edithatogo/corpus-nz-hathi

Historical New Zealand Parliamentary Debates from HathiTrust

Registry status

• Registry ID: edithatogo/corpus-nz-hathi
• Family: hathitrust-nz
• Repository role: derived_corpus
• Canonical dataset: edithatogo/nz-hansard-corpus
• Operational status: active
• Rights status: rights_vary_by_volume
• Authoritative catalog: edithatogo/dataset-estate-registry

Origin and provenance

• Origin repository: https://github.com/edithatogo/hathi-nz
• Upstream source: HathiTrust Digital Library collection 71329709.
• This card describes repository identity and relationships; it does not claim completeness beyond published manifests.

Rights boundary

Repository code, dataset compilation, and underlying source-record rights are separate. The registry currently records the rights state as rights_vary_by_volume. Where rights vary by record or source, downstream users must inspect the published provenance and rights fields.

Existing dataset documentation

Dataset Summary

This dataset contains 510 volumes of New Zealand Parliamentary Debates (Hansard) spanning 1854 to 1990, sourced from the HathiTrust Digital Library Collection ID 71329709 (source institution uc1 — University of California).

It is part of the NZ corpus-family under the edithatogo organization, designed for NLP research, legal analysis, historical text mining, and reproducible open science.

Source Provenance

Property	Value
Collection	NZ Parliamentary Debates (Hansard)
HathiTrust Collection ID	71329709
Source institution	uc1 (University of California)
Expected volumes	510
Date range	1854 – 1990
Primary content category	debates
Rights status	Public domain (pd) for most volumes

Dataset Structure

Configurations

Config	Description
debates	Default. Full corpus of 510 NZ Parliamentary Debates volumes (1854–1990).
legislation	Planned legislation subset for future HathiTrust-backed uploads under the same corpus namespace.

Data Fields

Each volume record in the manifest conforms to the manifests/schema.json schema (JSON Schema Draft 2020-12):

Field	Type	Description
htid	string	HathiTrust volume ID (e.g. uc1.b2889853) — primary identifier
category	string	Content category (debates, hansard, parliamentary-papers, etc.)
year	int|null	Publication year (1800–2100); null when indeterminate
volume	string	Volume number or identifier string (from enumcron)
title	string	Full bibliographic title
rights	string	Rights status code: pd, ic-world, undetermined, suppressed
collection_id	string	HathiTrust collection ID (71329709)
source	string	Source institution code (e.g. uc1)
enumcron	string	Raw enumeration/chronology string
oclc_num	string	OCLC control number
isbn	string	ISBN (if available)
issn	string	ISSN (if available)
lccn	string	Library of Congress Control Number
sha256	string	SHA-256 checksum of volume content
size_bytes	int	File size in bytes
pipeline_version	string	Pipeline version that produced the record
record_schema_version	string	"1.0" — schema version constant
corpus_id	string	"corpus-nz-hathi" — cross-corpus identifier

Collection Methodology

1. Discovery: Volumes are discovered by filtering HathiFile dumps (tab-delimited) for Collection ID 71329709.
2. Enumeration: Each matching line is parsed into a structured volume record with year extraction from imprint/title.
3. Metadata enrichment: Optionally enriched via the HathiTrust Data API (/volume/{htid}/json).
4. Download: Volume content is downloaded from HathiTrust as raw text/zip files.
5. Verification: SHA-256 checksums verify file integrity.
6. Staging: Files are organized into data/raw/ and data/processed/ with Parquet metadata tables.
7. Validation: Automated validation against JSON Schema Draft 2020-12, consistency checks, and file integrity verification.
8. Upload: Validated files are pushed to Hugging Face Hub via huggingface-hub.

Usage Examples

Python (Hugging Face Datasets)

from datasets import load_dataset

ds = load_dataset("edithatogo/corpus-nz-hathi", "debates", split="train", streaming=True)

# Preview first 5 volumes
for i, row in enumerate(ds.take(5)):
    print(f"[{row['year']}] {row['title']}")

Polars (local Parquet files)

import polars as pl

df = pl.read_parquet("data/processed/metadata/*.parquet")
print(df.head())

DuckDB (direct HF query)

Intended Use

• Legal NLP research: Training language models on historical parliamentary language
• Historical text mining: Analyzing political discourse across 136 years of New Zealand history
• Digital humanities: Studying legislative evolution, policy debates, and political trends
• Open science: Reproducible, versioned corpus with persistent identifiers

Limitations

• OCR quality: Early volumes (1854–1900) may have lower OCR quality due to typeface and page condition.
• Coverage: The corpus covers 1854–1990. Post-1990 debates and supplementary parliamentary papers are not included.
• Rights: Some volumes may have undetermined or restricted rights status (ic-world, undetermined, suppressed).
• Not legal advice: This dataset is provided for research and informational purposes, not as legal advice.

Update Cadence

• Live HF dataset: Updated daily or as new volumes are discovered/verified.
• Zenodo snapshots: Annual DOI-backed archival releases for academic citation.
• OSF mirrors: Release bundles mirrored to an OSF project for secondary redundancy, while reusing the existing Zenodo DOI in the mirrored metadata.

Release Mirrors

The OSF mirror reuses the same prepared release archives and manifest files as Zenodo. Publish steps read .osf.json for mirror-specific metadata, mirror the Zenodo DOI recorded in this card, and use OSF_TOKEN plus OSF_PROJECT_ID to upload into the configured OSF project.

Zenodo Release Workflow

GitHub Actions publishes Zenodo releases from the same packaged archive that is used for the OSF mirror. The zenodo_release.yml workflow runs on published GitHub releases or manual workflow_dispatch events with a version input, packages the archive, validates version consistency, uploads to Zenodo, and updates the citation line in this dataset card with the DOI.

Use the production input only when you want the workflow to switch away from the sandbox Zenodo API. The workflow expects ZENODO_TOKEN to be available as a repository secret.

If you need to re-check publication state for a specific publish run, dispatch .github/workflows/publication_status.yml with workflow_run_id set to the collection_publish.yml run ID. The workflow will download the hathitrust-nz-collection-publication artifact and reuse its reports/status/status_report.json snapshot.

Redundancy guidance:

• HathiFiles, OAI-PMH, and the Bibliographic API are metadata redundancy sources.
• HTRC Extracted Features and HTRC Analytics are derived-feature redundancy sources.
• Internet Archive and Open Library are interim overlap sources only.

Citation

@misc{togo2026corpusnzhathi,
  author = {Edith A. Togo},
  title = {corpus-nz-hathi: NZ Parliamentary Debates (1854--1990) from HathiTrust},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/edithatogo/corpus-nz-hathi}}
}

For live use, cite the Hugging Face repository with access date. For academic citation, use the Zenodo DOI 10.5281/zenodo.21107389. The collection-level publication map is in the HathiTrust-NZ archive registry.

Licensing

• Pipeline code: Licensed separately in the source GitHub repository.
• Source material: NZ Parliamentary Debates text is Crown copyright. This dataset card does not relicense the source material. Users should verify HathiTrust rights status for each volume before redistribution.
• Metadata: Project-created metadata (schemas, manifests, dataset cards) shared under CC-BY-4.0.
• This dataset is not legal advice.

Contact

• Hugging Face: edithatogo/corpus-nz-hathi
• GitHub: edithatogo/hathi-nz
• Organization: edithatogo

SELECT htid, year, title, volume
FROM read_parquet('hf://datasets/edithatogo/corpus-nz-hathi/metadata/*.parquet')
WHERE year BETWEEN 1890 AND 1900
ORDER BY year, volume
LIMIT 10;

PyArrow

import pyarrow.dataset as ds

dataset = ds.dataset("hf://datasets/edithatogo/corpus-nz-hathi", format="parquet")
table = dataset.to_table(columns=["htid", "year", "title", "volume"])
print(table.to_pandas().head())

Data Splits

The dataset is not pre-split. Use streaming=True and filter by year or htid for custom splits.

All volumes are enumerated from HathiFile dumps, with metadata enriched via the HathiTrust Data API. Each volume record includes provenance fields: source, collection_id, sha256 checksum, and pipeline_version.

Languages

• English (en): Primary language of the debates.
• New Zealand English (nz): Includes New Zealand-specific terminology, place names, and Māori loanwords.
• Māori (selected terms): Occasional use of te reo Māori in parliamentary speeches.

Reconciliation policy

This repository is retained only while its recorded role is distinct from the canonical dataset. Renames, deletion, visibility changes, and replacement of immutable DOI snapshots require explicit human review.