Data

The dataset

childes-db releases are hosted as a versioned dataset on Redivis: datapages.childes_db. Each childes-db release is a Redivis dataset version, so any table from any release can be browsed, queried, downloaded, or accessed programmatically — and analyses can be pinned to an exact version for reproducibility.

childes-db is updated periodically with a fresh parse of the CHILDES database, to reflect new corpora and to revise existing annotations (see the documentation for how CHAT transcripts are treated). Visualizations on this site use the most recent release.

Releases

childes-db version Redivis tag Released Notes
2018.1 v1.0 2018 Initial release.
2019.1 v1.1 2019 Re-parsed to reflect 2019 changes in CHILDES. Note that this excludes key datasets like Providence which were moved to PhonBank.
2020.1 v1.2 2020 Re-parsed to reflect 2020 changes in CHILDES, as well as the 2020 version of PhonBank.
2021.1 v1.3 2021 Re-parsed to reflect 2021 changes in CHILDES and PhonBank, using a new version of the corpus processing code and a better set of tests.
2026.1 v1.4 2026-07-25 New chatter-based pipeline with additional annotation tiers (UD morphology and the token_morpheme table, dependency parses, romanization, speech acts, phonology, media timestamps), ~73 new corpora, and Redivis hosting. See the documentation for details, including a comparability break in morpheme-based measures.

Accessing the data

  • R (childesr). The childesr package provides tidy accessors for collections, corpora, participants, transcripts, utterances, and tokens, with a db_version argument to select a release. The Redivis-backed version currently lives on a branch; install it with:

    # install.packages("remotes")
    remotes::install_github("langcog/childesr", ref = "redivis")
  • Redivis web interface. Browse the dataset directly: inspect any table’s schema, run SQL queries in the browser, and download full tables or query results (CSV, parquet, and other formats), all without writing code.

  • Redivis clients (R / Python). For programmatic access beyond childesr, the redivis-r and redivis-python client libraries can query and fetch any table in the dataset, at any version.

  • Direct exports. Full-table exports for local analysis (e.g., parquet for DuckDB or arrow) are available from each table’s page on Redivis — no database server required.

Identifiers

Numeric ids (transcript_id, utterance_id, token_id, …) are release-internal: they are stable within a release and consistent across tables of that release, but they are not guaranteed to match across releases. To link transcripts across releases, use the TalkBank PID carried on each transcript, which is TalkBank’s persistent identifier.

Reproducibility

Pin your analysis to a version: record the childes-db version (e.g., 2021.1) and, if you access Redivis directly, reference the dataset version tag (e.g., v1.3) in your code. Every release remains available on Redivis, so a pinned analysis can be re-run as-is even after new releases appear.

Citation

If you use childes-db, please note the database version you used (e.g., 2021.1) and cite both our paper in Behavior Research Methods and CHILDES itself — both the database and the corpora you use — following the TalkBank policy. See About for the full citation policy.