Imaginary Numbers! Evaluating Numerical Referring Expressions by Neural End-to-End Surface Realization Systems

Download PDF
Open Webpage

Rossana Cunha, Osuji Cynthia Chinonso, João Gabriel Moura Campos, Brian Timoney, Brian Davis, Fabio Cozman, Adriana Silvina Pagano, Thiago Castro Ferreira

Published: 20 May 2024, Last Modified: 11 Feb 2025Fifth Workshop on Insights from Negative Results in NLPEveryoneCC BY 4.0
Abstract: Neural end-to-end surface realizers output more fluent texts than classical architectures. However, they tend to suffer from adequacy problems, in particular hallucinations in numerical referring expression generation. This poses a problem to language generation in sensitive domains, as is the case of robot journalism covering COVID-19 and Amazon deforestation. We propose an approach whereby numerical referring expressions are converted from digits to plain word form descriptions prior to being fed to state-of-the-art Large Language Models. We conduct automatic and human evaluations to report the best strategy to numerical superficial realization. Code and data are publicly available.