Glossary · Arabic & RTL

One word, several spellings.

Arabic readers see أحمد and احمد as the same name. A computer sees two different strings. Normalization is the small step that closes that gap before you search, merge duplicates or match records.

Updated 19 November 2026Sources checked 16 November 2026Reviewed by Eng. Hasan Kassas

What it means

Arabic text normalization is the step that rewrites Arabic text into one consistent spelling before it is searched, compared or indexed. It folds variants that readers treat as the same word, such as أ, إ and آ into ا, ة into ه and ى into ي, and removes diacritics and tatweel, so a search for احمد also finds أحمد.

The name covers two related ideas. Unicode normalization, defined in UAX #15, deals with characters that can be encoded in more than one way: under the NFKC form, a presentation form such as the lam-alef ligature ﻻ becomes the two ordinary letters لا. Arabic orthographic normalization goes further and merges letters that Unicode treats as genuinely different, such as alef with hamza and bare alef. No Unicode form will ever make أ equal ا, which is why search code adds its own rules on top.

Both work on a copy of the text, never on what people read. The original keeps its hamzas and diacritics for display; the normalized copy exists only to be matched. And one function has to run on both sides: on the stored text when it is saved, and on every query as it comes in.

Names

What it is called

NameNotes
EnglishArabic text normalizationBritish spelling: normalisation. Search engineers also say "folding" or "orthographic normalization".
Modern Standard Arabicتوحيد النص العربيAlso «توحيد الإملاء» when the focus is on spelling variants.
Also heard in Syria«توحيد الهمزات», normalizeDevelopers often use the English verb; the hamza rule is usually the first one people name.
In code or standardsnormalize("NFKC"), UAX #15Unicode's normalization forms. A function such as normalizeArabic adds the Arabic folding rules on top.
The variants

What usually gets folded

  • Hamza and madda on alef. أ (hamza above), إ (hamza below), آ (madda) and ٱ (wasla) all become a bare ا, so إدلب and ادلب meet. The hamza is often left out in quick typing.
  • Ta marbuta. The tied ta ة at the end of words like مدرسة is folded into ه, the letter it is most often swapped with at the end of a word.
  • Alef maqsura. A final ى, as in مصطفى, becomes ي, since the two are often confused at the end of a word.
  • Hamza on waw and ya. ؤ becomes و and ئ becomes ي, so مؤسسة matches موسسه. This rule is optional: skip it if your users rely on the difference.
  • Tashkeel. The short-vowel marks fatha, damma and kasra, plus shadda, sukun and tanwin, are removed, so قِطْعَة becomes قطعه.
  • Tatweel. The stretching stroke ـ that lengthens words in headlines and ads, as in سيـــارة, is deleted.
  • Presentation forms. Text copied from old documents can contain ligature code points such as ﻻ; NFKC turns them back into ordinary letters.
  • Persian letters and digits. ک and ی from Persian keyboard layouts become ك and ي, and both sets of Arabic-Indic digits become 0 to 9; see Arabic-Indic digits.

An example from a Syrian app

A landlord on a property app in Damascus posts «شقّة للإيجار في المزّة», with a shadda on two letters and a hamza under the alef. A student looking for a room types «شقه ايجار المزه» on a phone. As raw text, not one word of the query appears in the listing.

Run both through the same normalizer and tokenizer and the listing becomes شقه, ايجار, في, مزه, while the query becomes شقه, ايجار, مزه: every word of the query is now in the listing. The shaddas are gone, ة became ه, the hamza fell off the alef, and the article ال and the prefix لل were trimmed from the start of words. That last step goes beyond normalization in the strict sense, but the tested helper does it in the same pass.

The listing page still shows «شقّة للإيجار في المزّة» exactly as the landlord wrote it.

Pitfalls

Common mistakes

  • Showing the normalized text. مدرسه and مستشفي work as search keys and look wrong on screen. Keep the original for display.
  • Normalizing one side only. If listings are folded and queries are not, or the reverse, matches still fail. Share one function between both.
  • Folding at query time. Normalizing every stored row during each search is slow on a phone and on a server; save the normalized copy in its own column and index that.
  • Expecting a stemmer. Normalization does not connect كتاب with كتب. Relating word forms is a separate job with its own errors.
  • Expecting transliteration. ايفون and iPhone stay different strings; a short synonym list covers brand names typed in both scripts.
  • Different rules on web and mobile. If the website and the app fold text differently, a search works on one and fails on the other. Our JavaScript and Dart versions share the same rules (checked with Node 24 and flutter test on 16 November 2026); Dart has no built-in NFKC, so it maps the lam-alef forms by hand.
FAQ

Common questions

Rewriting Arabic text into one standard spelling before matching it: hamza forms of alef become ا, ة becomes ه, ى becomes ي, and diacritics and tatweel are removed. It runs on a copy used for search, not on the text people read.

Sources

Where these facts come from

  1. Arabic block chart (U+0600 to U+06FF)Unicode Consortium · accessed 16 Nov 2026
  2. UAX #15: Unicode Normalization FormsUnicode Consortium · accessed 16 Nov 2026
  3. String.prototype.normalize()MDN Web Docs · accessed 16 Nov 2026

The folding rules are those in our tested JavaScript and Dart helpers (Node 24 and flutter test, 16 November 2026). The property-app example was run through the same JavaScript helper.

Users say search cannot find things?

Send us a few searches that should work. We will show you which spelling gets in the way.