bead

A Python framework for constructing, deploying, and analyzing large-scale linguistic judgment experiments with active learning.

Overview

bead implements a 6-stage pipeline for linguistic experiment design:

1. Resources: lexical items and templates with constraints
2. Templates: template filling strategies (exhaustive, random, stratified, MLM, mixed)
3. Items: experimental item construction (9 task types)
4. Lists: list partitioning with constraint satisfaction
5. Deployment: jsPsych 8.x batch experiment generation for JATOS
6. Training: active learning with GLMM support and convergence detection

Key Features

• Stand-off annotation with UUID-based references for provenance tracking
• 9 task types: forced-choice, ordinal scale, binary, categorical, multi-select, magnitude, free text, cloze, span labeling
• GLMM support: Generalized Linear Mixed Models with random effects
• Batch deployment: server-side list distribution via JATOS batch sessions
• Language-agnostic: works with any language supported by UniMorph
• Configuration-first: single YAML file orchestrates entire pipeline
• Type-safe: full Python 3.13 type hints with Pydantic v2 validation

Quick Links

• Installation: get started in 5 minutes
• Quick Start: complete tutorial in 15 minutes
• User Guide: in-depth documentation
• API Reference: complete API documentation
• Examples: gallery of example projects

Installation

uv pip install bead

For development installation:

git clone https://github.com/FACTSlab/bead.git
cd bead
uv sync --all-extras

Requirements

• Python 3.13+
• Operating Systems: macOS, Linux, Windows (WSL recommended)

Citation

If you use bead in your research, please cite:

@software{white_bead_2026,
  author = {White, Aaron Steven},
  title = {Bead: A python framework for linguistic judgment experiments with active learning},
  year = {2026},
  url = {https://github.com/FACTSlab/bead},
  version = {0.2.0}
}

License

MIT License. See LICENSE for details.