Abstract

Natural language (NL) to code suggestion systems assist developers in programming Integrated Development Environments (IDEs) by translating NL utterances into compilable code snippet. This facilitates the programmers to answer “how to” questions while using a new language or API, without the need to adequately familiarize themselves with the new programming languages, and the latest commonly used open source APIs. The development of these systems poses several challenges because of the lack of large (NL, code) annotated corpora, the misalignment between NL and code tokens, the syntactic and semantic requirements of the target code, and its contextual dependencies.

The current approaches mainly involve hard-coded, rule-based systems based on semantic parsing. These systems make heavy use of hand-crafted rules that map patterns in NL or elements in its syntax parse tree to various query constructs and can only work on a limited subset of NL with a restricted NL syntax. These systems are unable to extract semantic information from the coding intents of the developer, and fail to infer types, names, the context of the source code, and the developer’s intent to get accurate system-level code suggestions. Hence, these systems are non-scalable for real-time deployment.

In this master thesis, we present novel sequence-to-sequence deep learning models and training paradigms to map NL to general-purpose programming languages that can assist users with suggestions of source code snippets, given a NL intent, and also extend auto-completion functionality of the source code to users while they are writing a code. These developed architectures incorporate contextual awareness into neural models which generate source code tokens directly instead of generating parse trees/abstract meaning representations from the source code and converting them back to source code.

The proposed pretraining strategy and the data augmentation techniques improve the performance of the proposed architectures. An ablation study of the system was performed to gauge the important components of the developed AI system. The proposed Seq2Seq-BART has been found to exceed the performance of a neural semantic parser, TranX, based on the BLEU-4 metric by 10.82%. Thereafter, a finer analysis for the parsable code translations from the NL intent for CoNaLA¹ challenge was introduced. The proposed system is bidirectional as it can be also used to generate NL code documentation given source code. Lastly, a RoBERTa masked language model for Python code was proposed to extend the developed system for code completion.

Download paper here

Recommended citation: Mitodru Niyogi. (2021) “Neural Models for Source Code Synthesis and Completion”, IWR, Heidelberg University, Germany.