Compositional Generalization via Semantic Tagging

Hao Zheng, Mirella Lapata


Abstract
Although neural sequence-to-sequence models have been successfully applied to semantic parsing, they fail at compositional generalization, i.e., they are unable to systematically generalize to unseen compositions of seen components. Motivated by traditional semantic parsing where compositionality is explicitly accounted for by symbolic grammars, we propose a new decoding framework that preserves the expressivity and generality of sequence-to-sequence models while featuring lexicon-style alignments and disentangled information processing. Specifically, we decompose decoding into two phases where an input utterance is first tagged with semantic symbols representing the meaning of individual words, and then a sequence-to-sequence model is used to predict the final meaning representation conditioning on the utterance and the predicted tag sequence. Experimental results on three semantic parsing datasets show that the proposed approach consistently improves compositional generalization across model architectures, domains, and semantic formalisms.
Anthology ID:
2021.findings-emnlp.88
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2021
Month:
November
Year:
2021
Address:
Punta Cana, Dominican Republic
Venue:
Findings
SIG:
SIGDAT
Publisher:
Association for Computational Linguistics
Note:
Pages:
1022–1032
Language:
URL:
https://aclanthology.org/2021.findings-emnlp.88
DOI:
10.18653/v1/2021.findings-emnlp.88
Bibkey:
Cite (ACL):
Hao Zheng and Mirella Lapata. 2021. Compositional Generalization via Semantic Tagging. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 1022–1032, Punta Cana, Dominican Republic. Association for Computational Linguistics.
Cite (Informal):
Compositional Generalization via Semantic Tagging (Zheng & Lapata, Findings 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.findings-emnlp.88.pdf
Video:
 https://aclanthology.org/2021.findings-emnlp.88.mp4