Does Traversal Order Matter? A Systematic Study of Tree Traversal Methods in Transformer Grammars

The paper presents a systematic study of tree traversal methods in Transformer Grammars (TGs), introducing Breadth-First Traversal (BFT) and a novel Production-Rule Traversal (PRT) alongside the traditional Depth-First Traversal (DFT). It evaluates these methods across various tree configurations and masking strategies for language modeling, syntactic generalization, and summarization, highlighting trade-offs between nested composition and global lookahead. This research is significant for practitioners as it provides insights into optimizing traversal strategies for improved performance in TG applications.