摘要：Modern systems consist of large numbers of languages, frameworks, libraries, APIs, and more. Each has characteristic behavior and data. Capturing these in semantics is valuable not only for understanding them but also essential for formal treatment (such as proofs). Unfortunately, most of these systems are defined primarily through implementations, which means the semantics needs to be learned. We describe the problem of learning a semantics, provide a structuring process that is of potential value, and also outline our failed attempts at achieving this so far.
