What’s New at Unbabel

March 15, 2023

It’s been a time of great change and disruption in the world of artificial intelligence, while businesses are grappling with how to drive down the costs of their operations. 

Unbabel is focused on how we can help businesses manage the growing volume of content coming from generative AI by investing in technologies that improve the quality of translations. At the same time, we’re looking at how to drive efficiency in workflows to help organizations better scale. 

Announcing the LangOps Self-Service Trial

Currently, Unbabel’s customers receive the best results by leveraging the full potential of our platform through the support of the Unbabel LangOps Team. But we want to make the same experience available to more users — those who want to dive into what it means to be a LangOps expert, and experiment freely with our tools.

We’ve built the toolkit, and we’re now making it accessible to you through a self-service free trial.

Users can sign-up right now for our beta launch of the self-service LangOps platform, and within minutes, they can request translations combining human and AI, integrate with Zendesk for seamless translation, or even leverage advanced AI components such as our Quality Estimation technology.

To learn more about getting started, read this article.

The addition of new file formats

Unbabel can now translate your content in the most widely used file formats: PowerPoint, Excel, Word, XML, XLS, and CSV files, plus many more formats into multiple languages using our high-quality translation flows, without disrupting the formatting, layout, or structure of your document.

Having the ability to seamlessly submit and retrieve content for translation is paramount for any business seeking to scale. Content is generated across multiple systems and, after translation, needs to be input back while maintaining context, format, and quality. So businesses need a solution that is flexible and tailored specifically to each of their needs.

To read the full article, click here.

Significant upgrade to COMET

Since it was first released in 2020, the COMET framework has had a significant impact on the machine translation field by providing more accurate evaluation of machine translation, while helping the community at large move away from metrics that didn’t match up with human evaluation.

As part of our ongoing innovation in MT evaluation, we released a new and improved metric trained with more data covering more language pairs resulting in better performance across domains and language pairs. We used a large language model developed by Meta and then fine-tuned it with actual human judgments to produce MT quality assessments. COMET is unique in utilizing source sentences in the evaluation process, which is helpful to detect errors that simply cannot be detected by comparing with a reference translation.

This release was a big contributing factor in Unbabel’s recent win in the WMT 2022 Quality Estimation shared task.

For more, check out our blog on the COMET release.

Recognition of key terms in more languages

Over the last year, Unbabel has supercharged Named Entity Recognition (NER), the neural AI technique for identifying key words. It has multiple applications: Ensuring translation quality, protecting privacy, and making linguistic formatting customizable. In Q2 of 2022, we announced new functionality to respect entity style rules that customers define themselves (including date, number, and measurement formatting), and in Q4 of 2022, we shared that we now train industry-specific NER models, to boost accuracy and privacy even further for new domains, content-types, and customers.

We now have neural Named Entity Recognition (NER) available across 4 new source languages: French (FR), Brazilian Portuguese (pt-BR), LATAM Spanish (es-LATAM), and simplified Chinese (zh-CN). This is on top of existing neural models for English (EN), German (DE), and Polish (PL).

Advanced Translation Memories

Translation Memories (TMs) create a store of the same translation between a source and a target. The TM concept has been around for years because it���s a great tool for speeding up translation and delivering top-quality translations. However, we’re going further, with In-Context Exact (ICE) TM Matching. ICE TMs let us identify and utilize the context around a segment (i.e. the text before and after it) when we match a TM usually.

ICE TM Matching boosts quality by making sure translations are consistent between jobs. They’re another source of high-quality, pre-approved translations, but also mean that customers can resend documents for translation, and know only the updated segments will be re-translated.

If you have any questions about what we’ve been working on and how it can help your business, contact us here:

About the Author

Profile Photo of Phill Brougham
Phill Brougham

Director of Product Marketing at Unbabel, Phill Brougham spent the last five years working for SaaS businesses focused on applying artificial intelligence to solving real-world business and productivity problems. Throughout his roles, Phill’s focus has been on translating technological capability into clear, understandable value.