Part 1: Core concepts and workflow
Superpowers of generative AI
- Context
- Easy and flexible trainability
- Knowledge of the world
Necessity of new workflows
- All translation is now revision.
- Machine translation post-editing (MTPE) assumes a static workflow imposed from above.
- Neural machine translation (NMT) is a legacy technology.
- Translation memory (TM) was also designed for yesterday’s workflows.
- Today’s CAT tools aren’t ready for the future, either.
- “State of the art” in tools
- Legacy technologies still have a role.
- Generative AI Iterative Translation (GAIT) leverages the superpowers of generative AI at the translator level.
- Keep “slow” and “expensive” in perspective.
- Fast/free NMT is cheap, indeed!
- If you don’t yet see the potential of generative AI in your translation workflow, then stop here!
Workflow fundamentals of Generative AI Iterative Translation (GAIT)
- Set up the anchor prompt
- Get source text batch
- Send to the AI
- Get the AI output
- Edit the AI output
- Update the anchor prompt
Part 2: Advanced concepts and techniques
Context, tokens, and costs
Working in your CAT tool
- What your CAT tool is still good at
- Leveraging context in your CAT tool
- Why pre-translating with MT is a bad idea
- Preparing your files for the GAIT workflow
- Working with formatting in GAIT
Other advanced techniques
- How a project manager could use GAIT as a replacement for substandard MTPE
- Writing and structuring your anchor prompt
- Using GAIT with a lot of TM hits
- Managing tokens to balance quality and cost
- Impact of GAIT on the overall workflow