Automated Generation of Executable Cross-Language Background Knowledge

Anonymous

TL;DR: Framework to automate the creation of executeable background knowledge

Abstract: Certain problems require specific semanticknowledge to be solved, e.g. the transformation “January→1”.Without the requireddomain specific knowledge, a program synthesizer can’t generalize towards a program thatalso learns the other months of the year. Online repositories of code contain many of thesetransformations and they can be used to improve the semantic strength of a synthesis technique. By reusing previouslywritten code that contains parts of or the entiretransformation, these domain specific problemscan now be solved as well. This backgroundknowledge, in the form of method declarations,is however not easily available in a uniform format. The lack of formalization of this processresults in handcrafted crawling and parsing software for each program synthesis technique, ascan be found in recent works. Moreover, when the program space spans over multiple programminglanguages these techniques have to be adapted bythe synthesis researchers for each separate language. This work tries to formalize the processof generating executable background knowledgein multiple programming languages that can beused in different synthesis techniques.Suchframework would lower the amount of time spenton crawling, parsing and runtime environmentswithin future research projects.