chunked/nltk_chunking/model_doc_xls_r/chunk_7.txt

Moreover, we show that with sufficient model size, cross-lingual pretraining can outperform
English-only pretraining when translating English speech into other languages, a setting which favors monolingual
pretraining.