Khmer writing check
Spelling underlined where it is wrong, and the faults a dictionary would never catch.
Which dictionary, and why that matters
Spelling here is checked against 75,056 Khmer words from the SBBIC spelling list — built from the Chuon Nath and Headley dictionaries together with the KhmerOS and SBBIC word lists, and the same list behind the Khmer spelling checker for LibreOffice. It follows Chuon Nath, which is what Cambodian schools follow. The list is published under the GPL and ships here unchanged, with its notice at /khmer/LICENCE.txt.
Naming it is not a formality. A Khmer word list is a set of decisions about spellings people genuinely disagree on — ជំរុញ or ជម្រុញ, and a hundred more — so a checker that will not say whose list it uses is really saying "wrong because I say so". You can see the convention this one applies and decide whether it is yours.
A word the list does not hold is underlined, and that is not the same as a word being wrong: names, places and loanwords are not in any dictionary. The tool says which it is rather than guessing, and offers only spellings that are genuinely in the list — never anything invented.
The half a dictionary cannot see
A spelling checker reads words. Most of what goes wrong in Khmer documents is not a word at all: a space in front of ។, a full stop typed where ។ belongs because the keyboard was still in English, a vowel keyed before the subscript it belongs after, one number written half in Khmer digits and half in Arabic. Every one of those is invisible to the person who typed it, and every one is checked here as well.
That is why the two halves belong on one page. The dictionary catches the misspelt word; the rules catch the sentence that will break when somebody else opens it.
The space before ។
The most common fault in typed Khmer, and the one that survives longest. ។ attaches to the word before it with no gap. Put a space there and the mark becomes a separate item — so when the line wraps, ។ falls to the start of the next line on its own.
It looks like nothing on the screen you typed it on, at the width you typed it at. It shows up on somebody else’s phone, in a PDF, or in print, which is to say after it is too late to be embarrassing quietly.
A full stop where ។ belongs
This one is a keyboard slip: the layout was still in English when the sentence ended. A full stop after Khmer is nearly invisible to read past, and it means the sentence never ends as far as anything parsing the text is concerned — search, screen readers, and anything counting sentences.
The same goes the other way for numbers. ០ and 0 are different characters, so a figure written half in Khmer digits and half in Arabic ones will never match itself. That is how a phone number in a form fails to find its own record.
Word spacing, without sending your text anywhere
Khmer is written with no gap between words, so anything laying out a line has to know where the words end or it will break one in half. The fix is a zero-width space at each boundary — invisible, and the only place a break is then allowed.
Browsers no longer need the help: measured in Chrome, the same paragraph breaks at exactly the same points with and without them, because the browser carries a Khmer dictionary of its own. What still needs them is everything else — Word, PDF exports, phone apps, email clients, and anything that counts or indexes words. That is where a document goes wrong after it leaves your screen.
Finding those boundaries is normally done on a server: you paste your text, it is posted to an API, and the marked-up version comes back. There is no need for any of that. Every current browser already carries ICU’s Khmer dictionary, because it needs one to break lines at all, and this page asks the browser directly. The words it found are listed as chips so you can check the segmentation by eye before you accept it.
Fixing, carefully
The button puts right only what has one correct answer: spaces, doubled spaces, punctuation marks and the space that belongs after a sentence. It never touches a word.
That restraint is the point. A spelling is never changed in a batch: you click the word, you see what the dictionary offers, and you choose. Everything the tool cannot put right exactly is listed for a person to decide, because guessing quietly is the one thing a checker must never do.