Intermediate representation

It’s about reducing many-to-many relationships into many-to-one + one-to-many. Often, adding this extra step and going around can be more efficient.

The idea of intermediate representation is powerful. There are so many examples. There is a strong pull towards a Lingua franca when a diverse set of language meets. Many platforms by big tech companies have power because they offer a single place of transaction.

The idea is conspicuous in many “translation” tasks. Pandoc translates documents by creating an intermediate representation. LLVM turns code into an intermediate representation. Language Server Protocol defines a single protocol. Well, any “protocol” or “standard” are doing the same. Deep learning models make use of this a lot, especially in the translation models.