Sony Europe & Pangeanic Case Study (TAUS 2010): Pioneering Custom AI Translation

3:06

TAUS USER CONFERENCE 2010

Language Business Innovation – Shared Opportunities Governor Hotel, Portland (OR), USA / October 3 – 6, 2010

Editor's Note (2026 Update): This article documents our milestone presentation at TAUS 2010 alongside Sony Europe. While the industry has evolved from Statistical Machine Translation (SMT) to the Generative & Neural AI we develop today, this case study proves a timeless truth: clean data and customization are the keys to enterprise quality.

The principles of data hygiene we applied for Sony in 2010 are the foundation of our current Deep Adaptive AI technology, now recognized by Gartner in their 2023, 2024, and 2025 reports. See how we apply these principles today with Deep Adaptive AI >

Moving on with MT: Building open-source MT with your vendor and serving it

Salomé López-Lavado (Sony Europe) and Elia Yuste (Pangeanic/PangeaMT) presented a case study demonstrating the best practices of a long-term collaboration between Sony and Pangeanic, using TAUS Data Association data, which highlighted the benefits of PangeaMT as a domain-/client-specific MT solution range, its open-standard approach, and the necessary scalability and ease-of-use of the web-based interface as required by Sony.

This use case was part of the Moses Case Study Series. All speakers of these carefully selected presentations took part in a final Moses Use Case and Panel Discussion, led by Jaap Van der Meer, TAUS Director. This highly interactive discussion was recorded and can be checked out here as well. Salomé López-Lavado, Sony Europe's Translation Department Manager, stated: "Since we began working with Pangeanic's translation service, they have become a trusted translation vendor. We have started a collaboration in machine translation customization using our own translation memories and data. Pangeanic has cleaned, selected data prone for machine translation and, after several experiments, they have produced state-of-the-art engines to pre-translate documents. We also have a translation panel available (similar to Google Translate) to query sentence, word, or on-the-fly translations."

"We have primarily working from English, and PangeaMT pre-translates into French, Italian, Spanish, and German for our internal translation department to post-edit. The difficulty in creating our machine translation engines stems from the use of our very specific terminology, which cannot be mixed with other data. We also directed Pangeanic to develop engines that contained a large number of marketing documents, as we translate a lot of outbound documentation. Pangeanic harvested data within the domain to increase every engine's performance. We would strongly recommend working with Pangeanic engines. The company is flexible in its approach to machine translation customization and has proved itself capable of building an MT service, maintaining an SLA, and providing excellent results."