How to Build Transparent AI Language Support So Generations Can Talk Together

Why Read This from an HCI Perspective

This article is meaningful for HCI/UX practitioners and researchers because it frames AI translation not as a mere ‘tool for swapping words accurately,’ but as an interaction mechanism that reduces misunderstandings and helps sustain conversation between people. In particular, by comparing how different presentation styles—showing only the translation result versus showing both the original text and the translation—affect trust, conversational flow, and feelings of closeness, it prompts readers to think about what information should actually be disclosed in real products. It also clearly shows that, more than simply adding AI features, deciding ‘how much to show’ may be the more important design question.

CIT's Commentary

The core of this study is less about whether an LLM can ‘translate well’ and more about how much users can verify and intervene in that transformation. An interesting finding is that black-box translation can actually disrupt the flow of conversation, whereas a transparent approach that shows the original and the translation together improved conversation quality. In safety-critical systems, if users cannot see the system’s state, they may notice incorrect decisions only too late—and the same issue appears here as well. Even a feature that looks like ‘just translating a small message’ is ultimately about ‘who interprets meaning and takes responsibility.’ Therefore, the interface should not be a window that speaks for the system’s output; it should be a collaborative space where users can check meaning together. In contexts like domestic messenger environments—where speed and familiarity matter—there also seems to be an important balance: even if transparency is added, the design should be staged so the screen does not become heavy or cumbersome.

Questions to Consider While Reading

• Q.If transparent translation is always the better choice, how can we reduce the cognitive burden that users might experience when they see the original and the translation together?
• Q.In sensitive contexts like family conversations, when AI changes meaning, what information should be shown and what should be hidden?
• Q.If we apply this kind of translation interface to Korean messenger environments like Naver or Kakao, would we encounter interaction issues that differ from global research findings?