Common Key Terms Implementation for silnlp and Serval

[Enkidu93]

(From Michael)
Serval and silnlp rely on different implementations of the key terms logic. Differences have been identified in a number of areas, including:

• Pairing KT glosses (from Paratext datasets) and project-specific KT renderings
• Adding key term pairs to the training data (silnlp adds each KT twice; Serval adds each KT once).
• Extracting key term renderings from the project data.
• Supported gloss languages (Serval supports English, Spanish, French, Indonesian and Portuguese, while silnlp does not support Portuguese).
  A single shared implementation would simplify maintenance and enhancement of this feature as well as user support.

The most natural place for a common implementation is Machine. This will require updates to machine.py, Machine, silnlp, and Serval. Broadly, we need to:

• Expand machine.py's key term handling (extraction and alignment) to be consistent with/as flexible as silnlp's current behavior as needed.
• Replace silnlp's key term handling with machine.py's
• Port machine.py updates to Machine (C#)
• Make updates to Serval and machine.py's build_job classes as needed to use those updates

We should establish a small collection of key terms situations/projects that we can use to test consistency throughout this process.