PILLAR Group

The PILLAR Group is a natural language processing research lab at Peking University. We study how large language models understand, reason about, and act on the world — with current interests spanning large language models, reasoning and planning, LLM-based agents, and multilingual NLP. Our goal is to build language technologies that are capable, reliable, and useful across languages and domains.

We publish at venues such as ACL, EMNLP, and NeurIPS, and we maintain active collaborations with both academia and industry. If you are interested in joining the group as a PhD student, master student, or research intern, please see our join us page.

News

Sep 05, 2026	Our new lab website is online! Content is still being populated — stay tuned.
Sep 01, 2025	We are recruiting PhD and master students for Fall 2026, as well as undergraduate research interns. See the join us page for details.
May 15, 2025	Two papers from the group were accepted to ACL 2025! Congratulations to all the authors.
Mar 10, 2025	Our paper on cross-lingual chain-of-thought reasoning was presented at EMNLP 2024 in Miami. The code and data are available on our GitHub.

Latest Posts

Sep 01, 2025	Welcome to Lab Notes
Aug 20, 2025	A Practical Checklist for Evaluating LLM Reasoning

Selected Publications

ACL
Self-Consistent Multi-Agent Debate Improves Mathematical Reasoning in Large Language Models

Wei Zhang, Na Li, Jiawei Chen, and 1 more author

In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL), 2025

Abs Bib PDF Code

We study whether multiple LLM agents debating candidate solutions can improve mathematical reasoning. Placeholder abstract — replace with the real one.
@inproceedings{zhang2025multiagent, title = {Self-Consistent Multi-Agent Debate Improves Mathematical Reasoning in Large Language Models}, author = {Zhang, Wei and Li, Na and Chen, Jiawei and Wang, Fang}, booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL)}, year = {2025}, }
EMNLP
Cross-Lingual Transfer of Chain-of-Thought Reasoning in Low-Resource Languages

Jiawei Chen, Xinyu Liu, Wei Zhang, and 1 more author

In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2024

Abs Bib PDF Code

A placeholder abstract about cross-lingual chain-of-thought transfer. Replace with the real abstract.
@inproceedings{chen2024crosslingual, title = {Cross-Lingual Transfer of Chain-of-Thought Reasoning in Low-Resource Languages}, author = {Chen, Jiawei and Liu, Xinyu and Zhang, Wei and Wang, Fang}, booktitle = {Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP)}, year = {2024}, }
NeurIPS
Scalable Preference Optimization for Helpful and Harmless Dialogue Agents

Yiming Zhao, Na Li, and Fang Wang

In Advances in Neural Information Processing Systems 37 (NeurIPS), 2024

Abs Bib PDF Code

A placeholder abstract about preference optimization for dialogue agents. Replace with the real abstract.
@inproceedings{zhao2024preference, title = {Scalable Preference Optimization for Helpful and Harmless Dialogue Agents}, author = {Zhao, Yiming and Li, Na and Wang, Fang}, booktitle = {Advances in Neural Information Processing Systems 37 (NeurIPS)}, year = {2024}, }