Step-by-Step: Separating Planning from Realization in Neural Data-to-Text Generation

NAACL 2019 Amit MoryossefYoav GoldbergIdo Dagan

Data-to-text generation can be conceptually divided into two parts: ordering and structuring the information (planning), and generating fluent language describing the information (realization). Modern neural generation systems conflate these two steps into a single end-to-end differentiable system... (read more)

PDF Abstract

Evaluation Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK COMPARE
Data-to-Text Generation WebNLG Chimera - Plan Based BLEU 0.474 # 2
Data-to-Text Generation WebNLG Chimera - Neural End2End BLEU 0.465 # 3