Up next is Nalin Kumar presenting the Charles University Prague system for #WebNLG2023 "Better Translation + Split and Generate for Multilingual RDF-to-Text"

#MMNLG #SIGDIALxINLG2023

Lots of the "silver standard" training data was noisy, so the team used a state of the art MT system to retranslate some of the texts from English.

Kumar describes the Split and Generate approach which groups subsets of triples together to generate texts in smaller chunks, since longer texts exhibited more errors.