面向通用人工智能的形式语用学：理论联系、实践进路与研究价值Formal Pragmatics for Artificial General Intelligence: Theoretical Connections, Practical Approaches, and Research Significance
向明友,刘欣雅
XIANG Mingyou,LIU Xinya
摘要(Abstract)：
大语言模型在复杂语用推理环节暴露出的缺陷促使学界的关注重心由“大数据+弱规则”的研究路径逐渐转向“小数据+强规则”的研究路径，而该转向又迫切需要形式语用学成果的补位助力。本文针对大语言模型的语用推理短板，提出以形式语用学对语用推理的形式化建模为应对策略，探寻二者的理论关联；再以提升大语言模型语用推理能力的形式语用学实践进路为着力点，从语用推理过程的形式化、语用推理结果的可解释性和语用推理模型的可计算性三个层面，勾勒形式语用学赋能大语言模型的融合发展路径，彰显形式语用学对以大语言模型为代表的人工智能研发的价值和意义，以期助力大语言模型朝着能理解人类心智的通用人工智能的方向迈进。
The limitations of large language models(LLMs) in complex pragmatic inference tasks have prompted a shift of research focus from the big data+weak rules" toward the "small data+strong rules" paradigm,which calls for support from formal pragmatics.In response to LLMs' pragmatic inferential limitations,this study proposes formalizing pragmatic inference in formal pragmatics as a strategy to explore the theoretical connections between LLMs and formal pragmatics.To explore practical approaches in formal pragmatics to improving LLMs' pragmatic inferential capabilities,this study then outlines an integrative pathway to empowering LLMs with formal pragmatics based on three key dimensions:the formalization of pragmatic inference processes,the explainability of pragmatic inference outcomes,and the computability of the pragmatic inference model.This interdisciplinary integration highlights the value and significance of formal pragmatics for the development of artificial intelligence exemplified by LLMs and offers a promising path toward artificial general intelligence,which aims to understand the human mind.
关键词(KeyWords)：
形式语用学;语用推理;大语言模型;人工智能
formal pragmatics;pragmatic inference;large language models;artificial intelligence
基金项目(Foundation): 国家社科基金项目“面向人工智能的言语行为博弈机制研究”（21BYY112）的阶段性成果
作者(Author):
向明友,刘欣雅
XIANG Mingyou,LIU Xinya
DOI: 10.19923/j.cnki.fltr.2026.04.004
参考文献(References)：
- 冯志伟等，2024，“辛顿·乔姆斯基·语言学发展”多人谈[J]，《语言战略研究》（6）:5–17。
- 蒋严，2002，论语用推理的逻辑属性——形式语用学初探[J]，《外国语》（3）:18–29。
- 李佐文、梁国杰，2022，语言智能学科的内涵与建设路径[J]，《外语电化教学》（5）:88–93。
- 陆俭明，2025，大语言模型的“语言”跟自然语言性质迥然不同[J]，《语言战略研究》（1）:1。
- 毛眺源，2022，《语用寓义推理形式化研究》[M]。北京：科学出版社。
- 辛顿，2024，杰弗里·辛顿接受尤利西斯奖章时发表的获奖感言[J]，陈国华译，《当代语言学》（4）:489–495。
- 杨延宁，2024，语言学与人工智能：下一个十年[J]，《外语教学理论与实践》（3）:1–9。
- 张钹、朱军、苏航，2020，迈向第三代人工智能[J]，《中国科学：信息科学》（9）:1281–1302。
- Bhuyan, B., et al. 2024. Neuro-symbolic artificial intelligence:A survey[J]. Neural Computingand Applications 36(21):12809–12844.
- Collins, K., et al. 2024. Building machines that learn and think with people[J]. Nature HumanBehaviour 8(10):1851–1863.
- Elder, C.&M. Haugh. 2024. The role of inference and inferencing in pragmatic models ofcommunication[J]. Journal of Pragmatics 229:71–76.
- Franke, M.&G. J??ger. 2016. Probabilistic pragmatics, or why Bayes’ rule is probably importantfor pragmatics[J]. Zeitschrift für Sprachwissenschaft 35(1):3–44.
- Goodman, N.&M. Frank. 2016. Pragmatic language interpretation as probabilistic inference[J].Trends in Cognitive Sciences 20(11):818–829.
- Grice, H. 1975. Logic and conversation[A]. In P. Cole&J. Morgan(eds.), Syntax and Semantics(Vol. 3):Speech Acts[C]. New York:Academic Press. 41–58.
- Grundy, P. 2019. Doing Pragmatics(4th edition)[M]. London:Routledge.
- Hansen, M.&M. Terkourafi. 2023. We need to talk about Hearer’s Meaning![J]. Journal ofPragmatics 208:99–114.
- Jones, C., et al. 2025. People cannot distinguish GPT-4 from a human in a Turing test[A]. InProceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency[C].New York:Association for Computing Machinery. 1615–1639.
- LeCun, Y., Y. Bengio&G. Hinton. 2015. Deep learning[J]. Nature 521(7553):436–444.
- Tsvilodub, P., R. Hawkins&M. Franke. 2025. Integrating neural and symbolic componentsin a model of pragmatic question-answering[A]. In C. Anderson, F. Mailhot&G. Prasad(eds.), Proceedings of the Society for Computation in Linguistics 2025[C]. Kerrville, TX:Association for Computational Linguistics. 2–17.
- Wei, Jason, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models[A]. In S. Koyejo et al.(eds.), Proceedings of the 36th International Conference on NeuralInformation Processing Systems[C]. Red Hook, NY:Curran Associates. 24824–24837.