A Relation-Centric View of Semantic Representation Learning
Representation learning for semantics describes a class of methods that automatically induce features for semantic concepts of itnerests. These techniques hav eproven successful for many downstream MLP applications. But despite their success, model still fail to capture many basic properties of human languages, such as polysemy, antonymy and semantic composition.
Thesis Committee: Eduard Hovy (Chair) Chris Dyer Lori Levin Peter Turney (Independent Researcher)