Yi Fang, Wenjie Wang, Mingfeng Xue, Boyi Deng, Fengli Xu, Dayiheng Liu, Fuli Feng. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Chengpeng Li, Mingfeng Xue, Zhenru Zhang, Jiaxi Yang, Beichen Zhang, Bowen Yu, Binyuan Hui, Junyang Lin, Xiang Wang, Dayiheng Liu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Large reasoning models (LRMs) like OpenAI-o1 have shown impressive capabilities in natural language reasoning. However, these models frequently demonstrate inefficiencies or inaccuracies when tackling complex mathematical operations. While integrating computational tools such as Code Interpreters (CIs) offers a promising solution, it introduces a critical challenge: a conflict between the model's internal, probabilistic reasoning and the external, deterministic knowledge provided by the CI, which often leads models to unproductive deliberation. To overcome this, we introduce CoRT (Code-Optimized Reasoning Training), a post-training framework designed to teach LRMs to effectively utilize CIs. We propose **Hint-Engineering**, a new data synthesis strategy that strategically injects diverse hints at optimal points within reasoning paths. This approach generates high-quality, code-integrated reasoning data specifically tailored to optimize LRM-CI interaction. Using this method, we have synthesized 30 high-quality samples to post-train models ranging from 1.5B to 32B parameters through supervised fine-tuning. CoRT further refines the multi-round interleaving of external CI usage and internal thinking by employing rejection sampling and reinforcement learning. Our experimental evaluations demonstrate CoRT's effectiveness, yielding absolute improvements of 4\% and 8\% on DeepSeek-R1-Distill-Qwen-32B and DeepSeek-R1-Distill-Qwen-1.5B, respectively, across five challenging mathematical reasoning datasets. Moreover, CoRT significantly enhances efficiency, reducing token usage by approximately 30\% for the 32B model and 50\% for the 1.5B model compared to pure natural language reasoning baselines. The models and code are available at: [this url](https://github.com/ChengpengLi1003/CoRT).

Guanting Dong, Hongyi Yuan, Keming Lu, Chengpeng Li, Mingfeng Xue, Dayiheng Liu, Wei Wang, Zheng Yuan, Chang Zhou, Jingren Zhou. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Kexin Yang, Dayiheng Liu, Wenqiang Lei, Baosong Yang, Mingfeng Xue, Boxing Chen, Jun Xie. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Mingfeng Xue, Dayiheng Liu, Wenqiang Lei, Jie Fu, Jian Lan, Mei Li, Baosong Yang, Jun Xie, Yidan Zhang, Dezhong Peng, Jiancheng Lv. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

The deep neural network (DNN) has become prevailing for paraphrase generation (PG), a widely studied task in natural language processing. Even though the DNN models have shown impressive performances, they are still far from generating desired paraphrases that require sufficient diversity in wording while keeping the correct semantics. Meanwhile, the large DNN models suffer from significant cost of training, the difficulty of reproducibility, and the black-box training process, and thus make it hard to generalize and implement a novel model in other works. Therefore, we propose a simple but effective alternative: a post-hoc method called RulePG, which is based on syntactic rules for PG. It has the advantages of improving wording diversities while preserving semantics, and is designed post-hoc manner so that it is model-agnostic and would not require extra training when utilized. Moreover, instructed by the rules, the generation process is more interpretable with explicit control of sentence forms. To gain and make full use of rules, we construct SRPG (syntactic rules for paraphrase generation), the first syntactic rule set for PG. We novelly build semantic representations for SRPG and endow the RulePG with the capacity of semantic understanding. Automatic evaluation on various metrics and human evaluations demonstrate the effectiveness of our method and show that the RulePG reaches new state-of-the-art performances.