Posts

Showing posts from August, 2017

Can we machine-learn a programming language semantics?

In this post I will talk about whether denotational semantics of programming languages can be machine-learned and why that could be useful.

But let's set some background first.

Programming language semantics is a collection of mathematical theories about representing the meaning of programming languages in an abstract, machine-independent ways. The semantics of a language can be given either operationally, by a set of syntactic transformations which, loosely speaking, show how a program is evaluated, or can be given denotationally, by mapping any piece of code into a mathematical object, its "meaning". Operational specifications can be given to realistic languages and, as a method, is the workhorse of formal aspects of programming languages from verification to compilation. Denotational interpretations, on the other hand, are mathematically much more mathematically challenging to develop but are also more powerful. In particular, denotational models can be used to reason …