Towards Table-to-Text Generation with Numerical Reasoning

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 1451–1465 August 1–6, 2021. ©2021 Association for Computational Linguistics 1451 Towards Table-to-Text Generation with Numerical Reasoning Lya Hulliyyatus Suadaa1, Hidetaka Kamigaito1, Kotaro Funakoshi1, Manabu Okumura1 and Hiroya Takamura1,2 1Tokyo Institute of Technology 2National Institute of Advanced Industrial Science and Technology (AIST) lya@stis.ac.id {kamigaito,funakoshi,oku}@lr.pi.titech.ac.jp takamura@pi.titech.ac.jp Abstract Recent neural text generation models have shown significant improvement in generating descriptive text from structured data such as table formats. One of the remaining impor- tant challenges is generating more analytical descriptions that can be inferred from facts in a data source. The use of a template-based gen- erator and a pointer-generator is among the po- tential alternatives for table-to-text generators. In this paper, we propose a framework consist- ing of a pre-trained model and a copy mecha- nism. The pre-trained models are fine-tuned to produce fluent text that is enriched with numer- ical reasoning. However, it still lacks fidelity to the table contents. The copy mechanism is incorporated in the fine-tuning step by using general placeholders to avoid producing hallu- cinated phrases that are not supported by a ta- ble while preserving high fluency. In summary, our contributions are (1) a new dataset for nu- merical table-to-text generation using pairs of a table and a paragraph of a table description with richer inference from scientific papers, and (2) a table-to-text generation framework enriched with numerical reasoning. 1 Introduction Recent data-to-text generation studies have shown significant improvement in generating faithful text aligned with data sources. A copy mechanism has been widely explored to improve faithfulness in various ways. Wiseman et al. (2017) used joint probabilities to let models choose between copy- ing records from data sources or generating from a vocabulary. Puduppully et al. (2019) improved a similar approach by modeling entity represen- tations as a unit of copying. This approach has proven to be effective in generating descriptive text that explicitly mentions facts from sources. However, as introduced by Chen et al. (2020a), humans have the ability to produce more analyti- Model Precision Recall F1 Our full model 89.6 82.2 85.7 Lee et al. (2018) 86.2 83.7 84.9 Table 2: The overall mention detection results on the test set of OntoNotes. Target Header Our full model Description Table 2 shows the mention detection results on the test set. Similar to coreference linking results, our model achieves higher precision and F1 score, which indicates that our model can significantly reduce false positive mentions while it can still find a reasonable number of mentions. Figure 1: Example of table and description in numeric NLG dataset. cal text with richer inference, including numerical reasoning. Making inferences beyond texts is still an open question due to the limitation of language models in handling numeric operations. In this study, we further encourage research by elaborat- ing numerical tables to initialize the ability to inject reasoning while maintaining high fluency. Our contributions are summarized as follows. • …

Content truncated to 500 words