Official code for "Ketaba-OCR at AR-MS NakbaNLP 2026" — QLoRA fine-tuning of a specialized HTR model with Linear+Boost ensemble for Arabic manuscript recognition. 1st place per-line (CER 0.082) and 3rd place official leaderboard at NakbaNLP 2026 (LREC 2026).

Official code for "Fine-Tashkeel at KSAA-2026" — Systematic evaluation of 18 Seq2Seq, token classification, decoder LLM, and ASR models for automatic Arabic text diacritization. 5th place at KSAA-2026 Shared Task (OSACT7 @ LREC 2026).

Additional experimental model for NakbaNLP 2026 Shared Task (AR-MS) — LoRA/DoRA fine-tuning of Qari-OCR (Qwen2-VL-2B) for Arabic handwritten manuscript recognition on the Omar Al-Saleh Memoir Collection (1951-1965).

OasisSimp: An Open-source Asian-English Sentence Simplification Dataset

Code and data for our EvaLatin 2026 submission (first on poetry).