Fusing Part-of-Speech Information in Low-Resource Neural Paraphrase Generation

Figure 7

Dataset statistics for Quora datasets: (a) BLEU score computed between source and target sentences on Quora50K and Quora test set, (b) average ratio between the lengths of the longer sentence and the shorter sentence in source-target pairs, and (c) proportion each range represents.

(a)

(b)

(c)