ACL 2025 Papers

With the continuously expanding parameters, efficiently adapting large language models to downstream tasks is crucial in resource-limited conditions. Many parameter-efficient fine-tuning methods have emerged to address this challenge. However, they lack flexibility, like LoRA requires manually selecting trainable parameters and rank size, (IA)3 can only scale the activations along columns, yielding inferior results due to less precise fine-tuning. To address these issues, we propose a novel method named AdaDHP with fewer parameters and finer granularity, which can adaptively select important parameters for each task. Specifically, we introduce two trainable vectors for each parameter and fine-tune the parameters through Hadamard product along both rows and columns. This significantly reduces the number of trainable parameters, with our parameter count capped at the lower limit of LoRA. Moreover, we design an adaptive parameter selection strategy to select important parameters for downstream tasks dynamically. This allows our method to flexibly remove unimportant parameters for downstream tasks. Finally, we demonstrate the superiority of our method on the T5-base model across 17 NLU tasks and on complex mathematical tasks with the Llama series models.

Identifying factors that make ad text attractive is essential for advertising success. This study proposes AdParaphrase v2.0, a dataset for ad text paraphrasing, containing human preference data, to enable the analysis of the linguistic factors and to support the development of methods for generating attractive ad texts. Compared with v1.0, this dataset is 20 times larger, comprising 16,460 ad text paraphrase pairs, each annotated with preference data from ten evaluators, thereby enabling a more comprehensive and reliable analysis. Through the experiments, we identified multiple linguistic features of engaging ad texts that were not observed in v1.0 and explored various methods for generating attractive ad texts. Furthermore, our analysis demonstrated the relationships between human preference and ad performance, and highlighted the potential of reference-free metrics based on large language models for evaluating ad text attractiveness.The dataset is publicly available at: https://github.com/CyberAgentAILab/AdParaphrase-v2.0.

Large Vision-Language Models (LVLMs) have demonstrated impressive capabilities in multimodal understanding, but they frequently suffer from hallucination - generating content inconsistent with visual inputs. In this work, we explore a novel perspective on hallucination mitigation by examining the intermediate activations of LVLMs during generation. Our investigation reveals that hallucinated content manifests as distinct, identifiable patterns in the model’s hidden state space. Motivated by this finding, we propose Activation Steering Decoding (ASD), a training-free approach that mitigates hallucination through targeted intervention in the model’s intermediate activations. ASD operates by first identifying directional patterns of hallucination in the activation space using a small calibration set, then employing a contrast decoding mechanism that computes the difference between positive and negative steering predictions. This approach effectively suppresses hallucination patterns while preserving the model’s general capabilities. Extensive experiments demonstrate that our method significantly reduces hallucination across multiple benchmarks while maintaining performance on general visual understanding tasks. Notably, our approach requires no model re-training or architectural modifications, making it readily applicable to existing deployed models.

Large Vision-Language Models (LVLMs) typically learn visual capacity through visual instruction tuning, involving updates to both a projector and their LLM backbones. Inspired by the concept of a visual region in the human brain, we investigate the existence of an analogous visual region within LLMs that functions as a cognitive core, and explore the potential of efficient training of LVLMs via selective layers tuning. Using Bunny-Llama-3-8B-V for detailed analysis and other three LVLMs for validation across diverse visual and textual tasks, we find that selectively updating 25% of LLMs layers, when sparsely and uniformly distributed, can preserve nearly 99% of visual performance and maintain or improve textual task results, while effectively reducing training time. Based on this targeted training approach, we further propose a novel visual region-based pruning paradigm, removing non-critical layers outside the visual region, which can achieve minimal performance loss. This study offers an effective and efficient strategy for LVLM training and inference by activating a layer-wise visual region within LLMs, which proves consistently effective across different models.

Active perception, a crucial human capability, involves setting a goal based on the current understanding of the environment and performing actions to achieve that goal. Despite significant efforts in evaluating Multimodal Large Language Models (MLLMs), active perception has been largely overlooked. To address this gap, we propose a novel benchmark named ActiView to evaluate active perception in MLLMs. We focus on a specialized form of Visual Question Answering (VQA) that eases and quantifies the evaluation yet challenging for existing MLLMs. Meanwhile, intermediate reasoning behaviors of models are also discussed. Given an image, we restrict the perceptual field of a model, requiring it to actively zoom or shift its perceptual field based on reasoning to answer the question successfully. We conduct extensive evaluation over 30 models, including proprietary and open-source models, and observe that restricted perceptual fields play a significant role in enabling active perception. Results reveal a significant gap in the active perception capability of MLLMs, indicating that this area deserves more attention. We hope that ActiView could help develop methods for MLLMs to understand multimodal inputs in more natural and holistic ways.

In team communication, dialogue acts play a crucial role in helping team members understand each other’s intentions and revealing the roles and communication patterns within interactions. Although existing studies have focused on using Dialogue Act classification to capture the speaker’s intentions, few have explored the underlying power dynamics reflected by these dialogue acts. To this end, we present an online Dialogue Act Classification and Dynamic Power Analysis framework—Act2P, which is based on large language model. The framework combines the zero-shot learning capability of LLMs and introduces an online feedback classification method that allows for online classification with iterative feedback to previous stages, achieving efficient and accurate classification without the labeled data. Additionally, we also propose the PowerRank algorithm, which quantifies power dynamics through a graph-based structure. Through comparative experiments with existing methods, we demonstrate the significant superiority of Act2P in online scenarios and successfully visualize dialogue power in online, clearly presenting the distribution and dynamic transfer of power. This framework provides new scientific insights and practical tools for optimizing team collaboration.

Recent advancements in large language models (LLMs) have demonstrated their impressive generative capabilities, primarily due to their extensive parameterization, which enables them to encode vast knowledge. However, effectively integrating new knowledge into LLMs remains a major challenge. Current research typically first constructs novel knowledge datasets and then injects this knowledge into LLMs through various techniques. However, existing methods for constructing new datasets either rely on timestamps, which lack rigor, or use simple templates for synthesis, which are simplistic and do not accurately reflect the real world. To address this issue, we propose a novel knowledge dataset construction approach that simulates biological evolution using knowledge graphs to generate synthetic entities with diverse attributes, resulting in a dataset, NovelHuman. Systematic analysis on NovelHuman reveals that the intra-sentence position of knowledge significantly affects the acquisition of knowledge. Therefore, we introduce an intra-sentence permutation to enhance knowledge acquisition. Furthermore, given that potential conflicts exist between autoregressive (AR) training objectives and permutation-based learning, we propose PermAR, a permutation-based language modeling framework for AR models. PermAR seamlessly integrates with mainstream AR architectures, endowing them with bidirectional knowledge acquisition capabilities. Extensive experiments demonstrate the superiority of PermAR, outperforming knowledge augmentation methods by 3.3%-38%.

Acoustic individual identification of wild animals is an essential task for understanding animal vocalizations within their social contexts, and for facilitating conservation and wildlife monitoring efforts. However, most of the work in this space relies on human efforts, as the development of methods for automatic individual identification is hindered by the lack of data. In this paper, we explore cross-species pre-training to address the task of individual classification in white-faced capuchin monkeys. Using acoustic embeddings from birds and humans, we find that they can be effectively used to identify the calls from individual monkeys. Moreover, we find that joint multi-species representations can lead to further improvements over the use of one representation at a time. Our work demonstrates the potential of cross-species data transfer and multi-species representations, as strategies to address tasks on species with very limited data.

Quantizing large language models (LLMs) to 1-bit precision significantly reduces computational costs, but existing quantization techniques suffer from noticeable performance degradation when using weight and activation precisions below 4 bits (W4A4). In this paper, we propose a post-training quantization framework with W(1+1)A(1×4) configuration, where weights are quantized to 1 bit with an additional 1 bit for fine-grain grouping and activations are quantized to 1 bit with a 4-fold increase in the number of channels. For weight quantization, we propose utilizing Hessian-aware fine-grained grouping along with an EM-based quantization scheme. For activation quantization, we decompose INT4-quantized activations into a 4 × INT1 format equivalently and simultaneously smooth the scaling factors based on quantization errors, which further reduces the quantization errors in activations. Our method surpasses state-of-the-art (SOTA) LLM quantization baselines on W2A4 across multiple tasks, pushing the boundaries of existing LLM quantization methods toward fully binarized models. Code is available at https://github.com/JimmyCrave/LLM-PTQ-binarization.

In this paper, we introduce AceMath, a suite of frontier math models that excel in solving complex math problems, along with highly effective reward models capable of evaluating generated solutions and reliably identifying the correct ones. To develop the instruction-tuned math models, we propose a supervised fine-tuning (SFT) process that first achieves competitive performance across general domains, followed by targeted fine-tuning for the math domain using a carefully curated set of prompts and synthetically generated responses. The resulting model, AceMath-72B-Instruct greatly outperforms Qwen2.5-Math-72B-Instruct, GPT-4o and Claude-3.5 Sonnet. To develop math-specialized reward model, we first construct AceMath-RewardBench, a comprehensive and robust benchmark for evaluating math reward models across diverse problems and difficulty levels. After that, we present a systematic approach to build our math reward models. The resulting model, AceMath-72B-RM, consistently outperforms state-of-the-art reward models. Furthermore, when combining AceMath-72B-Instruct with AceMath-72B-RM, we achieve the highest average rm@8 score across the math reasoning benchmarks.

The impressive capabilities of Large Language Models (LLMs) come at the cost of substantial computational resources during deployment. While KV Cache can significantly reduce recomputation during inference, it also introduces additional memory overhead. KV Cache quantization presents a promising solution, striking a good balance between memory usage and accuracy. Previous research has shown that the Keys are distributed by channel, while the Values are distributed by token. Consequently, the common practice is to apply channel-wise quantization to the Keys and token-wise quantization to the Values. However, our further investigation reveals that a small subset of unusual tokens exhibit unique characteristics that deviate from this pattern, which can substantially impact quantization accuracy. To address this, we develop a simple yet effective method to identify these tokens accurately during the decoding process and exclude them from quantization as outlier tokens, significantly improving overall accuracy. Extensive experiments show that our method achieves significant accuracy improvements under 2-bit quantization and can deliver a 6.4 times reduction in memory usage and a 2.3 times increase in throughput.

Large language models (LLMs) achieve strong performance but suffer from slow and costly inference. Existing acceleration methods often lead to noticeable performance degradation, while Mixture-of-Experts (MoE) models require extensive computational resources. In this paper, we propose L0-MoE, a lightweight MoE approach using L0-regularization to accelerate dense LLMs nearly without performance loss. Our method introduces a cluster confusion matrix for domain-aware dataset curation and applies dynamic batching for efficient training. Experiments show that L0-MoE achieves up to 2.5x speedup over dense models while maintaining competitive performance, outperforming existing LLM acceleration baselines.

Retrieval-augmented generation (RAG) has emerged as a pivotal method for expanding the knowledge of large language models. To handle complex queries more effectively, researchers developed Adaptive-RAG (A-RAG) to enhance the generated quality through multiple interactions with external knowledge bases. Despite its effectiveness, A-RAG exacerbates the pre-existing efficiency challenges inherent in RAG, which are attributable to its reliance on multiple iterations of generation. Existing A-RAG approaches process all retrieved contents from scratch. However, they ignore the situation where there is a significant overlap in the content of the retrieval results across rounds. The overlapping content is redundantly represented, which leads to a large proportion of repeated computations, thus affecting the overall efficiency. To address this issue, this paper introduces a model-agnostic approach that can be generally applied to A-RAG methods, which is dedicated to reducing the redundant representation process caused by the overlapping of retrieval results. Specifically, we use cache access and parallel generation to speed up the prefilling and decoding stages respectively. Additionally, we also propose an instruction-driven module to further guide the model to more effectively attend to each part of the content in a more suitable way for LLMs. Experiments show that our approach achieves 2.79 and 2.33 times significant acceleration on average for prefilling and decoding respectively while maintaining equal generation quality.

We introduce AbGen, the first benchmark designed to evaluate the capabilities of LLMs in designing ablation studies for scientific research. AbGen consists of 2,000 expert-annotated examples derived from 677 NLP papers. In this benchmark, LLMs are tasked with generating detailed ablation study designs for a specified module or process based on the given research context. Our evaluation of leading LLMs, such as GPT-4o and Llama-3.1, highlights a significant performance gap between these models and human experts in terms of the importance, faithfulness, and soundness of the ablation study designs. Moreover, we demonstrate that current automated evaluation methods are not reliable for our task, as they show a significant discrepancy when compared to human assessment. To better investigate this, we develop AbGen-Eval, a meta-evaluation benchmark designed to assess the reliability of commonly used automated evaluation systems in measuring LLM performance on our task. We investigate various LLM-based evaluation methods on AbGen-Eval, providing insights for future research on developing more effective and reliable LLM-based evaluation systems for complex scientific tasks.

Multimodal large language models (MLLMs) have enabled LLM-based agents to directly interact with application user interfaces (UIs), enhancing agents’ performance in complex tasks. However, these agents often suffer from high latency and low reliability due to the extensive sequential UI interactions. To address this issue, we propose AXIS, a novel LLM-based agents framework that prioritize actions through application programming interfaces (APIs) over UI actions. This framework also facilitates the creation and expansion of APIs through automated exploration of applications. Our experiments on Microsoft Word demonstrate that AXIS reduces task completion time by 65%-70% and cognitive workload by 38%-53%, while maintaining accuracy of 97%-98% compared to humans. Our work contributes to a new human-agent-computer interaction (HACI) framework and explores a fresh UI design principle for application providers to turn applications into agents in the era of LLMs, paving the way towards an agent-centric operating system (Agent OS). The code and dataset will be available at https://aka.ms/haci_axis.

Recently, large multimodal models (LMMs) have achieved significant advancements. When dealing with high-resolution images, dominant LMMs typically divide them into multiple local images and a global image, leading to a large number of visual tokens. In this work, we introduce AVG-LLaVA, an LMM that can adaptively select the appropriate visual granularity based on the input image and instruction. Specifically, we first apply the multiple pooling layers to obtain visual tokens at different granularities. Then we propose a visual granularity router, which includes a Transformer layer, an MLP layer, and a voter layer, used to select the appropriate visual granularity based on the image and instruction. Furthermore, we put forward RGLF, a novel training paradigm that aims at aligning the granularity predicted by the router with the preferences of the LMM, without the need for additional manually annotated data. Extensive experiments and analysis show that AVG-LLaVA achieves superior performance across 11 benchmarks, as well as significantly reduces the number of visual tokens and speeds up inference (e.g., an 85.3% reduction in visual tokens and a 2.53× increase in inference speed on the AI2D benchmark).

As our awareness of autism and ableism continues to increase, so does our understanding of ableist language towards autistic people. Such language poses a significant challenge in NLP research due to its subtle and context-dependent nature. Yet, detecting anti-autistic ableist language remains underexplored, with existing NLP tools often failing to capture its nuanced expressions. We present AUTALIC, the first dataset dedicated to the detection of anti-autistic ableist language in context, addressing a significant gap in the field. AUTALIC comprises 2,400 autism-related sentences collected from Reddit, accompanied by surrounding context, and annotated by trained experts with backgrounds in neurodiversity. Our comprehensive evaluation reveals that current language models, including state-of-the-art LLMs, struggle to reliably identify anti-autistic ableism and diverge from human judgments, underscoring their limitations in this domain. We publicly release our dataset along with the individual annotations, providing an essential resource for developing more inclusive and context-aware NLP systems that better reflect diverse perspectives.

Multilingual audio-text retrieval (ML-ATR) is a challenging task that aims to retrieve audio clips or multilingual texts from databases. However, existing ML-ATR schemes suffer from inconsistencies for instance similarity matching across languages. To address the inconsistency issue in multilingual audio-text retrieval, we first identify two intuitive factors that contribute to inconsistency: misalignment between audio and multilingual text embeddings, and error propagation in model optimization. By systematically analyzing these factors, we derive theoretical weight error upper bounds for quantifying their effects and find that the main source of inconsistency is the data distribution error during training. This finding motivates our solution to reduce data distribution errors.We propose a consistent ML-ATR scheme using 1-to-k contrastive learning and audio-English co-anchor contrastive learning, aiming to mitigate the negative impact of data distribution error on recall and consistency in ML-ATR. Experimental results on the translated AudioCaps and Clotho datasets show that our scheme achieves state-of-the-art performance on recall and consistency metrics for eight mainstream languages, including English. Our code will be available at https://github.com/ATRI-ACL/ATRI-ACL.

Large Language Model (LLM) agents have demonstrated remarkable generalization capabilities across multi-domain tasks. Existing agent tuning approaches typically employ supervised finetuning on entire expert trajectories. However, behavior-cloning of full trajectories can introduce expert bias and weaken generalization to states not covered by the expert data. Additionally, critical steps—such as planning, complex reasoning for intermediate subtasks, and strategic decision-making—are essential to success in agent tasks, so learning these steps is the key to improving LLM agents. For more effective and efficient agent tuning, we propose ATLAS that identifies the critical steps in expert trajectories and finetunes LLMs solely on these steps with reduced costs. By steering the training’s focus to a few critical steps, our method mitigates the risk of overfitting entire trajectories and promotes generalization across different environments and tasks. In extensive experiments, an LLM finetuned on only 30% critical steps selected by ATLAS outperforms the LLM finetuned on all steps and recent open-source LLM agents. ATLAS maintains and improves base LLM skills as generalist agents interacting with diverse environments.