A new arXiv paper investigates how well cross-lingual alignment (CLA) scores predict the performance of large language models (LLMs) on both classification and machine translation tasks. The research compares 27 CLA score variants and introduces a new PMI-based translation metric to assess translation quality across different languages. Findings suggest that alignment with English is a strong predictor of performance, indicating that LLMs may use English as an internal pivot language. AI
IMPACT This research could lead to better methods for evaluating and improving multilingual LLM capabilities, particularly for translation tasks.
RANK_REASON The cluster contains an academic paper published on arXiv detailing research into LLM performance prediction. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Child Health Research Foundation
- Classification
- Cross-lingual alignment of biomedical acronyms and their expansions.
- English
- machine translation
- PMI-based translation metric
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
