[4Xin1] Poster session 2

[4Xin1-62] Text Generation on Important Features of Table Data

〇Yuri Murayama^1,2, Tatsuya Ishigaki², Yui Uehara², Yusuke Miyao^3,2, Hiroya Takamura², Ichiro Kobayashi^1,2 (1.Ochanomizu University, 2.National Institute of Advanced Industrial Science and Technology, 3.The University of Tokyo)

Keywords:Data-to-Text, Text Generation, Dataset creation

We collected text describing an important feature of table data as a data-to-text dataset and generated text by T5. The features of table data are defined as 1) Change in value, 2) The most prominent difference, 3) Similarities/differences among groups, 4) Exception, 5) No change, 6) The highest/lowest value, 7) Ranking. Our aim is to generate text describing not only superficial information but also meaningful information from table data by training the model with text grasping any of these seven features. We collected text using 7,392 table data of the LogicNLG dataset through Amazon Mechanical Turk. We also conducted an experiment with our dataset and discussed the results.