Considerations When Choosing Fonts for a Multilingual Website

Avatar of Omar Bourhaouta
Omar Bourhaouta on

AWS Amplify - the fastest, easiest way to develop mobile and web apps that scale.

As a front-end developer working for clients all over the world, I’ve always struggled to deal with multilingual websites — especially cases where both right-to-left (RTL) and left-to-right (LTR) are used. That said, I’ve learned a few things along the way and am going to share a few tips in this post.

Let’s work in Arabic and English, not just because Arabic is my native language, but because it’s a classic example of RTL in use.

Adding RTL support to a site

Before this though, we’ll want to add support for an RTL language on our site. There are two ways we can go about this, both of which aren’t exactly ideal.

Not ideal: Use a specific font for each direction

The first way we could go about this is to rely on the direction (dir) attribute on any given element (which is usually the <html> element so it sets the direction globally):

/* For LTR, use Roboto */
[dir=ltr] body {
  font-family: "Roboto", sans-serif;

/* For RTL, use Amiri */
[dir=rtl] body {
  font-family: "Amiri", sans-serif;

PostCSS-RTL makes it even easier to generate styles for each direction, but the issue with this method is that you are only using one font which is not ideal if you have multiple languages in one paragraph.

Here’s why. You’ll find that multi-lingual paragraphs mess up the UI because the Arabic glyphs are given a default font that doesn’t align with the element.

It might be worse in some browsers over others.

Also not ideal: Use one single font that supports both languages

The second option could be using fonts that offer support for both directions. However, in my personal experience, using just one font for both languages limits creativity and freedom to use a different font for each direction. It might not be that bad, depending on the design requirements. But I’ve definitely worked on projects where it makes a difference.

OK, so what then?

We need a simpler solution. According to MDN:

Font selection does not simply stop at the first font in the list that is on the user’s system. Rather, font selection is done one character at a time, so that if an available font does not have a glyph for a needed character, the latter fonts are tried.

Meaning, we can still use the font-family property but using a fallback in cases where the first font doesn’t have a glyph for a character. This actually solves both of the issues above, two birds with one stone style!

body {
  font-family: 'Roboto', 'Amiri', 'Galada', sans-serif;

Why does this work?

Just like the way flexbox and CSS grid, make CSS layouts a lot more flexible, the font matching algorithm makes it even easier to work with content in different languages. Here’s what W3C says about it matching characters to fonts:

When text contains characters such as combining marks, ideally the base character should be rendered using the same font as the mark, this assures proper placement of the mark. For this reason, the font matching algorithm for clusters is more specialized than the general case of matching a single character by itself. For sequences containing variation selectors, which indicate the precise glyph to be used for a given character, user agents always attempt system font fallback to find the appropriate glyph before using the default glyph of the base character.

(Emphasis mine)

And how are fonts matched? The spec outlines the steps the algorithm takes, which I’ll paraphrase here.

  • The browser looks at a cluster of text and tries to match it to the list of fonts that are declared in CSS.
  • If it finds a font that supports all of the characters, great! That’s what gets used.
  • If the browser doesn’t find a font that supports all of the characters, it re-reads the list of fonts to find one that supports the unmatched characters and applies it to those specific characters. 
  • If the browser doesn’t find a font in the list that matches neither all of the characters in the cluster nor individual ones, then it reaches for the default system font and checks that it supports all of the characters.
  • If the default system font matches, again, great! That’s what gets used.
  • If the system font doesn’t work, that’s where the browser renders a broken glyph.

Let’s talk performance

The sequence we just looked at could be taxing on a site’s performance. Imagine the browser having to loop through every defined fallback, match specific characters to glyphs, and download font files based on what it finds. That can add up to a lot of work, not to mention FOUT and other rendering weirdness.

The goal is to let the font matching algorithm decide which font to apply to each text instead of relying on one font for both languages or adding extra CSS to handle different directions. If a font is never applied to anything (say a particular page is in RTL and happens to not have any LTR text on it, or vice versa) the font further down the stack that isn’t used is never downloaded.

Making that happen requires selecting good multilingual fonts. Good multilingual fonts are ones that have glyphs for as many characters you anticipate using on a page. And if you are unable to find one that supports them all, using one that supports most of them and then falling back to another font that does is an efficient way to go. If that happens to be the default system font, that’s just as great because it’s one less downloaded font file.

The good thing about letting the font-family property decide the font for each glyph (instead of making extra CSS selectors for each direction) is that the behavior is already there as we outlined earlier — we simply need to make use of it.