A community seed exchange is preparing a bilingual seed catalog. Each listing has a variety name, planting directions, and a brief description. Translating the unique variety name differently across pages can confuse volunteers even if the surrounding paragraph reads naturally. Amazon Translate’s custom terminology feature lets a team upload preferred term equivalents, but the documentation warns that the chosen target term is not guaranteed in every context. This guide describes a fictional catalog and a human-reviewed production workflow, not translation quality we measured.
Decide which text is a name and which is an instruction
Begin with a small editorial glossary that separates fixed variety names from agricultural instructions. A fictional seed named ‘Harbor Lantern Bean’ might keep a recognized label in one language, while ‘plant after the last frost’ must be translated accurately for local conditions. The team should assign an owner and source record to each approved name. Do not assume botanical names, marketing names, and common-language terms all follow the same rule. The most dangerous error may be an altered instruction rather than a misspelled name.
Store the catalog’s source language and each target language explicitly. For each entry record the original variety spelling, approved target form, rationale, and last reviewer. Include an alternative display name only if the local gardening group has approved it. Save non-translatable codes, lot identifiers, and measurement units separately from preferred translated phrases. A machine-readable terminology file should reflect real editorial decisions rather than guesses created by a language model.
Prepare a terminology file in a supported format
Amazon Translate documents CSV, TSV, and TMX terminology files. A CSV file uses language codes as column headers, with the source term and corresponding target-language forms in each row. Confirm the character encoding and file format restrictions in the current API documentation before importing. A deliberate sample can contain the fictional name ‘Harbor Lantern Bean’ and its editorially approved equivalent, but do not present that invented equivalence as a real botanical standard.
Version the terminology file alongside the catalog content, and give it a human-readable changelog. If a variety is renamed after review, the old file must not silently remain active for new editions. Avoid putting member contact details or private supplier information into the glossary. Run a small dry check with text you are authorized to process, and ensure the intended terminology resource is selected in the translation request rather than assuming a file upload automatically affects every job.
Understand matching without promising perfect enforcement
The documentation says custom terminology matching is case-insensitive: a source term can be recognized regardless of capitalization. But the service also warns that it does not guarantee every desired target term will be used because translation context affects selection. Therefore, a passing terminology import is only proof that the resource exists, not proof that every name in a multilingual PDF is correct. Test term matches at the beginning, middle, and end of realistic catalog sentences.
Use a controlled list of spelling variants, hyphenation variants, and pluralized phrases to see what the workflow actually does. Do not claim a partial match will trigger a particular replacement without testing. Display source and target side by side to the bilingual reviewer and highlight fixed names for manual verification. Keep translation software’s suggestions as suggestions; the editorial glossary remains the authority for how the organization wants to label the seeds.
Compare a protected name with a dangerous instruction drift
In an illustrative three-panel comparison, the first translation changes a variety name, the second preserves the name while changing a planting instruction, and the third preserves both after editorial review. These are invented examples, not outputs from Amazon Translate. This distinction is useful because a perfectly stable name can conceal incorrect climate advice. A reviewer should check quantities, temperature thresholds, directions, negations, and references to treatment, especially if a list is meant for beginners.
For example, ‘do not soak treated seeds’ and ‘soak treated seeds’ are opposites with potentially different handling implications. This example is hypothetical; it is not horticultural guidance. Mark negations and safety or handling language as critical fields for human proofreading. The glossary may help with a proper noun, but it does not certify that instructions retain meaning. Correct the source document if the original wording is ambiguous before asking another system to translate it.
Use a review matrix rather than one overall quality score
Organize proofreading by error type: fixed names, lot codes, units, planting directions, warnings, typography, and cultural readability. Ask a competent speaker of each target language to inspect a sample and then every field where an error could create harm or confusion. A reviewer should see the term from the glossary beside the actual sentence. Record whether a mismatch resulted from source spelling, glossary version, unsupported match behavior, or an editorial change.
Do not let a reviewer silently overwrite the translation output without updating the source tracking record. If a desired term is consistently absent, revisit its spelling and context, consider a clearer source sentence, or apply an approved editorial post-processing rule with its own tests. Avoid broad string replacement that changes unrelated words or parts of longer names. Save the original translation and approved correction to keep future print revisions consistent.
Separate glossary maintenance from a published catalog
A terminology file is a working production asset, whereas a published catalog is a versioned artifact. Lock the catalog edition after bilingual review, then archive the exact source, glossary revision, target-language files, and approver. A later terminology change should trigger a deliberate republishing decision rather than altering older PDFs without notice. Provide a correction route for gardeners who find an inaccurate name or instruction after distribution.
Test the local-language search experience too. If a label is fixed in English but the public searches in another language, add discoverable approved aliases without altering the official display name. For accessibility, ensure screen readers can pronounce or at least navigate the code and variety name reasonably. A consistency rule is useful only when people can still find, identify, and safely use the seed listing.
Run small acceptance tests and document the limits
Create synthetic entries with repeated names, punctuation variation, multiple quantities, a negation, and a changed glossary revision. Process them through the planned translation settings only after checking current service requirements and data controls. Have a bilingual editor mark each field pass, revise, or hold. Log the service configuration and glossary version. These are proposed tests, not empirical claims about model accuracy or customer pricing.
Before releasing any translated catalog, recheck Amazon Translate’s current documentation for supported languages, file formats, terminology matching, and applicable charges. Review local agricultural advice with an appropriate expert where needed. The end goal is a traceable multilingual publication in which variety names are consistent and safety-relevant instructions remain meaningful—not a claim that a vocabulary list substitutes for human translation review.
How We Research
For this November 2026 source review, we read these named official references: Amazon Translate: Customizing with custom terminology; Amazon Translate: Creating custom terminology; Amazon Translate: Customizing translations. We compared documented inputs, outputs, limitations, supported formats, and workflow prerequisites against the specific scenario. The images and examples are original conceptual illustrations, not screenshots, measurements, or real outputs. We did not create a vendor account, submit actual community artwork, member forms, seed catalogs, recordings, or restricted archives, or run a live API performance test. We do not claim accuracy rates, latency, tested prices, legal consent, or verified translation, and we do not infer those from the drawings. Before operational use, confirm current feature availability, rate limits, regional and data-handling requirements, pricing, and organization policy. Where permissions, linguistic accuracy, historical quotations, or authorizations matter, a qualified person must review the original evidence.
Frequently Asked Questions
Amazon Translate documents CSV, TSV, and TMX formats for terminology files.
No. AWS says matching is case-insensitive, though output capitalization and context still require review.
No. AWS explicitly says custom terminology does not guarantee the target term will be selected in every context.
No. It provides preferred term translations, not agricultural verification or proof that negations and quantities remained correct.
Version the glossary and catalog, rerun affected pages, and have bilingual editors approve the changes before republishing.