Datasets:
Tasks:
Multiple Choice
Modalities:
Text
Formats:
parquet
Sub-tasks:
multiple-choice-qa
Languages:
English
Size:
1K - 10K
ArXiv:
License:
Clarification on Redistribution of RACE Data for Non-Commercial Research
#4
by
HyeongSoo - opened
Hello, I am a graduate student conducting non-commercial academic research. I would like to clarify the scope of the RACE dataset license regarding the redistribution of modality-transformed versions of the dataset. Specifically, our research involves converting the original text passages and questions into speech using a text-to-speech (TTS) system, for the purpose of building a speech-based benchmark. The textual content itself remains unchanged, but the modality is converted from text to audio. I would like to confirm the following:
- Is it permitted to convert the RACE text passages into audio for non-commercial academic research purposes?
- Is it permitted to publicly release the generated audio files on GitHub for academic use only, with proper attribution and a clear non-commercial notice?
- Would the audio files generated from the original passages be considered “derived data” under the license?
- If redistribution of audio files is not permitted, would it be acceptable to release only the processing code and require users to generate the audio locally from the original dataset? I want to ensure full compliance with the license terms and would greatly appreciate your clarification.