By Dr. Ibrahim Khalifa · Published 20 November 2026

Arabic linguistic quality assurance (LQA) is a structured review of Arabic content against a defined list of error types and severity levels. The reviewer reads the Arabic as a user would, records each problem with a category and a severity, and gives the team a list they can fix in order of importance.

This checklist is what I use as a starting point. I have worked in Arabic software localization for 33 years, 28 of them at Oracle, and I now review Arabic software and AI output as an independent reviewer. Treat the list as a framework to adapt, not a fixed standard: every product needs its own style guide and glossary.

What is Arabic LQA, and when do you need it?

LQA is a quality check done after translation or after an AI system produces Arabic, and before users see it. You need it when:

  • you are launching or updating an Arabic interface, help center, or marketing content;
  • you use machine translation or an LLM and want a human to measure how good the Arabic really is;
  • users report that the Arabic “feels wrong” and you need to know where and why.

A translator translates. An LQA reviewer does not rewrite your product; the reviewer finds and classifies problems so your team or vendor can fix them. That separation is also why an independent reviewer is useful: the verdict is not tied to selling the fix.

How errors are classified: categories and severity

Most LQA programs use a typology such as MQM (Multidimensional Quality Metrics), the established error taxonomy and severity scheme in the translation industry. The exact category names vary between programs, but the core is stable:

CategoryWhat it covers
AccuracyMeaning added, lost, or changed. Mistranslation, omission, addition, untranslated text
TerminologyWrong or inconsistent terms, ignoring the client glossary
FluencyGrammar, spelling, punctuation, agreement, unnatural phrasing
StyleRegister, tone, consistency with the style guide
Locale conventionsDates, numbers, currencies, units, calendars, address and name formats
Design and layoutText direction, alignment, truncation, mirrored UI, broken mixed text
OtherAnything that does not fit above

Each error gets a severity:

  • Minor: noticeable, but the user still understands and trusts the product.
  • Major: changes meaning, confuses the user, or makes the product look unprofessional.
  • Critical: blocks the task or creates legal, financial, safety or security risk. A wrong meaning on a payment, security or primary-action screen is Critical regardless of how short the text is.

Agree on these definitions with your reviewer before the project starts, so the scores mean the same thing every time.

The checklist

1. Accuracy

  • Does the Arabic say what the source says, nothing more and nothing less?
  • Are negations, conditions (“only if”, “unless”) and quantities preserved?
  • Are placeholders and variables kept intact, and do they still read correctly after the sentence is reordered? (Arabic word order often differs from English, so a sentence built from fragments can break.)
  • Is any text left in English by mistake?

2. Terminology

  • Does every term match the approved glossary?
  • Is one English term always translated with the same Arabic term?
  • Are product names, feature names and UI labels handled according to the agreed rule (translated, transliterated, or kept in English)?
  • Is the same term used in the interface, the help pages, and the marketing text?

3. Fluency: grammar and spelling

  • Gender and number agreement, including agreement with numbers and counted nouns.
  • Hamza, taa marbuta, alef maqsura and other spelling points.
  • Correct use of the definite article and of prepositions.
  • Punctuation: Arabic comma and question mark versus Latin ones, applied consistently.
  • Tashkeel (diacritics): used only where needed and applied consistently. Example: fathatan sits on the letter before the added alif (كتابًا), not on the alif itself; words ending in taa marbuta take it without an alif (مدرسةً).

4. Style and register

  • Is the register right for the audience (formal Modern Standard Arabic, or a deliberate dialect choice)?
  • Are address forms consistent (singular or plural “you”, masculine or feminine forms)?
  • Does the tone match the brand voice, and are literal English idioms avoided?
  • Do buttons and messages use short, standard UI wording rather than sentence-style translations?

5. Locale conventions

  • Numerals: Western (0-9) or Eastern Arabic-Indic digits, chosen on purpose and consistent within a screen.
  • Date and time format, and which calendar is used.
  • Currency symbol and position, thousands and decimal separators.
  • Units, names, addresses, phone formats.
  • Correct sort order for Arabic lists.

6. Right-to-left layout and mixed text

This is where many Arabic products fail even when the language is good.

  • Is the text direction right for each element, including titles, tables, forms and menus?
  • Do mixed Arabic-English strings (a brand name, a number, a file name, a URL) display in the right order, with punctuation in the right place?
  • Are numbers inside Arabic sentences readable and in the correct order?
  • Are icons and navigation arrows mirrored where they should be, and not mirrored where they should not (for example, media controls and clocks)?
  • Do alignment, padding, and truncation still work with longer Arabic strings?
  • Do letters connect properly, and does the font render Arabic clearly at small sizes?

7. Functional and context checks

  • Does the string make sense on the actual screen, not just in the spreadsheet?
  • Are error messages helpful, with the right tone, and complete?
  • Do the Arabic screenshots or recordings match the strings the reviewer checked?

8. For AI-generated Arabic (an extra layer)

When the Arabic comes from a language model rather than a translator, add these checks:

  • Does the answer contain confident but wrong statements (hallucinations) in Arabic?
  • Does it handle dialect, formal Arabic and mixed prompts as the user expects?
  • Are terms consistent across the conversation and across repeated runs?
  • Does it keep right-to-left formatting in lists, code, tables and citations?
  • Is the tone appropriate and culturally safe for the audience?

Human review does not replace automatic benchmarks. It answers a different question: how the Arabic reads and performs for a real user in your product.

A simple way to score a sample

  1. Pick a representative sample (for example, the screens and strings users meet most often).
  2. Have the reviewer log each error with: location, category, severity, and a suggested fix.
  3. Add up the weighted errors and compare them to your word count or screen count to get a score you can track over time.
  4. Re-check the same sample after fixes. Keep the scoring rule written down. Changing it between rounds makes results impossible to compare.

Common mistakes in LQA projects

  • No glossary or style guide, so the reviewer and the translator disagree about what is “wrong.”
  • Reviewing the text only in a spreadsheet and never in the running product.
  • Treating every error as equal. A minor spelling slip and a wrong meaning on a payment screen are not the same.
  • Skipping right-to-left checks because the strings “look fine” in the translation tool.

FAQ

What does LQA stand for? Linguistic quality assurance (also called linguistic quality assessment): a structured review of language quality using defined error categories and severity levels.

Is Arabic LQA different from proofreading? Yes. Proofreading corrects text. LQA classifies and scores problems against a standard, so a team can decide what to fix first and track quality over time.

How big should the sample be? It depends on the content and the risk. A small sample of the most-used screens can already show whether there is a systematic problem. Ask your reviewer to explain the sampling approach in writing.

Do I need LQA if I already use a good translation vendor? An independent second opinion is often useful because the reviewer has no stake in the original translation.

Can LQA be used for AI output? Yes. The same categories and severities apply, with extra checks for hallucinations, dialect handling and consistency.

Want a second opinion on your Arabic?

If you want a reviewer to look at one product or one set of screens, tell me the product and the Arabic problem you see, and I will start there. Contact me or read about Arabic quality review for software and AI and what I do for localization and AI teams.

I write about Arabic quality and AI in my LinkedIn newsletter.

Discussion on LinkedIn