Does syntax need to grow on trees? Sources of hierarchical inductive bias in sequence-to-sequence networks
This addresses the problem of understanding inductive biases in neural networks for syntactic generalization, which is incremental as it compares existing architectures rather than introducing new ones.
The study investigated which architectural factors in neural sequence-to-sequence models influence generalization on syntactic tasks like English question formation and tense reinflection, finding that only tree-structured models consistently induced a hierarchical bias, while others, such as LSTMs and GRUs, showed varied inductive biases.
Learners that are exposed to the same training data might generalize differently due to differing inductive biases. In neural network models, inductive biases could in theory arise from any aspect of the model architecture. We investigate which architectural factors affect the generalization behavior of neural sequence-to-sequence models trained on two syntactic tasks, English question formation and English tense reinflection. For both tasks, the training set is consistent with a generalization based on hierarchical structure and a generalization based on linear order. All architectural factors that we investigated qualitatively affected how models generalized, including factors with no clear connection to hierarchical structure. For example, LSTMs and GRUs displayed qualitatively different inductive biases. However, the only factor that consistently contributed a hierarchical bias across tasks was the use of a tree-structured model rather than a model with sequential recurrence, suggesting that human-like syntactic generalization requires architectural syntactic structure.