BabelFish Net: Hypothesis Test Protocol Summary
The BabelFish Net platform was created as a live, collaborative experiment to test a core linguistic and social hypothesis in a constrained digital environment.
1. What was the Hypothesis?
Hypothesis: Can social communication succeed without the reliance on native fluency?
The goal was to test the viability of a social network where individuals are deliberately forced to communicate primarily in a non-native language (L2) while having their thoughts immediately interpretable by others in their own native language (L1). This tests if the ease of translation offsets the difficulty of expression and maintains a functioning community.
2. How was the Hypothesis Tested?
The test was implemented by establishing and enforcing two core constraints for every user:
- L1 (Native Language): The language the user is most fluent in, which serves as the translation target.
- L2 (Posting Language): A designated language the user must attempt to post in.
Constraint Enforcement
- Gemini-Powered Fluency Check: Before transmission, the system uses the Gemini API to detect the language of the post. If the Detected Language matches the user's declared L1, the system flags a Protocol Violation and asks the user to confirm the post.
- Real-Time Assistance: Posts in the feed include an LLM-driven Translation feature, converting the L2 post back to the viewer's L1, complete with contextual explanations.
- Grammar Feedback: Users can request a Grammar and Fluency Check on their L2 drafts before posting, receiving corrections and explanations from the model.
3. What Worked? What Didn’t?
The LLM-based constraint enforcement for the protocol was highly successful, effectively identifying and flagging L1 intrusions, and making the user aware of hypothesis violations. Furthermore, the post translation feature was robust and reliable, offering both the direct translation and crucial contextual explanations to minimize miscommunication. Functionally, real-time data synchronization via Firestore enabled posts to appear instantly in the multi-user feed.
However, the major technical challenge encountered was a critical usability bug related to Input Focus Loss. The continuous re-rendering of the posting form (necessary to update the character count and grammar feedback) caused the user to lose their text and cursor position on every keystroke. This required implementing a complex, specific client-side fix to manually save and restore the cursor state. A minor challenge was that the grammar check feature, while functional, sometimes generated verbose feedback, which could potentially slow down the user's drafting process.
Log in or sign up for Devpost to join the conversation.