Neurosymbolic Repair for Low-Code Formula Languages

  • 2022-07-24 16:56:03
  • Rohan Bavishi, Harshit Joshi, José Pablo Cambronero Sánchez, Anna Fariha, Sumit Gulwani, Vu Le, Ivan Radicek, Ashish Tiwari
  • 0

Abstract

Most users of low-code platforms, such as Excel and PowerApps, write programsin domain-specific formula languages to carry out nontrivial tasks. Often userscan write most of the program they want, but introduce small mistakes thatyield broken formulas. These mistakes, which can be both syntactic andsemantic, are hard for low-code users to identify and fix, even though they canbe resolved with just a few edits. We formalize the problem of producing suchedits as the last-mile repair problem. To address this problem, we developedLaMirage, a LAst-MIle RepAir-engine GEnerator that combines symbolic and neuraltechniques to perform last-mile repair in low-code formula languages. LaMiragetakes a grammar and a set of domain-specific constraints/rules, which jointlyapproximate the target language, and uses these to generate a repair enginethat can fix formulas in that language. To tackle the challenges of localizingthe errors and ranking the candidate repairs, LaMirage leverages neuraltechniques, whereas it relies on symbolic methods to generate candidaterepairs. This combination allows LaMirage to find repairs that satisfy theprovided grammar and constraints, and then pick the most natural repair. Wecompare LaMirage to state-of-the-art neural and symbolic approaches on 400 realExcel and PowerFx formulas, where LaMirage outperforms all baselines. Werelease these benchmarks to encourage subsequent work in low-code domains.

 

Quick Read (beta)

loading the full paper ...