Abstract
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).- Anthology ID:
- 2023.findings-emnlp.740
- Volume:
- Findings of the Association for Computational Linguistics: EMNLP 2023
- Month:
- December
- Year:
- 2023
- Address:
- Singapore
- Editors:
- Houda Bouamor, Juan Pino, Kalika Bali
- Venue:
- Findings
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 11082–11092
- Language:
- URL:
- https://aclanthology.org/2023.findings-emnlp.740
- DOI:
- 10.18653/v1/2023.findings-emnlp.740
- Cite (ACL):
- Sid Wang, Akshat Shrivastava, and Aleksandr Livshits. 2023. Treepiece: Faster Semantic Parsing via Tree Tokenization. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 11082–11092, Singapore. Association for Computational Linguistics.
- Cite (Informal):
- Treepiece: Faster Semantic Parsing via Tree Tokenization (Wang et al., Findings 2023)
- PDF:
- https://preview.aclanthology.org/emnlp-22-attachments/2023.findings-emnlp.740.pdf