Skip to main content
halfSpace inserts the Zero-Width Non-Joiner (‌, U+200C, نیم‌فاصله) between Persian words and their attached prefixes/suffixes — turning می خواهم into the typographically correct می‌خواهم.

​
Function

halfSpace(persianText: string): string

import { halfSpace } from "@persian-tools/persian-tools";

halfSpace("می خواهم بروم"); // "می‌خواهم بروم"
halfSpace("کتاب ها"); // "کتاب‌ها"
halfSpace("بزرگ ترین"); // "بزرگ‌ترین"
halfSpace("نمی توانم"); // "نمی‌توانم"
The library exports only halfSpace. There is no addHalfSpaceChar or removeHalfSpaceChar.

​
What it does

The input is tokenised by whitespace; consecutive spaces are first collapsed. For each pair of adjacent tokens, three rule families are tried (in priority order):
  1. Compound rules — comparative/superlative formation (بزرگ تر → بزرگ‌تر, بزرگ ترین → بزرگ‌ترین).
  2. Suffix rules — common plural / possessive endings (کتاب ها → کتاب‌ها, خانه ام → خانه‌ام).
  3. Prefix rules — modal/verbal prefixes (می خواهم → می‌خواهم, نمی توانم → نمی‌توانم).
If no rule fires for a pair, the space stays as-is.

​
What it does NOT do

  • Remove existing ZWNJ — halfSpace only inserts. To strip ZWNJ, do s.replace(/‌/g, " ") in your own code.
  • Validate Persian-ness — non-Persian tokens pass through unchanged.
  • Fix Arabic-keyboard input — run toPersianChars or autoArabicToPersian first so the rule tables (which key off Persian code points) actually match.
  • Operate inside a single word with no space — only space-separated tokens are processed.
import { halfSpace, autoArabicToPersian, autoConvertDigitsToEN } from "@persian-tools/persian-tools";

const polish = (s: string) => halfSpace(autoArabicToPersian(autoConvertDigitsToEN(s)));

​
Why ZWNJ matters

FormIssue
می خواهم (full space)reads as two separate words
میخواهم (no space)glyphs join visually, unreadable
می‌خواهم (ZWNJ)correct: visually separate, treated as one word

​
Pitfalls

  • The rules are conservative — they don’t try to repair pre-existing wrong ZWNJ usage.
  • Stripping whitespace with replace(/\s/g, "") may or may not consume ZWNJ depending on the JS engine — be explicit with ‌.
  • For multi-MB documents, chunk the input by paragraph; tokenisation is O(n) per chunk.

​
Source

src/modules/halfSpace/index.ts, src/modules/halfSpace/utils.ts · Tests: test/halfSpace.spec.ts