What’s new in babel 25.8

2025-04-29

⚠ This release introduces a potentially breaking change. Please, read the section Metadata below.

Hebrew justification

Only lualatex.

Hebrew justification, unlike many Latin-based scripts that primarily adjust the spacing between words, frequently utilizes subtle variations in the spacing between individual letters within words. This can be achieved by inserting tiny amounts of extra space between letter combinations.

The transform justification.interletter in all locales in the Hebrew script (Hebrew, Ancient Hebrew, Ladino, Yiddish) activates this kind of justification, by inserting a \nobreak and a zero width glue with a ‘plus’ of .5 times the ‘plus’ inter-word space before a letter when preceded by another letter or a diacritic (remember in Unicode combining chars are placed after the corresponding letter). Punctuation spacing is left unchanged.

This interletter adjustment should be carefully controlled to avoid disrupting the natural flow and readability of the Hebrew script. It depends on several factors, like the font or the text width. There are three tools:

• \SetTransformValue{hebrew}{interletter.plus}{<value>}, which sets the factor for the glue ‘plus’ value. By default, .5. (Use the corresponding language name.)
• WordSpace as a fontspec option. This is not a babel feature, but is mentioned because it can be useful.
• The locale option justification=unhyphenated, which is a stronger version of \sloppy.

In addition, you can consider using microtype, too.

Here is an example:

\documentclass{article}

\usepackage[hebrew,
  provide={ transforms=justification.interletter }, 
  bidi=basic]{babel}
\babelfont{rm}[WordSpace={1,2,1}]{FreeSerif}

\SetTransformValue{hebrew}{interletter.plus}{.7}

\setlength{\textwidth}{4.5cm}
\setlength{\parindent}{0pt}

\begin{document}

בְּרֵאשִׁית בָּרָא אֱלֹהִים אֵת הַשָּׁמַיִם וְאֵת הָאָרֶץ׃ וְהָאָרֶץ
הָיְתָה תֹהוּ וָבֹהוּ וְחֹשֶׁךְ עַל־פְּנֵי תְהוֹם וְרוּחַ אֱלֹהִים
מְרַחֶפֶת עַל־פְּנֵי הַמָּיִם׃ וַיֹּאמֶר אֱלֹהִים יְהִי אוֹר וַיְהִי
אוֹר׃ וַיַּרְא אֱלֹהִים אֶת־הָאוֹר כִּי־טוֹב וַיַּבְדֵּל אֱלֹהִים בֵּין
הָאוֹר וּבֵין הַחֹשֶׁךְ׃ וַיִּקְרָא אֱלֹהִים לָאוֹר יוֹם וְלַחֹשֶׁךְ
קָרָא לָיְלָה וַיְהִי־עֶרֶב וַיְהִי־בֹקֶר יוֹם אֶחָד׃

\end{document}

Metadata

Babel now sets the main language based on the document metadata. Basically, what it does is to convert something like:

\DocumentMetadata{lang=da}
...
\usepackage[english]{babel}

to

\DocumentMetadata{lang=da}
...
\usepackage[main=danish, english, danish]{babel}

The basic tag lookup explained in the babel manual is applied here, so that fr-Latn-FR is valid and mapped to fr, which is in turn mapped to french, while en-Latn-US is mapped to en-US, which is mapped to american. This eases the localization of automatically generated documents.

Note the information in the ini file is kept. So, lang=es-CO is (currently) mapped to spanish and the field tag.bcp47 is still es, because it’s the resolved locale. However, \GetDocumentProperties{document/lang} returns es-CO, because it stores, as far as babel is concerned, the requested locale.

Tags are not validated, which means mistakes like en-UK or wrong tags like de-AUS or la-x-classical might be accepted without complaining.

This is a breaking change. But \DocumentMetadata is a recent LaTeX feature, and problems should be minimal. Moreover, the short example above didn’t make much sense, because settings are contradictory. Note also with the lang metadata, and if there are no other language declarations as class or package options, the language loaded is that set as metadata instead of the dummy language nil.

Note also setting the lang metadata and a language as a class option at the same time is not well defined, even without babel (which explains why main is set explicitly in the example, even if usually discouraged).

As part of this change, there is some cleanup and refactoring related to locales tags, which is a work in progress. Also, the info in ini files is now read when the ldf starts (instead of at the end of the babel package). \BabelEnsureInfo is now no-op, because it’s loaded always.