Datasets:

---

danish-foundation-models

/

norwegian-dynaword

Like 7

Follow

Danish Foundation Models 174

Hi! First, thank you for maintaining this dataset.

The current lovdata subset comes from the Norwegian Colossal Corpus, based on Lovdata CDs/DVDs last published in 2005. So it is historical material.

I maintain Lovverk — an open, Git-versioned corpus of current Norwegian law. It is rebuilt every day from Lovdata's official public-data API. Right now it contains about 5,880 documents: 759 laws (lover) and 5,121 central regulations (sentrale forskrifter). The license is NLOD 2.0 with attribution to Lovdata. The pipeline is deterministic and open source.

• Corpus: https://github.com/bartoszkobylinski/lovverk
• Engine: https://github.com/bartoszkobylinski/lovspor

I would like to contribute it as a new subset, and keep the existing historical lovdata subset unchanged.

Before I write any code, I want to ask two questions:

1. Naming. Should this be a new subset named lovverk? Or would you prefer something like lovdata-current?
2. Deduplication. Some laws from 2005 still exist today, so there is overlap with the historical lovdata subset. How do you want this handled? I can run exact and fuzzy overlap detection first and report the numbers before adding anything.

My plan for the contribution: a create.py that downloads a pinned Lovverk revision (Git commit SHA), selects current documents from the manifest, and maps them to the Dynaword schema (id, text, source, added, created, token_count). Because the revision is pinned, every Dynaword release stays reproducible. Daily updates upstream mean the subset can be refreshed in later releases.

This should complement #25 (court case law) — Lovverk covers statutes and regulations, not court decisions.

Happy to adjust anything to fit your process.