Neural models of factuality

TL;DR

This paper introduces two neural models for event factuality prediction that outperform previous models and expands a major dataset, significantly advancing the field of factuality understanding.

Contribution

The paper presents two neural models for factuality prediction and significantly enlarges the largest event factuality dataset to date.

Findings

Models outperform previous approaches on three datasets.

Expanded dataset provides a new benchmark for factuality prediction.

Model results demonstrate improved accuracy on the extended dataset.

Abstract

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