Machine Translation in the Wild: User Reaction to Xiaohongshu's Built-In Translation Feature

A new research paper titled 'Machine Translation in the Wild: User Reaction to Xiaohongshu's Built-In Translation Feature' provides a detailed look at how users interacted with the social media platform's AI translation tool after its January 2025 launch. Authored by Sui He and available on arXiv, the study analyzed 6,723 user comments collected from 11 official posts promoting the feature. Using a combination of sentiment analysis and thematic analysis, the research found that overall reactions were positive, particularly for the function's ability to translate posts and comments, bridging linguistic gaps on the platform.

However, the study also revealed user concerns regarding functionality, accessibility, and translation accuracy. Beyond just giving feedback, users actively 'stress-tested' the AI system with a wide variety of inputs. This included not only standard English and Chinese words and phrases but also abbreviations in pinyin, internet slang, emoji, kaomoji (Japanese emoticons), and other coded text forms. This real-world experimentation goes far beyond the controlled environments typically used to train and benchmark machine translation models.

The findings underscore a critical gap between how translation AI is developed and how it's actually used in messy, informal social media contexts. The paper concludes by emphasizing the importance of closer, interdisciplinary collaboration among computer scientists, translation scholars, and platform designers. This partnership is essential to better understand user behavior and to iteratively improve translation technologies so they function effectively within real-world, dynamic communicative environments, not just in lab settings.