Papers by Zhou Su

Alignment-Enhanced Decoding: Defending Jailbreaks via Token-Level Adaptive Refining of Probability Distributions (2024.emnlp-main)

Challenge:	despite growing interest in NL2GQL, benchmarking progress has been constrained by the lack of resources that are simultaneously large-scale, cross-domain, and cross-dialect.
Approach:	They propose a framework that integrates NL2SQL-to-NL2GQL conversion with graph-native data generation.
Outcome:	The proposed framework supports execution-based evaluation on Cypher and ISO-GQL, covering hundreds of graph databases and over 20k natural language questions for each dialect.

Copied to clipboard

Quan Liu, Zhenhong Zhou, Longzhu He, Yi Liu, Wei Zhang, Sen Su

Confidence Based Bidirectional Global Context Aware Training Framework for Neural Machine Translation (2022.acl-long)

Challenge:	Existing defenses against jailbreaks focus on perturbing or inspecting inputs, but ignore competing objectives, the underlying cause of alignment failures.
Approach:	They propose a novel defense that employs adaptive decoding to address the root causes of jailbreak issues.
Outcome:	The proposed defense improves safety alignment while maintaining helpfulness.

Copied to clipboard

Chulun Zhou, Fandong Meng, Jie Zhou, Min Zhang, Hongji Wang, Jinsong Su

Text Generation

Challenge:	Existing studies focus on how to effectively exploit bidirectional global contexts in neural machine translation models.
Approach:	They propose a Confidence Based Bidirectional Global Context Aware training framework for NMT . they incorporate bidirectional global context to the NMT model on unconfidently-predicted target words .
Outcome:	The proposed framework improves the NMT model on three large-scale translation datasets by +1.02, +0.57 BLEU scores.

Towards Making the Most of Dialogue Characteristics for Neural Chat Translation (2021.emnlp-main)

Copied to clipboard

Yunlong Liang, Chulun Zhou, Fandong Meng, Jinan Xu, Yufeng Chen, Jinsong Su, Jie Zhou

Efficient Continue Training of Temporal Language Model with Structural Information (2023.findings-emnlp)

Challenge:	Neural Chat Translation (NCT) models that use dialogue characteristics of chat are often incoherent and speakerirrelevant.
Approach:	They propose to introduce the modeling of dialogue characteristics into the NCT model by capturing the inherent dialogue characteristics.
Outcome:	The proposed model can translate conversational text between speakers of different languages.

Copied to clipboard

Zhaochen Su, Juntao Li, Zikang Zhang, Zihan Zhou, Min Zhang

Cross-Lingual Cross-Platform Rumor Verification Pivoting on Multimedia Content (D18-1)

Challenge:	Existing temporal language models are limited by the superficial temporal information brought by timestamps, which fails to learn the inherent changes of linguistic components.
Approach:	They propose a method that captures syntactically changed tokens and captures the relationship between the time prefix and tokens.
Outcome:	The proposed method outperforms existing temporal language models on two datasets and three tasks.

Copied to clipboard

Weiming Wen, Songwen Su, Zhou Yu

Towards Robust Neural Machine Translation with Iterative Scheduled Data-Switch Training (2022.coling-1)

Challenge:	rumors with multimedia content are becoming more and more common on social networks . a new feature set is proposed to verify rumors pivoting on multimedia content .
Approach:	They propose to use multimedia content to find external information on social media platforms . they propose to leverage semantic similarity between rumors and external information .
Outcome:	The proposed approach achieves state-of-the-art results on social networks . it leverages semantic similarity between rumors and external information .

Copied to clipboard

Zhongjian Miao, Xiang Li, Liyan Kang, Wen Zhang, Chulun Zhou, Yidong Chen, Bin Wang, Min Zhang, Jinsong Su

Robustness in NLP

Challenge:	Existing methods on robust neural machine translation (NMT) construct adversarial examples by injecting noise into authentic examples and indiscriminately exploit two types of examples.
Approach:	They propose an iterative scheduled data-switch training framework to mitigate this problem by injecting noise into authentic examples and indiscriminately exploiting two types of examples.
Outcome:	The proposed model outperforms several competitive benchmarks on four translation benchmarks.

Is this the real life? Is this just fantasy? The Misleading Success of Simulating Social Interactions With LLMs (2024.emnlp-main)

Copied to clipboard

Xuhui Zhou, Zhe Su, Tiwalayo Eisape, Hyunwoo Kim, Maarten Sap

Challenge:	Recent advances in large language models have enabled richer social simulations . however, the role of information asymmetry in these simulations has been overlooked .
Approach:	They develop an evaluation framework to simulate social interactions with LLMs in different settings.
Outcome:	The proposed framework performs better in unrealistic, omniscient simulation settings but struggles in those with information asymmetry.

Knowledge Inheritance for Pre-trained Language Models (2022.naacl-main)

Copied to clipboard

Yujia Qin, Yankai Lin, Jing Yi, Jiajie Zhang, Xu Han, Zhengyan Zhang, Yusheng Su, Zhiyuan Liu, Peng Li, Maosong Sun, Jie Zhou

A Variational Hierarchical Model for Neural Cross-Lingual Summarization (2022.acl-long)

Challenge:	Existing large-scale pre-trained language models are mainly trained from scratch individually, ignoring that many well-taught PLMs are available.
Approach:	They propose a pre-training framework called knowledge inheritance and propose auxiliary supervision to efficiently learn larger PLMs.
Outcome:	The proposed framework can be used to train large-scale language models with huge parameters and a large dataset can be adapted to domain adaptation and knowledge transfer.

Copied to clipboard

Yunlong Liang, Fandong Meng, Chulun Zhou, Jinan Xu, Yufeng Chen, Jinsong Su, Jie Zhou

Text Generation

Challenge:	Existing studies on cross-lingual summarization focus on pipeline methods or jointly training an end-to-end model through an auxiliary MT or MS objective.
Approach:	They propose a hierarchical model for the cross-lingual summarization task . the model is based on the conditional variational auto-encoder .
Outcome:	The proposed model generates better cross-lingual summaries than comparison models in the few-shot setting.

MT3: A Synergistic Multi-Task RL Framework for Specializing MLLMs in Text Image Machine Translation (2026.acl-long)

Copied to clipboard

Zhaopeng Feng, Yupu Liang, Shaosheng Cao, Jiayuan Su, Jiahan Ren, Zhijie Zhou, Wenxuan Huang, Jian Wu, Zuozhu Liu

An Evaluation Resource for Grounding Translation Errors (2025.findings-emnlp)

Challenge:	Text Image Machine Translation (TIMT) is a critical subfield of machine translation . it requires accurate optical character recognition, robust visual-text reasoning, and high-quality translation a challenge .
Approach:	They propose a multi-task optimization framework to specialize MLLMs into expert TIMT models.
Outcome:	The proposed model outperforms baselines on the latest in-domain MIT-10M benchmark.

Copied to clipboard

Sujin Chen, Kang Wang, Zixuan Zhou, Xiangyu Duan, Wanqun Zhang, Hao Yang, Jinsong Su, Min Zhang

Explainability & Interpretability in NLP

Challenge:	Current fine-grained error analyses do not ground the errors to the reasons why the annotated text spans are erroneous.
Approach:	They use a bi-directional grounding scheme to ground erroneous text in two directions . if the error spans of both directions are consistent, the explanation is valid .
Outcome:	The proposed grounding process improves translation error detection significantly.

Reinforcement Learning with Semantic Rewards Enables Low-Resource Language Expansion without Alignment Tax (2026.findings-acl)

Copied to clipboard

Zeli Su, Ziyin Zhang, Zhou Liu, Xuexian Song, Zhankai Xu, Longfei Zheng, Xiaolu Zhang, Rong Fu, Guixian Xu, Wentao Zhang

Exploiting Emotion-Semantic Correlations for Empathetic Response Generation (2023.findings-emnlp)

Challenge:	Extending large language models to low-resource languages often incurs an "alignment tax" token-level fine-tuning enforces token-level surface imitation on narrow and biased data distributions.
Approach:	They propose a semantic-space alignment paradigm powered by group-level semantic rewards instead of likelihood maximization.
Outcome:	The proposed model acquires low-resource capa- bilities while mitigating alignment tax on Tibetan–Chinese machine translation and Ti- betan headline generation.

Copied to clipboard

Zhou Yang, Zhaochun Ren, Wang Yufeng, Xiaofei Zhu, Zhihao Chen, Tiecheng Cai, Wu Yunbing, Yisong Su, Sibo Ju, Xiangwen Liao

Type-Driven Multi-Turn Corrections for Grammatical Error Correction (2022.findings-acl)

Challenge:	Empathetic response generation aims to generate empathetic responses by understanding the speaker’s emotional feelings from the language of dialogue.
Approach:	They propose a dynamical Emotion-Semantic Correlation Model (ESCM) which constructs dynamic emotion-semantics through the interaction of context and emotions.
Outcome:	The proposed model understands emotions more accurately and expresses fluent and informative empathetic responses.

Copied to clipboard

Shaopeng Lai, Qingyu Zhou, Jiali Zeng, Zhongli Li, Chao Li, Yunbo Cao, Jinsong Su

Syntactic Text Processing

Challenge:	Existing studies focus on data augmentation to combat exposure bias . but data augmented models lack the ability to recognize the procedure of gradual corrections .
Approach:	They propose a type-driven multi-turn corrections approach that uses multiple training instances to train dominant models.
Outcome:	The proposed model achieves state-of-the-art single-model performance on English GEC benchmarks.

Identifying Factual Inconsistencies in Summaries: Grounding LLM Inference via Task Taxonomy (2024.findings-emnlp)

Copied to clipboard

Liyan Xu, Zhenlin Su, Mo Yu, Jin Xu, Jinho Choi, Jie Zhou, Fei Liu

Improving Multi-turn Dialogue Modelling with Utterance ReWriter (P19-1)

Challenge:	Existing studies have focused on specialized BERT-variants and recent LLMs to reason inconsistencies.
Approach:	They propose to incorporate task-specific taxonomy into inferences to facilitate both zero-shot and supervised paradigms.
Outcome:	The proposed model outperforms specialized non-LLM and recent LLM models in a number of domains.

Copied to clipboard

Hui Su, Xiaoyu Shen, Rongzhi Zhang, Fei Sun, Pengwei Hu, Cheng Niu, Jie Zhou

Paraphrasing

Challenge:	Recent research has achieved impressive results in single-turn dialogue modelling, but multi-turn models still remain challenging.
Approach:	They propose to rewrite human utterances as a pre-process to help multi-turn dialgoue modelling.
Outcome:	The proposed architecture achieves remarkably good performance on the utterance rewriting task.

Towards General Agentic Intelligence via Environment Scaling (2026.findings-acl)

Copied to clipboard

Runnan Fang, Shihao Cai, Baixuan Li, Jialong Wu, Guangyu Li, Wenbiao Yin, Xinyu Wang, Xiaobin Wang, Liangcai Su, Zhen Zhang, Shibin Wu, Zhengwei Tao, Yong Jiang, Pengjun Xie, Ningyu Zhang, Fei Huang, Wentao Zhang, Jingren Zhou

Is He Extroverted? Identifying Missing Relevant Personas for Faithful User Simulation (2026.eacl-srw)

Challenge:	Diverse real-world APIs require precise, robust function-calling intelligence, which needs agents to develop these capabilities through interaction in varied environments.
Approach:	They propose a framework that scales up environments to enable agentic intelligence . they use a two-phase agent fine-tuning strategy to first endow agents with basic agentic capabilities, then specializing them for domain-specific contexts.
Outcome:	Experiments on -bench, -Bench, and ACEBench show that the model significantly enhances the models’ function-calling capability.

Copied to clipboard

Weiwen Su, Yuhan Zhou, Zihan Wang, Naoki Yoshinaga, Masashi Toyoda

Walking in Others’ Shoes: How Perspective-Taking Guides Large Language Models in Reducing Toxicity and Bias (2024.emnlp-main)

Challenge:	Existing user simulation approaches focus on generating user-like responses in dialogue without verifying whether critical personas are supplied.
Approach:	They propose a task of identifying persona dimensions that are relevant but missing in simulating a user's reply for a given dialogue context.
Outcome:	The proposed model identifies persona dimensions that are relevant but missing in simulating a user’s response for a given dialogue context.

Copied to clipboard

Rongwu Xu, Zian Zhou, Tianwei Zhang, Zehan Qi, Su Yao, Ke Xu, Wei Xu, Han Qiu

AgentCPM-GUI: Building Mobile-Use Agents with Reinforcement Fine-Tuning (2025.emnlp-demos)

Challenge:	Existing prompting methods that require white-box access to the model or substantial training fail to simultaneously lessen toxicity and bias.
Approach:	They propose a strategy that encourages LLMs to integrate diverse human perspectives and self-regulate their responses by incorporating diverse human viewpoints.
Outcome:	The proposed approach can significantly diminish toxicity (up to 89%) and bias (up 73%) in LLMs’ responses.

Copied to clipboard

Zhong Zhang, Yaxi Lu, Yikun Fu, Yupeng Huo, Shenzhi Yang, Yesai Wu, Han Si, Xin Cong, Haotian Chen, Yankai Lin, Xie Xie, Wei Zhou, Wang Xu, Zhou Su, Zhongwu Zhai, Xiaoming Liu, null Meiyudong, Jianming Xu, Hongyan Tian, Chongyi Wang, Chi Chen, Yuan Yao, Zhiyuan Liu, Maosong Sun

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

Challenge:	Large language model agents have enabled GUI-based automation, but their deployment is limited by noisy data, poor generalization, and lack of support for non-English GUIs.
Approach:	They propose an 8B-parameter GUI agent built for robust and efficient on-device GUI interaction.
Outcome:	The proposed GUI agent achieves promising performance on five public benchmarks and proposed Chinese benchmark CAGUI.

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%.

LLaVE: Large Language and Vision Embedding Models with Hardness-Weighted Contrastive Learning (2025.findings-emnlp)

Copied to clipboard

Zhibin Lan, Liqiang Niu, Fandong Meng, Jie Zhou, Jinsong Su

AVG-LLaVA: An Efficient Large Multimodal Model with Adaptive Visual Granularity (2025.findings-acl)

Challenge:	Existing LMM-based embedding models exhibit a high degree of overlap in similarity distribution between positive and negative pairs, making it challenging to distinguish hard negative pairs effectively.
Approach:	They propose a framework that improves the embedding model's representation learning for negative pairs based on their discriminative difficulty.
Outcome:	The proposed framework improves the embedding model's representation learning for negative pairs based on their discriminative difficulty.

Copied to clipboard

Zhibin Lan, Liqiang Niu, Fandong Meng, Wenbo Li, Jie Zhou, Jinsong Su

RMTBench: Benchmarking LLMs Through Multi-Turn User-Centric Role-Playing (2025.findings-emnlp)

Challenge:	Existing large multimodal models typically divide high-resolution images into multiple local images and a global image, leading to a large number of visual tokens.
Approach:	They propose an LMM that can adaptively select the appropriate visual granularity based on the input image and instruction.
Outcome:	The proposed model significantly reduces visual tokens and speeds up inference on 11 benchmarks.

Copied to clipboard

Hao Xiang, Tianyi Tang, Yang Su, Bowen Yu, An Yang, Fei Huang, Yichang Zhang, Yaojie Lu, Hongyu Lin, Xianpei Han, Jingren Zhou, Junyang Lin, Le Sun

Accurate KV Cache Quantization with Outlier Tokens Tracing (2025.acl-long)

Challenge:	Existing benchmarks focus on character-centric approach and fail to reflect real-world applications.
Approach:	RMTBench is a user-centric bilingual role-playing benchmark featuring 80 diverse characters and over 8,000 dialogue rounds.
Outcome:	RMTBench features 80 diverse characters and over 8,000 dialogue rounds.

Copied to clipboard

Yi Su, Yuechi Zhou, Quantong Qiu, Juntao Li, Qingrong Xia, Ping Li, Xinyu Duan, Zhefeng Wang, Min Zhang

Multi-Level Cross-Modal Alignment for Speech Relation Extraction (2024.emnlp-main)

Challenge:	Large Language Models (LLMs) require substantial computational resources during deployment.
Approach:	They propose a method to identify outlier tokens and exclude them from quantization . they find that the method can deliver a 6.4 times reduction in memory usage and a 2.5 times increase in throughput .
Outcome:	The proposed method delivers a 6.4 times reduction in memory usage and a 2.5 times increase in throughput under 2-bit quantization.

Copied to clipboard

Liang Zhang, Zhen Yang, Biao Fu, Ziyao Lu, Liangying Shao, Shiyu Liu, Fandong Meng, Jie Zhou, Xiaoli Wang, Jinsong Su

Many Heads Are Better Than One: Improved Scientific Idea Generation by A LLM-Based Multi-Agent System (2025.acl-long)

Challenge:	Existing studies use synthetic speech to train and evaluate SpeechRE models, hindering their development . modality gap issue limits performance of existing models, limiting future researches .
Approach:	They propose to use speech data to train and evaluate SpeechRE models by using real speech . they propose to train a cross-modal alignment model to bridge the modality gap .
Outcome:	The proposed model can train to bridge the modality gap between speech encoder and text decoder . the proposed model is based on two real SpeechRE datasets .

Copied to clipboard

Haoyang Su, Renqi Chen, Shixiang Tang, Zhenfei Yin, Xinzhe Zheng, Jinzhe Li, Biqing Qi, Qi Wu, Hui Li, Wanli Ouyang, Philip Torr, Bowen Zhou, Nanqing Dong

Beyond Single-Value Metrics: Evaluating and Enhancing LLM Unlearning with Cognitive Diagnosis (2025.findings-acl)

Challenge:	Recent AI methods have shown promise in tasks such as hypothesis generation and experimental design, but they fail to replicate the collaborative nature of real-world scientific practices.
Approach:	They propose a virtual scientific system that mimics the collaborative nature of scientific research by organizing a team of agents to generate, evaluate, and refine research ideas.
Outcome:	The proposed system outperforms the state-of-the-art method in producing new scientific ideas and offers valuable insights to guide future research.

Copied to clipboard

Yicheng Lang, Kehan Guo, Yue Huang, Yujun Zhou, Haomin Zhuang, Tianyu Yang, Yao Su, Xiangliang Zhang

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

Challenge:	Large Language Models (LLMs) have been used to remove harmful knowledge and undesirable capabilities.
Approach:	They propose a framework that leverages Cognitive Diagnosis Modeling to evaluate LLM unlearning.
Outcome:	The proposed framework enhances evaluation and facilitates removal of harmful abilities.

Copied to clipboard

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

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

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

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

MovieChats: Chat like Humans in a Closed Domain (2020.emnlp-main)

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

Hui Su, Xiaoyu Shen, Zhou Xiao, Zheng Zhang, Ernie Chang, Cheng Zhang, Cheng Niu, Jie Zhou

PD3F: A Pluggable and Dynamic DoS-Defense Framework against resource consumption attacks targeting Large Language Models (2025.findings-emnlp)

Challenge:	Currently, open-domain chatbots are far from satisfactory.
Approach:	They propose a unified, readily scalable neural approach which reconciles all subtasks like intent prediction and knowledge retrieval.
Outcome:	The proposed approach outperforms commercial systems replying on complex rules on static and interactive tests and shows that the results are remarkably good.

Copied to clipboard

Yuanhe Zhang, Xinyue Wang, Haoran Gao, Zhenhong Zhou, Fanyu Meng, Yuyao Zhang, Sen Su

RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining (2022.acl-long)

Challenge:	Existing work lacks mitigation strategies against resource consumption attacks . existing work does not provide mitigation strategies for real-world LLM deployments .
Approach:	They propose a pluggable and dynamic doS-Defense framework which employs a two-stage approach to defend against resource consumption attacks from both the input and output sides.
Outcome:	The proposed framework significantly mitigates resource consumption attacks, improving users’ access capacity by up to 500% during adversarial load.

Copied to clipboard

Hui Su, Weiwei Shi, Xiaoyu Shen, Zhou Xiao, Tuo Ji, Jiarui Fang, Jie Zhou

Diversifying Dialogue Generation with Non-Conversational Text (2020.acl-main)

Challenge:	Large-scale pretrained language models have achieved SOTA results on NLP tasks but are vulnerable to adversarial attacks especially for logographic languages like Chinese.
Approach:	They propose a pretrained Chinese Bert that is robust to various forms of adversarial attacks like word perturbation, synonyms, typos, etc.
Outcome:	The proposed model outperforms baselines on 5 Chinese NLU tasks without sacrificing performance on clean testsets.

Copied to clipboard

Hui Su, Xiaoyu Shen, Sanqiang Zhao, Zhou Xiao, Pengwei Hu, Randy Zhong, Cheng Niu, Jie Zhou

Entity-Aware Abstractive Multi-Document Summarization (2021.findings-acl)

Challenge:	Neural network-based sequence-to-sequence models suffer from low diversity in open-domain dialogue generation.
Approach:	They propose a way to diversify dialogue generation by leveraging non-conversational text . they collect large-scale corpus from forum comments, idioms and book snippets .
Outcome:	The proposed model produces significantly more diverse responses without sacrificing relevance with context.

Copied to clipboard

Hao Zhou, Weidong Ren, Gongshen Liu, Bo Su, Wei Lu

Summarization

Challenge:	Existing models for multidocument summarization do not focus on explicitly modeling the underlying semantic information across documents.
Approach:	They propose an entityaware model for abstractive multi-document summarization that augments the classical Transformer-based encoder-decoder framework with a heterogeneous graph consisting of text units and entities as nodes.
Outcome:	The proposed model can deal with saliency and redundancy issues explicitly and can be used with pre-trained language models, arriving at improved performance.

Augmenting Multi-Agent Communication with State Delta Trajectory (2025.emnlp-main)

Copied to clipboard

Yichen Tang, Weihang Su, Yujia Zhou, Yiqun Liu, Min Zhang, Shaoping Ma, Qingyao Ai

TwinVoice: A Multi-dimensional Benchmark Towards Digital Twins via LLM Persona Simulation (2026.findings-acl)

Challenge:	Multi-agent systems based on large language models (LLMs) have shown to be effective in downstream tasks.
Approach:	They propose a protocol that transfers both natural language tokens and token-wise state transition trajectory from one agent to another.
Outcome:	The proposed protocol can transfer both natural language tokens and token-wise state transition trajectory from one agent to another.

Copied to clipboard

Bangde Du, Minghao Guo, Songming He, Ziyi Ye, Xi Zhu, Weihang Su, Shuqi Zhu, Yujia Zhou, Yongfeng Zhang, Qingyao Ai, Yiqun Liu

Dual Context-Guided Continuous Prompt Tuning for Few-Shot Learning (2022.findings-acl)

Challenge:	Existing studies show that advanced LLMs produce text indistinguishable from human writing.
Approach:	They propose a benchmark to assess persona simulation across diverse contexts by decomposing the evaluation into six fundamental capabilities including opinion consistency, memory recall, logical reasoning, persona tone, and syntactic style.
Outcome:	The proposed model achieves moderate accuracy but falls short of the basic capabilities needed to simulate personas in real-world contexts.

Copied to clipboard

Jie Zhou, Le Tian, Houjin Yu, Zhou Xiao, Hui Su, Jie Zhou

On Transferability of Prompt Tuning for Natural Language Processing (2022.naacl-main)

Challenge:	Existing prompt-based paradigms have shown their competitive performance in many NLP tasks, but their effectiveness varies upon the model and training data.
Approach:	They propose a dual context-guided continuous prompt tuning method that integrates contextual information into the input input.
Outcome:	The proposed method outperforms existing prompt tuning methods in the few-shot setting and can be used in many NLP tasks.

Copied to clipboard

Yusheng Su, Xiaozhi Wang, Yujia Qin, Chi-Min Chan, Yankai Lin, Huadong Wang, Kaiyue Wen, Zhiyuan Liu, Peng Li, Juanzi Li, Lei Hou, Maosong Sun, Jie Zhou

Advancing SMoE for Continuous Domain Adaptation of MLLMs: Adaptive Router and Domain-Specific Loss (2025.acl-long)

Challenge:	Pre-trained language models (PLMs) can achieve comparable performance to full-parameter fine-tuning by tuning a few soft prompts, but require much more training time than fine-timing.
Approach:	They empirically investigate the transferability of soft prompts across different downstream tasks and PLMs to determine what decides prompt transferability.
Outcome:	The proposed method can achieve comparable performance to full-parameter fine-tuning by tuning a few soft prompts, but requires much more training time than fine-timing.

Copied to clipboard

Liang Zhang, Ziyao Lu, Fandong Meng, Hui Li, Jie Zhou, Jinsong Su

A Novel Graph-based Multi-modal Fusion Encoder for Neural Machine Translation (2020.acl-main)

Challenge:	Recent studies have explored Continual Instruction Tuning (CIT) in Multimodal Large Language Models (MLLMs), with a primary focus on Task-incremental CIT, where MLLM are required to continuously acquire new tasks.
Approach:	They propose a Sparse Mixture of Expert (SMoE) based method for domain-incremental CIT in Multimodal Large Language Models (MLLMs) . they equip the SMoA module with a domain-specific autoregressive loss (DSAL) they establish a new benchmark to evaluate the efficacy of their method .
Outcome:	The proposed method outperforms all baselines and is based on a Sparse Mixture of Experts (SMoE) module .

Copied to clipboard

Yongjing Yin, Fandong Meng, Jinsong Su, Chulun Zhou, Zhengyuan Yang, Jie Zhou, Jiebo Luo

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

Challenge:	Existing multi-modal neural machine translation models do not fully exploit fine-grained semantic correspondences between semantic units of different modalities.
Approach:	They propose a graph-based multi-modal fusion encoder that exploits fine-grained semantic correspondences between different modalities.
Outcome:	The proposed encoder significantly extends the conventional text-based translation by taking images as additional inputs.

Copied to clipboard

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

AI-LieDar : Examine the Trade-off Between Utility and Truthfulness in LLM Agents (2025.naacl-long)

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

Zhe Su, Xuhui Zhou, Sanketh Rangreji, Anubha Kabra, Julia Mendelsohn, Faeze Brahman, Maarten Sap

Challenge:	LieDar is a framework to study how LLM-based agents navigate these scenarios in a multi-turn interactive setting.
Approach:	They propose a framework to study how LLM-based agents navigate these scenarios in an interactive multi-turn setting.
Outcome:	The proposed framework shows that all models are truthful less than 50% of the time, although truthfulness and goal achievement rates vary across models.

Getting the Most out of Simile Recognition (2022.findings-emnlp)

Copied to clipboard

Xiaoyue Wang, Linfeng Song, Xin Liu, Chulun Zhou, Hualin Zeng, Jinsong Su

Challenge:	Recent work ignores features other than surface strings and suffers from data hunger issue.
Approach:	They propose to use simile sentence classification and simile component extraction to find simile components.
Outcome:	The proposed model outperforms current state-of-the-art systems and baselines.

InstructDiff: Domain-Adaptive Data Selection via Contrastive Entropy for Efficient LLM Fine-Tuning (2026.acl-long)

Copied to clipboard

Junyou Su, He Zhu, Xiao Luo, Liyu Zhang, Hong-Yu Zhou, Yun Chen, Peng Li, Yang Liu, Guanhua Chen

PRMBench: A Fine-grained and Challenging Benchmark for Process-Level Reward Models (2025.acl-long)

Challenge:	Existing data selection methods suffer from severe domain specificity . existing methods for general instruction-following fail on reasoning tasks .
Approach:	They propose a framework that operationalizes contrastive entropy as a domain-adaptive selection criterion through warmup calibration, bi-directional NLL filtering, and entropic-based ranking.
Outcome:	Experiments show that InstructDiff outperforms baseline training on reasoning tasks while using only 10% of the data.

Copied to clipboard

Mingyang Song, Zhaochen Su, Xiaoye Qu, Jiawei Zhou, Yu Cheng

Unsupervised Real-Time Hallucination Detection based on the Internal States of Large Language Models (2024.findings-acl)

Challenge:	Recent large language models (LLMs) have achieved significant performance in complex reasoning tasks such as mathematics and code generation.
Approach:	They propose a process-level benchmark specifically designed to assess the fine-grained error detection capabilities of PRMs.
Outcome:	The proposed model measures the accuracy, soundness, and sensitivity of 25 models across open-source and closed-source large language models.

Copied to clipboard

Weihang Su, Changyue Wang, Qingyao Ai, Yiran Hu, Zhijing Wu, Yujia Zhou, Yiqun Liu

CE-GPPO: Coordinating Entropy via Gradient-Preserving Clipping Policy Optimization in Reinforcement Learning (2026.acl-long)

Challenge:	Existing studies on hallucination detection for LLMs focus on how to identify possible factrelated errors in outputs.
Approach:	They propose an unsupervised training framework that leverages the internal states of LLMs for real-time hallucination detection without requiring manual annotations.
Outcome:	The proposed framework outperforms existing state-of-the-art methods in hallucination detection.

Copied to clipboard

Zhenpeng Su, Leiyu Pan, Minxuan Lv, Yuntao Li, Wenping Hu, Fuzheng Zhang, Kun Gai, Guorui Zhou

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

Challenge:	Existing methods for proximal policy optimization discard valuable gradient signals from low-probability tokens due to the clipping mechanism.
Approach:	They propose an algorithm that reintroduces gradients from clipped tokens in native PPO in a gentle and bounded manner.
Outcome:	The proposed algorithm outperforms strong baselines on reasoning benchmarks on different model scales.

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.

HyperNetwork-based Decoupling to Improve Model Generalization for Few-Shot Relation Extraction (2023.emnlp-main)

Copied to clipboard

Liang Zhang, Chulun Zhou, Fandong Meng, Jinsong Su, Yidong Chen, Jie Zhou

Information Extraction & Text Mining

Challenge:	Existing studies cannot generalize well to unseen relations using Prototypical Networks . current approaches are dependent on large amount of labeled data and cannot deal with unseense relations well.
Approach:	They propose a HyperNetwork-based Decoupling approach to improve FSRE generalization . they propose FSre models with an encoder, network generator and refined classifiers .
Outcome:	The proposed method improves the generalization of few-shot relation extraction models.

Exploring Layer-wise Information Effectiveness for Post-Training Quantization in Small Language Models (2026.findings-acl)

Copied to clipboard

He Xiao, Qingyao Yang, Dirui Xie, Wendong XU, Zunhai Su, Runming Yang, Haobo Liu, Wenyong Zhou, Zhengwu Liu, Ngai Wong

DSMoE: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense LLMs (2025.emnlp-main)

Challenge:	Large language models with billions of parameters are often over-provisioned . smaller models exhibit lower robustness under extreme low-bit quantization .
Approach:	They propose a hardware-native, metric-driven post-training quantization framework that keeps uniform bit-width within each layer while mixing precision across layers.
Outcome:	LieQ reduces large accuracy gap observed for large language models with billions of parameters while preserving standard multiplication kernels.

Copied to clipboard

Minxuan Lv, Zhenpeng Su, Leiyu Pan, Yizhe Xiong, Zijia Lin, Hui Chen, Wei Zhou, Jungong Han, Guiguang Ding, Wenwu Ou, Di Zhang, Kun Gai, Songlin Hu

Exploring Dynamic Selection of Branch Expansion Orders for Code Generation (2021.acl-long)

Challenge:	Existing sparsification methods like pruning can lose model knowledge through parameter removal.
Approach:	They propose a novel approach that achieves sparsification by partitioning pre-trained FFN layers into computational blocks.
Outcome:	The proposed approach achieves superior performance across language modeling and downstream tasks under equivalent computational constraints.

Copied to clipboard

Hui Jiang, Chulun Zhou, Fandong Meng, Biao Zhang, Jie Zhou, Degen Huang, Qingqiang Wu, Jinsong Su

Programming Languages in NLP

Challenge:	Existing code generation models model abstract syntax tree (AST) but not suitable for all multi-branch nodes.
Approach:	They propose to equip a Seq2Tree model with a branch selector to determine optimal expansion orders for multi-branch nodes.
Outcome:	The proposed model can determine optimal expansion orders of branches for multi-branch nodes.

Translatotron-V(ison): An End-to-End Model for In-Image Machine Translation (2024.findings-acl)

Copied to clipboard

Zhibin Lan, Liqiang Niu, Fandong Meng, Jie Zhou, Min Zhang, Jinsong Su

MiLe Loss: a New Loss for Mitigating the Bias of Learning Difficulties in Generative Language Models (2024.findings-naacl)

Challenge:	In-image machine translation (IIMT) aims to translate an image containing texts in source language into an image with translations in target language.
Approach:	They propose an end-to-end IIMT model with four modules that translate images . they propose a two-stage training framework to assist the model in learning alignment across languages .
Outcome:	The proposed model outperforms cascaded models with only 70.9% of parameters and is highly accurate.

Copied to clipboard

Zhenpeng Su, Zijia Lin, Baixue Baixue, Hui Chen, Songlin Hu, Wei Zhou, Guiguang Ding, Xing W

SURf: Teaching Large Vision-Language Models to Selectively Utilize Retrieved Information (2024.emnlp-main)

Challenge:	Existing generative language models neglect an inherent challenge in text corpus during training, i.e., the imbalance between frequent tokens and infrequent ones.
Approach:	They propose a function to mitigate the imbalance between frequent and infrequent tokens . authors propose 'MiLe Loss' function to assess learning difficulty of tokens during training .
Outcome:	Experiments show that models with proposed model can improve on downstream benchmarks.

Copied to clipboard

Jiashuo Sun, Jihai Zhang, Yucheng Zhou, Zhaochen Su, Xiaoye Qu, Yu Cheng

Dense Retrievers Can Fail on Simple Queries: Revealing The Granularity Dilemma of Embeddings (2025.findings-emnlp)

Challenge:	Existing studies focus on the text modality or are limited to specific tasks.
Approach:	They propose a framework to teach Large Vision-Language Models to selectively utilize retrieved information and improve their robustness against irrelevant or misleading references.
Outcome:	The proposed framework improves LVLMs’ ability to utilize retrieved multimodal references and their robustness against irrelevant or misleading information.

Copied to clipboard

Liyan Xu, Zhenlin Su, Mo Yu, Jiangnan Li, Fandong Meng, Jie Zhou

Dial-MAE: ConTextual Masked Auto-Encoder for Retrieval-based Dialogue Systems (2024.naacl-long)

Challenge:	a limited number of text encoders are able to recognize fine-grained entities or events within encoded semantics.
Approach:	They propose a new evaluation dataset to examine embeddings' ability to recognize fine-grained entities or events within encoded semantics.
Outcome:	The proposed dataset shows embeddings struggle with fine-grained matching . the proposed encoder outperforms the state-of-the-art 7B model in a small sample .

Copied to clipboard

Zhenpeng Su, Xing W, Wei Zhou, Guangyuan Ma, Songlin Hu

Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning (2026.findings-acl)

Challenge:	Existing studies on dialogue response selection focus on post-training and fine-tuning for cross-encoders.
Approach:	They propose a post-training technique tailored for dense encoders in dialogue response selection . they propose 'Dialogue Contextual Masking Auto-Encoder' which compresses dialogue semantics into dense vectors .
Outcome:	The proposed technique achieves state-of-the-art on two commonly evaluated benchmarks.

Copied to clipboard

Zhenpeng Su, Leiyu Pan, Minxuan Lv, Tiehua Mei, Zijia Lin, Yuntao Li, Wenping Hu, Ruiming Tang, Kun Gai, Guorui Zhou

Improving Graph-based Sentence Ordering with Iteratively Predicted Pairwise Orderings (2021.emnlp-main)

Challenge:	Large language model post-training often adopts an off-policy training paradigm . however, the off-poliicy training model introduces distribution shifts that push the policy beyond the trust region.
Approach:	They propose to use the entropy ratio as a global metric to measure the relative change in policy exploration throughout updates.
Outcome:	Experiments show that the proposed metric improves performance across multiple benchmarks.

Copied to clipboard

Shaopeng Lai, Ante Wang, Fandong Meng, Jie Zhou, Yubin Ge, Jiali Zeng, Junfeng Yao, Degen Huang, Jinsong Su

Structured Data in NLP

Challenge:	Existing sentence ordering models can be classified into pairwise ordering models and set-to-sequence models.
Approach:	They propose a novel sentence ordering framework which introduces two classifiers to make better use of pairwise orderings for graph-based sentence ordering.
Outcome:	The proposed model achieves state-of-the-art performance on five commonly-used datasets.

Chinese Court Simulation with LLM-Based Agents System (2026.findings-acl)

Copied to clipboard

Kaiyuan Zhang, Jiaqi Li, Yueyue Wu, Haitao Li, Cheng Luo, Shaokun Zou, Yujia Zhou, Weihang Su, Yiqun Liu, Qingyao Ai

A Self-Denoising Model for Robust Few-Shot Relation Extraction (2025.acl-long)

Challenge:	Existing studies have neglected the systematic design and procedure evaluation of court simulations, which are critical to the credibility and usage of court simulators in practice.
Approach:	They propose a court simulation paradigm based on the real-world procedure structure of Chinese courts and a framework that focuses on both legal judgment prediction and court procedure analysis.
Outcome:	The proposed model outperforms judges and lawyers from the real trials in many aspects.

Copied to clipboard

Liang Zhang, Yang Zhang, Ziyao Lu, Fandong Meng, Jie Zhou, Jinsong Su

ConceptMath: A Bilingual Concept-wise Benchmark for Measuring Mathematical Reasoning of Large Language Models (2024.findings-acl)

Challenge:	Existing studies assume that the support set contains only accurately labeled instances, but this assumption is often unrealistic.
Approach:	They propose a self-denoising model for FSRE which can automatically correct noisy labels of support instances.
Outcome:	The proposed model outperforms all baselines on two public datasets showing that it can correct mislabeled support instances.

Copied to clipboard

Yanan Wu, Jie Liu, Xingyuan Bu, Jiaheng Liu, Zhanhui Zhou, Yuanxing Zhang, Chenchen Zhang, ZhiqiBai ZhiqiBai, Haibin Chen, Tiezheng Ge, Wanli Ouyang, Wenbo Su, Bo Zheng

Query-Aware Knowledge Retrieval via Hyperbolic Structuring (2026.acl-long)

Challenge:	ConceptMath evaluates concept-wise mathematical reasoning of Large Language Models (LLMs) Existing benchmarks that evaluate general mathematical reasoning with an average accuracy fail to probe the fine-grained failure modes of mathematical reasoning on specific datasets.
Approach:	They introduce a bilingual, fine-grained benchmark that evaluates concept-wise mathematical reasoning of Large Language Models.
Outcome:	The proposed benchmarks evaluate concept-wise mathematical reasoning of Large Language Models with concept-based accuracies.

Copied to clipboard

Chuang Zhou, Junnan Dong, Yilin Xiao, Shengyuan Chen, Su Dong, di Yin, Xing Sun, Zhaozhuo Xu, Xiao Huang

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

Challenge:	Existing approaches focus primarily on retrieving isolated factual knowledge entities while neglecting the critical reasoning relationships.
Approach:	They propose a query-centric retrieval framework that explicitly integrates structured knowledge graphs to support complex reasoning tasks.
Outcome:	Extensive experiments on three benchmark datasets show that HyperRAG outperforms baselines.

Copied to clipboard

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

Lightweight Haar Wavelet Subband Pruning for LLMs (2026.findings-acl)

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.

Copied to clipboard

Jiang Li, Pengfei Cao, Chenxi Zhou, Tian Lan, Xiangdong Su, Kang Liu, Jun Zhao, Guanglai Gao

Challenge:	Large language models (LLMs) have impressive performance but require computational and memory resources.
Approach:	They propose a post-training framework that uses a Haar wavelet transform to prune weights.
Outcome:	The proposed pruning framework reduces pruning time and computational costs by removing less important weights while preserving model architecture.

Decoupling Reasoning and Knowledge Injection for In-Context Knowledge Editing (2025.findings-acl)

Copied to clipboard

Changyue Wang, Weihang Su, Qingyao Ai, Yujia Zhou, Yiqun Liu

MT-Bench-101: A Fine-Grained Benchmark for Evaluating Large Language Models in Multi-Turn Dialogues (2024.acl-long)

Challenge:	Existing knowledge editing approaches directly edit model context without isolating target knowledge from the reasoning path of model inference, resulting in unreliable and low-quality outputs, especially in multi-hop tasks.
Approach:	They propose a framework that separates model reasoning from knowledge editing and propose 'DecKER' that allows users to modify specific factual associations without retraining the entire model.
Outcome:	The proposed framework significantly improves multi-hop reasoning performance by mitigating knowledge conflicts and preserving reasoning integrity.

Copied to clipboard

Ge Bai, Jie Liu, Xingyuan Bu, Yancheng He, Jiaheng Liu, Zhanhui Zhou, Zhuoran Lin, Wenbo Su, Tiezheng Ge, Bo Zheng, Wanli Ouyang

Speaker Clustering in Textual Dialogue with Pairwise Utterance Relation and Cross-corpus Dialogue Act Supervision (2022.coling-1)

Challenge:	Large Language Models (LLMs) have greatly enhanced dialogue systems, but evaluation of their capabilities remains a challenge.
Approach:	They propose a model to evaluate the fine-grained abilities of Large Language Models in multi-turn dialogues.
Outcome:	The proposed model evaluates 21 popular chatbots based on MT-Bench-101 . it includes 3 overarching abilities and 13 distinct tasks within multi-turn dialogue scenarios.

Copied to clipboard

Zhihua Su, Qiang Zhou

Information Extraction & Text Mining

Challenge:	Existing models for textual dialogues do not include speaker annotations.
Approach:	They propose a speaker clustering model for textual dialogues that groups utterances without annotations so that the actual speakers are identical inside each cluster.
Outcome:	The proposed model outperforms the sequence classification baseline and benefits from the auxiliary dialogue act classification task.

OneRec-Think: In-Text Reasoning for Generative Recommendation (2026.acl-long)

Copied to clipboard

Zhanyu Liu, Shiyao Wang, Xingmei Wang, Rongzhou Zhang, Jiaxin Deng, Honghui Bao, Jinghao Zhang, Wuchao Li, PengFei Zheng, Xiangyu Wu, Yifei Hu, Qigen Hu, Xinchen Luo, Lejian Ren, Zhang Zixing, Qianqian Wang, Kuo Cai, Yunfan Wu, Hongtao Cheng, Zexuan Cheng, Lu Ren, Huanjie Wang, Yi Su, Ruiming Tang, Kun Gai, Guorui Zhou

Towards Robust k-Nearest-Neighbor Machine Translation (2022.emnlp-main)

Challenge:	Existing generative models lack the capacity for explicit and controllable reasoning, a key advantage of LLMs.
Approach:	They propose a framework that integrates dialogue, reasoning, and personalized recommendation.
Outcome:	Experiments across public benchmarks show state-of-the-art performance.

Copied to clipboard

Hui Jiang, Ziyao Lu, Fandong Meng, Chulun Zhou, Jie Zhou, Degen Huang, Jinsong Su

Text Generation

Challenge:	k-Nearest-Neighbor Machine Translation (kNN-MT) is a popular research paradigm in machine translation.
Approach:	They propose a confidence-enhanced kNN-MT model with robust training to reduce noise . they introduce NMT confidence to refine the modeling of important components of kN-MT .
Outcome:	The proposed model improves on four benchmark datasets and is robust to training.

RC3: Regularized Contrastive Cross-lingual Cross-modal Pre-training (2023.findings-acl)

Copied to clipboard

Chulun Zhou, Yunlong Liang, Fandong Meng, Jinan Xu, Jinsong Su, Jie Zhou

A Hybrid Detection and Generation Framework with Separate Encoders for Event Extraction (2023.eacl-main)

Challenge:	Existing V&L pre-training methods rely on strictly-aligned multilingual image-text pairs generated from English-centric datasets.
Approach:	They propose a regularized cross-lingual visual contrastive learning objective that constrains representation proximity of weakly-aligned multilingual image-text pairs.
Outcome:	The proposed model outperforms competing models with weak zero-shot capability on 5 multi-modal tasks across 6 languages.

Copied to clipboard

Ge Shi, Yunyue Su, Yongliang Ma, Ming Zhou