TWiST: Turkish-English Wikipedia & Thesis STEM Terminology Dataset

We introduce TWiST—the Turkish-English Wikipedia & Thesis STEM Terminology Dataset—an expertly curated, sentence-aligned bilingual resource that addresses a key gap in Turkish computational linguistics. TWiST is a 3,300-sentence Turkish–English parallel corpus of STEM terminology drawn from two sources: 1,185 sentences from Wikimedia Content Translation dump and 2,115 sentences from 287 graduate-thesis abstracts at Turkey’s top six universities. Focused on Mathematics, Physics, and Computer Science, every sentence pair was triple-annotated by 43 trained bilingual annotators following a 30-page guideline, achieving substantial agreement (Fleiss κ ≈ 0.7). TWiST ultimately captures 10,157 annotated term instances covering 1,223 distinct English technical terms, offering a high-quality benchmark for bilingual terminology extraction, translation consistency, and terminology-aware NLP.