Abstract

We present two neural models for event factuality prediction, which yieldsignificant performance gains over previous models on three event factualitydatasets: FactBank, UW, and MEANTIME. We also present a substantial expansionof the It Happened portion of the Universal Decompositional Semantics dataset,yielding the largest event factuality dataset to date. We report model resultson this extended factuality dataset as well.

 

Quick Read (beta)