SciGen subset

SciGen a corpus designed for reasoning-aware T2T generation, comprising tables from arXiv papers across fields such as Computation and Language, Machine Learning, Computer Science, Computational Geometry, etc.

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;
paper - (str) title of the source paper;
paper_id - (str) ID of the source paper;
table_caption - (str) caption of a given table;
table_column_names - (List[str]) table header names;
table_content_values - (List[List[str]]) table row values;
text - (str) gold description of a table;
image_id - (str) table image name;
license - (str) license type of the source paper;
venue - (str) paper source (arXiv or acl);
table_latex - (str) table in LaTeX format;
table_html - (str) table in HTML format;
table_xml - (str) table in XML format;
table_html_cleaned - (str) HTML of a table cleaned from irrelevant LaTeXML tags;
subset - (str) cl or other subset from SciGen;
field - (str) field tag for arXiv papers;
text_cleaned - (str) gold description of a table without special [CONTINUE] tokens;
table_xml_cleaned - (str) XML of a table cleaned from irrelevant LaTeXML tags.