Engine lifecycle

Engines need to be re-trained from time to time – and not only because change in corpora. A trained engine is also a snapshot of what Globalese was capable of at the time of training.

Since the technology and algorithms keep evolving (much like your corpora), it will become necessary to re-train your engines every now and then. Globalese keeps a record of when an engine was last trained, and if there have been considerable improvements in the meantime, it will let you know.

The meaning of the various indicators is as follows:

Current (green)

An engine that has been trained using the current minor version of Globalese includes everything Globalese is capable of doing at the time. (Read here about semantic versioning.)

Example: The latest released Globalese version is 4.3.3, and an engine was trained on 4.3.1. The engine is considered current, because only the patch version is different.

Supported (yellow)

As time passes and improvements are made in Globalese, engines will start losing their competitive edge. They will not be performing worse or slower compared to themselves, but their quality compared to more recently trained engines will slowly decrease over time. Retraining these engines will result in improved quality, and is therefore recommended.

While an engine is being retrained, you can still use its current version for translation.

If the Globalese version superseding the version the engine was trained on was released less than 6 months ago, the engine is considered supported.

Example: The latest released Globalese version is 4.3.3, and an engine was trained on version 4.2.1. If version 4.3.0 was released less than 6 months ago, the engine is considered supported.

Unsupported (grey)

Bugs are no longer fixed for unsupported versions.

Over time, changes that “favour” more recently trained engines will get introduced. This means that some of the “older” engines may start running into errors during translation, and being slower in general. Retraining these engines will result in significantly improved quality and speed.

While an engine is being retrained, you can still use its current version for translation.

If the Globalese version superseding the version the engine was trained on was released more than 6 months but less than 9 months ago, the engine is considered unsupported.

Example: The latest released Globalese version is 4.3.3, and an engine was trained on version 4.1.4. If version 4.2.0 (which followed 4.1.x) was released more than 6 months ago, the engine is considered unsupported.

Deprecated (red)

Deprecated engines must be fully retrained before they can be used for translation.

If the Globalese version superseding the version the engine was trained on was released more than 9 months ago, the engine is considered deprecated.

Example: The latest released Globalese version is 4.3.3, and an engine was trained on version 4.0.1. If version 4.1.0 (which followed 4.0.x) was released more than 9 months ago, the engine is considered deprecated.

The number of days until deprecation is visible on the engine overview page under the column Days until deprecation.

 

To update an engine to the latest version, a full re-training is required. Quick training does not change the version of an engine. However, a full re-training is only recommended when the status is unsupported. Engines with supported status can be still safely used.