It's a great work. I was wondering if there is any alignment or shared representation space among the four types of embeddings generated by the respective encoders?

Looking forward to your reply, thanks.