Learning to Write with Cooperative Discriminators

TL;DR

This paper introduces a method that combines RNN language models with multiple discriminative models to improve the quality, coherence, and informativeness of generated natural language text.

Contribution

It proposes a novel cooperative discriminative approach that guides RNN decoding, addressing issues of repetitiveness and lack of communicative goals in language generation.

Findings

Human evaluation shows improved preference for generated text.

Enhanced coherence, style, and information content in outputs.

Significant improvement over baseline models.

Abstract

Recurrent Neural Networks (RNNs) are powerful autoregressive sequence models, but when used to generate natural language their output tends to be overly generic, repetitive, and self-contradictory. We postulate that the objective function optimized by RNN language models, which amounts to the overall perplexity of a text, is not expressive enough to capture the notion of communicative goals described by linguistic principles such as Grice's Maxims. We propose learning a mixture of multiple discriminative models that can be used to complement the RNN generator and guide the decoding process. Human evaluation demonstrates that text generated by our system is preferred over that of baselines by a large margin and significantly enhances the overall coherence, style, and information content of the generated text.