Papers by Xingyu Li

32 papers
Deceptive Semantic Shortcuts on Reasoning Chains: How Far Can Models Go without Hallucination? (2024.naacl-long)

Copied to clipboard

Challenge: Existing large language models (LLMs) suffer from hallucinations and unfaithful reasoning due to keyword/entity biases.
Approach: They propose a new probing method and benchmark to quantify this phenomenon by using a keyword/entity biases-based probing technique called EUREQA.
Outcome: The proposed method achieves 62% accuracy on multi-hop and complex QA benchmarks.
Research Replication Prediction Using Weakly Supervised Learning (2020.findings-emnlp)

Copied to clipboard

Challenge: Existing methods to predict scientific claims’ replicability use only hand-extracted statistics features without utilizing research papers’ text information.
Approach: They propose two weakly supervised learning approaches that use automatically extracted text information of research papers to improve the prediction accuracy of research replication using both labeled and unlabeled datasets.
Outcome: The proposed methods achieve an accuracy of 75.76% over real-world datasets.
JiraiBench: A Bilingual Benchmark for Evaluating Large Language Models’ Detection of Human risky health behavior Content in Jirai Community (2026.eacl-long)

Copied to clipboard

Challenge: a cross-lingual dataset captures a transnational cultural phenomenon . risky health behaviors (RHB) are often linked to complex mental health conditions .
Approach: They present the first cross-lingual dataset that captures a transnational cultural phenomenon . their dataset of more than 15,000 annotated social media posts forms the core of JiraiBench .
Outcome: The study shows that cultural context can be more influential than linguistic similarity . the study also shows that the Japanese prompts better handle Chinese content .
DocMMIR: A Framework for Document Multi-modal Information Retrieval (2025.findings-emnlp)

Copied to clipboard

Challenge: Existing multi-modal information retrieval models lack a comprehensive exploration of document-level retrieval . existing models suffer from the absence of cross-domain datasets at this granularity.
Approach: They propose a multi-modal document retrieval framework to unify diverse document formats and domains with a comprehensive retrieval scenario.
Outcome: The proposed framework improves document retrieval performance on a large multimodal dataset.
RealChart2Code: Bridging the Gap in Real-World Chart-to-Code Generation via Multi-Task Evaluation (2026.acl-long)

Copied to clipboard

Challenge: Vision-Language Models (VLMs) have demonstrated impressive capabilities in code generation across various domains, but their ability to replicate complex, multi-panel visualizations remains largely unassessed.
Approach: They propose a large-scale benchmark to evaluate chart generation from large- scale raw data and assess iterative code refinement in a multi-turn conversational setting.
Outcome: The new benchmark evaluates 14 leading VLMs on real-world data and shows they struggle with complex plot structures and authentic data.
Topic-Aware Evidence Reasoning and Stance-Aware Aggregation for Fact Verification (2021.acl-long)

Copied to clipboard

Challenge: Existing methods for fact verification focus on analyzing semantic interaction between claim and evidence but fail to capture their topical consistency . Existing models focus on the aggregation of multiple pieces of evidence without considering their implicit stances to the claim, thereby introducing spurious information.
Approach: They propose a topic-aware evidence reasoning and stance-again aggregation model that checks topical consistency between claims and evidence.
Outcome: The proposed model outperforms state-of-the-art models on two benchmark datasets.
ReasonMed: A 370K Multi-Agent Generated Dataset for Advancing Medical Reasoning (2025.emnlp-main)

Copied to clipboard

Challenge: Existing medical reasoning datasets are limited in scale and typically rely on incomplete data.
Approach: They propose to use ReasonMed to train medical reasoning models using a multi-agent generation, verification, and refinement pipeline.
Outcome: The largest medical reasoning dataset to date surpasses the prior best sub-10B models by 4.17% and even exceeds LLaMA3.1-70B on PubMedQA by 4.60%.
Improving Contrastive Learning of Sentence Embeddings with Focal InfoNCE (2023.findings-emnlp)

Copied to clipboard

Challenge: SimCSE does not fully exploit the potential of hard negative samples in contrastive learning.
Approach: They propose an unsupervised contrastive learning framework that combines SimCSE with hard negative mining to enhance the quality of sentence embeddings.
Outcome: The proposed framework improves sentence embeddings on various STS benchmarks in terms of Spearman’s correlation, representation alignment and uniformity.
Social Welfare Function Leaderboard: On the Emergence of LLM Agents as the Welfare Dictator (2026.findings-acl)

Copied to clipboard

Challenge: Large language models (LLMs) are increasingly entrusted with high-stakes decisions that affect human welfare.
Approach: They evaluate 20 state-of-the-art Large language models (LLMs) and 20 LLM dictators to create a social welfare function benchmark.
Outcome: The proposed model creates dilemma between maximizing collective efficiency and ensuring distributive fairness.
MODDP: A Multi-modal Open-domain Chinese Dataset for Dialogue Discourse Parsing (2024.findings-acl)

Copied to clipboard

Challenge: Existing benchmark datasets for discourse parsing are domain-specific and contain only textual modality . this makes it difficult to accurately understand the dialogue without multi-modal clues .
Approach: They propose a multi-modal Chinese discourse parsing dataset based on open-domain dialogues . they propose to integrate multi-modality into the original textual unimodal DDP model .
Outcome: The proposed dataset improves on the existing unimodal model by adding multimodalities to the model.
Too Good to be Bad: On the Failure of LLMs to Role-Play Villains (2026.findings-acl)

Copied to clipboard

Challenge: Large Language Models (LLMs) are increasingly tasked with creative generation, but their ability to portray non-prosocial, antagonistic personas remains largely unexamined.
Approach: They propose a moral alignment benchmark to test the safety of large language models . they find that models struggle with traits directly antithetical to safety principles .
Outcome: The proposed model fails to accurately portray morally ambiguous or villainous characters . the model fails most with traits directly antithetical to safety principles .
Sentient Agent as a Judge: Evaluating Higher-Order Social Cognition in Large Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Large language models (LLMs) have evolved from statistical sequence predictors to sophisticated autonomous agents capable of reasoning, planning, and sustaining multi-turn conversa-tions.
Approach: They propose a system that instantiates a "Sentient Agent" that simulates human-like emotional changes and inner thoughts to provide a more realistic evaluation of the model in multi-turn conversations.
Outcome: The proposed framework measures the agent's higher-order social cognition in multi-turn conversations.
Agent-Dice: Disentangling Knowledge Updates via Geometric Consensus for Agent Continual Learning (2026.findings-acl)

Copied to clipboard

Challenge: Large Language Model (LLM)-based agents extend the utility of LLMs by interacting with dynamic environments.
Approach: They propose a parameter fusion framework based on directional consensus evaluation that disentangles knowledge updates through a two-stage process.
Outcome: The proposed framework disentangles knowledge updates through a two-stage process with minimal computational overhead and parameter updates.
TrajGuard: Streaming Hidden-state Trajectory Detection for Decoding-time Jailbreak Defense (2026.findings-acl)

Copied to clipboard

Challenge: Existing jailbreak defense paradigms rely on static detection of prompts, outputs, or internal states . hidden states in critical layers during decoding carry stronger and more stable risk signals .
Approach: They propose a decoding-time defense framework that aggregates hidden-state trajectories via a sliding window to quantify risk in real time.
Outcome: The proposed framework achieves an average defense rate of 95% in 12 jailbreak attacks and open-source LLMs.
TPTU-v2: Boosting Task Planning and Tool Usage of Large Language Model-based Agents in Real-world Industry Systems (2024.emnlp-industry)

Copied to clipboard

Challenge: Large language models have demonstrated proficiency in addressing tasks that necessitate a combination of task planning and the usage of external tools.
Approach: They propose a framework to enhance the task planning and tool usage abilities of LLMs in industrial systems.
Outcome: The proposed framework enhances the task planning and tool usage abilities of LLM-based agents in industrial systems.
Multi-stage Distillation Framework for Cross-Lingual Semantic Similarity Matching (2022.findings-naacl)

Copied to clipboard

Challenge: Existing studies have shown that cross-lingual knowledge distillation can improve the performance of pre-trained models for cross-linguistic similarity matching tasks.
Approach: They propose a multi-stage distillation framework for constructing a small-size but high-performance cross-lingual model using contrastive learning, bottleneck, and parameter recurrent strategies.
Outcome: The proposed model can compress the size of XLM-R and MiniLM by more than 50% while the performance is only reduced by about 1%.
MoleculeQA: A Dataset to Evaluate Factual Accuracy in Molecular Comprehension (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing models generate erroneous information and evaluations fail to assess factual correctness of models.
Approach: They propose to use MoleculeQA to evaluate molecular factual correctness in large language models by organizing molecules into a taxonomy and building QA pairs through human and LLM efforts.
Outcome: The proposed model improves the factual correctness of generated information and enables the development of new models.
Multimodal Coreference Resolution for Chinese Social Media Dialogues: Dataset and Benchmark Approach (2025.acl-long)

Copied to clipboard

Challenge: Multimodal coreference resolution (MCR) aims to identify mentions referring to the same entity across different modalities, such as text and visuals.
Approach: They propose a Chinese multimodal coreference dataset based on Douyin short-video platform to help researchers understand multimodal content.
Outcome: The proposed dataset pairs short videos with corresponding textual dialogues from user comments and includes manually annotated coreference clusters for person mentions in the text and the coreferential person head regions in the corresponding video frames.
Interleaved Latent Visual Reasoning with Selective Perceptual Modeling (2026.acl-long)

Copied to clipboard

Challenge: Existing approaches to interleaved reasoning are limited by the cost of re-encoding pixel-dense images.
Approach: They propose a framework that unifies dynamic state evolution with precise perceptual modeling.
Outcome: The proposed framework outperforms existing approaches on multimodal reasoning benchmarks.
ReAgent: Reversible Multi-Agent Reasoning for Knowledge-Enhanced Multi-Hop QA (2025.emnlp-main)

Copied to clipboard

Challenge: Multi-hop question answering (QA) is a central challenge in natural language processing . early mistakes can cause errors and undermine the final result, authors say .
Approach: They propose a reversible multi-agent reasoning framework that backtracks to earlier valid states when conflicts arise.
Outcome: Empirical evaluation shows that the framework improves on forward-only benchmarks by 6% . the approach enables agents to backtrack to valid states when conflicts arise .
One Agent to Serve All: a Lite-Adaptive Stylized AI Assistant for Millions of Multi-Style Official Accounts (2026.findings-acl)

Copied to clipboard

Challenge: Existing methods for prompting in official accounts are computationally prohibitive and lack contextually grounded responses.
Approach: They propose a lite-adaptive framework for stylized contextual question answering that scales to millions of official accounts.
Outcome: The proposed framework can serve large volumes of official accounts with minimal overhead while maintaining stylistic diversity.
Neural Topic Modeling via Contextual and Graph Information Fusion (2025.emnlp-main)

Copied to clipboard

Challenge: Existing topic models generate uninformative and incoherent topics that hinder interpretable insights from managing textual data.
Approach: They propose to incorporate contextual and graph information to improve the variational autoencoder framework by combining contextual and bag-of-words information.
Outcome: The proposed framework generates more coherent and diverse topics on three benchmark datasets and achieves strong performance on automatic and manual evaluations.
CogDual: Enhancing Dual Cognition of LLMs via Reinforcement Learning with Implicit Rule-Based Rewards (2025.emnlp-main)

Copied to clipboard

Challenge: Existing approaches to role-playing language models rely on prompt engineering or supervised fine-tuning to emulate character behaviors but neglect the underlying cognitive mechanisms driving these behaviors.
Approach: They propose a novel RPLA adopting a cognize-then-respond reasoning paradigm that leverages dual cognition for more contextually grounded and psychologically coherent responses.
Outcome: The proposed RPLA outperforms baselines and generalizes effectively across diverse role-playing tasks.
End-to-End Open-Domain Question Answering with BERTserini (N19-4)

Copied to clipboard

Challenge: a new open-domain question answering system integrates best practices from IR with a BERT-based reader to identify answers from a large corpus of Wikipedia articles.
Approach: They propose an end-to-end question answering system that integrates BERT with an IR reader.
Outcome: The proposed system improves on a standard benchmark test collection.
PAR2-RAG: Planned Active Retrieval and Reasoning for Multi-Hop Question Answering (2026.acl-industry)

Copied to clipboard

Challenge: Multi-hop question answering is a practical bottleneck in industry applications . large language models (LLMs) fail frequently when evidence coverage is incomplete or reasoning trajectories drift .
Approach: They propose a training-free two-stage framework that separates coverage from commitment . it performs breadth-first anchoring to build a high-recall evidence frontier . compared with IRCoT, it achieves 23.5% higher answer accuracy .
Outcome: The proposed framework outperforms strong baselines in MHQA benchmarks and achieves 23.5% higher answer accuracy and 10.5% NDCG gains in retrieval quality.
Massive End-to-end Speech Recognition Models with Time Reduction (2024.naacl-long)

Copied to clipboard

Challenge: Using the neural architecture of Google’s universal speech model, we reduce the frame rate and speed up training and inference.
Approach: They propose to use the neural architecture of Google’s universal speech model with additional funnel pooling layers to significantly reduce the frame rate and speed up training and inference.
Outcome: The proposed methods work with both connectionist temporal classification (CTC) and RNN-Transducer (RNN-T) and over two domains.
Scaling Laws for Fact Memorization of Large Language Models (2024.findings-emnlp)

Copied to clipboard

Challenge: Fact knowledge memorization is crucial for Large Language Models (LLMs) to generate factual and reliable responses.
Approach: They analyze scaling laws for LLM’s fact knowledge and LLMs’ behaviors of memorizing different types of facts.
Outcome: The proposed model can generalize on unseen facts and its scaling law is similar to general pre-training.
BatonVoice: An Operationalist Framework for Enhancing Controllable Speech Synthesis with Linguistic Intelligence from LLMs (2026.acl-long)

Copied to clipboard

Challenge: Existing approaches often fail to leverage the linguistic intelligence of Large Language Models (LLMs) Existing models lack the ability to follow text instructions for controllable Text-to-Speech (TTS).
Approach: They propose a framework where an LLM acts as a conductor, understanding user instructions and generating a textual plan - explicit vocal features.
Outcome: The proposed model outperforms open- and closed-source models in speech synthesis and achieves zero-shot cross-lingual generalization.
Generate then Select: Open-ended Visual Question Answering Guided by World Knowledge (2023.findings-acl)

Copied to clipboard

Challenge: Open-ended Visual Question Answering (VQA) requires models to reason over visual and natural language inputs using world knowledge.
Approach: They propose a new VQA pipeline that deploys a generate-then-select strategy guided by world knowledge for the first time.
Outcome: The proposed pipeline expands the knowledge coverage from in-domain training data by 4.1% on OK-VQA, without additional computation cost.
UniCreative: Unifying Long-form Logic and Short-form Sparkle via Reference-Free Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Challenge: Existing alignment paradigms for creative writing use static reward signals and supervised data.
Approach: They propose a constraint-aware reward model that synthesizes query-specific criteria to provide fine-grained preference judgments.
Outcome: The proposed framework aligns models with human preferences across content quality and structural paradigms without supervised fine-tuning and ground-truth references.
Logic-of-Thought: Injecting Logic into Contexts for Full Reasoning in Large Language Models (2025.naacl-long)

Copied to clipboard

Challenge: Large Language Models (LLMs) have demonstrated remarkable capabilities across various tasks but their performance in complex logical reasoning tasks remains unsatisfactory.
Approach: They propose a propositional logic prompting method which generates expanded logical information descriptions and utilizes them as an additional augmentation to original contexts.
Outcome: Extensive experiments show that Logic-of-Thought boosts the performance of various prompting methods with a striking margin across five logical reasoning tasks.
InstructMol: Multi-Modal Integration for Building a Versatile and Reliable Molecular Assistant in Drug Discovery (2025.coling-main)

Copied to clipboard

Challenge: Large Language Models (LLMs) can attain professional-level proficiency in specific domains through fine-tuning.
Approach: They propose a multi-modal LLM that aligns molecular structures with natural language via an instruction-tuning approach.
Outcome: InstructMol surpasses existing models and reduces the gap with specialists in drug discovery tasks.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations