7 Papers Accepted at ACL 2026

Held from 2–7 July 2026 in San Diego, California, the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) is one of the top global conferences in natural language processing (NLP) and computational linguistics.

Congratulations to the following researchers from A*STAR Centre for Frontier AI Research (A*STAR CFAR) on having their papers accepted at ACL 2026:

• Prof Ivor Tsang, Director, A*STAR CFAR
• Dr Nancy F. Chen, Principal Investigator
• Dr Qu Bohao, Scientist
• Dr Yu Xingrui, Scientist
• Dr Zhang Jie, Scientist

1. CORBA: Contagious Recursive Blocking Attacks on Multi-Agent Systems Based on Large Language Models
   Zhenhong Zhou*, Zherui Li, Jie Zhang, Yuanhe Zhang, Kun Wang, Yang Liu, Qing Guo**
   
   We introduce Denial-of-Collaboration (DoC), a new attack class that disrupts the collaborative structure of LLM multi-agent systems rather than individual components. Our proposed CORBA attack induces benign but recursive communication loops, causing resource exhaustion and system paralysis.
2. From Language to Driving: A Dual-Loop SLM-Enhanced Framework for Multi-Planner Scheduling via a Domain-Specific Language
   Jiawei Liu***, Xun Gong, Muli Yang, Xingrui Yu, Fen Fang, Xulei Yang, Ivor Tsang, Yunfeng Hu, Hong Chen, Qing Guo**
   
   This work frames instruction realisation for autonomous driving as a scheduling problem across multiple motion planners and introduces a dual-loop framework that translates natural language into safe, reliable vehicle control. A small language model handles high-level reasoning and scheduling, while a fast inner loop executes control, with receding-horizon planning, a constrained DSL, and reinforcement learning jointly improving long-horizon performance, safety, and instruction completion.
   
   
3. Safety Sidecar: Reflection-Driven Runtime Control for Safer Agents (ACL Findings)
   Bin Wang***, Jiazheng Quan, Xingrui Yu, Hansen Hu, Hao Yu, Anjun Gao, Zhenglin Wan*, Hui Li, Ivor Tsang
   
   Autonomous LLM agents are powerful but fragile, as small reasoning or retrieval errors can cascade into unsafe actions, and existing defenses lack real-time control and portability. We introduce Safety Sidecar, a model-agnostic runtime module that uses reflective, evidence-driven intervention with external verification to enforce safe execution, improving secure-solution rates across multiple CWE scenarios while maintaining efficiency and correctness.
   
   
4. Learn Like Humans: Use Meta-cognitive Reflection for Efficient Self-Improvement
   Xinmeng Hou*, Peiliang Gong, Bohao Qu, Wuqi Wang, Qing Guo**, Yang Liu
   
   MARS is a self-improving agent framework that enables efficient self-evolution within a single recurrence cycle, avoiding the high computational cost of multi-turn recursive refinement. By combining principle-based and procedural reflection inspired by human learning, it generates optimized instructions that significantly improve reasoning performance across multiple benchmarks.
   
   
5. Can Persona-Prompted LLMs Emulate Subgroup Values? An Empirical Analysis of Generalisability and Fairness in Cultural Alignment
   Bryan Chen Zhengyu Tan*, Zhengyuan Liu, Xiaoyuan Yi, Jing Yao, Xing Xie, Nancy F. Chen, Roy Ka-Wei Lee
   
   We studied whether LLMs can align with the distinct cultural values of demographic subgroups, using Singapore as a case study. It finds that strong LLMs like GPT-4.1 performs poorly on this task, while fine-tuning on structured value data improves accuracy and partially transfers to open-ended generation. However, models exhibit demographic biases, and fine-tuning increases performance disparities across subgroups, highlighting fairness challenges in subgroup-level cultural alignment.
   
   
6. MMAC: A Multilingual, Multimodal Alignment Framework for Cultural Grounding Evaluation
   Weihua Zheng, Zhengyuan Liu, Tanmoy Chakraborty, Weiwen Xu, Xiaoxue Gao, Bryan Chen Zhengyu Tan, Bowei Zou, Chang Liu, Yujia Hu, Xing Xie, Xiaoyuan Yi, Jing Yao, Chaojun Wang, Long Li, Rui Liu, Huiyao Liu, Koji Inoue, Ryuichi Sumida, Tatsuya Kawahara, Fan Xu, Lingyu Ye, Wei Tian, Dongjun Kim, Jimin Jung, Jaehyung Seo, Nadya Yuki Wangsajaya, Pham Minh Duc, Ojasva Saxena, Palash Nandi, Xiyan Tao, Wiwik Karlina, Tuan Luong, Keertana Arun Vasan, Roy Ka-Wei Lee, Nancy F. Chen
   
   We proposed MMAC, a systematic framework that encompasses a tri-modally aligned cultural benchmark creation pipeline and a five-dimensional evaluation protocol to assess cross-country awareness disparities, evaluate cross-lingual and cross-modal consistency, and verify cultural knowledge generalisation and grounding validity.
   
   
7. Programming over Thinking: Efficient and Robust Multi-Constraint Planning
   Derrick Goh Xin Deik*, Quanyu Long, Zhengyuan Liu, Nancy F. Chen, Wenya Wang
   
   We introduce the Scalable Code Planning Engine (SCOPE), a systematic framework that disentangles query-specific problem reasoning from generic code execution. SCOPE first transforms input queries into optimized structured representations, capturing the interdependent constraints, and then autonomously generates reusable solver functions (Combination, Filter, and Deliver) that provide consistent and reliable execution across diverse problems.
   
   

* denotes former CFAR student
** denotes former CFAR researcher
*** denotes current CFAR student
(accurate at time of posting)

More on ACL 2026.