Requirements | Providers | Features | Integration | FAQ | Events | More reading
What is translation quality prediction?
This new category of AI predicts if a translation is good or bad, to help humans translate more efficiently.
Quality prediction can provide scores for new content, including machine translations. A quality prediction engine takes a source segment and a target segment, and it returns a score from 0 to 100.
Translation management systems integrate quality prediction to make post-editing workflows up to 5x more efficient.
Translation quality prediction or machine translation quality prediction (MTQP) is also known machine translation quality estimation (MTQE or QE) or machine translation confidence scoring.
For translation buyers with high-volume or high-speed workflows, the highest-value use cases of quality prediction create efficiency gains of up to 5x in production:
These highest-value use cases require segment-level accuracy. They are a great fit for high-volume or high-speed content types like technical documentation, product titles and descriptions or supplier-generated content, user-generated reviews or customer support emails and chats.
There are also offline use cases of quality prediction, like comparing machine translation engines or filtering translation memories.
“the missing link for machine translation adoption”
— João Graça, CTO and co-founder, Unbabel
“the next evolution of human-quality translation”
— Adam Bittlingmayer, CEO and co-founder, ModelFront
For translation buyers to get both efficiency and human quality with their current setup, there are key requirements that a quality prediction provider or feature needs to fulfill.
The first provider of a translation quality prediction API product launched in 2019.
In 2020, Gartner started covering ModelFront and quality prediction, for example in its market guide AI-Enabled Translation Services.
Besides dedicated providers, there are a few traditional products or services that include a machine translation quality prediction feature.
Translation management systems with a translation quality prediction feature
Human translation services with a machine translation quality prediction feature
Machine translation providers with a machine translation quality prediction feature
Translation management systems are also adding official integrations with machine translation quality prediction APIs.
Buyers can also integrate machine translation quality prediction providers into translation management systems by using APIs.
Quality evaluation metrics, like BLEU, require human reference translations and do not provide segment-level scores.
So they cannot be used in production, before human post-editing or translation. Quality evaluation is used for comparing engines.
Yes, quality prediction can be applied to human translation.
For example, it can be applied to human translated segments before the review step in a translation workflow.
To be accurate, this requires quality prediction that is customized on the workflow data and engine-independent.
On Machine Translate, open resources for machine translation
Articles
Books
© qualityprediction.org 2023