HiTab-StatCan-NSF

A commercially-permissive subset of Microsoft HiTab (ACL 2022) containing only the Statistics Canada and NSF table sources. The Wikipedia/ToTTo subset (CC-BY-SA-4.0, share-alike) is excluded.

License

• NSF tables (~14% of rows): sourced from US National Science Foundation reports. US federal government works are public domain under 17 U.S.C. § 105.
• StatCan tables (~86% of rows): sourced from Statistics Canada reports under the Statistics Canada Open Government Licence, which permits commercial use with attribution. Cite as: "Adapted from Statistics Canada. Reproduced and distributed on an 'as is' basis with the permission of Statistics Canada."

The original HiTab dataset is released under C-UDA-1.0. That agreement's Section 1.3 preserves your rights under any other applicable licence, so the per-source terms above govern the StatCan and NSF subsets.

Source

Built from microsoft/HiTab by filtering table_source to statcan and nsf.

Content

Hierarchical table question answering and table-to-text generation covering Canadian statistical reports (demographics, immigration, economics, health) and US NSF science and engineering statistics. Tables have multi-level row and column headers encoded as HTML with rowspan/colspan and as raw JSON.

Splits

Split	Rows	StatCan	NSF
train	6,041	5,486	555
validation	1,378	1,268	110
test	1,286	1,189	97
total	8,705	7,943	762

Schema

Field	Type	Description
id	string	Original HiTab MD5 row identifier
table_id	string	Table file key (matches HiTab tables/hmt/)
table_source	string	statcan or nsf
question	string	Natural language question
answer	list[string]	Answer values (numbers and strings both as strings)
aggregation	list[string]	Aggregation operation(s) required
sub_sentence	string	Reference table-to-text sentence
sentence_id	string	HiTab sentence group identifier
sub_sentence_id	string	HiTab sub-sentence identifier
linked_cells	string (JSON)	Cell-to-entity/quantity linkage map
answer_formulas	list[string]	Spreadsheet-style formulas for answers
reference_cells_map	string (JSON)	Formula cell reference map
table_title	string	Full table title
table_html	string	Table rendered as HTML with hierarchical rowspan/colspan headers
table_json	string	Full table as JSON (title, top_root, left_root, data trees)

Use table_source to filter to a single provenance if your licence constraints require it.

Usage

from datasets import load_dataset

# All clean rows
ds = load_dataset("rootsautomation/HiTab-StatCan-NSF")

# Public-domain only (NSF)
nsf = ds.filter(lambda x: x["table_source"] == "nsf")

# StatCan only
statcan = ds.filter(lambda x: x["table_source"] == "statcan")

Citation

If you use this dataset please cite the original HiTab paper:

@inproceedings{cheng-etal-2022-hitab,
  title     = {{HiTab}: A Hierarchical Table Dataset for Question Answering
               and Natural Language Generation},
  author    = {Cheng, Zhoujun and Dong, Haoyu and Wang, Zhiruo and Jia,
               Ran and Guo, Jiaqi and Gao, Yan and Han, Shi and Lou, Jian-Guang
               and Zhang, Dongmei},
  booktitle = {Proceedings of the 60th Annual Meeting of the Association
               for Computational Linguistics (Volume 1: Long Papers)},
  year      = {2022},
  pages     = {867--880},
}