Explorations on Multi-lingual Neural Machine Translation

Deep (recurrent) neural networks has been shown to successfully learn complex mappings between arbitrary length input and output sequences, within the effective framework of encoder-decoder networks. We investigate the extensions of this sequence to sequence models, to handle multiple sequences at the same time, within the same model. This reduces to the problem of multi-lingual machine translation (MLNMT), as we explore applicability and the benefits of MLNMT on, (1) large scale machine translation tasks, between all six languages of WMT’15 shared task, (2) low-resource language transfer problems, for Finnish, Uzbek and Turkish into English translation, (3) multi-source translation tasks where we have multi-way parallel text available, and (4) Zero-resource translation tasks where we don’t have any available bi-text between two languages. We will further discuss about the natural extensions of MLNMT model for system combination (of SMT and NMT models) and larger-context NMT (given the entire documents during translation).

Date:
Speakers:
Orhan Firat
Affiliation:
Middle East Technical University

Watch Next