Exploring Compositional Generalization of Large Language Models

Haoran Yang, Hongyuan Lu, Wai Lam, Deng Cai


Abstract
In this paper, we study the generalization ability of large language models (LLMs) with respect to compositional instructions, which are instructions that can be decomposed into several sub-instructions. We argue that the ability to generalize from simple instructions to more intricate compositional instructions represents a key aspect of the out-of-distribution generalization for LLMs. Since there are no specialized datasets for studying this phenomenon, we first construct a dataset with the help of ChatGPT, guided by the self-instruct technique. Then, we fine-tune and evaluate LLMs on these datasets. Interestingly, our experimental results indicate that training LLMs on higher-order compositional instructions enhances their performance on lower-order ones, but the reverse does not hold true.
Anthology ID:
2024.naacl-srw.3
Volume:
Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 4: Student Research Workshop)
Month:
June
Year:
2024
Address:
Mexico City, Mexico
Editors:
Yang (Trista) Cao, Isabel Papadimitriou, Anaelia Ovalle, Marcos Zampieri, Francis Ferraro, Swabha Swayamdipta
Venue:
NAACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
16–24
Language:
URL:
https://aclanthology.org/2024.naacl-srw.3
DOI:
10.18653/v1/2024.naacl-srw.3
Bibkey:
Cite (ACL):
Haoran Yang, Hongyuan Lu, Wai Lam, and Deng Cai. 2024. Exploring Compositional Generalization of Large Language Models. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 4: Student Research Workshop), pages 16–24, Mexico City, Mexico. Association for Computational Linguistics.
Cite (Informal):
Exploring Compositional Generalization of Large Language Models (Yang et al., NAACL 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.naacl-srw.3.pdf