Exploring the impact of data representation on neural data-to-text generation

Howcroft, David M.; Watson, Lewis N.; Nedopas, Olesia; Gkatzia, Dimitra

Abstract

A relatively under-explored area in research on neural natural language generation is the impact of the data representation on text quality. Here we report experiments on two leading input representations for data-to-text generation: attribute-value pairs and Resource Description Framework (RDF) triples. Evaluating the performance of encoder-decoder seq2seq models as well as recent large language models (LLMs) with both automated metrics and human evaluation, we find that the input representation does not seem to have a large impact on the performance of either purpose-built seq2seq models or LLMs. Finally, we present an error analysis of the texts generated by the LLMs and provide some insights into where these models fail.

Citation

Howcroft, D. M., Watson, L. N., Nedopas, O., & Gkatzia, D. (2024, September). Exploring the impact of data representation on neural data-to-text generation. Presented at INLG 2024, Tokyo, Japan

• Presentation Conference Type: Conference Paper (published)
• Conference Name: INLG 2024
• Start Date: Sep 23, 2024
• End Date: Sep 27, 2024
• Acceptance Date: Jul 15, 2024
• Online Publication Date: Oct 1, 2024
• Publication Date: 2024
• Deposit Date: Jul 16, 2024
• Publicly Available Date: Oct 3, 2024
• Publisher: Association for Computational Linguistics (ACL)
• Peer Reviewed: Peer Reviewed
• Pages: 243–253
• Book Title: Proceedings of the 17th International Natural Language Generation Conference
• ISBN: 9798891761223
• Publisher URL: https://aclanthology.org/2024.inlg-main.20/
• External URL: https://inlg2024.github.io/