Papers by Liang Song

Two Pathways to Truthfulness: On the Intrinsic Encoding of LLM Hallucinations (2026.acl-long)

Challenge:	Existing benchmarks for role-playing agents only evaluate surface-level fidelity and provide limited insight into decision making under role–alignment value conflicts.
Approach:	They propose a benchmark to evaluate RPAs under role–alignment value conflicts . they use 8k diverse role profiles and 240k dilemma instances to evaluate role-aware decision making .
Outcome:	The proposed benchmark covers 8k diverse role profiles and scenarios and nearly 240k dilemma instances across three difficulty levels and eight role categories.

Copied to clipboard

Wen Luo, Guangyue Peng, Wei Li, Shaohang Wei, Feifan Song, Liang Wang, Nan Yang, Xingxing Zhang, Jing Jin, Furu Wei, Houfeng Wang

UHGEval: Benchmarking the Hallucination of Chinese Large Language Models via Unconstrained Generation (2024.acl-long)

Challenge:	Previous work shows that large language models generate hallucinations, yet the origins and mechanisms of these signals remain unclear.
Approach:	They propose to validate and disentangle two different pathways for truthfulness cues . they also propose to use the same mechanism to derive self-contained evidence from the generated answer .
Outcome:	The proposed applications improve hallucination detection performance by integrating two different inputs.

Copied to clipboard

Xun Liang, Shichao Song, Simin Niu, Zhiyu Li, Feiyu Xiong, Bo Tang, Yezhaohui Wang, Dawei He, Cheng Peng, Zhonghao Wang, Haiying Deng

CLOMO: Counterfactual Logical Modification with Large Language Models (2024.acl-long)

Challenge:	Large language models (LLMs) produce hallucinated text, compromising their practical utility in professional contexts.
Approach:	They have developed an unconstrained hallucination generation evaluation benchmark that contains hallucines generated by large language models with minimal restrictions.
Outcome:	The proposed benchmarks are based on a Chinese-language dataset that is lacking in the field.

Copied to clipboard

Yinya Huang, Ruixin Hong, Hongming Zhang, Wei Shao, Zhicheng Yang, Dong Yu, Changshui Zhang, Xiaodan Liang, Linqi Song

ServImage: An Image Generation and Editing Benchmark from Real-world Commercial Imaging Services (2026.acl-long)

Challenge:	Existing studies on evaluating model reasoning are limited in both form and content.
Approach:	They propose a task to cultivate counterfactual thought processes within large language models and an evaluation metric to evaluate their natural language output instead of modeling the task as a multiple-choice problem.
Outcome:	The proposed evaluation metric aligns well with human preference.

Copied to clipboard

Fengxian Ji, Jingpu Yang, Zirui Song, Lang Gao, Junhong Liang, Zhenhao Chen, Jinghui Zhang, Xiuying Chen

Knowledge-Augmented Multimodal Clinical Rationale Generation for Disease Diagnosis with Small Language Models (2025.acl-long)

Challenge:	Recent image generation and editing models demonstrate robust adherence to instructions and high visual quality on academic benchmarks.
Approach:	They propose a benchmark that correlates image outputs with economic value in commercial design projects.
Outcome:	ServImage benchmarks show image generation models perform well on academic benchmarks but are uncertain on commercial projects.

Copied to clipboard

Shuai Niu, Jing Ma, Hongzhan Lin, Liang Bai, Zhihua Wang, Richard Yi Da Xu, Yunya Song, Xian Yang

Reconstruct Before Summarize: An Efficient Two-Step Framework for Condensing and Summarizing Meeting Transcripts (2023.emnlp-main)

Challenge:	Existing models struggle to balance predictive accuracy with human-understandable rationales.
Approach:	They propose to enhance LLMs by leveraging rationale distillation and domain knowledge injection for trustworthy multimodal rationale generation.
Outcome:	Experiments on real-world medical datasets show that ClinRaGen achieves state-of-the-art performance in disease diagnosis and rationale generation.

Copied to clipboard

Haochen Tan, Han Wu, Wei Shao, Xinyun Zhang, Mingjie Zhan, Zhaohui Hou, Ding Liang, Linqi Song

Context-aware Watermark with Semantic Balanced Green-red Lists for Large Language Models (2024.emnlp-main)

Challenge:	Existing approaches to meeting summarization are limited due to noise, lengthy transcripts, and scattered salient information.
Approach:	They propose a two-step framework for meeting summarization that leverages a self-supervised paradigm to reconstruct transcripts and a relative positional bucketing algorithm to equip models to generate the summary.
Outcome:	The proposed method significantly reduces memory consumption and processing time on two meeting summarization datasets.

Copied to clipboard

Yuxuan Guo, Zhiliang Tian, Yiping Song, Tianlun Liu, Liang Ding, Dongsheng Li

Robustness in NLP

Challenge:	Recent research suggests that watermarking methods cause degradation of text quality due to semantic disparities between the watermarked text and the unwatermarked text.
Approach:	They propose a semantic-aware watermark method that generates a watermark key considering contexts to split a green/red list for watermark injection.
Outcome:	The proposed method reduces performance drop due to adding bias on green lists . it also allows green lists to cover almost all semantics .

PsyScam: A Benchmark for Psychological Techniques in Real-World Scams (2025.findings-emnlp)

Copied to clipboard

Shang Ma, Tianyi Ma, Jiahao Liu, Wei Song, Zhenkai Liang, Xusheng Xiao, Yanfang Ye

Challenge:	PTs are employed by scammers to manipulate victims and cause lasting psychological trauma.
Approach:	They propose a benchmark to capture the PTs employed in real-worldscam reports and investigate how LLMs can be utilized to generate variants of scams based on the pts and the contexts provided by thesescams.
Outcome:	The proposed model can generate variants of scams based on the PTs employed in real-world scam reports and the contexts provided by these scams.

Circuit Complexity Bounds for RoPE-based Transformer Architecture (2025.emnlp-main)

Copied to clipboard

Bo Chen, Xiaoyu Li, Yingyu Liang, Jiangxuan Long, Zhenmei Shi, Zhao Song, Jiahao Zhang

Relation Discovery with Out-of-Relation Knowledge Base as Supervision (N19-1)

Challenge:	Recent studies provide the circuit complexity bounds to Transformer-like architectures. position embedding has emerged as a crucial technique in modern large language models.
Approach:	They propose to use position embedding to improve Transformer-like architectures by analyzing their circuits and analyzing the results.
Outcome:	The proposed model is able to solve canonical tasks without embedding positional information.

Copied to clipboard

Yan Liang, Xin Liu, Jianwen Zhang, Yangqiu Song

From Hypothesis to Publication: A Comprehensive Survey of AI-Driven Research Support Systems (2025.findings-emnlp)

Challenge:	Existing methods to extract relations from text corpus without annotated data are violated by up to 31%.
Approach:	They propose to use out-of-relation knowledge bases to supervise the discovery of unseen relations where relations to discover from the text corpus and those in knowledge bases are not overlapped.
Outcome:	The proposed method improves the state-of-the-art relation discovery performance by a large margin.

Copied to clipboard

Zekun Zhou, Xiaocheng Feng, Lei Huang, Xiachong Feng, Ziyun Song, Ruihan Chen, Liang Zhao, Weitao Ma, Yuxuan Gu, Baoxin Wang, Dayong Wu, Guoping Hu, Ting Liu, Bing Qin

TRACE: Two-Phase RL for Causal Graph Exploration and Deeper Psychological Intervention in Dynamic Counseling Scenarios (2026.findings-acl)

Challenge:	rapid development of artificial intelligence (AI) technologies has inspired researchers to explore how AI can accelerate and enhance research.
Approach:	They organize the relevant studies into three main categories: hypothesis formulation, hypothesis validation, and manuscript publication.
Outcome:	The authors summarize the current state of research in three main areas: hypothesis formulation, hypothesis validation, and manuscript publication.

Copied to clipboard

Shilin Tang, Zunyi Yin, Xuefeng Liang, Guanghui Shi, Song Tong, Chen Guangyu

Low-code LLM: Graphical User Interface over Large Language Models (2024.naacl-demo)

Challenge:	Existing models lack the ability to actively explore the underlying causes of psychological distress.
Approach:	They propose a two-phase reinforcement learning framework that implements a causal-graph-driven reward scheme across two phases: an exploration phase that rewards the causal graph reconstruction following a surface-to-deep path, and an intervention phase that supports targeted restructuring of irrational beliefs.
Outcome:	Extensive experiments show that TRACE outperforms existing models, enabling causal-chain-aware psychological intervention beyond surface-level empathy.

Copied to clipboard

Yuzhe Cai, Shaoguang Mao, Wenshan Wu, Zehua Wang, Yaobo Liang, Tao Ge, Chenfei Wu, WangYou WangYou, Ting Song, Yan Xia, Nan Duan, Furu Wei

Empathy in Diversity: Personalized Depression and Anxiety Therapy via Dialogue State Tracking and Patient-Aware Planning (2026.acl-long)

Challenge:	Low-code LLM is a visual programming interface that allows users to incorporate their ideas into the process without writing trivial prompts.
Approach:	They propose a human-LLM interaction framework that incorporates low-code visual programming interactions to achieve more controllable and stable responses.
Outcome:	The proposed framework enables users to incorporate ideas into the process without writing trivial prompts.

Copied to clipboard

Xinwei Yang, Junyi Fan, Yuqing Liu, Jiaxuan Wang, Jiashuai Zhang, Hongru Liang, Wenqiang Lei, Yao Song

Dialogue Systems & Conversational Agents

Challenge:	Recent efforts have turned to large language models (LLMs) as therapeutic agents for psychological therapy tasks, yet robustness across diverse patients remains underexplored.
Approach:	They propose a realistic role-play protocol for evaluating therapeutic dialogue agents and a de-identified, expert-annotated corpus of therapist–patient dialogues.
Outcome:	The proposed framework outperforms baselines on therapeutic outcomes and dialogue quality while improving conversational efficiency.

VizoMem: A Visual-Textual Memory Framework for Efficient Long-Horizon Reasoning (2026.findings-acl)

Copied to clipboard

Weijie Liang, Yuanfeng Song, Xing Chen, Caleb Chen Cao, Sirui Han, Yike Guo

Visual Data in NLP

Challenge:	Existing systems that use long-context modeling incur computational and memory overhead.
Approach:	They propose a visual memory framework that pre-rendered text into structured images and stored as visual notes for agentic systems.
Outcome:	The proposed system reduces token consumption while preserving effective long-term memory recall.

Controlled Text Generation for Large Language Model with Dynamic Attribute Graphs (2024.findings-acl)

Copied to clipboard

Xun Liang, Hanyu Wang, Shichao Song, Mengting Hu, Xunzhi Wang, Zhiyu Li, Feiyu Xiong, Bo Tang

Improving Deep Embedded Clustering via Learning Cluster-level Representations (2022.coling-1)

Challenge:	Controlled Text Generation (CTG) aims to produce texts that exhibit specific desired attributes.
Approach:	They propose a pluggable CTG framework for Large Language Models to control text . they use attribute scorers to evaluate attributes of sentences and construct dynamic attribute graphs .
Outcome:	The proposed framework achieves a peak improvement of 19.29% over baseline methods in two tasks.

Copied to clipboard

Qing Yin, Zhihua Wang, Yunya Song, Yida Xu, Shuai Niu, Liang Bai, Yike Guo, Xian Yang

Information Extraction & Text Mining

Challenge:	Existing efforts to learn meaningful representations at the instance level are limited.
Approach:	They propose a deep embedded clustering model with cluster-level representation learning to jointly learn cluster and instance level representations.
Outcome:	The proposed model produces meaningful clusters on real-world short text datasets.

Your Reasoning Model is Secretly a Reward Model - Optimization-Free Verification from Experience (2026.acl-long)

Copied to clipboard

Zhenwen Liang, Ruosen Li, Yujun Zhou, Linfeng Song, Dian Yu, Xinya Du, Haitao Mi, Dong Yu

Extracting and Combining Abilities For Building Multi-lingual Ability-enhanced Large Language Models (2025.emnlp-main)

Challenge:	Existing verifiers operate on the surface text or on confidence proxies derived from token probabilities, which can be brittle.
Approach:	They propose a training-free, non-parametric verifier that summarizes each reasoning trace by an activation delta and compares it to two class centroids computed from labeled experience.
Outcome:	The proposed model improves selection and reranking on large and less-calibrated models.

Copied to clipboard

Zhipeng Chen, Kun Zhou, Liang Song, Xin Zhao, Bingning Wang, Weipeng Chen, Ji-Rong Wen

Self-Evolution Knowledge Distillation for LLM-based Machine Translation (2025.coling-main)

Challenge:	Existing work relies on training with multi-lingual ability-related data, which may not be available for low-resource languages.
Approach:	They propose a multi-lingual ability-enhanced LLM that extracts language-agnostic ability-related weights from LLMs and combine them across different languages by simple addition and subtraction operations without training.
Outcome:	The proposed approach extracts language-agnostic ability-related weights from LLMs and combine them across different languages without training.

Copied to clipboard

Yuncheng Song, Liang Ding, Changtong Zan, Shujian Huang

DeepPlanner: Scaling Planning Capability for Deep Research Agents via Advantage Shaping (2026.findings-acl)

Challenge:	Existing knowledge distillation strategies for large language models minimize output distributions between student and teacher models indiscriminately for each token.
Approach:	They propose a distillation strategy that integrates teacher and one-hot distribution of ground truth into the student distribution as prior knowledge, which promotes the distillation process.
Outcome:	The proposed method brings an average improvement of approximately 1.4 SacreBLEU points across four translation directions in the WMT22 test sets.

Copied to clipboard

Wei Fan, Wenlin Yao, Zheng Li, Feng Yao, Xin Liu, Liang Qiu, Qingyu Yin, Yangqiu Song, Bing Yin

Improving Semantic Matching through Dependency-Enhanced Pre-trained Model with Adaptive Fusion (2022.findings-emnlp)

Challenge:	Existing approaches to planning involve implicit planning or introduce explicit planners without systematically optimizing the planning stage.
Approach:	They propose an end-to-end RL framework that enhances the planning capabilities of deep research agents.
Outcome:	Experiments show that DeepPlanner improves planning quality and achieves state-of-the-art results under a lower training budget.

Copied to clipboard

Jian Song, Di Liang, Rumei Li, Yuntao Li, Sirui Wang, Minlong Peng, Wei Wu, Yongxin Yu

FTibSuite: A Comprehensive Resource Suite for Tibetan Vision–Language Modeling (2026.findings-acl)

Challenge:	Existing work on dependency prior structure integration into pre-trained models is still unclear.
Approach:	They propose a dependency-based fusion attention paradigm which explicitly introduces dependency prior structure into pre-trained models and adaptively fuses it with semantic information.
Outcome:	The proposed model achieves state-of-the-art or competitive performance on 10 public datasets, demonstrating the benefits of adaptively fusing dependency structure in semantic matching task.

Copied to clipboard

Guixian Xu, Yide Liang, Zeli Su, Xuexian Song, Ziyin Zhang, Yushuang Dong, Ting Zhang, Xu Han

Toolscaler: Scalable Generative Tool Calling via Structure-Aware Semantic Tokenization (2025.findings-emnlp)

Challenge:	FTibSuite provides an end-to-end training-and-evaluation workflow for vision–language models . Tibetan is underserved due to the lack of infrastructure for reproducible training and evaluation.
Approach:	They propose a resource-centric workflow for Tibetan VLMs that provides an end-to-end training-and-evaluation workflow and human-verified multimodal annotations.
Outcome:	FTibSuite provides an end-to-end training-and-evaluation workflow and human-verified multimodal annotations.

Copied to clipboard

Yunyue Su, Zhang Jinshuai, Bowen Fang, Wen Ye, Jinghao Zhang, Bowen Song, Weiqiang Wang, Qiang Liu, Liang Wang

Syntactic Text Processing

Challenge:	Extensive experiments demonstrate the effectiveness of SGTC across various tasks.
Approach:	They propose a generative tool invocation framework that introduces structure-aware semantic tokenization to encode tools as discrete code sequences.
Outcome:	The proposed framework reduces the size of the representation space and underutilizes collaborative signals among tools in downstream tasks.

Conv-Basis: A New Paradigm for Efficient Attention Inference and Gradient Computation in Transformers (2025.findings-emnlp)

Copied to clipboard

Yingyu Liang, Heshan Liu, Zhenmei Shi, Zhao Song, Zhuoyan Xu, Jiale Zhao, Zhen Zhuang

Never Lost in the Middle: Mastering Long-Context Question Answering with Position-Agnostic Decompositional Training (2024.acl-long)

Challenge:	a large computational cost for attention computation in large language models is a major obstacle .
Approach:	They propose a convolution-like structure for attention computation using convolution matrices . they then propose an efficient approximation method to approximate the attention matrix .
Outcome:	The proposed method achieves nearly linear time complexity in n1+o(1) time.

Copied to clipboard

Junqing He, Kunhao Pan, Xiaoqun Dong, Zhuoyang Song, LiuYiBo LiuYiBo, Qianguosun Qianguosun, Yuxin Liang, Hao Wang, Enming Zhang, Jiaxing Zhang

Natural Language Interfaces

Challenge:	Large language models suffer from severe hallucinations, compromising performance in knowledge-oriented QA, dialogue, and writing.
Approach:	They propose to enhance the information searching and reflection ability of large language models by training them in position-agnostic multi-step QA tasks to improve their model's accuracy.
Outcome:	The proposed model improves in multi-doc QA and other benchmarks by 13.7% absolute gain in shuffled settings, by 21.5% in passage retrieval task.

Towards Infinite-Long Prefix in Transformer (2025.emnlp-main)

Copied to clipboard

Yingyu Liang, Zhenmei Shi, Zhao Song, Chiwun Yang

VCSUM: A Versatile Chinese Meeting Summarization Dataset (2023.findings-acl)

Challenge:	Prefix Learning is an empirically efficient and effective method for language models . but the theoretical understandings are limited on the performance of such methods .
Approach:	They propose a method that can train an ultra-long prefix in a stylized setting using the Neural Tangent Kernel framework.
Outcome:	The proposed method can achieve superior performance on vision, natural language, and math data.

Copied to clipboard

Han Wu, Mingjie Zhan, Haochen Tan, Zhaohui Hou, Ding Liang, Linqi Song

Text Generation

Challenge:	Compared to news and chat summarization, meeting summarizing is decelerated by the limited data.
Approach:	They propose a Chinese meeting summarization dataset that provides annotations for each transcript and a set of benchmark models to facilitate further research.
Outcome:	The proposed model can be used to summarize the content of meeting transcripts in Chinese.

EconProver: Towards More Economical Test-Time Scaling for Automated Theorem Proving (2026.acl-long)

Copied to clipboard

Mukai Li, Linfeng Song, Zhenwen Liang, Jiahao Xu, Shansan Gong, Qi Liu, Haitao Mi, Dong Yu

MetaGPT: Merging Large Language Models Using Model Exclusive Task Arithmetic (2024.emnlp-main)

Challenge:	Large Language Models (LLMs) have recently advanced the field of Automated Theorem Proving (ATP) Existing cost analyses regulate only the number of sampling passes, ignoring the substantial disparities in sampling costs.
Approach:	They propose to integrate two complementary methods into a unified EconRL pipeline to increase pass rates under constrained sampling passes.
Outcome:	The proposed method reduces token usage and sample passes while maintaining the original performance.

Copied to clipboard

Yuyan Zhou, Liang Song, Bingning Wang, Weipeng Chen

PibE-MPP: A Play-it-by-Ear Masking Performance Plug-in for LLMs (2026.findings-acl)

Challenge:	Existing methods face the trilemma of performance, data privacy, and computational costs, which hinders their application to LLMs.
Approach:	They propose a model-exclusive task arithmetic method for merging GPT-scale models which is data-agnostic and bypasses the heavy search process.
Outcome:	The proposed method achieves state-of-the-art performance on multiple tasks while minimizing the average loss difference between the merged model and each individual task model.

Copied to clipboard

Mengwei Wang, Simin Niu, Xun Liang, Yuefeng Ma, Sensen Zhang, Jiawei Yang, Shichao Song, Hanyu Wang, Huayi Lai

Efficient Data Labeling by Hierarchical Crowdsourcing with Large Language Models (2025.coling-main)

Challenge:	Random masking is a widely adopted classic baseline in large language models (LLMs).
Approach:	They propose a play-it-by-ear masking performance plug-in which enables LLMs to adaptively select masking target combinations for each task.
Outcome:	The proposed performance plug-in retains the advantages and mitigates the drawbacks of random masking in large language models.

Copied to clipboard

Haodi Zhang, Junyu Yang, Jinyin Nie, Peirou Liang, Kaishun Wu, Defu Lian, Rui Mao, Yuanfeng Song

DABERT: Dual Attention Enhanced BERT for Semantic Matching (2022.coling-1)

Challenge:	Large language models (LLMs) have been gaining attention for their impressive performance in in-context dialogues.
Approach:	They propose a hierarchical framework that leverages multiple LLMs for efficient data labeling under budget constraints.
Outcome:	The proposed framework outperforms human labelers and GPT-4 in terms of accuracy and efficiency.

Copied to clipboard

Sirui Wang, Di Liang, Jian Song, Yuntao Li, Wei Wu

Following the Autoregressive Nature of LLM Embeddings via Compression and Alignment (2025.emnlp-main)

Challenge:	Existing models for semantic sentence matching lack the ability to capture subtle differences.
Approach:	They propose to use a Transformer-based pre-trained language model to capture fine-grained differences in sentence pairs by introducing a dual attention module and a fusion module to learn the aggregation of difference and affinity features.
Outcome:	The proposed method is able to capture fine-grained differences in sentence pairs.

Copied to clipboard

Jingcheng Deng, Zhongtao Jiang, Liang Pang, Zihao Wei, Liwei Chen, Kun Xu, Yang Song, Huawei Shen, Xueqi Cheng