Language is already abstract and eliminates much unpredictable information, allowing models to bypass explicit abstraction steps like joint embedding.
And in language, we can get away without doing this because language is already to some level abstract and already has eliminated a lot of information that is not predictable.