Benchmarking Language Models for Code Syntax Understanding
This work identifies a key limitation in pre-training methods for programming languages, which is important for researchers and developers in code understanding and generation, though it is incremental as it builds on prior benchmarking efforts.
The authors tackled the problem of assessing how well pre-trained language models understand code syntax by benchmarking state-of-the-art models on a new dataset, CodeSyntax, and found that these models fail to match simple baselines, highlighting their limitations in capturing program structure.
Pre-trained language models have demonstrated impressive performance in both natural language processing and program understanding, which represent the input as a token sequence without explicitly modeling its structure. Some prior works show that pre-trained language models can capture the syntactic rules of natural languages without finetuning on syntax understanding tasks. However, there is limited understanding of how well pre-trained models understand the code structure so far. In this work, we perform the first thorough benchmarking of the state-of-the-art pre-trained models for identifying the syntactic structures of programs. Specifically, we introduce CodeSyntax, a large-scale dataset of programs annotated with the syntactic relationships in their corresponding abstract syntax trees. Our key observation is that existing language models pretrained on code still lack the understanding of code syntax. In fact, these pre-trained programming language models fail to match the performance of simple baselines based on positional offsets and keywords. We also present a natural language benchmark to highlight the differences between natural languages and programming languages in terms of syntactic structure understanding. Our findings point out key limitations of existing pre-training methods for programming languages, and suggest the importance of modeling code syntactic structures.