Facilitator guide: Translation Check Lab
This is a practice lab and human-review training environment. It is not a privacy tool, an AI fact checker or a safe translator. The software never shows a green "safe", "verified" or "approved" badge.
French version: prepare a parallel briefing using the kit's French materials. Interface strings are available in FR via the lab language switch.
Learning objective
Participants practise a structured check of a short humanitarian broadcast message and its translation before it would be sent to affected people. They see how machine translation can change meaning, why back-translation can hide errors, and why a qualified bilingual reviewer is required.
Maps to kit Session 2
This activity supports Verified work by role (Session 2) and connects to the kit's synthetic evacuation SMS translation case. It reinforces source checking, negation checks, and human approval before external use.
Duration
20 to 30 minutes in a live remote session (can extend to 40 minutes with full practice set).
Activity plan
| Time | Activity | Notes |
|---|---|---|
| 0–3 min | Introduce the lab link | One sentence: MT can draft; humans must verify before sending. SMS is not confidential. Share link in chat. |
| 3–12 min | Practice items 1–3 (pairs) | Each pair reviews one EN→FR item. Mark errors and safe/unsafe. Submit and discuss answer key. |
| 12–18 min | Practice item 15 (back-translation trap) | Show how back-translation can match the source while the French is wrong. |
| 18–25 min | Practice items 8–10 (FR→EN) | Rotate pairs. Focus on negation and modality. |
| 25–28 min | Optional: "Check my own text" | Complete the human review record before copying final text. Pattern checks only if bandwidth is low. |
| 28–30 min | Debrief | See questions below. |
Debrief questions
- Which error type was easiest to miss? Why?
- When would you require a qualified translator instead of a bilingual staff member?
- What is one check you will add to your team's workflow this week?
- How does practice item 15 show that back-translation is not proof?
- How does this connect to the evacuation SMS case in Session 2?
Running without an AI account
The practice mode uses pre-written illustrative translations. No login or generative AI tool is required. Pattern checks run entirely in the browser.
Low bandwidth
- Share the lab URL before the session so participants can open it early.
- Skip optional on-device model download (~100 MB per language pair).
- Use practice mode only; paste text from the kit case study.
- Co-host can screen-share one person's results if individual loading fails.
Accessibility notes
- Lab supports keyboard navigation, visible focus, and screen reader announcements for results.
- Offer to read practice items aloud for participants who prefer audio.
- Allow extra time for reading in a second language.
- Consider audio or visual channels alongside SMS for people with low literacy.
- See kit technology accessibility checklist.