Learning to Blame

Localizing type errors is challenging in languages with global type inference, as the type checker must make assumptions about what the programmer intended to do. We introduce N���, a data-driven approach to error localization based on supervised learning. N��� analyzes a large corpus of training data — pairs of ill-typed programs and their “�xed” versions — to automatically learn a model of where the error is most likely to be found. Given a new ill-typed program, N��� executes the model to generate a list of potential blame assignments ranked by likelihood. We evaluate N��� by comparing its precision to the state of the art on a set of over 5,000 ill-typed OC��� programs drawn from two instances of an introductory programming course. We show that when the top-ranked blame assignment is considered, N���’s data-driven model is able to correctly predict the exact sub-expression that should be changed 72% of the time, 28 points higher than OC��� and 16 points higher than the state-of-the-art SHE��L�� tool. Furthermore, N���’s accuracy surpasses 85% when we consider the top two locations and reaches 91% if we consider the top three.