Langdrift
Measures how differently an original and its translation land in a listener's head, using a brain-activation model.
Creator / Starside Labs
TypeScript Python Modal Meta FAIR TRIBE v2
Translation that feels right, not just reads right
Langdrift compares audio of an original and its translation and returns a drift score: how far apart the two land cognitively. It catches translations that are technically accurate but emotionally or contextually off.
What it does
- Compares original vs translation audio with Meta FAIR’s TRIBE v2 model (trained on fMRI data)
- Maps brain-activation patterns across the cortex
- Returns a drift score for the cognitive gap between versions
Stack
TypeScript/Node.js CLI with a Python backend on Modal (A10G GPU). Core model: Meta FAIR TRIBE v2 with Wav2Vec-BERT 2.0 encoding and Pearson correlation scoring. Part of the Starside Labs initiative.