LARC
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/samacqua/larc
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是对ARC的扩展，它包含了由人类参与者为清晰传达任务而生成的自然语言任务描述。LARC数据集为ARC任务中的88%提供了描述，并评估了能够从特定领域语言生成符号程序的模型。该任务旨在研究具有自然语言描述的抽象推理任务的神经程序合成。
This dataset is an extension of the ARC, which encompasses natural language task descriptions generated by human participants to explicitly clarify and communicate the corresponding tasks. The LARC dataset offers descriptions for 88% of ARC tasks, and serves as a benchmark to evaluate models capable of generating symbolic programs from domain-specific languages. This task is specifically designed to investigate neural program synthesis for abstract reasoning tasks paired with natural language descriptions.
提供机构：
Acquaviva et al.
