The Unreasonable Effectiveness of Random Target Embeddings for Continuous-Output Neural Machine Translation
This finding is significant for machine translation researchers as it questions established practices and offers a simpler, effective alternative, though it may be incremental in its specific domain.
The paper tackles the problem of continuous-output neural machine translation by challenging the assumption that semantic structure in target embeddings is crucial, showing that random embeddings outperform pretrained ones, especially for rare words on larger datasets.
Continuous-output neural machine translation (CoNMT) replaces the discrete next-word prediction problem with an embedding prediction. The semantic structure of the target embedding space (i.e., closeness of related words) is intuitively believed to be crucial. We challenge this assumption and show that completely random output embeddings can outperform laboriously pretrained ones, especially on larger datasets. Further investigation shows this surprising effect is strongest for rare words, due to the geometry of their embeddings. We shed further light on this finding by designing a mixed strategy that combines random and pre-trained embeddings for different tokens.