Abstract

With a familiar language, programmers require little assistance to correct syntactic errors. But with unfamiliar languages, correcting syntactic errors without expert guidance, in the form of error messages, can be a frustrating experience. Crafting these error messages using traditional techniques requires extensive effort beyond building a parser and is rarely undertaken. Instead of building error messages manually, we are inspired by natural language processing techniques to use Bayesian statistical analysis to automatically produce high-quality parse error messages with almost no implementation effort. Our technique directly applies to combinator parsers, GLR parsers and Earley parsers. We exemplify our technique with a combinator parser for Java, where our automatically generated messages are of similar or superior quality to those created by existing techniques.