An NLP Framework for Detecting, Aligning, and Normalizing French-Origin Lexical Borrowings in Jordanian Arabic
Year: 2026
Type: conference-paper
Abstract: This paper presents a pilot NLP framework for detecting, phoneticizing, source-aligning, and translation-normalizing French-origin lexical borrowings in Jordanian Arabic. The framework targets low-resource, mixed-script input and combines phonographic rules, IPA-informed grapheme-to-phoneme normalization, feature-based source-lemma ranking, and a context-sensitive translation decision module. A 40-item pilot lexicon was evaluated against 80 negative-control tokens, including unrelated dialectal forms, English-origin loans, named entities, brand names, pure French code-switching, and transliteration-only examples. On this 120-token calibration set, the hybrid detector achieved 88.4% precision, 95.0% recall, and 91.5% F1-score; source alignment reached 82.5% top-1 and 95.0% top-3 accuracy; and translation-strategy selection agreed with expert labels in 85.0% of cases. The findings suggest that phonetic normalization improves the detection of adapted Arabic-script borrowings, while context and integration cues reduce false positives. The study contributes a reproducible, borrowing-aware NLP workflow for dialectal Arabic processing, machine translation, and multilingual lexical search.
Authors: Authoul Abdul Hay, Taghreed Altamimi
Institutions: Al-Zaytoonah University of Jordan, Alfaisal University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7221659897