Javascript RegExp + Word boundaries + unicode characters

There appears to be a problem with Regex and the word boundary \b matching the beginning of a string with a starting character out of the normal 256 byte range.

Instead of using \b, try using (?:^|\\s)

var title = "this is simple string with finnish word tämä on ääkköstesti älkää ihmetelkö";
// Does not work
var searchterm = "äl";

// does not work
//var searchterm = "ää";

// Works
//var searchterm = "wi";

if ( new RegExp("(?:^|\\s)"+searchterm, "gi").test(title) ) {
    $("#result").html("Match: ("+searchterm+"): "+title);
} else {
    $("#result").html("nothing found with term: "+searchterm);   
}

Breakdown:

(?: parenthesis () form a capture group in Regex. Parenthesis started with a question mark and colon ?: form a non-capturing group. They just group the terms together

^ the caret symbol matches the beginning of a string

| the bar is the "or" operator.

\s matches whitespace (appears as \\s in the string because we have to escape the backslash)

) closes the group

So instead of using \b, which matches word boundaries and doesn't work for unicode characters, we use a non-capturing group which matches the beginning of a string OR whitespace.

The \b character class in JavaScript RegEx is really only useful with simple ASCII encoding. \b is a shortcut code for the boundary between \w and \W sets or \w and the beginning or end of the string. These character sets only take into account ASCII "word" characters, where \w is equal to [a-zA-Z0-9_] and \W is the negation of that class.

This makes the RegEx character classes largely useless for dealing with any real language.

\s should work for what you want to do, provided that search terms are only delimited by whitespace.

this question is old, but I think I found a better solution for boundary in regular expressions with unicode letters. Using XRegExp library you can implement a valid \b boundary expanding this

XRegExp('(?=^|$|[^\\p{L}])')

the result is a 4000+ char long, but it seems to work quite performing.

Some explanation: (?= ) is a zero-length lookahead that looks for a begin or end boundary or a non-letter unicode character. The most important think is the lookahead, because the \b doesn't capture anything: it is simply true or false.

\b is a shortcut for the transition between a letter and a non-letter character, or vice-versa.

Updating and improving on max_masseti's answer:

With the introduction of the /u modifier for RegExs in ES2018, you can now use \p{L} to represent any unicode letter, and \P{L} (notice the uppercase P) to represent anything but.

EDIT: Previous version was incomplete.

As such:

const text = 'A Fé, o Império, e as terras viciosas';

text.split(/(?<=\p{L})(?=\P{L})|(?<=\P{L})(?=\p{L})/);

// ['A', ' Fé', ',', ' o', ' Império', ',', ' e', ' as', ' terras', ' viciosas']

We're using a lookbehind (?<=...) to find a letter and a lookahead (?=...) to find a non-letter, or vice versa.

I would recommend you to use XRegExp when you have to work with a specific set of characters from Unicode, the author of this library mapped all kind of regional sets of characters making the work with different languages easier.

Despite the fact the issue seems to be 8 years old, I run into a similar problem (I had to match Cyrillic letters) not so far ago. I spend a whole day on this and could not find any appropriate answer here on StackOverflow. So, to avoid others making lots of effort, I'd like to share my solution.

Yes, \b word boundary works only with Latin letters (Word boundary: \b):

Word boundary \b doesn’t work for non-Latin alphabets The word boundary test \b checks that there should be \w on the one side from the position and "not \w" – on the other side. But \w means a Latin letter a-z (or a digit or an underscore), so the test doesn’t work for other characters, e.g. Cyrillic letters or hieroglyphs.

Yes, JavaScript RegExp implementation hardly supports UTF-8 encoding.

So, I tried implementing own word boundary feature with the support of non-Latin characters. To make word boundary work just with Cyrillic characters I created such regular expression:

new RegExp(`(?<![\u0400-\u04ff])${cyrillicSearchValue}(?![\u0400-\u04ff])`,'gi')

Where \u0400-\u04ff is a range of Cyrillic characters provided in the table of codes. It is not an ideal solution, however, it works properly in most cases.

To make it work in your case, you just have to pick up an appropriate range of codes from the list of Unicode characters.

To try out my example run the code snippet below.

function getMatchExpression(cyrillicSearchValue) {
  return new RegExp(
    `(?<![\u0400-\u04ff])${cyrillicSearchValue}(?![\u0400-\u04ff])`,
    'gi',
  );
}

const sentence = 'Будь-який текст кирилицею, де необхідно знайти слово з контексту';

console.log(sentence.match(getMatchExpression('текст')));
// expected output: ["текст"]


console.log(sentence.match(getMatchExpression('но')));
// expected output: null

I noticed something really weird with \b when using Unicode:

/\bo/.test("pop"); // false (obviously)
/\bä/.test("päp"); // true (what..?)

/\Bo/.test("pop"); // true
/\Bä/.test("päp"); // false (what..?)

It appears that meaning of \b and \B are reversed, but only when used with non-ASCII Unicode? There might be something deeper going on here, but I'm not sure what it is.

In any case, it seems that the word boundary is the issue, not the Unicode characters themselves. Perhaps you should just replace \b with (^|[\s\\/-_&]), as that seems to work correctly. (Make your list of symbols more comprehensive than mine, though.)

I had a similar problem, where I was trying to replace all of a particular unicode word with a different unicode word, and I cannot use lookbehind because it's not supported in the JS engine this code will be used in. I ultimately resolved it like this:

const needle = "КАРТОПЛЯ";
const replace = "БАРАБОЛЯ";
const regex = new RegExp(
  String.raw`(^|[^\n\p{L}])`
    + needle
    + String.raw`(?=$|\P{L})`,
   "gimu",
);

const result = (
    'КАРТОПЛЯ сдффКАРТОПЛЯдадф КАРТОПЛЯ КАРТОПЛЯ КАРТОПЛЯ??? !!!КАРТОПЛЯ ;!;!КАРТОПЛЯ/#?#?'
    + '\n\nКАРТОПЛЯ КАРТОПЛЯ - - -КАРТОПЛЯ--'
  )
    .replace(regex, function (match, ...args) {
      return args[0] + replace;
    });
console.log(result)

output:

БАРАБОЛЯ сдффКАРТОПЛЯдадф БАРАБОЛЯ БАРАБОЛЯ БАРАБОЛЯ??? !!!БАРАБОЛЯ ;!;!БАРАБОЛЯ/#?#?

БАРАБОЛЯ БАРАБОЛЯ - - -БАРАБОЛЯ--

Breaking it apart

The first regex: (^|[^\n\p{L}])

• ^| = Start of the line or
• [^\n\p{L}] = Any character which is not a letter or a newline

The second regex: (?=$|\P{L})

• ?= = Lookahead
• $| = End of the line or
• \P{L} = Any character which is not a letter

The first regex captures the group and is then used via args[0] to put it back into the string during replacement, thereby avoiding a lookbehind. The second regex utilized lookahead.

Note that the second one MUST be a lookahead because if we capture it then overlapping regex matches will not trigger (e.g. КАРТОПЛЯ КАРТОПЛЯ КАРТОПЛЯ would only match on the 1st and 3rd ones).

Trying to find text "myTest":

/(?<![\p{L}\p{N}_])myTest(?![\p{L}\p{N}_])/gu

Similar to NetBeans or Notepad++ form. Trying to find the expression without any letter or number or underscore (like \w characters of word boundary \b) in any unicode characters of letter and number before or after the expression.

My idea is to search with codes representing the Finnish letters

new RegExp("\\b"+asciiOnly(searchterm), "gi").test(asciiOnly(title))

My original idea was to use plain encodeURI but the % sign seemed to interfere with the regexp.

http://jsfiddle.net/7TsxB/5/

I wrote a crude function using encodeURI to encode every character with code over 128 but removing its % and adding 'QQ' in the beginning. It is not the best marker but I couldn't get non alphanumeric to work.

What you are looking for is the Unicode word boundaries standard:

http://unicode.org/reports/tr29/tr29-9.html#Word_Boundaries

There is a JavaScript implementation here (unciodejs.wordbreak.js)

https://github.com/wikimedia/unicodejs

None of the current answers are suitable for general-purpose use, so below is what I'm using.

Important things to note:

• We use positive lookaround for the word sides of the boundary, and negative lookaround for the non-word sides. This is different from using positive lookaround with a negated character class, as the latter won't match the start or end of the input string.
• w is roughly equivalent to the concept of a "word-like character" (including numbers and diacritics), but you might want to use a different definition depending on your use case or the characteristics of your target languages.
• s matches the start of a word, and e matches the end. b matches either and so is the closest in semantics to \b, but usually it's better to just use s or e alone for clarity and performance reasons, because they're mutually exlusive.

// Word character: letter, mark (diacritics), or number.
// Add/remove more characters and character classes as desired,
// e.g. you might want to add _ for greater equivalence with \b
const w = /[\p{L}\p{M}\p{N}]/u.source

// Start of word
const s = new RegExp(`(?:(?<!${w})(?=${w}))`, 'u').source
// End of word
const e = new RegExp(`(?:(?<=${w})(?!${w}))`, 'u').source
// Word break
const b = new RegExp(`(?:${s}|${e})`, 'u').source

console.log('Compiled source:', b)

// Usage:
const regex = new RegExp(`${s}(?:word|单?词)${e}`, 'giu')

const text = `Word more content word-with-hyphen
foreword, wordless, forewordless (not matched)
单词, comma,
end of line: 词
at end: WORD`

console.log('Results:', text.replaceAll(regex, '[[$&]]'))

Depending on your use case, you might also find that using Intl.Segmenter gives the best results. You can even try using a character that isn't present in the input string as a makeshift delimiter, then matching on that:

const DELIM = '⍼'

function addDelims(text) {
    const segments = [...new Intl.Segmenter('en-US', { granularity: 'word' }).segment(text)]
    
    return DELIM + segments.map((s) => s.segment).join(DELIM) + DELIM
}

function stripDelims(text) {
    return text.replaceAll(DELIM, '')
}

const text = `word foreword wordlike WORD 词 单词 word`
const withDelims = addDelims(text)
const replaced = withDelims.replaceAll(new RegExp(`${DELIM}(?:${'word|单?词'})${DELIM}`, 'giu'), '[[$&]]')
const stripped = stripDelims(replaced)

console.log({
    withDelims,
    replaced,
    stripped,
})

I have had a similar problem, but I had to replace an array of terms. All solutions, which I have found did not worked, if two terms were in the text next to each other (because their boundaries overlaped). So I had to use a little modified approach:

var text = "Ještě. že; \"už\" à. Fürs, 'anlässlich' že že že.";
var terms = ["à","anlässlich","Fürs","už","Ještě", "že"];
var replaced = [];
var order = 0;
for (i = 0; i < terms.length; i++) {
    terms[i] = "(^\|[ \n\r\t.,;'\"\+!?-])(" + terms[i] + ")([ \n\r\t.,;'\"\+!?-]+\|$)";
}
var re = new RegExp(terms.join("|"), "");
while (true) {
    var replacedString = "";
    text = text.replace(re, function replacer(match){
        var beginning = match.match("^[ \n\r\t.,;'\"\+!?-]+");
        if (beginning == null) beginning = "";
        var ending = match.match("[ \n\r\t.,;'\"\+!?-]+$");
        if (ending == null) ending = "";
        replacedString = match.replace(beginning,"");
        replacedString = replacedString.replace(ending,"");
        replaced.push(replacedString);
        return beginning+"{{"+order+"}}"+ending;
    });
if (replacedString == "") break;
order += 1;
}

See the code in a fiddle: http://jsfiddle.net/antoninslejska/bvbLpdos/1/

The regular expression is inspired by: http://breakthebit.org/post/3446894238/word-boundaries-in-javascripts-regular

I can't say, that I find the solution elegant...

The correct answer to the question is given by andrefs. I will only rewrite it more clearly, after putting all required things together.

For ASCII text, you can use \b for matching a word boundary both at the start and the end of a pattern. When using Unicode text, you need to use 2 different patterns for doing the same:

• Use (?<=^|\P{L}) for matching the start or a word boundary before the main pattern.
• Use (?=\P{L}|$) for matching the end or a word boundary after the main pattern.
• Additionally, use (?i) in the beginning of everything, to make all those matchings case-insensitive.

So the resulting answer is: (?i)(?<=^|\P{L})xxx(?=\P{L}|$), where xxx is your main pattern. This would be the equivalent of (?i)\bxxx\b for ASCII text.

For your code to work, you now need to do the following:

• Assign to your variable "searchterm", the pattern or words you want to find.
• Escape the variable's contents. For example, replace '\' with '\\' and also do the same for any reserved special character of regex, like '\^', '\$', '\/', etc. Check here for a question on how to do this.
• Insert the variable's contents to the pattern above, in the place of "xxx", by simply using the string.replace() method.

bad but working:

var text = " аб аб АБ абвг ";
var ttt = "(аб)"
var p = "(^|$|[^A-Za-zА-Я-а-я0-9()])"; // add other word boundary symbols here
var exp = new RegExp(p+ttt+p,"gi");
text = text.replace(exp, "$1($2)$3").replace(exp, "$1($2)$3");
const t1 = performance.now();
console.log(text);

result (without qutes):

" (аб) (аб) (АБ) абвг "  

I struggled hard on this. Working with French accented characters, and I managed to find this solution :

const myString = "MyString";
const regex = new RegExp(
    "(?:[^À-ú]|^)\\b(" + myString + ")\\b(?:[^À-ú]|$)",
    "ig"
);

What id does : It keeps checking word-boundaries with \b before and after "MyString". In addition to that, (?:[^À-ú]|^) and (?:[^À-ú]|$) will check if MyString is not surrounded by any accented characters

It will not work with cyrillic but it may be possible to find the range of cirillic charactes and edit [^À-ú] in consequence.

Warning, it captures only the group (MyString) but the total match contains previous and next characters

See example : https://regex101.com/r/5P0ZIe/1

Match examples :

• MyString
  • match : "MyString"
  • group 1 : "MyString"
• Lorem ipsum. MyString dolor sit amet
  • match : " MyString "
  • group 1 : "MyString"
• (MyString)
  • match : "(MyString)"
  • group 1 : "MyString"
• BetweenCharactersMyStringIsNotFound
  • match : Nothing
  • group 1 : Nothing
• éMyStringé
  • match : Nothing
  • group 1 : Nothing
• ùMyString
  • match : Nothing
  • group 1 : Nothing
• MyStringÖ
  • match : Nothing
  • group 1 : Nothing