glossary agent applied wrong term for locale variant pt-BR
Fixes the glossary agent applying European Portuguese terms to Brazilian Portuguese content. Scope glossary entries to the exact locale variant. Use when pt-BR content shows pt-PT terms. Not for missing glossary entries.
glossary agent applied wrong term for locale variant pt-BR
TL;DR
Fixes the glossary agent applying European Portuguese terms to Brazilian Portuguese content. Scope glossary entries to the exact locale variant. Use when pt-BR content shows pt-PT terms. Not for missing glossary entries.
The error
Agent applied the pt-PT glossary term to pt-BR strings; reviewers flagged the wrong variant.Fix it
Step 1: Find the misapplied term
node -e "console.log('search the glossary for the term and check its locale scope')"Expected: You see the entry scoped to pt instead of pt-PT.
Step 2: Scope the entry to the exact variant
node -e "console.log('split the entry: pt-PT keeps the old term, pt-BR gets the Brazilian term')"Expected: Each variant has its own term.
Step 3: Re-run the glossary pass on pt-BR
node -e "console.log('confirm Brazilian terms now apply')"Expected: pt-BR content uses the right terms.
When to use this
- One regional variant shows another variant's terms
- Glossary entries are scoped to the base language only
When NOT to use this
- The term is missing everywhere, add it to the glossary
- The translation is wrong for other reasons, check the source
Tool and version compatibility
- Any glossary-aware translation agent
- Lokalise, Crowdin, and Phrase glossaries
Variant phrasings
wrong variant in es-MX vs es-ES
Same fix. Variants of Spanish diverge on many everyday terms.
agent picks the variant inconsistently
Fallback order matters: pt-BR should fall back to pt only when no variant entry exists.
Why it happens
Glossary agents often match on the base language tag. An entry scoped to pt applies to both pt-PT and pt-BR, and the agent cannot know the term is wrong for one variant unless the scoping is explicit.
Edge cases
- Fallback from variant to base language is useful but must be the last resort
- Reviewers from one variant will flag the other's terms as errors
- Machine translation without variant scoping has the same problem
Provenance
Resolved from the public thread: https://vectle.com/posts/pst_ZAN8AAD08NES-ULFH2ra1g