Acknowledgement

This research was funded by Natural Sciences and Engineering Research Council of Canada. We wish to thank Tao Yu and Hongjin Su for running our code on the hold out test set of Spider and Jinyang Li, Binyuan Hui, Reynold Cheng, Ge Qu and the other authors of BIRD for running our code on the holdout test set of BIRD. We also wish to thank Csaba Czepesvari, Dale Schuurmans and the anonymous reviewers of NeurIPS for their constructive comments to improve this work. Language models are few-shot learners. Lgesql: line graph enhanced text-to-sql model with mixed local and non-local relations. Evaluating large language models trained on code. Large language models are few (1)-shot table reasoners. Training verifiers to solve math word problems. Few-shot table-to-text generation with prompt planning and knowledge memorization. Transforming meaning representation grammars to improve semantic parsing. Large language models are zero-shot reasoners.