ドキュメント翻訳の拡張 – Microsoft Buildで発表された新機能Expanding the Reach of Document Translation – New Capabilities Announced at Microsoft Build

Microsoft used its Build 2026 conference to detail an expanded set of Document Translation capabilities in Azure Translator, now surfaced to developers through Foundry Tools. The update matters because document-level translation has long been more difficult than translating loose strings of text: real-world files carry layout, formatting, embedded images, and structured markup that a translation engine must preserve while it swaps one language for another. By widening the range of supported formats and routing access through Foundry, Microsoft appears to be positioning translation as a standard building block within its broader AI development stack rather than a standalone service.

At the center of the announcement is broader file-type coverage. The enhanced Document Translation service can now handle images, PDFs, and Office files such as Word, PowerPoint, and Excel, alongside two formats that matter specifically to technical writing and localization workflows: DITA and XLIFF. Support for scanned or image-based content typically relies on optical character recognition to extract text before translation, after which the service attempts to reconstruct the original document with the translated content in place. For PDFs and Office documents, the goal is to return a file that looks like the source but reads in the target language, sparing developers the work of parsing and reassembling complex documents themselves.

The inclusion of DITA and XLIFF is notable for organizations that produce documentation at scale. DITA, the Darwin Information Typing Architecture, is an XML-based standard widely used for modular technical documentation, where content is broken into reusable topics. XLIFF, the XML Localization Interchange File Format, is a standard designed specifically to move translatable content between authoring tools and translation systems while keeping track of what still needs work. Supporting these formats directly means the service can fit into established localization pipelines and translation management systems instead of forcing teams to convert files into a simpler format and lose structural metadata in the process.

Delivering these features through Foundry Tools reflects how Microsoft has been consolidating its AI offerings. Azure AI Foundry is the company's platform for building, evaluating, and deploying AI applications, and Foundry Tools represents a collection of capabilities that developers and agents can call as reusable components. Exposing Document Translation here means that translation can be composed alongside other Foundry functions, potentially as a tool an AI agent invokes when a workflow requires moving content across languages. This aligns with a wider industry pattern in which discrete AI capabilities are packaged as callable tools that larger systems orchestrate, rather than features accessed only through separate portals or one-off APIs.

Microsoft also signaled that large language model-based translation is on its roadmap. The existing Document Translation service has historically been built on neural machine translation, the approach that has powered Azure Translator for years and that remains strong for high-volume, predictable translation tasks. LLM-based translation is likely intended to complement this by offering more context-aware output, better handling of tone and terminology, and the ability to follow instructions about style or domain. Because the company has framed this as a future capability, the specifics, including availability, supported languages, and pricing, were not fully detailed, so the practical impact will become clearer as it ships.

For context, document translation sits within a competitive field. Google Cloud, Amazon Web Services, and specialized providers such as DeepL all offer document or API-based translation, and many enterprises pair these engines with translation management platforms and human review steps for quality assurance. The growing interest in LLM-driven translation across the industry suggests that vendors increasingly see general-purpose models as a path to higher-quality, more adaptable localization, though neural machine translation remains cost-effective and fast for many scenarios.

Developers evaluating these capabilities will want to weigh format fidelity, latency, and cost against their existing tooling, and to test how well structured formats like DITA round-trip through the service without losing markup. As with any machine translation, human review is still advisable for high-stakes or published content. Overall, the Build announcement points to a steady expansion of translation as an integrated, developer-facing service within Azure AI Foundry, with the LLM-based work representing the next phase to watch.