📌 DOI: 10.34894/A0NLTQ

This repository introduces MCSD, containing 10.57 hours of video (aligned with post-edited error-free transcriptions). The dataset achieved a Fleiss’ kappa of 0.74 (unweighted) and 0.79 (annotation certainty-weighted). Validation using a SVM achieves a 76.64% F1-score. The standardized annotation pipeline can be used to replicate the dataset curation in other languages (see Section 2.2 Annotation process below). Please see the wiki page for detailed information of the dataset: https://github.com/x-y-g/MCSD/wiki