I'm new to azure and am trying to train a translator model. When creating a project, it is asked to choose a source language and a target language. In the list, it can be seen that L1->L2 can be taken but also L2->1. From this raises my question: if I want a model that can translate from one language to another interchangeably L1<->L2, do I need to train 2 models ? One L1->L2 and the other L2->L1. Training is quite expensive and having to do it 2 times seems unpractical.
Each Custom Translator project represents a translation system in one direction. You may choose to build one direction or both directions. You can still translate in both directions if you have built the system only in one direction, Translator will use your custom system in the direction you have built it, and the generic system in the other. All projects in your workspace access the same data store, so that you can use the same set of documents building a translation system in either direction.