Skip to main content

What’s New at Unbabel

The new year has brought in a host of changes, most notably with Unbabel's acquisition of Lingo24. Our global team has expanded, and we’ve all been hard at work pushing our Language Operations (LangOps) vision forward. This has inspired our product development team to continue to innovate, build creative solutions for more use cases, and ultimately, improve the customer service industry. 

This quarter, we’ve built a ton of improvements and features into the Unbabel platform that would help you take your customer support to the next level. As always, we’ve prioritized and invested in quality, visibility, and agent experience. 

More visibility and control

The Unbabel Portal is your hub for visibility and control of how you use Unbabel. To improve the user experience for our customers, we’ve made some important design updates.

  • Reviewing your bill is now even simpler: You can easily track your costs to determine how your usage translates to your plan.

  • Quality Reporting insights are surfaced on the homepage to ensure that everything is performing well.

  • Prior On-demand requests are viewable on the Portal homepage, making it easier to review the status of orders and retrieve translations.

Simplified and enhanced agent experience 

Unbabel integrations define the agent experience — the simpler it is for the agent, the faster the resolution and the better the experience for the end customer. With that in mind, we’ve made a few improvements to simplify translation workflows.

Oracle 

We’ve launched a new integration with Oracle B2C Service. Now you can request the app in Oracle Cloud Marketplace and embed Unbabel directly in Oracle's Browser User Interface (BUI). From there, Unbabel acts as an invisible layer between the Oracle B2C Service and end customers, automatically providing translations directly within agents’ existing workflows.

Zendesk

  • We’re rolling out an Unbabel Translate button on Zendesk Support to make it straightforward for agents to initiate translations. Say goodbye to the infamous #unbabel. 

  • The FAQ translation process will get that much easier by separating category and translation titles.

Privacy protection and error reduction

We deploy Natural Language Processing (NLP) in various parts of our translation pipelines for privacy, linguistic quality, and translation efficiency for our global community of editors. To ensure the privacy of your customers, we use an NLP tool called Named Entity Recognition (NER) to pick out personally identifiable information and hide it. Without NER, machine translation tools can change the key entities in a sentence, and communicate something incorrectly. We’ve made some changes to boost the quality of the customer interaction and help to avoid machine translation errors. 

  • We’ve reduced errors by 33% in personal names, user names, products, services, organizations, dates, addresses, cities, and countries

  • We’ve now added NER categories for the date, address, city, and country.

  • Our NLP team has also worked hard at fixing the final document quality for HTML files. Not only have we made improvements to make sure the HTML output of our translated files looks professional, but we’ve also put in place validation tests to make sure we catch outliers. 

Enhanced translation performance

Our global community of Editors is one of the key pillars of Unbabel quality. They ensure that our translation quality is high with careful reviews and provide the ongoing scoring of translations to refine and improve performance.  

  • The Unbabel Community team now automatically detect and mitigate translation peaks. When there’s a long backlog of translations, we can automatically notify our Community Editors that there’s a spike in tasks, and we can also lower the Quality Estimation threshold to automate more messages. This helps us reduce turnaround time and lower the need to monitor volume manually. This is for Tier 1 Language Pairs only. 

  • We’ve also made it possible for your LangOps Manager, who manages your quality performance, to proactively select the proficiency level of the Editor to make sure we assign the right level to the right work, depending on the complexity of your customer service conversations. This takes place in the Portal. 

Boost in quality and speed

Our AI Technology team has always strived to improve model quality and proficiency. Based on advancements in model retraining logic, we deliver quality improvements across all language pairs. 

  • For our most widely used languages (Spanish, French, and German), we were able to meaningfully improve the MQM quality score. (You can read more about Unbabel translation quality here.) 

  • We also made breakthroughs in quality for two of our trickier languages, Polish and Russian. Based on changes to translation workflows, we increased our CUA green score (Good or Excellent translations) substantially.

We’ve been busy! For any questions, contact us here


 

About the Author

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.

Profile Photo of Phill Brougham