Quality checks
What transglot checks in every translation, when the checks run, and how to read the Quality page.
transglot checks translations for specific, nameable problems: a placeholder that went missing, a plural form the language needs, a glossary term that was not used, a sentence that changes the meaning. Each problem becomes a finding on the string, and a string with a serious finding is flagged for a person to check.
What is checked
There are two kinds of check. The automatic checks follow fixed rules and cost nothing. The AI review reads the translation against its source the way a reviewer would.
| Check | Kind | What it catches |
|---|---|---|
| Placeholders | Automatic | A placeholder or variable from the source (such as {name} or %s) that is missing, extra or altered in the translation. |
| Plurals | Automatic | A plural string that lacks a form its language needs. |
| Length | Automatic | A translation longer than the maximum length set on its key. |
| Overflow | Automatic, advisory | A translation so much longer than its source that it may not fit where the source fits. It is reported but never blocks. |
| Terminology | Automatic | A glossary term translated differently from its required wording, or a forbidden term that appears. |
| Accuracy | AI review | The translation does not say what the source says: an omission, an addition or a mistranslation. |
| Fluency | AI review | Grammar, spelling or punctuation that is wrong in the target language. |
| Style | AI review | Wording that is awkward, inconsistent, or off the project's voice. |
If the project's Screenshot context setting is on, the AI review also sees the screenshot captured for a key, so it can flag text that overflows the space it renders in. See The screenshot library.
When the checks run
- As each AI translation is written. Every translation Glossa produces goes through the automatic checks before it is saved. Automatic checks always win: a translation with a broken placeholder, the wrong plural forms, or a glossary conflict is always sent to review no matter how confident it was.
- While you type in the editor. The editor checks your draft against the same rules before you save it.
- When you run a quality check. A full check runs the automatic checks and the AI review, and gives each translation a quality score. See Run a quality check.
- When a source string changes, if Source drift repair is on. See Configure checks and automation.
Severity
Every finding has one of three severities: minor, major or critical. Critical findings are counted as blocking on the Quality page. A translation is flagged for review when it has a finding of major severity or above, or when its quality score is low.
A flag does not stop a translation from being published. It puts the string in front of a reviewer: flagged strings appear under Needs checking in the editor and on Releases. See Why was my translation flagged for review?
The Quality page
Open the project and select Quality in the sidebar. Everyone who can open the project can read it.
Before the first check has run, the page lists each language with how many of its translated strings are waiting to be scored, under Nothing here has been checked yet.
After a check, the page header shows:
- the mqm · project average score, with its change and trend over the last 12 weeks,
- how many translated strings have been checked (checked 120 of 140 translated strings),
- the count of blocking findings and warning findings,
- how many strings need checking, and when the last check ran.
Below the header:
| Section | What it shows |
|---|---|
| Every check, every locale | A grid of languages against checks, with the number of findings in each cell. A language added since the last check reads not scored. Select a cell to filter the findings list to that language and check. |
| Findings | The strings behind the selected cell: each finding's check, severity and message. |
| MQM dimensions | The selected language's findings grouped into the five MQM categories. |
| All languages | Each language's score. |
The buttons in the header:
- Configure opens the project's check settings (Owners and Admins).
- Re-run checks starts a quality check. See Run a quality check.
- Fix 12 with Glossa repairs the failing AI translations. See Self-heal a translation.
The Scorecard tab beside Overview opens the per-language scorecard.
Findings in the editor
In the editor, a string with findings shows how many it has. Open them to see Quality findings for that string: each finding's category, severity and message. When a finding comes with a suggested fix, Accept suggestion saves it as your own edit.
Why was my translation flagged for review?
A translation needs checking because an automatic check found a problem, because it was not confident enough to skip review, or because a person asked for a second look.
Run a quality check
Score a project's translations and get a list of specific problems to fix, one language or all of them at once.