Hi all, regarding creating LPs for CJK using pretrained embeddings my plan is to implement the approach from [1] using commons-maths (read models, perform matrix multiplication / inverse, etc.) for getting a first word level translation capability; then I'd like to dig into how to extend that to phrases without incurring into having to train a neural net, because that would require adding an explicit dependency to a DL framework (which I'd like to avoid for now). Thoughts ?
Regards, Tommaso [1] : https://arxiv.org/abs/1309.4168
