Metadata-Version: 2.5
Name: denckring-en-data
Version: 0.6.0
Summary: English data for denckring: pronunciation, nouns, glosses, graded words, frequency
Author-email: Dennis Senzel <dennis.senzel@posteo.com>
License-Expression: Apache-2.0 AND CC-BY-4.0 AND BSD-2-Clause AND HPND-sell-variant
License-File: LICENSE
License-File: LICENSE-CMUDICT
License-File: LICENSE-LEIPZIG
License-File: LICENSE-SCOWL
License-File: LICENSE-WORDNET
License-File: NOTICE
Classifier: Intended Audience :: Science/Research
Classifier: Topic :: Text Processing :: Linguistic
Classifier: Typing :: Typed
Requires-Python: >=3.11
Requires-Dist: denckring==0.6.0
Description-Content-Type: text/markdown

# denckring-en-data

English data for [denckring](https://github.com/senzelden/denckring): exact syllable
counts and rhyme from a pronouncing dictionary, and the nouns, glosses, graded words and
word frequencies the lexicon rows read.

```console
pip install denckring[en]
```

Installing this package replaces the core English pack's spelling heuristic with
dictionary lookups. Nothing in the API changes — the same procedures simply answer more
accurately, and the `estimated_words` metric on every syllabic report drops to zero for
words the dictionary knows.

## Data

Three vendored sources, all redistributed rather than downloaded, so neither
installation nor use touches the network:

- The CMU Pronouncing Dictionary, 135,166 entries, under its BSD-2-Clause licence —
  see `LICENSE-CMUDICT`.
- The Open English WordNet 2024, under its CC BY 4.0 licence — see `LICENSE-WORDNET`.
  `nouns.txt` (56,468 single-word noun lemmas, in dictionary order) and
  `glosses.txt.gz` (78,866 lemmas, 134,298 glosses across every part of speech) are
  both derived from it. `metadata.json` beside the data records the exact version and
  generation date; `scripts/build_lexicon.py` regenerates both files from OEWN alone.
  `synonyms.json.gz` and `antonyms.json.gz` add strict single-word relations
  (44,948 and 6,039 source lemmas). `scripts/build_relations.py` reproduces them
  with wn 1.1.1 and OEWN 2024; metadata records rules, counts and SHA-256 hashes.
  Dictionary relations do not resolve word sense in context.
- SCOWL 2020.12.07, under the terms in `LICENSE-SCOWL`. `graded_words.txt.gz` holds
  77,078 words, each with the SCOWL size band it first appears at, built from the
  `english-words` and `american-words` lists at sizes 10 through 60 — 60 being the
  largest size SCOWL's own documentation is confident carries no misspellings. The
  band is the point: `room` at 10 and `tinsel` at 35 are equally words, and only one
  of them is a common one, which is what a word list built for membership alone
  cannot say. `scripts/build_graded_words.py` regenerates it.
  `everyday_exclusions.txt` is this project's own: seven band-10 entries (`payed`,
  `numbest`, `cs`, …) that `denckring.words()` leaves out of its everyday view, each
  with its reason. The graded table itself keeps them (ADR 0051).
- The Leipzig Corpora Collection's `eng_news_2023_1M`, under CC BY 4.0 — see
  `LICENSE-LEIPZIG`. `frequencies.txt.gz` holds 49,631 of the graded words, each with
  how often it occurs in lowercase in a million sentences of 2023 news: a frequency,
  where SCOWL's bands are size classes. `scripts/build_frequencies.py` regenerates it
  from a pinned download (ADR 0052).

The code in this package is Apache-2.0, like denckring itself. The data is not — see
`NOTICE` for which file carries which terms.

## Proverb starter corpus

`proverbs()` exposes three sayings with editorial prefix/suffix boundaries for
perverb. Their attestation is Thomas Preston's *A Dictionary of English Proverbs
and Proverbial Phrases*, entries 123, 1367 and 1532, available in public-domain text
at https://www.gutenberg.org/files/39281/39281-h/39281-h.htm. ADR 0048 in the core
repository records the selected boundaries and limited six-pairing scope.
