Обучение на основе согласия

We introduce an agreement-based ap- proach to learning parallel lexicons and phrases from non-parallel corpora. The basic idea is to encourage two asym- metric latent-variable translation models (i.e., source-to-target and …...