Self-supervised neural machine translation (SSNMT) jointly learns to identify and select suitable training data from comparable (rather than parallel) corpora and to translate, in a way that the two tasks support each other in a virtuous circle. In this study, we provide an in-depth analysis of the sampling choices the SSNMT model makes during training. We show how, without it having been told to do so, the model self-selects samples of increasing (i) complexity and (ii) task-relevance in combination with (iii) performing a denoising curriculum. We observe that the dynamics of the mutual-supervision signals of both system internal representation types are vital for the extraction and translation performance. We show that in terms of the Gunning-Fog Readability index, SSNMT starts extracting and learning from Wikipedia data suitable for high school students and quickly moves towards content suitable for first year undergraduate students.
@article{arxiv.2004.03151,
title = {Self-Induced Curriculum Learning in Self-Supervised Neural Machine Translation},
author = {Dana Ruiter and Josef van Genabith and Cristina España-Bonet},
journal= {arXiv preprint arXiv:2004.03151},
year = {2020}
}