CAT-probing: A Metric-based Approach to Interpret How Pre-trained Models for Programming Language Attend Code Structure

TL;DR

This paper introduces CAT-probing, a novel metric-based method to interpret how pre-trained models for programming languages attend to code structure, addressing limitations of previous probing methods by considering code characteristics.

Contribution

The paper proposes CAT-probing, a new approach that quantitatively measures how well CodePTMs capture code structure using a novel CAT-score metric.

Findings

CAT-probing effectively interprets CodePTMs across multiple programming languages.

Experimental results demonstrate the high effectiveness of CAT-probing in understanding code attention.

The method provides insights into the models' ability to attend to code structure.

Abstract

Code pre-trained models (CodePTMs) have recently demonstrated significant success in code intelligence. To interpret these models, some probing methods have been applied. However, these methods fail to consider the inherent characteristics of codes. In this paper, to address the problem, we propose a novel probing method CAT-probing to quantitatively interpret how CodePTMs attend code structure. We first denoise the input code sequences based on the token types pre-defined by the compilers to filter those tokens whose attention scores are too small. After that, we define a new metric CAT-score to measure the commonality between the token-level attention scores generated in CodePTMs and the pair-wise distances between corresponding AST nodes. The higher the CAT-score, the stronger the ability of CodePTMs to capture code structure. We conduct extensive experiments to integrate CAT-probing with representative CodePTMs for different programming languages. Experimental results show the effectiveness of CAT-probing in CodePTM interpretation. Our codes and data are publicly available at https://github.com/nchen909/CodeAttention.