Arabic Letters Hold Hands. Your Computer Might Not Know That.
You know how Scrabble tiles sit there, separate and self-contained? Each letter its own little rectangle, minding its own business. Arabic doesn't work that way. Arabic letters hold hands. They join, shift shape, merge into ligatures depending on who their neighbors are. A letter at the start of a word looks different from the same letter in the middle, or at the end, or standing alone. Get the joining wrong and you don't have misspelled text. You have gibberish that looks like letters fell out of a word and scattered across the floor.
This turns out to be a surprisingly hard problem to solve at scale. And the history of how digital text has handled it is full of choices that word nerds will find either fascinating or maddening, depending on how much you enjoy the idea that language refuses to be simple.
The 1991 Problem Nobody Fixed Until It Was Too Late
Here's a puzzle: what do you do when you have two sets of Unicode codepoints representing Arabic characters, one set from 1991 and one from 1995, and someone runs a data import that grabs the wrong vintage?
You get twelve thousand names encoded with fossil codepoints from 1991 instead of the regular ones from 1995. A 2017 data import did exactly this. The regular Arabic codepoints had existed for over two decades by then. The fossil ones predated them by four years. Both technically in the Unicode standard. One of them making rendering engines do increasingly confused things.
Unicode introduced regular Arabic codepoints in 1995. The fossil versions date from 1991. Using the fossil variants in modern contexts is the rough equivalent of writing someone's address with an obsolete postal format: technically traceable, but likely to get mangled somewhere in transit.
When Letters Won't Join
Arabic text flows from the right margin, with the rag falling on the left. That's already a mental flip if you grew up reading English. But the joining is the really interesting part. Arabic text depends on shaping engines, software that looks at each character in context and decides what form it takes and how it connects to its neighbors.
If your PDF library predates shaping engine support in its language runtime, you get Arabic letters that render unjoined. Every character isolated. The visual equivalent of reading every word with spaces between each letter. Technically the right letters. Completely wrong as text.
The World Digital Library, which localized its platform into Arabic and had member libraries across multiple regions, ran into exactly this kind of infrastructure problem. Building for a global audience means inheriting every assumption baked into your tools, including assumptions made before anyone thought hard about bidirectional text.
Arabic Isn't Even the Only One
About 100 million people read Javanese. Javanese script, like Arabic, requires connecting ligatures to be coherent. You can't just line up discrete glyphs and call it text. The characters need to flow into each other, and software that doesn't understand that produces something that looks like a writing system having a breakdown.
This is the part that should make every word lover's brain light up: writing systems are not interchangeable. The assumption that you can build text infrastructure around one model of how letters work and apply it universally is a category error. Some scripts go left to right. Some right to left. Some connect; some don't. Some have characters that only exist in combination.
The JavaScript Intl API, which makes internationalization in web browsers much more tractable, didn't exist during the World Digital Library project's early development. People were solving these problems without the current tools, which explains a lot of the workarounds still embedded in legacy systems today.
Brill Spent $750,000 on a Font and Then Gave It Away
Academic publisher Brill needed to handle Semitic philology. That means Hebrew, Arabic, Aramaic, and related scripts in scholarly contexts where precision matters enormously. They spent $750,000 creating fonts that could do it correctly, then released them for free.
That's a remarkable act of generosity toward the broader digital text ecosystem. Typography at this level of complexity is expensive to produce correctly. The kind of precision that lets a scholar cite an ancient manuscript without the text garbling in transmission doesn't come cheap. Brill ate the cost and shared the result.
The Digit Twist
Here's the genuinely surprising one. At an Egypt partners meeting, Arabic-speaking participants chose Western 0-9 numerals over Arabic-Indic digits for a site redesign.
Arabic-Indic digits are the numeral forms that developed alongside Arabic script. They look different from Western numerals, and they have extended variants used in a handful of countries. You'd think an audience selecting numerals for an Arabic-language site would default to Arabic-Indic. They didn't.
This probably reflects how thoroughly Western numerals have become the practical standard for digital contexts, even among native speakers of languages with their own numeral traditions. Language makes pragmatic accommodations to how the world actually works, not just how it worked historically. Even word nerds make compromises.
Why Word People Should Care
If you love words, you probably love them as living things: shaped by use, carrying history, moving through contexts and changing slightly with each one. Arabic text online is a visible record of what happens when you try to represent a living writing system in infrastructure built by people who didn't fully account for it.
Every joined letter that renders correctly is a small engineering victory. Every unjoined letter is a ghost of an assumption someone made years ago. And somewhere out there, twelve thousand names are sitting in a database encoded with Unicode codepoints from 1991, waiting for someone to care enough to migrate them.
Words are more complicated than games let on. That's not a flaw. That's the whole point.
Source: Languagehat