REPLICA is a toolkit for replication and reproduction in computational psycholinguistics. It combines dataset loading, feature extraction and cross-validated model comparison through configurable interfaces.
Website · Datasets · Tutorials · Architecture · Contributing
Use Python 3.11 or newer:
python -m pip install replicaverseThe Python import and command are named replica. The base package supports
tabular analysis and precomputed predictors. The getting-started example uses
replicaverse[models,r] and requires R with lme4 for the default mixed-effects
backend. See the environment guide for setup,
optional dependencies and source installation.
Start with Getting started to load a dataset, evaluate a language-model predictor and interpret the results.
| Notebook | Topics |
|---|---|
| Benchmarking | Compare datasets and models |
| Loading datasets | Catalog, variants and loader options |
| Segmentation | Character, word and sentence units |
| Surprisal methods | Compare predictor calculations |
| Adding a study | Integrate and validate a study recipe |
Dataset walkthroughs: eye-tracking, self-paced reading, EEG and fMRI.
Explore available datasets with replica datasets or the
catalog. Download a supported source with
replica fetch DATASET; some sources require manual acquisition or licensed
materials, as described in their catalog entries.
Data and caches use automatic defaults; no path settings are required. The environment guide explains how to change their locations. The predictor-store guide covers precomputed predictors.
The architecture guide describes the package and its extension points. Follow the contributor guide and test guide to develop and validate changes, then open a PR.
Related projects: ModelBlocks, pymovements and EyeBench.