Latent Relation Representations for Universal Schemas

Supervised relation extraction uses a pre-defined schema of relation types (such as born-in or employed-by). This approach requires labeling textual relations, a time-consuming and difficult process. This has led to significant interest in distantly-supervised learning. Here one aligns existing database records with the sentences in which these records have been “rendered”, and from this labeling one can train a machine learning system as before [1, 2]. However, this method relies on the availability of a large database that has the desired schema.