Quality prediction - the next evolution of human-quality translation
- Presentation
- GALA
- 15. November
- 11:30 - 12:15 AM (CET)
- C7.2
- finished
- No presentation slides uploaded
-
Adam Bittlingmayer
- ModelFront
-
Shamus Dermody
Contents
In the three decades since the internet started to change the way billions of people live and work, only two technologies had significant impact on the efficiency of human-quality translation: the translation memory and machine translation.
Now the next evolution is here: machine translation quality prediction (MTQP).
Global companies and institutions are using quality prediction to translate up to 5x more efficiently - at the same human quality - by accurately deciding which segments require human post-editing, and which don't.
Like GPT, quality prediction is built with Transformer-based multilingual LLMs (large language models) first researched in academia and companies like Microsoft and Google, but it took startups to make it accessible as a cloud product, available in translation management systems and trusted by enterprise teams translating millions or billions of words a year.
We'll cover the problem, what quality prediction is, the use cases and how to get starte
Takeaways
1. The problem: Why human-quality translation - even machine translation post-editing - is so inefficient
2. Quality prediction: What it is, and how it is different than quality evaluation like BLEU
3. Use cases: The highest value use cases, when they're a fit, ROI
4. How to get started: Build or buy, choosing a provider, customization, rollout