Equivariant Transduction through Invariant Alignment
This work addresses compositional generalization in NLP models, which is crucial for handling infinite sentence constructions from finite words, but it appears incremental as it builds on prior group-equivariant approaches.
The authors tackled the problem of compositional generalization in NLP by introducing a novel group-equivariant architecture with group-invariant hard alignment, which outperformed previous group-equivariant networks on the SCAN task.
The ability to generalize compositionally is key to understanding the potentially infinite number of sentences that can be constructed in a human language from only a finite number of words. Investigating whether NLP models possess this ability has been a topic of interest: SCAN (Lake and Baroni, 2018) is one task specifically proposed to test for this property. Previous work has achieved impressive empirical results using a group-equivariant neural network that naturally encodes a useful inductive bias for SCAN (Gordon et al., 2020). Inspired by this, we introduce a novel group-equivariant architecture that incorporates a group-invariant hard alignment mechanism. We find that our network's structure allows it to develop stronger equivariance properties than existing group-equivariant approaches. We additionally find that it outperforms previous group-equivariant networks empirically on the SCAN task. Our results suggest that integrating group-equivariance into a variety of neural architectures is a potentially fruitful avenue of research, and demonstrate the value of careful analysis of the theoretical properties of such architectures.