcut reliance can be diagnosed and mitigated in small instruction-tuned language models using a simple representation-level quantity. We fine-tune Gemma 3 1B Instruct and Llama 3.2 1B on two synthetic sentiment shortcuts in SST-2 and one natural shortcut in MNLI based on lexical overlap. During training, we fit linear probes for the task label and the shortcut attribute at every layer and define CDRE as the absolute cosine similarity between the two probe directions. Across settings, increasing shortcut prevalence produces a sharp rise in the robustness gap between shortcut-aligned and shortcut-free test sets, and higher deep-layer CDRE tracks this degradation. At a 99% shortcut ratio, Llama’s clean accuracy on capitalization-biased SST-2 drops from 93.2% at 0% bias to 49.0%, while Gemma drops from 91.8% to 60.2%. A CDRE-regularized objective substantially improves robustness for capitalization and lexical-overlap shortcuts, but offers little benefit for a speaker-prefix shortcut whose learned directions are already nearly orthogonal. These results show that probe-derived representation entanglement provides a reliable signal of harmful shortcut reliance and offers a practical criterion for determining when shortcut mitigation is likely to be effective.

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.6ACL Anthology · ACL 2026

Detecting Hallucinations in Large Language Models via Internal Attention Divergence Signals

Gijs van Dijk

原文摘要

We propose a lightweight and single-pass uncertainty quantification method for detecting hallucinations in Large Language Models. The method uses attention matrices to estimate uncertainty without requiring repeated sampling or external models. Specifically, we measure the Kullback–Leibler divergence between each attention head’s distribution and a uniform reference distribution, and use these features in a logistic regression probe. Across multiple datasets, task types, and model families, attention divergence is strongly predictive of answer correctness and performs competitively with existing uncertainty estimation methods. We find that this signal is concentrated in middle layers and on factual tokens such as named entities and numbers, suggesting that attention dynamics provides an efficient and interpretable white-box signal of model uncertainty.

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.60ACL Anthology · ACL 2026

LAMP-MedQA: A Lightweight Multi-Agent System for Patient-Oriented Medical Question Answering

Jack A. Johnson, Meghali Banerjee University of Oxford, Joseph Crawford University of Oxford, James Welch University of Oxford, Jim Davies NIHR Oxford Biomedical Research Centre, Oxford, Tingyan Wang University of Oxford

原文摘要

Patient health literacy is critical to health outcomes, yet medical discharge summaries remain inaccessible to many patients due to jargon and complex language. Large language models (LLMs) offer a promising means of bridging this gap, but their deployment in resource-constrained hospital environments demands lightweight, privacy-preserving solutions. We evaluate a range of open- and closed-source LLMs on the MeDiSumQA dataset, comprising real patient discharge summaries paired with lay questions and clinician-verified answers, and demonstrate that larger open-source models achieve accuracy and semantic similarity performance comparable to GPT-5. We then introduce LAMP-MedQA, a lightweight multi-agent framework for patient-oriented medical question answering. The framework decomposes the task into two sequential stages: question-relevant evidence extraction and patient-facing answer simplification. Each stage is governed by an automated, metric-driven feedback loop that enables iterative self-correction without human-in-the-loop supervision. Using Qwen2.5-7B-Instruct for generation agents and Phi-3.5-Mini-Instruct for reviewer/verifier agents, it achieves significantly lower FKGL than zero-shot GPT-5, indicating better readability, and obtains the highest simplification quality (SARI) among all evaluated models, while remaining broadly competitive on accuracy and semantic similarity. This competitiveness is further improved by an offline medical glossary, which narrows the gap in n-gram overlap and contextual-similarity metrics. These results suggest that collaborative lightweight agents represent a viable approach to improving health literacy in clinical settings. Our code is available at: https://github.com/JackJ3636/LAMP-MedQA

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.61ACL Anthology · ACL 2026

Inference-Time Feedback for Reasoning Controllability in Diffusion Language Models

Clovis Barbour New Mexico Institute of Mining & Technology, Sandia National Laboratories and New Mexico Institute of Mining & Technology, Huixin Zhan New Mexico Institute of Mining & Technology

原文摘要

Scientific NLP systems often require outputs that satisfy strict, machine-checkable constraints. In this work, we study structured-generation controllability along three axes: structural control, iterative correction, and decoding dynamics. Diffusion decoding is of particular interest because its iterative refinement may improve global structure and revision behavior, but may also introduce distinct failure modes such as termination instability and repetition. To quantify controllability, we evaluate compliance with five machine-checkable constraints: (i) required headings and (ii) correct ordering, which reflect global structural control; (iii) explicit end markers and (iv) per-section bullet constraints, which probe local constraint adherence; and (v) repetition avoidance, which captures generation stability under different decoding dynamics. We use these metrics to assess both single-pass generation and changes under iterative correction. Our goal is to isolate structural reliability under parser-facing requirements rather than to directly measure scientific correctness. Across our benchmark, diffusion models tend to better preserve global structure, while iterative improvement substantially improves explicit termination and other local control constraints. Hybrid systems show mixed behavior depending on decoding order. These results suggest that machine-checkable controllability can be usefully decomposed into global structure versus local control, and that the two may benefit from different inference-time strategies.

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.62ACL Anthology · ACL 2026

Disentangling Linguistic Relatedness from Task Alignment in Cross-Lingual Transfer

Ahmed Haj Ahmed, Ruochen Zhang Brown University, Alvin C Grissom II Haverford College

原文摘要

We study cross-lingual transfer by fine-tuning seven large language models (4B–671B parameters) on Arabic and evaluating zero-shot reading comprehension on Semitic languages and non-Semitic controls. Across dense and Mixture-of-Experts architectures, we find no evidence of Semitic-specific transfer: models with weak baselines improve dramatically across all languages, while strong-baseline models show only marginal gains regardless of language family. A chain-of-thought ablation reinforces this finding: the same models that benefit most from fine-tuning benefit equally from inference-time reasoning, suggesting both mechanisms address task-format alignment rather than cross-lingual knowledge transfer.

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.63ACL Anthology · ACL 2026

PE-QAT: Parameter-Efficient Quantization-Aware Training for Large Language Models

Shresth Mishra

原文摘要

As large language models (LLMs) grow, their compute and memory demands become prohibitive for on-device deployment. Quantization is a crucial technique to shrink model footprint and accelerate inference, but aggressively low-bit weight-activation quantization schemes often sacrifice accuracy. Quantization Aware Training (QAT) is a commonly used paradigm to minimize quantization noise, but is extremely expensive to train and often unscalable to large models. We introduce PE-QAT, a parameter-efficient framework targeting per-channel 4-bit weight-activation quantization of LLMs, which aims to preserve model accuracy while significantly reducing resource requirements. The proposed method freezes the base model and trains lightweight LoRA adapters by fake quantizing the merged-weight model, enabling PE-QAT to scale efficiently unlike full QAT. We apply fake quantization with Straight-Through Estimators (STE) to the merged weights, allowing the adapters to explicitly compensate for quantization noise during training. One of the biggest challenges with quantizing activations alongside weights is addressing outliers that are orders of magnitude larger than other activations, which inflate quantization scales and suppress lower-magnitude values. To mitigate the impact of severe activation outliers, PE-QAT jointly learns per-channel smoothing factors and symmetric activation clipping thresholds. PE-QAT retains accuracy within 0.11 percentage points of the full-precision baseline on Llama-2-7B zero-shot tasks while training only 1.26% of total parameters.

cs.CL

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