Improving neural machine translation of languages with little data and rich morphology

Abstract

PoS Classification Accuracy on Layer 0 and the effect of vocabulary splitting. Also included are the standard deviations on the 5 runs. . . . 4.7 Size of the test dataset on the application of a vocabulary splitting technique. Also included are the standard deviations on the 5 runs. The bars represent how many of the tokens remain (in percentages) after splitting. Number of unknown tokens (in percentages) in the development set vsDictionary size for each language and segmentation type. Also systems that reported the lowest and highest CHRF scores (on the development set) for each language and segmentation are marked. . . . . . . . . . 6.2The Probability density function plot showing the distribution of the average subword length for a given segmenter and language on the training sets. The colored boxes denote the mean of the respective distributions. Also included are the differences in the CHRF scores between