View on GitHub

Babel

The multilingual framework to localize LaTeX, LuaLaTeX, XeLaTeX

Non-standard hyphenation with luatex

Hyphenation in LaTeX is accomplished by means of the so-called discretionaries. You can find a brief description here.

This article describes an extension which can serve to several purposes, particularly dealing with non-standard hyphenation rules, including changes in letters and weighted hyphenation points. (Note luatex currently provides built-in ways the deal with some frequent cases, too. Please, refer to its manual for further information.)

The basic syntax is explained in the babel manual. This article complements it with an explanation of the second and third arguments of \babelposthyphenation, which also applies to \babelprehyphenation with the differences explained in the manual.

Here is a simple example of a declaration, which tells LaTeX to change the group ‘ck’ to ‘kk’ with an optional hyphenation point inside this group (it’s not meant as a full or realistic rule for German, but just a starting point; a more detailed rule is shown below).

\babelposthyphenation{german}{ck}{
  { no = c, pre = k- },
  {}
}

It consists of:

The language here refers to a set of hyphenation rules, ie, to \language. So, the first letter in the pattern is replaced with the first item in the list, the second letter with the second item and so on. (This is not strictly true, because the replace list is filled with nil’s if shorter.)

Replacement list

The items in the replacement list are the following:

  1. An empty group {} leaves the corresponding item untouched. For example, in the rule above the ‘k’ in the pattern (the second element) provides the context, because in the replacement list the second item is {} and therefore the character is just kept.
  2. A list like { no = c, pre = k-, post = } replaces the letter by the corresponding discretionary. Only one of the keys is necessary, and the rest defaults to empty. By default the penalty is \hyphenpenalty or \exhyphenpenalty (TeXbook, p96), but a different value can be set with the key penalty. A further field is data - automatic hyphens contain no information about the font and the like, and with this key you can set which element in the list (as captured) they will the taken from. In the rule above the ‘c’ is replaced by a discretionary, but no data is required because the item to be replaced is a character, which already contains the required data. (Remember discretionaries are not allowed in \babelprehyphenation.)
  3. The key string replaces the character with the string. If empty, the item (in TeX jargon, the node) is removed; to insert chars, just use a multi-character string. The nodes created are literal copies of the original (the same font, language, and so on), but with the new characters.
  4. With remove the node is, well, removed. A synonymous is string=.
  5. Spaces are declared with something like space =.2 .1 0. The values are in em units, and they are the natural width, the plus, and the minus. Here, you may need data, too. With spacefactor the unit is the font size of the current font (if the node is a glyph; you may need a data= pointing to a specific glyph).
  6. Penalties are declared with penalty.

Some keys can be used in conjunction with insert, which must be the very first one in the replacement. With it the item is not replaced, but inserted. The following rule inserts a penalty in the middle of the group ‘ff’:

\babelposthyphenation{english}{ ff }
  { {},
    { insert, penalty = 10 },
    {}
  }

In the replacement list, there is an extended syntax which allows to map the captured characters (see below). For example, {2|ΐΰῒῢ|ίύὶὺ} means: if the second captured char is ΐ replace it with ί, ύ with ύ, and so on. This feature is particularly useful when a letter changes if there is a hyphen, and also when transliterating. Here is a partial example of the latter (the full example is here, with digraphs and trigraphs):

\babelprehyphenation{transrussian}
  {([ABVGDEËZIJKLMNOPRSTUFHÈY"abvgdeëzijklmnoprstufhèy'])}{
  string = {1|ABVGDEËZIJKLMNOPRSTUFHÈY"abvgdeëzijklmnoprstufhèy'%
             |АБВГДЕЁЗИЙКЛМНОПРСТУФХЭЫЬабвгдеёзийклмнопрстуфхэыь}
}

babel traverses the strings to be processed with the help of a pointer. Another key available in the replacements is step = <num>, which moves this pointer forward (if positive) or backwards (if negative). By default it’s, of course, 0, which leaves the pointer just after the last replacement. It can be set in any non-empty replacement (eg, { string = a, step = -1 }).

A further key is kashida, for Arabic justification. See What’s new in babel 3.59.

Patterns

‘Regular’ hyphenation points, as inserted automatically by the hyphenation algorithm, are entered in the pattern as vertical bars (|), as the short examples below show. Explicit hyphens are entered as =. Spaces are allowed for clarity, and they are discarded. If you are not sure where the hyphenation points fall, use \showhyphens. (Also, remember | in \babelprehyphenation is a space.)

The pattern is matched with lua empty captures, which are automatically added before and after the string. You may set explicitly different empty captures, to reduce the number of items in the replacement list:

\babelprehyphenation{english}{very()long()pattern}{
  string = L,
  string = OOO,
  string = N,
  string = G
}

With this rule, the string ‘verylongpattern’ is replaced with ‘veryLOOONGpattern’.

Lua patterns with dots, characters classes (with %, but see below for an alternative TeX-friendly syntax) and char-sets (with [], including complementing and ranges) are allowed, too. When using the dot, be aware it matches | and =, too. +, -, ? and * are allowed outside the ()…() block, but not inside. So, {a}|?()Á() is a letter followed optionally by a discretionary, but only Á is actually transformed (in these cases, you may want to go back with the key step).

Captures with () are allowed, too. Ordinary captures are allowed inside the empty captures (they must resolve to exactly one character). In the pattern, the syntax {n} is a backreference matching the n-th capture inside the empty captures. This syntax can be used in the replacement strings, with the corresponding capture:

\babelposthyphenation{german}
  { [AEIOUÄÖÜaeiouäöü] ([cC]) ([kK]) [AEIOUÄÖÜaeiouäöü] }{
    {},                        % Keep first vowel
    { no = {1}, pre = {2}- },  % c or C → discretionary
    {},                        % Keep k or K             
    {}                         % Keep second vowel
}

There are two captures, namely, [cC] (which means either ‘c’ or ‘C’) and [kK], used in the second replacement as {1} and {2}. With this rule, \showhyphens{Trockenerzeugnis Druckeinstellung} will display something like:

Underfull \hbox (badness 10000) in paragraph at lines 15--15
[] \TU/lmr/m/n/10 Trok-ken-erzeug-nis  Druck-ein-stel-lung

Note \showhyphens actually hyphenates the first word, and therefore the rule is applied. On the other hand, discretionaries are taken into account, which means the rule isn’t applied to Druckeinstellung because the second vowel doesn’t immediately follow the ‘k’ (there is a soft hyphen in between). Very often what we need is a combination of hyphenation patterns with post-hyphenation rules. (Remember also german isn’t current German, but the 1901 variant.)

Another example is:

\babelposthyphenation{german}{([fmtrp]) | {1}}{
  { no = {1}, pre = {1}{1}- }, 
  remove,
  {}
}

No attempt has been done here to follow the full German rules. For a more realistic example of double consonants, in Norwegian, see the guide for this language.

Since the percent sign has a quite different meaning in lua and tex, as a convenience the {} syntax can be used to enter character classes in the pattern, too (ie, {d} becomes %d, but note {1} is not internally the same as %1).

Another extension with {} is the possibility to enter a character by its hex code (at least 4 digits). So, {007C} and {003D} are the characters ‘|’ and ‘=’, in case you need their literal meaning.

Short examples