Neural Text Generation from Structured Data with Application to the Biography Domain

TL;DR

This paper presents a neural text generation model for biographies from structured data, demonstrating significant improvements over classical models on a large, diverse dataset with over 700,000 samples.

Contribution

The paper introduces a scalable neural model for concept-to-text generation that effectively handles large vocabularies by combining fixed vocabulary with copy actions, applied to a new extensive biography dataset.

Findings

Neural model outperforms classical language models by nearly 15 BLEU.

The dataset is over 700k samples, vastly larger and more diverse than previous resources.

The model effectively transfers sample-specific words from input to output.

Abstract

This paper introduces a neural model for concept-to-text generation that scales to large, rich domains. We experiment with a new dataset of biographies from Wikipedia that is an order of magnitude larger than existing resources with over 700k samples. The dataset is also vastly more diverse with a 400k vocabulary, compared to a few hundred words for Weathergov or Robocup. Our model builds upon recent work on conditional neural language model for text generation. To deal with the large vocabulary, we extend these models to mix a fixed vocabulary with copy actions that transfer sample-specific words from the input database to the generated output sentence. Our neural model significantly out-performs a classical Kneser-Ney language model adapted to this task by nearly 15 BLEU.