The Sanctuary

Writing about interests; Computer Science, Philosophy, Mathematics and AI.

Publications

2026 · Oral presentation

MoriscoNER: A Native Moroccan-Darija Latin-Script NER Corpus and an Encoder-Conditional Native-vs-Transliteration Study

Achraf Soltani — 1st International Conference on Artificial Intelligence and Natural Language Processing (IANLP 2026), Hassan II University of Casablanca, Morocco, 29–30 June 2026
Springer proceedings, to appear

Introduces MoriscoNER, a native Moroccan-Darija Latin-script (Arabizi) NER training corpus — 1,531 sentences and 2,643 entities, with a native-validated 400-sentence test set — together with a controlled ablation of native supervision against transliterate-then-NER, whose outcome proves conditional on the encoder, and MoRoBERTa, a permissively released encoder continue-pretrained on Moroccan Arabizi text.

2026 · Poster presentation

Cross-Lingual Teacher–Student Distillation of Job-Market Entity Recognition for French and Arabic

Achraf Soltani — 1st International Conference on Artificial Intelligence and Natural Language Processing (IANLP 2026), Hassan II University of Casablanca, Morocco, 29–30 June 2026
Springer proceedings, to appear

A teacher–student distillation recipe for job-market entity recognition — skills, certifications, education and experience requirements — in French and Arabic, anchored to the multilingual ESCO taxonomy. Builds the first Arabic job-market entity dataset to the author's knowledge, and shows the best per-language Arabic student performing on par with its LLM teacher.