TreePiece: Faster Semantic Parsing via Tree Tokenization
This addresses efficiency issues for NLP practitioners working on semantic parsing tasks, though it appears incremental as it builds on existing AR and NAR methods.
The paper tackles the slow decoding speed of autoregressive models in semantic parsing by introducing TreePiece, which tokenizes parse trees into subtrees and generates one subtree per step, achieving 4.6 times faster decoding than standard AR on the TopV2 benchmark.
Autoregressive (AR) encoder-decoder neural networks have proved successful in many NLP problems, including Semantic Parsing -- a task that translates natural language to machine-readable parse trees. However, the sequential prediction process of AR models can be slow. To accelerate AR for semantic parsing, we introduce a new technique called TreePiece that tokenizes a parse tree into subtrees and generates one subtree per decoding step. On TopV2 benchmark, TreePiece shows 4.6 times faster decoding speed than standard AR, and comparable speed but significantly higher accuracy compared to Non-Autoregressive (NAR).