NeuralREG: An end-to-end approach to referring expression generation

TL;DR

NeuralREG introduces a deep neural network model for referring expression generation that jointly decides form and content without explicit features, significantly outperforming traditional baselines.

Contribution

The paper presents NeuralREG, a novel end-to-end neural approach for REG that simplifies the process and improves performance over existing methods.

Findings

NeuralREG outperforms baseline models on WebNLG corpus.

Joint decision-making enhances REG quality.

Model and data are publicly available.

Abstract

Traditionally, Referring Expression Generation (REG) models first decide on the form and then on the content of references to discourse entities in text, typically relying on features such as salience and grammatical function. In this paper, we present a new approach (NeuralREG), relying on deep neural networks, which makes decisions about form and content in one go without explicit feature extraction. Using a delexicalized version of the WebNLG corpus, we show that the neural model substantially improves over two strong baselines. Data and models are publicly available.