Papers by Cunxiang Wang
Can Generative Pre-trained Language Models Serve As Knowledge Bases for Closed-book QA? (2021.acl-long)
Copied to clipboard
| Challenge: | Existing work is limited in using small benchmarks with high test-train overlaps. |
| Approach: | They construct a dataset of closed-book QA using SQuAD and investigate the performance of BART. |
| Outcome: | Experiments show that pre-trained language models can achieve high performance on closed-book QA tasks. |
TRAMS: Training-free Memory Selection for Long-range Language Modeling (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods like Transformer-XL are plagued by ineffective memory selections due to the high number of tokens involved in attention calculation. |
| Approach: | They propose a plug-and-play strategy that selects tokens participating in attention calculation based on one simple metric and ignores the other ones. |
| Outcome: | The proposed strategy keeps tokens with high attention scores and ignores the other ones on word-level and character-level benchmarks without additional training or adding additional parameters. |
Training Language Model to Critique for Better Refinement (2025.findings-acl)
Copied to clipboard
Tianshu Yu, Chao Xiang, Mingchuan Yang, Pei Ke, Bosi Wen, Cunxiang Wang, Jiale Cheng, Li Zhang, Xinyu Mu, Chuxiong Sun, Minlie Huang
| Challenge: | Large language models (LLMs) have remarkable evaluation and critique capabilities, providing insightful feedback and identifying flaws in various tasks. |
| Approach: | They propose a framework to train critic models using refinement signals to generate feedback loops where critiques guide the model in refining its responses. |
| Outcome: | The proposed framework outperforms traditional methods and open-source models in terms of critique quality and refinement outcomes. |
Exploiting Abstract Meaning Representation for Open-Domain Question Answering (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing work attempts to address these challenges using Pretrained Language Models (PLMs) but the diversity of surface form expressions can hinder the model’s ability to capture accurate correlations, especially when the context is lengthy and complex. |
| Approach: | They propose a method known as Graph-as-Token (GST) to incorporate AMRs into PLMs to assist the model in understanding complex semantic information. |
| Outcome: | The proposed method outperforms existing methods and significantly improves performance on both Natural Questions and TriviaQA. |
RFiD: Towards Rational Fusion-in-Decoder for Open-Domain Question Answering (2023.findings-acl)
Copied to clipboard
| Challenge: | Open-domain Question Answering (ODQA) systems rely on spurious features instead of genuine causal relationships to generate answers. |
| Approach: | They propose a model that leverages the encoders of FiD to distinguish between causal relationships and spurious features and guides the decoder to generate answers informed by this discernment. |
| Outcome: | The proposed model improves on two ODQA datasets and shows that it can identify causal relationships and identify spurious features. |
SHIELD: Evaluation and Defense Strategies for Copyright Compliance in LLM Text Generation (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have transformed machine learning but have raised significant legal concerns due to their potential to produce text that infringes on copyrights. |
| Approach: | They propose a lightweight, real-time defense mechanism to prevent the generation of copyrighted text by evaluating methods and testing attack strategies. |
| Outcome: | The proposed defense significantly reduces the volume of copyrighted text generated by LLMs by effectively refusing malicious requests. |
Beyond Literal Mapping: Benchmarking and Improving Non-Literal Translation Evaluation (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have advanced machine translation (MT) a meta-evaluation dataset focused on non-literal translations is lacking . experimental results show the inaccuracies of traditional MT metrics and the limitations of LLM-as-a-Judge. |
| Approach: | They propose a meta-evaluation framework that leverages sub-agents to evaluate machine translation metrics. |
| Outcome: | The proposed framework improves on the knowledge cutoff and score inconsistency problem. |
Does it Make Sense? And Why? A Pilot Study for Sense Making and Explanation (P19-1)
Copied to clipboard
| Challenge: | Existing benchmarks measure common sense knowledge indirectly or without reasoning. |
| Approach: | They propose a benchmark to test whether a system can differentiate natural language statements that make sense from those that do not make sense. |
| Outcome: | The proposed benchmarks show that models trained on large corpora perform better than humans on some benchmarks. |
IF-CRITIC: Towards a Fine-Grained LLM Critic for Instruction-Following Evaluation (2026.acl-long)
Copied to clipboard
Bosi Wen, Yilin Niu, Cunxiang Wang, Pei Ke, Xiaoying Ling, Ying Zhang, Aohan Zeng, Hongning Wang, Minlie Huang
| Challenge: | Existing evaluation models for instruction-following have many shortcomings, such as substantial costs and unreliable assessments. |
| Approach: | They propose an LLM critic for fine-grained instruction-following evaluation using a checklist generator and a constraint-level preference optimization method. |
| Outcome: | The proposed model beats strong LLM-as-a-Judge baselines in evaluations under lower computational overhead compared to baselines. |
IF-RewardBench: Benchmarking Judge Models for Instruction-Following Evaluation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks for instruction-following lack data coverage and oversimplified pairwise evaluation paradigms that misalign with model optimization scenarios. |
| Approach: | They propose a meta-evaluation benchmark for instruction-following that covers diverse instruction and constraint types and a preference graph for each instruction. |
| Outcome: | Extensive experiments on IF-RewardBench show that the proposed benchmark achieves a stronger positive correlation with downstream task performance compared to existing benchmarks. |
Self-DC: When to Reason and When to Act? Self Divide-and-Conquer for Compositional Unknown Questions (2025.naacl-long)
Copied to clipboard
Hongru Wang, Boyang Xue, Baohang Zhou, Tianhua Zhang, Cunxiang Wang, Huimin Wang, Guanhua Chen, Kam-Fai Wong
| Challenge: | Existing studies focus on leveraging internal knowledge of Large Language Models (LLMs) to answer known questions. |
| Approach: | They propose a framework that allows LLMs to choose between internal and external knowledge . they use a dataset to analyze compositional questions that are composed of unknown sub-questions . |
| Outcome: | The proposed framework can achieve comparable or even better performance with much fewer external calls compared with several strong baselines. |
LongSafety: Evaluating Long-Context Safety of Large Language Models (2025.acl-long)
Copied to clipboard
Yida Lu, Jiale Cheng, Zhexin Zhang, Shiyao Cui, Cunxiang Wang, Xiaotao Gu, Yuxiao Dong, Jie Tang, Hongning Wang, Minlie Huang
| Challenge: | Large Language Models (LLMs) have demonstrated remarkable capabilities in understanding and generating long sequences. |
| Approach: | They propose a benchmark to evaluate LLM safety in open-ended long-context tasks . they find that relevant context and extended input sequences can exacerbate safety risks . |
| Outcome: | The proposed benchmark identifies significant safety vulnerabilities in 16 LLMs . strong safety performance in short-context scenarios does not correlate with safety in long-contact tasks . |
Knowledge Conflicts for LLMs: A Survey (2024.emnlp-main)
Copied to clipboard
| Challenge: | This survey examines knowledge conflicts for large language models (LLMs) this survey aims to shed light on strategies for improving the robustness of LLMs . |
| Approach: | They focus on three categories of knowledge conflicts: context-memory, inter-context, and intra-membry conflict. |
| Outcome: | The findings highlight the challenges faced by large language models when blending contextual and parametric knowledge. |
How Likely Do LLMs with CoT Mimic Human Reasoning? (2025.coling-main)
Copied to clipboard
| Challenge: | Using chain-of-thought to elicit reasoning capabilities is not always effective and accurate. |
| Approach: | They compare the reasoning process of LLMs with humans to understand the causal chain . they find that LLM deviates from the ideal causal chain, resulting in spurious correlations . |
| Outcome: | The proposed method does not improve performance or accurately represent reasoning processes in LLMs. |
Unlocking Recursive Thinking of LLMs: Alignment via Refinement (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for recursive reasoning are limited due to lack of expert-curated data. |
| Approach: | They propose a method that unlocks the potential of Large Language Models for recursive reasoning through long-form Chain of Thought. |
| Outcome: | The proposed method outperforms preference optimization methods on the openAI o1-series models by 20% on 3k synthetic samples. |
HPSS: Heuristic Prompting Strategy Search for LLM Evaluators (2025.findings-acl)
Copied to clipboard
Bosi Wen, Pei Ke, Yufei Sun, Cunxiang Wang, Xiaotao Gu, Jinfeng Zhou, Jie Tang, Hongning Wang, Minlie Huang
| Challenge: | Existing efforts to optimize text evaluation prompts neglect the combinatorial impact of multiple factors, leading to insufficient optimization of the evaluation pipeline. |
| Approach: | They propose to integrate 8 key factors for evaluation prompts and integrate them into an algorithm that searches for well-behaved prompting strategies for LLM evaluators. |
| Outcome: | The proposed method outperforms existing methods and human-designed evaluation prompts on four evaluation tasks. |
HoWToBench: Holistic Evaluation for LLM’s Capability in Human-level Writing using Tree of Writing (2026.acl-long)
Copied to clipboard
Andrew Zhuoer Feng, Cunxiang Wang, Yu Luo, Lin Fan, Irene Zhou, Zikang Wang, Xiaotao Gu, Jie Tang, Hongning Wang, Minlie Huang
| Challenge: | Evaluating the writing capabilities of large language models remains a significant challenge due to the multidimensional nature of writing skills and the limitations of existing metrics. |
| Approach: | They propose to model the aggregation weights of sub-features in a tree-structured workflow and propose a Chinese writing benchmark that mitigates biases. |
| Outcome: | The proposed tree-of-writing (ToW) measures the writing capabilities of large language models (LLMs) in Chinese and shows that it mitigates biases and achieves a *0.93* Pearson correlation with human judgments. |
CPRM: A LLM-based Continual Pre-training Framework for Relevance Modeling in Commercial Search (2025.naacl-industry)
Copied to clipboard
Kaixin Wu, Yixin Ji, Zeyuan Chen, Qiang Wang, Cunxiang Wang, Hong Liu, Baijun Ji, Xu Jia, Zhongyi Liu, Jinjie Gu, Yuan Zhou, Linjian Mo
| Challenge: | Relevance modeling between queries and items is a key component of commercial search engines. |
| Approach: | They propose a framework for continual pre-training of LLMs to enhance domain knowledge . they employ queries and multi-field item to jointly pre-train for enhancing domain knowledge. |
| Outcome: | The proposed model achieves convincing performance compared to strong baselines. |
LONG2RAG: Evaluating Long-Context & Long-Form Retrieval-Augmented Generation with Key Point Recall (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Retrieval-augmented generation (RAG) is a promising approach to address limitations of fixed knowledge in large language models. |
| Approach: | They propose a benchmark and a metric to assess LLMs' ability to generate long-form responses that exploit retrieved information. |
| Outcome: | The proposed benchmarks lack a comprehensive evaluation method to assess LLMs' ability to generate long-form responses that effectively exploits retrieved information. |
Nash CoT: Multi-Path Inference with Preference Equilibrium (2024.emnlp-main)
Copied to clipboard
| Challenge: | Multi-path inference is an improvement on multi-path reasoning, but there is no optimal setting for the number of inference paths. |
| Approach: | They propose to use question-related role templates to guide LLMs into relevant roles to reduce the dependence on the number of inference paths. |
| Outcome: | The proposed system can achieve comparable or better results than self-consistency with the same number of paths. |