Skip to content

Arabic-SQUAD train/dev/test splits #22

@spookyQubit

Description

@spookyQubit

Hi @husseinmozannar , thanks a lot for sharing the code and the data.

Had a small question regarding Arabic-Squad: In the paper, it is mentioned that Arabic-SQuAD is split 80-10-10% into three parts for training, development and testing: Arabic-SQuad-Test is composed of 2,966 questions on 24 articles; note that articles are distinct between the parts.

Is there an official split which one should use for Table 5 of the paper? The reason I ask is that the Arabic-SQuAD.json comes without any train/dev/test markings (unless I am not looking at the correct file). Will it be possible to please share the splits?

Thanks a lot in advance.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions