Sketch and Refine: Towards Faithful and Informative Table-to-Text Generation

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4831–4843 August 1–6, 2021. ©2021 Association for Computational Linguistics 4831 Sketch and Refine: Towards Faithful and Informative Table-to-Text Generation Peng Wang , Junyang Lin , An Yang , Chang Zhou , Yichang Zhang , Jingren Zhou , Hongxia Yang† DAMO Academy, Alibaba Group {zheluo.wp, junyang.ljy, ya235025, ericzhou.zc, yichang.zyc, jingren.zhou, yang.yhx }@alibaba-inc.com Abstract Table-to-text generation refers to generating a descriptive text from a key-value table. Traditional autoregressive methods, though can generate text with high fluency, suffer from low coverage and poor faithfulness prob- lems. To mitigate these problems, we pro- pose a novel Skeleton-based two-stage method that combines both Autoregressive and Non- Autoregressive generation (SANA). Our ap- proach includes: (1) skeleton generation with an autoregressive pointer network to select key tokens from the source table; (2) edit-based non-autoregressive generation model to pro- duce texts via iterative insertion and deletion operations. By integrating hard constraints from the skeleton, the non-autoregressive model improves the generation’s coverage over the source table and thus enhances its faithfulness. We conduct experiments on both the WikiPerson and WikiBio datasets. Exper- imental results demonstrate that our method outperforms the previous state-of-the-art meth- ods in both automatic and human evaluation, especially on coverage and faithfulness. In par- ticular, we achieve PARENT-T recall of 99.47 in WikiPerson, improving over the existing best results by more than 10 points. 1 Introduction Table-to-text generation is a challenging task which aims at generating a descriptive text from a key- value table. There have been a broad range of applications in this field, such as the generation of weather forecast (Mei et al., 2016), sports news (Wiseman et al., 2017), biography (Lebret et al., 2016; Wang et al., 2018), etc. Figure 1 illustrates a typical input and output example of this task. Previous methods (Liu et al., 2018; Nie et al., 2018; Bao et al., 2018) are usually trained in an †Corresponding author. Figure 1: An example of table-to-text generation. The case of poor faithfulness hallucinates content not en- tailed by the table (marked in red color). The case of low coverage misses the information of the person’s birth place (marked in blue color). end-to-end fashion with the encoder-decoder ar- chitecture (Bahdanau et al., 2015). Despite gener- ating text with high fluency, their lack of control in the generation process leads to poor faithful- ness and low coverage. As shown in Figure 1, the case of poor faithfulness hallucinates the oc- cupation “singer” not entailed by the source table, and the case of low coverage misses the informa- tion of the place of birth. Even if trained with a cleaned dataset, end-to-end methods still encounter these problems as it is too complicated to learn the probability distribution under the table constraints (Parikh et al., 2020). To alleviate these problems, recent studies (Shao et al., 2019; Puduppully et al., 2019; Ma et al., 2019) propose two-stage methods to control the generation process. In the first stage, a pointer net- work selects the salient key-value pairs …

Content truncated to 500 words