Latent Tree Decomposition Parsers for AMR-to-Text Generation
This work addresses the challenge of efficiently processing tree-like structures in graph-based tasks like AMR-to-text generation and molecular prediction, offering incremental improvements over existing methods.
The paper tackled the problem of encoding Abstract Meaning Representations (AMRs) for text generation by proposing a model that uses tree decomposition forests instead of standard graph encoders, resulting in a BLEU increase of 0.7 and chrF++ increase of 0.3, and it also improved molecular property prediction by 1.92% ROC-AUC.
Graph encoders in AMR-to-text generation models often rely on neighborhood convolutions or global vertex attention. While these approaches apply to general graphs, AMRs may be amenable to encoders that target their tree-like structure. By clustering edges into a hierarchy, a tree decomposition summarizes graph structure. Our model encodes a derivation forest of tree decompositions and extracts an expected tree. From tree node embeddings, it builds graph edge features used in vertex attention of the graph encoder. Encoding TD forests instead of shortest-pairwise paths in a self-attentive baseline raises BLEU by 0.7 and chrF++ by 0.3. The forest encoder also surpasses a convolutional baseline for molecular property prediction by 1.92% ROC-AUC.