United States

When asked a question in a language less seen in its training data, current reasoning large language models (RLMs) often exhibit dramatically lower performance than when asked the same question in English. In response, we introduce \texttt{SP3F} (Self-Play with Privileged Pairwise Feedback), a two-stage framework for enhancing multilingual reasoning without \textit{any} data in the target language(s). First, we supervise fine-tune (SFT) on translated versions of English question-answer pairs to raise base model correctness. Second, we perform RL with feedback from a pairwise judge in a self-play fashion, with the judge receiving the English reference response as \textit{privileged information}. Thus, even when none of the model&#39;s responses are completely correct, the privileged pairwise judge can still tell which response is better. End-to-end, \texttt{SP3F} greatly improves base model performance, even outperforming fully post-trained models on multiple math and non-math tasks with less than 1/8 of the training data across the single-language, multilingual, and generalization to unseen language settings.

ACL 2026

Gained in Translation: Privileged Pairwise Judges Enhance Multilingual Reasoning

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields 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.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. <br>
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) <br><br><br>
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>
<br><br><br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to ACL 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

Registration Required

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

When asked a question in a language less seen in its training data, current reasoning large language models (RLMs) often exhibit dramatically lower performance than when asked the same question in English. In response, we introduce \texttt{SP3F} (Self-Play with Privileged Pairwise Feedback), a two-stage framework for enhancing multilingual reasoning without \textit{any} data in the target language(s). First, we supervise fine-tune (SFT) on translated versions of English question-answer pairs to raise base model correctness. Second, we perform RL with feedback from a pairwise judge in a self-play fashion, with the judge receiving the English reference response as \textit{privileged information}. Thus, even when none of the model's responses are completely correct, the privileged pairwise judge can still tell which response is better. End-to-end, \texttt{SP3F} greatly improves base model performance, even outperforming fully post-trained models on multiple math and non-math tasks with less than 1/8 of the training data across the single-language, multilingual, and generalization to unseen language settings.

technical paper

Recent progress in large language models (LLMs) is largely driven by scaling training compute through either pre-training with next-token prediction (NTP) or post-training with reinforcement learning (RL). The former contributes to learning broad knowledge and skills from general data, while struggling with data inefficiency and catastrophic forgetting in continual learning settings. The latter incentivizes reasoning capabilities with strong generalization, but is constrained by limited data availability due to its reliance on human annotation. To alleviate these issues, we propose Reinforcement Learning on Pre-Training data (RLPT), which combines the advantages of learning from general data and RL. In particular, RLPT derives reward signals directly from general text data through a next-segment reasoning objective, rewarding the policy for correctly predicting next text segments conditioned on the prefix text. Experiments across multiple benchmarks and models demonstrate the effectiveness of \ours. For example, RLPT yields substantial improvements in continual pre-training ($+4.6\%$) and provides a strong foundation for post-training ($+3.4\%$) on Qwen3-8B-Base.

Reinforcement Learning on Pre-Training Data

Retrieval-augmented generation (RAG) extends large language models (LLMs) with external knowledge, but it must balance limited effective context, redundant retrieved evidence, and the loss of fine-grained facts under aggressive compression. Pure compression-based approaches reduce input size but often discard fine-grained details essential for factual accuracy. We propose SARA, a hybrid RAG framework that targets answer quality under fixed token budgets by combining natural-language snippets with semantic compression vectors. SARA retains a small set of passages in text form to preserve entities and numerical values, compresses the remaining evidence into interpretable vectors for broader coverage, and uses those vectors for iterative evidence reranking. Across 9 datasets and 5 open-source LLMs spanning 3 model families (Mistral, Llama, and Gemma), SARA consistently improves answer relevance (+17.71), answer correctness (+13.72), and semantic similarity (+15.53), demonstrating the importance of integrating textual and compressed representations for robust, context-efficient RAG.

SARA: Selective and Adaptive Retrieval-augmented Generation with Context Compression

While large language models have revolutionized Text-to-SQL tasks, translating natural language into Graph Query Languages (Text2GQL) remains underexplored due to the topological heterogeneity and syntactic diversity of graph query languages (e.g., Cypher, Gremlin, SPARQL). Existing approaches often struggle with structural hallucinations and lack adaptability in cold-start scenarios. 
In this paper, we present a unified, training-free Text2GQL framework. First, Structural Twig Linking elevates schema grounding to the identification of semantic substructures (``twigs''), providing robust topological priors. Second, addressing data scarcity, Evolutionary In-Context Learning operates in a Tabula Rasa setting to implicitly construct a self-growing repository of verified examples driven by syntactic utility. Finally, our Adversarial Execution-Guided Correction agent enforces fidelity through synergistic static critique and dynamic verification. Experiments demonstrate significant improvements over baselines in both accuracy and executability across diverse GQLs. The code is available at https://github.com/nf202/Text2Graph.

Adaptive Text2GQL: Integrating Structural Twig Linking and Evolutionary In-Context Learning

Large language models (LLMs) are increasingly used by end users, yet existing personalization methods relying on static profiles or text-only signals fail to capture query-specific expertise variation. We present ExPerT, a query-wise personalization framework that adapts LLM responses to users' query domain expertise by combining semantic and behavioral cues. ExPerT consists of two key components: (i) a semantic–behavioral expertise inference module that jointly interprets query text and keystroke dynamics via in-context LLM prompting, and (ii) an expertise-conditioned response generation that adapts the level of detail, terminology, and conceptual complexity. Our user study with 40 participants and 1270 queries demonstrated that ExPerT reduced expertise inference error by 65.7% compared to the strongest baseline (MAE = 0.398 vs. 1.162) and improved response satisfaction by 17.52% (from 3.71 to 4.36) on a 5‑point Likert scale.

ExPerT: Personalizing LLM Responses to Users' Domain Expertise via Query-Wise Semantic and Keystroke Behavioral Cues

Binary Code Similarity Detection (BCSD) plays a vital role in various security applications, including vulnerability identification, malware analysis, and code plagiarism detection. With the growing adoption of deep neural networks (DNNs), substantial progress has been made in recognizing and classifying similar code segments. However, DNN-based BCSD methods often exhibit low accuracy and robustness because they struggle to capture fine-grained and high-level program semantics. In contrast, such semantics are typically captured through natural language interpretations of source code by large language models (LLMs). Yet, LLM-based BCSD methods are constrained by their large model sizes and high inference latency. To alleviate these limitations, this paper proposes BinSKD. The key idea is to leverage an LLM-based BCSD method as the teacher model and transfer its knowledge of high-level program semantics to various DNN-based student models. Specifically, to avoid propagating errors from the teacher to the student, we introduce selective distillation, selecting targets with accurate semantics according to their detection retrieval. In addition, to mitigate the noise introduced by a number of negative samples during distillation, we further propose discrepancy-weighted sampling to focus on the samples
where the student’s prediction notably deviates from the teacher’s. Our experiments show that BinSKD yields Recall@1 improvements of 14.5\%–91.2\% for DNN-based BCSD methods and enables HermesSim to match the teacher's performance with orders-of-magnitude efficiency.

Selective Knowledge Distillation: Fusing LLM Semantic Strengths with DNN Efficiency for Binary Code Similarity Detection

Enzyme–reaction retrieval is a fundamental problem in computational biology, underpinning enzyme characterization, reaction mechanism elucidation, and the rational design of metabolic pathways and biocatalysts. As a bidirectional task, it entails both enzyme-to-reaction and reaction-to-enzyme mapping. However, existing approaches suffer from poor generalization across tasks and distributions, with performance highly sensitive to dataset splits and substantial asymmetry between retrieval directions. To address these challenges, we present TIGER, a Text-Informed Generalized Enzyme-Reaction Retrieval framework that leverages protein-to-text generation models to distill textual semantic knowledge from enzyme sequences, providing a generalized representation that bridges enzymes and biochemical reactions. To ensure the quality and reliability of textual semantics, we design a Dynamic Gating Network that adaptively fuses text-derived knowledge with sequence features, enabling more consistent and informative enzyme representations, while a Structure-Shared Feature Projector aligns enzyme and reaction representations within a unified latent space. Extensive experiments demonstrate that, under bidirectional retrieval supervision, TIGER significantly outperforms state-of-the-art baselines across diverse distributions and exhibits strong robustness and transferability across tasks.

TIGER: Text-Informed Generalized Enzyme-Reaction Retrieval

Diffusion large language models (dLLMs) generate text by repeatedly unmasking a partially noised sequence in parallel, promising lower latency than autoregressive decoding. However, most discrete dLLMs still rely on fixed denoising schedules, which are non‑adaptive to input difficulty and cannot learn efficient unmasking orders. This paper introduces a reinforcement learning (RL) framework that transforms dLLM decoding into a trajectory-aware, learnable policy. We propose a confidence-gated denoising strategy that dynamically decides which tokens to unmask and how many to unmask per step, enabling adaptive exploration of denoising trajectories. Building on Group Relative Policy Optimization, we reformulate it into a trajectory-aware variant, TA-GRPO-\textit{d}, which combines a trajectory-level signal—captured as the z‑score of the AUC over intermediate rewards—with a token-level unmasking-time weight. This design allows the model to learn not only the final output quality but also the efficiency of the decoding path itself. Experiments on MATH-500, Countdown, Sudoku, and code benchmarks (HumanEval, MBPP) show that TA-GRPO-\textit{d} maintains or improves accuracy while reducing average denoising steps by up to half, achieving both faster inference and lower computational cost. Our approach provides an RL framework for optimizing dLLM decoding policies toward adaptive, efficient reasoning. Code is available at our GitHub.

TA-GRPO-d: Trajectory-Aware GRPO for Optimizing Denoising Trajectories in Diffusion LLMs

Idioms are difficult to transfer across languages due to their non-compositionality and weak surface-form grounding, making literal mappings unreliable. We present G-IdiomAlign, a gloss-pivoted benchmark where each idiom is anchored by an English gloss from Wiktionary. We further construct a high-confidence reference alignment set for reproducible evaluation. G-IdiomAlign supports two protocols: (1) a controlled Multiple-Choice Idiom Equivalence with typed distractors for error attribution; and (2) a Gloss-Contrastive Generation contrasting No-gloss and With-gloss inputs to isolate the effect of an explicit semantic pivot. Across diverse LLMs, a bias to literal translation is a dominant failure mode, especially when the target is a low-resource language. Glosses consistently improve Gloss-Contrastive Generation under an embedding-based semantic proxy, but performance remains modest, indicating substantial headroom in the open output space. Subsequent analysis on Qwen3-8B further suggests that cross-condition differences are concentrated more in attention heads than in layers, while better With-gloss generations coincide with stronger gloss anchoring.

G-IdiomAlign: A Gloss-Pivoted Benchmark for Cross-Lingual Idiom Alignment

Despite representing nearly one-third of the world's languages, African languages remain critically underserved by modern NLP technologies, with 88\% classified as severely underrepresented or completely ignored in computational linguistics. We present the African Languages Lab (All Lab), a comprehensive research initiative that addresses this technological gap through systematic data collection, model development, and empirical analysis. 

Our contributions include: (1) a quality-controlled data collection pipeline, yielding the largest validated African multi-modal speech and text dataset spanning 40 languages with 19 billion text tokens and 12,628 hours of aligned speech data; (2) extensive experimental validation demonstrating that even modest-scale models, when fine-tuned on targeted language data, achieve substantial improvements over untrained baselines, averaging +23.69 ChrF++, +0.33 COMET, and +15.34 BLEU points across 31 evaluated languages; and (3) a comparative analysis against Google Translate in which a 1B-parameter model matched or surpassed the commercial system in several languages including Yoruba and Twi, revealing that data scarcity, rather than model scale, constitutes the primary bottleneck for low-resource NLP, and suggesting that systematic dataset development yields disproportionate returns for low-resource languages.

The African Languages Lab: A Collaborative Approach to Advancing Low-Resource African NLP

Video Large Language Models (VideoLLMs) excel at video understanding tasks where outputs are textual, such as Video Question Answering and Video Captioning. However, they underperform specialized embedding-based models in Retrieval tasks, such as Text-toVideo Retrieval and Moment Retrieval. We introduce ViLL-E (Video-LLM-Embed), a unified VideoLLM architecture endowed with a novel embedding generation mechanism that allows the model to "think longer" for complex videos and stop early for easy ones. We train this model with a three-stage training methodology combining generative and contrastive learning: initial large-scale pre-training with video-caption pairs; followed by continual training on a smaller, detailed-caption dataset; and concluding with task-specific fine-tuning on a novel multi-task dataset covering Video QA, Temporal Localization, Video Retrieval, and Video-Text Matching. Our model significantly improves temporal localization (on avg. 7% over other VideoLLMs) and video retrieval (up to 4% over dual encoder models), achieving performance comparable to state-of-the-art specialized embedding models while remaining competitive on VideoQA tasks. Furthermore, our joint contrastive-generative training unlocks new zero-shot capabilities, significantly outperforming state-of-the-art methods in composed video retrieval (+5% over SotA) and retrieval from long text (+2% over SotA).

Downloads

Next from ACL 2026

Reinforcement Learning on Pre-Training Data

Similar lecture

Masked by Consensus: Disentangling Privileged Knowledge in LLM Correctness