Neural Machine Translation for Low Resource Languages

Details

This project was done as part of the course CS671: Introduction to Natural Language Processing, in the Spring ‘18 term at IIT Kanpur under Prof. Harish Karnick, Department of Computer Science and Engineering, IIT Kanpur.

Abstract

Much work has been done on Neural Machine Translation, but most of it requires the availability of a large parallel corpora to train on, which is not always readily available. Unsupervised methods also exist that work by encoding both the languages in consideration to a common latent space, and then decoding them to the target language as required. This work aims at an accurate implementation of the work presented in Lample et al. (2017), and train it on a non-parallel corpora of English and Catalan languages. Baseline implementation is evaluated using BLEU score metric. We also hypothesize that adding the GCN layers would optimize the existing representations