Reconstructing Signing Avatars from Video Using Linguistic Priors