Austria

Faced with an expensive human annotation process, creators of NLP systems increasingly turn to synthetic data generation. While this method shows promise, the extent to which synthetic data can replace human annotation is poorly understood. We investigate the use of synthetic data in Fact Verification (FV) and Evidence-based Question Answering (QA) by incrementally replacing human-generated data with synthetic points on eight diverse datasets. Strikingly, replacing up to 90% of the training data only marginally decreases performance, but replacing the final 10% leads to severe declines. We find that models trained on purely synthetic data can be improved by including as few as 125 human-generated data points. We show that matching the performance gain of a little human data requires an order of magnitude more synthetic data, and then estimate price ratios at which human annotation would be a more cost-effective solution. Our results suggest that even when human annotation at scale is infeasible, there is great value to having a small proportion of the dataset being human-generated.

ACL 2025

A Little Human Data Goes A Long Way

synthetic data

fact verification

question answering

### 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

Fine-tuning is widely recognized as a crucial process for aligning large language models (LLMs) with human intentions. However, the substantial memory requirements associated with fine-tuning pose a significant barrier to extending the applicability of LLMs. While parameter-efficient fine-tuning can be a promising approach by reducing trainable parameters, intermediate activations still need to be cached to compute gradients during the backward pass, thereby limiting overall memory efficiency. In this work, we propose Saliency-Guided Gradient Flow (SAGE), a memory-efficient fine-tuning method designed to minimize the memory specifically associated with cached intermediate activations. The key strategy is to selectively cache activations based on their saliency during the forward pass and then use these activations for the backward pass. This process transforms the dense backward pass into a sparse one, thereby enhancing memory efficiency. To verify whether SAGE can serve as an efficient alternative for fine-tuning, we conduct comprehensive experiments across diverse fine-tuning scenarios and setups. The experimental results show that SAGE substantially improves memory efficiency without a significant loss in accuracy, highlighting its broad value in real-world applications

Forward Knows Efficient Backward Path: Saliency-Guided Memory-Efficient Fine-tuning of Large Language Models

Medical Large Vision-Language Models (Med-LVLMs) often exhibit suboptimal attention distribution on visual inputs, leading to hallucinated or inaccurate outputs. Existing methods primarily rely on inference-time interventions, which are limited in attention adaptation or require additional supervision. To address this, we propose A$^3$Tune, a novel fine-tuning framework for Automatic Attention Alignment Tuning. A
Tune leverages zero-shot weak labels from SAM, refines them into prompt-aware labels using BioMedCLIP, and then selectively modifies visually-critical attention heads to improve alignment while minimizing interference. Additionally, we introduce a A$^3$MoE module, enabling adaptive parameter selection for attention tuning across diverse prompts and images. Extensive experiments on medical VQA and report generation benchmarks show that A$^3$Tune outperforms state-of-the-art baselines, achieving enhanced attention distributions and performance in Med-LVLMs.

Focus on What Matters: Enhancing Medical Vision-Language Models with Automatic Attention Alignment Tuning

Processing long contexts poses a significant challenge for large language models (LLMs) due to their inherent context window limitations and the computational burden of extensive key-value (KV) activations, which severely impact efficiency. For information-seeking tasks, full context perception is often unnecessary, as a query's information needs can dynamically range from localized details to a global perspective, depending on its complexity. However, existing methods struggle to adapt effectively to this dynamic information needs.

In the paper, we propose a method for processing long-context information-seeking tasks via query-guided ACtivation REfilling (ACRE). ACRE constructs a Bi-layer KV Cache for long contexts, where the layer-1 (L1) cache compactly captures global information, and the layer-2 (L2) cache provides detailed, localized information. ACRE establishes a proxying relationship between the two caches, allowing the input query to attend to the L1 cache and dynamically refill it with relevant entries from the L2 cache. This mechanism integrates global understanding with query-specific local details, thereby enhancing answer decoding. Experiments on a variety of long-context information-seeking datasets demonstrate ACRE's effectiveness, achieving significant improvements in both performance and efficiency.

Boosting Long-Context Information Seeking via Query-Guided Activation Refilling

While language models are increasingly utilized in materials science, typical models rely on frequency-centric tokenization methods originally developed for natural language processing. However, these methods frequently produce excessive fragmentation and semantic loss, failing to maintain the structural and semantic integrity of material concepts. To address this issue, we propose MATTER, a novel tokenization approach that integrates material knowledge into tokenization. Based on MatDetector trained on our materials knowledge base and re-ranking method prioritizing material terms in token merging, MATTER maintains the structural integrity of identified materials concepts and prevents fragmentation during tokenization, ensuring their semantic meaning remains intact. The experimental results demonstrate that MATTER outperforms existing tokenization methods, achieving an average performance gain of 4% and 2% in the generation and classification tasks, respectively. These results underscore the importance of domain knowledge for tokenization strategies in scientific text processing.

Incorporating Domain Knowledge into Materials Tokenization

Vision-language models (VLMs) aligned with general human objectives, such as being harmless and hallucination-free, have become valuable assistants of humans in managing visual tasks. However, people with diversified backgrounds have different cognition even in the same situation. Consequently, they may have personalized expectations for VLM assistants. This highlights the urgent need to align VLM assistants with personalized situated cognition for real-world assistance. To study this problem, we first simplify it by characterizing individuals based on the sociological concept of Role-Set. Then, we propose to evaluate the individuals' actions to examine whether the personalized alignment is achieved. Further, we construct a benchmark named PCogAlignBench, which includes 18k instances and 20 individuals with different Role-Sets. Finally, we present a framework called PCogAlign, which constructs a cognition-aware and action-based reward model for personalized alignment. Experimental results and human evaluations demonstrate the reliability of the PCogAlignBench and the effectiveness of our proposed PCogAlign. We will open-source the constructed benchmark and code after being accepted.

Aligning VLM Assistants with Personalized Situated Cognition

Speculative Decoding (SD) is a promising method for reducing the inference latency of large language models (LLMs). A well-designed draft model and an effective draft candidate tree construction method are key to enhancing the acceleration effect of SD. In this paper, we first propose the Effective Draft Decoder (EDD), which treats the LLM as a powerful encoder and generates more accurate draft tokens by leveraging the encoding results as soft prompts. Furthermore, we use KL divergence instead of the standard cross-entropy loss to better align the draft model's output with the LLM. Next, we introduce the Pruned Candidate Tree (PCT) algorithm to construct a more efficient candidate tree. Specifically, we found that the confidence scores predicted by the draft model are well-calibrated with the acceptance probability of draft tokens. Therefore, PCT estimates the expected time gain for each node in the candidate tree based on confidence scores and retains only the nodes that contribute to acceleration, pruning away redundant nodes. We conducted extensive experiments with various LLMs across four datasets. The experimental results verify the effectiveness of our proposed method, which significantly improves the performance of SD and reduces the inference latency of LLMs.

Faster Speculative Decoding via Effective Draft Decoder with Pruned Candidate Tree

Generative Retrieval (GR) introduces a new information retrieval paradigm that directly generates unique document identifiers (DocIDs). The key challenge of GR lies in creating effective yet discrete DocIDs that preserve semantic relevance for similar documents while differentiating dissimilar ones. However, existing methods generate DocIDs solely based on the textual content of documents, which may result in DocIDs with weak semantic connections for similar documents due to variations in expression. Therefore, we propose using queries as a bridge to connect documents with varying relevance levels for learning improved DocIDs. In this paper, we propose **M**ulti-l**E**vel **R**elevance document identifier learning for **G**enerative r**E**trieval (MERGE), a novel approach that utilizes multi-level document relevance to learn high-quality DocIDs. MERGE incorporates three modules: a multi-relevance query-document alignment module to effectively align document representations with related queries, an outer-level contrastive learning module to capture binary-level relevance, and an inner-level multi-level relevance learning module to distinguish documents with different relevance levels. Our approach encodes rich hierarchical semantic information and maintains uniqueness across documents. Experimental results on real-world multilingual e-commerce search datasets demonstrate that MERGE significantly outperforms existing methods, underscoring its effectiveness. The source code is available at <https://github.com/zhangfw123/MERGE>.

Multi-level Relevance Document Identifier Learning for Generative Retrieval

Summarizing long-form narratives—such as books, movies, and TV scripts—requires capturing intricate plotlines, character interactions, and thematic coherence, a task that remains challenging for existing LLMs. We introduce NexusSum, a multi-agent LLM framework for narrative summarization that processes long-form text through a structured, sequential pipeline—without requiring fine-tuning. Our approach introduces two key innovations: **(1) Dialogue-to-Description Transformation**: A narrative-specific preprocessing method that standardizes character dialogue and descriptive text into a unified format, improving coherence. **(2) Hierarchical Multi-LLM Summarization**: A structured summarization pipeline that optimizes chunk processing and controls output length for accurate, high-quality summaries. Our method establishes a new state-of-the-art in narrative summarization, achieving up to **a 30.0% improvement in BERTScore (F1)** across books, movies, and TV scripts. These results demonstrate the effectiveness of multi-agent LLMs in handling long-form content, offering a scalable approach for structured summarization in diverse storytelling domains.

NexusSum: Hierarchical LLM Agents for Long-Form Narrative Summarization

Recent Multi-modal Large Language Models (MLLMs) have made great progress in video understanding. However, their performance on videos involving human actions is still limited by the lack of high-quality data. To address this, we introduce a two-stage data annotation pipeline. First, we design strategies to accumulate videos featuring clear human actions from the Internet. Second, videos are annotated in a standardized caption format that uses human attributes to distinguish individuals and chronologically details their actions and interactions. Through this pipeline, we curate two datasets, namely HAICTrain and HAICBench. **HAICTrain** comprises 126K video-caption pairs generated by Gemini-Pro and verified for training purposes. Meanwhile, **HAICBench** includes 412 manually annotated video-caption pairs and 2,000 QA pairs, for a comprehensive evaluation of human action understanding. Experimental results demonstrate that training with HAICTrain not only significantly enhances human understanding abilities across 4 benchmarks, but can also improve text-to-video generation results. Both the HAICTrain and HAICBench will be made open-source to facilitate further research.

HAIC: Improving Human Action Understanding and Generation with Better Captions for Multi-modal Large Language Models

Personalized AI assistants, a hallmark of the human-like capabilities of Large Language Models (LLMs), are a challenging application that intertwines multiple problems in LLM research. Despite the growing interest in the development of personalized assistants, the lack of an open-source conversational dataset tailored for personalization remains a significant obstacle for researchers in the field. To address this research gap, we introduce HiCUPID, a new benchmark to probe and unleash the potential of LLMs to deliver personalized responses. Alongside a conversational dataset, HiCUPID provides a Llama-3.2-based automated evaluation model whose assessment closely mirrors human preferences. We release our dataset, evaluation model, and code at https://github.com/12kimih/HiCUPID.

Downloads

Next from ACL 2025

Forward Knows Efficient Backward Path: Saliency-Guided Memory-Efficient Fine-tuning of Large Language Models

Similar lecture

Understanding the Influence of Synthetic Data for Text Embedders