Azure AI Translator glossary: no inflections or derived forms recognized and casing issues for headings (DE → EN)

Hi Stefan Müller

Glossary matching for plurals, inflected, and derived forms:

Azure AI Translator glossaries (Document Translation, Dynamic Dictionary, and phrase dictionaries) work as exact, literal find‑and‑replace rules. They do not perform morphological analysis, stemming, lemmatization, or plural/inflection expansion.

This means a glossary entry such as “Server → Server” will match only that exact surface form; related forms like Servers, Servern, Serversysteme, or other derived/compound forms are not matched automatically. This behavior is documented and intentional, not a defect, and applies equally to German→English translations. [learn.microsoft.com], [learn.microsoft.com]

limitation exists:

Glossaries in Azure Translator are designed to be deterministic and predictable. They are applied before or alongside the neural model as forced replacements, not as linguistic rules. Because of this design, glossaries deliberately avoid “smart” linguistic expansion that could introduce ambiguity or inconsistent output. Morphology and fluency are instead handled by the base neural MT model, not by glossary logic. [learn.microsoft.com], [learn.microsoft.com]

Recommended workarounds for morphological variants :

The officially recommended workaround is to explicitly enumerate all required variants in the glossary (for example: Server, Servers, Serversysteme, etc.). In advanced enterprise pipelines, teams often generate these variants using external German NLP tooling and expand the glossary offline before submitting it to Translator. Using Custom Translator with a neural phrase dictionary can improve overall fluency, but glossary enforcement itself still remains exact‑match and case‑sensitive.

Casing problems in Word headings vs body text

Azure Translator applies glossary replacements without awareness of document structure (for example, Word headings vs body paragraphs). The service does not know whether a matched term appears in a heading, title, or normal sentence. As a result, if your glossary defines a term in lowercase, it will also be inserted in lowercase in Word headings, even though the heading style visually expects capitalization. This behavior follows the documented rule that glossary matching is case‑sensitive and literal. [learn.microsoft.com]

context‑aware casing is not supported:

Document Translation preserves layout and formatting styles (like Word heading styles), but text casing is controlled by the glossary entry itself, not by document context. There is currently no feature to apply different glossary casing rules based on structural context (heading vs body text) across Word, PowerPoint, PDF, or HTML documents. This is a known limitation of the service design. [learn.microsoft.com], [docs.azure.cn]

Best practices for handling capitalization with glossaries :

Microsoft‑aligned best practices are:

1. Case‑match your glossary entries to the expected source casing (for example, include both Server and server entries if needed).
2. Post‑process translated Word documents to enforce heading capitalization (Title Case) using Word styles or automation, while leaving body text unchanged.
3. In high‑fidelity publishing workflows, translate headings and body text separately and then reassemble the document. These approaches are commonly used in regulated and technical translation pipelines. [github.com]

Roadmap and future support: As of early 2026, Microsoft has not published a roadmap commitment for morphology‑aware glossary matching or context‑aware casing control in Azure Translator. Current documentation and public statements continue to position glossaries as exact, deterministic rules, with linguistic intelligence handled by the neural MT model instead. [learn.microsoft.com], [docs.azure.cn]

Morphological/plural/derived form matching: Not supported; exact match only. •

Different casing for headings vs body text: Not supported in glossaries.

Recommended approach: Enumerate variants + case‑matched entries + post‑translation Word formatting. • Roadmap: No public commitment to change this behavior.

• Azure AI Custom Translator Overview
• Dynamic Dictionary Guidance

Do let me know if you have any further queries.

Thankyou!