Classifiers are Better Experts for Controllable Text Generation

TL;DR

This paper introduces CAIF sampling, a simple and flexible method for controllable text generation that adjusts language model logits using classifiers, outperforming existing methods in toxicity and sentiment control tasks.

Contribution

The paper presents CAIF sampling, a novel approach that leverages arbitrary classifiers to guide text generation, offering improved performance and ease of implementation over prior methods.

Findings

Outperforms PPLM, GeDi, and DExperts in toxicity and sentiment tasks

Easier to implement and tune with fewer restrictions

Significantly better metrics on external classifier evaluations

Abstract

This paper proposes a simple method for controllable text generation based on weighting logits with a free-form classifier, namely CAIF sampling. Using an arbitrary text classifier, we adjust a small part of a language model's logits and guide text generation towards or away from classifier prediction. We experimented with toxicity avoidance and sentiment control tasks and showed that the proposed method significantly outperforms recent PPLM, GeDi, and DExperts on PPL and task accuracy metrics based on the external classifier of generated texts. In addition, compared to other approaches, it is easier to implement and tune and has significantly fewer restrictions and requirements.