Yi-Mandarin Focus Prosody

Acoustic analysis pipeline for focus production in Yi-Mandarin bilingual speakers.

Yi (Nuosu) is a Tibeto-Burman language that lacks prosodic focus marking (no post-focus compression). When Yi-dominant bilinguals speak Mandarin -- a language that uses both on-focus F0 expansion and post-focus compression (PFC) -- do they transfer the full Mandarin prosodic system, or produce expansion without compression? This repo contains the extraction, storage, and statistical analysis pipeline for investigating that question.

Corpus

	Yi speakers	Beijing Mandarin controls
N	20 (10F, 10M)	5 (3F, 2M)
L1	Yi (Nuosu)	Beijing Mandarin
Task	Read Mandarin sentences	Read Mandarin sentences

4 focus conditions (broad / subject / verb / object focus) x 4 sentences x ~5 repetitions = ~80 recordings per speaker, ~2,000 total.

Each recording includes .wav, Praat .TextGrid (words + phones tiers), and .Pitch files.

Pipeline

wav + TextGrid
      |
      v
[extract_f0.py]  parselmouth: word-level F0 contours (10-point normalised),
                 duration, intensity, speaker-normalised semitones
      |
      v
[build_database.py]  SQLite: speakers / recordings / f0_word / f0_contour
      |               + example queries (group aggregation, delta computation,
      v                missing-data checks)
[focus_lmm.R]    RSQLite -> lme4 LMMs:
                   F0 delta ~ focus_region * group + (1|speaker)
                   + emmeans contrasts, contour plots

Usage

# 1. extract F0 features from raw corpus
python src/extract_f0.py /path/to/yi_language/ -o data/

# 2. build SQLite database
python src/build_database.py data/df_word.csv data/df_contour.csv -o data/corpus.db

# 3. run statistical analysis (requires R with lme4, RSQLite)
Rscript analysis/focus_lmm.R

Key finding

Yi-Mandarin bilinguals show on-focus F0 expansion comparable to Beijing Mandarin controls, but without post-focus compression -- consistent with L1 transfer of Yi's prosodic typology.

Citation

Li, C., et al. (2026). Prosodic realisation of focus in Yi-Mandarin bilingual speakers: on-focus expansion without post-focus compression. Under review, Interspeech 2026.

Requirements

Python >= 3.9: pip install -r requirements.txt

R: lme4, lmerTest, emmeans, RSQLite, ggplot2

Raw audio data is not included in this repository.