CROSSWIRE explores consequential ambiguity in multilingual conversations. Before translation, it asks a model for possible readings and shows those that could lead to different actions; the speaker chooses a meaning before it is sent.
The Node.js prototype integrates NVIDIA Nemotron through Nebius Token Factory. It analyzes ambiguity, translates candidate interpretations and back-translates them; a deterministic word-overlap check compares the round trips. The recipient can then indicate how they understood the message.
The approach can miss ambiguity or flag paraphrase noise. The prototype uses an in-memory room, and English and Spanish examples; it is not a substitute for professional interpretation. Further evaluation with bilingual participants and real use cases is needed.
Log in or sign up for Devpost to join the conversation.