Diverse Demonstrations Improve In-context Compositional Generalization
This addresses a key limitation in in-context learning for semantic parsing, enabling better generalization to novel compositions, though it is incremental as it builds on existing demonstration selection methods.
The paper tackles the problem of compositional generalization in semantic parsing, where models struggle with unseen output structures, by proposing a method to select diverse demonstrations that collectively cover required structures, resulting in substantial performance improvements across three datasets.
In-context learning has shown great success in i.i.d semantic parsing splits, where the training and test sets are drawn from the same distribution. In this setup, models are typically prompted with demonstrations that are similar to the input utterance. However, in the setup of compositional generalization, where models are tested on outputs with structures that are absent from the training set, selecting similar demonstrations is insufficient, as often no example will be similar enough to the input. In this work, we propose a method to select diverse demonstrations that aims to collectively cover all of the structures required in the output program, in order to encourage the model to generalize to new structures from these demonstrations. We empirically show that combining diverse demonstrations with in-context learning substantially improves performance across three compositional generalization semantic parsing datasets in the pure in-context learning setup and when combined with finetuning.