Unobserved Local Structures Make Compositional Generalization Hard
This work addresses the problem of compositional generalization in semantic parsing for AI researchers, providing insights into instance-level difficulty and methods for creating adversarial splits and improving training efficiency, though it is incremental in nature.
The paper investigates why sequence-to-sequence models struggle with compositional generalization, identifying that test instances are hard if they contain unobserved local structures, and shows this criterion predicts instance-level generalization well across 5 semantic parsing datasets.
While recent work has convincingly showed that sequence-to-sequence models struggle to generalize to new compositions (termed compositional generalization), little is known on what makes compositional generalization hard on a particular test instance. In this work, we investigate what are the factors that make generalization to certain test instances challenging. We first substantiate that indeed some examples are more difficult than others by showing that different models consistently fail or succeed on the same test instances. Then, we propose a criterion for the difficulty of an example: a test instance is hard if it contains a local structure that was not observed at training time. We formulate a simple decision rule based on this criterion and empirically show it predicts instance-level generalization well across 5 different semantic parsing datasets, substantially better than alternative decision rules. Last, we show local structures can be leveraged for creating difficult adversarial compositional splits and also to improve compositional generalization under limited training budgets by strategically selecting examples for the training set.