9

Projects

11

Datasets

200K+

Annotations

Universal Decompositional Semantics

The Universal Decompositional Semantics (UDS) dataset represents a new approach to semantic annotation, breaking down complex meaning into simple, answerable questions.

Access the complete dataset through our decomp toolkit, or download individual datasets below. Learn more about our methodology in our LREC 2020 paper.

Available Datasets

Project Dataset Corpus Train Dev Test Download
Semantic Proto-Roles v1 Penn TreeBank 7800 969 969 TAR.GZ
v2 English Web TreeBank 4877 632 582 TAR.GZ
Factuality v1 English Web TreeBank 5668 652 600 TAR.GZ
v2 English Web TreeBank 22279 2660 2561 TAR.GZ
Genericity pred English Web TreeBank 26721 3274 3119 ZIP
arg English Web TreeBank 30035 3611 3500 ZIP
Event Structure pred English Web TreeBank 26701 9864 9419 ZIP
pred-arg English Web TreeBank 8878 3012 2970 ZIP
pred-pred English Web TreeBank 32975 24387 21264 ZIP
Time v1 English Web TreeBank 59593 16914 15411 ZIP
Word Sense v1 English Web TreeBank 17202 1943 1876 TAR.GZ