@article{assis-etal-2026-esg,
    title = "{ESG}-{QA}: Building a Dataset for Question Answering on Environmental, Social, and Governance Pillars",
    author = "Assis, Gabriel  and
      Surica, Ayrton  and
      Kroll, Pedro  and
      Mendes, Gabriela Aires  and
      Rabbani, Darian  and
      Bollis, Edson  and
      Pellicer, Lucas Francisco Amaral Orosco  and
      Paes, Aline",
    editor = "Piperidis, Stelios  and
      Bel, N{\'u}ria  and
      van den Heuvel, Henk  and
      Ide, Nancy  and
      Krek, Simon  and
      Toral, Antonio",
    journal = "International Conference on Language Resources and Evaluation",
    volume = "main",
    month = may,
    year = "2026",
    address = "Palma de Mallorca, Spain",
    publisher = "ELRA Language Resource Association",
    url = "https://preview.aclanthology.org/ingest-lrec/2026.lrec-main.420/",
    pages = "5377--5388",
    abstract = "Environmental, Social, and Governance (ESG) factors are becoming increasingly central to corporate accountability and sustainable development. However, benchmarks for evaluating large language models (LLMs) in this domain remain scarce. To alleviate this gap, we present ESG-QA, a dataset of 87,261 question{--}answer{--}context triplets spanning the three ESG pillars. ESG-QA was built using an LLM-based Question Answer (QA) generation pipeline, enhanced through rule-based and semantic filtering, and validated by human inspection, enabling both abstractive QA and retrieval-augmented setups. We benchmark three open-weight LLM families (Llama-3, Gemma-3, and Qwen-3) across multiple dimensions, including correctness, environmental impact, and readability. Results show that Qwen-3 with retrieval achieves the highest absolute QA performance, while Gemma-3 provides the strongest overall balance between correctness, efficiency, and clarity. By releasing ESG-QA and its generation framework, this work establishes a comprehensive benchmark for advancing ESG-oriented QA and promoting more transparent and responsible AI evaluation."
}