Mapping interpretable concepts and features in a latent space of language