Guides

How to Catch Localization Bugs Before Release

Maksym Ostapenko
Head of Product Growth, SmartcatUpdated · 6 min read
AI summary

A German button that doesn’t fit, a sentence glued together from fragments, a placeholder that crashes the Arabic build: localization bugs show up late because nobody sees the translated interface until the release candidate.

Localization testing checks that a product works and reads correctly in every language it ships in. The short answer: pseudo-localize while you build, run linguistic QA on every string, and test the localized build. Smartcat covers the first two.

This guide covers the three testing layers, how to run pseudo-localization, how much room to leave for text expansion, linguistic QA, and a checklist organized by bug type. The examples use Smartcat.


Key takeaways

  • Three layers. Pseudo-localization while you build, linguistic QA on every string, functional tests on the build.
  • Pseudo-localize early. Accented, padded fake text exposes hard-coded strings and tight layouts before translation starts.
  • Plan for expansion. Strings of up to 10 characters can grow 200–300% in translation (W3C).
  • Automate the string checks. Smartcat runs automated QA and reviewer approval; 92% of AI-translated words need no edit.
  • Fix bugs where they’re cheapest. A hard-coded string is one line of code during development and a patch release after it.

How do you catch localization bugs before they reach production?

Test in three layers: pseudo-localization while the feature is built, linguistic QA on every translated string, and functional tests on the localized build. Smartcat covers the first two, with a pseudo-localization engine, automated QA checks and reviewer approval for each of 280+ languages, so strings reach your QA environment already checked.

LayerWhat it findsWhenWhere
Pseudo-localizationHard-coded strings, clipped labels, missing glyphsAs soon as a feature’s strings are externalizedSmartcat or your platform’s pseudolocales
Linguistic QAWrong terms, broken placeholders, inconsistent translationsOn every translated stringSmartcat: automated checks and reviewers
Functional testingCrashes, wrong formats, broken flows in another languageOn the build you’re about to shipYour QA environment and devices

The earlier a bug is found, the cheaper the fix: a hard-coded string caught in a pseudo-localized build is one line of code, the same string found after translation is a patch release. More on the testing side of the product: localization testing in Smartcat.


What is pseudo-localization, and how do you run it?

Pseudo-localization replaces every UI string with longer, accented look-alike text, so layout and encoding bugs show before any real translation exists. Windows defines three pseudo-locales for this (Microsoft Learn: Pseudolocalization), and Smartcat has a pseudo-localization engine; a Smartcat specialist can show you how it fits your project.

English          Pseudo-localized
Settings         [Šéţţîñĝš ~~~~~]
Save             [Šåvé ~~~]
Delete account   [Ðéļéţé åççöûñţ ~~~~~~]
Brackets show where a string is cut off; accents show whether the font and encoding cope; the tildes pad each string so tight layouts break now, not after translation.

Any string that still reads as plain English in a pseudo-localized build is hard-coded, which means it will never be translated.


How much room should your UI leave for translated text?

Plan for short labels to double or triple: English strings of up to 10 characters can grow 200–300% in translation, while text over 70 characters grows about 130% (W3C: Text size in translation). Smartcat enforces length limits where the format carries them, such as XML @max attributes and LocJSON maximum lengths.

English source lengthTypical size after translation
Up to 10 characters200–300%
11–20 characters180–200%
21–30 characters160–180%
31–50 characters140–160%
Over 70 charactersAbout 130%
  • Buttons and tabs: the shortest strings expand the most, so test them first.
  • Hard limits: push notifications, store titles and table headers need a maximum length translators can see.
  • Layout: let containers grow or wrap instead of fixing their width to English text.

How do you run linguistic QA on software strings?

Combine automated checks with a human reviewer who sees each string in context. In Smartcat, automated QA flags placeholders, numbers and terminology, the Quality Assurance Coworker checks every language, and 92% of words translated by Smartcat AI are accepted without a single human edit, so reviewers focus on the rest.

A diagram of automated QA checks in Smartcat flagging two segments and passing one
Automated QA checks flag problem segments and pass the rest.
  • Context: key IDs and developer comments tell the reviewer where a string appears.
  • Terminology: a glossary keeps product terms the same across the app, the help center and release notes.
  • Review at scale: the Reviewer Coworker helps reviewers work through many languages.

“When people hear how quickly I can run 31 languages through the Reviewer Coworker, and that I can do in hours what would traditionally take weeks of work, they really perk up. That kind of speed changes what’s possible.”

Michelle Quirke, Program Engagement Manager, Topcon

What belongs on a localization testing checklist?

Group the checks by the bug they catch, and assign each to the layer that finds it cheapest. Smartcat handles the string checks before the build; the 8 checks below cover what most product teams miss.

CheckBug it catchesLayer
Every visible string comes from a resource fileHard-coded textPseudo-localization
No sentence is built from fragmentsWord order that can’t be translatedPseudo-localization, code review
Labels fit at 200% lengthTruncation and overlapPseudo-localization
Placeholders match the sourceCrashes or wrong valuesLinguistic QA (automated)
Plural forms exist for every language“1 items”, missing formsLinguistic QA
Terms match the glossaryInconsistent product namesLinguistic QA
Dates, numbers and currency use the localeWrong formatsFunctional testing
Right-to-left screens mirror correctlyBroken layouts in Arabic or HebrewPseudo-localization, functional testing

For a step-by-step testing process, see localization testing: a step-by-step guide and checklist.


See it on your product

Bring one feature’s strings and see pseudo-localization, QA checks and review in Smartcat (localization testing in Smartcat).


Frequently asked questions

What is the difference between pseudo-localization and localization testing?

Pseudo-localization is one localization test: it runs the interface with fake, longer text before translation. Localization testing also includes linguistic QA of real translations and functional tests of the localized build. Smartcat covers pseudo-localization and linguistic QA.

Does Smartcat have a pseudo-localization engine?

Yes. Smartcat has a pseudo-localization engine, and a Smartcat specialist can show you how it fits your project and release process.

What does linguistic QA check in software strings?

Linguistic QA checks that each translation is correct, consistent with the glossary and safe for the code: placeholders, numbers, tags and length. In Smartcat, automated checks run on every language and reviewers approve in context.

When should localization testing start?

As soon as a feature’s strings are externalized. Pseudo-localization at that point finds hard-coded text and tight layouts while developers are still working on the feature, which is when they are cheapest to fix.

Can Smartcat enforce maximum string lengths?

Yes, where the format carries a limit. Smartcat reads @max attributes in XML and maximum lengths in LocJSON, so translators see the limit while they work.

Is Smartcat affiliated with the companies named in this guide?

No. Smartcat is an independent platform and is not affiliated with, endorsed by, sponsored by or otherwise associated with any company or product named in this guide. All product names and trademarks are the property of their respective owners and are used solely for identification and comparison.


Sources

  1. Microsoft Learn: Pseudolocalization
  2. Android Developers: Test your app with pseudolocales
  3. Apple: Testing your internationalized app (archived guide)
  4. W3C: Text size in translation
Share this guide@
Maksym Ostapenko

· Head of Product Growth, Smartcat

Maksym Ostapenko is Head of Product Growth at Smartcat, where he works on AI translation workflows, software localization and enterprise localization. He writes about translation technology and localization operations from hands-on product experience.

LinkedIn