Compositional Generalization via Neural-Symbolic Stack Machines

Part of Advances in Neural Information Processing Systems 33 pre-proceedings (NeurIPS 2020)

Bibtex »Paper »Supplemental »

Bibtek download is not availble in the pre-proceeding


Xinyun Chen, Chen Liang, Adams Wei Yu, Dawn Song, Denny Zhou


<p>Despite achieving tremendous success, existing deep learning models have exposed limitations in compositional generalization, the capability to learn compositional rules and apply them to unseen cases in a systematic manner. To tackle this issue, we propose the Neural-Symbolic Stack Machine (NeSS). It contains a neural network to generate traces, which are then executed by a symbolic stack machine enhanced with sequence manipulation operations. NeSS combines the expressive power of neural sequence models with the recursion supported by the symbolic stack machine. Without training supervision on execution traces, NeSS achieves 100% generalization performance in four domains: the SCAN benchmark of language-driven navigation tasks, the task of few-shot learning of compositional instructions, the compositional machine translation benchmark, and context-free grammar parsing tasks.</p>