ibm-research/NQTablesRetrieval

NQTables Retrieval

This dataset is part of a Table + Text retrieval benchmark. Includes queries and relevance judgments across train, dev, test split(s), with corpus in 3 format(s): corpus_linearized, corpus_md, corpus_structure.

Configs

Config	Description	Split(s)
default	Relevance judgments (qrels): qid, did, score	train, dev, test
queries	Query IDs and text	train_queries, dev_queries, test_queries
corpus_linearized	Linearized table representation	corpus_linearized
corpus_md	Markdown table representation	corpus_md
corpus_structure	Structured corpus with headers, cells, meta_data. text field corresponds to linearized Text + Table.	corpus_structure

corpus_structure additional fields

Field	Type	Description
meta_data	string	Table metadata / caption
headers	list[string]	Column headers
cells	list[string]	Flattened cell values

TableIR Benchmark Statistics

Dataset	Structured	#Train	#Dev	#Test	#Corpus
OpenWikiTables	✓	53.8k	6.6k	6.6k	24.7k
NQTables	✓	9.6k	1.1k	1k	170k
FeTaQA	✓	7.3k	1k	2k	10.3k
OTT-QA (small)	✓	41.5k	2.2k	--	8.8k
MultiHierTT	✗	--	929	--	9.9k
AIT-QA	✗	--	--	515	1.9k
StatcanRetrieval	✗	--	--	870	5.9k
watsonxDocsQA	✗	--	--	30	1.1k

Citation

If you use TableIR Eval: Table-Text IR Evaluation Collection, please cite:

@misc{doshi2026tableir,
  title        = {TableIR Eval: Table-Text IR Evaluation Collection},
  author       = {Doshi, Meet and Boni, Odellia and Kumar, Vishwajeet and Sen, Jaydeep and Joshi, Sachindra},
  year         = {2026},
  institution  = {IBM Research},
  howpublished = {https://huggingface.co/collections/ibm-research/table-text-ir-evaluation},
  note         = {Hugging Face dataset collection}
}

All credit goes to original authors. Please cite their work:

@inproceedings{herzig-etal-2021-open,
    title = "Open Domain Question Answering over Tables via Dense Retrieval",
    author = {Herzig, Jonathan  and
      M{\"u}ller, Thomas  and
      Krichene, Syrine  and
      Eisenschlos, Julian},
    editor = "Toutanova, Kristina  and
      Rumshisky, Anna  and
      Zettlemoyer, Luke  and
      Hakkani-Tur, Dilek  and
      Beltagy, Iz  and
      Bethard, Steven  and
      Cotterell, Ryan  and
      Chakraborty, Tanmoy  and
      Zhou, Yichao",
    booktitle = "Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = jun,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.naacl-main.43/",
    doi = "10.18653/v1/2021.naacl-main.43",
    pages = "512--519",
    abstract = "Recent advances in open-domain QA have led to strong models based on dense retrieval, but only focused on retrieving textual passages. In this work, we tackle open-domain QA over tables for the first time, and show that retrieval can be improved by a retriever designed to handle tabular context. We present an effective pre-training procedure for our retriever and improve retrieval quality with mined hard negatives. As relevant datasets are missing, we extract a subset of Natural Questions (Kwiatkowski et al., 2019) into a Table QA dataset. We find that our retriever improves retrieval results from 72.0 to 81.1 recall@10 and end-to-end QA results from 33.8 to 37.7 exact match, over a BERT based retriever."
}