Automatic program repair (APR) is crucial to improve software reliability. Recently, neural machine translation (NMT) techniques have been proposed to automatically fix bugs. While promising, these approaches have two major issues. Their search space often does not contain the correct fix and their search strategy ignores software knowledge such as strict code syntax. Due to these two issues, existing NMT-based techniques fail to outperform template-based approaches.

We propose CURE, a new NMT-based APR technique with three major novelties. First, CURE uses a subword tokenization technique to generate a smaller search space that contains more correct fixes. Second, CURE uses a new unsupervised programming language (PL) model to learn the correct syntax of source code before the APR task. Finally, CURE introduces a new code-aware search strategy that finds more correct fixes by focusing on searching for compilable patches and patches that are close in length to the buggy code.

Our evaluation on two widely-used benchmarks shows that CURE correctly fixes 57 Defects4J bugs and 26 QuixBugs bugs, outperforming all existing APR techniques on both benchmarks.