eduhk-compling/GroupJ_Project

Project Title

From Kana to Cantonese: Segmental and Suprasegmental Challenges for Japanese Learners

Abstract

This dataset captures segmental and suprasegmental challenges faced by Japanese learners of Cantonese. Approximately one hour of learner speech was collected from two social media creators (@aa_mung and @meiceon51 on Threads). Videos were downloaded, audio tracks separated, and segmented into single‑sentence clips using Praat. Transcriptions were manually produced, checked for accuracy, and converted into CSV format with aligned metadata. Annotations highlight segmental errors (e.g., articulation) and suprasegmental issues (e.g., tone shifts, pauses). The dataset is openly available on Hugging Face and offers reuse potential for linguistic analysis, pedagogy, and NLP error detection.

The categories of tags are as follows:

%: Indicates a pause within the utterance.

( ): Denotes a wrong articulation or phonetic error.

↑ / ↓: Describes a directional shift in tone, indicating where the pitch becomes notably higher or lower, respectively.