hafsamenaz1/Group_H_Chinese-English-Code-Mixing-Prosody

A Multimodal Dataset of Phonological Shift and Prosodic Reset in Chinese-English Code-Mixing

Abstract

Existing code-mixing corpora primarily rely on text transcripts, lacking the precise acoustic alignments necessary to study prosody at the switch boundary. This dataset provides 3 hours of carefully curated, naturalistic Chinese-English code-mixed speech sourced from diverse social media video content. We utilize a dual-level annotation scheme: manual token-level labeling for phonological realization (native-like vs. local accent), interaction target, affective state, and kinesics (gestures), alongside automated extraction of fundamental frequency and intensity metrics within a ±200ms boundary window. This corpus enables linguists and NLP researchers to quantitatively evaluate the extent to which phonological accent shifts correlate with performative prosodic resets during code-switching.

Dataset Structure

• Lines.csv: Contains the token-level annotations, timestamps, and extracted acoustic features (Pitch/Intensity) for each code-mixed event.
• Clips.csv: Contains the metadata and source links for the YouTube videos used.
• audio/: Contains the raw audio files used for extraction.

Project Details

• Course: LIN3046
• Group: Group H
• Data Collection: Sourced from 6 varied YouTube vloggers.