Adapting BigScience Multilingual Model to Unseen Languages
arXiv:2204.04873v15 citationsh-index: 17
Originality Synthesis-oriented
AI Analysis
This work addresses the problem of extending multilingual models to new languages for NLP practitioners, but it is incremental as it builds on existing methods.
The study benchmarked strategies for adding German and Korean to a 1.3-billion-parameter multilingual model, identifying factors affecting adaptability and trade-offs between computational costs and performance.
We benchmark different strategies of adding new languages (German and Korean) into the BigScience's pretrained multilingual language model with 1.3 billion parameters that currently supports 13 languages. We investigate the factors that affect the language adaptability of the model and the trade-offs between computational costs and expected performance.