Table-to-Text Generation with Effective Hierarchical Encoder on Three Dimensions (Row, Column and Time)

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 3143–3152, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 3143 Table-to-Text Generation with Effective Hierarchical Encoder on Three Dimensions (Row, Column and Time) Heng Gong, Xiaocheng Feng, Bing Qin∗, Ting Liu Harbin Institute of Technology, China {hgong, xcfeng, qinb, tliu}@ir.hit.edu.cn Abstract Although Seq2Seq models for table-to-text generation have achieved remarkable progress, modeling table representation in one dimen- sion is inadequate. This is because (1) the table consists of multiple rows and columns, which means that encoding a table should not de- pend only on one dimensional sequence or set of records and (2) most of the tables are time series data (e.g. NBA game data, stock mar- ket data), which means that the description of the current table may be affected by its histori- cal data. To address aforementioned problems, not only do we model each table cell consid- ering other records in the same row, we also enrich table’s representation by modeling each table cell in context of other cells in the same column or with historical (time dimension) data respectively. In addition, we develop a ta- ble cell fusion gate to combine representations from row, column and time dimension into one dense vector according to the saliency of each dimension’s representation. We evaluated our methods on ROTOWIRE, a benchmark dataset of NBA basketball games. Both automatic and human evaluation results demonstrate the ef- fectiveness of our model with improvement of 2.66 in BLEU over the strong baseline and out- performance of state-of-the-art model. 1 Introduction Table-to-text generation is an important and chal- lenging task in natural language processing, which aims to produce the summarization of numeri- cal table (Reiter and Dale, 2000; Gkatzia, 2016). The related methods can be empirically divided into two categories, pipeline model and end-to- end model. The former consists of content selec- tion, document planning and realisation, mainly for early industrial applications, such as weather ∗ Email corresponding. Player PTS AST REB … Wizards Paul Pierce 11 1 3 … Nene 8 1 7 … Bradley Beal 18 1 11 … John Wall 16 10 1 … … … … … … Kris Humphries 13 1 5 … Hornets Michael Kidd-Gilchrist 13 3 13 … Al Jefferson 18 1 12 … Gerald Henderson 17 5 2 … Brian Roberts 18 3 1 … … … … … … Gary Neal 12 1 0 … The Charlotte Hornets ( 21 - 27 ) defeated the Washington Wizards ( 31 - 18 ) 92 - 88 on Wednesday … The Hornets were led by the duo of John Wall and Bradley Beal . Wall went 4 - for - 14 from the field and 1 - for - 4 from the three - point line to score a game - high of 16 point … Gerald Henderson had a solid showing as well , finishing with 17 points ( 6 …

Content truncated to 500 words