The invention discloses a method, a device, and a system for translation training. For example, a translation training method applied in a distribution server of a distributed system can comprise: distributing a first translation training task and a to-be-trained corpus to each mapping computation sever of a mapping computation sever cluster, so that each mapping computation sever executes the first translation training task on the received to-be-trained corpus concurrently, to output a first translation training result; distributing a second translation training task and the first translation training result output by each mapping computation sever to each reduction computation sever of a reduction computation server cluster, wherein the first translation training results corresponding to same language elements are distributed to the same reduction computation server, so that each reduction computation server uses the received first translation training result to execute a second translation training task concurrently, to output a second translation training result. In addition, the invention also discloses a method and a device for online translation.