Jisr — English ↔ 13 Arabic dialects
Two 49M-parameter MarianMT models trained from scratch. Pick a direction below.
Egyptian · Moroccan · Algerian · Tunisian · Libyan · Saudi · Bahraini · Iraqi · Syrian · Lebanese · Palestinian · Sudanese · Yemeni — plus Modern Standard Arabic.
Jisr-MT-50M-AllDialects — one model, 13 dialect tags. Beats a frontier model on Moroccan Darija (32.23 vs 20.86 chrF++).
Jisr-MT-50M-Dialects2En — one model for all 13 dialects, no tag needed. Reaches 82% of gpt-5.1 on this direction.
Source dialects above, in order: 🇱🇧 Lebanese · 🇮🇶 Iraqi · 🇱🇾 Libyan · 🇧🇭 Bahraini · 🇲🇦 Moroccan · 🇩🇿 Algerian · 🇪🇬 Egyptian · 🇸🇦 Saudi · 🇸🇩 Sudanese · 🇵🇸 Palestinian
One English sentence, rendered in all 13 dialects plus MSA at once.
Scope. English → Arabic and Arabic → English only. Dialect data is ASR transcripts with machine-translated English, so the register skews conversational. Algerian, Bahraini and Iraqi have no external benchmark — their quality is unverified.
Dialectal Arabic has no standard orthography; spellings here are one convention among several.