New Arabic-English scientific translation benchmark ASCAT enables advanced evaluation across physics, AI, and mathematics with 67,293 English tokens validated by domain experts.

ASCAT corpus contains full scientific abstracts averaging 141.7 English words and 111.78 Arabic words, five times longer than existing Arabic-English datasets

Covers five scientific domains: physics, mathematics, computer science, quantum mechanics, and artificial intelligence

Each abstract translated by three complementary systems: Gemini, Hugging Face quickmt-en-ar transformer model, and commercial APIs (Google Translate, DeepL)

All translations validated by domain experts at lexical, syntactic, and semantic levels to ensure high quality

Dataset contains 67,293 English tokens and 60,026 Arabic tokens for robust benchmarking of scientific translation models