BEIR ArguAna (orgrctera/beir_arguana)

Overview

This release packages the ArguAna slice of the BEIR (Benchmarking IR) benchmark as a flat, table-oriented dataset for retrieval evaluation and tooling. ArguAna targets counterargument retrieval: given an argument as the query, a system must rank passages from a corpus so that the best counterargument (human-judged) appears at the top.

The underlying text comes from the ArguAna Counterargument Corpus (debate-style arguments, originally tied to content from the online debate portal idebate.org). Wachsmuth et al. (ACL 2018) frame the task as finding a counterargument without prior topic knowledge—models must balance topical overlap with the query and stance opposition, which makes purely lexical matching brittle and rewards semantic understanding.

In the BEIR benchmark suite, ArguAna is listed under argument retrieval. The standard BEIR statistics report on the order of ~1.4k queries, a corpus of ~8.7k argument passages, and one gold relevant document per query on average.

This Hub dataset is a tabular export aligned with other CTERA benchmark releases: one row per query with expected_output as a JSON string of relevant corpus document IDs (BEIR qrels-style supervision).

Task

• Task type: Retrieval for counterargument finding (dense/sparse retrieval, re-ranking, or RAG-style pipelines over the BEIR ArguAna corpus).
• Input (input): Natural-language argument text (the retrieval query).
• Reference (expected_output): A JSON string listing relevant corpus document IDs with scores (typically binary 1 for the counterargument), e.g. [{"id": "test-law-...", "score": 1}]. Evaluators retrieve against the full BEIR ArguAna corpus (distributed with BEIR; not duplicated in every row) and score with standard IR metrics (nDCG@k, MRR, Recall@k, etc.).
• Metadata: Original BEIR query_id and split name are preserved.

Background

ArguAna Counterargument Corpus (source data)

The corpus was introduced to study automatic retrieval of counterarguments—a core problem in computational argumentation and deliberation support. Arguments are paired with human-selected best counterarguments from opposing-stance candidates; the task stresses stance-aware similarity (same issue, opposite position) rather than generic semantic similarity.

• Original resource (Webis / dataset landing): ArguAna data · Zenodo record (community data publication for the counterargument experiments)

BEIR reformulation

BEIR (Thakur et al., 2021) standardizes many retrieval datasets into a common corpus + queries + qrels layout so that zero-shot and cross-domain comparisons of lexical, sparse, dense, and re-ranking systems are straightforward. For ArguAna, queries and corpus passages are both argumentative texts; in the BEIR packaging, query texts also appear in the corpus (the full pool of arguments), which is expected for this benchmark and discussed in the BEIR community (e.g. related discussion).

• Code / resources: UKPLab/beir
• Related Hub layout: BeIR/arguana · BeIR/beir-corpus

Data fields

Column	Type	Description
id	string	Stable UUID for this row in this Hub release.
input	string	Query text (an argument seeking a counterargument).
expected_output	string	JSON array of {"id": "<corpus-doc-id>", "score": 1} objects — relevant (counterargument) document IDs in the BEIR corpus.
metadata.query_id	string	BEIR ArguAna query identifier.
metadata.split	string	Split name (see below).

Splits

Split	Rows
test	1,406
Total	1,406

Examples

Illustrative rows as stored (IDs and text from the dataset).

Example 1 — counterargument retrieval (test)

• input: It better enables Attorneys to advance their client's case An attorney's main duty is owed to their client. Under Rule 1.04 of the Solicitors' Code of Conduct a solicitor "must act in the best interests of each client". [1] It is part of the adversarial system that we have that two opposing parties in litigation argue for their best interests. The whole working of the adversarial system of justice is that each party knows the facts but argues the facts that most support their case. To take away client-attorney privilege is to undermine this way of achieving justice. [1] Rule 1: Core duties, Solicitors' Code of Conduct 2007, accessed 18/05/11
• expected_output:

[{"id": "test-law-lghbacpsba-con02b", "score": 1}]

• metadata.query_id: test-law-lghbacpsba-con02a · metadata.split: test

Example 2 — economic / policy argument (test)

• input (excerpt): Public and private institutions should hire people based on skills not gender to achieve positive economic impact Businesses advance when they hire the best person for a job who can unite people and create value. These qualities are individual and enhanced through training rather than not gender-specific. [...] The empirical data from Norway, for example, reveals that after being exposed to a severe limitation on their choice of directors, boards experienced large declines in value. [1] [...]
• expected_output:

[{"id": "test-society-ghbgqeaaems-con02b", "score": 1}]

• metadata.query_id: test-society-ghbgqeaaems-con02a · metadata.split: test

References

BEIR (benchmark)

Abstract (arXiv:2104.08663): Existing neural information retrieval (IR) models have often been studied in homogeneous and narrow settings, which has considerably limited insights into their out-of-distribution (OOD) generalization capabilities. To address this, and to facilitate researchers to broadly evaluate the effectiveness of their models, we introduce Benchmarking-IR (BEIR), a robust and heterogeneous evaluation benchmark for information retrieval. We leverage a careful selection of 18 publicly available datasets from diverse text retrieval tasks and domains…

• BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models — arXiv:2104.08663
• NeurIPS 2021 Datasets & Benchmarks (OpenReview)

ArguAna / counterargument retrieval (source task)

Paper (ACL 2018): Retrieval of the Best Counterargument without Prior Topic Knowledge — proposes a topic-independent approach that combines evidence of shared aspects with opposing stance when ranking counterarguments.

• ACL Anthology — P18-1023
• PDF

Citation

If you use the BEIR benchmark formulation, cite:

@inproceedings{thakur2021beir,
  title={{BEIR}: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models},
  author={Thakur, Nandan and Reimers, Nils and R{\"u}ckl{\'e}, Andreas and
          Srivastava, Abhishek and Gurevych, Iryna},
  booktitle={Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2)},
  year={2021},
  url={https://openreview.net/forum?id=wCu6T5xFjeJ}
}

For the original counterargument retrieval work and corpus lineage, cite:

@inproceedings{wachsmuth-etal-2018-retrieval,
  title = {Retrieval of the Best Counterargument without Prior Topic Knowledge},
  author = {Wachsmuth, Henning and Syed, Shahbaz and Stein, Benno},
  editor = {Gurevych, Iryna and Miyao, Yusuke},
  booktitle = {Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  month = jul,
  year = {2018},
  address = {Melbourne, Australia},
  publisher = {Association for Computational Linguistics},
  url = {https://aclanthology.org/P18-1023/},
  doi = {10.18653/v1/P18-1023},
  pages = {241--251}
}

Provenance

Exported for retrieval evaluation (e.g. Langfuse / internal tooling) with ArguAna as the BEIR sub-benchmark arguana. Corpus argument texts are not inlined in each row; join expected_output document IDs to the BEIR ArguAna corpus when building an index.