Up next Ehud Reiter is presenting work on "Enhancing Factualness and Controllability of Data-to-Text Generation via Data Views and Constraints" #INLG2023
Long-form #DataToText generation remains difficult for neural #NLG systems or #LLMs to do with high accuracy, and this work proposes bringing #DocumentPlanning back, to guide the generation process.
At the #MMNLG workshop today at #SIGDIALxINLG2023, currently enjoying Dimitra Gkatzia's keynote about "Dealing with Data Challenges in Low-Resource NLG"
She began by highlighting ways in which NLG can be low-resource: it's not just about languages, but domains can be low-resource as well, and even our "large" #DataToText datasets are small by modern #NLP standards.
If you missed the #SIGGEN webinar last week by Ross Turner on "7 Characteristics of Highly Effective Data-to-Text Systems", here's the link: https://www.youtube.com/watch?v=1o4PVhfl3rE
Dr Turner works at #ArriaNLG, a leading company delivering #NaturalLanguageGeneration tools and frequent sponsor of research events relating to the topic.
Dave in 🏴
Dave 🦔