What does Transformer learn about source code?

Abstract

In the field of source code processing, the transformer-based representation models have shown great powerfulness and have achieved state-of-the-art (SOTA) performance in many tasks. Although the transformer models process the sequential source code, pieces of evidence show that they may capture the structural information (e.g., in the syntax tree, data flow, control flow, etc.) as well. We propose the aggregated attention score, a method to investigate the structural information learned by the transformer. We also put forward the aggregated attention graph, a new way to extract program graphs from the pre-trained models automatically. We measure our methods from multiple perspectives. Furthermore, based on our empirical findings, we use the automatically extracted graphs to replace those ingenious manual designed graphs in the Variable Misuse task. Experimental results show that the semantic graphs we extracted automatically are greatly meaningful and effective, which provide a new perspective for us to understand and use the information contained in the model.

What do pre-trained code models know about code?

“…They find that CAT-scores for source code models are correlated with their performance on code summarization, and that the CAT-scores vary per layer and per language, with a tendency for higher scores in the earlier layers. Zhang et al [43] define a similar metric over the attention matrix, an aggregated attention score, with which they can derive a graph of relationships between tokens.…”