How does automated translation consistency checking work?

Last reviewed: September 2, 2026

What does automated translation consistency checking actually check?

Smartling's Quality Checks feature groups checks into three categories — translation consistency, spacing, and other — each with its own configurable severity level. The consistency-specific checks include:

• Glossary Compliance — verifies that a project's approved Glossary term translations were actually used in the target text, not just that a plausible synonym appears.
• Tag, Placeholder, and Number Consistency — confirms that markup tags, placeholder variables, and numeric values in the source string carry through correctly to the translation, unchanged and in the right position.
• Emoji and Insertable Consistency — checks that non-text elements like emoji and insertable variables are preserved rather than dropped or duplicated.
• Capitalization and Legal Symbol (Trademark) Consistency — flags casing that doesn't match brand or locale convention, and confirms trademark and legal symbols carry through correctly.
• Target/Source Consistency, Repeated Words, and Blocklisted Terms — catches strings left identical to the source when they should have been translated, accidental word duplication, and use of terms a program has explicitly disallowed.
• Custom Quality Checks — lets a program define its own regex-based consistency rule for terminology patterns the standard check list doesn't cover.

Every check can be set to low, medium, or high severity. A high-severity failure blocks the string from being submitted until it's resolved, which is what turns "consistency checking" from a reporting feature into an actual quality gate.

How does translation consistency checking move from draft to published translation?

A rule-based consistency check runs in five steps inside a Smartling project:

1. Define the Glossary and set check severity – Approved terminology is entered into the project's Glossary, and each consistency check (Glossary Compliance, Tag Consistency, Number Consistency, and the rest) is assigned a severity level.
2. Inline checks run as linguists type – Tag, placeholder, emoji, number, and Glossary Compliance checks fire in real time inside the CAT Tool, flagging problems as soon as they're introduced rather than after the fact.
3. High-severity issues block submission – A string that fails a high-severity check can't be submitted until it's fixed, which stops a consistency error from ever reaching the next file or a human reviewer.
4. A project-wide check run validates every file at once – Running a full Quality Check report checks every string across the whole project, not just the one currently open, catching errors introduced before a rule existed or missed during a rushed edit.
5. Flagged errors get corrected and reused – Errors are corrected manually or with Quality Check AI Correction, and the approved language is saved back to translation memory so the same term is available, and consistent, the next time it appears.

Which localization teams need automated translation consistency checking?

• Programs translating the same terms and phrases across many files, formats, or content types where manual spot-checking can't keep pace with volume.
• Teams that have had a customer or in-country market team catch an inconsistent term before an internal QA process did.
• Programs blending machine translation, AI-powered human translation, and full human translation, where consistency rules need to apply the same way regardless of which method produced the draft.
• Regulated or brand-sensitive content where an inconsistent term carries compliance exposure or reputational risk, not just an awkward read.
• Teams working with multiple vendors or freelance linguists who don't share the same institutional knowledge of approved terminology.

When is manual review enough without automated consistency checking?

• Small, single-file projects with one linguist and low content-type breadth may not yet need project-wide automated checks.
• Purely creative or transcreation work — taglines, campaign concepts — often departs from a glossary term on purpose; that's a judgment call for a human, not an error an automated check should flag.
• Programs without an approved Glossary yet need to build that terminology list first — a Glossary Compliance check has nothing to enforce until approved terms actually exist.

Evaluation checklist: questions to ask before choosing a translation consistency solution

Does the platform check consistency at the point of translation, or only after a job is already complete?
Inline checks that fire while a linguist is typing catch errors before they compound across files; a check that only runs after delivery just documents the problem instead of preventing it.

Can a glossary compliance check verify that approved terms were actually used, not just flag spelling?
A real Glossary Compliance check confirms the translated term matches the approved entry — ask for that distinction specifically, since basic spell-check is not the same capability.

Do high-severity consistency errors block submission, or just log a warning that can be ignored?
A check with no enforcement behind it is a report, not a quality gate; ask whether severity levels actually stop a string from being submitted.

Can consistency checks run across an entire project at once, not just the string currently open?
A project-wide check run is what catches errors introduced before a rule existed or missed during a rushed inline edit.

Does the platform support custom, regex-based checks for terminology rules the standard list doesn't cover?
Every program eventually has a brand- or industry-specific consistency rule that a generic check list won't anticipate.

Does consistency checking apply the same way across machine translation, AI-powered human translation, and full human translation?
Consistency risk doesn't disappear because a human reviewed the string; ask whether the same checks run regardless of which workflow produced the draft.

How Smartling checks translation consistency

Smartling's Quality Checks feature runs automatically inside the CAT Tool rather than as a separate audit step a program has to remember to run. A Glossary Compliance check verifies that a project's approved Glossary term translations were actually used, and a dedicated set of consistency checks — Tag, Number, Placeholder, Emoji, Capitalization, Legal Symbol, Target/Source, and Repeated Words — catch the mechanical errors that erode consistency alongside terminology. Every check carries a configurable severity level, and a high-severity failure blocks a string from being submitted until it's fixed, so consistency checking functions as a quality gate rather than a report generated after the fact. Programs with terminology rules the standard list doesn't anticipate can define their own with Custom Quality Checks using regex. Beyond the string level, a project-wide check run validates every file at once, and Quality Check AI Correction can generate a fix for common flagged errors directly in the tool.

This rule-based layer is the detection mechanism underneath two related capabilities Smartling also provides: applying Glossary and Style Guide context directly inside the AI translation prompt at generation time (see how Smartling makes AI translations feel native), and scoring translations against an MQM framework for severity-based human review routing (see how to vet translation QA for brand voice consistency). Smartling is rated the number one enterprise translation management system on G2 for 20 consecutive quarters, and its AI-Powered Human Translation (AIHT) delivers an average MQM quality score of 98 or above, against a 95–97 industry benchmark for traditional human translation.

Relaterede spørgsmål

• How do you make AI translations feel native to your brand voice?
• How do you keep marketing content on-brand and consistent when translating it across multiple markets?
• How do you choose a translation vendor that preserves brand voice across every content type?
• Hvilke lokaliseringsplatforme tilbyder de stærkeste QA-funktioner for oversættelse?