On this page
Step 1: collect candidate termsStep 2: decide what each entry holdsStep 3: decide the Korean equivalentsStep 4: put it where the work happensStep 5: keep it aliveWhere this fitsA SaaS product calls its main object a "workspace". In the Korean app it is a transliteration of the English word. In the help center it is a native Korean word meaning roughly "work area". In the sales deck it is a third term the Korean partner prefers. Korean customers ask support whether these are three different features. Nobody made a mistake; nobody made a decision either.
A termbase is a list of decisions: for this concept, in this product, we use this word in each language. It is the cheapest consistency tool in localization and the one most often left half-built. This post walks through building one for Korean, step by step, in a form your translators, reviewers and machine translation engines can all use.
Step 1: collect candidate terms
Do not start from an industry glossary. Start from your own content and your own problems:
- Product, plan and feature names, from the UI and the pricing page.
- Nouns that appear in UI strings more than a few times (export the strings and count).
- Terms used in contracts, terms of service and the privacy policy.
- Words reviewers have already corrected or that appear in Korean support tickets.
- Terms that must not be translated: brand names, code identifiers, some technical acronyms.
For a first version, 50 to 150 terms is plenty. A short list that people use beats a long one they ignore.
Step 2: decide what each entry holds
A term pair alone is not enough. Translators and engines need to know why a term was chosen and when it applies.
| Field | Purpose | Example (illustrative) |
|---|---|---|
| Source term | The English term as it appears in the product | workspace |
| Definition | What it means in your product, in one sentence | A shared area where a team's projects, members and settings live |
| Korean term | The approved equivalent | Transliterated loanword of "workspace" |
| Forbidden variants | Words that must not be used for this concept | Native word for "work area"; the partner's alternative term |
| Part of speech and usage | Noun, verb, button label; any grammar notes | Noun; ends in a vowel in Korean, so takes the vowel form of particles |
| Context and source | Where it appears; a screenshot or link | Left navigation, settings page |
| Status, owner, date | Proposed, approved or deprecated; who decided; when | Approved by product lead, with date |
The particle note is a Korean-specific detail worth keeping. Korean attaches particles to nouns, and their form depends on whether the noun ends in a consonant or a vowel. Recording it helps developers who build strings with variables.
Step 3: decide the Korean equivalents
For each concept, a Korean linguist proposes options and the reasoning. The common choices in Korean software are:
Transliterated loanword
The English word written in Korean scriptFamiliar in tech; matches what users see in other softwareGood for widely used product concepts. Can feel cold or opaque for less technical audiences.Native or Sino-Korean term
An established Korean word for the conceptClear to every reader; can sound formalGood for general actions and for regulated or legal contexts, where established terms matter.Keep in English
Latin script inside Korean textPreserves brand and code namesRight for brand names and identifiers; wrong for anything a user needs to understand.Check each candidate against what Korean users actually say: search data, competitors' interfaces and your own support tickets. Then have the product or domain owner confirm that the chosen word means the right thing, and let one named person make the final call.
Step 4: put it where the work happens
A termbase only works if it appears while people and machines translate. Load it into your translation management system so approved terms show up next to each segment, and turn on the QA check that flags a source term whose approved translation is missing.
Load it into machine translation as well. DeepL supports glossaries and style rules for formatting and stylistic preferences, and has announced a tool that proposes glossary entries from existing translations. Google Cloud Translation supports glossaries and an adaptive translation mode that steers output using your example translations. For LLM prompts, include the relevant entries with each request. Treat generated term suggestions as proposals for a human to approve, not as decisions.
Step 5: keep it alive
- New features get their terms decided before the strings go to translation, not after.
- Every reviewer correction that changes a term goes back into the termbase the same week.
- Deprecated terms stay in the termbase, marked deprecated, so people can see what replaced them.
- Review the list each quarter with product, support and marketing; terms drift when teams do.
- Make sure you can export the termbase if you change tools. Two years of decisions are an asset.
Where this fits
The termbase is one component of a wider stack; see the translation tooling stack. For why inconsistent terms hurt, see translation mistakes that damage brand trust and inconsistent brand voice across languages. In patents, where terminology carries legal weight, the same discipline matters even more: see how incorrect IP terminology can jeopardize patent rights.
If you buy translation from a vendor, put termbase ownership in the contract too; our guide to running large document translation programs covers that clause.
Need a Korean termbase that people actually use?
We build Korean termbases from your product, tickets and contracts, with reasons for each decision, ready to load into your TMS and MT engine. Get in touch →