Learning Syntax Without Planting Trees: Understanding Hierarchical Generalization in Transformers

TL;DR

This paper investigates how transformer models trained on language data naturally learn hierarchical syntactic structures, focusing on the influence of training objectives and internal subnetworks, revealing a preference for hierarchical generalization.

Contribution

The study identifies that language modeling objectives promote hierarchical generalization in transformers and uncovers internal subnetworks responsible for different generalization behaviors.

Findings

Language modeling objectives lead to hierarchical generalization.

Pruning reveals subnetworks with different structural biases.

Transformers prefer hierarchical explanations when they exist.

Abstract

Transformers trained on natural language data have been shown to learn its hierarchical structure and generalize to sentences with unseen syntactic structures without explicitly encoding any structural bias. In this work, we investigate sources of inductive bias in transformer models and their training that could cause such generalization behavior to emerge. We extensively experiment with transformer models trained on multiple synthetic datasets and with different training objectives and show that while other objectives e.g. sequence-to-sequence modeling, prefix language modeling, often failed to lead to hierarchical generalization, models trained with the language modeling objective consistently learned to generalize hierarchically. We then conduct pruning experiments to study how transformers trained with the language modeling objective encode hierarchical structure. When pruned, we find joint existence of subnetworks within the model with different generalization behaviors (subnetworks corresponding to hierarchical structure and linear order). Finally, we take a Bayesian perspective to further uncover transformers' preference for hierarchical generalization: We establish a correlation between whether transformers generalize hierarchically on a dataset and whether the simplest explanation of that dataset is provided by a hierarchical grammar compared to regular grammars exhibiting linear generalization.