Syntactic Scaffolds

Use a small amount to syntax to do better on semantics.

Model architecture for jointly training syntax and semantics.

Syntactic scaffolds avoid expensive syntactic processing at runtime, only making use of a treebank during training, through a multitask objective. We improve over strong baselines on PropBank semantics, frame semantics, and coreference resolution, achieving competitive performance on all three tasks.

Our paper was presented at EMNLP 2018:

@inproceedings{Swayamdipta2018SyntacticSF,
  title={Syntactic Scaffolds for Semantic Structures},
  author={Swabha Swayamdipta and Sam Thomson and Kenton Lee and Luke Zettlemoyer and Chris Dyer and Noah A. Smith},
  booktitle={EMNLP},
  year={2018},
  url={https://arxiv.org/abs/1808.10485}
}