MADLAD-400: A Multilingual And Document-Level Large Audited Dataset

About

We introduce MADLAD-400, a manually audited, general domain 3T token monolingual dataset based on CommonCrawl, spanning 419 languages. We discuss the limitations revealed by self-auditing MADLAD-400, and the role data auditing had in the dataset creation process. We then train and release a 10.7B-parameter multilingual machine translation model on 250 billion tokens covering over 450 languages using publicly available data, and find that it is competitive with models that are significantly larger, and report the results on different domains. In addition, we train a 8B-parameter language model, and assess the results on few-shot translation. We make the baseline models available to the research community.

Sneha Kudugunta, Isaac Caswell, Biao Zhang, Xavier Garcia, Christopher A. Choquette-Choo, Katherine Lee, Derrick Xin, Aditya Kusupati, Romi Stella, Ankur Bapna, Orhan Firat · 2023

Related benchmarks

Task	Dataset	Result
Speech Translation Metric Evaluation	Europarl-ST (test)	Average Correlation0.9951	84
Machine Translation	Wikinews-25 en->it	sacreBLEU49.6611	35
Machine Translation	Wikinews-25 it->en	sacreBLEU44.7538	35
Translation	FLORES-200 en-it (devtest)	sacreBLEU31.6561	35
Translation	FLORES-200 it-en (devtest)	sacreBLEU35.0758	35
Machine Translation	NTREX (en->it) 128 (test)	sacreBLEU38.9525	35
Machine Translation	NTREX it->en 128 (test)	sacreBLEU42.5244	35
Machine Translation	Tatoeba en->it	sacreBLEU59.2901	33
Machine Translation	Tatoeba it->en	sacreBLEU70.2099	33
Subtitle Translation	subtitle dataset en-zh	Translation Score59.7	24