The Universal Decompositional Semantics (UDS) dataset represents a unified approach to semantic annotation that combines multiple semantic annotation layers into a single graph-based representation. UDS graphs consist of directed acyclic structures where nodes and edges carry real-valued attributes, with these semantic graphs anchored to Universal Dependencies syntactic trees.
The Decomp toolkit provides Python 3 libraries for working with UDS data, offering SPARQL-based querying capabilities for semantic graph exploration. The toolkit enables researchers to access and analyze the normalized semantic annotations that span multiple linguistic phenomena including semantic roles, factuality, genericity, and temporal relations.
UDS 1.0 integrates five annotation datasets collected using crowdsourcing methodologies, where complex semantic phenomena are decomposed into simple binary questions. This decompositional approach allows naive annotators to provide reliable judgments that are then aggregated and normalized into continuous-valued semantic properties.
Publications
-
White, Aaron Steven, Elias Stengel-Eskin, Siddharth Vashishtha, Venkata Subrahmanyan Govindarajan, Dee Ann Reisinger, Tim Vieira, Keisuke Sakaguchi, Sheng Zhang, Francis Ferraro, Rachel Rudinger, Kyle Rawlins, and Benjamin Van Durme. 2020. The Universal Decompositional Semantics Dataset and Decomp Toolkit. In Proceedings of The 12th Language Resources and Evaluation Conference, 5698–5707. Marseille, France: European Language Resources Association.