Hyphenate🔗ℹ

A simple hyphenation engine that uses the Knuth–Liang hyphenation algorithm originally developed for TeX. I have added little to their work. Accordingly, I take little credit.

1 Installation🔗ℹ

2 Importing the module🔗ℹ

The module can be invoked two ways: fast or safe.

Fast mode is the default, which you get by importing the module in the usual way: (require hyphenate).

Safe mode enables the function contracts documented below. Use safe mode by importing the module as (require (submod hyphenate safe)).

3 Interface🔗ℹ

The REPL displays a soft hyphen as \u00AD. But in ordinary use, you’ll only see a soft hyphen when it appears at the end of a line or page as part of a hyphenated word. Otherwise it’s not displayed. In most of the examples here, I use a standard hyphen for clarity (by adding #\- as an argument).

The #:min-left-length and #:min-right-length keyword arguments set the minimum distance between a potential hyphen and the left or right ends of the word. The default is 2 characters. Larger values will reduce hyphens, but also prevent small words from breaking. These values will override a smaller #:min-length value.

Another way of controlling hyphen frequency is with the #:min-hyphens keyword argument, which sets the minimum number of hyphens in a hyphenatable word. (It has no effect on non-hyphenatable words.) The default is 1 hyphen. Larger values will reduce hyphens, but also prevent small words from breaking.

Because the hyphenation is based on an algorithm rather than a dictionary, it makes good guesses with unusual words:

Using the #:exceptions keyword, you can pass hyphenation exceptions as a list of words with hyphenation points marked with regular hyphens ("-"). If an exception word contains no hyphens, that word will never be hyphenated.

Knuth & Liang were sufficiently confident about their algorithm that they originally released it with only 14 exceptions: associate[s], declination, obligatory, philanthropic, present[s], project[s], reciprocity, recognizance, reformation, retribution, and table. Admirable bravado, but it’s not hard to discover others that need adjustment.

The Knuth–Liang algorithm is designed to omit legitimate hyphenation points (i.e., generate false negatives) more often than it creates erroneous hyphenation points (i.e., false positives). This is good policy. Perfect hyphenation — that is, hyphenation that represents an exact linguistic syllabification of each word — is superfluous for typesetting. Hyphenation simply seeks to mark possible line-break and page-break locations for whatever layout engine is drawing the text. The ultimate goal is to permit more even text flow. Like horseshoes and hand grenades, close is good enough. And a word wrongly hyphenated is more likely to be noticed by a reader than a word inefficiently hyphenated.

For this reason, certain words can’t be hyphenated algorithmically, because the correct hyphenation depends on meaning, not merely on spelling. For instance:

This is the right result. If you used adder to mean the machine, it would be hyphenated add-er; if you meant the snake, it would be ad-der. Better to avoid hyphenation than to hyphenate incorrectly.

You can send HTML-style X-expressions through hyphenate. It will recursively hyphenate the text strings, while leaving the tags and attributes alone, as well as non-hyphenatable material (like character entities and CDATA).

Don’t send raw HTML or XML through hyphenate. It can’t distinguish tags and attributes from textual content, so everything will be hyphenated, thus goofing up your file. But you can easily convert your HTML or XML to an X-expression, hyphenate it, and then convert back.

If you’re working with HTML, be careful not to include any <script> or <style> blocks, which contain non-hyphenatable data. You can protect that data by using the #:omit-txexpr keyword to specify a txexpr-test. The test will be applied to all tagged X-expressions (see txexpr?). When txexpr-test evaluates to true, the item will be skipped.

You can also use #:omit-txexpr to omit tagged X-expressions with particular attributes. This can be used to selectively suppress hyphenation at the markup level.

Similarly, you can use the #:omit-word argument to avoid words that match word-test. Convenient if you want to prevent hyphenation of certain sets of words, like proper names:

Sometimes you need #:omit-word to prevent unintended consequences. For instance, if you’re using ligatures in CSS, certain groups of characters (fi, fl, ffi, et al.) will be replaced by a single glyph. That looks snazzy, but adding soft hyphens between any of these pairs will defeat the ligature substitution, creating inconsistent results. With #:omit-word, you can skip these words:

“Wouldn’t it be better to exclude certain pairs of letters rather than whole words?” Yes. But for now, that’s not supported.

A side effect of using hyphenate is that soft hyphens (or whatever the joiner is) will be embedded in the output text. If you need to support copying of text, for instance in a GUI application, you’ll probably want to strip out the hyphenation before the copied text is moved to the clipboard.

Use this function cautiously — if joiner appeared in the original input to hyphenate, the output from unhyphenate won’t be the same string.

Keep in mind that soft hyphens could appear in your input string. Certain word processors allow users to insert soft hyphens in their text.

4 French🔗ℹ

French hyphenation is available by importing the module as hyphenate/fr or (submod hyphenate/fr safe) and using the hyphenate function normally. Below, notice that the word “formidable” hyphenates differently in French.

The two languages are in separate submodules for performance reasons. That way, they can maintain separate caches of hyphenated words.

There is no way to use hyphenate in “polyglot” mode, where English and French are detected automatically. It is possible, however, to mix both the English and French hyphenate functions in a single file, and apply them as needed. To avoid a name conflict between the two hyphenate functions, you’ll need to use prefix-in:

5 License & source code🔗ℹ

This module is released under the MIT license.

Source repository at http://git.matthewbutterick.com/mbutterick/typesetting. Suggestions & corrections welcome.