Project description
Recent years have seen a number of technical advances in methods that use unannotated corpora to automatically learn representations of meaning, primarily of single words but increasingly often also of larger linguistic units. On the other hand, very significant efforts have been invested in the development of structured linguistic resources such as grammars and lexicons. The project will explore synergies between these two approaches to the acquisition of linguistic knowledge: we will apply corpus-driven methods as a way to expand and correct existing linguistic resources, and conversely we will use hand-crafted resources as additional sources of supervision when learning meaning representations automatically.Software
- SCOUSE, a tool that embeds a sense network into a word space.
