Comparing Computational Architectures for Automated Journalism

Abstract

The majority of NLG systems have been designed following either a template-based or a pipeline-based architecture. Recent neural models for datato-text generation have been proposed with an end-to-end deep learning flavor, which handles non-linguistic input in natural language without explicit intermediary representations. This study compares the most often employed methods for generating Brazilian Portuguese texts from structured data. Results suggest that explicit intermediate steps in the generation process produce better texts than the ones generated by neural end-to-end architectures, avoiding data hallucination while better generalizing to unseen inputs. Code and corpus are publicly available.

Um sistema de perguntas e respostas com aprendizado por reforço profundo para perguntas complexas utilizando textos e tabelas.

“…This work was presented at the "AI: Modeling Oceans and Climate Change (IJCAI-ECAI), 2022"(PIROZELLI et al, 2022).• The author contributed to the development to a project focused on creating a journalist robot for reporting on the Blue Amazon using three of the most important Natural Language Generation techniques: template-based, pipeline-based, and neural end-to-end. Results were published at "Encontro Nacional de Inteligência Artificial e Computacional" (ENIAC 22)(SYM et al, 2022).• The author participated in a research paper titled "Coordination within Conversational Agents with Multiple Sources" which describes how a Large Language Model can effectively coordinate various sources and systems to develop a conversational agent. This work was presented in the proceedings of "XIX Encontro Nacional de Inteligência Artificial e Computacional" (ENIAC 23)(MATOS et al, 2023).…”