A new research paper on arXiv challenges common assumptions about multilingual transformer models, particularly for coreference resolution in low-resource languages like Bengali. The study found that a small amount of high-quality annotated data significantly outperforms larger, mixed-quality datasets. Furthermore, data augmentation techniques like back-translation did not improve performance, and cross-lingual transfer from distant languages like English even degraded results. The research also identified issues with collapsed embedding spaces in several widely used multilingual models, making them unsuitable for similarity-based clustering tasks. AI
IMPACT Highlights the critical need for high-quality data in low-resource NLP tasks and questions the generalizability of current multilingual models.
RANK_REASON Academic paper published on arXiv detailing research findings. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
