Papers by Su Yan

The Illusion of Specialization: Unveiling the Domain-Invariant "Standing Committee" in Mixture-of-Experts Models (2026.acl-long)

Challenge:	Current physics benchmarks focus on text-only inputs or only on problem-solving . current physics reasoning benchmarks neglect critical intermediate steps of variable identification and process formulation.
Approach:	a new benchmark evaluates multimodal large language models in physics reasoning . the benchmark measures variables, process formulations, and solution derivation .
Outcome:	PhysicsArena is the first multimodal physics reasoning benchmark . it evaluates MLLMs across three critical dimensions: variable identification, process formulation, and solution derivation.

Copied to clipboard

Yan Wang, Yitao Xu, Nanhan Shen, Jinyan Su, Jimin Huang, Zining Zhu

Challenge:	Mixture of Experts models are widely assumed to achieve domain specialization through sparse routing.
Approach:	They propose a framework that analyzes routing behavior at the level of expert groups rather than individual experts.
Outcome:	The proposed framework analyzes routing behavior at the level of expert groups rather than individual experts.

Topology Matters: Measuring Memory Leakage in Multi-Agent LLMs (2026.findings-acl)

Copied to clipboard

Jinbo Liu, Defu Cao, Yifei Wei, Tianyao Su, Yuan Liang, Yushun Dong, Yan Liu, Yue Zhao, Xiyang Hu

Robustness in NLP

Challenge:	Multi-agent LLMs are rapidly moving from prototype to real-world use . network topology is a first-order security parameter in multi-aggent systems .
Approach:	They propose a framework for comparing topology-conditioned memory leakage in multi-agent LLM systems.
Outcome:	The proposed framework evaluates topology-conditioned memory leakage in multi-agent LLM systems.

Battle of the Large Language Models: Dolly vs LLaMA vs Vicuna vs Guanaco vs Bard vs ChatGPT - A Text-to-SQL Parsing Comparison (2023.findings-emnlp)

Copied to clipboard

Shuo Sun, Yuchen Zhang, Jiahuan Yan, Yuze Gao, Donovan Ong, Bin Chen, Jian Su

Programming Languages in NLP

Challenge:	a number of open-source large language models claim to be performing better than commercial ones . however, these models fall short of the performance achieved by closed-source models like GPT-3.5 .
Approach:	They evaluate six popular large language models against each other to evaluate their performance . authors say open-source models are not as effective as those built by commercial models .
Outcome:	a new set of models claim to match or surpass the language understanding abilities of commercial models . the results show that the models performed far below the performance of closed-source models compared to open-source ones .

Multi-hop Question Generation with Graph Convolutional Network (2020.findings-emnlp)

Copied to clipboard

Dan Su, Yan Xu, Wenliang Dai, Ziwei Ji, Tiezheng Yu, Pascale Fung

ErrorRadar: Benchmarking Complex Mathematical Reasoning of Multimodal Large Language Models Via Error Detection (2026.findings-acl)

Challenge:	Existing studies on text-based QG focus on generating SQuAD-style questions.
Approach:	They propose a multi-hop question generation model that does context encoding in multiple hops with Graph Convolutional Network and encoder fusion via an Encoder Reasoning Gate.
Outcome:	Empirical results show that the proposed model generates fluent questions with high completeness and outperforms baselines on automatic evaluation metrics.

Copied to clipboard

Yibo Yan, Shen Wang, Jiahao Huo, Hang Li, Boyan Li, Jiamin Su, Xiong Gao, YiFan Zhang, Tianlong Xu, Zhendong Chu, Aoxiao Zhong, Kun Wang, Hui Xiong, Philip S. Yu, Xuming Hu, Qingsong Wen

KGPT: Knowledge-Grounded Pre-Training for Data-to-Text Generation (2020.emnlp-main)

Challenge:	Current mathematical benchmarks focus on evaluating MLLMs’ problem-solving ability, yet there is a crucial gap in addressing more complex scenarios such as error detection.
Approach:	They propose to evaluate multimodal error detection by evaluating two sub-tasks error step identification and error categorization.
Outcome:	The proposed task evaluates MLLMs' ability to handle multimodal questions compared to text-only models.

Copied to clipboard

Wenhu Chen, Yu Su, Xifeng Yan, William Yang Wang

DemonAgent: Dynamically Encrypted Multi-Backdoor Implantation Attack on LLM-based Agent (2025.findings-emnlp)

Challenge:	Existing methods for data-to-text generation rely on labeled data, which is costly to acquire and limits their application to new tasks and domains.
Approach:	They propose to leverage pre-training and transfer learning to address this problem by leveraging a general knowledge-grounded generation model and a knowledge-based model.
Outcome:	The proposed model can generate knowledge-enriched text on a knowledge-grounded text corpus crawled from the web in three settings.

Copied to clipboard

Pengyu Zhu, Zhenhong Zhou, Yuanhe Zhang, Shilinlu Yan, Kun Wang, Sen Su

Robustness in NLP

Challenge:	a new method for detecting advanced backdoors is proposed to bypass safety audits.
Approach:	They propose a backdoor implantation strategy that introduces dynamic encryption to bypass safety audits.
Outcome:	The proposed method achieves an attack success rate approaching 100% while maintaining a detection rate of 0%.

ShopSimulator: Evaluating and Exploring RL-Driven LLM Agent for Shopping Assistants (2026.acl-long)

Copied to clipboard

Pei Wang, Yanan Wu, Xiaoshuai Song, Weixun Wang, Gengru Chen, Zhongwen Li, Kezhong Yan, Qi Liu, Ken Deng, Shuaibing Zhao, Shaopan Xiong, Xuepeng Liu, Xuefeng Chen, Wanxi Deng, Wenbo Su, Bo Zheng

Challenge:	Existing studies on large language model-based agents focus on evaluation benchmarks without training support.
Approach:	They propose a large-scale Chinese shopping simulation environment that uses large language models to train agents.
Outcome:	The proposed model performs poorly in a large-scale and challenging shopping environment in China.

What It Takes to Achieve 100% Condition Accuracy on WikiSQL (D18-1)

Copied to clipboard

Semih Yavuz, Izzeddin Gur, Yu Su, Xifeng Yan

Generalizing Question Answering System with Pre-trained Language Model Fine-tuning (D19-58)

Challenge:	despite of its simplicity, none of the publicly reported structured query generation models can achieve an accuracy beyond 62%, which is far from enough for practical use.
Approach:	They propose a model that can achieve 88.6% condition accuracy on WikiSQL . they ask: why is the accuracy still low for such simple queries?
Outcome:	The proposed solution can reach up to 88.6% condition accuracy on the WikiSQL dataset.

Copied to clipboard

Dan Su, Yan Xu, Genta Indra Winata, Peng Xu, Hyeondey Kim, Zihan Liu, Pascale Fung

Natural Language Interfaces

Challenge:	Existing methods focus on improving in-domain performance, leaving open the question of how they can generalize to out-of-domain and unseen RC tasks.
Approach:	They propose a multi-task learning framework that learns the shared representation across different tasks and builds on a large pre-trained language model and fine-tuned on multiple RC datasets.
Outcome:	The proposed framework improves the BERT-Large baseline by 8.39 and 7.22 respectively.

Keep the Primary, Rewrite the Secondary: A Two-Stage Approach for Paraphrase Generation (2021.findings-acl)

Copied to clipboard

Yixuan Su, David Vandyke, Simon Baker, Yan Wang, Nigel Collier

EssayJudge: A Multi-Granular Benchmark for Assessing Automated Essay Scoring Capabilities of Multimodal Large Language Models (2025.findings-acl)

Challenge:	Existing approaches to generate paraphrases are decomposable, but some use a sequence-to-sequence model to generate each word in a uniform way.
Approach:	They propose a framework for identification then aggregation of input tokens and a custom decoder to generate paraphrases.
Outcome:	The proposed framework outperforms previous studies on two benchmark datasets and generates paraphrases in interpretable and controllable way.

Copied to clipboard

Jiamin Su, Yibo Yan, Fangteng Fu, Zhang Han, Jingheng Ye, Xiang Liu, Jiahao Huo, Huiyu Zhou, Xuming Hu

NusaCrowd: Open Source Initiative for Indonesian NLP Resources (2023.findings-acl)

Challenge:	Automated Essay Scoring (AES) systems face three major challenges: reliance on handcrafted features that limit generalizability, difficulty in capturing fine-grained traits like coherence and argumentation, and inability to handle multimodal contexts.
Approach:	They propose a multimodal benchmark to evaluate AES capabilities across lexical-, sentence-, and discourse-level traits without manual feature engineering.
Outcome:	The proposed system can evaluate AES capabilities across lexical-, sentence-, and discourse-level traits without manual feature engineering.

Copied to clipboard

Samuel Cahyawijaya, Holy Lovenia, Alham Fikri Aji, Genta Winata, Bryan Wilie, Fajri Koto, Rahmad Mahendra, Christian Wibisono, Ade Romadhony, Karissa Vincentio, Jennifer Santoso, David Moeljadi, Cahya Wirawan, Frederikus Hudi, Muhammad Satrio Wicaksono, Ivan Parmonangan, Ika Alfina, Ilham Firdausi Putra, Samsul Rahmadani, Yulianti Oenang, Ali Septiandri, James Jaya, Kaustubh Dhole, Arie Suryani, Rifki Afina Putri, Dan Su, Keith Stevens, Made Nindyatama Nityasya, Muhammad Adilazuarda, Ryan Hadiwijaya, Ryandito Diandaru, Tiezheng Yu, Vito Ghifari, Wenliang Dai, Yan Xu, Dyah Damapuspita, Haryo Wibowo, Cuk Tho, Ichwanul Karo Karo, Tirana Fatyanosa, Ziwei Ji, Graham Neubig, Timothy Baldwin, Sebastian Ruder, Pascale Fung, Herry Sujaini, Sakriani Sakti, Ayu Purwarianti

Challenge:	Existing NLP research in Indonesian languages has been held back by factors such as language diversity, orthographic variation, resource limitation and other societal challenges.
Approach:	They present a collaborative initiative to collect and unify existing resources for Indonesian languages and open access to previously non-public resources.
Outcome:	The results show that the datasets are highly reliable and can be used to generate the first zero-shot benchmarks for natural language understanding and generation in Indonesian and the local languages of Indonesia.

Demonstration Augmentation for Zero-shot In-context Learning (2024.findings-acl)

Copied to clipboard

Yi Su, Yunpeng Tai, Yixin Ji, Juntao Li, Yan Bowen, Min Zhang

Low-Resource NLP

Challenge:	Large Language Models (LLMs) have demonstrated an impressive capability known as In-context Learning (ICL), which enables them to acquire knowledge from textual demonstrations without the need for parameter updates.
Approach:	They propose to use model’s previously predicted historical samples as demonstrations for subsequent ones to improve model’ s performance.
Outcome:	The proposed method significantly outperforms the previous method and its predecessors in terms of inference cost and time.

SEE: Signal Embedding Energy for Quantifying Noise Interference in Large Audio Language Models (2026.acl-long)

Copied to clipboard

Yuanhe Zhang, Jiayu Tian, Yibo Zhang, Shilinlu Yan, Liang Lin, Zhenhong Zhou, Li Sun, Sen Su

MT-R1-Zero: Advancing LLM-based Machine Translation via R1-Zero-like Reinforcement Learning (2025.findings-emnlp)

Challenge:	Existing studies on noise lack quantitative analysis and rely on intuition and empirical observation, thus failing to understand practical robustness.
Approach:	They propose a method for quantifying the impact of noise intensity on LALM inputs by using a structured activation subspace derived from the model's internal representations.
Outcome:	The proposed method outperforms existing denoising methods and demonstrates that noise is perceived more accurately than raw audio features.

Copied to clipboard

Zhaopeng Feng, Shaosheng Cao, Jiahan Ren, Jiayuan Su, Ruizhe Chen, Yan Zhang, Jian Wu, Zuozhu Liu

Machine Translation

Challenge:	Large-scale reinforcement learning (RL) methods have proven effective in enhancing the reasoning abilities of large language models.
Approach:	They propose an open-source adaptation of the R1-Zero RL framework for machine translation (MT) their code is available at https://github.com/fzp0424/MT-R1-zero.
Outcome:	The proposed framework surpasses towerinstruct-7B-v0.2 on the english-chinese benchmark by 1.26 points.

DialSQL: Dialogue Based Structured Query Generation (P18-1)

Copied to clipboard

Izzeddin Gur, Semih Yavuz, Yu Su, Xifeng Yan

Natural Language Interfaces

Challenge:	Recent advances in deep learning and semantic parsing have improved the translation accuracy of natural language questions to structured queries.
Approach:	They propose a dialogue-based structured query generation framework that leverages human intelligence to boost performance of existing algorithms via user interaction.
Outcome:	The proposed framework improves on a WikiSQL dataset from 61.3% to 69.0% using only 2.4 validation questions per dialogue.

EpLSA: Synergy of Expert-prefix Mixtures and Task-Oriented Latent Space Adaptation for Diverse Generative Reasoning (2024.lrec-main)

Copied to clipboard

Fujun Zhang, Xiangdong Su, Jiang Li, Rong Yan, Guanglai Gao

How Large a Vocabulary Does Text Classification Need? A Variational Approach to Vocabulary Selection (N19-1)

Challenge:	Existing models for diverse generative reasoning struggle to generate multiple unique and plausible results.
Approach:	They propose a model based on expert-prefix mixtures and task-oriented latent space adaptation for diverse generative reasoning.
Outcome:	The proposed model outperforms existing models on three types of generative reasoning tasks.

Copied to clipboard

Wenhu Chen, Yu Su, Yilin Shen, Zhiyu Chen, Xifeng Yan, William Yang Wang

Information Extraction & Text Mining

Challenge:	Using a pre-defined vocabulary is a common approach to selecting text inputs . however, using a large vocabulary is not economical, as it limits the model's applicability on computation-or memoryconstrained scenarios.
Approach:	They propose a more sophisticated variational vocabulary dropout to perform vocabulary selection . they propose two new metrics to measure area under accuracy-vocab curve and Vocab Size under X% accuracy drop .
Outcome:	The proposed framework outperforms the baselines on the vocabulary selection problem on multiple NLP classification tasks.

Global Textual Relation Embedding for Relational Understanding (P19-1)

Copied to clipboard

Zhiyu Chen, Hanwen Zha, Honglei Liu, Wenhu Chen, Xifeng Yan, Yu Su

Representation Learning

Challenge:	Existing methods to learn textual relation embeddings are lacking in large open-domain corpora.
Approach:	They propose to learn a general-purpose embedding of textual relations using a large dataset from Freebase.
Outcome:	The proposed embedding can facilitate downstream tasks requiring relational understanding of the text.

BAPO: Boundary-Aware Policy Optimization for Reliable Agentic Search (2026.findings-acl)

Copied to clipboard

Shiyu Liu, Yongjing Yin, Jianhao Yan, Yunbo Tang, Qinggang Zhang, Bei Li, Xin Chen, Jingang Wang, Xunliang Cai, Jinsong Su

Living in the Moment: Can Large Language Models Grasp Co-Temporal Reasoning? (2024.acl-long)

Challenge:	Existing RL-based agentic search models fail to recognize reasoning boundaries and rarely admit "I DON'T KNOW" lack of reliability leads to plausible but unreliable answers, introducing significant risks .
Approach:	They propose a framework to cultivate reliable boundary awareness without compromising accuracy.
Outcome:	Experiments show that the proposed framework improves the reliability of agentic search models.

Copied to clipboard

Zhaochen Su, Juntao Li, Jun Zhang, Tong Zhu, Xiaoye Qu, Pan Zhou, Yan Bowen, Yu Cheng, Min Zhang

AscendKernelGen: LLM-Driven Kernel Generation for NPUs (2026.findings-acl)

Challenge:	Current temporal reasoning datasets are limited to questions about single or isolated events, falling short in mirroring the realistic temporal characteristics involving concurrent nature and intricate temporal interconnections.
Approach:	They propose a co-temporal Question Answering benchmark that contains four co-time scenarios with 4,748 samples for evaluating the co-timing abilities of large language models.
Outcome:	The proposed benchmarks show that current LLMs struggle on CoTempQA tasks even when enhanced with Chain of Thought methodologies.

Copied to clipboard

Xinzi Cao, Jianyang Zhai, Pengfei Li, Zhiheng Hu, Cen Yan, null Mubingxu, Guanghuan Fang, Bin She, Jiayu Li, Yihan Su, Dongyang Tao, Feidiao Yang, Chang-Dong Wang, Yutong Lu, Weicheng Xue, Bin Zhou, Yonghong Tian

Challenge:	Neural Processing Units (NPUs) are critical for AI infrastructure, but their development remains a bottleneck due to vendor-specific Domain-Specific Languages (DSLs).
Approach:	They propose a framework for NPU kernel development that bridges the gap in hardware-specific coding . compiler success on complex Level-2 kernels improves from 0% to 95.5%, they say .
Outcome:	The proposed framework bridges the gap in hardware-specific coding, showing a near-zero success rate on complex kernels.

Context-aware Difference Distilling for Multi-change Captioning (2024.acl-long)

Copied to clipboard

Yunbin Tu, Liang Li, Li Su, Zheng-Jun Zha, Chenggang Yan, Qingming Huang

Language Models

Challenge:	Existing methods for multi-change captioning are difficult because it requires a higher level of cognition to reason an arbitrary number of changes.
Approach:	They propose a context-aware difference distilling network to capture all genuine changes for yielding sentences.
Outcome:	The proposed network captures all genuine changes for yielding sentences on three public datasets.

Uncovering and Categorizing Social Biases in Text-to-SQL (2023.acl-long)

Copied to clipboard

Yan Liu, Yan Gao, Zhe Su, Xiaokang Chen, Elliott Ash, Jian-Guang Lou

Programming Languages in NLP

Challenge:	Existing Text-to-SQL models are trained on clean, neutral datasets, such as Spider and WikiSQl, but these models contain social bias at different rates.
Approach:	They propose to use data to map natural language utterances to SQL queries.
Outcome:	The proposed model can contain social bias at different rates in the downstream Text-to-SQL task.

Vision-Language Introspection: Mitigating Overconfident Hallucinations in MLLMs via Interpretable Bi-Causal Steering (2026.acl-long)

Copied to clipboard

Shuliang Liu, Songbo Yang, Dong Fang, Sihang Jia, Yuqi Tang, Lingfeng Su, Ruoshui Peng, Yibo Yan, Xin Zou, Xuming Hu

Responsible & Trustworthy NLP

Challenge:	Existing approaches to overcome object hallucination are limited . Existing mitigations include costly retraining and a training-free inference framework .
Approach:	They propose a training-free inference framework that simulates a metacognitive self-correction process.
Outcome:	The proposed framework reduces object hallucination rates by 12.67% on MMHal-Bench and improves accuracy by 5.8% on POPE.

XL-NBT: A Cross-lingual Neural Belief Tracking Framework (D18-1)

Copied to clipboard

Wenhu Chen, Jianshu Chen, Yu Su, Xin Wang, Dong Yu, Xifeng Yan, William Yang Wang

Natural Language Interfaces

Challenge:	a multi-lingual approach to training dialog systems is expensive and tedious, but it can be useful for cross-lingual support.
Approach:	They propose to annotate data for multiple languages and train a multi-lingual dialog system for each language.
Outcome:	The proposed framework bypasses the expensive human annotation and achieves promising results.

Efficient k-Nearest-Neighbor Machine Translation with Dynamic Retrieval (2024.findings-acl)

Copied to clipboard

Yan Gao, Zhiwei Cao, Zhongjian Miao, Baosong Yang, Shiyu Liu, Min Zhang, Jinsong Su

Information Retrieval

Challenge:	Existing models for non-parametric domain adaptation lack kNN retrieval at each timestep, leading to substantial time overhead.
Approach:	They propose a kNN-MT-based model that uses a domain-specific translation knowledge store to interpolate the prediction distribution of the model.
Outcome:	The proposed model significantly extends kNN-MT with dynamic retrieval on widely-used datasets.

Rotation Control Unlearning: Quantifying and Controlling Continuous Unlearning for LLM with The Cognitive Rotation Space (2026.findings-acl)

Copied to clipboard

Xiang Zhang, Kun Wei, Xu Yang, Jiahua Li, Su Yan, Cheng Deng

Challenge:	Existing methods to remove undesirable data from Large Language Models suffer from cumulative catastrophic utility loss under continuous unlearning requests.
Approach:	They propose a method that leverages the rotational salience weight of RCU to quantify and control the unlearning degree in the continuous unlearning process.
Outcome:	The proposed method achieves SOTA performance without a retained dataset.

CEDAR: A Chinese Evaluation Dataset for Computational Argumentation (2026.acl-long)

Copied to clipboard

Tian Lan, Jiang Li, Rong Yan, Feilong Bao, Weihua Wang, Guanglai Gao, Xiangdong Su

Learning to Edit Knowledge via Instruction-based Chain-of-Thought Prompting (2026.acl-long)

Challenge:	Existing debate datasets neglect important labels for argument mining, generation, and evaluation.
Approach:	They propose a Chinese Evaluation Dataset for Computational Argumentation that includes key arguments and key rhetorical figures, debater roles, modal words, debate results and transcripts.
Outcome:	The proposed dataset covers 600 debates about 318 topics from Chinese debate competitions.

Copied to clipboard

Jinhu Fu, Yan Bai, Longzhu He, Yihang Lou, Yanxiao Zhao, Li Sun, Sen Su

A Survey of Mathematical Reasoning in the Era of Multimodal Large Language Model: Benchmark, Method & Challenges (2025.findings-acl)

Challenge:	Existing knowledge editing methods focus on structured fact triples, overlooking diverse unstructured forms of factual information.
Approach:	They propose a method that allows LLMs to edit knowledge via Chain of Thoughts reasoning.
Outcome:	The proposed method achieves strong generalization across six diverse knowledge editing scenarios with a single round of training on three open-source language models.

Copied to clipboard

Yibo Yan, Jiamin Su, Jianxiang He, Fangteng Fu, Xu Zheng, Yuanhuiyi Lyu, Kun Wang, Shen Wang, Qingsong Wen, Xuming Hu

Global Relation Embedding for Relation Extraction (N18-1)

Challenge:	This survey provides the first comprehensive analysis of mathematical reasoning in the era of multimodal large language models . integrating large language model with mathematical reasoning tasks is becoming significant as AI advances .
Approach:	They review over 200 studies published since 2021 and examine the state-of-the-art developments in Math-LLMs . they identify five major challenges hindering the realization of AGI in this domain .
Outcome:	The authors examine the state-of-the-art developments in Math-LLMs with a focus on multimodal settings.

Copied to clipboard

Yu Su, Honglei Liu, Semih Yavuz, Izzeddin Gür, Huan Sun, Xifeng Yan

Information Extraction & Text Mining

Challenge:	Existing methods to extract textual relations with distant supervision are limited by their reliance on supervised training data.
Approach:	They propose to embed relations with global statistics of relations to combat the wrong labeling problem of distant supervision.
Outcome:	The proposed method is more robust to training noise introduced by distant supervision and improves relation extraction models.

Taming Pre-trained Language Models with N-gram Representations for Low-Resource Domain Adaptation (2021.acl-long)

Copied to clipboard

Shizhe Diao, Ruijia Xu, Hongjin Su, Yilei Jiang, Yan Song, Tong Zhang

Dialogue Response Selection with Hierarchical Curriculum Learning (2021.acl-long)

Challenge:	Existing methods to train pre-trained models require domain-specific data and computational resources.
Approach:	They propose a domain-aware N-gram Adaptor to incorporate unseen and domain-specific words into a generic pretrained model.
Outcome:	The proposed model can improve on eight low-resource tasks using limited data with lower computational costs.

Copied to clipboard

Yixuan Su, Deng Cai, Qingyu Zhou, Zibo Lin, Simon Baker, Yunbo Cao, Shuming Shi, Nigel Collier, Yan Wang

Dialogue Systems & Conversational Agents

Challenge:	Empirical studies on three benchmark datasets with three state-of-the-art matching models demonstrate that the proposed learning framework significantly improves the model performance across various evaluation metrics.
Approach:	They propose a hierarchical curriculum learning framework that trains matching models in an “easy-to-difficult” scheme.
Outcome:	The proposed framework significantly improves the model performance across evaluation metrics on three benchmark datasets with three state-of-the-art matching models.

TrendFact: A Benchmark Towards Hotspot Perception in Automatic Fact-Checking (2026.acl-long)

Copied to clipboard

Xiaocheng Zhang, Xi Wang, Yifei Lu, Jianing Wang, Zhuangzhuang Ye, Mengjiao Bao, Peng Yan, Xiaohong Su