Al-ʻAranjiyya: An Expert-Annotated Corpus of Covert Linguistic Hybridization in Modern Arabic News Text

Al-ʻAranjiyya is the first publicly released, expert-annotated corpus of covert linguistic hybridization (al-ʻaranjiyya, العَرَنْجِيَّة): the covert adoption of foreign structural and semantic patterns into Arabic prose that, unlike code-switching, leaves no surface marker and requires structural and semantic reasoning to detect. It comprises 1,000 Modern Standard Arabic news articles (78,601 running words) sampled from the publicly available SANAD dataset and balanced across six domains (Culture, Finance, Medical, Politics, Sports, Technology), annotated by two professional Arabic linguists into 6,093 spans and 7,881 label assignments across six categories (Syntactic Template, Stylistic Template, Semantic Template, Lexical Error, Lexical Borrowing, Morphological Error) with substantial inter-annotator agreement (Cohen's κ = 0.67); each span carries character offsets, one or more labels, and a free-text rationale.