Why AI in other languages is genuinely harder
Almost all public AI guidance is quietly English-first, and a great many projects fall over the first time they meet a real non-English user. This lesson is why that happens and what to do about it. Arabic is the running example because it makes every problem concrete, but none of this is specific to Arabic.
Spot the language assumptions buried in an AI plan before they become a failed launch.
Lesson 1 for tokens, lesson 2 for retrieval, lesson 4 for evaluation. The rest is self-contained.
Plain-English mode
Each section opens with a short summary before the detail. Turn this off once the vocabulary feels familiar.
Three reasons: less training data, more complicated scripts, and the gap between the written language and the way people actually speak. The third causes the most failed launches.
Three things stack up, and the third is the one that catches teams out.
First, most languages appear far less than English in the material these models learned from, so the models are simply less good at them. Second, many writing systems are harder: marks that get left out in everyday writing, and words that pack several meanings into a single form through prefixes and suffixes.
Third, and most important: in many languages the formal written version and the version people actually speak are substantially different. Modern Standard Arabic is what appears in official documents and news broadcasts. It is not how anyone orders coffee. Gulf, Egyptian, Levantine, and Maghrebi speech differ from it and from each other. A model can be entirely fluent in the formal register and still fail to understand a real person describing their problem. Add to that the habit of switching languages mid-sentence, which is normal in most bilingual populations and unusual in training data.
So if you only test on formal written text, you will ship something that beautifully and falls over on contact with real users. Designing for another language is a stack of deliberate choices, not a setting somebody enables. The rest of this lesson is part of the paid course.
Answer in your head first, then reveal. Getting it wrong is the part that makes it stick.
A system scores 94% on a standard Arabic test set and fails badly in a pilot. What is the most likely explanation?
The rest of this lesson
- The token penalty
- The stack of choices
- One worked example: a multilingual phone agent
Unlock “Why AI in other languages is genuinely harder” and the whole playbook
One payment unlocks all 12 AI lessons in full, plus every career map, 90-day plan, and template in the Clinical Pivot Playbook.
This is educational career guidance, not a promise of a job, a salary, or any specific outcome, and not financial or legal advice. Your results depend on your own effort and circumstances. Make big decisions with proper professional advice where it matters.
Unlock every AI lesson
One payment unlocks all 12 AI lessons in full, plus the entire Clinical Pivot Playbook: 8 career maps, 8 personalised 90-day plans, CV and LinkedIn templates, the cold-message bank, the interview answers, the glossary, and the startup list.