As most of the pipeline becomes neural and through Dependency Tree-> dependency forest (tree -into-> 3D tensor), how will the workflow change?
dependency forest: differentiable parser