Arabic
How to get natural Arabic out of Syntic models, and what to check before you ship. The general rules for every language are on Work in other languages.
At a glance
| Language tags | ar, with regional tags such as ar-SA, ar-EG, ar-AE |
| Script and direction | Arabic script, right to left |
| Token use | Measure with token counting on your own text; non-Latin scripts and long compounds usually take more tokens per sentence than English. |
Register and form of address
Modern Standard Arabic (فصحى) is the safe default for written replies. Dialects (Gulf, Egyptian, Levantine, Maghrebi) differ enough that customers notice; ask for one only if your audience shares it.
Numbers, dates and currency
Decide between Western digits (123) and Eastern Arabic digits (١٢٣) and say which in the prompt. Dates are usually day/month/year; the Hijri calendar matters for some religious and government contexts.
Common pitfalls
- Mixed-direction text: product codes, URLs and English names inside Arabic sentences can render out of order. Wrap them in Unicode isolates (U+2068 … U+2069) when you display them, or keep them in separate UI fields.
- Gender agreement: verbs and adjectives agree with the addressee. If you know the user’s preferred form, pass it; otherwise ask the model to use neutral phrasing or the masculine plural convention your brand uses.
- Diacritics (tashkeel) are normally omitted in business text. Ask for them only for teaching or religious content.
A system prompt line that works
Writing the language instruction in Arabic itself is a strong signal:
أجب باللغة العربية الفصحى، وبأسلوب مهذب وموجز، واستخدم الأرقام الغربية.(Answer in Modern Standard Arabic, politely and briefly, using Western digits.)
Keep the rest of your instructions in whichever language your team maintains best; the model follows mixed-language system prompts reliably, and the reply language follows this line.
Test before launch
- Collect 30 or more real Arabic messages, including short, misspelt and mixed-language ones.
- Have a fluent speaker grade replies for correctness, register and naturalness; then calibrate a rubric grader on their grades (Evaluate).
- Check rendering in your actual UI: line breaks, truncation, right-to-left layout and fonts.
- Re-run the set whenever you change the prompt or the model.