Neural Program Generation Modulo Static Analysis

TL;DR

This paper introduces a neurosymbolic approach that combines neural models with static analysis to improve the generation of entire Java methods, effectively capturing long-range semantic relationships and reducing semantic errors.

Contribution

The paper presents a novel method integrating static analysis with neural generation, enabling better long-horizon program synthesis and semantic correctness in generated code.

Findings

Outperforms state-of-the-art transformers in Java method generation

Produces code with fewer semantic errors

Achieves higher syntactic accuracy compared to previous models

Abstract

State-of-the-art neural models of source code tend to be evaluated on the generation of individual expressions and lines of code, and commonly fail on long-horizon tasks such as the generation of entire method bodies. We propose to address this deficiency using weak supervision from a static program analyzer. Our neurosymbolic method allows a deep generative model to symbolically compute, using calls to a static-analysis tool, long-distance semantic relationships in the code that it has already generated. During training, the model observes these relationships and learns to generate programs conditioned on them. We apply our approach to the problem of generating entire Java methods given the remainder of the class that contains the method. Our experiments show that the approach substantially outperforms state-of-the-art transformers and a model that explicitly tries to learn program semantics on this task, both in terms of producing programs free of basic semantic errors and in terms of syntactically matching the ground truth.