Strengthening Structural Inductive Biases by Pre-training to Perform Syntactic Transformations

TL;DR

This paper proposes a pre-training method that enhances Transformers' structural inductive biases by training on syntactic transformations, improving few-shot learning and generalization in syntactic and semantic tasks.

Contribution

Introducing intermediate pre-training on syntactic transformations to strengthen Transformers' structural biases for better performance on syntactic and semantic tasks.

Findings

Pre-training improves few-shot syntactic task performance.

Enhanced structural generalization in semantic parsing.

Attention heads track syntactic transformations effectively.

Abstract

Models need appropriate inductive biases to effectively learn from small amounts of data and generalize systematically outside of the training distribution. While Transformers are highly versatile and powerful, they can still benefit from enhanced structural inductive biases for seq2seq tasks, especially those involving syntactic transformations, such as converting active to passive voice or semantic parsing. In this paper, we propose to strengthen the structural inductive bias of a Transformer by intermediate pre-training to perform synthetically generated syntactic transformations of dependency trees given a description of the transformation. Our experiments confirm that this helps with few-shot learning of syntactic tasks such as chunking, and also improves structural generalization for semantic parsing. Our analysis shows that the intermediate pre-training leads to attention heads that keep track of which syntactic transformation needs to be applied to which token, and that the model can leverage these attention heads on downstream tasks.