Combining OCL and Natural Language: a Call for a Community Effort

Abstract

The growing popularity and availability of pretrained natural language models opens the door to many interesting applications combining natural language (NL) with software artefacts. A couple of examples are the generation of code excerpts from NL instructions or the verbalization of programs in NL to facilitate their comprehension.

Many of these language models have been trained with open source software datasets and therefore understand a variety of programming languages, but not OCL.

We argue that OCL needs to jump into the machine learning bandwagon or it will risk losing its appeal as a constraint specification language. For that, the key first task is to create together an OCL corpus dataset amenable for natural language processing.