OMTMT4PE

This app's name stands for something like "machine translation for post-editing in OmegaT".

This is a service that runs in Ur for internal consumption of the cApStAn team. This is a proof of concept, and a web service can be created to allow partners to send requests to this service from outside Ur.

Usage

The files folder where you must drop your input files is /media/data/data/company/Apps/omtmt4pe (probably mounted as U:\Apps\omtmt4pe on a Windows machine).

Any other folder can be used if it's more convenient for you, just let us know.

Your input is different depending on what you'd like to machine-translate:

• Offline project packages
• Online team projects

The output, in any case, will be a TM containing MT translations, added to the project under tm/auto/mt (or the path provided in the configuration as the value of key destination-for-mt-tm).

The TM containing the translation from DeepL is in tm/auto/mt/deepl_YYYYMMDD.tmx.

Project packages

Drop your OMT package(s) in the files folder.

Each project will be machine-translated and moved to /media/data/data/company/Apps/omtmt4pe/done, ready to be dispatched to the post-editor.

Team projects

Drop your a repos.txt file in files folder containing one line for each repo url, e.g.

https://github.com/capstanlqc-test/TEST_bul-BGR_OMT.git
https://github.com/capstanlqc-test/TEST_ara-ZZZ_OMT.git

Each team project will be machine-translated online. The MT translations will be visible when the user downloads the project in OmegaT.

Configuration

1. Tweak configuration file config.json as suitable.
2. Put it in the files folder or inside the omegat folder of the project before you drop your files.

Options:

• keep-existing-translations (false by default): Depending on whether this key is set to true or false, the output TM containing MT translations will include existing translations from the working TM.
• destination-for-mt-tm (tm/auto/mt by default): The TM containing MT translations will be added in this location of the project before re-packing it or committing the file.

TODO (dev info)

• Log some stats: project, wordcount, language pair, engine, expense, etc.
• Segments for which there is a translation should not be sent to the MT engine if keep-existing-translations is set to true
• Fine tune the script to translate only segments that don't have a translation coming from...
  • project_save.tmx?
  • tm/auto*?
  • tm/enforce*?
  • etc.
• API-based web service for other partners or platforms to request MT for an OmegaT project (WIP)
• Add Google Translate, etc. for languages that DeepL does not support (in a list, from higher to lower preference)
• Add quality estimation (WIP)
• Use destination-for-mt-tm value (instead of hard-coded tm/auto/mt)
• Read config file mt_config.json or mt_config.xlsx from project_root/omegat (if found, ignore the one in the files folder)
• Add user options to config, e.g. user options (form of address, etc.)
• Tweak the groovy script to include such options, glossaries, etc.

Deployment (dev info)

A .env file in the app root must contain valid paths to all parameters. Copy .env-ur to .env for convenience.

Script /path/to/run.sh is scheduled to run every minute.

---

TODO:

• put config.json inside omegat package(s), or next to it in template, or in omtpkgs if found inside the project, it takes precedence over the one found next to packages

-- DIferent ways to run the thing:

• from an XX omegat project template -> instances for all languages specified in config
• from one or more omegat project packages (instances)
• frmo a team project template (todo?)
• from a list of git-hosted team projects (repos.txt)

-> create offline projects (folders) that omegat can interact with

-> output:

• if packages -> output packages moved to done
• if team projects -> commit files and add url to done/repos.txt

config:

• modules: -- mt: providers -- qe: true, models...

Tips:

• add region subtag (for mt it's easier to remove it than to add it)