Industry field guide
Translation glossary and terminology decision data for AI
The same source term can have different approved translations for different audiences or regions.
- 2022 NAICS
- 541930
- Practice pack
- Example and answer key (JSON)
- Updated
What this guide tests
Which entry supports the preferred physical-part term for TR1, and should it be applied unchanged to TR3?
See the specimen and fields · Work through the fictional records
Task and source context
The same source term can have different approved translations for different audiences or regions. The American Translators Association’s professional code addresses confidentiality and accurate communication. A language service can write a small original example with a bilingual reviewer and explicit terminology rules. ATA Code of Ethics
Write a short fictional equipment label and two glossary entries for distinct locales. Change the preferred term in one revised entry and include a sentence where the term has a different sense. Ask the model to choose and cite the applicable entry and flag the ambiguous case for review.
Use independently written source and target text. Exclude client translation memories, unpublished files and interpretation audio. Obtain consent and written reuse rights for any recorded example.
Industry research
The U.S. Census Bureau’s industry definitions establish which activities these codes cover. The workflow sources in this guide describe the task or record type. Fiund’s pilot below uses fictional material to test a specific decision within that scope.
541930 · Translation and Interpretation Services
This industry comprises establishments primarily engaged in translating written material and interpreting speech from one language to another and establishments primarily engaged in providing sign language services.
Build a pilot sample
Set up this pilot
Write original phrases with one term used for a physical part and another for a metaphor. Approve a glossary entry by locale and sense, then grade reuse.
Prepare the test packets
- Start with the worked case. Use the fictional record types below as a guide to the source files you need. Give each packet a stable ID and save its source versions with it.
- Change one fact. Make one packet with enough evidence to answer the question and one where the evidence points to a different answer. Keep the other fields fixed so a reviewer can tell what changed.
- Remove the decisive evidence. Give the reviewer a third packet with a plausible near-match in its place. Have them write what can still be established and what evidence would resolve the rest.
Fields the reviewer needs
- Source span and intended sense
- Limits the terminology rule's applicability.
- Locale and approved glossary revision
- Identifies the controlling preference.
- Review status and unmatched sense
- Preserves cases requiring language judgment.
Use Build an AI evaluation from professional work to turn the specimen into a brief with a defined task, source and review owner.
Fictional exercise
Locale, sense and glossary revision jointly determine a preferred term
A fictional equipment label and glossary are independently authored. A bilingual reviewer would validate a real language evaluation; this text-only example illustrates terminology evidence without using client translation memories.
| Record ID | Source | What it records |
|---|---|---|
| TR1 | Original label | Source: 'Place the cover over the display.' Target locale L-A. Here cover means the physical protective part. |
| TR2 | Approved glossary | L-A v1: cover/protective part = cubierta. L-A v2 explicitly replaces that entry with tapa. L-B v2 retains cubierta for the same sense. |
| TR3 | Ambiguous sentence | Source: 'The agreement will cover the display.' No sense annotation or approved agreement-context entry supplied; target locale L-A. |
Question for the model. Which entry supports the preferred physical-part term for TR1, and should it be applied unchanged to TR3?
Show the reference answer
TR1 specifies locale L-A and the protective-part sense, so TR2's approved L-A v2 entry supports tapa. The L-B entry and superseded L-A v1 entry do not control that choice. TR3 uses cover in an agreement context without a matching approved sense, making automatic substitution unsupported. Flag the sentence for bilingual review instead of converting a physical-part glossary preference into a complete translation of an unrelated sense.
How to check the result
Ask for the answer, the record IDs that support it and any fact the evidence leaves unresolved. Have a reviewer compare the response against the reference before deciding whether the sample is ready to expand.
Errors to flag in this case
- Selecting a term from the correct revision but the wrong target locale.
- Applying a noun's physical-part glossary entry to an ambiguous agreement-context verb.
Change the inputs and run it again
Switch TR1 to locale L-B while retaining the same physical part. Require cubierta with the L-B source, rather than globally adopting L-A's revision.
Score the decision and the evidence separately
Save the reviewer’s expected answer, supporting record IDs and unresolved fields in a separate answer key. Score the model on the decision, its citations and its handling of the deliberate gap. A correct answer citing a different case counts as an evidence error.
Before expanding collection, have a second reviewer run the same checks on the held-out packet. Keep related versions, views and exports of the same case together so they stay on one side of the training and evaluation split.
Check permission and scope
Use original source and target examples with reviewer authority. Exclude client memories, interpretation recordings and confidential text.
Use original or licensed practice material. Leave client work product, confidentiality obligations and regulated outcomes out of the sample.
Describe an authorized archive through Fiund’s existing-data assessment, or describe a new capture capability through the collection brief. The contributor agreement covers the terms for participation.
Sources and further reading
Resources for this workflow
Company and program reviews
Use these reviews to compare sourcing, capture or evaluation approaches. Confirm the applicable program, accepted sources and terms with the provider.
Compare the approach
Invisible Technologies review for expert-led AI training
Compare expert calibration and reviewer agreement before commissioning a professional-work evaluation.
Compare the approach
Mercor review for expert AI training work
Read how an individual expert role differs from licensing an employer or client’s work product.
Related industry examples
Industry guide · NAICS 541990
Weather forecast issue and verification window data for AI
A weather provider can test whether a model matches a forecast’s valid period to the correct observations.
Industry guide · NAICS 541910
Survey wording and weighting records for AI
Two surveys can report similar percentages while asking different questions or using different weights.
Industry guide · NAICS 541921
How photographers can license original images for AI training
Start with a shoot you controlled from planning through delivery.