Sequence-to-Sequence Models for Data-to-Text Natural Language Generation: Word- vs. Character-based Processing and Output Diversity

TL;DR

This paper compares word-based and character-based sequence-to-sequence models for data-to-text generation, showing their effectiveness, differences in output diversity, and the models' ability to generalize beyond training data.

Contribution

It provides a comprehensive comparison of input representations and demonstrates neural models' capacity to learn and generalize from synthetic template data.

Findings

Models achieve comparable or better results than challenge submissions.

Character-based models produce more diverse outputs.

Neural models can generalize beyond training structures.

Abstract

We present a comparison of word-based and character-based sequence-to-sequence models for data-to-text natural language generation, which generate natural language descriptions for structured inputs. On the datasets of two recent generation challenges, our models achieve comparable or better automatic evaluation results than the best challenge submissions. Subsequent detailed statistical and human analyses shed light on the differences between the two input representations and the diversity of the generated texts. In a controlled experiment with synthetic training data generated from templates, we demonstrate the ability of neural models to learn novel combinations of the templates and thereby generalize beyond the linguistic structures they were trained on.