Austria

Due to the sensitive nature of personally identifiable information (PII), its owners may have the authority to control its inclusion or request its removal from large-language model (LLM) training. Beyond this, PII may be added or removed from training datasets due to evolving dataset curation techniques, because they were newly scraped for retraining, or because they were included in a new downstream fine-tuning stage. We find that the amount and ease of PII memorization is a dynamic property of a model that evolves throughout training pipelines and depends on commonly altered design choices. We characterize three such novel phenomena: (1) similar-appearing PII seen later in training can elicit memorization of earlier-seen sequences in what we call assisted memorization, and this is a significant factor (in our settings, up to 1/3); (2) adding PII can increase memorization of other PII; and (3) removing PII can lead to other PII being memorized.

ACL 2025

Privacy Ripple Effects from Adding or Removing Personal Information in Language Model Training

personal information

memorization

language models

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

poster

Vision-language models (VLMs) work well in tasks ranging from image captioning to visual question answering (VQA), yet they struggle with spatial reasoning, a key skill for understanding our physical world that humans excel at. We find that spatial relations are generally rare in widely used VL datasets, with only a few being well represented, while most form a long tail of underrepresented relations. This gap leaves VLMs ill-equipped to handle diverse spatial relationships. To bridge it, we construct a synthetic VQA dataset focused on spatial reasoning generated from hyper-detailed image descriptions in Localized Narratives, DOCCI, and PixMo-Cap. Our dataset consists of 455k samples containing 3.4 million QA pairs. Trained on this dataset, our Spatial-Reasoning Enhanced (SpaRE) VLMs show strong improvements on spatial reasoning benchmarks, achieving up to a 49% performance gain on the What's Up benchmark, while maintaining strong results on general tasks. Our work narrows the gap between human and VLM spatial reasoning and makes VLMs more capable in real-world tasks such as robotics and navigation. We plan to share our code and dataset in due course.

SpaRE: Enhancing Spatial Reasoning in Vision-Language Models with Synthetic Data

Effectively selecting data from subgroups where a model performs poorly is crucial for improving its performance. Traditional methods for identifying these subgroups often rely on sensitive information, raising privacy issues. Additionally, gathering such information at runtime might be impractical. This paper introduces a cost-effective strategy that addresses these concerns. We identify underperforming subgroups and train a model to predict if an utterance belongs to these subgroups without needing sensitive information. This model helps mitigate bias by selecting and adding new data, which is labeled as challenging, for re-training the speech model. Experimental results on intent classification and automatic speech recognition tasks show the effectiveness of our approach in reducing biases and enhancing performance, with improvements in reducing error rates of up to 39% for FSC, 16% for ITALIC, and 22% for LibriSpeech.

Privacy Preserving Data Selection for Bias Mitigation in Speech Models

In the context of large language models (LLMs), current advanced reasoning methods have made impressive strides in various reasoning tasks. However, when it comes to logical reasoning tasks, significant challenges remain in both efficacy and efficiency. This is rooted in the fact that these systems fail to fully leverage the inherent structure of logical tasks throughout the reasoning processes, including decomposition, search, and resolution. To address this, this paper proposes a logic-complete reasoning framework, Aristotle. The framework consists of three key components: Logical Decomposer, Logical Search Router, and Logical Resolver, in which symbolic expressions and logical rules are comprehensively integrated into the entire reasoning process, significantly alleviating the bottlenecks of logical reasoning, i.e., reducing sub-task complexity, minimizing search errors, and resolving logical contradictions. Experimental results demonstrate that Aristotle consistently outperforms state-of-the-art reasoning frameworks in both accuracy and efficiency, particularly excelling in complex logical reasoning scenarios.

Aristotle: Mastering Logical Reasoning with A Logic-Complete Decompose-Search-Resolve Framework

Dogs communicate intelligently but little is known about the phonetic properties of their vocalization communication. For the first time, this paper presents an iterative algorithm inspired by human phonetic discovery, which is based on minimal pairs that determine phonemes by distinguishing different words in human language, and is able to produce a complete alphabet of distinct canine phoneme-like units. In addition, the algorithm produces a number of canine repeated acoustic units, which may correspond to specific environments and activities of a dog, composed exclusively of the canine phoneme-like units in the alphabet. The framework outlined in this paper is expected to function not only on canines but other animal species.

Toward Automatic Discovery of a Canine Phonetic Alphabet

Large Language Models (LLMs) have shown promising potential in the medical domain, assisting with tasks like clinical note generation and patient communication. However, current LLMs are limited to text-based communication, hindering their ability to interact with diverse forms of information in clinical environments. Despite clinical agents succeeding in diverse signal interaction, they are oriented to a single clinical scenario and hence fail for broader applications. To evaluate clinical agents holistically, we propose ClinicalAgent Bench~(CAB), a comprehensive medical agent benchmark consisting of 18 tasks across five key realistic clinical dimensions. Building on this, we introduce ReflectTool, a novel framework that excels at utilizing domain-specific tools within two stages. The first optimization stage progressively enlarges a long-term memory by saving successful solving processes and tool-wise experience of agents in a tiny pre-defined training set. In the following inference stage, ReflectTool can search for supportive successful demonstrations from already built long-term memory to guide the tool selection strategy, and a verifier improves the tool usage according to the tool-wise experience with two verification methods--iterative refinement and candidate selection. Extensive experiments on CAB demonstrate that ReflectTool surpasses the pure LLMs with more than 10 points and the well-established agent-based methods with 3 points, highlighting its adaptability and effectiveness in solving complex clinical tasks. Our code and datasets are available at https://github.com/BlueZeros/ReflecTool.

ReflecTool: Towards Reflection-Aware Tool-Augmented Clinical Agents

With the development of large language models, they are widely used as agents in various fields. A key component of agents is memory, which stores vital information but is susceptible to jailbreak attacks. Existing research mainly focuses on single-agent attacks and shared memory attacks. However, real-world scenarios often involve independent memory. In this paper, we propose the Troublemaker Makes Chaos in Honest Town (TMCHT) task, a large-scale, multi-agent, multi-topology text-based attack evaluation framework. TMCHT involves one attacker agent attempting to mislead an entire society of agents. We identify two major challenges in multi-agent attacks: (1) Non-complete graph structure, (2) Large-scale systems. We attribute these challenges to a phenomenon we term toxicity disappearing. To address these issues, we propose an Adversarial Replication Contagious Jailbreak (ARCJ) method, which optimizes the retrieval suffix to make poisoned samples more easily retrieved and optimizes the replication suffix to make poisoned samples have contagious ability. We demonstrate the superiority of our approach in TMCHT, with 23.51%, 18.95%, and 52.93% improvements in line, star topologies, and 100-agent settings. It reveals potential contagion risks in widely used multi-agent architectures.

A Troublemaker with Contagious Jailbreak Makes Chaos in Honest Towns

Mitigating positional bias of language models (LMs) for listwise inputs is a well-known and important problem (e.g., lost-in-the-middle). While zero-shot order-invariant LMs have been proposed to solve this issue, their success on practical listwise problems has been limited. In this work, as a first contribution, we identify and overcome two limitations to make zero-shot invariant LMs more practical: (1) training and inference distribution mismatch arising from modifying positional ID assignments to enforce invariance, and (2) failure to adapt to mixture of order-invariant and sensitive inputs in practical listwise problems. Then, to overcome these issues we propose (1) RoToR, a zero-shot invariant LM for genuinely order-invariant inputs with minimal modifications of positional IDs, and (2) Selective Routing, an adaptive framework that handles both order-invariant and order-sensitive inputs in listwise tasks. On the Lost in the middle (LitM), Knowledge Graph QA (KGQA), and MMLU benchmarks, we show that RoToR with Selective Routing can effectively handle practical listwise input tasks in a zero-shot manner (https://github.com/soyoung97/RoToR)

RoToR: Towards More Reliable Responses for Order-Invariant Inputs

Best-of-N decoding methods instruct large language models (LLMs) to generate multiple solutions, score each using a scoring function, and select the highest scored as the final answer to mathematical reasoning problems. However, this repeated independent process often leads to the same mistakes, making the selected solution still incorrect. We propose a novel prompting method named Stepwise Correction (StepCo) that helps LLMs identify and revise incorrect steps in their generated reasoning paths. It iterates verification and revision phases that employ a process-supervised verifier. The verify-then-revise process not only improves answer correctness but also reduces token consumption with fewer paths needed to generate. With StepCo, a series of LLMs demonstrate exceptional performance. Notably, using GPT-4o as the backend LLM, StepCo achieves an average accuracy of 94.1 across eight datasets, significantly outperforming the state-of-the-art Best-of-N method by +2.4, while reducing token consumption by 77.8\%. Our implementation is made publicly available at https://wzy6642.github.io/stepco.github.io.

Enhancing Mathematical Reasoning in LLMs by Stepwise Correction

Originally developed to reduce the manual burden of grading standardised language tests, Automated Essay Scoring (AES) research has long focused on holistic scoring methods which offer minimal formative feedback in the classroom. With the increasing demand for technological tools that support language acquisition, the field is turning to analytic AES (evaluating essays according to different linguistic traits). This approach holds promise for generating more detailed essay feedback, but relies on analytic scoring data that is both more cognitively demanding for humans to produce, and prone to bias. The dominant paradigm in AES is to aggregate disagreements between raters into a single gold-standard label, which fails to account for genuine examiner variability. In an attempt to make AES more representative and trustworthy, we propose to explore the sources of disagreements and lay out a novel AES system design that learns from individual raters instead of the gold standard labels.

Beyond the Gold Standard in Analytic Automated Essay Scoring

With the advent of highly capable instruction-tuned neural language models, benchmarking in natural language processing (NLP) is increasingly shifting towards pairwise comparison leaderboards, such as LMSYS Arena, from traditional global pointwise scores (e.g., GLUE, BIG-bench, SWE-bench). This paper empirically investigates the strengths and weaknesses of both global scores and pairwise comparisons to aid decision-making in selecting appropriate model evaluation strategies. Through computational experiments on synthetic and real-world datasets using standard global metrics and the popular Bradley–Terry model for pairwise comparisons, we found that while global scores provide more reliable overall rankings, they can underestimate strong models with rare, significant errors or low confidence. Conversely, pairwise comparisons are particularly effective for identifying strong contenders among models with lower global scores, especially where quality metrics are hard to define (e.g., text generation), though they require more comparisons to converge if ties are frequent.

Downloads

Next from ACL 2025

SpaRE: Enhancing Spatial Reasoning in Vision-Language Models with Synthetic Data

Similar lecture

Privacy Ripple Effects from Adding or Removing Personal Information in Language Model Training