Exploring Diversity in Back Translation for Low-Resource Machine Translation

Back translation is one of the most widely used methods for improving the performance of neural machine translation systems. Recent research has sought to enhance the effective- ness of this method by increasing the ‘diversity’ of the generated translations. We argue that the definitions and metrics used to quantify ‘diversity’ in previous work have been insuf- ficient. This work puts forward a more nu- anced framework for understanding diversity in training data, splitting it into lexical diversity and syntactic diversity. We present novel met- rics for measuring these different aspects of diversity and carry out empirical analysis into the effect of these types of diversity on fi- nal neural machine translation model perform- ance for low-resource English↔Turkish and mid-resource English↔Icelandic. Our findings show that generating back translation using nuc- leus sampling results in higher final model per- formance, and that this method of generation has high levels of both lexical and syntactic diversity. We also find evidence that lexical diversity is more important than syntactic for back translation performance