Using pretrained embeddings for additional features

Hi,
I’ve added some additional linguistic features to the words of the source side of my parallel corpus. But I want to use pretrained embeddings for them. Is it possible?
As I understood for the documentation the pretrained embeddings could be used only for the words’ surface form. I want the model to learn the word embeddings but to use my pretained embedding vectors for other features.

Hi,

Which OpenNMT implementation are you using?