Austria

Synthetic data has driven recent state-of-the-art advancements, but reliance on a single oracle teacher model can lead to model collapse and bias propagation. These issues are particularly severe in multilingual settings, where no single model excels across all languages. In this study, we propose multilingual arbitration, which exploits performance variations among multiple models for each language. By strategically routing samples through a diverse set of models, each with unique strengths, we mitigate these challenges and enhance multilingual performance. Extensive experiments with state-of-the-art models demonstrate that our approach significantly surpasses single-teacher distillation, achieving up to 80% win rates over proprietary and open-weight models like Gemma 2, Llama 3.1, and Mistral v0.3, with the largest improvements in low-resource languages.

ACL 2025

Multilingual Arbitration: Optimizing Data Pools to Accelerate Multilingual Progress

synthetic data generation

distillation

multilingual

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

ACL 2024

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

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)

Existing LLM-based agents have achieved strong performance on held-in tasks, but their generalizability to unseen tasks remains poor. Hence, some recent work focus on fine-tuning the policy model with more diverse tasks to improve the generalizability. In this work, 
we find that finetuning a reward model to guide the policy model is more robust than directly finetuning the policy model.
Based on this finding, we propose AgentRM, a 8B generalizable reward model, to guide the policy model for effective test-time search.
We comprehensively investigate three approaches to construct the reward model, including explicit reward modeling, implicit reward modeling and LLM-as-a-judge.
We then use AgentRM to guide the answer generation with Best-of-N sampling and beam search.
We show that AgentRM is robust to paraphrasings of task instructions and can generalize to unseen tasks that require novel optimal behavior.
Through extensive evaluation across nine tasks spanning four categories, AgentRM enhances the non-finetuned 8B policy model by 8.8 points on average, surpassing the top general agent by 4.0.
Moreover, it demonstrates weak-to-strong generalization, yielding greater improvement on more powerful policy models.
As for the specializability, AgentRM can also boost a finetuned policy model and outperform the top specialized agent by 11.4 on three held-in tasks.
Further analysis verifies its effectiveness in test-time scaling.
We release the code and data at https://github.com/thunlp/AgentRM.

AgentRM: Enhancing Agent Generalization with Reward Modeling

Causal masking is a fundamental component in Generative Pre-Trained Transformer (GPT) models, playing a crucial role during training. Although GPTs can process the entire user prompt at once, the causal masking is applied to all input tokens step-by-step, mimicking the generation process. This imposes an unnecessary constraint during the initial "prefill" phase when the model processes the input prompt and generates the internal representations before producing any output tokens. In this work, attention is masked based on the known block structure at the prefill phase, followed by the conventional token-by-token autoregressive process after that. For example, in a typical chat prompt, the system prompt is treated as one block, and the user prompt as the next one. Each of these is treated as a unit for the purpose of masking, such that the first tokens in each block can access the subsequent tokens in a non-causal manner. Then, the model answer is generated in the conventional causal manner. The Segment-by-Segment scheme entails no additional computational overhead. When integrated using a lightweight fine-tuning into already trained models such as Llama and Qwen, MAS quickly increases models' performances.

Segment-Based Attention Masking for GPTs

High computation costs and latency of large language models such as GPT-4 have limited their deployment in clinical settings. Small language models (SLMs) offer a cost-effective alternative, but their limited capacity requires biomedical domain adaptation, which remains challenging. An additional bottleneck is the unavailability and high sensitivity of clinical data. To address these challenges, we propose a novel framework for adapting SLMs into high-performing clinical models. We introduce the MediPhi collection of 3.8B-parameter SLMs developed with our novel framework: pre-instruction tuning of experts on relevant medical and clinical corpora (PMC, Medical Guideline, MedWiki, etc.), model merging, and clinical-tasks alignment. To cover most clinical tasks, we extended the CLUE benchmark to CLUE+, doubling its size. Our expert models deliver relative improvements on this benchmark over the base model without any task-specific fine-tuning: 64.3% on medical entities, 49.5% on radiology reports, and 44% on ICD-10 coding (outperforming GPT-4-0125 by 14%). We unify the expert models into MediPhi via model merging, preserving gains across benchmarks. Furthermore, we built the MediFlow collection, a synthetic dataset of 2.5 million high-quality instructions on 14 medical NLP tasks, 98 fine-grained document types, and JSON format support. Alignment of MediPhi using supervised fine-tuning and direct preference optimization achieves further gains of 18.9% on average.

A Modular Approach for Clinical SLMs Driven by Synthetic Data with Pre-Instruction Tuning, Model Merging, and Clinical-Tasks Alignment

Incorporating automatic speech recognition (ASR) into field linguistics workflows for language documentation has become increasingly common. While ASR performance has seen improvements in low-resource settings, obstacles remain when training models on data collected by documentary linguists. One notable challenge lies in the way that this data is curated. ASR datasets built from spontaneous speech are typically recorded in consistent settings and transcribed by native speakers following a set of well designed guidelines. In contrast, field linguists collect data in whatever format it is delivered by their language consultants and transcribe it as best they can given their language skills and the quality of the recording. This approach to data curation, while valuable for linguistic research, does not always align with the standards required for training robust ASR models. In this paper, we explore methods for identifying speech transcriptions in fieldwork data that may be unsuitable for training ASR models. We focus on two complimentary automated measures of transcription quality that can be used to identify transcripts with characteristics that are common in field data but could be detrimental to ASR training. We show that one of the metrics is highly effective at retrieving these types of transcriptions. Additionally, we find that filtering datasets using this metric of transcription quality reduces WER both in controlled experiments using simulated fieldwork with artificially corrupted data and in real fieldwork corpora.

That doesn’t sound right: Evaluating speech transcription quality in field linguistics corpora

Scaling the test-time compute of large language models has demonstrated impressive performance on reasoning benchmarks. However, existing evaluations of test-time scaling make the strong assumption that a reasoning system should always give an answer to any question provided. This overlooks concerns about whether a model is confident in its answer, and whether it is appropriate to always provide a response. To address these concerns, we extract confidence scores during reasoning for thresholding model responses. We find that increasing compute budget at inference time not only helps models answer more questions correctly, but also increases confidence in correct responses. We then extend the current paradigm of zero-risk responses during evaluation by considering settings with non-zero levels of response risk, and suggest a recipe for reporting evaluations under these settings.

Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering

Multimodal learning has celebrated remarkable success across diverse areas, yet faces the challenge of prohibitively expensive data collection and annotation when adapting models to new environments. In this context, domain adaptation has gained growing popularity as a technique for knowledge transfer, which, however, remains underexplored in multimodal settings compared with unimodal ones. This paper investigates multimodal domain adaptation, focusing on a practical partially shifting scenario where some modalities (referred to as anchors) remain domain-stable, while others (referred to as drifts) undergo a domain shift. We propose a bi-alignment scheme to simultaneously perform drift-drift and anchor-drift matching. The former is achieved through adversarial learning, aligning the representations of the drifts across source and target domains; the latter corresponds to an "anchor dragging drift" strategy, which matches the distributions of the drifts and anchors within the target domain using the optimal transport (OT) method. The overall design principle features \textbf{A}dversarial \textbf{A}lignment with \textbf{A}nchor \textbf{D}ragging \textbf{D}rift, abbreviated as \textbf{$A^3D^2$}, for multimodal domain adaptation with partially shifted modalities. Comprehensive empirical results verify the effectiveness of the proposed approach, and demonstrate that $A^3D^2$ achieves superior performance compared with state-of-the-art approaches. The code is available at: \url{https://github.com/sunjunaimer/A3D2.git}.

Adversarial Alignment with Anchor Dragging Drift (A3D2): Multimodal Domain Adaptation with Partially Shifted Modalities

Retrieval-augmented generation (RAG) helps address the limitations of parametric knowledge embedded within a language model (LM). In real world settings, retrieved information can vary in complexity, yet most investigations of LM utilisation of context has been limited to synthetic text. We introduce DRUID (Dataset of Retrieved Unreliable, Insufficient and Difficult-to-understand contexts) with real-world queries and contexts manually annotated for stance. The dataset is based on the prototypical task of automated claim verification, for which automated retrieval of real-world evidence is crucial. We compare DRUID to synthetic datasets (CounterFact, ConflictQA) and find that artificial datasets often fail to represent the complexity and diversity of realistically retrieved context. We show that synthetic datasets exaggerate context characteristics rare in real retrieved data, which leads to inflated context utilisation results, as measured by our novel ACU score. Moreover, while previous work has mainly focused on singleton context characteristics to explain context utilisation, correlations between singleton context properties and ACU on DRUID are surprisingly small compared to other properties related to context source. Overall, our work underscores the need for real-world aligned context utilisation studies to represent and improve performance in real-world RAG settings.

A Reality Check on Context Utilisation for Retrieval-Augmented Generation

Large Language Models (LLMs) have revolutionized various domains, including natural language processing, data analysis, and software development, by enabling automation. In software engineering, LLM-powered coding agents have garnered significant attention due to their potential to automate complex development tasks, assist in debugging, and enhance productivity. However, existing approaches often struggle with sub-optimal decision-making, requiring either extensive manual intervention or inefficient compute scaling strategies. To improve coding agent performance, we present Dynamic Action Re-Sampling (DARS), a novel inference time compute scaling approach for coding agents, that is faster and more effective at recovering from sub-optimal decisions compared to baselines. While traditional agents either follow linear trajectories or rely on random sampling for scaling compute, our approach DARS works by branching out a trajectory at certain key decision points by taking an alternative action given the history of the trajectory and execution feedback of the previous attempt from that point. We evaluate our approach on SWE-Bench Lite benchmark, demonstrating that this scaling strategy achieves a pass@k score of 55% with Claude 3.5 Sonnet V2. Our framework achieves a pass@1 rate of 47%, outperforming state-of-the-art (SOTA) open-source frameworks.

DARS: Dynamic Action Re-Sampling to Enhance Coding Agent Performance by Adaptive Tree Traversal

Multi-task learning (MTL) enhances efficiency by sharing representations across tasks, but task dissimilarities often cause partial learning, where some tasks dominate while others are neglected. Existing methods mainly focus on balancing loss or gradients but fail to fundamentally address this issue due to the representation discrepancy in latent space. In this paper, we propose variance-invariant probabilistic decoding for multi-task learning (VIP-MTL), a framework that ensures impartial learning by harmonizing representation spaces across tasks. VIP-MTL decodes shared representations into task-specific probabilistic distributions and applies variance normalization to constrain these distributions to a consistent scale. Experiments on two language benchmarks show that VIP-MTL outperforms 12 representative methods under the same multi-task settings, especially in heterogeneous task combinations and data-constrained scenarios. Further analysis shows that VIP-MTL is robust to sampling distributions, efficient on optimization process, and scale-invariant to task losses. Additionally, the learned task-specific representations are more informative, enhancing the language understanding abilities of pre-trained language models under the multi-task paradigm.

Impartial Multi-task Representation Learning via Variance-invariant Probabilistic Decoding

Large language models (LLMs) have recently pushed open-domain question answering (ODQA) to new frontiers. However, prevailing retriever–reader pipelines often depend on multiple rounds of prompt-level instructions, leading to high computational overhead, instability, and suboptimal retrieval coverage. In this paper, we propose EmbQA, an embedding-level framework that alleviates these shortcomings by enhancing both the retriever and the reader. Specifically, we refine query representations via lightweight linear layers under an unsupervised contrastive learning objective, thereby reordering retrieved passages to highlight those most likely to contain correct answers. Additionally, we introduce an exploratory embedding that broadens the model’s latent semantic space to diversify candidate generation and employs an entropy-based selection mechanism to choose the most confident answer automatically. Extensive experiments across three open-source LLMs, three retrieval methods, and four ODQA benchmarks demonstrate that EmbQA substantially outperforms recent baselines in both accuracy and efficiency.

Downloads

Next from ACL 2025

AgentRM: Enhancing Agent Generalization with Reward Modeling

Similar lecture

Knowledge Distillation from Monolingual to Multilingual Models for Intelligent and Interpretable Multilingual Emotion Detection

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES