
… smaller number of parameters that will still be expressive enough. In order to do this, we can use pre-trained sentence encoders like Facebook’s InferSent or Google’s Universal sentence encoder. We can also train a sentence encoder on unlabeled data from our domain using methods like skip-thou…