Two Sessions: US vs. China AI Translation Showdown Unveiled

At the globally watched Two Sessions, effective language communication is paramount. Mainland blogger Yu Yuan Tan Tian made an intriguing discovery on-site: many foreign journalists opted for large language models to translate Chinese idiomatic expressions. But can these models accurately capture the rich cultural nuances of such phrases? Intrigued, Tan Tian launched a unique challenge.

Two Sessions 2025 venue during Yu Yuan Tan Tian’s AI translation challenge

Foreign journalists evaluating AI translations at Two Sessions 2025

During the press conference before the Two Sessions, the spokesperson referenced Sino-European relations with the phrase, “To develop Sino-European ties, the key is confidence—come what may, remain steady as a rock.” Similarly, Foreign Minister Wang Yi at the Munich Security Conference quoted, “Let others be overbearing; the moon still shines over the river.” These expressions embody distinct Chinese wisdom, but can foreigners truly grasp them? Tan Tian found that journalists favored AI models for translation, using both foreign and Chinese ones.

To investigate, Tan Tian consulted a friend experienced in press conference translations and contacted experts from the China Industrial Internet Research Institute specializing in AI model evaluation. Based on their recommendations, 10 top-tier US and Chinese models were selected. To ensure fairness, testing conditions were standardized, using the highest model versions and providing identical tasks and context.

Foreign journalists evaluating AI translations at Two Sessions 2025

Two Sessions 2025 venue during Yu Yuan Tan Tian’s AI translation challenge

Soon, the 10 models delivered varied results. Tan Tian organized the US and Chinese models into two groups for a “group discussion,” yielding their best translations. To determine which version resonated more with foreigners, Tan Tian interviewed six seasoned journalists at the Two Sessions. The competition was fierce, with scores neck-and-neck, but the Chinese group edged out a narrow victory.

Translation outcomes revealed distinct styles: the US group excelled in idiomatic English, often using phrases like “come what may,” but sometimes resulting in verbosity. The Chinese group prioritized conveying the metaphors’ deeper meanings, using vivid, concise language while preserving imagery.

As Tan Tian continued validating the results, a Kazakh journalist shared his perspective, expressing a desire for his country to have its own open-source AI model to amplify its voice. He noted that he paid for AI translation software in Indonesia but had never used it. This underscored Tan Tian’s realization of the privilege of having world-class domestic models, enabling China to narrate its story through its own technology and data.

At the Foreign Minister’s press conference, Wang Yi stated, “Green hills cannot obstruct; the river flows eastward nonetheless.” Chinese models, blending historical context with modern relevance, produced a seamless translation in just 25 seconds. In an era where nations yearn to be heard, China’s domestic AI models are advancing, shouldering the mission of conveying the nation’s voice, and are poised to play an even greater role in global communication.