A new paper on arXiv explores the effectiveness of source languages in cross-lingual in-context learning (ICL). The study challenges the assumption that insights from traditional supervised fine-tuning directly apply to ICL, finding that conventional heuristics for selecting source languages may not be optimal. The research analyzes language confusion as a key obstacle for generative tasks in cross-lingual ICL and proposes alternative methods for effective source language selection. AI
IMPACT Challenges existing assumptions about cross-lingual transfer in few-shot learning, potentially guiding future research and model development.
RANK_REASON The cluster contains a research paper submitted to arXiv detailing empirical study results.
- alphaXiv
- arXiv
- CatalyzeX
- CORE Recommender
- DagsHub
- English
- Few-shot learning
- Gotit.pub
- Hugging Face
- Influence Flower
- natural language processing
- ScienceCast
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →