Skip to content

mideind/GreynirTerms

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Greynir

GreynirTerms

GreynirTerms is a utility program for generating synthetic parallel corpora of English and Icelandic sentences that contain rare terms. These rare terms are typically specialized vocabularies or glossaries for particular topic domains.

GreynirTerms is used to help Neural Machine Translation (NMT) networks learn vocabulary that occurs relatively infrequently in authentic parallel corpora.

GreynirTerms works in two distinct phases:

  1. An authentic parallel corpus and a glossary of common terms are combined to generate a template file.

  2. A template file and a dictionary of rare terms are combined to generate a synthetic parallel corpus file.

The synthetic parallel corpus can then be mixed into NMT training corpora to ensure that the network sees and learns the rare terms.

Generating a template file (phase 1)

In phase 1, to generate a template file by using the common terms as placeholders:

python main.py authentic.tsv template.tsv --template --glossary=resources/glossary.txt

The first parameter is the input (authentic corpus) file; if left out, stdin is read.

The second parameter is the output (template) file; if left out, the output is written to stdout.

The --glossary parameter identifies a glossary file containing translations of common nouns. Its format is explained below.

Generating a synthetic corpus (phase 2)

In phase 2, to generate a synthetic parallel corpus containing the rare terms:

python main.py template.tsv synthetic.tsv --generate --glossary=resources/terms.txt --count=50

The first parameter is the input (template) file generated in phase 1; if left out, stdin is read.

The second parameter is the output (synthetic corpus) file; if left out, the output is written to stdout.

The --glossary parameter identifies a terms file containing the rare terms (nouns) that should occur in the synthetic corpus, as well as their translations.

The --count parameter specifies the number of (randomly sampled) synthetic pairs that should be generated for each rare term. Since the grammatical forms of Icelandic nouns can be up to 16 (4 cases * 2 numbers * 2 definite/indefinite forms), it is advisable to generate enough examples so that the synthetic corpus contains most or all of those forms. The default value of --count is 10, which is enough for testing purposes but probably not enough for real use cases.

File formats

Authentic and synthetic corpus files, as well as template files, are tab-separated UTF-8 text files. Each line contains an English sentence, a tab character, and an Icelandic sentence, terminated by a newline character.

The common noun glossary file has the following format:

icelandic-noun-lemma/category, english-lemma [, alternative-english-lemma]*

Example:

rannsókn/kvk, investigation, study

This entry causes all forms of the lemma rannsókn, along with the corresponding English form (investigation/investigations/study/studies), to be eligible as placeholders in templates, for feminine nouns.

The rare term glossary file has the following format:

icelandic-noun-lemma/category, english-lemma-singular [, english-lemma-plural]

Example:

rauðdvergur/kk, red dwarf, red dwarves

If the English lemma is given only in the singular, a plural form is automatically generated via heuristics. These are pretty good so an explicit plural should only be required in very irregular cases.

The GreynirTerms software is careful about maintaining correspondence between genders, cases, numbers, determiners and casing (upper/lower case) within the sentence pairs in the generated corpus file. The result should thus remain grammatically correct (as long as the authentic source is correct) on both the Icelandic and the English side. However, the meaning may of course become more or less nonsensical as nouns are replaced with different ones.

Copyright and licensing

GreynirTerms is Copyright (C) 2021 Miðeind ehf.

Parts of this software are developed under the auspices of the Icelandic Government's 5-year Language Technology Programme for Icelandic, which is described here (English version here).

This software is licensed under the MIT License:

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Miðeind ehf.

If you would like to use this software in ways that are incompatible with the standard MIT license, contact Miðeind ehf. to negotiate custom arrangements.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages