Compression-based learning for OT is incompatible with Richness of the Base∗