Replication Package:

Repository: https://github.com/DevSecOpsLab-CSIE-NPU/austronesian-replication

---

Overview

This repository provides the complete replication package for the above paper. It contains all analysis scripts, pre-computed results, pre-registration documents, and figures for the six-family cross-linguistic migration-gradient workflow.

Families covered: Austronesian, Indo-European, Sino-Tibetan, Bantu, Uralic (pre-registered null), Turkic (pre-registered null)

Core method: CI-disjoint signal classification — each (family, concept) pair is mechanically labelled as selective anchor, founder-effect baseline, borrowing signature, ceiling effect, or null/untestable based on 500-bootstrap 95% confidence intervals.

---

Repository Structure

├── scripts/
│   ├── xfam_an_migration_gradient.py        # Austronesian
│   ├── xfam_an_expanded_concepts.py
│   ├── xfam_ie_migration_gradient.py        # Indo-European
│   ├── xfam_ie_cultural_concepts.py
│   ├── xfam_ie_homeland_sensitivity.py
│   ├── xfam_st_migration_gradient.py        # Sino-Tibetan
│   ├── xfam_st_pooled_mountain.py
│   ├── xfam_st_stedt_full_extraction.py
│   ├── xfam_st_stedt_expanded_extraction.py
│   ├── xfam_st_stedt_rice_poc.py
│   ├── xfam_bantu_migration_gradient.py     # Bantu
│   ├── xfam_bantu_cultural_concepts.py
│   ├── xfam_5th_uralic_migration_gradient.py   # Pre-registered null
│   ├── xfam_6th_turkic_migration_gradient.py   # Pre-registered null
│   ├── xfam_meta_analysis_v4.py            # Cross-family summary (primary)
│   ├── xfam_robustness_alt_tests.py        # Inferential robustness
│   ├── xfam_check_data_availability.py
│   ├── draw_stedt_pipeline.py              # Figure: STEDT pipeline
│   ├── generate_decision_figure.py         # Figure: forest plot
│   ├── generate_new_figures.py
│   └── visualization.py
├── src/austronesian/                        # Python package (distance functions, ABVD client)
├── results/                                 # Pre-computed outputs (JSON + CSV)
│   ├── xfam_meta_v4.json                   # Primary cross-family results
│   ├── xfam_meta_v4_long.csv
│   ├── xfam_robustness.json                # Robustness check results
│   ├── xfam_robustness_table.csv
│   └── xfam_*.json / xfam_*.csv           # Per-family outputs
├── figures/
│   ├── fig_xfam_v4_forest.png             # Figure 1: forest plot
│   ├── fig_xfam_v4_scorecard.png          # Figure 2: scorecard
│   └── fig_stedt_pipeline.png             # Figure 3: STEDT pipeline
├── preregistration/
│   ├── paper-2-5th-family-preregistration.md   # Uralic (null verdict)
│   └── paper-2-6th-family-preregistration.md   # Turkic (null verdict)
└── pyproject.toml

---

Requirements

• Python 3.9+

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

Key dependencies: pandas, numpy, scipy, matplotlib, requests, tqdm

---

Data Sources

The workflow ingests data from Lexibank CLDF datasets. These must be downloaded separately:

Family	Dataset	Source
Austronesian	ABVD	https://abvd.eva.mpg.de/
Indo-European	IECoR	https://github.com/lexibank/iecor
Sino-Tibetan	sagartst / zhangst / STEDT	https://github.com/lexibank/sagartst ; https://stedt.berkeley.edu/
Bantu	Grollemund 2015	https://github.com/lexibank/grollemund
Uralic	DIACL	https://github.com/diacl
Turkic	(see script for source)	—

Pre-computed results for all families are included in results/ — re-running ingestion scripts is not required for replication.

---

Pipeline Execution

Per-family gradient scripts

# Austronesian
python scripts/xfam_an_migration_gradient.py

# Indo-European (primary Kurgan homeland)
python scripts/xfam_ie_migration_gradient.py
# Sensitivity: Anatolian alternative homeland
python scripts/xfam_ie_homeland_sensitivity.py

# Sino-Tibetan (STEDT + Lexibank)
python scripts/xfam_st_stedt_full_extraction.py
python scripts/xfam_st_migration_gradient.py

# Bantu
python scripts/xfam_bantu_migration_gradient.py

# Pre-registered families (Uralic, Turkic)
python scripts/xfam_5th_uralic_migration_gradient.py
python scripts/xfam_6th_turkic_migration_gradient.py

Cross-family meta-analysis

python scripts/xfam_meta_analysis_v4.py   # → results/xfam_meta_v4.json

Robustness checks

python scripts/xfam_robustness_alt_tests.py  # → results/xfam_robustness.json

Figure regeneration

python scripts/generate_decision_figure.py   # forest plot
python scripts/draw_stedt_pipeline.py        # STEDT pipeline diagram

---

Key Constants (fixed across all scripts)

Constant	Value	Description
RANDOM_SEED	42	np.random.default_rng(42)
N_BOOTSTRAP	500	Bootstrap resamples for CI
NLD_THRESHOLD	0.4	STEDT Method B cognate clustering

---

Pre-Registration

Two families (Uralic, Turkic) were pre-registered before gradient computation. The pre-registration files in preregistration/ are version-controlled with no edits after the "Pre-registration signed off" line. Both produced null verdicts — the workflow is not a yes-machine.

---

Software Versions

Python 3.12; pandas 2.3; numpy 2.4; scipy 1.17; matplotlib 3.10.

---

License

Code: MIT License
Data: Subject to respective source database terms (ABVD, STEDT, Lexibank — academic, non-commercial)