Helps translation layers where we expect inputs to be multiple float vectors rather than an indexed matrix.