Experimental Linguistics Analysis Pipelines

Comprehensive data analysis pipelines for experimental linguistic research, covering EEG, speech recordings, eye-tracking, and reaction time studies. These notebooks demonstrate reproducible workflows for processing and analyzing diverse experimental data types commonly used in phonetics, psycholinguistics, and cognitive science research.

Author: Chem Vatho, PhD
Affiliation: University of Cologne, IfL – Phonetik
Contact: chemvatho@gmail.com

---

📋 Overview

This repository contains analysis pipelines developed for the Experimental Service Hub at the Data Center for the Humanities (DCH), University of Cologne. Each notebook provides a complete, reproducible workflow from data loading to statistical analysis and visualization.

Pipeline	Data Type	Key Methods
EEG Analysis	EEG recordings	ERP analysis, time-frequency decomposition
Speech Recording	Audio files	F0 extraction, formant analysis, voice quality
Eye-tracking	Gaze data	Fixation analysis, time course visualization
Reaction Time	Behavioral data	RT distributions, mixed-effects modeling

---

🧠 EEG Analysis

Notebook: EEG_Analysis_Pipeline.ipynb

A comprehensive pipeline for processing electroencephalography (EEG) data in linguistic experiments, with focus on event-related potentials (ERPs) relevant to language processing.

Features

• Preprocessing: Bandpass filtering (0.1–30 Hz), baseline correction, artifact rejection
• ERP Analysis: N400 and P600 component extraction for semantic/syntactic processing
• Time-Frequency Analysis: Morlet wavelet decomposition for oscillatory dynamics
• Statistical Analysis: T-tests, effect sizes (Cohen's d), condition comparisons
• Visualization: Multi-channel ERP plots, topographic maps, TFR spectrograms

Sample Output

Statistical Results at Electrode Pz
══════════════════════════════════════════════════════════
N400 Component (300-500 ms):
  Congruent: M = -2.15 µV (SD = 1.82)
  Incongruent: M = -5.43 µV (SD = 2.01)
  t = 8.234, p < .001, Cohen's d = 1.72

Dependencies

mne, numpy, pandas, scipy, matplotlib, seaborn, scikit-learn

---

🎤 Speech Recording Analysis

Notebook: Speech_Recording_Analysis.ipynb

A complete acoustic phonetics pipeline using Praat (via Parselmouth) and librosa for analyzing speech recordings.

Features

• F0 Extraction: Multiple algorithms (Praat autocorrelation, pYIN, CREPE)
• Formant Analysis: F1–F4 tracking with bandwidth measurements
• Voice Quality: Jitter, shimmer, harmonics-to-noise ratio (HNR)
• Visualization: Spectrograms, formant tracks, F1-F2 vowel space plots
• Batch Processing: Automated pipeline for multiple audio files

Sample Output

Voice Quality Measures:
══════════════════════════════════════════════════════════
Jitter (local): 0.892%
Shimmer (local): 3.241%
HNR: 18.45 dB

Formant Statistics:
  F1: M = 512 Hz (SD = 89)
  F2: M = 1543 Hz (SD = 234)
  F3: M = 2651 Hz (SD = 187)

Dependencies

parselmouth, librosa, numpy, pandas, scipy, matplotlib, seaborn, soundfile

---

👁️ Eye-tracking Analysis

Notebook: Peekbank_Eyetracking_Analysis.ipynb

Analysis pipeline for visual world paradigm eye-tracking experiments using the Peekbank database framework.

Features

• Data Processing: Fixation extraction, area-of-interest (AOI) mapping
• Time Course Analysis: Proportion of looks over time
• Growth Curve Analysis: Polynomial and GAM modeling of looking behavior
• Statistical Analysis: Cluster-based permutation tests, bootstrapped CIs
• Visualization: Time course plots, heatmaps, individual differences

Repository

See the dedicated repository: github.com/chemvatho/Peekbank-Analysis

---

⏱️ Reaction Time Analysis

Notebook: Reaction_Time_Analysis.ipynb

A psycholinguistics-focused pipeline for analyzing reaction time data from lexical decision and similar paradigms.

Features

• Preprocessing: Outlier removal (absolute bounds + SD-based), accuracy filtering
• Distribution Analysis: RT histograms, Q-Q plots, log transformation assessment
• Effect Analysis: Lexicality, word frequency, semantic priming effects
• ANOVA: Repeated measures with pairwise comparisons (Bonferroni correction)
• Mixed-Effects Models: Linear mixed models with random slopes for subjects/items
• Visualization: Interaction plots, by-subject distributions, effect size plots

Sample Output

Repeated Measures ANOVA (Frequency × Priming):
══════════════════════════════════════════════════════════════════════
Source          SS        DF       MS         F       p-unc    η²
──────────────────────────────────────────────────────────────────────
frequency    125432.1      1   125432.1   89.234    <.001   0.124
priming       45621.8      1    45621.8   32.451    <.001   0.045
freq*prime     2341.2      1     2341.2    1.665     .207   0.002

Dependencies

pandas, numpy, scipy, statsmodels, pingouin, matplotlib, seaborn

---

..............................

🚀 Quick Start

..............................

Option 1: Google Colab (Recommended)

Click the "Open in Colab" badge on any notebook to run directly in your browser—no installation required.

Option 2: Local Installation

# Clone the repository
git clone https://github.com/chemvatho/experimental-linguistics-pipelines.git
cd experimental-linguistics-pipelines

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Launch Jupyter
jupyter notebook

Requirements File

# requirements.txt
numpy>=1.21.0
pandas>=1.3.0
scipy>=1.7.0
matplotlib>=3.4.0
seaborn>=0.11.0
statsmodels>=0.13.0
pingouin>=0.5.0
mne>=1.0.0
librosa>=0.9.0
parselmouth>=0.4.0
scikit-learn>=1.0.0
soundfile>=0.10.0

---

📊 Datasets

These pipelines are designed to work with the following open datasets:

Data Type	Dataset	Source
EEG	EEG Dataset	Kaggle
EEG	ZuCo (reading EEG)	OSF
Speech	Speech Accent Archive	Kaggle
Speech	LJSpeech	Kaggle
Eye-tracking	Peekbank	peekbank.stanford.edu
Reaction Time	British Lexicon Project	crr.ugent.be
Reaction Time	MALD Database	Springer

---

📁 Repository Structure

experimental-linguistics-pipelines/
│
├── README.md                          # This file
├── requirements.txt                   # Python dependencies
├── LICENSE                            # MIT License
│
├── notebooks/
│   ├── EEG_Analysis_Pipeline.ipynb
│   ├── Speech_Recording_Analysis.ipynb
│   └── Reaction_Time_Analysis.ipynb
│
├── data/                              # Sample data (gitignored for large files)
│   └── .gitkeep
│
├── outputs/                           # Analysis outputs
│   ├── figures/
│   └── results/
│
└── utils/                             # Helper functions
    ├── __init__.py
    ├── preprocessing.py
    └── visualization.py

---

📚 Related Publications

• Chem, Vatho. (in prep.). Adapting forced alignment for Khmer, a low-resource language.
• Chem, Vatho. (in prep.). The Illustration of IPA: Khmer (Phnom Penh Dialect). Journal of the IPA.
• Chem, Vatho. (2020). Khmer Vowel System: Structure and Variation. CJBAR, 2(2).

---

🔗 Related Resources

• Khmer G2P Tool: huggingface.co/spaces/Vatho/Khmer-g2p
• Peekbank Analysis: github.com/chemvatho/Peekbank-Analysis
• University Profile: ifl.phil-fak.uni-koeln.de/phonetik/personen/vatho-chem-ma

---

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

🙏 Acknowledgments

• Supervisors: Prof. Dr. Martine Grice & PD Dr. Constantijn Kaland (University of Cologne)
• Mentors: Prof. Dr. Reinhold Greisbach & T. Mark Ellison (University of Cologne)
• Funding: KAAD (Katholischer Akademischer Ausländer-Dienst)
• Tools: Praat, MNE-Python, librosa

---

🙏 AI Assisted tools

The repository was developed with the assistance of Claude (Anthropic), which supported logical reasoning and algorithm design. Its integration with NotebookLM led to the creation of a pilot project, which is documented on GitHub.

---

Developed for the Experimental Service Hub, Data Center for the Humanities (DCH)
University of Cologne, Germany