It was a busy year-end for the product team. We were working diligently behind the scenes to set up success in 2023.
The TL;DR: We released new Named Entity Recognition (NER) Models and expanded the functionalities of Polyglot, our Computer-Assisted Translation (CAT) Tool. Both enhancements remove manual steps in the translation process, making translations faster and more accurate.
Expansion into new domains
In Q4 we introduced additional versions of domain-adapted Named Entity Recognition (NER) with the release of two new models. As iterations of Unbabel’s existing Entity Processing catalog, the new models detect and adapt customer entities for customer-specific and marketing industry domains. This lets us detect and input key customer and industry-specific terms into things like press releases, support articles, catalogs, and more, raising the accuracy of the final translation, and ensuring greater privacy should translations require human post-editing.
Customers benefit from an 87% improvement in Unbabel’s ability to detect errors in entities, leading to greater translation accuracy and more secure translations.
Increased editor productivity
Our global community of editors plays a key role in refining translations and delivering the quality customers demand. Last quarter, we released two new CAT Tool features to enhance editor productivity.
Not all glossary terms are like-for-like translations — sometimes an editor requires extra context for ambiguous terms. Take the word “spike,” for example: It can refer to either a spike in productivity or a cactus spike. To ensure the editor understands the intended meaning for accurate translation, additional context must be at their fingertips.
Our solution flags and detects ambiguous glossary terms and surfaces additional context right in the editor’s workflow. This ensures the editors are provided the right terms, at the right time, without complicated or time-consuming back-and-forth.
Formatting tags placed on rich text documents indicate things like font, font size, spacing, hyperlinks, image spacing, and anything else relevant to the structure of a document, to ensure the translated document mirrors the format of the original.
In the translation process, a document’s character count and spacing often fluctuate depending on the target language. Take a longer language like French for instance: When translated from English the French copy is usually around 15-20% longer. That said, formatting tags must remain true to their original placement. We wouldn’t want the header of a customer’s article to be the last sentence of the conclusion, would we?
Here’s where editors come in: Editors can now realign, reposition, and edit tags to ensure format accuracy. They have the option to batch edit identical formatting tags and, because no Unbabel feature would ever be complete without some AI automation, editors benefit from the automatic flagging of any changes that might have thrown tags out of sync.
With these two releases, we’ve been able to help editors focus on the most challenging translations while allowing them to act more independently so translations are on-brand and on time.
If you have any questions about what we’ve been working on and how it can help your business, contact us here: