Deeper Shallow Embeddings

Deep and shallow embeddings are two popular techniques for embedding a language in a host language with complementary strengths and weaknesses. In a deep embedding, embedded constructs are defined as data in the host: this allows for syntax manipulation and facilitates metatheoretic reasoning, but is challenging to implement - especially in the case of dependently typed embedded languages. In a shallow embedding, by contrast, constructs are encoded using features of the host: this makes them quite straightforward to implement, but limits their use in practice.
In this paper, we attempt to bridge the gap between the two, by presenting a general technique for extending a shallow embedding of a type theory with a deep embedding of its typing derivations. Such embeddings are almost as straightforward to implement as shallow ones, but come with capabilities traditionally associated with deep ones. We demonstrate these increased capabilities in a number of case studies; including a DSL that only holds affine terms, and a dependently typed core language with computational beta reduction that leverages function extensionality.