HiTZ/elkarhizketak-RAG

Dataset Card for ElkarHizketak RAG and its Disruptor Variants

Base and disruptor variants of ElkarHizketak, built to stress-test conversational RAG systems in Basque under realistic interaction patterns (conversational openings, topic shifts).

Dataset Details

Dataset Description

This dataset extends ElkarHizketak with a base variant (rewritten opening queries, retrieval-needed labels, retrieved chunks) and disruptor variants that inject conversational openings and topic shifts, to evaluate robustness of retrieval-augmented generation pipelines.

• Language(s): Basque (eu)
• License: Copyright (C) by Ixa Taldea, University of the Basque Country UPV/EHU. This dataset is licensed under the Creative Commons Attribution-ShareAlike 4.0 International Public License (CC BY-SA 4.0). To view a copy of this license, visit https://creativecommons.org/licenses/by-sa/4.0/legalcode.

Dataset Sources

• Original dataset: ElkarHizketak, created by Ixa Taldea, University of the Basque Country (UPV/EHU), licensed under CC BY-SA 4.0.
• Original paper: Otegi et al. (2020), "Conversational Question Answering in Low Resource Scenarios: A Dataset and Case Study for Basque," LREC.

Uses

Direct Use

Intended for academic research and benchmarking of conversational RAG systems in Basque. Non-commercial use only, per the CC BY-NC 4.0 license.

Out-of-Scope Use

This dataset is not intended, and should not be used, for:

• Training or fine-tuning AI models intended for commercial deployment.
• Any military, defense, surveillance, or law enforcement application.
• Any use that violates the non-commercial terms of the CC BY-NC 4.0 license.

Dataset Structure

The dataset has 4 configurations, each with dev and test splits:

• base: minimally modified ElkarHizketak with rewritten opening queries, retrieval-needed labels, and retrieved chunks.
• disruptor-1: adds conversational opening turns.
• disruptor-2a: adds topic shifts at the end of conversations.
• disruptor-2b: adds topic shifts mid-conversation.

Citation

If you use this dataset, please cite the accompanying thesis:

@mastersthesis{briones2026elkarhizketak,
  author = {Briones Basauri, Telmo},
  title  = {When Retrieval Meets Conversation: Uncovering the Limitations of Self-Reflective Conversational RAG in Basque},
  school = {University of the Basque Country (UPV/EHU)},
  type   = {Master's thesis},
  year   = {2026},
  note   = {Advisors: Ander Barrena Madinabeitia and Joseba Fernández de Landa Aguirre. HiTZ, Basque Center for Language Technology. Available via ADDI},
  url    = {http://academico.ehu.eus/handle/123456789/74383}
}

Please also cite the original ElkarHizketak dataset, on which this work is based:

@inproceedings{otegi2020conversational,
  title     = {Conversational Question Answering in Low Resource Scenarios: A Dataset and Case Study for Basque},
  author    = {Otegi, Arantxa and Agirre, Aitor and Campos, Jon Ander and Soroa, Aitor and Agirre, Eneko},
  booktitle = {Proceedings of the 12th Language Resources and Evaluation Conference (LREC 2020)},
  year      = {2020},
  url       = {https://aclanthology.org/2020.lrec-1.55.pdf}
}