Strengthening Structural Inductive Biases by Pre-training to Perform Syntactic Transformations
This work addresses the need for better generalization in NLP models for syntactic tasks, but it is incremental as it builds on existing pre-training methods to enhance biases rather than introducing a new paradigm.
The paper tackles the problem of enhancing structural inductive biases in Transformers for seq2seq tasks involving syntactic transformations, such as active-to-passive voice conversion and semantic parsing, by proposing intermediate pre-training on synthetically generated syntactic transformations of dependency trees, which results in improved few-shot learning for tasks like chunking and better structural generalization in semantic parsing.
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.