SAP Integrates Large Language Models Into Translation Hub for AI Translation

SAP has added large language model (LLM) technology to its cloud-based Translation Hub, expanding its translation offering to over 2,800 language pairs across 54 languages, as announced in a recent blog post.

The new option for AI translation — alongside SAP’s existing traditional machine translation (MT) — is available via the SAP Business Technology Platform (BTP), where the Translation Hub enables enterprises to localize user interfaces, documentation, and other business content directly through SAP-managed APIs and UIs.

SAP emphasized that this is an additional option — not a replacement — to its domain-trained MT, which remains the default for terminology-sensitive SAP content. The new capability is aimed primarily at translating general, non-SAP material, “where adherence to SAP terminology is not required/less important.”

According to SAP, the hybrid approach gives users flexibility based on content type and use case:

• SAP MT is optimized for SAP-specific content, offering faster turnaround and accuracy based on proprietary terminology.
• SAP AI Translation provides better handling of more generic or less structured content, especially for languages or language pairs not covered by SAP MT. It also performs better on so-called “noisy” content, including typos, incomplete sentences, or mixed-language inputs — commonly seen in chat messages or blog comments.

“With generative AI bringing about new opportunities, we decided to explore […] LLMs as an additional translation provider to support in broadening our language scope and to improve our ability to handle content outside of the SAP domain,” SAP said.

Users are clearly informed when content is machine- or AI-translated. A quality index of 25% is provided for both SAP MT and AI-translated texts, reinforcing the company’s guidance to review translations, and, where necessary, correct or adjust them before use.

Like its MT engine, the AI translation offering meets SAP’s standards for security and data governance, allowing it to be used with confidential content.