1. Tailored Translation (Easy)

Translate subtitles for the TED video talk. Specify requirements to achieve a performance close to the human translation provided. Protect the time stamps from translation, and enforce Netflix timed text style guide character limit of 42 characters. Materials:

Subtitles

Original video: Amy Cuddy – Your Body Language May Shape Who You Are

2. Terminology (Easy)

3. Error labeling / Translation Quality Estimation (Intermediate)

Write a prompt that measures the quality of translation, identifies and labels errors in it. Measure quality as “Good, Passable, and Bad” or use a numerical scale, for example, from 1 to 10. We recommend using MQM error typology if you are familiar with it, or simply error severities (Minor, Major, Critical) if not. 

– Task sample – 10 segments 
– DEMETR dataset – if you need more sentences 
– DEMETR sample export to Excel
– MQM error typology

Advanced: use Chain of Thought Prompting to improve accuracy, and split error annotation from quality estimation.

4. Screenshot testing – Vision (Easy)

Image Folder

5. RAG: Multilingual content generation (Complex)

B) Research assistant – write a prompt that provides answers based on the website data (Clear Global)

C) Repurpose content – reuse existing content to creates new articles (Clear Global)

6. More prompt ideas

This workshop includes one hour for self-guided work on Friday, and prompt presentations by up to 10 teams. We encourage you to design a practical prompt using some of the techniques learned and to share it with the other participants. Consider one of the ideas below or design something completely different – there is no limit to what you can explore!

6.1 Translating with context (full subtitle text, image, game character gender)

6.2 Cultural and geopolitical issue detector – try it on news from the DEMETR dataset

6.3 A translation editor prompt that fixes the errors identified at the LQA step

6.4 An SEO web page rewriting tool. It will inject SEO keywords from your RAG storage

6.5 A writing assistant that improves text readability and makes it inclusive (in a language of your choice)

6.6 A translation memory cleaner that fixes spelling errors and outdated terminology using RAG 

6.7 A coding assistant that explains internationalization errors to your developers.

Good luck!