numericNLG/README.md

numericNLG subset

numericNLG a dataset focusing on the T2T generation task with numerical reasoning based on tables and their textual descriptions extracted from ACL Anthology articles.

Images of tables are stored in the .zip folders, while textual formats of tables and all metadata are availabe in .json files.

Data Fields

• instance_id - (str) unique ID of a given instance;
• table_id_paper - (str) table and source paper IDs;
• caption - (str) caption of a given table;
• row_header_level - (int);
• row_headers - (List[List[str]]) table row header names;
• column_header_level - (int);
• column_headers - (List[List[str]]) table column header names;
• contents - (List[List[str]]) table row values;
• metrics_loc - (str);
• metrics_type - (List[str]);
• target_entity - (List[str]);
• table_html - (str) table in HTML format from the source numericNLG without tables' captions;
• table_name - (str) title of a table (e.g., Table 5);
• table_id - (str) ID of a given table;
• paper_id - (str) ID of the source paper;
• page_no - (str);
• dir - (str);
• description - (str) gold description of a table;
• class_sentence - (List[str]);
• sentences - (List[str]);
• header_mention - (List[str]);
• valid - (int);
• image_id - (str) table image name;
• venue - (str) paper source (arXiv or acl);
• license - (str) license type of the source paper;
• table_xml - (str) table in XML format;
• table_latex - (str) table in LaTeX format;
• table_html_cleaned - (str) HTML of a table with tables' captions and cleaned from format indicators.