cldr-segmentation

cldr-segmentation

Text segmentation library for JavaScript.

What is this thing?

This library provides CLDR-based text segmentation capabilities in JavaScript. Text segmentation is the process of identifying word, sentence, and other boundaries in a text. The segmentation rules are published by the Unicode consortium as part of the Common Locale Data Repository, or CLDR, and made freely available to the public.

Why not just split on spaces or periods?

Good question. Most of the time, that'll probably work fine. However, it's not always obvious where words or sentences should start or end. Consider this sentence:

I like Mrs. Murphy. She's nice.

Splitting only on periods will give you ["I like Mrs. ", "Murphy. ", "She's nice."], which probably isn't what you wanted - the period after Mrs doesn't indicate the end of the sentence.

In addition, other languages use different segmentation rules than English. For example, identifying sentence boundaries in Japanese is a little more difficult because sentences tend to end with \u3002 - the ideographic full stop - as opposed to a period. The CLDR contains support for hundreds of languages, meaning you don't have to consider every language when dealing with international text.

Examples

Cldr-segmentation is published as a UMD module meaning it should work in both node via require and the browser via a <script> tag. In the browser, use window.cldrSegmentation to access the library's functionality.

Sentence Segmentation

cldrSegmentation.sentenceSplit("I like Mrs. Murphy. She's nice.");
// => ["I like Mrs. ", "Murphy. ", "She's nice."]

You'll notice that Mrs. was treated as the end of a sentence. To avoid this, use the suppressions for the language you care about. Suppressions are essentially arrays of strings. Each string represents a series of characters after which there should not be a break. Using the English suppressions for the example above yields better results:

var supp = cldrSegmentation.suppressions.en;
cldrSegmentation.sentenceSplit("I like Mrs. Murphy. She's nice.", supp);
// => ["I like Mrs. Murphy. ", "She's nice."]

If you'd like to iterate over each sentence instead of splitting, use a BreakIterator:

var breakIter = new cldrSegmentation.BreakIterator(supp);
var str = "I like Mrs. Murphy, she's nice.";

breakIter.eachSentence(str, (sentence, start, stop) => {
  // do something
});

Suppressions for all languages are available via cldrSegmentation.suppressions.all.

Other Types of Segmentation

Word, line, and grapheme cluster segmentation are supported:

cldrSegmentation.wordSplit("I like Mrs. Murphy. She's nice.");
// => ["I", " ", "like", " ", "Mrs",  ".", " ", "Murphy", ".", "She's", " ", "nice", "."]

Also available are the lineSplit and graphemeSplit functions.

When using a break iterator:

var breakIter = new cldrSegmentation.BreakIterator(supp);
var str = "I like Mrs. Murphy, she's nice.";

breakIter.eachWord(str, (word, start, stop) => {
  // do something
});

Also available are the eachLine and eachGraphemeCluster functions.

Custom Suppressions

Suppressions are just objects with a single shouldBreak function that returns a boolean. The function is passed a cursor object positioned at the index of the proposed break. Cursors deal exclusively with Unicode codepoints, meaning your custom suppression logic will need to be implemented in those terms. For example, let's create a custom suppression function that doesn't allow breaks after sentences that end with the letter 't'.

class TeeSuppression {
  shouldBreak(cursor) {
    var position = cursor.logicalPosition;

    // skip backwards past spaces and periods
    do {
      let cp = cursor.getCodePoint(position);
      position --;
    } while (cp === 32 || cp === 46);

    // we skipped one too many in the loop
    position ++;

    // if the ending character is 't', return false;
    // otherwise return true
    return cursor.getCodePoint(position) !== 116;
  }
}

Note that you don't have to use ES6 classes. It's equally valid to create a simple object:

let teeSuppression = {
  shouldBreak: (cursor) => {
    // logic here
  }
}

Running Tests

Tests are written in Jasmine and can be executed via jasmine-node:

1. npm install -g jasmine-node
2. jasmine-node spec

Authors

Written and maintained by Cameron C. Dutro (@camertron).

License

Copyright 2017 Cameron Dutro, licensed under the MIT license.

Changelog

2.0.2

• Ensure suppressions are preceded by a space, i.e. are whole words.
  • Fixes bug where "Phillip. Is my name." would contain no breaks because 'pp.' is an English suppression.