Restricted archive / case files
Four case files. Same question each time: when language passes through a system, what survives, what gets cleaned up, and who pays for the loss?
A person gives testimony through an interpreter. The English rendering becomes the record. The record decides credibility. Procurement, qualifications, domain knowledge, and transcripts decide whether the person sounds coherent or impossible.
The interpreter was the variable in every case we examined.
When the machine gets the words right but loses the person.
Accurate — but not personal, and not connective.
Field material
→ Refugee EngagementAI learns from an uneven internet. Languages with less data become harder to model, less accurate, more expensive, and easier to ignore. Automation keeps the hierarchy.
The token tax is a language tax.
Students read translations and think they are reading the text. They are reading someone else's interpretation. Most of the time, nobody tells them.
We are teaching the interpretation as if it were the original.