Samielakkad/AI-NLP-Darija-Resources

AI + NLP · curated corpora, papers & tools for low-resource Moroccan Arabic (Darija) # AI + NLP · darija-nlp-resources > Curated list of public corpora, papers, tools, and benchmarks for Moroccan Arabic (Darija) NLP — the operating language behind jak.ma. --- ## Why this list Moroccan Arabic (Darija) is the operating language of ~37 million people but is severely under-resourced in mainstream NLP infrastructure. Most "Arabic NLP" work targets Modern Standard Arabic (MSA), which Moroccan users do not actually use for service-discovery, day-to-day messaging, or marketplace transactions. This repo is the curated reading + tooling list I built while shipping jak.ma — a live Moroccan marketplace running grounded retrieval in Darija. If you're starting a Darija NLP project, start here. --- ## Public corpora **Full annotated list — per-entry sizes, licenses, caveats, and a "what to use it for in production" note: corpora.md.** ### Curated multi-dialect Arabic corpora that include Darija - **PADIC** (Parallel Arabic Dialect Corpus) — multi-dialect Arabic including Maghrebi varieties. Useful for cross-dialect alignment. - **MADAR** (Multi-Arabic Dialect Applications and Resources) — 25 city-level Arabic dialects across 25 Arab cities. Includes Rabat, Casablanca, Fes for Morocco. - **OSACT** (Open-Source Arabic Corpora and Processing Tools) — repeating shared task with Arabic dialect tracks. - **TUNIZI** — Tunisian Latin-script Arabic. Code-switching patterns transfer to Darija Arabizi. ### Darija-specific projects - **DODa** (Darija Open Dataset) — community-curated Darija dictionary mapping to MSA + English. - **MADAR-26** — Rabat + Casablanca + Fes subsets, sentence-level transliteration. - **Maghrebi-Arabic-NLP** — community resource list for Maghreb dialects (Algeria, Tunisia, Morocco). ### Script-mixed (Arabizi) corpora Latin-script Darija ("Arabizi") is ~30% of jak.ma's user-generated text. Resources are sparser: - Twitter-mined Arabizi datasets (search arxiv for "Arabizi" + year) - Tunisian Arabizi corpora (TUNIZI) — closest analog - …