Researchers have introduced PERCEPT, the first large-scale corpus for analyzing Persian-English code-mixing, specifically targeting part-of-speech (POS) tagging for code-mixed words. This new dataset, comprising 6,800 posts from platforms like X, Instagram, and Digikala, includes Universal Dependencies POS annotations. An LLM-assisted framework was employed for annotation, showing high agreement with human evaluations, which validates its reliability. The corpus enables the first comprehensive linguistic analysis of Persian-English code-mixing, revealing that nouns are the most frequent code-mixed word category and that code-mixed words appear consistently across platforms, with a stronger triggering effect observed on Digikala. AI
IMPACT Enables development of NLP models for code-mixed languages and deeper linguistic understanding of multilingual social media communication.
RANK_REASON The item describes a new corpus and associated analysis for a specific linguistic phenomenon, fitting the research category. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
