cstr/grundwortschatz-voc-en

WortUniversum English Vocabulary Database

Status: published at cstr/grundwortschatz-voc-en (CC-BY-SA 4.0). The shipped app asset is assets/grundwortschatz_en.db.gz in the WortUniversum / words-universe repository; this dataset is its CC-BY-SA 4.0 re-distribution form. Rebuild + re-push with pipeline/build_hf_datasets.py --upload.

Dataset summary

A UK-English lexical database of 11,539 lemmas covering primary-school vocabulary (CEFR-J A1–B2, YLE Starters/Movers/Flyers, UK Year 1–6 statutory lists), with multi-source enrichment per word:

• Wiktionary definitions, IPA pronunciation, inflections, examples
• Open English WordNet (OEWN) sense records: synonyms, antonyms, hypernyms, hyponyms
• wordfreq Zipf score, occurrences-per-million, frequency band (1–5)
• CEFR-J v1.5 level tags (A1–B2) for 6,879 entries
• Cambridge YLE Starters / Movers / Flyers vocabulary tags (833 entries)
• UK DfE statutory spelling lists: Year 1–2, 3–4, 5–6 (318 entries)
• Curriculum-derived gradeLevelEstimate (1–6) for all entries
• 27,363 common learner error pairs on 4,653 entries (Norvig + Wikipedia)
• 264 UK ↔ US dialect spelling variants across 150 entries (SCOWL)
• Grade-differentiated example sentences (LLM-generated, all 6 grades)
• Project Gutenberg example sentences from 72 public-domain EN books
• SQLite FTS5 search index compatible with the app runtime

The vocabulary is a superset of curriculum targets; tags control which subset the app surfaces per user level.

Languages

• en (English) — primary; all lemma, definition, enrichment, and example content
• de (German) — translations (empty in current artifact; added by future pipeline step)

Dataset structure

The dataset ships as a SQLite database (grundwortschatz_en.db.gz). The Parquet companion files (words.parquet, translations.parquet, examples.parquet) are produced by the export script; see Loading below.

Table: words

Column	Type	Description
id	INTEGER PK	Surrogate key
original_id	TEXT UNIQUE	Stable build identifier (word + pos slug)
word	TEXT	English surface form (UK spelling canonical)
lemma	TEXT	Lemma
article	TEXT	a / an hint for nouns, nullable
genus	TEXT	Always NULL (English has no grammatical genus)
word_type	TEXT	POS token: noun, verb, adjective, adverb, etc.
grade_level	INTEGER	1–6 grade estimate (synced from gradeLevelEstimate)
audio_path	TEXT	Audio URL/path, nullable
frequency_json	TEXT JSON	{"zipf": float, "per_million": float, "frequency_band": 1–5}
enrichment_json	TEXT JSON	Wiktionary data, OEWN senses, misspellings, variants, tags
metadata_json	TEXT JSON	CEFR level, grade examples, Gutenberg examples, curriculum sources

enrichment_json shape

{
  "enrichment_status": "success" | "minimal" | "no_data",
  "definitions": ["..."],
  "pronunciation": {"ipa": "...", "audio": "..."},
  "inflections": [...],
  "examples": [...],
  "synonyms": [...],
  "antonyms": [...],
  "hypernyms": [...],
  "hyponyms": [...],
  "wordnetSenses": [...],          // OEWN per-synset records
  "commonLearnerErrors": [         // common misspellings
    {"error": "recieve", "source": "norvig"},
    {"error": "recieve", "source": "wikipedia"}
  ],
  "spellingVariants": [            // UK / US dialect variants
    {"variant": "color", "dialect": "american"},
    {"variant": "colour", "dialect": "british"}
  ],
  "tags": ["source:uk_y1_y2", "source:cambridge_yle_starters", ...],
  "sources": [...]
}

metadata_json shape

{
  "cefr_level": "A1" | "A2" | "B1" | "B2",
  "yle_level": "starters" | "movers" | "flyers",
  "gradeLevelEstimate": 1,          // 1–6 (UK Year grade equivalent)
  "gutenberg_examples": ["..."],    // up to 5 Gutenberg sentences
  "grade_examples": {               // LLM-generated per-grade examples
    "1": ["...", "..."],
    "2": [...], "3": [...], "4": [...], "5": [...], "6": [...]
  },
  "tags": ["source:uk_y3_y4", ...],
  "sources": [...]
}

Table: translations

Empty in the current artifact (no DE translations step for EN words). Schema: id, word_id, lang_code, translation.

Table: examples

18,642 sentences from Wiktionary and Project Gutenberg example extraction. Schema: id, word_id, sentence.

Table: phrasal_verbs

400 English phrasal verbs (added 2026-05-29) powering the in-app phrasal-verb games. Source: Wiktionary "English phrasal verbs" categories (CC BY-SA 4.0) + LLM grade-leveled example sentences. Content-filtered for a K-6 audience.

Column	Type	Notes
id	INTEGER PK
word_id	INTEGER FK → words.id	the base verb's row
phrasal	TEXT	e.g. give up
base_verb / particle	TEXT	give / up
meaning	TEXT	short kid-friendly definition
senses_json	TEXT (JSON)	up to 3 Wiktionary glosses
distractors_json	TEXT (JSON)	wrong particles (real PVs of the same base verb)
examples_json	TEXT (JSON)	{"1":[…],…,"6":[…]} grade-leveled sentences (398/400)
wiktionary_examples_json	TEXT (JSON)	CC-BY-SA fallback sentences
grade_band	INTEGER	3–6
base_zipf	REAL	base-verb frequency (ranking signal)
source / license	TEXT	wiktionary / CC-BY-SA-4.0

gradeLevelEstimate decision tree

Priority: lowest cap wins.

1. source:uk_y1_y2 or source:cambridge_yle_starters → cap ≤ 2
2. source:uk_y3_y4 or source:cambridge_yle_movers → cap ≤ 3
3. source:uk_y5_y6 or source:cambridge_yle_flyers → cap ≤ 4
4. CEFR A1 → cap ≤ 2; A2 → ≤ 3; B1 → ≤ 4; B2 → ≤ 5
5. Frequency band baseline: band1→1, band2→2, band3→3, band4→5, band5→6
6. Result: min(cap, freq_baseline). Stored in both metadata_json.gradeLevelEstimate and the grade_level column.

Loading

Export the SQLite asset to Parquet with:

cd pipeline/voc-en
python3 export_to_parquet.py --db grundwortschatz_en.db

This writes hf_export/words.parquet, hf_export/translations.parquet, and hf_export/examples.parquet.

Load with pandas:

import pandas as pd
words = pd.read_parquet("words.parquet")

Load with DuckDB:

import duckdb
conn = duckdb.connect()
conn.execute("SELECT word, grade_level, frequency_json FROM 'words.parquet' LIMIT 10").fetchdf()

Citation

If you use this dataset in research, please cite the primary upstream sources:

• Wiktionary: https://www.wiktionary.org/ (CC BY-SA 4.0)
• OEWN: McCrae et al., Open English WordNet 2024, https://en-word.net/ (CC BY 4.0)
• CEFR-J: Tono, Y. & Negishi, M. (2012). CEFR-J Wordlist, TUFS (CC BY-SA 4.0)
• wordfreq: Speer, R. (2023). wordfreq, Luminoso Technologies (Apache-2.0 + CC BY-SA 4.0)

Source attribution

Source	License	What it contributes
Wiktionary EN	CC BY-SA 4.0	Definitions, IPA, inflections, examples
Open English WordNet (oewn:2024)	CC BY 4.0	Sense records, synonyms, hypernyms
wordfreq (Luminoso)	Apache-2.0 + CC BY-SA 4.0	frequency_json (Zipf, per_million, band)
CEFR-J v1.5 (Tono & Negishi, TUFS)	CC BY-SA 4.0	cefr_level tags
Cambridge YLE word lists	Factual (not copyrightable)	yle_level tags
UK DfE statutory spelling lists	OGL v3 — commercial OK	source:uk_y* tags, grade signal
Project Gutenberg texts (72 books)	Public domain (pre-1928)	Gutenberg example sentences
Norvig spell-errors.txt	MIT code + CC BY-SA data	commonLearnerErrors
Wikipedia common misspellings	CC BY-SA 4.0	commonLearnerErrors
SCOWL / en-wl (Kevin Atkinson)	MIT-like permissive	spellingVariants (UK↔US)

License

Effective license: CC BY-SA 4.0.

The dominant copyleft inputs are CC BY-SA 4.0 (Wiktionary, wordfreq, CEFR-J, Wikipedia/Norvig misspellings). OGL v3.0, public-domain (Gutenberg), and CC BY 4.0 (OEWN) sources are all compatible with redistribution under CC BY-SA 4.0.

The Flutter application code is under a separate proprietary license. CC BY-SA on the shipped data does not affect the application source code — it only applies when redistributing derivative data works.

Changes made from upstream sources: filtered to ~11.5k lemmas covering UK primary-school curriculum; data reshaped into normalized SQLite schema; grade-level estimates computed from multi-source decision tree; common learner errors deduplicated and attributed per source.