sangyun Kwon retweeted
🙌We're at NAACL’25 presenting four fresh papers pushing MT, speech, embeddings & cultural eval forward:
📜Effective Self-Mining of In-Context Examples for Unsupervised MT with LLMs
🎓 Abdellah El Mekki & Muhammad Abdul-Mageed
TL;DR: Self-mined word→sentence exemplars let LLMs translate 288 pairs, +7 BLEU vs prior UMT.
🔗 aclanthology.org/2025.findin…
📜uDistil-Whisper: Label-Free Data Filtering for Knowledge Distillation in Low-Data Regimes
🎓 Abdul Waheed, Karima Kadaoui, Bhiksha Raj, and Muhammad Abdul-Mageed
TL;DR: Unlabeled filtering distils Whisper into 25-50 % smaller ASR models that beat the teacher by 5-7 WER.
🔗 aclanthology.org/2025.naacl-…
📜Swan & ArabicMTEB: Dialect-Aware, Arabic-Centric Embeddings & Benchmarks
🎓 Gagan Bhatia, El Moatez Billah Nagoudi, Abdellah El Mekki, Fakhraddin Alwajih, and Muhammad Abdul-Mageed
TL;DR: Swan-Small/Large + MTEB (8 tasks, 94 datasets); Swan-Large leads on Arabic embedding tasks. Models offer significant monetary gains compared to Open AI and other models
🔗 aclanthology.org/2025.findin…
📜Jawaher: A Multidialectal Dataset of Arabic Proverbs for LLM Benchmarking
🎓 Samar Mohamed Magdy, Sang Yun Kwon, Fakhraddin Alwajih, Safaa Taher Abdelfadil, Shady Shehata, and Muhammad Abdul-Mageed
TL;DR: 10k Arabic proverbs across 20 dialects show LLMs translate but miss cultural nuance and struggle with producing explanations.
🔗 aclanthology.org/2025.naacl-…
#NAACL2025 #AIResearch #NLP #MultilingualAI #SpeechRecognition #ArabicNLP #MachineTranslation
sangyun Kwon retweeted
We’re off to Singapore 🇸🇬 ✈️ to present 12 papers, 4 at #EMNLP2023 and 8 at #ArabicNLP2023. We’re presenting work on #LLMs, #speech, and #vision_language models, both on #massively_multilingual NLP and #Arabic NLp (Arabic varieties and #dialects). Please say “hi” if you see any of us.
سنكون بمشيئة الله في سنغافورة لتقديم ١٢ ورقة بحثية حول النماذج اللغوية الضخمة، تحليل الكلام، ونماذج اللغة-الرؤية الحاسوبية.
@UBC_CS @UBCiSchool @UBCLinguistics @mbzuai
@e_nagoudi @arelmadany @KarimaKadaoui @MacAbdul9 @ChiyuZhang0851
@mmagdy_samar @tawkat97
#NLProc #AI
sangyun Kwon retweeted
#LLMs like #ChatGPT can be powerful educational tools in English. Can they also be in #Arabic? Based on our new 📜, "YES". We also introduce a new SoTA model for Arabic grammatical error correction.
🔗 tinyurl.com/argec With @SangyoonKwon @gaganbhatiaml @e_nagoudi #NLProc
sangyun Kwon retweeted
Hello Twitter!
How do you guys like the name?
(Early version).
هل يعجبكم الاسم؟
#AI #ArabicNLG #ArabicNLProc #NLProc
#الذكاء_الاصطناعي_التوليدي #اللغة_العربية
#معالجة_اللغة_العربية

















