English TN sentence-break rules

This artifact is derived from the English portion of the Google Text
Normalization dataset (tn-en_with_types), created for:

Richard Sproat and Navdeep Jaitly. 2016. "RNN Approaches to Text
Normalization: A Challenge."

Corpus source:
https://github.com/rwsproat/text-normalization-data/tree/master/en_with_types

The corpus is based on English Wikipedia. Attribution is also given to the
Wikipedia contributors whose work supplied the underlying text:
https://www.wikipedia.org/

Modifications: icukit's ordered rules were mined from the corpus's
sentence-boundary labels, converted to the icukit break-rules format, and
supplemented with validation witnesses synthesized from each rule's predicates,
which include lexical values mined from the corpus (e.g. `lower` token values);
no corpus sentence or row was read or copied.

The cartlet model file `sentence-tn-cart.json.gz` is a second artifact derived
the same way: a CART model trained on the same corpus's sentence-boundary labels;
no corpus text is included.

English real-text sentence-break model

The opt-in cartlet model file `sentence-real-cart.json.gz` was trained on
sentence-boundary labels from commercially compatible documents in the
Georgetown University Multilayer Corpus (GUM) and on candidate-restricted
labels from the UK House of Commons Hansard. No corpus sentence or source
record is included in the model.

GUM source repository and licensing information:
https://github.com/amir-zeldes/gum

GUM annotation repository:
https://github.com/UniversalDependencies/UD_English-GUM

The included GUM documents have source-specific CC BY 2.5, CC BY 4.0, or
CC BY-SA 3.0 terms. `REAL_TEXT_RECEIPT.json` identifies every training and
development document, its source URL, and its verified license. Documents with
noncommercial or unverified terms were excluded.

Hansard source:
https://hansard-api.parliament.uk

Contains Parliamentary information licensed under the Open Parliament Licence
v3.0:
https://www.parliament.uk/site-information/copyright-parliament/open-parliament-licence/

Modifications: the source documents were converted to candidate-boundary
feature rows; a depth-10 CART was fitted on training documents; its eight
features were selected by held-out permutation importance; and the model was
compiled for icukit's flat equality evaluator. Test data was not used for
training or selection.

The real-text model is distributed under CC BY-SA 4.0 because part of its
training text is CC BY-SA; ShareAlike is applied to the model, with attribution
as listed above and in `REAL_TEXT_RECEIPT.json`.

The rule artifacts, model artifacts, this notice, and the accompanying receipts
are distributed under the Creative Commons Attribution-ShareAlike 4.0
International license. The full license text is in LICENSE beside this file,
and the canonical license URL is
https://creativecommons.org/licenses/by-sa/4.0/.
