Malawian CiTonga Tone Production Study

An exploratory phonetic study investigating how consonant context influences the pitch (F₀) of high (H) and low (L) tones in Malawian CiTonga verb stems. We present two experiments:

1. Experiment A: Compare mean F₀ of H tones following sonorant vs. voiced-obstruent onsets.
2. Experiment B: Compare mean F₀ of H vs. L tones following individual voiced obstruents (b, d, j, g, v, z).

---

Table of Contents

1. Project Overview
2. Data Description
3. Experiment A
   1. Hypotheses
   2. Visualization
   3. Statistical Test
   4. Conclusions
4. Experiment B
   1. Visualization
   2. Statistical Test
   3. Conclusions
5. Overall Discussion & Follow-up Questions
6. Prerequisites
7. Installation & Usage
8. Script Breakdown
9. Extending the Analysis
10. Data Source & Acknowledgments
11. License

---

Project Overview

We examine how the phonetic context of a verb stem onset (C1.class or individual C1) affects the realized pitch of H and L tones in CiTonga:

• Experiment A: Do H‐tone verbs exhibit different mean F₀ when the first consonant is a sonorant vs. a voiced obstruent?
• Experiment B: For voiced‐obstruent onsets (b, d, j, g, v, z), is there a difference in mean F₀ between H‐tone and L‐tone verbs?

---

Data Description

File: citonga.csv
Columns of interest:

• Tone: target tone of the verb stem ("H" or "L")
• C1, C2: first and second consonants
• C1.class: phonetic class of the first consonant ("sonorants" or "voiced obstruents")
• meanf0: mean fundamental frequency of the stem vowel (Hz)
• Additional acoustic measures (maxf0, minf0, intensity, duration, etc.)

---

Experiment A

Hypotheses

• H₀: Mean F₀ of H tones does not differ between sonorant vs. voiced-obstruent onsets.
• H₁: Mean F₀ of H tones does differ between these onset classes.

Visualization (Boxplot)

df_H <- subset(df, Tone == "H" & C1.class %in% c("sonorants","voiced obstruents"))
ggplot(df_H, aes(C1.class, meanf0)) +
  geom_boxplot() +
  labs(
    title = "Mean F₀ of H Tone by C1 Class",
    x     = "C1 Class",
    y     = "Mean F₀ (Hz)"
  ) +
  theme_minimal()

Statistical Test

sonorants_H <- subset(df, Tone=="H" & C1.class=="sonorants")$meanf0
voicedObs_H  <- subset(df, Tone=="H" & C1.class=="voiced obstruents")$meanf0
t.test(sonorants_H, voicedObs_H, var.equal=TRUE)

• t = 5.1426
• df = 23
• p-value ≈ 3.28×10⁻⁵

Since p < 0.05, we reject H₀: there is a significant difference in mean F₀.

Conclusions

1. H-tone verbs preceded by sonorants have significantly higher mean F₀ than those preceded by voiced obstruents.
2. Onset consonant class influences pitch realization of H tones in CiTonga.

---

Experiment B

Visualization (Boxplot by Individual C1)

voiced_obs <- subset(df, C1 %in% c("b","d","j","g","v","z"))
ggplot(voiced_obs, aes(C1, meanf0, fill=Tone)) +
  geom_boxplot() +
  labs(
    title = "Mean F₀ by Voiced Obstruent and Tone",
    x     = "Voiced Obstruent (C1)",
    y     = "Mean F₀ (Hz)"
  ) +
  theme_minimal()

Observation: Contrary to expectation, L-tone words often show higher mean F₀ than H-tone words after voiced obstruents.

Statistical Test

H_vo <- subset(df, Tone=="H" & C1 %in% voiced_set)$meanf0
L_vo <- subset(df, Tone=="L" & C1 %in% voiced_set)$meanf0
t.test(H_vo, L_vo, var.equal=TRUE)

• t = –5.4622
• df = 153
• p-value ≈ 1.86×10⁻⁷

p < 0.05 → reject H₀: mean F₀ differs significantly between H vs. L tones in this context.

Conclusions

1. After voiced obstruents, L-tone stems have significantly higher mean F₀ than H-tone stems.
2. This reversal suggests a context‐dependent pitch enhancement of L tones or suppression of H tones, contrary to neutralization.

---

Overall Discussion & Follow-up Questions

• Phonetic interaction: Onset consonant voicing class modulates pitch realization of both H and L tones.
• Unexpected reversal: L-tone exaggeration after voiced onsets raises questions about tonal implementation and downstep.
• Follow‐up: What phonological or aerodynamic mechanisms cause L tones to surface higher than H tones after voiced obstruents? How does this shape the tonal system and perceptual contrasts in CiTonga?

---

Prerequisites

• R ≥ 4.0
• RStudio (optional)
• Packages: ggplot2, ggmosaic (for mosaic if extended)

---

Installation & Usage

1. Clone the repository:

   git clone https://github.com/yourusername/citonga-tone-analysis.git
   cd citonga-tone-analysis

2. Place citonga.csv in the project root.

3. Run the analysis script:

   source("citonga_tone_analysis.R")

4. View the generated figures in figures/.

---

Script Breakdown

• Load data from citonga.csv.
• Experiment A: filter for H tones & compare by C1.class.
• Experiment B: filter for voiced‐obstruent onsets & compare by Tone.
• Visualize with ggplot2 boxplots.
• Test with t.test(..., var.equal=TRUE).
• Print results and interpret.

---

Extending the Analysis

• Perform two‐way ANOVA on meanf0 ~ Tone * C1.class.
• Include other acoustic measures (intensity, duration).
• Use mixed‐effects models to account for speaker or lexical random effects.
• Visualize with mosaic plots for categorical interactions.

---

Data Source & Acknowledgments

• Acoustic measurements collected in Malawian CiTonga fieldwork (unpublished).
• Thanks to field assistants and native speakers for data collection.

---

License

This project is released under the MIT License. See LICENSE for details.