Point Precisely: Towards Ensuring the Precision of Data in Generated Texts Using Delayed Copy Mechanism

Proceedings of the 27th International Conference on Computational Linguistics, pages 1044–1055 Santa Fe, New Mexico, USA, August 20-26, 2018. 1044 Point Precisely: Towards Ensuring the Precision of Data in Generated Texts Using Delayed Copy Mechanism Liunian Li Xiaojun Wan Institute of Computer Science and Technology, Peking University The MOE Key Laboratory of Computational Linguistics, Peking University {liliunian, wanxiaojun}@pku.edu.cn Abstract The task of data-to-text generation aims to generate descriptive texts conditioned on a number of database records, and recent neural models have shown significant progress on this task. The attention based encoder-decoder models with copy mechanism have achieved state-of-the-art results on a few data-to-text datasets. However, such models still face the problem of putting incorrect data records in the generated texts, especially on some more challenging datasets like ROTOWIRE. In this paper, we propose a two-stage approach with a delayed copy mechanism to improve the precision of data records in the generated texts. Our approach first adopts an encoder-decoder model to generate a template text with data slots to be filled and then leverages a proposed delayed copy mechanism to fill in the slots with proper data records. Our delayed copy mechanism can take into account all the information of the input data records and the full generated template text by using double attention, position-aware attention and a pairwise ranking loss. The two models in the two stages are trained separately. Evaluation results on the ROTOWIRE dataset verify the efficacy of our proposed approach to generate better templates and copy data records more precisely. 1 Introduction One important task in the natural language generation (NLG) area is to generate a natural language de- scription for some structured data (i.e., a number of database records), a.k.a., the data-to-text generation task. Applicable to wide areas such as weather forecasting, financial reporting, game broadcasting, data- to-text generation systems have been studied for decades (Gatt and Krahmer, 2018). Traditionally this task is addressed by using templates or statistically learned shallow models with hand-crafted features (van Deemter et al., 2005; Belz, 2008; Konstas and Lapata, 2012). Recently, neural generation systems have shown significant progress on this task. The attention based encoder-decoder models with the copy mechanism have achieved state-of-the-art results on a few data-to-text datasets (Wiseman et al., 2017; Sha et al., 2017). In this paper, we try to address the problem of faithfully describing the data in the generated texts, i.e., how to generate a descriptive text that contains correct entities and numeric values. Intuitively speaking, when a human writer writes weather forecasts or game summaries, he may not consider the exact entities or numbers right away. Rather, he may have finished the whole sentence (without the actual entities or numbers) in his mind, then look up the data table and fill in those exact entities and numbers at last. Likewise, we don’t have to train our models to construct sentences and fill in the data at the same time. If we take out the entities and numeric values from the text, what is left …

Content truncated to 500 words