Improving Compositional Generalization with Latent Structure and Data Augmentation
About
Generic unstructured neural networks have been shown to struggle on out-of-distribution compositional generalization. Compositional data augmentation via example recombination has transferred some prior knowledge about compositionality to such black-box neural models for several semantic parsing tasks, but this often required task-specific engineering or provided limited gains. We present a more powerful data recombination method using a model called Compositional Structure Learner (CSL). CSL is a generative model with a quasi-synchronous context-free grammar backbone, which we induce from the training data. We sample recombined examples from CSL and add them to the fine-tuning data of a pre-trained sequence-to-sequence model (T5). This procedure effectively transfers most of CSL's compositional bias to T5 for diagnostic tasks, and results in a model even stronger than a T5-CSL ensemble on two real world compositional generalization tasks. This results in new state-of-the-art performance for these challenging semantic parsing tasks requiring generalization to both natural language variation and novel compositions of elements.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Semantic Parsing | GeoQuery (i.i.d.) | Exact Match Accuracy93.3 | 32 | |
| Semantic Parsing | GeoQuery compositional | Accuracy89.3 | 29 | |
| Semantic Parsing | COGS (generalization) | Accuracy (Generalization)99 | 25 | |
| Semantic Parsing | SMCalFlow-CS (16-C) | Accuracy61.4 | 20 | |
| Instruction Following | SCAN jump | Accuracy99.7 | 18 | |
| Semantic Parsing | GeoQuery (Len.) | Exact Match Accuracy67.8 | 17 | |
| Semantic Parsing | COGS | Accuracy99.5 | 9 | |
| Semantic Parsing | SMCalFlow-CS (32-C) | Accuracy70.4 | 8 | |
| Semantic Parsing | SMCalFlow CS (8-C) | -- | 8 | |
| Semantic Parsing | SMCalFlow-CS (0-C) | -- | 8 |