Key Fact as Pivot: A Two-Stage Model for Low Resource Table-to-Text Generation

Key Fact as Pivot: A Two-Stage Model for Low Resource Table-to-Text Generation Shuming Ma,1,3 Pengcheng Yang,1,2 Tianyu Liu,1 Peng Li,3 Jie Zhou,3 Xu Sun1,2 1MOE Key Lab of Computational Linguistics, School of EECS, Peking University 2Deep Learning Lab, Beijing Institute of Big Data Research, Peking University 3Pattern Recognition Center, WeChat AI, Tencent Inc, China {shumingma,yang pc,tianyu0421,xusun}@pku.edu.cn {patrickpli,withtomzhou}@tencent.com Abstract Table-to-text generation aims to translate the structured data into the unstructured text. Most existing methods adopt the encoder-decoder framework to learn the transformation, which requires large-scale training samples. How- ever, the lack of large parallel data is a ma- jor practical problem for many domains. In this work, we consider the scenario of low resource table-to-text generation, where only limited parallel data is available. We propose a novel model to separate the generation into two stages: key fact prediction and surface re- alization. It first predicts the key facts from the tables, and then generates the text with the key facts. The training of key fact prediction needs much fewer annotated data, while surface re- alization can be trained with pseudo parallel corpus. We evaluate our model on a biogra- phy generation dataset. Our model can achieve 27.34 BLEU score with only 1, 000 parallel data, while the baseline model only obtain the performance of 9.71 BLEU score.1 1 Introduction Table-to-text generation is to generate a descrip- tion from the structured table. It helps readers to summarize the key points in the table, and tell in the natural language. Figure 1 shows an exam- ple of table-to-text generation. The table provides some structured information about a person named “Denise Margaret Scott”, and the corresponding text describes the person with the key information in the table. Table-to-text generation can be ap- plied in many scenarios, including weather report generation (Liang et al., 2009), NBA news writ- ing (Barzilay and Lapata, 2005), biography gener- ation (Dubou´e and McKeown, 2002; Lebret et al., 2016), and so on. Moreover, table-to-text genera- 1The codes are available at https://github.com/ lancopku/Pivot. Denise Margaret Scott Born 24 April 1955 Melbourne, Victoria Nationality Australian Other names Scotty Occupation Comedian, actor, television and radio presenter Known for Studio 10 Partner(s) John Lane Children 2 Denise Margaret Scott (born 24 April 1955) is an Australian comedian, actor and television presenter. Denise Margaret Scott 24 April 1955 Australian Comedian, actor, television and radio presenter Key Fact Prediction Surface Realization Figure 1: An example of table-to-text generation, and also a flow chart of our method. tion is a good testbed of a model’s ability of un- derstanding the structured knowledge. Most of the existing methods for table-to- text generation are based on the encoder-decoder framework (Sutskever et al., 2014; Bahdanau et al., 2014). They represent the source tables with a neural encoder, and generate the text word-by- word with a decoder conditioned on the source ta- ble representation. Although the encoder-decoder framework has proven successful in the area of natural language generation (NLG) (Luong et al., 2015; Chopra et al., 2016; Lu et al., 2017; …

Content truncated to 500 words