Abstract Syntax Networks for Code Generation and Semantic Parsing
arXiv:1704.07535
Abstract
Tasks like code generation and semantic parsing require mapping unstructured (or partially structured) inputs to well-formed, executable outputs. We introduce abstract syntax networks, a modeling framework for these problems. The outputs are represented as abstract syntax trees (ASTs) and constructed by a decoder with a dynamically-determined modular structure paralleling the structure of the output tree. On the benchmark Hearthstone dataset for code generation, our model obtains 79.2 BLEU and 22.7% exact match accuracy, compared to previous state-of-the-art values of 67.1 and 6.1%. Furthermore, we perform competitively on the Atis, Jobs, and Geo semantic parsing datasets with no task-specific engineering.
ACL 2017. MR and MS contributed equally
References in corpus (6)
- Learning to Map Sentences to Logical Form: Structured Classification with Probabilistic Categorial Grammars
- DyNet: The Dynamic Neural Network Toolkit
- DeepCoder: Learning to Write Programs
- Language to Logical Form with Neural Attention
- Latent Predictor Networks for Code Generation
- Span-Based Constituency Parsing with a Structure-Label System and Provably Optimal Dynamic Oracles
Cited by in corpus (11)
- GraphCodeBERT: Pre-training Code Representations with Data Flow
- Unsupervised Translation of Programming Languages
- Neural Program Synthesis with Priority Queue Training
- Semantic Parsing with Syntax- and Table-Aware SQL Generation
- Multi-Task Learning for Conversational Question Answering over a Large-Scale Knowledge Base
- Natural Language Processing for Requirements Formalization: How to Derive New Approaches?
- Autoencoders as Tools for Program Synthesis
- How could Neural Networks understand Programs?
- A Neural-based Program Decompiler
- Linguacodus: A Synergistic Framework for Transformative Code Generation in Machine Learning Pipelines
- Learning to Extend Program Graphs to Work-in-Progress Code