Austria

Scaling laws are a critical component of the LLM development pipeline, most famously as a way to forecast training decisions such as &#39;compute-optimally&#39; trading-off parameter count and dataset size, alongside a more recent growing list of other crucial decisions. In this work, we ask whether compute-optimal scaling behaviour can be skill-dependent. In particular, we examine knowledge and reasoning-based skills such as knowledge-based QA and code generation, and we answer this question in the affirmative: scaling laws are skill-dependent. Next, to understand whether skill-dependent scaling is an artefact of the pretraining datamix, we conduct an extensive ablation of different datamixes and find that, also when correcting for datamix differences, knowledge and code exhibit fundamental differences in scaling behaviour. We conclude with an analysis of how our findings relate to standard compute-optimal scaling using a validation set, and find that a misspecified validation set can impact compute-optimal parameter count by nearly 50%, depending on its skill composition.

ACL 2025

Compute Optimal Scaling of Skills: Knowledge vs Reasoning

compute optimal

llms

scaling laws

skills

pretraining

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community." 

*Message from the General Chair, Dirk Hovy*

EMNLP 2025

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

Scaling laws are a critical component of the LLM development pipeline, most famously as a way to forecast training decisions such as 'compute-optimally' trading-off parameter count and dataset size, alongside a more recent growing list of other crucial decisions. In this work, we ask whether compute-optimal scaling behaviour can be skill-dependent. In particular, we examine knowledge and reasoning-based skills such as knowledge-based QA and code generation, and we answer this question in the affirmative: scaling laws are skill-dependent. Next, to understand whether skill-dependent scaling is an artefact of the pretraining datamix, we conduct an extensive ablation of different datamixes and find that, also when correcting for datamix differences, knowledge and code exhibit fundamental differences in scaling behaviour. We conclude with an analysis of how our findings relate to standard compute-optimal scaling using a validation set, and find that a misspecified validation set can impact compute-optimal parameter count by nearly 50%, depending on its skill composition.

poster

### 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)

Long-document QA presents challenges with large-scale text and long-distance dependencies. Recent advances in Large Language Models (LLMs) enable entire documents to be processed in a single pass. However, their computational cost is significantly high. Retrieval-Augmented Generation (RAG) methods split text into smaller chunks, but they often yield inferior results and may lose global context. Recent approaches that integrate LLMs into RAG via iterative summarization either underutilize LLM capabilities or still incur high computational costs. In this paper, we combine the high accuracy of LLMs with the efficiency of RAG and propose LLM-Guided Dynamic Progress Control with Attention-Based Hierarchical Weighted Graph (PECAN). Our method introduces two key improvements: (1) LLM-Guided Dynamic Progress Control: We leverage LLMs to dynamically control the retrieval process, adjusting the amount of retrieved information based on different queries to achieve a better balance of effectiveness and efficiency. (2) Attention-Guided Retrieval: We propose a novel retrieval method that constructs a hierarchical graph where edges are derived by LLM attention weights. Experimental results demonstrate that PECAN achieves LLM-level performance while maintaining computational complexity comparable to that of RAG methods on two single-document and two multi-document QA datasets.

PECAN: LLM-Guided Dynamic Progress Control with Attention-Guided Hierarchical Weighted Graph for Long-Document QA

Detecting AI-generated text is a difficult problem to begin with; detecting AI-generated text on social media is made even more difficult due to the short text length and informal, idiosyncratic language of the internet. It is nonetheless important to tackle this problem, as social media represents a significant attack vector in online influence campaigns, which may be bolstered through the use of mass-produced AI-generated posts supporting (or opposing) particular policies, decisions, or events. We approach this problem with the mindset and resources of a reasonably sophisticated threat actor, and create a dataset of 505,159 AI-generated social media posts from a combination of open-source, closed-source, and fine-tuned LLMs, covering 11 different controversial topics. We show that while the posts can be detected under typical research assumptions about knowledge of and access to the generating models, under the more realistic assumption that an attacker will not release their fine-tuned model to the public, detectability drops dramatically. This result is confirmed with a human study. Ablation experiments highlight the vulnerability of various detection algorithms to fine-tuned LLMs. This result has implications across all detection domains, since fine-tuning is a generally applicable and realistic LLM use case.

When Detection Fails: The Power of Fine-Tuned Models to Generate Human-Like Social Media Text

With the growing prevalence of large language models (LLMs), the safety of LLMs has raised significant concerns. However, evaluating their safety is still lack of definitive standards due to the subjective nature of current safety benchmarks. To address this gap, we conducted the first exploration of LLMs safety evaluation from a legal perspective by proposing the SafeLawBench benchmark. The SafeLawBench categorizes safety risks into three levels based on legal standards, providing a systematic and comprehensive framework for evaluation. It comprises 24,860 multi-choice questions and 1,107 open-domain question-answering (QA) tasks. Our evaluation encompassed 17 LLMs using both zero-shot and few-shot prompting, highlighting the advantages of closed-source models over open-source ones. We also evaluated the safety-related reasoning stability of the LLMs and their refusal behavior. Additionally, we found that a majority voting mechanism can enhance model performance. Notably, even leading SOTA models like GPT-4o and Claude-3.5-Sonnet have not exceeded 80.3% performance, while the average performance of 17 LLMs remains at 67% on SafeLawBench. We urge the community to prioritize research on the safety of LLMs.

SafeLawBench: Towards Safe Alignment of Large Language Models

Most Large language models (LLMs) adopt an autoregressive architecture, predicting the next word token based on the preceding context. While this approach is robust for language generation tasks such as writing and summarization, it has limitations for high-level reasoning tasks, such as prediction and decision-making. To overcome these limitations, we introduce a new method called Prediction-Augmented Generation (PAG). PAG can improve the generation quality and predictive accuracy of large language models in inference-driven tasks by integrating task-specific predictive models as external tools, enabling more structured and precise reasoning. Moreover, our method does not simply copy the inferences of a predictive model, but improves the inference results with knowledge from the large language model to create better predictions. We comprehensively evaluate our proposed method on diverse datasets for automatic diagnosis tasks requiring extensive domain knowledge and advanced reasoning.

Prediction-Augmented Generation for Automatic Diagnosis Tasks

Data watermarking in language models injects traceable signals, such as specific token sequences or stylistic patterns, into copyrighted text, allowing copyright holders to track and verify training data ownership. Previous data watermarking techniques primarily focus on effective memorization after pretraining, while overlooking challenges that arise in other stages of the LLM pipeline, such as the risk of watermark filtering during data preprocessing, or potential forgetting through post-training, or verification difficulties due to API-only access. We propose a novel data watermarking approach that injects coherent and plausible yet fictitious knowledge into training data using generated passages describing a fictitious entity and its associated attributes. Our watermarks are designed to be memorized by the LLM through seamlessly integrating in its training data, making them harder to detect lexically during preprocessing. We demonstrate that our watermarks can be effectively memorized by LLMs, and that increasing our watermarks' density, length, and diversity of attributes strengthens their memorization. We further show that our watermarks remain robust throughout LLM development, maintaining their effectiveness after continual pretraining and supervised finetuning. Finally, we show that our data watermarks can be evaluated even under API-only access via question answering.

Robust Data Watermarking in Language Models by Injecting Fictitious Knowledge

Text watermarking, which modify tokens to embed watermark, has proven effective in detecting machine-generated texts. Yet its application to low-entropy texts like code and mathematics presents significant challenges. A fair number of tokens in these texts are hardly modifiable without changing the intended meaning, causing statistical measures to falsely indicate the absence of a watermark. Existing research addresses this issue by rely mainly on a limited number of high-entropy tokens, which are considered flexible for modification, and accurately reflecting watermarks. However, their detection accuracy remains suboptimal, as they neglect strong watermark evidences embedded in low entropy tokens modified through watermarking. To overcome this limitation, we introduce Bayes’ Rule derived Watermark Detector (BRWD), which exploit watermark information from every token, by leveraging the posterior probability of watermark’s presence. We theoretically prove the optimality of our method in terms of detection accuracy, and demonstrate its superiority across various datasets, models, and watermark injection strategies. Notably, our method achieves up to 50% and 70% relative improvements in detection accuracy over the best baselines in code generation and math problem-solving tasks, respectively. Our code is available at https://github.com/cczslp/BRWD.

Low-Entropy Watermark Detection via Bayes’ Rule Derived Detector

Majority voting is considered an effective method to enhance chain-of-thought reasoning, as it selects the answer with the highest ''self-consistency'' among different reasoning paths (Wang et al., 2023). However, previous chain-of-thought reasoning methods typically generate only a single answer in each trial, thereby ignoring the possibility of other potential answers. As a result, these alternative answers are often overlooked in subsequent voting processes. In this work, we propose to generate ranked answers in each reasoning process and conduct ranked voting among multiple ranked answers from different responses, thereby making the overall self-consistency more reliable. Specifically, we use three ranked voting methods: Instant-runoff voting, Borda count voting, and mean reciprocal rank voting. We validate our methods on six datasets, including three multiple-choice and three open-ended question-answering tasks, using both advanced open-source and closed-source large language models. Extensive experimental results indicate that our proposed method outperforms the baselines, showcasing the potential of leveraging the information of ranked answers and using ranked voting to improve reasoning performance. Code and logs will be released.

Ranked Voting based Self-Consistency of Large Language Models

The breakthrough of OpenAI o1 highlights the potential of enhancing reasoning to improve LLM. Yet, most research in reasoning has focused on mathematical tasks, leaving domains like medicine underexplored. The medical domain, though distinct from mathematics, also demands robust reasoning to provide reliable answers, given the high standards of healthcare. However, verifying medical reasoning is challenging, unlike those in mathematics. To address this, we propose **Medical Verifiable Problems** with a medical verifier to check the correctness of model outputs. This verifiable nature enables advancements in medical reasoning through **a two-stage approach**: (1) using the verifier to guide the search for a complex reasoning trajectory for fine-tuning LLMs, (2) applying reinforcement learning (RL) with verifier-based rewards to enhance complex reasoning further. Finally, we introduce HuatuoGPT-o1, a medical LLM capable of complex reasoning, which outperforms general and medical-specific baselines using only 40K verifiable problems. Experiments show complex reasoning improves medical problem-solving and benefits more from RL. We hope our approach inspires advancements in reasoning across medical and other specialized domains. Code, datasets, and models are publicly available at https://github.com/FreedomIntelligence/HuatuoGPT-o1.

Towards Medical Complex Reasoning with LLMs through Medical Verifiable Problems

In recent progress, mathematical verifiers have achieved success in mathematical reasoning tasks by validating the correctness of solutions generated by policy models. However, existing verifiers are trained with binary classification labels, which are not informative enough for the model to accurately assess the solutions. To mitigate the aforementioned insufficiency of binary labels, we introduce step-wise natural language feedback as rationale labels, that is, the correctness of each step and the detailed explanations. In this paper, we propose Math-Minos, a natural language feedback-enhanced verifier by constructing automatically generated training data and a two-stage training paradigm for effective training and efficient inference. Our experiments reveal that a small set of natural language feedback can significantly boost the performance of the verifier in both verification and reinforcement learning and also significantly alleviates the data-demanding problems of the reward model with an over 700% data efficiency improvement.

LLM Critics Help Catch Bugs in Mathematics: Towards a Better Mathematical Verifier with Natural Language Feedback

Question Answering (QA) with large language models has shown impressive performance, yet hallucinations still persist, particularly when user queries carry incorrect premises, insufficient context, or linguistic ambiguity. To address this issue, we propose Lightweight Query Checkpoint (LQC), a small classification model that detects verification-required queries before the LLM generates a potentially faulty answer. LQC leverages hidden states extracted from intermediate layers of a smaller-scale, non-instruct-tuned LLM to effectively distinguish queries requiring verification from clear queries. We first systematically define categories of queries that need verification, construct a dataset comprising both defective and clear queries, and train a binary contrastive learning model. Through extensive experiments on various QA datasets, we demonstrate that incorporating LQC into QA pipelines reduces hallucinations while preserving strong answer quality.

Downloads

Next from ACL 2025

PECAN: LLM-Guided Dynamic Progress Control with Attention-Guided Hierarchical Weighted Graph for Long-Document QA

Similar lecture

Scaling Laws Are Unreliable for Downstream Tasks: A Reality Check

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES