Shrinking language models by robust approximation | IEEE Conference Publication | IEEE Xplore