I wonder how much memory this translation via an intermediate representation of a sentence takes. It seems like representing the semantic meaning of a sentence in a language-independent way would take a huge amount of data.